sys/netpfil/pf/pf.c

   1 /*-
   2  * Copyright (c) 2001 Daniel Hartmeier
   3  * Copyright (c) 2002 - 2008 Henning Brauer
   4  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  *    - Redistributions of source code must retain the above copyright
  12  *      notice, this list of conditions and the following disclaimer.
  13  *    - Redistributions in binary form must reproduce the above
  14  *      copyright notice, this list of conditions and the following
  15  *      disclaimer in the documentation and/or other materials provided
  16  *      with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  21  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  22  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  26  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  *
  31  * Effort sponsored in part by the Defense Advanced Research Projects
  32  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  33  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  34  *
  35  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43 #include "opt_bpf.h"
  44 #include "opt_pf.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/bus.h>
  48 #include <sys/endian.h>
  49 #include <sys/hash.h>
  50 #include <sys/interrupt.h>
  51 #include <sys/kernel.h>
  52 #include <sys/kthread.h>
  53 #include <sys/limits.h>
  54 #include <sys/mbuf.h>
  55 #include <sys/md5.h>
  56 #include <sys/random.h>
  57 #include <sys/refcount.h>
  58 #include <sys/socket.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/taskqueue.h>
  61 #include <sys/ucred.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_var.h>
  65 #include <net/if_types.h>
  66 #include <net/if_vlan_var.h>
  67 #include <net/route.h>
  68 #include <net/radix_mpath.h>
  69 #include <net/vnet.h>
  70
  71 #include <net/pfvar.h>
  72 #include <net/if_pflog.h>
  73 #include <net/if_pfsync.h>
  74
  75 #include <netinet/in_pcb.h>
  76 #include <netinet/in_var.h>
  77 #include <netinet/in_fib.h>
  78 #include <netinet/ip.h>
  79 #include <netinet/ip_fw.h>
  80 #include <netinet/ip_icmp.h>
  81 #include <netinet/icmp_var.h>
  82 #include <netinet/ip_var.h>
  83 #include <netinet/tcp.h>
  84 #include <netinet/tcp_fsm.h>
  85 #include <netinet/tcp_seq.h>
  86 #include <netinet/tcp_timer.h>
  87 #include <netinet/tcp_var.h>
  88 #include <netinet/udp.h>
  89 #include <netinet/udp_var.h>
  90
  91 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  92
  93 #ifdef INET6
  94 #include <netinet/ip6.h>
  95 #include <netinet/icmp6.h>
  96 #include <netinet6/nd6.h>
  97 #include <netinet6/ip6_var.h>
  98 #include <netinet6/in6_pcb.h>
  99 #include <netinet6/in6_fib.h>
 100 #include <netinet6/scope6_var.h>
 101 #endif /* INET6 */
 102
 103 #include <machine/in_cksum.h>
 104 #include <security/mac/mac_framework.h>
 105
 106 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 107
 108 /*
 109  * Global variables
 110  */
 111
 112 /* state tables */
 113 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 114 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 115 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 116 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 117 VNET_DEFINE(struct pf_kstatus,           pf_status);
 118
 119 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 120 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 121 VNET_DEFINE(int,                         altqs_inactive_open);
 122 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 123
 124 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 125 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 126 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 127 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 128 VNET_DEFINE(int,                         pf_tcp_secret_init);
 129 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 130 VNET_DEFINE(int,                         pf_tcp_iss_off);
 131 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 132 VNET_DECLARE(int,                        pf_vnet_active);
 133 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 134
 135 static VNET_DEFINE(uint32_t, pf_purge_idx);
 136 #define V_pf_purge_idx  VNET(pf_purge_idx)
 137
 138 /*
 139  * Queue for pf_intr() sends.
 140  */
 141 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 142 struct pf_send_entry {
 143         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 144         struct mbuf                     *pfse_m;
 145         enum {
 146                 PFSE_IP,
 147                 PFSE_IP6,
 148                 PFSE_ICMP,
 149                 PFSE_ICMP6,
 150         }                               pfse_type;
 151         struct {
 152                 int             type;
 153                 int             code;
 154                 int             mtu;
 155         } icmpopts;
 156 };
 157
 158 STAILQ_HEAD(pf_send_head, pf_send_entry);
 159 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 160 #define V_pf_sendqueue  VNET(pf_sendqueue)
 161
 162 static struct mtx pf_sendqueue_mtx;
 163 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 164 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 165 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 166
 167 /*
 168  * Queue for pf_overload_task() tasks.
 169  */
 170 struct pf_overload_entry {
 171         SLIST_ENTRY(pf_overload_entry)  next;
 172         struct pf_addr                  addr;
 173         sa_family_t                     af;
 174         uint8_t                         dir;
 175         struct pf_rule                  *rule;
 176 };
 177
 178 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 179 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 180 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 181 static VNET_DEFINE(struct task, pf_overloadtask);
 182 #define V_pf_overloadtask       VNET(pf_overloadtask)
 183
 184 static struct mtx pf_overloadqueue_mtx;
 185 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 186     "pf overload/flush queue", MTX_DEF);
 187 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 188 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 189
 190 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 191 struct mtx pf_unlnkdrules_mtx;
 192 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 193     MTX_DEF);
 194
 195 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 196 #define V_pf_sources_z  VNET(pf_sources_z)
 197 uma_zone_t              pf_mtag_z;
 198 VNET_DEFINE(uma_zone_t,  pf_state_z);
 199 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 200
 201 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 202 #define PFID_CPUBITS    8
 203 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 204 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 205 #define PFID_MAXID      (~PFID_CPUMASK)
 206 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 207
 208 static void              pf_src_tree_remove_state(struct pf_state *);
 209 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 210                             u_int32_t);
 211 static void              pf_add_threshold(struct pf_threshold *);
 212 static int               pf_check_threshold(struct pf_threshold *);
 213
 214 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 215                             u_int16_t *, u_int16_t *, struct pf_addr *,
 216                             u_int16_t, u_int8_t, sa_family_t);
 217 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 218                             struct tcphdr *, struct pf_state_peer *);
 219 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 220                             struct pf_addr *, struct pf_addr *, u_int16_t,
 221                             u_int16_t *, u_int16_t *, u_int16_t *,
 222                             u_int16_t *, u_int8_t, sa_family_t);
 223 static void              pf_send_tcp(struct mbuf *,
 224                             const struct pf_rule *, sa_family_t,
 225                             const struct pf_addr *, const struct pf_addr *,
 226                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 227                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 228                             u_int16_t, struct ifnet *);
 229 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 230                             sa_family_t, struct pf_rule *);
 231 static void              pf_detach_state(struct pf_state *);
 232 static int               pf_state_key_attach(struct pf_state_key *,
 233                             struct pf_state_key *, struct pf_state *);
 234 static void              pf_state_key_detach(struct pf_state *, int);
 235 static int               pf_state_key_ctor(void *, int, void *, int);
 236 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 237 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 238                             int, struct pfi_kif *, struct mbuf *, int,
 239                             struct pf_pdesc *, struct pf_rule **,
 240                             struct pf_ruleset **, struct inpcb *);
 241 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 242                             struct pf_rule *, struct pf_pdesc *,
 243                             struct pf_src_node *, struct pf_state_key *,
 244                             struct pf_state_key *, struct mbuf *, int,
 245                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 246                             struct pf_state **, int, u_int16_t, u_int16_t,
 247                             int);
 248 static int               pf_test_fragment(struct pf_rule **, int,
 249                             struct pfi_kif *, struct mbuf *, void *,
 250                             struct pf_pdesc *, struct pf_rule **,
 251                             struct pf_ruleset **);
 252 static int               pf_tcp_track_full(struct pf_state_peer *,
 253                             struct pf_state_peer *, struct pf_state **,
 254                             struct pfi_kif *, struct mbuf *, int,
 255                             struct pf_pdesc *, u_short *, int *);
 256 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 257                             struct pf_state_peer *, struct pf_state **,
 258                             struct pf_pdesc *, u_short *);
 259 static int               pf_test_state_tcp(struct pf_state **, int,
 260                             struct pfi_kif *, struct mbuf *, int,
 261                             void *, struct pf_pdesc *, u_short *);
 262 static int               pf_test_state_udp(struct pf_state **, int,
 263                             struct pfi_kif *, struct mbuf *, int,
 264                             void *, struct pf_pdesc *);
 265 static int               pf_test_state_icmp(struct pf_state **, int,
 266                             struct pfi_kif *, struct mbuf *, int,
 267                             void *, struct pf_pdesc *, u_short *);
 268 static int               pf_test_state_other(struct pf_state **, int,
 269                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 270 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 271                             sa_family_t);
 272 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 273                             sa_family_t);
 274 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 275                                 int, u_int16_t);
 276 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 277                             u_int8_t, sa_family_t);
 278 static void              pf_print_state_parts(struct pf_state *,
 279                             struct pf_state_key *, struct pf_state_key *);
 280 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 281                             struct pf_addr_wrap *);
 282 static struct pf_state  *pf_find_state(struct pfi_kif *,
 283                             struct pf_state_key_cmp *, u_int);
 284 static int               pf_src_connlimit(struct pf_state **);
 285 static void              pf_overload_task(void *v, int pending);
 286 static int               pf_insert_src_node(struct pf_src_node **,
 287                             struct pf_rule *, struct pf_addr *, sa_family_t);
 288 static u_int             pf_purge_expired_states(u_int, int);
 289 static void              pf_purge_unlinked_rules(void);
 290 static int               pf_mtag_uminit(void *, int, int);
 291 static void              pf_mtag_free(struct m_tag *);
 292 #ifdef INET
 293 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 294                             struct ifnet *, struct pf_state *,
 295                             struct pf_pdesc *);
 296 #endif /* INET */
 297 #ifdef INET6
 298 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 299                             struct pf_addr *, u_int8_t);
 300 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 301                             struct ifnet *, struct pf_state *,
 302                             struct pf_pdesc *);
 303 #endif /* INET6 */
 304
 305 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 306
 307 extern int pf_end_threads;
 308 extern struct proc *pf_purge_proc;
 309
 310 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 311
 312 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 313                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 314
 315 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 316         do {                                                            \
 317                 (s) = pf_find_state((i), (k), (d));                     \
 318                 if ((s) == NULL)                                        \
 319                         return (PF_DROP);                               \
 320                 if (PACKET_LOOPED(pd))                                  \
 321                         return (PF_PASS);                               \
 322                 if ((d) == PF_OUT &&                                    \
 323                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 324                     (s)->rule.ptr->direction == PF_OUT) ||              \
 325                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 326                     (s)->rule.ptr->direction == PF_IN)) &&              \
 327                     (s)->rt_kif != NULL &&                              \
 328                     (s)->rt_kif != (i))                                 \
 329                         return (PF_PASS);                               \
 330         } while (0)
 331
 332 #define BOUND_IFACE(r, k) \
 333         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 334
 335 #define STATE_INC_COUNTERS(s)                                           \
 336         do {                                                            \
 337                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 338                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 339                 if (s->anchor.ptr != NULL) {                            \
 340                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 341                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 342                 }                                                       \
 343                 if (s->nat_rule.ptr != NULL) {                          \
 344                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 345                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 346                 }                                                       \
 347         } while (0)
 348
 349 #define STATE_DEC_COUNTERS(s)                                           \
 350         do {                                                            \
 351                 if (s->nat_rule.ptr != NULL)                            \
 352                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 353                 if (s->anchor.ptr != NULL)                              \
 354                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 355                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 356         } while (0)
 357
 358 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 359 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 360 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 361 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 362
 363 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 364
 365 u_long  pf_hashmask;
 366 u_long  pf_srchashmask;
 367 static u_long   pf_hashsize;
 368 static u_long   pf_srchashsize;
 369
 370 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 371     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 372 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 373     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 374
 375 VNET_DEFINE(void *, pf_swi_cookie);
 376
 377 VNET_DEFINE(uint32_t, pf_hashseed);
 378 #define V_pf_hashseed   VNET(pf_hashseed)
 379
 380 int
 381 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 382 {
 383
 384         switch (af) {
 385 #ifdef INET
 386         case AF_INET:
 387                 if (a->addr32[0] > b->addr32[0])
 388                         return (1);
 389                 if (a->addr32[0] < b->addr32[0])
 390                         return (-1);
 391                 break;
 392 #endif /* INET */
 393 #ifdef INET6
 394         case AF_INET6:
 395                 if (a->addr32[3] > b->addr32[3])
 396                         return (1);
 397                 if (a->addr32[3] < b->addr32[3])
 398                         return (-1);
 399                 if (a->addr32[2] > b->addr32[2])
 400                         return (1);
 401                 if (a->addr32[2] < b->addr32[2])
 402                         return (-1);
 403                 if (a->addr32[1] > b->addr32[1])
 404                         return (1);
 405                 if (a->addr32[1] < b->addr32[1])
 406                         return (-1);
 407                 if (a->addr32[0] > b->addr32[0])
 408                         return (1);
 409                 if (a->addr32[0] < b->addr32[0])
 410                         return (-1);
 411                 break;
 412 #endif /* INET6 */
 413         default:
 414                 panic("%s: unknown address family %u", __func__, af);
 415         }
 416         return (0);
 417 }
 418
 419 static __inline uint32_t
 420 pf_hashkey(struct pf_state_key *sk)
 421 {
 422         uint32_t h;
 423
 424         h = murmur3_32_hash32((uint32_t *)sk,
 425             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 426             V_pf_hashseed);
 427
 428         return (h & pf_hashmask);
 429 }
 430
 431 static __inline uint32_t
 432 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 433 {
 434         uint32_t h;
 435
 436         switch (af) {
 437         case AF_INET:
 438                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 439                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 440                 break;
 441         case AF_INET6:
 442                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 443                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 444                 break;
 445         default:
 446                 panic("%s: unknown address family %u", __func__, af);
 447         }
 448
 449         return (h & pf_srchashmask);
 450 }
 451
 452 #ifdef ALTQ
 453 static int
 454 pf_state_hash(struct pf_state *s)
 455 {
 456         u_int32_t hv = (intptr_t)s / sizeof(*s);
 457
 458         hv ^= crc32(&s->src, sizeof(s->src));
 459         hv ^= crc32(&s->dst, sizeof(s->dst));
 460         if (hv == 0)
 461                 hv = 1;
 462         return (hv);
 463 }
 464 #endif
 465
 466 #ifdef INET6
 467 void
 468 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 469 {
 470         switch (af) {
 471 #ifdef INET
 472         case AF_INET:
 473                 dst->addr32[0] = src->addr32[0];
 474                 break;
 475 #endif /* INET */
 476         case AF_INET6:
 477                 dst->addr32[0] = src->addr32[0];
 478                 dst->addr32[1] = src->addr32[1];
 479                 dst->addr32[2] = src->addr32[2];
 480                 dst->addr32[3] = src->addr32[3];
 481                 break;
 482         }
 483 }
 484 #endif /* INET6 */
 485
 486 static void
 487 pf_init_threshold(struct pf_threshold *threshold,
 488     u_int32_t limit, u_int32_t seconds)
 489 {
 490         threshold->limit = limit * PF_THRESHOLD_MULT;
 491         threshold->seconds = seconds;
 492         threshold->count = 0;
 493         threshold->last = time_uptime;
 494 }
 495
 496 static void
 497 pf_add_threshold(struct pf_threshold *threshold)
 498 {
 499         u_int32_t t = time_uptime, diff = t - threshold->last;
 500
 501         if (diff >= threshold->seconds)
 502                 threshold->count = 0;
 503         else
 504                 threshold->count -= threshold->count * diff /
 505                     threshold->seconds;
 506         threshold->count += PF_THRESHOLD_MULT;
 507         threshold->last = t;
 508 }
 509
 510 static int
 511 pf_check_threshold(struct pf_threshold *threshold)
 512 {
 513         return (threshold->count > threshold->limit);
 514 }
 515
 516 static int
 517 pf_src_connlimit(struct pf_state **state)
 518 {
 519         struct pf_overload_entry *pfoe;
 520         int bad = 0;
 521
 522         PF_STATE_LOCK_ASSERT(*state);
 523
 524         (*state)->src_node->conn++;
 525         (*state)->src.tcp_est = 1;
 526         pf_add_threshold(&(*state)->src_node->conn_rate);
 527
 528         if ((*state)->rule.ptr->max_src_conn &&
 529             (*state)->rule.ptr->max_src_conn <
 530             (*state)->src_node->conn) {
 531                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 532                 bad++;
 533         }
 534
 535         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 536             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 537                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 538                 bad++;
 539         }
 540
 541         if (!bad)
 542                 return (0);
 543
 544         /* Kill this state. */
 545         (*state)->timeout = PFTM_PURGE;
 546         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 547
 548         if ((*state)->rule.ptr->overload_tbl == NULL)
 549                 return (1);
 550
 551         /* Schedule overloading and flushing task. */
 552         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 553         if (pfoe == NULL)
 554                 return (1);     /* too bad :( */
 555
 556         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 557         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 558         pfoe->rule = (*state)->rule.ptr;
 559         pfoe->dir = (*state)->direction;
 560         PF_OVERLOADQ_LOCK();
 561         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 562         PF_OVERLOADQ_UNLOCK();
 563         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 564
 565         return (1);
 566 }
 567
 568 static void
 569 pf_overload_task(void *v, int pending)
 570 {
 571         struct pf_overload_head queue;
 572         struct pfr_addr p;
 573         struct pf_overload_entry *pfoe, *pfoe1;
 574         uint32_t killed = 0;
 575
 576         CURVNET_SET((struct vnet *)v);
 577
 578         PF_OVERLOADQ_LOCK();
 579         queue = V_pf_overloadqueue;
 580         SLIST_INIT(&V_pf_overloadqueue);
 581         PF_OVERLOADQ_UNLOCK();
 582
 583         bzero(&p, sizeof(p));
 584         SLIST_FOREACH(pfoe, &queue, next) {
 585                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 586                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 587                         printf("%s: blocking address ", __func__);
 588                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 589                         printf("\n");
 590                 }
 591
 592                 p.pfra_af = pfoe->af;
 593                 switch (pfoe->af) {
 594 #ifdef INET
 595                 case AF_INET:
 596                         p.pfra_net = 32;
 597                         p.pfra_ip4addr = pfoe->addr.v4;
 598                         break;
 599 #endif
 600 #ifdef INET6
 601                 case AF_INET6:
 602                         p.pfra_net = 128;
 603                         p.pfra_ip6addr = pfoe->addr.v6;
 604                         break;
 605 #endif
 606                 }
 607
 608                 PF_RULES_WLOCK();
 609                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 610                 PF_RULES_WUNLOCK();
 611         }
 612
 613         /*
 614          * Remove those entries, that don't need flushing.
 615          */
 616         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 617                 if (pfoe->rule->flush == 0) {
 618                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 619                         free(pfoe, M_PFTEMP);
 620                 } else
 621                         counter_u64_add(
 622                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 623
 624         /* If nothing to flush, return. */
 625         if (SLIST_EMPTY(&queue)) {
 626                 CURVNET_RESTORE();
 627                 return;
 628         }
 629
 630         for (int i = 0; i <= pf_hashmask; i++) {
 631                 struct pf_idhash *ih = &V_pf_idhash[i];
 632                 struct pf_state_key *sk;
 633                 struct pf_state *s;
 634
 635                 PF_HASHROW_LOCK(ih);
 636                 LIST_FOREACH(s, &ih->states, entry) {
 637                     sk = s->key[PF_SK_WIRE];
 638                     SLIST_FOREACH(pfoe, &queue, next)
 639                         if (sk->af == pfoe->af &&
 640                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 641                             pfoe->rule == s->rule.ptr) &&
 642                             ((pfoe->dir == PF_OUT &&
 643                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 644                             (pfoe->dir == PF_IN &&
 645                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 646                                 s->timeout = PFTM_PURGE;
 647                                 s->src.state = s->dst.state = TCPS_CLOSED;
 648                                 killed++;
 649                         }
 650                 }
 651                 PF_HASHROW_UNLOCK(ih);
 652         }
 653         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 654                 free(pfoe, M_PFTEMP);
 655         if (V_pf_status.debug >= PF_DEBUG_MISC)
 656                 printf("%s: %u states killed", __func__, killed);
 657
 658         CURVNET_RESTORE();
 659 }
 660
 661 /*
 662  * Can return locked on failure, so that we can consistently
 663  * allocate and insert a new one.
 664  */
 665 struct pf_src_node *
 666 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 667         int returnlocked)
 668 {
 669         struct pf_srchash *sh;
 670         struct pf_src_node *n;
 671
 672         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 673
 674         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 675         PF_HASHROW_LOCK(sh);
 676         LIST_FOREACH(n, &sh->nodes, entry)
 677                 if (n->rule.ptr == rule && n->af == af &&
 678                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 679                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 680                         break;
 681         if (n != NULL) {
 682                 n->states++;
 683                 PF_HASHROW_UNLOCK(sh);
 684         } else if (returnlocked == 0)
 685                 PF_HASHROW_UNLOCK(sh);
 686
 687         return (n);
 688 }
 689
 690 static int
 691 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 692     struct pf_addr *src, sa_family_t af)
 693 {
 694
 695         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 696             rule->rpool.opts & PF_POOL_STICKYADDR),
 697             ("%s for non-tracking rule %p", __func__, rule));
 698
 699         if (*sn == NULL)
 700                 *sn = pf_find_src_node(src, rule, af, 1);
 701
 702         if (*sn == NULL) {
 703                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 704
 705                 PF_HASHROW_ASSERT(sh);
 706
 707                 if (!rule->max_src_nodes ||
 708                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 709                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 710                 else
 711                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 712                             1);
 713                 if ((*sn) == NULL) {
 714                         PF_HASHROW_UNLOCK(sh);
 715                         return (-1);
 716                 }
 717
 718                 pf_init_threshold(&(*sn)->conn_rate,
 719                     rule->max_src_conn_rate.limit,
 720                     rule->max_src_conn_rate.seconds);
 721
 722                 (*sn)->af = af;
 723                 (*sn)->rule.ptr = rule;
 724                 PF_ACPY(&(*sn)->addr, src, af);
 725                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 726                 (*sn)->creation = time_uptime;
 727                 (*sn)->ruletype = rule->action;
 728                 (*sn)->states = 1;
 729                 if ((*sn)->rule.ptr != NULL)
 730                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 731                 PF_HASHROW_UNLOCK(sh);
 732                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 733         } else {
 734                 if (rule->max_src_states &&
 735                     (*sn)->states >= rule->max_src_states) {
 736                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 737                             1);
 738                         return (-1);
 739                 }
 740         }
 741         return (0);
 742 }
 743
 744 void
 745 pf_unlink_src_node(struct pf_src_node *src)
 746 {
 747
 748         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 749         LIST_REMOVE(src, entry);
 750         if (src->rule.ptr)
 751                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 752 }
 753
 754 u_int
 755 pf_free_src_nodes(struct pf_src_node_list *head)
 756 {
 757         struct pf_src_node *sn, *tmp;
 758         u_int count = 0;
 759
 760         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 761                 uma_zfree(V_pf_sources_z, sn);
 762                 count++;
 763         }
 764
 765         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 766
 767         return (count);
 768 }
 769
 770 void
 771 pf_mtag_initialize()
 772 {
 773
 774         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 775             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 776             UMA_ALIGN_PTR, 0);
 777 }
 778
 779 /* Per-vnet data storage structures initialization. */
 780 void
 781 pf_initialize()
 782 {
 783         struct pf_keyhash       *kh;
 784         struct pf_idhash        *ih;
 785         struct pf_srchash       *sh;
 786         u_int i;
 787
 788         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 789                 pf_hashsize = PF_HASHSIZ;
 790         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 791                 pf_srchashsize = PF_HASHSIZ / 4;
 792
 793         V_pf_hashseed = arc4random();
 794
 795         /* States and state keys storage. */
 796         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 797             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 798         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 799         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 800         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 801
 802         V_pf_state_key_z = uma_zcreate("pf state keys",
 803             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 804             UMA_ALIGN_PTR, 0);
 805         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 806             M_PFHASH, M_WAITOK | M_ZERO);
 807         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 808             M_PFHASH, M_WAITOK | M_ZERO);
 809         pf_hashmask = pf_hashsize - 1;
 810         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 811             i++, kh++, ih++) {
 812                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 813                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 814         }
 815
 816         /* Source nodes. */
 817         V_pf_sources_z = uma_zcreate("pf source nodes",
 818             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 819             0);
 820         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 821         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 822         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 823         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 824           M_PFHASH, M_WAITOK|M_ZERO);
 825         pf_srchashmask = pf_srchashsize - 1;
 826         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 827                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 828
 829         /* ALTQ */
 830         TAILQ_INIT(&V_pf_altqs[0]);
 831         TAILQ_INIT(&V_pf_altqs[1]);
 832         TAILQ_INIT(&V_pf_pabuf);
 833         V_pf_altqs_active = &V_pf_altqs[0];
 834         V_pf_altqs_inactive = &V_pf_altqs[1];
 835
 836         /* Send & overload+flush queues. */
 837         STAILQ_INIT(&V_pf_sendqueue);
 838         SLIST_INIT(&V_pf_overloadqueue);
 839         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 840
 841         /* Unlinked, but may be referenced rules. */
 842         TAILQ_INIT(&V_pf_unlinked_rules);
 843 }
 844
 845 void
 846 pf_mtag_cleanup()
 847 {
 848
 849         uma_zdestroy(pf_mtag_z);
 850 }
 851
 852 void
 853 pf_cleanup()
 854 {
 855         struct pf_keyhash       *kh;
 856         struct pf_idhash        *ih;
 857         struct pf_srchash       *sh;
 858         struct pf_send_entry    *pfse, *next;
 859         u_int i;
 860
 861         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 862             i++, kh++, ih++) {
 863                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 864                     __func__));
 865                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 866                     __func__));
 867                 mtx_destroy(&kh->lock);
 868                 mtx_destroy(&ih->lock);
 869         }
 870         free(V_pf_keyhash, M_PFHASH);
 871         free(V_pf_idhash, M_PFHASH);
 872
 873         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 874                 KASSERT(LIST_EMPTY(&sh->nodes),
 875                     ("%s: source node hash not empty", __func__));
 876                 mtx_destroy(&sh->lock);
 877         }
 878         free(V_pf_srchash, M_PFHASH);
 879
 880         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 881                 m_freem(pfse->pfse_m);
 882                 free(pfse, M_PFTEMP);
 883         }
 884
 885         uma_zdestroy(V_pf_sources_z);
 886         uma_zdestroy(V_pf_state_z);
 887         uma_zdestroy(V_pf_state_key_z);
 888 }
 889
 890 static int
 891 pf_mtag_uminit(void *mem, int size, int how)
 892 {
 893         struct m_tag *t;
 894
 895         t = (struct m_tag *)mem;
 896         t->m_tag_cookie = MTAG_ABI_COMPAT;
 897         t->m_tag_id = PACKET_TAG_PF;
 898         t->m_tag_len = sizeof(struct pf_mtag);
 899         t->m_tag_free = pf_mtag_free;
 900
 901         return (0);
 902 }
 903
 904 static void
 905 pf_mtag_free(struct m_tag *t)
 906 {
 907
 908         uma_zfree(pf_mtag_z, t);
 909 }
 910
 911 struct pf_mtag *
 912 pf_get_mtag(struct mbuf *m)
 913 {
 914         struct m_tag *mtag;
 915
 916         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 917                 return ((struct pf_mtag *)(mtag + 1));
 918
 919         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 920         if (mtag == NULL)
 921                 return (NULL);
 922         bzero(mtag + 1, sizeof(struct pf_mtag));
 923         m_tag_prepend(m, mtag);
 924
 925         return ((struct pf_mtag *)(mtag + 1));
 926 }
 927
 928 static int
 929 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 930     struct pf_state *s)
 931 {
 932         struct pf_keyhash       *khs, *khw, *kh;
 933         struct pf_state_key     *sk, *cur;
 934         struct pf_state         *si, *olds = NULL;
 935         int idx;
 936
 937         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 938         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 939         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 940
 941         /*
 942          * We need to lock hash slots of both keys. To avoid deadlock
 943          * we always lock the slot with lower address first. Unlock order
 944          * isn't important.
 945          *
 946          * We also need to lock ID hash slot before dropping key
 947          * locks. On success we return with ID hash slot locked.
 948          */
 949
 950         if (skw == sks) {
 951                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 952                 PF_HASHROW_LOCK(khs);
 953         } else {
 954                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 955                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 956                 if (khs == khw) {
 957                         PF_HASHROW_LOCK(khs);
 958                 } else if (khs < khw) {
 959                         PF_HASHROW_LOCK(khs);
 960                         PF_HASHROW_LOCK(khw);
 961                 } else {
 962                         PF_HASHROW_LOCK(khw);
 963                         PF_HASHROW_LOCK(khs);
 964                 }
 965         }
 966
 967 #define KEYS_UNLOCK()   do {                    \
 968         if (khs != khw) {                       \
 969                 PF_HASHROW_UNLOCK(khs);         \
 970                 PF_HASHROW_UNLOCK(khw);         \
 971         } else                                  \
 972                 PF_HASHROW_UNLOCK(khs);         \
 973 } while (0)
 974
 975         /*
 976          * First run: start with wire key.
 977          */
 978         sk = skw;
 979         kh = khw;
 980         idx = PF_SK_WIRE;
 981
 982 keyattach:
 983         LIST_FOREACH(cur, &kh->keys, entry)
 984                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 985                         break;
 986
 987         if (cur != NULL) {
 988                 /* Key exists. Check for same kif, if none, add to key. */
 989                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 990                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 991
 992                         PF_HASHROW_LOCK(ih);
 993                         if (si->kif == s->kif &&
 994                             si->direction == s->direction) {
 995                                 if (sk->proto == IPPROTO_TCP &&
 996                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 997                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 998                                         /*
 999                                          * New state matches an old >FIN_WAIT_2
1000                                          * state. We can't drop key hash locks,
1001                                          * thus we can't unlink it properly.
1002                                          *
1003                                          * As a workaround we drop it into
1004                                          * TCPS_CLOSED state, schedule purge
1005                                          * ASAP and push it into the very end
1006                                          * of the slot TAILQ, so that it won't
1007                                          * conflict with our new state.
1008                                          */
1009                                         si->src.state = si->dst.state =
1010                                             TCPS_CLOSED;
1011                                         si->timeout = PFTM_PURGE;
1012                                         olds = si;
1013                                 } else {
1014                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1015                                                 printf("pf: %s key attach "
1016                                                     "failed on %s: ",
1017                                                     (idx == PF_SK_WIRE) ?
1018                                                     "wire" : "stack",
1019                                                     s->kif->pfik_name);
1020                                                 pf_print_state_parts(s,
1021                                                     (idx == PF_SK_WIRE) ?
1022                                                     sk : NULL,
1023                                                     (idx == PF_SK_STACK) ?
1024                                                     sk : NULL);
1025                                                 printf(", existing: ");
1026                                                 pf_print_state_parts(si,
1027                                                     (idx == PF_SK_WIRE) ?
1028                                                     sk : NULL,
1029                                                     (idx == PF_SK_STACK) ?
1030                                                     sk : NULL);
1031                                                 printf("\n");
1032                                         }
1033                                         PF_HASHROW_UNLOCK(ih);
1034                                         KEYS_UNLOCK();
1035                                         uma_zfree(V_pf_state_key_z, sk);
1036                                         if (idx == PF_SK_STACK)
1037                                                 pf_detach_state(s);
1038                                         return (EEXIST); /* collision! */
1039                                 }
1040                         }
1041                         PF_HASHROW_UNLOCK(ih);
1042                 }
1043                 uma_zfree(V_pf_state_key_z, sk);
1044                 s->key[idx] = cur;
1045         } else {
1046                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1047                 s->key[idx] = sk;
1048         }
1049
1050 stateattach:
1051         /* List is sorted, if-bound states before floating. */
1052         if (s->kif == V_pfi_all)
1053                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1054         else
1055                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1056
1057         if (olds) {
1058                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1059                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1060                     key_list[idx]);
1061                 olds = NULL;
1062         }
1063
1064         /*
1065          * Attach done. See how should we (or should not?)
1066          * attach a second key.
1067          */
1068         if (sks == skw) {
1069                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1070                 idx = PF_SK_STACK;
1071                 sks = NULL;
1072                 goto stateattach;
1073         } else if (sks != NULL) {
1074                 /*
1075                  * Continue attaching with stack key.
1076                  */
1077                 sk = sks;
1078                 kh = khs;
1079                 idx = PF_SK_STACK;
1080                 sks = NULL;
1081                 goto keyattach;
1082         }
1083
1084         PF_STATE_LOCK(s);
1085         KEYS_UNLOCK();
1086
1087         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1088             ("%s failure", __func__));
1089
1090         return (0);
1091 #undef  KEYS_UNLOCK
1092 }
1093
1094 static void
1095 pf_detach_state(struct pf_state *s)
1096 {
1097         struct pf_state_key *sks = s->key[PF_SK_STACK];
1098         struct pf_keyhash *kh;
1099
1100         if (sks != NULL) {
1101                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1102                 PF_HASHROW_LOCK(kh);
1103                 if (s->key[PF_SK_STACK] != NULL)
1104                         pf_state_key_detach(s, PF_SK_STACK);
1105                 /*
1106                  * If both point to same key, then we are done.
1107                  */
1108                 if (sks == s->key[PF_SK_WIRE]) {
1109                         pf_state_key_detach(s, PF_SK_WIRE);
1110                         PF_HASHROW_UNLOCK(kh);
1111                         return;
1112                 }
1113                 PF_HASHROW_UNLOCK(kh);
1114         }
1115
1116         if (s->key[PF_SK_WIRE] != NULL) {
1117                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1118                 PF_HASHROW_LOCK(kh);
1119                 if (s->key[PF_SK_WIRE] != NULL)
1120                         pf_state_key_detach(s, PF_SK_WIRE);
1121                 PF_HASHROW_UNLOCK(kh);
1122         }
1123 }
1124
1125 static void
1126 pf_state_key_detach(struct pf_state *s, int idx)
1127 {
1128         struct pf_state_key *sk = s->key[idx];
1129 #ifdef INVARIANTS
1130         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1131
1132         PF_HASHROW_ASSERT(kh);
1133 #endif
1134         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1135         s->key[idx] = NULL;
1136
1137         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1138                 LIST_REMOVE(sk, entry);
1139                 uma_zfree(V_pf_state_key_z, sk);
1140         }
1141 }
1142
1143 static int
1144 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1145 {
1146         struct pf_state_key *sk = mem;
1147
1148         bzero(sk, sizeof(struct pf_state_key_cmp));
1149         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1150         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1151
1152         return (0);
1153 }
1154
1155 struct pf_state_key *
1156 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1157         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1158 {
1159         struct pf_state_key *sk;
1160
1161         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1162         if (sk == NULL)
1163                 return (NULL);
1164
1165         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1166         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1167         sk->port[pd->sidx] = sport;
1168         sk->port[pd->didx] = dport;
1169         sk->proto = pd->proto;
1170         sk->af = pd->af;
1171
1172         return (sk);
1173 }
1174
1175 struct pf_state_key *
1176 pf_state_key_clone(struct pf_state_key *orig)
1177 {
1178         struct pf_state_key *sk;
1179
1180         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1181         if (sk == NULL)
1182                 return (NULL);
1183
1184         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1185
1186         return (sk);
1187 }
1188
1189 int
1190 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1191     struct pf_state_key *sks, struct pf_state *s)
1192 {
1193         struct pf_idhash *ih;
1194         struct pf_state *cur;
1195         int error;
1196
1197         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1198             ("%s: sks not pristine", __func__));
1199         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1200             ("%s: skw not pristine", __func__));
1201         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1202
1203         s->kif = kif;
1204
1205         if (s->id == 0 && s->creatorid == 0) {
1206                 /* XXX: should be atomic, but probability of collision low */
1207                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1208                         V_pf_stateid[curcpu] = 1;
1209                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1210                 s->id = htobe64(s->id);
1211                 s->creatorid = V_pf_status.hostid;
1212         }
1213
1214         /* Returns with ID locked on success. */
1215         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1216                 return (error);
1217
1218         ih = &V_pf_idhash[PF_IDHASH(s)];
1219         PF_HASHROW_ASSERT(ih);
1220         LIST_FOREACH(cur, &ih->states, entry)
1221                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1222                         break;
1223
1224         if (cur != NULL) {
1225                 PF_HASHROW_UNLOCK(ih);
1226                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1227                         printf("pf: state ID collision: "
1228                             "id: %016llx creatorid: %08x\n",
1229                             (unsigned long long)be64toh(s->id),
1230                             ntohl(s->creatorid));
1231                 }
1232                 pf_detach_state(s);
1233                 return (EEXIST);
1234         }
1235         LIST_INSERT_HEAD(&ih->states, s, entry);
1236         /* One for keys, one for ID hash. */
1237         refcount_init(&s->refs, 2);
1238
1239         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1240         if (pfsync_insert_state_ptr != NULL)
1241                 pfsync_insert_state_ptr(s);
1242
1243         /* Returns locked. */
1244         return (0);
1245 }
1246
1247 /*
1248  * Find state by ID: returns with locked row on success.
1249  */
1250 struct pf_state *
1251 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1252 {
1253         struct pf_idhash *ih;
1254         struct pf_state *s;
1255
1256         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1257
1258         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1259
1260         PF_HASHROW_LOCK(ih);
1261         LIST_FOREACH(s, &ih->states, entry)
1262                 if (s->id == id && s->creatorid == creatorid)
1263                         break;
1264
1265         if (s == NULL)
1266                 PF_HASHROW_UNLOCK(ih);
1267
1268         return (s);
1269 }
1270
1271 /*
1272  * Find state by key.
1273  * Returns with ID hash slot locked on success.
1274  */
1275 static struct pf_state *
1276 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1277 {
1278         struct pf_keyhash       *kh;
1279         struct pf_state_key     *sk;
1280         struct pf_state         *s;
1281         int idx;
1282
1283         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1284
1285         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1286
1287         PF_HASHROW_LOCK(kh);
1288         LIST_FOREACH(sk, &kh->keys, entry)
1289                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1290                         break;
1291         if (sk == NULL) {
1292                 PF_HASHROW_UNLOCK(kh);
1293                 return (NULL);
1294         }
1295
1296         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1297
1298         /* List is sorted, if-bound states before floating ones. */
1299         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1300                 if (s->kif == V_pfi_all || s->kif == kif) {
1301                         PF_STATE_LOCK(s);
1302                         PF_HASHROW_UNLOCK(kh);
1303                         if (s->timeout >= PFTM_MAX) {
1304                                 /*
1305                                  * State is either being processed by
1306                                  * pf_unlink_state() in an other thread, or
1307                                  * is scheduled for immediate expiry.
1308                                  */
1309                                 PF_STATE_UNLOCK(s);
1310                                 return (NULL);
1311                         }
1312                         return (s);
1313                 }
1314         PF_HASHROW_UNLOCK(kh);
1315
1316         return (NULL);
1317 }
1318
1319 struct pf_state *
1320 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1321 {
1322         struct pf_keyhash       *kh;
1323         struct pf_state_key     *sk;
1324         struct pf_state         *s, *ret = NULL;
1325         int                      idx, inout = 0;
1326
1327         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1328
1329         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1330
1331         PF_HASHROW_LOCK(kh);
1332         LIST_FOREACH(sk, &kh->keys, entry)
1333                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1334                         break;
1335         if (sk == NULL) {
1336                 PF_HASHROW_UNLOCK(kh);
1337                 return (NULL);
1338         }
1339         switch (dir) {
1340         case PF_IN:
1341                 idx = PF_SK_WIRE;
1342                 break;
1343         case PF_OUT:
1344                 idx = PF_SK_STACK;
1345                 break;
1346         case PF_INOUT:
1347                 idx = PF_SK_WIRE;
1348                 inout = 1;
1349                 break;
1350         default:
1351                 panic("%s: dir %u", __func__, dir);
1352         }
1353 second_run:
1354         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1355                 if (more == NULL) {
1356                         PF_HASHROW_UNLOCK(kh);
1357                         return (s);
1358                 }
1359
1360                 if (ret)
1361                         (*more)++;
1362                 else
1363                         ret = s;
1364         }
1365         if (inout == 1) {
1366                 inout = 0;
1367                 idx = PF_SK_STACK;
1368                 goto second_run;
1369         }
1370         PF_HASHROW_UNLOCK(kh);
1371
1372         return (ret);
1373 }
1374
1375 /* END state table stuff */
1376
1377 static void
1378 pf_send(struct pf_send_entry *pfse)
1379 {
1380
1381         PF_SENDQ_LOCK();
1382         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1383         PF_SENDQ_UNLOCK();
1384         swi_sched(V_pf_swi_cookie, 0);
1385 }
1386
1387 void
1388 pf_intr(void *v)
1389 {
1390         struct pf_send_head queue;
1391         struct pf_send_entry *pfse, *next;
1392
1393         CURVNET_SET((struct vnet *)v);
1394
1395         PF_SENDQ_LOCK();
1396         queue = V_pf_sendqueue;
1397         STAILQ_INIT(&V_pf_sendqueue);
1398         PF_SENDQ_UNLOCK();
1399
1400         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1401                 switch (pfse->pfse_type) {
1402 #ifdef INET
1403                 case PFSE_IP:
1404                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1405                         break;
1406                 case PFSE_ICMP:
1407                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1408                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1409                         break;
1410 #endif /* INET */
1411 #ifdef INET6
1412                 case PFSE_IP6:
1413                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1414                             NULL);
1415                         break;
1416                 case PFSE_ICMP6:
1417                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1418                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1419                         break;
1420 #endif /* INET6 */
1421                 default:
1422                         panic("%s: unknown type", __func__);
1423                 }
1424                 free(pfse, M_PFTEMP);
1425         }
1426         CURVNET_RESTORE();
1427 }
1428
1429 void
1430 pf_purge_thread(void *unused __unused)
1431 {
1432         VNET_ITERATOR_DECL(vnet_iter);
1433
1434         sx_xlock(&pf_end_lock);
1435         while (pf_end_threads == 0) {
1436                 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1437
1438                 VNET_LIST_RLOCK();
1439                 VNET_FOREACH(vnet_iter) {
1440                         CURVNET_SET(vnet_iter);
1441
1442
1443                         /* Wait until V_pf_default_rule is initialized. */
1444                         if (V_pf_vnet_active == 0) {
1445                                 CURVNET_RESTORE();
1446                                 continue;
1447                         }
1448
1449                         /*
1450                          *  Process 1/interval fraction of the state
1451                          * table every run.
1452                          */
1453                         V_pf_purge_idx =
1454                             pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1455                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1456
1457                         /*
1458                          * Purge other expired types every
1459                          * PFTM_INTERVAL seconds.
1460                          */
1461                         if (V_pf_purge_idx == 0) {
1462                                 /*
1463                                  * Order is important:
1464                                  * - states and src nodes reference rules
1465                                  * - states and rules reference kifs
1466                                  */
1467                                 pf_purge_expired_fragments();
1468                                 pf_purge_expired_src_nodes();
1469                                 pf_purge_unlinked_rules();
1470                                 pfi_kif_purge();
1471                         }
1472                         CURVNET_RESTORE();
1473                 }
1474                 VNET_LIST_RUNLOCK();
1475         }
1476
1477         pf_end_threads++;
1478         sx_xunlock(&pf_end_lock);
1479         kproc_exit(0);
1480 }
1481
1482 void
1483 pf_unload_vnet_purge(void)
1484 {
1485
1486         /*
1487          * To cleanse up all kifs and rules we need
1488          * two runs: first one clears reference flags,
1489          * then pf_purge_expired_states() doesn't
1490          * raise them, and then second run frees.
1491          */
1492         pf_purge_unlinked_rules();
1493         pfi_kif_purge();
1494
1495         /*
1496          * Now purge everything.
1497          */
1498         pf_purge_expired_states(0, pf_hashmask);
1499         pf_purge_expired_fragments();
1500         pf_purge_expired_src_nodes();
1501
1502         /*
1503          * Now all kifs & rules should be unreferenced,
1504          * thus should be successfully freed.
1505          */
1506         pf_purge_unlinked_rules();
1507         pfi_kif_purge();
1508 }
1509
1510
1511 u_int32_t
1512 pf_state_expires(const struct pf_state *state)
1513 {
1514         u_int32_t       timeout;
1515         u_int32_t       start;
1516         u_int32_t       end;
1517         u_int32_t       states;
1518
1519         /* handle all PFTM_* > PFTM_MAX here */
1520         if (state->timeout == PFTM_PURGE)
1521                 return (time_uptime);
1522         KASSERT(state->timeout != PFTM_UNLINKED,
1523             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1524         KASSERT((state->timeout < PFTM_MAX),
1525             ("pf_state_expires: timeout > PFTM_MAX"));
1526         timeout = state->rule.ptr->timeout[state->timeout];
1527         if (!timeout)
1528                 timeout = V_pf_default_rule.timeout[state->timeout];
1529         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1530         if (start) {
1531                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1532                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1533         } else {
1534                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1535                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1536                 states = V_pf_status.states;
1537         }
1538         if (end && states > start && start < end) {
1539                 if (states < end)
1540                         return (state->expire + timeout * (end - states) /
1541                             (end - start));
1542                 else
1543                         return (time_uptime);
1544         }
1545         return (state->expire + timeout);
1546 }
1547
1548 void
1549 pf_purge_expired_src_nodes()
1550 {
1551         struct pf_src_node_list  freelist;
1552         struct pf_srchash       *sh;
1553         struct pf_src_node      *cur, *next;
1554         int i;
1555
1556         LIST_INIT(&freelist);
1557         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1558             PF_HASHROW_LOCK(sh);
1559             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1560                 if (cur->states == 0 && cur->expire <= time_uptime) {
1561                         pf_unlink_src_node(cur);
1562                         LIST_INSERT_HEAD(&freelist, cur, entry);
1563                 } else if (cur->rule.ptr != NULL)
1564                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1565             PF_HASHROW_UNLOCK(sh);
1566         }
1567
1568         pf_free_src_nodes(&freelist);
1569
1570         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1571 }
1572
1573 static void
1574 pf_src_tree_remove_state(struct pf_state *s)
1575 {
1576         struct pf_src_node *sn;
1577         struct pf_srchash *sh;
1578         uint32_t timeout;
1579
1580         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1581             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1582             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1583
1584         if (s->src_node != NULL) {
1585                 sn = s->src_node;
1586                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1587                 PF_HASHROW_LOCK(sh);
1588                 if (s->src.tcp_est)
1589                         --sn->conn;
1590                 if (--sn->states == 0)
1591                         sn->expire = time_uptime + timeout;
1592                 PF_HASHROW_UNLOCK(sh);
1593         }
1594         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1595                 sn = s->nat_src_node;
1596                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1597                 PF_HASHROW_LOCK(sh);
1598                 if (--sn->states == 0)
1599                         sn->expire = time_uptime + timeout;
1600                 PF_HASHROW_UNLOCK(sh);
1601         }
1602         s->src_node = s->nat_src_node = NULL;
1603 }
1604
1605 /*
1606  * Unlink and potentilly free a state. Function may be
1607  * called with ID hash row locked, but always returns
1608  * unlocked, since it needs to go through key hash locking.
1609  */
1610 int
1611 pf_unlink_state(struct pf_state *s, u_int flags)
1612 {
1613         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1614
1615         if ((flags & PF_ENTER_LOCKED) == 0)
1616                 PF_HASHROW_LOCK(ih);
1617         else
1618                 PF_HASHROW_ASSERT(ih);
1619
1620         if (s->timeout == PFTM_UNLINKED) {
1621                 /*
1622                  * State is being processed
1623                  * by pf_unlink_state() in
1624                  * an other thread.
1625                  */
1626                 PF_HASHROW_UNLOCK(ih);
1627                 return (0);     /* XXXGL: undefined actually */
1628         }
1629
1630         if (s->src.state == PF_TCPS_PROXY_DST) {
1631                 /* XXX wire key the right one? */
1632                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1633                     &s->key[PF_SK_WIRE]->addr[1],
1634                     &s->key[PF_SK_WIRE]->addr[0],
1635                     s->key[PF_SK_WIRE]->port[1],
1636                     s->key[PF_SK_WIRE]->port[0],
1637                     s->src.seqhi, s->src.seqlo + 1,
1638                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1639         }
1640
1641         LIST_REMOVE(s, entry);
1642         pf_src_tree_remove_state(s);
1643
1644         if (pfsync_delete_state_ptr != NULL)
1645                 pfsync_delete_state_ptr(s);
1646
1647         STATE_DEC_COUNTERS(s);
1648
1649         s->timeout = PFTM_UNLINKED;
1650
1651         PF_HASHROW_UNLOCK(ih);
1652
1653         pf_detach_state(s);
1654         refcount_release(&s->refs);
1655
1656         return (pf_release_state(s));
1657 }
1658
1659 void
1660 pf_free_state(struct pf_state *cur)
1661 {
1662
1663         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1664         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1665             cur->timeout));
1666
1667         pf_normalize_tcp_cleanup(cur);
1668         uma_zfree(V_pf_state_z, cur);
1669         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1670 }
1671
1672 /*
1673  * Called only from pf_purge_thread(), thus serialized.
1674  */
1675 static u_int
1676 pf_purge_expired_states(u_int i, int maxcheck)
1677 {
1678         struct pf_idhash *ih;
1679         struct pf_state *s;
1680
1681         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1682
1683         /*
1684          * Go through hash and unlink states that expire now.
1685          */
1686         while (maxcheck > 0) {
1687
1688                 ih = &V_pf_idhash[i];
1689 relock:
1690                 PF_HASHROW_LOCK(ih);
1691                 LIST_FOREACH(s, &ih->states, entry) {
1692                         if (pf_state_expires(s) <= time_uptime) {
1693                                 V_pf_status.states -=
1694                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1695                                 goto relock;
1696                         }
1697                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1698                         if (s->nat_rule.ptr != NULL)
1699                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1700                         if (s->anchor.ptr != NULL)
1701                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1702                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1703                         if (s->rt_kif)
1704                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1705                 }
1706                 PF_HASHROW_UNLOCK(ih);
1707
1708                 /* Return when we hit end of hash. */
1709                 if (++i > pf_hashmask) {
1710                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1711                         return (0);
1712                 }
1713
1714                 maxcheck--;
1715         }
1716
1717         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1718
1719         return (i);
1720 }
1721
1722 static void
1723 pf_purge_unlinked_rules()
1724 {
1725         struct pf_rulequeue tmpq;
1726         struct pf_rule *r, *r1;
1727
1728         /*
1729          * If we have overloading task pending, then we'd
1730          * better skip purging this time. There is a tiny
1731          * probability that overloading task references
1732          * an already unlinked rule.
1733          */
1734         PF_OVERLOADQ_LOCK();
1735         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1736                 PF_OVERLOADQ_UNLOCK();
1737                 return;
1738         }
1739         PF_OVERLOADQ_UNLOCK();
1740
1741         /*
1742          * Do naive mark-and-sweep garbage collecting of old rules.
1743          * Reference flag is raised by pf_purge_expired_states()
1744          * and pf_purge_expired_src_nodes().
1745          *
1746          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1747          * use a temporary queue.
1748          */
1749         TAILQ_INIT(&tmpq);
1750         PF_UNLNKDRULES_LOCK();
1751         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1752                 if (!(r->rule_flag & PFRULE_REFS)) {
1753                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1754                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1755                 } else
1756                         r->rule_flag &= ~PFRULE_REFS;
1757         }
1758         PF_UNLNKDRULES_UNLOCK();
1759
1760         if (!TAILQ_EMPTY(&tmpq)) {
1761                 PF_RULES_WLOCK();
1762                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1763                         TAILQ_REMOVE(&tmpq, r, entries);
1764                         pf_free_rule(r);
1765                 }
1766                 PF_RULES_WUNLOCK();
1767         }
1768 }
1769
1770 void
1771 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1772 {
1773         switch (af) {
1774 #ifdef INET
1775         case AF_INET: {
1776                 u_int32_t a = ntohl(addr->addr32[0]);
1777                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1778                     (a>>8)&255, a&255);
1779                 if (p) {
1780                         p = ntohs(p);
1781                         printf(":%u", p);
1782                 }
1783                 break;
1784         }
1785 #endif /* INET */
1786 #ifdef INET6
1787         case AF_INET6: {
1788                 u_int16_t b;
1789                 u_int8_t i, curstart, curend, maxstart, maxend;
1790                 curstart = curend = maxstart = maxend = 255;
1791                 for (i = 0; i < 8; i++) {
1792                         if (!addr->addr16[i]) {
1793                                 if (curstart == 255)
1794                                         curstart = i;
1795                                 curend = i;
1796                         } else {
1797                                 if ((curend - curstart) >
1798                                     (maxend - maxstart)) {
1799                                         maxstart = curstart;
1800                                         maxend = curend;
1801                                 }
1802                                 curstart = curend = 255;
1803                         }
1804                 }
1805                 if ((curend - curstart) >
1806                     (maxend - maxstart)) {
1807                         maxstart = curstart;
1808                         maxend = curend;
1809                 }
1810                 for (i = 0; i < 8; i++) {
1811                         if (i >= maxstart && i <= maxend) {
1812                                 if (i == 0)
1813                                         printf(":");
1814                                 if (i == maxend)
1815                                         printf(":");
1816                         } else {
1817                                 b = ntohs(addr->addr16[i]);
1818                                 printf("%x", b);
1819                                 if (i < 7)
1820                                         printf(":");
1821                         }
1822                 }
1823                 if (p) {
1824                         p = ntohs(p);
1825                         printf("[%u]", p);
1826                 }
1827                 break;
1828         }
1829 #endif /* INET6 */
1830         }
1831 }
1832
1833 void
1834 pf_print_state(struct pf_state *s)
1835 {
1836         pf_print_state_parts(s, NULL, NULL);
1837 }
1838
1839 static void
1840 pf_print_state_parts(struct pf_state *s,
1841     struct pf_state_key *skwp, struct pf_state_key *sksp)
1842 {
1843         struct pf_state_key *skw, *sks;
1844         u_int8_t proto, dir;
1845
1846         /* Do our best to fill these, but they're skipped if NULL */
1847         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1848         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1849         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1850         dir = s ? s->direction : 0;
1851
1852         switch (proto) {
1853         case IPPROTO_IPV4:
1854                 printf("IPv4");
1855                 break;
1856         case IPPROTO_IPV6:
1857                 printf("IPv6");
1858                 break;
1859         case IPPROTO_TCP:
1860                 printf("TCP");
1861                 break;
1862         case IPPROTO_UDP:
1863                 printf("UDP");
1864                 break;
1865         case IPPROTO_ICMP:
1866                 printf("ICMP");
1867                 break;
1868         case IPPROTO_ICMPV6:
1869                 printf("ICMPv6");
1870                 break;
1871         default:
1872                 printf("%u", proto);
1873                 break;
1874         }
1875         switch (dir) {
1876         case PF_IN:
1877                 printf(" in");
1878                 break;
1879         case PF_OUT:
1880                 printf(" out");
1881                 break;
1882         }
1883         if (skw) {
1884                 printf(" wire: ");
1885                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1886                 printf(" ");
1887                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1888         }
1889         if (sks) {
1890                 printf(" stack: ");
1891                 if (sks != skw) {
1892                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1893                         printf(" ");
1894                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1895                 } else
1896                         printf("-");
1897         }
1898         if (s) {
1899                 if (proto == IPPROTO_TCP) {
1900                         printf(" [lo=%u high=%u win=%u modulator=%u",
1901                             s->src.seqlo, s->src.seqhi,
1902                             s->src.max_win, s->src.seqdiff);
1903                         if (s->src.wscale && s->dst.wscale)
1904                                 printf(" wscale=%u",
1905                                     s->src.wscale & PF_WSCALE_MASK);
1906                         printf("]");
1907                         printf(" [lo=%u high=%u win=%u modulator=%u",
1908                             s->dst.seqlo, s->dst.seqhi,
1909                             s->dst.max_win, s->dst.seqdiff);
1910                         if (s->src.wscale && s->dst.wscale)
1911                                 printf(" wscale=%u",
1912                                 s->dst.wscale & PF_WSCALE_MASK);
1913                         printf("]");
1914                 }
1915                 printf(" %u:%u", s->src.state, s->dst.state);
1916         }
1917 }
1918
1919 void
1920 pf_print_flags(u_int8_t f)
1921 {
1922         if (f)
1923                 printf(" ");
1924         if (f & TH_FIN)
1925                 printf("F");
1926         if (f & TH_SYN)
1927                 printf("S");
1928         if (f & TH_RST)
1929                 printf("R");
1930         if (f & TH_PUSH)
1931                 printf("P");
1932         if (f & TH_ACK)
1933                 printf("A");
1934         if (f & TH_URG)
1935                 printf("U");
1936         if (f & TH_ECE)
1937                 printf("E");
1938         if (f & TH_CWR)
1939                 printf("W");
1940 }
1941
1942 #define PF_SET_SKIP_STEPS(i)                                    \
1943         do {                                                    \
1944                 while (head[i] != cur) {                        \
1945                         head[i]->skip[i].ptr = cur;             \
1946                         head[i] = TAILQ_NEXT(head[i], entries); \
1947                 }                                               \
1948         } while (0)
1949
1950 void
1951 pf_calc_skip_steps(struct pf_rulequeue *rules)
1952 {
1953         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1954         int i;
1955
1956         cur = TAILQ_FIRST(rules);
1957         prev = cur;
1958         for (i = 0; i < PF_SKIP_COUNT; ++i)
1959                 head[i] = cur;
1960         while (cur != NULL) {
1961
1962                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1963                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1964                 if (cur->direction != prev->direction)
1965                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1966                 if (cur->af != prev->af)
1967                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1968                 if (cur->proto != prev->proto)
1969                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1970                 if (cur->src.neg != prev->src.neg ||
1971                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1972                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1973                 if (cur->src.port[0] != prev->src.port[0] ||
1974                     cur->src.port[1] != prev->src.port[1] ||
1975                     cur->src.port_op != prev->src.port_op)
1976                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1977                 if (cur->dst.neg != prev->dst.neg ||
1978                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1979                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1980                 if (cur->dst.port[0] != prev->dst.port[0] ||
1981                     cur->dst.port[1] != prev->dst.port[1] ||
1982                     cur->dst.port_op != prev->dst.port_op)
1983                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1984
1985                 prev = cur;
1986                 cur = TAILQ_NEXT(cur, entries);
1987         }
1988         for (i = 0; i < PF_SKIP_COUNT; ++i)
1989                 PF_SET_SKIP_STEPS(i);
1990 }
1991
1992 static int
1993 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1994 {
1995         if (aw1->type != aw2->type)
1996                 return (1);
1997         switch (aw1->type) {
1998         case PF_ADDR_ADDRMASK:
1999         case PF_ADDR_RANGE:
2000                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2001                         return (1);
2002                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2003                         return (1);
2004                 return (0);
2005         case PF_ADDR_DYNIFTL:
2006                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2007         case PF_ADDR_NOROUTE:
2008         case PF_ADDR_URPFFAILED:
2009                 return (0);
2010         case PF_ADDR_TABLE:
2011                 return (aw1->p.tbl != aw2->p.tbl);
2012         default:
2013                 printf("invalid address type: %d\n", aw1->type);
2014                 return (1);
2015         }
2016 }
2017
2018 /**
2019  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2020  * header isn't always a full checksum. In some cases (i.e. output) it's a
2021  * pseudo-header checksum, which is a partial checksum over src/dst IP
2022  * addresses, protocol number and length.
2023  *
2024  * That means we have the following cases:
2025  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2026  *      checksums, we need to update the checksum whenever we change anything.
2027  *  * Output (i.e. the checksum is a pseudo-header checksum):
2028  *      x The field being updated is src/dst address or affects the length of
2029  *      the packet. We need to update the pseudo-header checksum (note that this
2030  *      checksum is not ones' complement).
2031  *      x Some other field is being modified (e.g. src/dst port numbers): We
2032  *      don't have to update anything.
2033  **/
2034 u_int16_t
2035 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2036 {
2037         u_int32_t       l;
2038
2039         if (udp && !cksum)
2040                 return (0x0000);
2041         l = cksum + old - new;
2042         l = (l >> 16) + (l & 65535);
2043         l = l & 65535;
2044         if (udp && !l)
2045                 return (0xFFFF);
2046         return (l);
2047 }
2048
2049 u_int16_t
2050 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2051         u_int16_t new, u_int8_t udp)
2052 {
2053         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2054                 return (cksum);
2055
2056         return (pf_cksum_fixup(cksum, old, new, udp));
2057 }
2058
2059 static void
2060 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2061         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2062         sa_family_t af)
2063 {
2064         struct pf_addr  ao;
2065         u_int16_t       po = *p;
2066
2067         PF_ACPY(&ao, a, af);
2068         PF_ACPY(a, an, af);
2069
2070         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2071                 *pc = ~*pc;
2072
2073         *p = pn;
2074
2075         switch (af) {
2076 #ifdef INET
2077         case AF_INET:
2078                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2079                     ao.addr16[0], an->addr16[0], 0),
2080                     ao.addr16[1], an->addr16[1], 0);
2081                 *p = pn;
2082
2083                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2084                     ao.addr16[0], an->addr16[0], u),
2085                     ao.addr16[1], an->addr16[1], u);
2086
2087                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2088                 break;
2089 #endif /* INET */
2090 #ifdef INET6
2091         case AF_INET6:
2092                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2093                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2094                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2095                     ao.addr16[0], an->addr16[0], u),
2096                     ao.addr16[1], an->addr16[1], u),
2097                     ao.addr16[2], an->addr16[2], u),
2098                     ao.addr16[3], an->addr16[3], u),
2099                     ao.addr16[4], an->addr16[4], u),
2100                     ao.addr16[5], an->addr16[5], u),
2101                     ao.addr16[6], an->addr16[6], u),
2102                     ao.addr16[7], an->addr16[7], u);
2103
2104                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2105                 break;
2106 #endif /* INET6 */
2107         }
2108
2109         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2110             CSUM_DELAY_DATA_IPV6)) {
2111                 *pc = ~*pc;
2112                 if (! *pc)
2113                         *pc = 0xffff;
2114         }
2115 }
2116
2117 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2118 void
2119 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2120 {
2121         u_int32_t       ao;
2122
2123         memcpy(&ao, a, sizeof(ao));
2124         memcpy(a, &an, sizeof(u_int32_t));
2125         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2126             ao % 65536, an % 65536, u);
2127 }
2128
2129 void
2130 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2131 {
2132         u_int32_t       ao;
2133
2134         memcpy(&ao, a, sizeof(ao));
2135         memcpy(a, &an, sizeof(u_int32_t));
2136
2137         *c = pf_proto_cksum_fixup(m,
2138             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2139             ao % 65536, an % 65536, udp);
2140 }
2141
2142 #ifdef INET6
2143 static void
2144 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2145 {
2146         struct pf_addr  ao;
2147
2148         PF_ACPY(&ao, a, AF_INET6);
2149         PF_ACPY(a, an, AF_INET6);
2150
2151         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2152             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2153             pf_cksum_fixup(pf_cksum_fixup(*c,
2154             ao.addr16[0], an->addr16[0], u),
2155             ao.addr16[1], an->addr16[1], u),
2156             ao.addr16[2], an->addr16[2], u),
2157             ao.addr16[3], an->addr16[3], u),
2158             ao.addr16[4], an->addr16[4], u),
2159             ao.addr16[5], an->addr16[5], u),
2160             ao.addr16[6], an->addr16[6], u),
2161             ao.addr16[7], an->addr16[7], u);
2162 }
2163 #endif /* INET6 */
2164
2165 static void
2166 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2167     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2168     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2169 {
2170         struct pf_addr  oia, ooa;
2171
2172         PF_ACPY(&oia, ia, af);
2173         if (oa)
2174                 PF_ACPY(&ooa, oa, af);
2175
2176         /* Change inner protocol port, fix inner protocol checksum. */
2177         if (ip != NULL) {
2178                 u_int16_t       oip = *ip;
2179                 u_int32_t       opc;
2180
2181                 if (pc != NULL)
2182                         opc = *pc;
2183                 *ip = np;
2184                 if (pc != NULL)
2185                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2186                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2187                 if (pc != NULL)
2188                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2189         }
2190         /* Change inner ip address, fix inner ip and icmp checksums. */
2191         PF_ACPY(ia, na, af);
2192         switch (af) {
2193 #ifdef INET
2194         case AF_INET: {
2195                 u_int32_t        oh2c = *h2c;
2196
2197                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2198                     oia.addr16[0], ia->addr16[0], 0),
2199                     oia.addr16[1], ia->addr16[1], 0);
2200                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2201                     oia.addr16[0], ia->addr16[0], 0),
2202                     oia.addr16[1], ia->addr16[1], 0);
2203                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2204                 break;
2205         }
2206 #endif /* INET */
2207 #ifdef INET6
2208         case AF_INET6:
2209                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2210                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2211                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2212                     oia.addr16[0], ia->addr16[0], u),
2213                     oia.addr16[1], ia->addr16[1], u),
2214                     oia.addr16[2], ia->addr16[2], u),
2215                     oia.addr16[3], ia->addr16[3], u),
2216                     oia.addr16[4], ia->addr16[4], u),
2217                     oia.addr16[5], ia->addr16[5], u),
2218                     oia.addr16[6], ia->addr16[6], u),
2219                     oia.addr16[7], ia->addr16[7], u);
2220                 break;
2221 #endif /* INET6 */
2222         }
2223         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2224         if (oa) {
2225                 PF_ACPY(oa, na, af);
2226                 switch (af) {
2227 #ifdef INET
2228                 case AF_INET:
2229                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2230                             ooa.addr16[0], oa->addr16[0], 0),
2231                             ooa.addr16[1], oa->addr16[1], 0);
2232                         break;
2233 #endif /* INET */
2234 #ifdef INET6
2235                 case AF_INET6:
2236                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2237                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2238                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2239                             ooa.addr16[0], oa->addr16[0], u),
2240                             ooa.addr16[1], oa->addr16[1], u),
2241                             ooa.addr16[2], oa->addr16[2], u),
2242                             ooa.addr16[3], oa->addr16[3], u),
2243                             ooa.addr16[4], oa->addr16[4], u),
2244                             ooa.addr16[5], oa->addr16[5], u),
2245                             ooa.addr16[6], oa->addr16[6], u),
2246                             ooa.addr16[7], oa->addr16[7], u);
2247                         break;
2248 #endif /* INET6 */
2249                 }
2250         }
2251 }
2252
2253
2254 /*
2255  * Need to modulate the sequence numbers in the TCP SACK option
2256  * (credits to Krzysztof Pfaff for report and patch)
2257  */
2258 static int
2259 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2260     struct tcphdr *th, struct pf_state_peer *dst)
2261 {
2262         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2263         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2264         int copyback = 0, i, olen;
2265         struct sackblk sack;
2266
2267 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2268         if (hlen < TCPOLEN_SACKLEN ||
2269             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2270                 return 0;
2271
2272         while (hlen >= TCPOLEN_SACKLEN) {
2273                 olen = opt[1];
2274                 switch (*opt) {
2275                 case TCPOPT_EOL:        /* FALLTHROUGH */
2276                 case TCPOPT_NOP:
2277                         opt++;
2278                         hlen--;
2279                         break;
2280                 case TCPOPT_SACK:
2281                         if (olen > hlen)
2282                                 olen = hlen;
2283                         if (olen >= TCPOLEN_SACKLEN) {
2284                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2285                                     i += TCPOLEN_SACK) {
2286                                         memcpy(&sack, &opt[i], sizeof(sack));
2287                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2288                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2289                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2290                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2291                                         memcpy(&opt[i], &sack, sizeof(sack));
2292                                 }
2293                                 copyback = 1;
2294                         }
2295                         /* FALLTHROUGH */
2296                 default:
2297                         if (olen < 2)
2298                                 olen = 2;
2299                         hlen -= olen;
2300                         opt += olen;
2301                 }
2302         }
2303
2304         if (copyback)
2305                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2306         return (copyback);
2307 }
2308
2309 static void
2310 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2311     const struct pf_addr *saddr, const struct pf_addr *daddr,
2312     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2313     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2314     u_int16_t rtag, struct ifnet *ifp)
2315 {
2316         struct pf_send_entry *pfse;
2317         struct mbuf     *m;
2318         int              len, tlen;
2319 #ifdef INET
2320         struct ip       *h = NULL;
2321 #endif /* INET */
2322 #ifdef INET6
2323         struct ip6_hdr  *h6 = NULL;
2324 #endif /* INET6 */
2325         struct tcphdr   *th;
2326         char            *opt;
2327         struct pf_mtag  *pf_mtag;
2328
2329         len = 0;
2330         th = NULL;
2331
2332         /* maximum segment size tcp option */
2333         tlen = sizeof(struct tcphdr);
2334         if (mss)
2335                 tlen += 4;
2336
2337         switch (af) {
2338 #ifdef INET
2339         case AF_INET:
2340                 len = sizeof(struct ip) + tlen;
2341                 break;
2342 #endif /* INET */
2343 #ifdef INET6
2344         case AF_INET6:
2345                 len = sizeof(struct ip6_hdr) + tlen;
2346                 break;
2347 #endif /* INET6 */
2348         default:
2349                 panic("%s: unsupported af %d", __func__, af);
2350         }
2351
2352         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2353         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2354         if (pfse == NULL)
2355                 return;
2356         m = m_gethdr(M_NOWAIT, MT_DATA);
2357         if (m == NULL) {
2358                 free(pfse, M_PFTEMP);
2359                 return;
2360         }
2361 #ifdef MAC
2362         mac_netinet_firewall_send(m);
2363 #endif
2364         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2365                 free(pfse, M_PFTEMP);
2366                 m_freem(m);
2367                 return;
2368         }
2369         if (tag)
2370                 m->m_flags |= M_SKIP_FIREWALL;
2371         pf_mtag->tag = rtag;
2372
2373         if (r != NULL && r->rtableid >= 0)
2374                 M_SETFIB(m, r->rtableid);
2375
2376 #ifdef ALTQ
2377         if (r != NULL && r->qid) {
2378                 pf_mtag->qid = r->qid;
2379
2380                 /* add hints for ecn */
2381                 pf_mtag->hdr = mtod(m, struct ip *);
2382         }
2383 #endif /* ALTQ */
2384         m->m_data += max_linkhdr;
2385         m->m_pkthdr.len = m->m_len = len;
2386         m->m_pkthdr.rcvif = NULL;
2387         bzero(m->m_data, len);
2388         switch (af) {
2389 #ifdef INET
2390         case AF_INET:
2391                 h = mtod(m, struct ip *);
2392
2393                 /* IP header fields included in the TCP checksum */
2394                 h->ip_p = IPPROTO_TCP;
2395                 h->ip_len = htons(tlen);
2396                 h->ip_src.s_addr = saddr->v4.s_addr;
2397                 h->ip_dst.s_addr = daddr->v4.s_addr;
2398
2399                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2400                 break;
2401 #endif /* INET */
2402 #ifdef INET6
2403         case AF_INET6:
2404                 h6 = mtod(m, struct ip6_hdr *);
2405
2406                 /* IP header fields included in the TCP checksum */
2407                 h6->ip6_nxt = IPPROTO_TCP;
2408                 h6->ip6_plen = htons(tlen);
2409                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2410                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2411
2412                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2413                 break;
2414 #endif /* INET6 */
2415         }
2416
2417         /* TCP header */
2418         th->th_sport = sport;
2419         th->th_dport = dport;
2420         th->th_seq = htonl(seq);
2421         th->th_ack = htonl(ack);
2422         th->th_off = tlen >> 2;
2423         th->th_flags = flags;
2424         th->th_win = htons(win);
2425
2426         if (mss) {
2427                 opt = (char *)(th + 1);
2428                 opt[0] = TCPOPT_MAXSEG;
2429                 opt[1] = 4;
2430                 HTONS(mss);
2431                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2432         }
2433
2434         switch (af) {
2435 #ifdef INET
2436         case AF_INET:
2437                 /* TCP checksum */
2438                 th->th_sum = in_cksum(m, len);
2439
2440                 /* Finish the IP header */
2441                 h->ip_v = 4;
2442                 h->ip_hl = sizeof(*h) >> 2;
2443                 h->ip_tos = IPTOS_LOWDELAY;
2444                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2445                 h->ip_len = htons(len);
2446                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2447                 h->ip_sum = 0;
2448
2449                 pfse->pfse_type = PFSE_IP;
2450                 break;
2451 #endif /* INET */
2452 #ifdef INET6
2453         case AF_INET6:
2454                 /* TCP checksum */
2455                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2456                     sizeof(struct ip6_hdr), tlen);
2457
2458                 h6->ip6_vfc |= IPV6_VERSION;
2459                 h6->ip6_hlim = IPV6_DEFHLIM;
2460
2461                 pfse->pfse_type = PFSE_IP6;
2462                 break;
2463 #endif /* INET6 */
2464         }
2465         pfse->pfse_m = m;
2466         pf_send(pfse);
2467 }
2468
2469 static int
2470 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2471 {
2472         struct m_tag *mtag;
2473
2474         KASSERT(prio <= PF_PRIO_MAX,
2475             ("%s with invalid pcp", __func__));
2476
2477         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2478         if (mtag == NULL) {
2479                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2480                     sizeof(uint8_t), M_NOWAIT);
2481                 if (mtag == NULL)
2482                         return (ENOMEM);
2483                 m_tag_prepend(m, mtag);
2484         }
2485
2486         *(uint8_t *)(mtag + 1) = prio;
2487         return (0);
2488 }
2489
2490 static int
2491 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2492 {
2493         struct m_tag *mtag;
2494         u_int8_t mpcp;
2495
2496         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2497         if (mtag == NULL)
2498                 return (0);
2499
2500         if (prio == PF_PRIO_ZERO)
2501                 prio = 0;
2502
2503         mpcp = *(uint8_t *)(mtag + 1);
2504
2505         return (mpcp == prio);
2506 }
2507
2508 static void
2509 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2510     struct pf_rule *r)
2511 {
2512         struct pf_send_entry *pfse;
2513         struct mbuf *m0;
2514         struct pf_mtag *pf_mtag;
2515
2516         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2517         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2518         if (pfse == NULL)
2519                 return;
2520
2521         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2522                 free(pfse, M_PFTEMP);
2523                 return;
2524         }
2525
2526         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2527                 free(pfse, M_PFTEMP);
2528                 return;
2529         }
2530         /* XXX: revisit */
2531         m0->m_flags |= M_SKIP_FIREWALL;
2532
2533         if (r->rtableid >= 0)
2534                 M_SETFIB(m0, r->rtableid);
2535
2536 #ifdef ALTQ
2537         if (r->qid) {
2538                 pf_mtag->qid = r->qid;
2539                 /* add hints for ecn */
2540                 pf_mtag->hdr = mtod(m0, struct ip *);
2541         }
2542 #endif /* ALTQ */
2543
2544         switch (af) {
2545 #ifdef INET
2546         case AF_INET:
2547                 pfse->pfse_type = PFSE_ICMP;
2548                 break;
2549 #endif /* INET */
2550 #ifdef INET6
2551         case AF_INET6:
2552                 pfse->pfse_type = PFSE_ICMP6;
2553                 break;
2554 #endif /* INET6 */
2555         }
2556         pfse->pfse_m = m0;
2557         pfse->icmpopts.type = type;
2558         pfse->icmpopts.code = code;
2559         pf_send(pfse);
2560 }
2561
2562 /*
2563  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2564  * If n is 0, they match if they are equal. If n is != 0, they match if they
2565  * are different.
2566  */
2567 int
2568 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2569     struct pf_addr *b, sa_family_t af)
2570 {
2571         int     match = 0;
2572
2573         switch (af) {
2574 #ifdef INET
2575         case AF_INET:
2576                 if ((a->addr32[0] & m->addr32[0]) ==
2577                     (b->addr32[0] & m->addr32[0]))
2578                         match++;
2579                 break;
2580 #endif /* INET */
2581 #ifdef INET6
2582         case AF_INET6:
2583                 if (((a->addr32[0] & m->addr32[0]) ==
2584                      (b->addr32[0] & m->addr32[0])) &&
2585                     ((a->addr32[1] & m->addr32[1]) ==
2586                      (b->addr32[1] & m->addr32[1])) &&
2587                     ((a->addr32[2] & m->addr32[2]) ==
2588                      (b->addr32[2] & m->addr32[2])) &&
2589                     ((a->addr32[3] & m->addr32[3]) ==
2590                      (b->addr32[3] & m->addr32[3])))
2591                         match++;
2592                 break;
2593 #endif /* INET6 */
2594         }
2595         if (match) {
2596                 if (n)
2597                         return (0);
2598                 else
2599                         return (1);
2600         } else {
2601                 if (n)
2602                         return (1);
2603                 else
2604                         return (0);
2605         }
2606 }
2607
2608 /*
2609  * Return 1 if b <= a <= e, otherwise return 0.
2610  */
2611 int
2612 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2613     struct pf_addr *a, sa_family_t af)
2614 {
2615         switch (af) {
2616 #ifdef INET
2617         case AF_INET:
2618                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2619                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2620                         return (0);
2621                 break;
2622 #endif /* INET */
2623 #ifdef INET6
2624         case AF_INET6: {
2625                 int     i;
2626
2627                 /* check a >= b */
2628                 for (i = 0; i < 4; ++i)
2629                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2630                                 break;
2631                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2632                                 return (0);
2633                 /* check a <= e */
2634                 for (i = 0; i < 4; ++i)
2635                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2636                                 break;
2637                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2638                                 return (0);
2639                 break;
2640         }
2641 #endif /* INET6 */
2642         }
2643         return (1);
2644 }
2645
2646 static int
2647 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2648 {
2649         switch (op) {
2650         case PF_OP_IRG:
2651                 return ((p > a1) && (p < a2));
2652         case PF_OP_XRG:
2653                 return ((p < a1) || (p > a2));
2654         case PF_OP_RRG:
2655                 return ((p >= a1) && (p <= a2));
2656         case PF_OP_EQ:
2657                 return (p == a1);
2658         case PF_OP_NE:
2659                 return (p != a1);
2660         case PF_OP_LT:
2661                 return (p < a1);
2662         case PF_OP_LE:
2663                 return (p <= a1);
2664         case PF_OP_GT:
2665                 return (p > a1);
2666         case PF_OP_GE:
2667                 return (p >= a1);
2668         }
2669         return (0); /* never reached */
2670 }
2671
2672 int
2673 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2674 {
2675         NTOHS(a1);
2676         NTOHS(a2);
2677         NTOHS(p);
2678         return (pf_match(op, a1, a2, p));
2679 }
2680
2681 static int
2682 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2683 {
2684         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2685                 return (0);
2686         return (pf_match(op, a1, a2, u));
2687 }
2688
2689 static int
2690 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2691 {
2692         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2693                 return (0);
2694         return (pf_match(op, a1, a2, g));
2695 }
2696
2697 int
2698 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2699 {
2700         if (*tag == -1)
2701                 *tag = mtag;
2702
2703         return ((!r->match_tag_not && r->match_tag == *tag) ||
2704             (r->match_tag_not && r->match_tag != *tag));
2705 }
2706
2707 int
2708 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2709 {
2710
2711         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2712
2713         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2714                 return (ENOMEM);
2715
2716         pd->pf_mtag->tag = tag;
2717
2718         return (0);
2719 }
2720
2721 #define PF_ANCHOR_STACKSIZE     32
2722 struct pf_anchor_stackframe {
2723         struct pf_ruleset       *rs;
2724         struct pf_rule          *r;     /* XXX: + match bit */
2725         struct pf_anchor        *child;
2726 };
2727
2728 /*
2729  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2730  */
2731 #define PF_ANCHORSTACK_MATCH    0x00000001
2732 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2733
2734 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2735 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2736                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2737 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2738                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2739 } while (0)
2740
2741 void
2742 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2743     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2744     int *match)
2745 {
2746         struct pf_anchor_stackframe     *f;
2747
2748         PF_RULES_RASSERT();
2749
2750         if (match)
2751                 *match = 0;
2752         if (*depth >= PF_ANCHOR_STACKSIZE) {
2753                 printf("%s: anchor stack overflow on %s\n",
2754                     __func__, (*r)->anchor->name);
2755                 *r = TAILQ_NEXT(*r, entries);
2756                 return;
2757         } else if (*depth == 0 && a != NULL)
2758                 *a = *r;
2759         f = stack + (*depth)++;
2760         f->rs = *rs;
2761         f->r = *r;
2762         if ((*r)->anchor_wildcard) {
2763                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2764
2765                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2766                         *r = NULL;
2767                         return;
2768                 }
2769                 *rs = &f->child->ruleset;
2770         } else {
2771                 f->child = NULL;
2772                 *rs = &(*r)->anchor->ruleset;
2773         }
2774         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2775 }
2776
2777 int
2778 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2779     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2780     int *match)
2781 {
2782         struct pf_anchor_stackframe     *f;
2783         struct pf_rule *fr;
2784         int quick = 0;
2785
2786         PF_RULES_RASSERT();
2787
2788         do {
2789                 if (*depth <= 0)
2790                         break;
2791                 f = stack + *depth - 1;
2792                 fr = PF_ANCHOR_RULE(f);
2793                 if (f->child != NULL) {
2794                         struct pf_anchor_node *parent;
2795
2796                         /*
2797                          * This block traverses through
2798                          * a wildcard anchor.
2799                          */
2800                         parent = &fr->anchor->children;
2801                         if (match != NULL && *match) {
2802                                 /*
2803                                  * If any of "*" matched, then
2804                                  * "foo/ *" matched, mark frame
2805                                  * appropriately.
2806                                  */
2807                                 PF_ANCHOR_SET_MATCH(f);
2808                                 *match = 0;
2809                         }
2810                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2811                         if (f->child != NULL) {
2812                                 *rs = &f->child->ruleset;
2813                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2814                                 if (*r == NULL)
2815                                         continue;
2816                                 else
2817                                         break;
2818                         }
2819                 }
2820                 (*depth)--;
2821                 if (*depth == 0 && a != NULL)
2822                         *a = NULL;
2823                 *rs = f->rs;
2824                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2825                         quick = fr->quick;
2826                 *r = TAILQ_NEXT(fr, entries);
2827         } while (*r == NULL);
2828
2829         return (quick);
2830 }
2831
2832 #ifdef INET6
2833 void
2834 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2835     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2836 {
2837         switch (af) {
2838 #ifdef INET
2839         case AF_INET:
2840                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2841                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2842                 break;
2843 #endif /* INET */
2844         case AF_INET6:
2845                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2846                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2847                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2848                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2849                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2850                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2851                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2852                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2853                 break;
2854         }
2855 }
2856
2857 void
2858 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2859 {
2860         switch (af) {
2861 #ifdef INET
2862         case AF_INET:
2863                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2864                 break;
2865 #endif /* INET */
2866         case AF_INET6:
2867                 if (addr->addr32[3] == 0xffffffff) {
2868                         addr->addr32[3] = 0;
2869                         if (addr->addr32[2] == 0xffffffff) {
2870                                 addr->addr32[2] = 0;
2871                                 if (addr->addr32[1] == 0xffffffff) {
2872                                         addr->addr32[1] = 0;
2873                                         addr->addr32[0] =
2874                                             htonl(ntohl(addr->addr32[0]) + 1);
2875                                 } else
2876                                         addr->addr32[1] =
2877                                             htonl(ntohl(addr->addr32[1]) + 1);
2878                         } else
2879                                 addr->addr32[2] =
2880                                     htonl(ntohl(addr->addr32[2]) + 1);
2881                 } else
2882                         addr->addr32[3] =
2883                             htonl(ntohl(addr->addr32[3]) + 1);
2884                 break;
2885         }
2886 }
2887 #endif /* INET6 */
2888
2889 int
2890 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2891 {
2892         struct pf_addr          *saddr, *daddr;
2893         u_int16_t                sport, dport;
2894         struct inpcbinfo        *pi;
2895         struct inpcb            *inp;
2896
2897         pd->lookup.uid = UID_MAX;
2898         pd->lookup.gid = GID_MAX;
2899
2900         switch (pd->proto) {
2901         case IPPROTO_TCP:
2902                 if (pd->hdr.tcp == NULL)
2903                         return (-1);
2904                 sport = pd->hdr.tcp->th_sport;
2905                 dport = pd->hdr.tcp->th_dport;
2906                 pi = &V_tcbinfo;
2907                 break;
2908         case IPPROTO_UDP:
2909                 if (pd->hdr.udp == NULL)
2910                         return (-1);
2911                 sport = pd->hdr.udp->uh_sport;
2912                 dport = pd->hdr.udp->uh_dport;
2913                 pi = &V_udbinfo;
2914                 break;
2915         default:
2916                 return (-1);
2917         }
2918         if (direction == PF_IN) {
2919                 saddr = pd->src;
2920                 daddr = pd->dst;
2921         } else {
2922                 u_int16_t       p;
2923
2924                 p = sport;
2925                 sport = dport;
2926                 dport = p;
2927                 saddr = pd->dst;
2928                 daddr = pd->src;
2929         }
2930         switch (pd->af) {
2931 #ifdef INET
2932         case AF_INET:
2933                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2934                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2935                 if (inp == NULL) {
2936                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2937                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2938                            INPLOOKUP_RLOCKPCB, NULL, m);
2939                         if (inp == NULL)
2940                                 return (-1);
2941                 }
2942                 break;
2943 #endif /* INET */
2944 #ifdef INET6
2945         case AF_INET6:
2946                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2947                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2948                 if (inp == NULL) {
2949                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2950                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2951                             INPLOOKUP_RLOCKPCB, NULL, m);
2952                         if (inp == NULL)
2953                                 return (-1);
2954                 }
2955                 break;
2956 #endif /* INET6 */
2957
2958         default:
2959                 return (-1);
2960         }
2961         INP_RLOCK_ASSERT(inp);
2962         pd->lookup.uid = inp->inp_cred->cr_uid;
2963         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2964         INP_RUNLOCK(inp);
2965
2966         return (1);
2967 }
2968
2969 static u_int8_t
2970 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2971 {
2972         int              hlen;
2973         u_int8_t         hdr[60];
2974         u_int8_t        *opt, optlen;
2975         u_int8_t         wscale = 0;
2976
2977         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2978         if (hlen <= sizeof(struct tcphdr))
2979                 return (0);
2980         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2981                 return (0);
2982         opt = hdr + sizeof(struct tcphdr);
2983         hlen -= sizeof(struct tcphdr);
2984         while (hlen >= 3) {
2985                 switch (*opt) {
2986                 case TCPOPT_EOL:
2987                 case TCPOPT_NOP:
2988                         ++opt;
2989                         --hlen;
2990                         break;
2991                 case TCPOPT_WINDOW:
2992                         wscale = opt[2];
2993                         if (wscale > TCP_MAX_WINSHIFT)
2994                                 wscale = TCP_MAX_WINSHIFT;
2995                         wscale |= PF_WSCALE_FLAG;
2996                         /* FALLTHROUGH */
2997                 default:
2998                         optlen = opt[1];
2999                         if (optlen < 2)
3000                                 optlen = 2;
3001                         hlen -= optlen;
3002                         opt += optlen;
3003                         break;
3004                 }
3005         }
3006         return (wscale);
3007 }
3008
3009 static u_int16_t
3010 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3011 {
3012         int              hlen;
3013         u_int8_t         hdr[60];
3014         u_int8_t        *opt, optlen;
3015         u_int16_t        mss = V_tcp_mssdflt;
3016
3017         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3018         if (hlen <= sizeof(struct tcphdr))
3019                 return (0);
3020         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3021                 return (0);
3022         opt = hdr + sizeof(struct tcphdr);
3023         hlen -= sizeof(struct tcphdr);
3024         while (hlen >= TCPOLEN_MAXSEG) {
3025                 switch (*opt) {
3026                 case TCPOPT_EOL:
3027                 case TCPOPT_NOP:
3028                         ++opt;
3029                         --hlen;
3030                         break;
3031                 case TCPOPT_MAXSEG:
3032                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3033                         NTOHS(mss);
3034                         /* FALLTHROUGH */
3035                 default:
3036                         optlen = opt[1];
3037                         if (optlen < 2)
3038                                 optlen = 2;
3039                         hlen -= optlen;
3040                         opt += optlen;
3041                         break;
3042                 }
3043         }
3044         return (mss);
3045 }
3046
3047 static u_int16_t
3048 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3049 {
3050 #ifdef INET
3051         struct nhop4_basic      nh4;
3052 #endif /* INET */
3053 #ifdef INET6
3054         struct nhop6_basic      nh6;
3055         struct in6_addr         dst6;
3056         uint32_t                scopeid;
3057 #endif /* INET6 */
3058         int                      hlen = 0;
3059         uint16_t                 mss = 0;
3060
3061         switch (af) {
3062 #ifdef INET
3063         case AF_INET:
3064                 hlen = sizeof(struct ip);
3065                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3066                         mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3067                 break;
3068 #endif /* INET */
3069 #ifdef INET6
3070         case AF_INET6:
3071                 hlen = sizeof(struct ip6_hdr);
3072                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3073                 if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3074                         mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3075                 break;
3076 #endif /* INET6 */
3077         }
3078
3079         mss = max(V_tcp_mssdflt, mss);
3080         mss = min(mss, offer);
3081         mss = max(mss, 64);             /* sanity - at least max opt space */
3082         return (mss);
3083 }
3084
3085 static u_int32_t
3086 pf_tcp_iss(struct pf_pdesc *pd)
3087 {
3088         MD5_CTX ctx;
3089         u_int32_t digest[4];
3090
3091         if (V_pf_tcp_secret_init == 0) {
3092                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3093                 MD5Init(&V_pf_tcp_secret_ctx);
3094                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3095                     sizeof(V_pf_tcp_secret));
3096                 V_pf_tcp_secret_init = 1;
3097         }
3098
3099         ctx = V_pf_tcp_secret_ctx;
3100
3101         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3102         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3103         if (pd->af == AF_INET6) {
3104                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3105                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3106         } else {
3107                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3108                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3109         }
3110         MD5Final((u_char *)digest, &ctx);
3111         V_pf_tcp_iss_off += 4096;
3112 #define ISN_RANDOM_INCREMENT (4096 - 1)
3113         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3114             V_pf_tcp_iss_off);
3115 #undef  ISN_RANDOM_INCREMENT
3116 }
3117
3118 static int
3119 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3120     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3121     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3122 {
3123         struct pf_rule          *nr = NULL;
3124         struct pf_addr          * const saddr = pd->src;
3125         struct pf_addr          * const daddr = pd->dst;
3126         sa_family_t              af = pd->af;
3127         struct pf_rule          *r, *a = NULL;
3128         struct pf_ruleset       *ruleset = NULL;
3129         struct pf_src_node      *nsn = NULL;
3130         struct tcphdr           *th = pd->hdr.tcp;
3131         struct pf_state_key     *sk = NULL, *nk = NULL;
3132         u_short                  reason;
3133         int                      rewrite = 0, hdrlen = 0;
3134         int                      tag = -1, rtableid = -1;
3135         int                      asd = 0;
3136         int                      match = 0;
3137         int                      state_icmp = 0;
3138         u_int16_t                sport = 0, dport = 0;
3139         u_int16_t                bproto_sum = 0, bip_sum = 0;
3140         u_int8_t                 icmptype = 0, icmpcode = 0;
3141         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3142
3143         PF_RULES_RASSERT();
3144
3145         if (inp != NULL) {
3146                 INP_LOCK_ASSERT(inp);
3147                 pd->lookup.uid = inp->inp_cred->cr_uid;
3148                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3149                 pd->lookup.done = 1;
3150         }
3151
3152         switch (pd->proto) {
3153         case IPPROTO_TCP:
3154                 sport = th->th_sport;
3155                 dport = th->th_dport;
3156                 hdrlen = sizeof(*th);
3157                 break;
3158         case IPPROTO_UDP:
3159                 sport = pd->hdr.udp->uh_sport;
3160                 dport = pd->hdr.udp->uh_dport;
3161                 hdrlen = sizeof(*pd->hdr.udp);
3162                 break;
3163 #ifdef INET
3164         case IPPROTO_ICMP:
3165                 if (pd->af != AF_INET)
3166                         break;
3167                 sport = dport = pd->hdr.icmp->icmp_id;
3168                 hdrlen = sizeof(*pd->hdr.icmp);
3169                 icmptype = pd->hdr.icmp->icmp_type;
3170                 icmpcode = pd->hdr.icmp->icmp_code;
3171
3172                 if (icmptype == ICMP_UNREACH ||
3173                     icmptype == ICMP_SOURCEQUENCH ||
3174                     icmptype == ICMP_REDIRECT ||
3175                     icmptype == ICMP_TIMXCEED ||
3176                     icmptype == ICMP_PARAMPROB)
3177                         state_icmp++;
3178                 break;
3179 #endif /* INET */
3180 #ifdef INET6
3181         case IPPROTO_ICMPV6:
3182                 if (af != AF_INET6)
3183                         break;
3184                 sport = dport = pd->hdr.icmp6->icmp6_id;
3185                 hdrlen = sizeof(*pd->hdr.icmp6);
3186                 icmptype = pd->hdr.icmp6->icmp6_type;
3187                 icmpcode = pd->hdr.icmp6->icmp6_code;
3188
3189                 if (icmptype == ICMP6_DST_UNREACH ||
3190                     icmptype == ICMP6_PACKET_TOO_BIG ||
3191                     icmptype == ICMP6_TIME_EXCEEDED ||
3192                     icmptype == ICMP6_PARAM_PROB)
3193                         state_icmp++;
3194                 break;
3195 #endif /* INET6 */
3196         default:
3197                 sport = dport = hdrlen = 0;
3198                 break;
3199         }
3200
3201         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3202
3203         /* check packet for BINAT/NAT/RDR */
3204         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3205             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3206                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3207                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3208
3209                 if (pd->ip_sum)
3210                         bip_sum = *pd->ip_sum;
3211
3212                 switch (pd->proto) {
3213                 case IPPROTO_TCP:
3214                         bproto_sum = th->th_sum;
3215                         pd->proto_sum = &th->th_sum;
3216
3217                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3218                             nk->port[pd->sidx] != sport) {
3219                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3220                                     &th->th_sum, &nk->addr[pd->sidx],
3221                                     nk->port[pd->sidx], 0, af);
3222                                 pd->sport = &th->th_sport;
3223                                 sport = th->th_sport;
3224                         }
3225
3226                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3227                             nk->port[pd->didx] != dport) {
3228                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3229                                     &th->th_sum, &nk->addr[pd->didx],
3230                                     nk->port[pd->didx], 0, af);
3231                                 dport = th->th_dport;
3232                                 pd->dport = &th->th_dport;
3233                         }
3234                         rewrite++;
3235                         break;
3236                 case IPPROTO_UDP:
3237                         bproto_sum = pd->hdr.udp->uh_sum;
3238                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3239
3240                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3241                             nk->port[pd->sidx] != sport) {
3242                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3243                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3244                                     &nk->addr[pd->sidx],
3245                                     nk->port[pd->sidx], 1, af);
3246                                 sport = pd->hdr.udp->uh_sport;
3247                                 pd->sport = &pd->hdr.udp->uh_sport;
3248                         }
3249
3250                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3251                             nk->port[pd->didx] != dport) {
3252                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3253                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3254                                     &nk->addr[pd->didx],
3255                                     nk->port[pd->didx], 1, af);
3256                                 dport = pd->hdr.udp->uh_dport;
3257                                 pd->dport = &pd->hdr.udp->uh_dport;
3258                         }
3259                         rewrite++;
3260                         break;
3261 #ifdef INET
3262                 case IPPROTO_ICMP:
3263                         nk->port[0] = nk->port[1];
3264                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3265                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3266                                     nk->addr[pd->sidx].v4.s_addr, 0);
3267
3268                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3269                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3270                                     nk->addr[pd->didx].v4.s_addr, 0);
3271
3272                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3273                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3274                                     pd->hdr.icmp->icmp_cksum, sport,
3275                                     nk->port[1], 0);
3276                                 pd->hdr.icmp->icmp_id = nk->port[1];
3277                                 pd->sport = &pd->hdr.icmp->icmp_id;
3278                         }
3279                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3280                         break;
3281 #endif /* INET */
3282 #ifdef INET6
3283                 case IPPROTO_ICMPV6:
3284                         nk->port[0] = nk->port[1];
3285                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3286                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3287                                     &nk->addr[pd->sidx], 0);
3288
3289                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3290                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3291                                     &nk->addr[pd->didx], 0);
3292                         rewrite++;
3293                         break;
3294 #endif /* INET */
3295                 default:
3296                         switch (af) {
3297 #ifdef INET
3298                         case AF_INET:
3299                                 if (PF_ANEQ(saddr,
3300                                     &nk->addr[pd->sidx], AF_INET))
3301                                         pf_change_a(&saddr->v4.s_addr,
3302                                             pd->ip_sum,
3303                                             nk->addr[pd->sidx].v4.s_addr, 0);
3304
3305                                 if (PF_ANEQ(daddr,
3306                                     &nk->addr[pd->didx], AF_INET))
3307                                         pf_change_a(&daddr->v4.s_addr,
3308                                             pd->ip_sum,
3309                                             nk->addr[pd->didx].v4.s_addr, 0);
3310                                 break;
3311 #endif /* INET */
3312 #ifdef INET6
3313                         case AF_INET6:
3314                                 if (PF_ANEQ(saddr,
3315                                     &nk->addr[pd->sidx], AF_INET6))
3316                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3317
3318                                 if (PF_ANEQ(daddr,
3319                                     &nk->addr[pd->didx], AF_INET6))
3320                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3321                                 break;
3322 #endif /* INET */
3323                         }
3324                         break;
3325                 }
3326                 if (nr->natpass)
3327                         r = NULL;
3328                 pd->nat_rule = nr;
3329         }
3330
3331         while (r != NULL) {
3332                 r->evaluations++;
3333                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3334                         r = r->skip[PF_SKIP_IFP].ptr;
3335                 else if (r->direction && r->direction != direction)
3336                         r = r->skip[PF_SKIP_DIR].ptr;
3337                 else if (r->af && r->af != af)
3338                         r = r->skip[PF_SKIP_AF].ptr;
3339                 else if (r->proto && r->proto != pd->proto)
3340                         r = r->skip[PF_SKIP_PROTO].ptr;
3341                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3342                     r->src.neg, kif, M_GETFIB(m)))
3343                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3344                 /* tcp/udp only. port_op always 0 in other cases */
3345                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3346                     r->src.port[0], r->src.port[1], sport))
3347                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3348                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3349                     r->dst.neg, NULL, M_GETFIB(m)))
3350                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3351                 /* tcp/udp only. port_op always 0 in other cases */
3352                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3353                     r->dst.port[0], r->dst.port[1], dport))
3354                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3355                 /* icmp only. type always 0 in other cases */
3356                 else if (r->type && r->type != icmptype + 1)
3357                         r = TAILQ_NEXT(r, entries);
3358                 /* icmp only. type always 0 in other cases */
3359                 else if (r->code && r->code != icmpcode + 1)
3360                         r = TAILQ_NEXT(r, entries);
3361                 else if (r->tos && !(r->tos == pd->tos))
3362                         r = TAILQ_NEXT(r, entries);
3363                 else if (r->rule_flag & PFRULE_FRAGMENT)
3364                         r = TAILQ_NEXT(r, entries);
3365                 else if (pd->proto == IPPROTO_TCP &&
3366                     (r->flagset & th->th_flags) != r->flags)
3367                         r = TAILQ_NEXT(r, entries);
3368                 /* tcp/udp only. uid.op always 0 in other cases */
3369                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3370                     pf_socket_lookup(direction, pd, m), 1)) &&
3371                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3372                     pd->lookup.uid))
3373                         r = TAILQ_NEXT(r, entries);
3374                 /* tcp/udp only. gid.op always 0 in other cases */
3375                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3376                     pf_socket_lookup(direction, pd, m), 1)) &&
3377                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3378                     pd->lookup.gid))
3379                         r = TAILQ_NEXT(r, entries);
3380                 else if (r->prio &&
3381                     !pf_match_ieee8021q_pcp(r->prio, m))
3382                         r = TAILQ_NEXT(r, entries);
3383                 else if (r->prob &&
3384                     r->prob <= arc4random())
3385                         r = TAILQ_NEXT(r, entries);
3386                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3387                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3388                         r = TAILQ_NEXT(r, entries);
3389                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3390                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3391                     pf_osfp_fingerprint(pd, m, off, th),
3392                     r->os_fingerprint)))
3393                         r = TAILQ_NEXT(r, entries);
3394                 else {
3395                         if (r->tag)
3396                                 tag = r->tag;
3397                         if (r->rtableid >= 0)
3398                                 rtableid = r->rtableid;
3399                         if (r->anchor == NULL) {
3400                                 match = 1;
3401                                 *rm = r;
3402                                 *am = a;
3403                                 *rsm = ruleset;
3404                                 if ((*rm)->quick)
3405                                         break;
3406                                 r = TAILQ_NEXT(r, entries);
3407                         } else
3408                                 pf_step_into_anchor(anchor_stack, &asd,
3409                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3410                                     &match);
3411                 }
3412                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3413                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3414                         break;
3415         }
3416         r = *rm;
3417         a = *am;
3418         ruleset = *rsm;
3419
3420         REASON_SET(&reason, PFRES_MATCH);
3421
3422         if (r->log || (nr != NULL && nr->log)) {
3423                 if (rewrite)
3424                         m_copyback(m, off, hdrlen, pd->hdr.any);
3425                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3426                     ruleset, pd, 1);
3427         }
3428
3429         if ((r->action == PF_DROP) &&
3430             ((r->rule_flag & PFRULE_RETURNRST) ||
3431             (r->rule_flag & PFRULE_RETURNICMP) ||
3432             (r->rule_flag & PFRULE_RETURN))) {
3433                 /* undo NAT changes, if they have taken place */
3434                 if (nr != NULL) {
3435                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3436                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3437                         if (pd->sport)
3438                                 *pd->sport = sk->port[pd->sidx];
3439                         if (pd->dport)
3440                                 *pd->dport = sk->port[pd->didx];
3441                         if (pd->proto_sum)
3442                                 *pd->proto_sum = bproto_sum;
3443                         if (pd->ip_sum)
3444                                 *pd->ip_sum = bip_sum;
3445                         m_copyback(m, off, hdrlen, pd->hdr.any);
3446                 }
3447                 if (pd->proto == IPPROTO_TCP &&
3448                     ((r->rule_flag & PFRULE_RETURNRST) ||
3449                     (r->rule_flag & PFRULE_RETURN)) &&
3450                     !(th->th_flags & TH_RST)) {
3451                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3452                         int              len = 0;
3453 #ifdef INET
3454                         struct ip       *h4;
3455 #endif
3456 #ifdef INET6
3457                         struct ip6_hdr  *h6;
3458 #endif
3459
3460                         switch (af) {
3461 #ifdef INET
3462                         case AF_INET:
3463                                 h4 = mtod(m, struct ip *);
3464                                 len = ntohs(h4->ip_len) - off;
3465                                 break;
3466 #endif
3467 #ifdef INET6
3468                         case AF_INET6:
3469                                 h6 = mtod(m, struct ip6_hdr *);
3470                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3471                                 break;
3472 #endif
3473                         }
3474
3475                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3476                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3477                         else {
3478                                 if (th->th_flags & TH_SYN)
3479                                         ack++;
3480                                 if (th->th_flags & TH_FIN)
3481                                         ack++;
3482                                 pf_send_tcp(m, r, af, pd->dst,
3483                                     pd->src, th->th_dport, th->th_sport,
3484                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3485                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3486                         }
3487                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3488                     r->return_icmp)
3489                         pf_send_icmp(m, r->return_icmp >> 8,
3490                             r->return_icmp & 255, af, r);
3491                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3492                     r->return_icmp6)
3493                         pf_send_icmp(m, r->return_icmp6 >> 8,
3494                             r->return_icmp6 & 255, af, r);
3495         }
3496
3497         if (r->action == PF_DROP)
3498                 goto cleanup;
3499
3500         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3501                 REASON_SET(&reason, PFRES_MEMORY);
3502                 goto cleanup;
3503         }
3504         if (rtableid >= 0)
3505                 M_SETFIB(m, rtableid);
3506
3507         if (!state_icmp && (r->keep_state || nr != NULL ||
3508             (pd->flags & PFDESC_TCP_NORM))) {
3509                 int action;
3510                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3511                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3512                     hdrlen);
3513                 if (action != PF_PASS)
3514                         return (action);
3515         } else {
3516                 if (sk != NULL)
3517                         uma_zfree(V_pf_state_key_z, sk);
3518                 if (nk != NULL)
3519                         uma_zfree(V_pf_state_key_z, nk);
3520         }
3521
3522         /* copy back packet headers if we performed NAT operations */
3523         if (rewrite)
3524                 m_copyback(m, off, hdrlen, pd->hdr.any);
3525
3526         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3527             direction == PF_OUT &&
3528             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3529                 /*
3530                  * We want the state created, but we dont
3531                  * want to send this in case a partner
3532                  * firewall has to know about it to allow
3533                  * replies through it.
3534                  */
3535                 return (PF_DEFER);
3536
3537         return (PF_PASS);
3538
3539 cleanup:
3540         if (sk != NULL)
3541                 uma_zfree(V_pf_state_key_z, sk);
3542         if (nk != NULL)
3543                 uma_zfree(V_pf_state_key_z, nk);
3544         return (PF_DROP);
3545 }
3546
3547 static int
3548 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3549     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3550     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3551     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3552     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3553 {
3554         struct pf_state         *s = NULL;
3555         struct pf_src_node      *sn = NULL;
3556         struct tcphdr           *th = pd->hdr.tcp;
3557         u_int16_t                mss = V_tcp_mssdflt;
3558         u_short                  reason;
3559
3560         /* check maximums */
3561         if (r->max_states &&
3562             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3563                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3564                 REASON_SET(&reason, PFRES_MAXSTATES);
3565                 goto csfailed;
3566         }
3567         /* src node for filter rule */
3568         if ((r->rule_flag & PFRULE_SRCTRACK ||
3569             r->rpool.opts & PF_POOL_STICKYADDR) &&
3570             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3571                 REASON_SET(&reason, PFRES_SRCLIMIT);
3572                 goto csfailed;
3573         }
3574         /* src node for translation rule */
3575         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3576             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3577                 REASON_SET(&reason, PFRES_SRCLIMIT);
3578                 goto csfailed;
3579         }
3580         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3581         if (s == NULL) {
3582                 REASON_SET(&reason, PFRES_MEMORY);
3583                 goto csfailed;
3584         }
3585         s->rule.ptr = r;
3586         s->nat_rule.ptr = nr;
3587         s->anchor.ptr = a;
3588         STATE_INC_COUNTERS(s);
3589         if (r->allow_opts)
3590                 s->state_flags |= PFSTATE_ALLOWOPTS;
3591         if (r->rule_flag & PFRULE_STATESLOPPY)
3592                 s->state_flags |= PFSTATE_SLOPPY;
3593         s->log = r->log & PF_LOG_ALL;
3594         s->sync_state = PFSYNC_S_NONE;
3595         if (nr != NULL)
3596                 s->log |= nr->log & PF_LOG_ALL;
3597         switch (pd->proto) {
3598         case IPPROTO_TCP:
3599                 s->src.seqlo = ntohl(th->th_seq);
3600                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3601                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3602                     r->keep_state == PF_STATE_MODULATE) {
3603                         /* Generate sequence number modulator */
3604                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3605                             0)
3606                                 s->src.seqdiff = 1;
3607                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3608                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3609                         *rewrite = 1;
3610                 } else
3611                         s->src.seqdiff = 0;
3612                 if (th->th_flags & TH_SYN) {
3613                         s->src.seqhi++;
3614                         s->src.wscale = pf_get_wscale(m, off,
3615                             th->th_off, pd->af);
3616                 }
3617                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3618                 if (s->src.wscale & PF_WSCALE_MASK) {
3619                         /* Remove scale factor from initial window */
3620                         int win = s->src.max_win;
3621                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3622                         s->src.max_win = (win - 1) >>
3623                             (s->src.wscale & PF_WSCALE_MASK);
3624                 }
3625                 if (th->th_flags & TH_FIN)
3626                         s->src.seqhi++;
3627                 s->dst.seqhi = 1;
3628                 s->dst.max_win = 1;
3629                 s->src.state = TCPS_SYN_SENT;
3630                 s->dst.state = TCPS_CLOSED;
3631                 s->timeout = PFTM_TCP_FIRST_PACKET;
3632                 break;
3633         case IPPROTO_UDP:
3634                 s->src.state = PFUDPS_SINGLE;
3635                 s->dst.state = PFUDPS_NO_TRAFFIC;
3636                 s->timeout = PFTM_UDP_FIRST_PACKET;
3637                 break;
3638         case IPPROTO_ICMP:
3639 #ifdef INET6
3640         case IPPROTO_ICMPV6:
3641 #endif
3642                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3643                 break;
3644         default:
3645                 s->src.state = PFOTHERS_SINGLE;
3646                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3647                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3648         }
3649
3650         if (r->rt) {
3651                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3652                         REASON_SET(&reason, PFRES_MAPFAILED);
3653                         pf_src_tree_remove_state(s);
3654                         STATE_DEC_COUNTERS(s);
3655                         uma_zfree(V_pf_state_z, s);
3656                         goto csfailed;
3657                 }
3658                 s->rt_kif = r->rpool.cur->kif;
3659         }
3660
3661         s->creation = time_uptime;
3662         s->expire = time_uptime;
3663
3664         if (sn != NULL)
3665                 s->src_node = sn;
3666         if (nsn != NULL) {
3667                 /* XXX We only modify one side for now. */
3668                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3669                 s->nat_src_node = nsn;
3670         }
3671         if (pd->proto == IPPROTO_TCP) {
3672                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3673                     off, pd, th, &s->src, &s->dst)) {
3674                         REASON_SET(&reason, PFRES_MEMORY);
3675                         pf_src_tree_remove_state(s);
3676                         STATE_DEC_COUNTERS(s);
3677                         uma_zfree(V_pf_state_z, s);
3678                         return (PF_DROP);
3679                 }
3680                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3681                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3682                     &s->src, &s->dst, rewrite)) {
3683                         /* This really shouldn't happen!!! */
3684                         DPFPRINTF(PF_DEBUG_URGENT,
3685                             ("pf_normalize_tcp_stateful failed on first pkt"));
3686                         pf_normalize_tcp_cleanup(s);
3687                         pf_src_tree_remove_state(s);
3688                         STATE_DEC_COUNTERS(s);
3689                         uma_zfree(V_pf_state_z, s);
3690                         return (PF_DROP);
3691                 }
3692         }
3693         s->direction = pd->dir;
3694
3695         /*
3696          * sk/nk could already been setup by pf_get_translation().
3697          */
3698         if (nr == NULL) {
3699                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3700                     __func__, nr, sk, nk));
3701                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3702                 if (sk == NULL)
3703                         goto csfailed;
3704                 nk = sk;
3705         } else
3706                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3707                     __func__, nr, sk, nk));
3708
3709         /* Swap sk/nk for PF_OUT. */
3710         if (pf_state_insert(BOUND_IFACE(r, kif),
3711             (pd->dir == PF_IN) ? sk : nk,
3712             (pd->dir == PF_IN) ? nk : sk, s)) {
3713                 if (pd->proto == IPPROTO_TCP)
3714                         pf_normalize_tcp_cleanup(s);
3715                 REASON_SET(&reason, PFRES_STATEINS);
3716                 pf_src_tree_remove_state(s);
3717                 STATE_DEC_COUNTERS(s);
3718                 uma_zfree(V_pf_state_z, s);
3719                 return (PF_DROP);
3720         } else
3721                 *sm = s;
3722
3723         if (tag > 0)
3724                 s->tag = tag;
3725         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3726             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3727                 s->src.state = PF_TCPS_PROXY_SRC;
3728                 /* undo NAT changes, if they have taken place */
3729                 if (nr != NULL) {
3730                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3731                         if (pd->dir == PF_OUT)
3732                                 skt = s->key[PF_SK_STACK];
3733                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3734                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3735                         if (pd->sport)
3736                                 *pd->sport = skt->port[pd->sidx];
3737                         if (pd->dport)
3738                                 *pd->dport = skt->port[pd->didx];
3739                         if (pd->proto_sum)
3740                                 *pd->proto_sum = bproto_sum;
3741                         if (pd->ip_sum)
3742                                 *pd->ip_sum = bip_sum;
3743                         m_copyback(m, off, hdrlen, pd->hdr.any);
3744                 }
3745                 s->src.seqhi = htonl(arc4random());
3746                 /* Find mss option */
3747                 int rtid = M_GETFIB(m);
3748                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3749                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3750                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3751                 s->src.mss = mss;
3752                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3753                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3754                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3755                 REASON_SET(&reason, PFRES_SYNPROXY);
3756                 return (PF_SYNPROXY_DROP);
3757         }
3758
3759         return (PF_PASS);
3760
3761 csfailed:
3762         if (sk != NULL)
3763                 uma_zfree(V_pf_state_key_z, sk);
3764         if (nk != NULL)
3765                 uma_zfree(V_pf_state_key_z, nk);
3766
3767         if (sn != NULL) {
3768                 struct pf_srchash *sh;
3769
3770                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3771                 PF_HASHROW_LOCK(sh);
3772                 if (--sn->states == 0 && sn->expire == 0) {
3773                         pf_unlink_src_node(sn);
3774                         uma_zfree(V_pf_sources_z, sn);
3775                         counter_u64_add(
3776                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3777                 }
3778                 PF_HASHROW_UNLOCK(sh);
3779         }
3780
3781         if (nsn != sn && nsn != NULL) {
3782                 struct pf_srchash *sh;
3783
3784                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3785                 PF_HASHROW_LOCK(sh);
3786                 if (--nsn->states == 0 && nsn->expire == 0) {
3787                         pf_unlink_src_node(nsn);
3788                         uma_zfree(V_pf_sources_z, nsn);
3789                         counter_u64_add(
3790                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3791                 }
3792                 PF_HASHROW_UNLOCK(sh);
3793         }
3794
3795         return (PF_DROP);
3796 }
3797
3798 static int
3799 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3800     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3801     struct pf_ruleset **rsm)
3802 {
3803         struct pf_rule          *r, *a = NULL;
3804         struct pf_ruleset       *ruleset = NULL;
3805         sa_family_t              af = pd->af;
3806         u_short                  reason;
3807         int                      tag = -1;
3808         int                      asd = 0;
3809         int                      match = 0;
3810         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3811
3812         PF_RULES_RASSERT();
3813
3814         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3815         while (r != NULL) {
3816                 r->evaluations++;
3817                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3818                         r = r->skip[PF_SKIP_IFP].ptr;
3819                 else if (r->direction && r->direction != direction)
3820                         r = r->skip[PF_SKIP_DIR].ptr;
3821                 else if (r->af && r->af != af)
3822                         r = r->skip[PF_SKIP_AF].ptr;
3823                 else if (r->proto && r->proto != pd->proto)
3824                         r = r->skip[PF_SKIP_PROTO].ptr;
3825                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3826                     r->src.neg, kif, M_GETFIB(m)))
3827                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3828                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3829                     r->dst.neg, NULL, M_GETFIB(m)))
3830                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3831                 else if (r->tos && !(r->tos == pd->tos))
3832                         r = TAILQ_NEXT(r, entries);
3833                 else if (r->os_fingerprint != PF_OSFP_ANY)
3834                         r = TAILQ_NEXT(r, entries);
3835                 else if (pd->proto == IPPROTO_UDP &&
3836                     (r->src.port_op || r->dst.port_op))
3837                         r = TAILQ_NEXT(r, entries);
3838                 else if (pd->proto == IPPROTO_TCP &&
3839                     (r->src.port_op || r->dst.port_op || r->flagset))
3840                         r = TAILQ_NEXT(r, entries);
3841                 else if ((pd->proto == IPPROTO_ICMP ||
3842                     pd->proto == IPPROTO_ICMPV6) &&
3843                     (r->type || r->code))
3844                         r = TAILQ_NEXT(r, entries);
3845                 else if (r->prio &&
3846                     !pf_match_ieee8021q_pcp(r->prio, m))
3847                         r = TAILQ_NEXT(r, entries);
3848                 else if (r->prob && r->prob <=
3849                     (arc4random() % (UINT_MAX - 1) + 1))
3850                         r = TAILQ_NEXT(r, entries);
3851                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3852                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3853                         r = TAILQ_NEXT(r, entries);
3854                 else {
3855                         if (r->anchor == NULL) {
3856                                 match = 1;
3857                                 *rm = r;
3858                                 *am = a;
3859                                 *rsm = ruleset;
3860                                 if ((*rm)->quick)
3861                                         break;
3862                                 r = TAILQ_NEXT(r, entries);
3863                         } else
3864                                 pf_step_into_anchor(anchor_stack, &asd,
3865                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3866                                     &match);
3867                 }
3868                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3869                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3870                         break;
3871         }
3872         r = *rm;
3873         a = *am;
3874         ruleset = *rsm;
3875
3876         REASON_SET(&reason, PFRES_MATCH);
3877
3878         if (r->log)
3879                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3880                     1);
3881
3882         if (r->action != PF_PASS)
3883                 return (PF_DROP);
3884
3885         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3886                 REASON_SET(&reason, PFRES_MEMORY);
3887                 return (PF_DROP);
3888         }
3889
3890         return (PF_PASS);
3891 }
3892
3893 static int
3894 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3895         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3896         struct pf_pdesc *pd, u_short *reason, int *copyback)
3897 {
3898         struct tcphdr           *th = pd->hdr.tcp;
3899         u_int16_t                win = ntohs(th->th_win);
3900         u_int32_t                ack, end, seq, orig_seq;
3901         u_int8_t                 sws, dws;
3902         int                      ackskew;
3903
3904         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3905                 sws = src->wscale & PF_WSCALE_MASK;
3906                 dws = dst->wscale & PF_WSCALE_MASK;
3907         } else
3908                 sws = dws = 0;
3909
3910         /*
3911          * Sequence tracking algorithm from Guido van Rooij's paper:
3912          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3913          *      tcp_filtering.ps
3914          */
3915
3916         orig_seq = seq = ntohl(th->th_seq);
3917         if (src->seqlo == 0) {
3918                 /* First packet from this end. Set its state */
3919
3920                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3921                     src->scrub == NULL) {
3922                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3923                                 REASON_SET(reason, PFRES_MEMORY);
3924                                 return (PF_DROP);
3925                         }
3926                 }
3927
3928                 /* Deferred generation of sequence number modulator */
3929                 if (dst->seqdiff && !src->seqdiff) {
3930                         /* use random iss for the TCP server */
3931                         while ((src->seqdiff = arc4random() - seq) == 0)
3932                                 ;
3933                         ack = ntohl(th->th_ack) - dst->seqdiff;
3934                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3935                             src->seqdiff), 0);
3936                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3937                         *copyback = 1;
3938                 } else {
3939                         ack = ntohl(th->th_ack);
3940                 }
3941
3942                 end = seq + pd->p_len;
3943                 if (th->th_flags & TH_SYN) {
3944                         end++;
3945                         if (dst->wscale & PF_WSCALE_FLAG) {
3946                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3947                                     pd->af);
3948                                 if (src->wscale & PF_WSCALE_FLAG) {
3949                                         /* Remove scale factor from initial
3950                                          * window */
3951                                         sws = src->wscale & PF_WSCALE_MASK;
3952                                         win = ((u_int32_t)win + (1 << sws) - 1)
3953                                             >> sws;
3954                                         dws = dst->wscale & PF_WSCALE_MASK;
3955                                 } else {
3956                                         /* fixup other window */
3957                                         dst->max_win <<= dst->wscale &
3958                                             PF_WSCALE_MASK;
3959                                         /* in case of a retrans SYN|ACK */
3960                                         dst->wscale = 0;
3961                                 }
3962                         }
3963                 }
3964                 if (th->th_flags & TH_FIN)
3965                         end++;
3966
3967                 src->seqlo = seq;
3968                 if (src->state < TCPS_SYN_SENT)
3969                         src->state = TCPS_SYN_SENT;
3970
3971                 /*
3972                  * May need to slide the window (seqhi may have been set by
3973                  * the crappy stack check or if we picked up the connection
3974                  * after establishment)
3975                  */
3976                 if (src->seqhi == 1 ||
3977                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3978                         src->seqhi = end + MAX(1, dst->max_win << dws);
3979                 if (win > src->max_win)
3980                         src->max_win = win;
3981
3982         } else {
3983                 ack = ntohl(th->th_ack) - dst->seqdiff;
3984                 if (src->seqdiff) {
3985                         /* Modulate sequence numbers */
3986                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3987                             src->seqdiff), 0);
3988                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3989                         *copyback = 1;
3990                 }
3991                 end = seq + pd->p_len;
3992                 if (th->th_flags & TH_SYN)
3993                         end++;
3994                 if (th->th_flags & TH_FIN)
3995                         end++;
3996         }
3997
3998         if ((th->th_flags & TH_ACK) == 0) {
3999                 /* Let it pass through the ack skew check */
4000                 ack = dst->seqlo;
4001         } else if ((ack == 0 &&
4002             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4003             /* broken tcp stacks do not set ack */
4004             (dst->state < TCPS_SYN_SENT)) {
4005                 /*
4006                  * Many stacks (ours included) will set the ACK number in an
4007                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4008                  */
4009                 ack = dst->seqlo;
4010         }
4011
4012         if (seq == end) {
4013                 /* Ease sequencing restrictions on no data packets */
4014                 seq = src->seqlo;
4015                 end = seq;
4016         }
4017
4018         ackskew = dst->seqlo - ack;
4019
4020
4021         /*
4022          * Need to demodulate the sequence numbers in any TCP SACK options
4023          * (Selective ACK). We could optionally validate the SACK values
4024          * against the current ACK window, either forwards or backwards, but
4025          * I'm not confident that SACK has been implemented properly
4026          * everywhere. It wouldn't surprise me if several stacks accidentally
4027          * SACK too far backwards of previously ACKed data. There really aren't
4028          * any security implications of bad SACKing unless the target stack
4029          * doesn't validate the option length correctly. Someone trying to
4030          * spoof into a TCP connection won't bother blindly sending SACK
4031          * options anyway.
4032          */
4033         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4034                 if (pf_modulate_sack(m, off, pd, th, dst))
4035                         *copyback = 1;
4036         }
4037
4038
4039 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4040         if (SEQ_GEQ(src->seqhi, end) &&
4041             /* Last octet inside other's window space */
4042             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4043             /* Retrans: not more than one window back */
4044             (ackskew >= -MAXACKWINDOW) &&
4045             /* Acking not more than one reassembled fragment backwards */
4046             (ackskew <= (MAXACKWINDOW << sws)) &&
4047             /* Acking not more than one window forward */
4048             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4049             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4050             (pd->flags & PFDESC_IP_REAS) == 0)) {
4051             /* Require an exact/+1 sequence match on resets when possible */
4052
4053                 if (dst->scrub || src->scrub) {
4054                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4055                             *state, src, dst, copyback))
4056                                 return (PF_DROP);
4057                 }
4058
4059                 /* update max window */
4060                 if (src->max_win < win)
4061                         src->max_win = win;
4062                 /* synchronize sequencing */
4063                 if (SEQ_GT(end, src->seqlo))
4064                         src->seqlo = end;
4065                 /* slide the window of what the other end can send */
4066                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4067                         dst->seqhi = ack + MAX((win << sws), 1);
4068
4069
4070                 /* update states */
4071                 if (th->th_flags & TH_SYN)
4072                         if (src->state < TCPS_SYN_SENT)
4073                                 src->state = TCPS_SYN_SENT;
4074                 if (th->th_flags & TH_FIN)
4075                         if (src->state < TCPS_CLOSING)
4076                                 src->state = TCPS_CLOSING;
4077                 if (th->th_flags & TH_ACK) {
4078                         if (dst->state == TCPS_SYN_SENT) {
4079                                 dst->state = TCPS_ESTABLISHED;
4080                                 if (src->state == TCPS_ESTABLISHED &&
4081                                     (*state)->src_node != NULL &&
4082                                     pf_src_connlimit(state)) {
4083                                         REASON_SET(reason, PFRES_SRCLIMIT);
4084                                         return (PF_DROP);
4085                                 }
4086                         } else if (dst->state == TCPS_CLOSING)
4087                                 dst->state = TCPS_FIN_WAIT_2;
4088                 }
4089                 if (th->th_flags & TH_RST)
4090                         src->state = dst->state = TCPS_TIME_WAIT;
4091
4092                 /* update expire time */
4093                 (*state)->expire = time_uptime;
4094                 if (src->state >= TCPS_FIN_WAIT_2 &&
4095                     dst->state >= TCPS_FIN_WAIT_2)
4096                         (*state)->timeout = PFTM_TCP_CLOSED;
4097                 else if (src->state >= TCPS_CLOSING &&
4098                     dst->state >= TCPS_CLOSING)
4099                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4100                 else if (src->state < TCPS_ESTABLISHED ||
4101                     dst->state < TCPS_ESTABLISHED)
4102                         (*state)->timeout = PFTM_TCP_OPENING;
4103                 else if (src->state >= TCPS_CLOSING ||
4104                     dst->state >= TCPS_CLOSING)
4105                         (*state)->timeout = PFTM_TCP_CLOSING;
4106                 else
4107                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4108
4109                 /* Fall through to PASS packet */
4110
4111         } else if ((dst->state < TCPS_SYN_SENT ||
4112                 dst->state >= TCPS_FIN_WAIT_2 ||
4113                 src->state >= TCPS_FIN_WAIT_2) &&
4114             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4115             /* Within a window forward of the originating packet */
4116             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4117             /* Within a window backward of the originating packet */
4118
4119                 /*
4120                  * This currently handles three situations:
4121                  *  1) Stupid stacks will shotgun SYNs before their peer
4122                  *     replies.
4123                  *  2) When PF catches an already established stream (the
4124                  *     firewall rebooted, the state table was flushed, routes
4125                  *     changed...)
4126                  *  3) Packets get funky immediately after the connection
4127                  *     closes (this should catch Solaris spurious ACK|FINs
4128                  *     that web servers like to spew after a close)
4129                  *
4130                  * This must be a little more careful than the above code
4131                  * since packet floods will also be caught here. We don't
4132                  * update the TTL here to mitigate the damage of a packet
4133                  * flood and so the same code can handle awkward establishment
4134                  * and a loosened connection close.
4135                  * In the establishment case, a correct peer response will
4136                  * validate the connection, go through the normal state code
4137                  * and keep updating the state TTL.
4138                  */
4139
4140                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4141                         printf("pf: loose state match: ");
4142                         pf_print_state(*state);
4143                         pf_print_flags(th->th_flags);
4144                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4145                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4146                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4147                             (unsigned long long)(*state)->packets[1],
4148                             pd->dir == PF_IN ? "in" : "out",
4149                             pd->dir == (*state)->direction ? "fwd" : "rev");
4150                 }
4151
4152                 if (dst->scrub || src->scrub) {
4153                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4154                             *state, src, dst, copyback))
4155                                 return (PF_DROP);
4156                 }
4157
4158                 /* update max window */
4159                 if (src->max_win < win)
4160                         src->max_win = win;
4161                 /* synchronize sequencing */
4162                 if (SEQ_GT(end, src->seqlo))
4163                         src->seqlo = end;
4164                 /* slide the window of what the other end can send */
4165                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4166                         dst->seqhi = ack + MAX((win << sws), 1);
4167
4168                 /*
4169                  * Cannot set dst->seqhi here since this could be a shotgunned
4170                  * SYN and not an already established connection.
4171                  */
4172
4173                 if (th->th_flags & TH_FIN)
4174                         if (src->state < TCPS_CLOSING)
4175                                 src->state = TCPS_CLOSING;
4176                 if (th->th_flags & TH_RST)
4177                         src->state = dst->state = TCPS_TIME_WAIT;
4178
4179                 /* Fall through to PASS packet */
4180
4181         } else {
4182                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4183                     (*state)->src.state == TCPS_SYN_SENT) {
4184                         /* Send RST for state mismatches during handshake */
4185                         if (!(th->th_flags & TH_RST))
4186                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4187                                     pd->dst, pd->src, th->th_dport,
4188                                     th->th_sport, ntohl(th->th_ack), 0,
4189                                     TH_RST, 0, 0,
4190                                     (*state)->rule.ptr->return_ttl, 1, 0,
4191                                     kif->pfik_ifp);
4192                         src->seqlo = 0;
4193                         src->seqhi = 1;
4194                         src->max_win = 1;
4195                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4196                         printf("pf: BAD state: ");
4197                         pf_print_state(*state);
4198                         pf_print_flags(th->th_flags);
4199                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4200                             "pkts=%llu:%llu dir=%s,%s\n",
4201                             seq, orig_seq, ack, pd->p_len, ackskew,
4202                             (unsigned long long)(*state)->packets[0],
4203                             (unsigned long long)(*state)->packets[1],
4204                             pd->dir == PF_IN ? "in" : "out",
4205                             pd->dir == (*state)->direction ? "fwd" : "rev");
4206                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4207                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4208                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4209                             ' ': '2',
4210                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4211                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4212                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4213                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4214                 }
4215                 REASON_SET(reason, PFRES_BADSTATE);
4216                 return (PF_DROP);
4217         }
4218
4219         return (PF_PASS);
4220 }
4221
4222 static int
4223 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4224         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4225 {
4226         struct tcphdr           *th = pd->hdr.tcp;
4227
4228         if (th->th_flags & TH_SYN)
4229                 if (src->state < TCPS_SYN_SENT)
4230                         src->state = TCPS_SYN_SENT;
4231         if (th->th_flags & TH_FIN)
4232                 if (src->state < TCPS_CLOSING)
4233                         src->state = TCPS_CLOSING;
4234         if (th->th_flags & TH_ACK) {
4235                 if (dst->state == TCPS_SYN_SENT) {
4236                         dst->state = TCPS_ESTABLISHED;
4237                         if (src->state == TCPS_ESTABLISHED &&
4238                             (*state)->src_node != NULL &&
4239                             pf_src_connlimit(state)) {
4240                                 REASON_SET(reason, PFRES_SRCLIMIT);
4241                                 return (PF_DROP);
4242                         }
4243                 } else if (dst->state == TCPS_CLOSING) {
4244                         dst->state = TCPS_FIN_WAIT_2;
4245                 } else if (src->state == TCPS_SYN_SENT &&
4246                     dst->state < TCPS_SYN_SENT) {
4247                         /*
4248                          * Handle a special sloppy case where we only see one
4249                          * half of the connection. If there is a ACK after
4250                          * the initial SYN without ever seeing a packet from
4251                          * the destination, set the connection to established.
4252                          */
4253                         dst->state = src->state = TCPS_ESTABLISHED;
4254                         if ((*state)->src_node != NULL &&
4255                             pf_src_connlimit(state)) {
4256                                 REASON_SET(reason, PFRES_SRCLIMIT);
4257                                 return (PF_DROP);
4258                         }
4259                 } else if (src->state == TCPS_CLOSING &&
4260                     dst->state == TCPS_ESTABLISHED &&
4261                     dst->seqlo == 0) {
4262                         /*
4263                          * Handle the closing of half connections where we
4264                          * don't see the full bidirectional FIN/ACK+ACK
4265                          * handshake.
4266                          */
4267                         dst->state = TCPS_CLOSING;
4268                 }
4269         }
4270         if (th->th_flags & TH_RST)
4271                 src->state = dst->state = TCPS_TIME_WAIT;
4272
4273         /* update expire time */
4274         (*state)->expire = time_uptime;
4275         if (src->state >= TCPS_FIN_WAIT_2 &&
4276             dst->state >= TCPS_FIN_WAIT_2)
4277                 (*state)->timeout = PFTM_TCP_CLOSED;
4278         else if (src->state >= TCPS_CLOSING &&
4279             dst->state >= TCPS_CLOSING)
4280                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4281         else if (src->state < TCPS_ESTABLISHED ||
4282             dst->state < TCPS_ESTABLISHED)
4283                 (*state)->timeout = PFTM_TCP_OPENING;
4284         else if (src->state >= TCPS_CLOSING ||
4285             dst->state >= TCPS_CLOSING)
4286                 (*state)->timeout = PFTM_TCP_CLOSING;
4287         else
4288                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4289
4290         return (PF_PASS);
4291 }
4292
4293 static int
4294 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4295     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4296     u_short *reason)
4297 {
4298         struct pf_state_key_cmp  key;
4299         struct tcphdr           *th = pd->hdr.tcp;
4300         int                      copyback = 0;
4301         struct pf_state_peer    *src, *dst;
4302         struct pf_state_key     *sk;
4303
4304         bzero(&key, sizeof(key));
4305         key.af = pd->af;
4306         key.proto = IPPROTO_TCP;
4307         if (direction == PF_IN) {       /* wire side, straight */
4308                 PF_ACPY(&key.addr[0], pd->src, key.af);
4309                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4310                 key.port[0] = th->th_sport;
4311                 key.port[1] = th->th_dport;
4312         } else {                        /* stack side, reverse */
4313                 PF_ACPY(&key.addr[1], pd->src, key.af);
4314                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4315                 key.port[1] = th->th_sport;
4316                 key.port[0] = th->th_dport;
4317         }
4318
4319         STATE_LOOKUP(kif, &key, direction, *state, pd);
4320
4321         if (direction == (*state)->direction) {
4322                 src = &(*state)->src;
4323                 dst = &(*state)->dst;
4324         } else {
4325                 src = &(*state)->dst;
4326                 dst = &(*state)->src;
4327         }
4328
4329         sk = (*state)->key[pd->didx];
4330
4331         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4332                 if (direction != (*state)->direction) {
4333                         REASON_SET(reason, PFRES_SYNPROXY);
4334                         return (PF_SYNPROXY_DROP);
4335                 }
4336                 if (th->th_flags & TH_SYN) {
4337                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4338                                 REASON_SET(reason, PFRES_SYNPROXY);
4339                                 return (PF_DROP);
4340                         }
4341                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4342                             pd->src, th->th_dport, th->th_sport,
4343                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4344                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4345                         REASON_SET(reason, PFRES_SYNPROXY);
4346                         return (PF_SYNPROXY_DROP);
4347                 } else if (!(th->th_flags & TH_ACK) ||
4348                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4349                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4350                         REASON_SET(reason, PFRES_SYNPROXY);
4351                         return (PF_DROP);
4352                 } else if ((*state)->src_node != NULL &&
4353                     pf_src_connlimit(state)) {
4354                         REASON_SET(reason, PFRES_SRCLIMIT);
4355                         return (PF_DROP);
4356                 } else
4357                         (*state)->src.state = PF_TCPS_PROXY_DST;
4358         }
4359         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4360                 if (direction == (*state)->direction) {
4361                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4362                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4363                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4364                                 REASON_SET(reason, PFRES_SYNPROXY);
4365                                 return (PF_DROP);
4366                         }
4367                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4368                         if ((*state)->dst.seqhi == 1)
4369                                 (*state)->dst.seqhi = htonl(arc4random());
4370                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4371                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4372                             sk->port[pd->sidx], sk->port[pd->didx],
4373                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4374                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4375                         REASON_SET(reason, PFRES_SYNPROXY);
4376                         return (PF_SYNPROXY_DROP);
4377                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4378                     (TH_SYN|TH_ACK)) ||
4379                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4380                         REASON_SET(reason, PFRES_SYNPROXY);
4381                         return (PF_DROP);
4382                 } else {
4383                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4384                         (*state)->dst.seqlo = ntohl(th->th_seq);
4385                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4386                             pd->src, th->th_dport, th->th_sport,
4387                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4388                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4389                             (*state)->tag, NULL);
4390                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4391                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4392                             sk->port[pd->sidx], sk->port[pd->didx],
4393                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4394                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4395                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4396                             (*state)->src.seqlo;
4397                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4398                             (*state)->dst.seqlo;
4399                         (*state)->src.seqhi = (*state)->src.seqlo +
4400                             (*state)->dst.max_win;
4401                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4402                             (*state)->src.max_win;
4403                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4404                         (*state)->src.state = (*state)->dst.state =
4405                             TCPS_ESTABLISHED;
4406                         REASON_SET(reason, PFRES_SYNPROXY);
4407                         return (PF_SYNPROXY_DROP);
4408                 }
4409         }
4410
4411         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4412             dst->state >= TCPS_FIN_WAIT_2 &&
4413             src->state >= TCPS_FIN_WAIT_2) {
4414                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4415                         printf("pf: state reuse ");
4416                         pf_print_state(*state);
4417                         pf_print_flags(th->th_flags);
4418                         printf("\n");
4419                 }
4420                 /* XXX make sure it's the same direction ?? */
4421                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4422                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4423                 *state = NULL;
4424                 return (PF_DROP);
4425         }
4426
4427         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4428                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4429                         return (PF_DROP);
4430         } else {
4431                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4432                     &copyback) == PF_DROP)
4433                         return (PF_DROP);
4434         }
4435
4436         /* translate source/destination address, if necessary */
4437         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4438                 struct pf_state_key *nk = (*state)->key[pd->didx];
4439
4440                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4441                     nk->port[pd->sidx] != th->th_sport)
4442                         pf_change_ap(m, pd->src, &th->th_sport,
4443                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4444                             nk->port[pd->sidx], 0, pd->af);
4445
4446                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4447                     nk->port[pd->didx] != th->th_dport)
4448                         pf_change_ap(m, pd->dst, &th->th_dport,
4449                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4450                             nk->port[pd->didx], 0, pd->af);
4451                 copyback = 1;
4452         }
4453
4454         /* Copyback sequence modulation or stateful scrub changes if needed */
4455         if (copyback)
4456                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4457
4458         return (PF_PASS);
4459 }
4460
4461 static int
4462 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4463     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4464 {
4465         struct pf_state_peer    *src, *dst;
4466         struct pf_state_key_cmp  key;
4467         struct udphdr           *uh = pd->hdr.udp;
4468
4469         bzero(&key, sizeof(key));
4470         key.af = pd->af;
4471         key.proto = IPPROTO_UDP;
4472         if (direction == PF_IN) {       /* wire side, straight */
4473                 PF_ACPY(&key.addr[0], pd->src, key.af);
4474                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4475                 key.port[0] = uh->uh_sport;
4476                 key.port[1] = uh->uh_dport;
4477         } else {                        /* stack side, reverse */
4478                 PF_ACPY(&key.addr[1], pd->src, key.af);
4479                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4480                 key.port[1] = uh->uh_sport;
4481                 key.port[0] = uh->uh_dport;
4482         }
4483
4484         STATE_LOOKUP(kif, &key, direction, *state, pd);
4485
4486         if (direction == (*state)->direction) {
4487                 src = &(*state)->src;
4488                 dst = &(*state)->dst;
4489         } else {
4490                 src = &(*state)->dst;
4491                 dst = &(*state)->src;
4492         }
4493
4494         /* update states */
4495         if (src->state < PFUDPS_SINGLE)
4496                 src->state = PFUDPS_SINGLE;
4497         if (dst->state == PFUDPS_SINGLE)
4498                 dst->state = PFUDPS_MULTIPLE;
4499
4500         /* update expire time */
4501         (*state)->expire = time_uptime;
4502         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4503                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4504         else
4505                 (*state)->timeout = PFTM_UDP_SINGLE;
4506
4507         /* translate source/destination address, if necessary */
4508         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4509                 struct pf_state_key *nk = (*state)->key[pd->didx];
4510
4511                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4512                     nk->port[pd->sidx] != uh->uh_sport)
4513                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4514                             &uh->uh_sum, &nk->addr[pd->sidx],
4515                             nk->port[pd->sidx], 1, pd->af);
4516
4517                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4518                     nk->port[pd->didx] != uh->uh_dport)
4519                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4520                             &uh->uh_sum, &nk->addr[pd->didx],
4521                             nk->port[pd->didx], 1, pd->af);
4522                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4523         }
4524
4525         return (PF_PASS);
4526 }
4527
4528 static int
4529 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4530     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4531 {
4532         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4533         u_int16_t        icmpid = 0, *icmpsum;
4534         u_int8_t         icmptype;
4535         int              state_icmp = 0;
4536         struct pf_state_key_cmp key;
4537
4538         bzero(&key, sizeof(key));
4539         switch (pd->proto) {
4540 #ifdef INET
4541         case IPPROTO_ICMP:
4542                 icmptype = pd->hdr.icmp->icmp_type;
4543                 icmpid = pd->hdr.icmp->icmp_id;
4544                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4545
4546                 if (icmptype == ICMP_UNREACH ||
4547                     icmptype == ICMP_SOURCEQUENCH ||
4548                     icmptype == ICMP_REDIRECT ||
4549                     icmptype == ICMP_TIMXCEED ||
4550                     icmptype == ICMP_PARAMPROB)
4551                         state_icmp++;
4552                 break;
4553 #endif /* INET */
4554 #ifdef INET6
4555         case IPPROTO_ICMPV6:
4556                 icmptype = pd->hdr.icmp6->icmp6_type;
4557                 icmpid = pd->hdr.icmp6->icmp6_id;
4558                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4559
4560                 if (icmptype == ICMP6_DST_UNREACH ||
4561                     icmptype == ICMP6_PACKET_TOO_BIG ||
4562                     icmptype == ICMP6_TIME_EXCEEDED ||
4563                     icmptype == ICMP6_PARAM_PROB)
4564                         state_icmp++;
4565                 break;
4566 #endif /* INET6 */
4567         }
4568
4569         if (!state_icmp) {
4570
4571                 /*
4572                  * ICMP query/reply message not related to a TCP/UDP packet.
4573                  * Search for an ICMP state.
4574                  */
4575                 key.af = pd->af;
4576                 key.proto = pd->proto;
4577                 key.port[0] = key.port[1] = icmpid;
4578                 if (direction == PF_IN) {       /* wire side, straight */
4579                         PF_ACPY(&key.addr[0], pd->src, key.af);
4580                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4581                 } else {                        /* stack side, reverse */
4582                         PF_ACPY(&key.addr[1], pd->src, key.af);
4583                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4584                 }
4585
4586                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4587
4588                 (*state)->expire = time_uptime;
4589                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4590
4591                 /* translate source/destination address, if necessary */
4592                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4593                         struct pf_state_key *nk = (*state)->key[pd->didx];
4594
4595                         switch (pd->af) {
4596 #ifdef INET
4597                         case AF_INET:
4598                                 if (PF_ANEQ(pd->src,
4599                                     &nk->addr[pd->sidx], AF_INET))
4600                                         pf_change_a(&saddr->v4.s_addr,
4601                                             pd->ip_sum,
4602                                             nk->addr[pd->sidx].v4.s_addr, 0);
4603
4604                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4605                                     AF_INET))
4606                                         pf_change_a(&daddr->v4.s_addr,
4607                                             pd->ip_sum,
4608                                             nk->addr[pd->didx].v4.s_addr, 0);
4609
4610                                 if (nk->port[0] !=
4611                                     pd->hdr.icmp->icmp_id) {
4612                                         pd->hdr.icmp->icmp_cksum =
4613                                             pf_cksum_fixup(
4614                                             pd->hdr.icmp->icmp_cksum, icmpid,
4615                                             nk->port[pd->sidx], 0);
4616                                         pd->hdr.icmp->icmp_id =
4617                                             nk->port[pd->sidx];
4618                                 }
4619
4620                                 m_copyback(m, off, ICMP_MINLEN,
4621                                     (caddr_t )pd->hdr.icmp);
4622                                 break;
4623 #endif /* INET */
4624 #ifdef INET6
4625                         case AF_INET6:
4626                                 if (PF_ANEQ(pd->src,
4627                                     &nk->addr[pd->sidx], AF_INET6))
4628                                         pf_change_a6(saddr,
4629                                             &pd->hdr.icmp6->icmp6_cksum,
4630                                             &nk->addr[pd->sidx], 0);
4631
4632                                 if (PF_ANEQ(pd->dst,
4633                                     &nk->addr[pd->didx], AF_INET6))
4634                                         pf_change_a6(daddr,
4635                                             &pd->hdr.icmp6->icmp6_cksum,
4636                                             &nk->addr[pd->didx], 0);
4637
4638                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4639                                     (caddr_t )pd->hdr.icmp6);
4640                                 break;
4641 #endif /* INET6 */
4642                         }
4643                 }
4644                 return (PF_PASS);
4645
4646         } else {
4647                 /*
4648                  * ICMP error message in response to a TCP/UDP packet.
4649                  * Extract the inner TCP/UDP header and search for that state.
4650                  */
4651
4652                 struct pf_pdesc pd2;
4653                 bzero(&pd2, sizeof pd2);
4654 #ifdef INET
4655                 struct ip       h2;
4656 #endif /* INET */
4657 #ifdef INET6
4658                 struct ip6_hdr  h2_6;
4659                 int             terminal = 0;
4660 #endif /* INET6 */
4661                 int             ipoff2 = 0;
4662                 int             off2 = 0;
4663
4664                 pd2.af = pd->af;
4665                 /* Payload packet is from the opposite direction. */
4666                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4667                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4668                 switch (pd->af) {
4669 #ifdef INET
4670                 case AF_INET:
4671                         /* offset of h2 in mbuf chain */
4672                         ipoff2 = off + ICMP_MINLEN;
4673
4674                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4675                             NULL, reason, pd2.af)) {
4676                                 DPFPRINTF(PF_DEBUG_MISC,
4677                                     ("pf: ICMP error message too short "
4678                                     "(ip)\n"));
4679                                 return (PF_DROP);
4680                         }
4681                         /*
4682                          * ICMP error messages don't refer to non-first
4683                          * fragments
4684                          */
4685                         if (h2.ip_off & htons(IP_OFFMASK)) {
4686                                 REASON_SET(reason, PFRES_FRAG);
4687                                 return (PF_DROP);
4688                         }
4689
4690                         /* offset of protocol header that follows h2 */
4691                         off2 = ipoff2 + (h2.ip_hl << 2);
4692
4693                         pd2.proto = h2.ip_p;
4694                         pd2.src = (struct pf_addr *)&h2.ip_src;
4695                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4696                         pd2.ip_sum = &h2.ip_sum;
4697                         break;
4698 #endif /* INET */
4699 #ifdef INET6
4700                 case AF_INET6:
4701                         ipoff2 = off + sizeof(struct icmp6_hdr);
4702
4703                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4704                             NULL, reason, pd2.af)) {
4705                                 DPFPRINTF(PF_DEBUG_MISC,
4706                                     ("pf: ICMP error message too short "
4707                                     "(ip6)\n"));
4708                                 return (PF_DROP);
4709                         }
4710                         pd2.proto = h2_6.ip6_nxt;
4711                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4712                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4713                         pd2.ip_sum = NULL;
4714                         off2 = ipoff2 + sizeof(h2_6);
4715                         do {
4716                                 switch (pd2.proto) {
4717                                 case IPPROTO_FRAGMENT:
4718                                         /*
4719                                          * ICMPv6 error messages for
4720                                          * non-first fragments
4721                                          */
4722                                         REASON_SET(reason, PFRES_FRAG);
4723                                         return (PF_DROP);
4724                                 case IPPROTO_AH:
4725                                 case IPPROTO_HOPOPTS:
4726                                 case IPPROTO_ROUTING:
4727                                 case IPPROTO_DSTOPTS: {
4728                                         /* get next header and header length */
4729                                         struct ip6_ext opt6;
4730
4731                                         if (!pf_pull_hdr(m, off2, &opt6,
4732                                             sizeof(opt6), NULL, reason,
4733                                             pd2.af)) {
4734                                                 DPFPRINTF(PF_DEBUG_MISC,
4735                                                     ("pf: ICMPv6 short opt\n"));
4736                                                 return (PF_DROP);
4737                                         }
4738                                         if (pd2.proto == IPPROTO_AH)
4739                                                 off2 += (opt6.ip6e_len + 2) * 4;
4740                                         else
4741                                                 off2 += (opt6.ip6e_len + 1) * 8;
4742                                         pd2.proto = opt6.ip6e_nxt;
4743                                         /* goto the next header */
4744                                         break;
4745                                 }
4746                                 default:
4747                                         terminal++;
4748                                         break;
4749                                 }
4750                         } while (!terminal);
4751                         break;
4752 #endif /* INET6 */
4753                 }
4754
4755                 switch (pd2.proto) {
4756                 case IPPROTO_TCP: {
4757                         struct tcphdr            th;
4758                         u_int32_t                seq;
4759                         struct pf_state_peer    *src, *dst;
4760                         u_int8_t                 dws;
4761                         int                      copyback = 0;
4762
4763                         /*
4764                          * Only the first 8 bytes of the TCP header can be
4765                          * expected. Don't access any TCP header fields after
4766                          * th_seq, an ackskew test is not possible.
4767                          */
4768                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4769                             pd2.af)) {
4770                                 DPFPRINTF(PF_DEBUG_MISC,
4771                                     ("pf: ICMP error message too short "
4772                                     "(tcp)\n"));
4773                                 return (PF_DROP);
4774                         }
4775
4776                         key.af = pd2.af;
4777                         key.proto = IPPROTO_TCP;
4778                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4779                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4780                         key.port[pd2.sidx] = th.th_sport;
4781                         key.port[pd2.didx] = th.th_dport;
4782
4783                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4784
4785                         if (direction == (*state)->direction) {
4786                                 src = &(*state)->dst;
4787                                 dst = &(*state)->src;
4788                         } else {
4789                                 src = &(*state)->src;
4790                                 dst = &(*state)->dst;
4791                         }
4792
4793                         if (src->wscale && dst->wscale)
4794                                 dws = dst->wscale & PF_WSCALE_MASK;
4795                         else
4796                                 dws = 0;
4797
4798                         /* Demodulate sequence number */
4799                         seq = ntohl(th.th_seq) - src->seqdiff;
4800                         if (src->seqdiff) {
4801                                 pf_change_a(&th.th_seq, icmpsum,
4802                                     htonl(seq), 0);
4803                                 copyback = 1;
4804                         }
4805
4806                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4807                             (!SEQ_GEQ(src->seqhi, seq) ||
4808                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4809                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4810                                         printf("pf: BAD ICMP %d:%d ",
4811                                             icmptype, pd->hdr.icmp->icmp_code);
4812                                         pf_print_host(pd->src, 0, pd->af);
4813                                         printf(" -> ");
4814                                         pf_print_host(pd->dst, 0, pd->af);
4815                                         printf(" state: ");
4816                                         pf_print_state(*state);
4817                                         printf(" seq=%u\n", seq);
4818                                 }
4819                                 REASON_SET(reason, PFRES_BADSTATE);
4820                                 return (PF_DROP);
4821                         } else {
4822                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4823                                         printf("pf: OK ICMP %d:%d ",
4824                                             icmptype, pd->hdr.icmp->icmp_code);
4825                                         pf_print_host(pd->src, 0, pd->af);
4826                                         printf(" -> ");
4827                                         pf_print_host(pd->dst, 0, pd->af);
4828                                         printf(" state: ");
4829                                         pf_print_state(*state);
4830                                         printf(" seq=%u\n", seq);
4831                                 }
4832                         }
4833
4834                         /* translate source/destination address, if necessary */
4835                         if ((*state)->key[PF_SK_WIRE] !=
4836                             (*state)->key[PF_SK_STACK]) {
4837                                 struct pf_state_key *nk =
4838                                     (*state)->key[pd->didx];
4839
4840                                 if (PF_ANEQ(pd2.src,
4841                                     &nk->addr[pd2.sidx], pd2.af) ||
4842                                     nk->port[pd2.sidx] != th.th_sport)
4843                                         pf_change_icmp(pd2.src, &th.th_sport,
4844                                             daddr, &nk->addr[pd2.sidx],
4845                                             nk->port[pd2.sidx], NULL,
4846                                             pd2.ip_sum, icmpsum,
4847                                             pd->ip_sum, 0, pd2.af);
4848
4849                                 if (PF_ANEQ(pd2.dst,
4850                                     &nk->addr[pd2.didx], pd2.af) ||
4851                                     nk->port[pd2.didx] != th.th_dport)
4852                                         pf_change_icmp(pd2.dst, &th.th_dport,
4853                                             saddr, &nk->addr[pd2.didx],
4854                                             nk->port[pd2.didx], NULL,
4855                                             pd2.ip_sum, icmpsum,
4856                                             pd->ip_sum, 0, pd2.af);
4857                                 copyback = 1;
4858                         }
4859
4860                         if (copyback) {
4861                                 switch (pd2.af) {
4862 #ifdef INET
4863                                 case AF_INET:
4864                                         m_copyback(m, off, ICMP_MINLEN,
4865                                             (caddr_t )pd->hdr.icmp);
4866                                         m_copyback(m, ipoff2, sizeof(h2),
4867                                             (caddr_t )&h2);
4868                                         break;
4869 #endif /* INET */
4870 #ifdef INET6
4871                                 case AF_INET6:
4872                                         m_copyback(m, off,
4873                                             sizeof(struct icmp6_hdr),
4874                                             (caddr_t )pd->hdr.icmp6);
4875                                         m_copyback(m, ipoff2, sizeof(h2_6),
4876                                             (caddr_t )&h2_6);
4877                                         break;
4878 #endif /* INET6 */
4879                                 }
4880                                 m_copyback(m, off2, 8, (caddr_t)&th);
4881                         }
4882
4883                         return (PF_PASS);
4884                         break;
4885                 }
4886                 case IPPROTO_UDP: {
4887                         struct udphdr           uh;
4888
4889                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4890                             NULL, reason, pd2.af)) {
4891                                 DPFPRINTF(PF_DEBUG_MISC,
4892                                     ("pf: ICMP error message too short "
4893                                     "(udp)\n"));
4894                                 return (PF_DROP);
4895                         }
4896
4897                         key.af = pd2.af;
4898                         key.proto = IPPROTO_UDP;
4899                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4900                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4901                         key.port[pd2.sidx] = uh.uh_sport;
4902                         key.port[pd2.didx] = uh.uh_dport;
4903
4904                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4905
4906                         /* translate source/destination address, if necessary */
4907                         if ((*state)->key[PF_SK_WIRE] !=
4908                             (*state)->key[PF_SK_STACK]) {
4909                                 struct pf_state_key *nk =
4910                                     (*state)->key[pd->didx];
4911
4912                                 if (PF_ANEQ(pd2.src,
4913                                     &nk->addr[pd2.sidx], pd2.af) ||
4914                                     nk->port[pd2.sidx] != uh.uh_sport)
4915                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4916                                             daddr, &nk->addr[pd2.sidx],
4917                                             nk->port[pd2.sidx], &uh.uh_sum,
4918                                             pd2.ip_sum, icmpsum,
4919                                             pd->ip_sum, 1, pd2.af);
4920
4921                                 if (PF_ANEQ(pd2.dst,
4922                                     &nk->addr[pd2.didx], pd2.af) ||
4923                                     nk->port[pd2.didx] != uh.uh_dport)
4924                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4925                                             saddr, &nk->addr[pd2.didx],
4926                                             nk->port[pd2.didx], &uh.uh_sum,
4927                                             pd2.ip_sum, icmpsum,
4928                                             pd->ip_sum, 1, pd2.af);
4929
4930                                 switch (pd2.af) {
4931 #ifdef INET
4932                                 case AF_INET:
4933                                         m_copyback(m, off, ICMP_MINLEN,
4934                                             (caddr_t )pd->hdr.icmp);
4935                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4936                                         break;
4937 #endif /* INET */
4938 #ifdef INET6
4939                                 case AF_INET6:
4940                                         m_copyback(m, off,
4941                                             sizeof(struct icmp6_hdr),
4942                                             (caddr_t )pd->hdr.icmp6);
4943                                         m_copyback(m, ipoff2, sizeof(h2_6),
4944                                             (caddr_t )&h2_6);
4945                                         break;
4946 #endif /* INET6 */
4947                                 }
4948                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4949                         }
4950                         return (PF_PASS);
4951                         break;
4952                 }
4953 #ifdef INET
4954                 case IPPROTO_ICMP: {
4955                         struct icmp             iih;
4956
4957                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4958                             NULL, reason, pd2.af)) {
4959                                 DPFPRINTF(PF_DEBUG_MISC,
4960                                     ("pf: ICMP error message too short i"
4961                                     "(icmp)\n"));
4962                                 return (PF_DROP);
4963                         }
4964
4965                         key.af = pd2.af;
4966                         key.proto = IPPROTO_ICMP;
4967                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4968                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4969                         key.port[0] = key.port[1] = iih.icmp_id;
4970
4971                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4972
4973                         /* translate source/destination address, if necessary */
4974                         if ((*state)->key[PF_SK_WIRE] !=
4975                             (*state)->key[PF_SK_STACK]) {
4976                                 struct pf_state_key *nk =
4977                                     (*state)->key[pd->didx];
4978
4979                                 if (PF_ANEQ(pd2.src,
4980                                     &nk->addr[pd2.sidx], pd2.af) ||
4981                                     nk->port[pd2.sidx] != iih.icmp_id)
4982                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4983                                             daddr, &nk->addr[pd2.sidx],
4984                                             nk->port[pd2.sidx], NULL,
4985                                             pd2.ip_sum, icmpsum,
4986                                             pd->ip_sum, 0, AF_INET);
4987
4988                                 if (PF_ANEQ(pd2.dst,
4989                                     &nk->addr[pd2.didx], pd2.af) ||
4990                                     nk->port[pd2.didx] != iih.icmp_id)
4991                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4992                                             saddr, &nk->addr[pd2.didx],
4993                                             nk->port[pd2.didx], NULL,
4994                                             pd2.ip_sum, icmpsum,
4995                                             pd->ip_sum, 0, AF_INET);
4996
4997                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4998                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4999                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5000                         }
5001                         return (PF_PASS);
5002                         break;
5003                 }
5004 #endif /* INET */
5005 #ifdef INET6
5006                 case IPPROTO_ICMPV6: {
5007                         struct icmp6_hdr        iih;
5008
5009                         if (!pf_pull_hdr(m, off2, &iih,
5010                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5011                                 DPFPRINTF(PF_DEBUG_MISC,
5012                                     ("pf: ICMP error message too short "
5013                                     "(icmp6)\n"));
5014                                 return (PF_DROP);
5015                         }
5016
5017                         key.af = pd2.af;
5018                         key.proto = IPPROTO_ICMPV6;
5019                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5020                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5021                         key.port[0] = key.port[1] = iih.icmp6_id;
5022
5023                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5024
5025                         /* translate source/destination address, if necessary */
5026                         if ((*state)->key[PF_SK_WIRE] !=
5027                             (*state)->key[PF_SK_STACK]) {
5028                                 struct pf_state_key *nk =
5029                                     (*state)->key[pd->didx];
5030
5031                                 if (PF_ANEQ(pd2.src,
5032                                     &nk->addr[pd2.sidx], pd2.af) ||
5033                                     nk->port[pd2.sidx] != iih.icmp6_id)
5034                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5035                                             daddr, &nk->addr[pd2.sidx],
5036                                             nk->port[pd2.sidx], NULL,
5037                                             pd2.ip_sum, icmpsum,
5038                                             pd->ip_sum, 0, AF_INET6);
5039
5040                                 if (PF_ANEQ(pd2.dst,
5041                                     &nk->addr[pd2.didx], pd2.af) ||
5042                                     nk->port[pd2.didx] != iih.icmp6_id)
5043                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5044                                             saddr, &nk->addr[pd2.didx],
5045                                             nk->port[pd2.didx], NULL,
5046                                             pd2.ip_sum, icmpsum,
5047                                             pd->ip_sum, 0, AF_INET6);
5048
5049                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5050                                     (caddr_t)pd->hdr.icmp6);
5051                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5052                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5053                                     (caddr_t)&iih);
5054                         }
5055                         return (PF_PASS);
5056                         break;
5057                 }
5058 #endif /* INET6 */
5059                 default: {
5060                         key.af = pd2.af;
5061                         key.proto = pd2.proto;
5062                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5063                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5064                         key.port[0] = key.port[1] = 0;
5065
5066                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5067
5068                         /* translate source/destination address, if necessary */
5069                         if ((*state)->key[PF_SK_WIRE] !=
5070                             (*state)->key[PF_SK_STACK]) {
5071                                 struct pf_state_key *nk =
5072                                     (*state)->key[pd->didx];
5073
5074                                 if (PF_ANEQ(pd2.src,
5075                                     &nk->addr[pd2.sidx], pd2.af))
5076                                         pf_change_icmp(pd2.src, NULL, daddr,
5077                                             &nk->addr[pd2.sidx], 0, NULL,
5078                                             pd2.ip_sum, icmpsum,
5079                                             pd->ip_sum, 0, pd2.af);
5080
5081                                 if (PF_ANEQ(pd2.dst,
5082                                     &nk->addr[pd2.didx], pd2.af))
5083                                         pf_change_icmp(pd2.dst, NULL, saddr,
5084                                             &nk->addr[pd2.didx], 0, NULL,
5085                                             pd2.ip_sum, icmpsum,
5086                                             pd->ip_sum, 0, pd2.af);
5087
5088                                 switch (pd2.af) {
5089 #ifdef INET
5090                                 case AF_INET:
5091                                         m_copyback(m, off, ICMP_MINLEN,
5092                                             (caddr_t)pd->hdr.icmp);
5093                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5094                                         break;
5095 #endif /* INET */
5096 #ifdef INET6
5097                                 case AF_INET6:
5098                                         m_copyback(m, off,
5099                                             sizeof(struct icmp6_hdr),
5100                                             (caddr_t )pd->hdr.icmp6);
5101                                         m_copyback(m, ipoff2, sizeof(h2_6),
5102                                             (caddr_t )&h2_6);
5103                                         break;
5104 #endif /* INET6 */
5105                                 }
5106                         }
5107                         return (PF_PASS);
5108                         break;
5109                 }
5110                 }
5111         }
5112 }
5113
5114 static int
5115 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5116     struct mbuf *m, struct pf_pdesc *pd)
5117 {
5118         struct pf_state_peer    *src, *dst;
5119         struct pf_state_key_cmp  key;
5120
5121         bzero(&key, sizeof(key));
5122         key.af = pd->af;
5123         key.proto = pd->proto;
5124         if (direction == PF_IN) {
5125                 PF_ACPY(&key.addr[0], pd->src, key.af);
5126                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5127                 key.port[0] = key.port[1] = 0;
5128         } else {
5129                 PF_ACPY(&key.addr[1], pd->src, key.af);
5130                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5131                 key.port[1] = key.port[0] = 0;
5132         }
5133
5134         STATE_LOOKUP(kif, &key, direction, *state, pd);
5135
5136         if (direction == (*state)->direction) {
5137                 src = &(*state)->src;
5138                 dst = &(*state)->dst;
5139         } else {
5140                 src = &(*state)->dst;
5141                 dst = &(*state)->src;
5142         }
5143
5144         /* update states */
5145         if (src->state < PFOTHERS_SINGLE)
5146                 src->state = PFOTHERS_SINGLE;
5147         if (dst->state == PFOTHERS_SINGLE)
5148                 dst->state = PFOTHERS_MULTIPLE;
5149
5150         /* update expire time */
5151         (*state)->expire = time_uptime;
5152         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5153                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5154         else
5155                 (*state)->timeout = PFTM_OTHER_SINGLE;
5156
5157         /* translate source/destination address, if necessary */
5158         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5159                 struct pf_state_key *nk = (*state)->key[pd->didx];
5160
5161                 KASSERT(nk, ("%s: nk is null", __func__));
5162                 KASSERT(pd, ("%s: pd is null", __func__));
5163                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5164                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5165                 switch (pd->af) {
5166 #ifdef INET
5167                 case AF_INET:
5168                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5169                                 pf_change_a(&pd->src->v4.s_addr,
5170                                     pd->ip_sum,
5171                                     nk->addr[pd->sidx].v4.s_addr,
5172                                     0);
5173
5174
5175                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5176                                 pf_change_a(&pd->dst->v4.s_addr,
5177                                     pd->ip_sum,
5178                                     nk->addr[pd->didx].v4.s_addr,
5179                                     0);
5180
5181                                 break;
5182 #endif /* INET */
5183 #ifdef INET6
5184                 case AF_INET6:
5185                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5186                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5187
5188                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5189                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5190 #endif /* INET6 */
5191                 }
5192         }
5193         return (PF_PASS);
5194 }
5195
5196 /*
5197  * ipoff and off are measured from the start of the mbuf chain.
5198  * h must be at "ipoff" on the mbuf chain.
5199  */
5200 void *
5201 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5202     u_short *actionp, u_short *reasonp, sa_family_t af)
5203 {
5204         switch (af) {
5205 #ifdef INET
5206         case AF_INET: {
5207                 struct ip       *h = mtod(m, struct ip *);
5208                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5209
5210                 if (fragoff) {
5211                         if (fragoff >= len)
5212                                 ACTION_SET(actionp, PF_PASS);
5213                         else {
5214                                 ACTION_SET(actionp, PF_DROP);
5215                                 REASON_SET(reasonp, PFRES_FRAG);
5216                         }
5217                         return (NULL);
5218                 }
5219                 if (m->m_pkthdr.len < off + len ||
5220                     ntohs(h->ip_len) < off + len) {
5221                         ACTION_SET(actionp, PF_DROP);
5222                         REASON_SET(reasonp, PFRES_SHORT);
5223                         return (NULL);
5224                 }
5225                 break;
5226         }
5227 #endif /* INET */
5228 #ifdef INET6
5229         case AF_INET6: {
5230                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5231
5232                 if (m->m_pkthdr.len < off + len ||
5233                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5234                     (unsigned)(off + len)) {
5235                         ACTION_SET(actionp, PF_DROP);
5236                         REASON_SET(reasonp, PFRES_SHORT);
5237                         return (NULL);
5238                 }
5239                 break;
5240         }
5241 #endif /* INET6 */
5242         }
5243         m_copydata(m, off, len, p);
5244         return (p);
5245 }
5246
5247 #ifdef RADIX_MPATH
5248 static int
5249 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5250     int rtableid)
5251 {
5252         struct radix_node_head  *rnh;
5253         struct sockaddr_in      *dst;
5254         int                      ret = 1;
5255         int                      check_mpath;
5256 #ifdef INET6
5257         struct sockaddr_in6     *dst6;
5258         struct route_in6         ro;
5259 #else
5260         struct route             ro;
5261 #endif
5262         struct radix_node       *rn;
5263         struct rtentry          *rt;
5264         struct ifnet            *ifp;
5265
5266         check_mpath = 0;
5267         /* XXX: stick to table 0 for now */
5268         rnh = rt_tables_get_rnh(0, af);
5269         if (rnh != NULL && rn_mpath_capable(rnh))
5270                 check_mpath = 1;
5271         bzero(&ro, sizeof(ro));
5272         switch (af) {
5273         case AF_INET:
5274                 dst = satosin(&ro.ro_dst);
5275                 dst->sin_family = AF_INET;
5276                 dst->sin_len = sizeof(*dst);
5277                 dst->sin_addr = addr->v4;
5278                 break;
5279 #ifdef INET6
5280         case AF_INET6:
5281                 /*
5282                  * Skip check for addresses with embedded interface scope,
5283                  * as they would always match anyway.
5284                  */
5285                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5286                         goto out;
5287                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5288                 dst6->sin6_family = AF_INET6;
5289                 dst6->sin6_len = sizeof(*dst6);
5290                 dst6->sin6_addr = addr->v6;
5291                 break;
5292 #endif /* INET6 */
5293         default:
5294                 return (0);
5295         }
5296
5297         /* Skip checks for ipsec interfaces */
5298         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5299                 goto out;
5300
5301         switch (af) {
5302 #ifdef INET6
5303         case AF_INET6:
5304                 in6_rtalloc_ign(&ro, 0, rtableid);
5305                 break;
5306 #endif
5307 #ifdef INET
5308         case AF_INET:
5309                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5310                 break;
5311 #endif
5312         }
5313
5314         if (ro.ro_rt != NULL) {
5315                 /* No interface given, this is a no-route check */
5316                 if (kif == NULL)
5317                         goto out;
5318
5319                 if (kif->pfik_ifp == NULL) {
5320                         ret = 0;
5321                         goto out;
5322                 }
5323
5324                 /* Perform uRPF check if passed input interface */
5325                 ret = 0;
5326                 rn = (struct radix_node *)ro.ro_rt;
5327                 do {
5328                         rt = (struct rtentry *)rn;
5329                         ifp = rt->rt_ifp;
5330
5331                         if (kif->pfik_ifp == ifp)
5332                                 ret = 1;
5333                         rn = rn_mpath_next(rn);
5334                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5335         } else
5336                 ret = 0;
5337 out:
5338         if (ro.ro_rt != NULL)
5339                 RTFREE(ro.ro_rt);
5340         return (ret);
5341 }
5342 #endif
5343
5344 int
5345 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5346     int rtableid)
5347 {
5348 #ifdef INET
5349         struct nhop4_basic      nh4;
5350 #endif
5351 #ifdef INET6
5352         struct nhop6_basic      nh6;
5353 #endif
5354         struct ifnet            *ifp;
5355 #ifdef RADIX_MPATH
5356         struct radix_node_head  *rnh;
5357
5358         /* XXX: stick to table 0 for now */
5359         rnh = rt_tables_get_rnh(0, af);
5360         if (rnh != NULL && rn_mpath_capable(rnh))
5361                 return (pf_routable_oldmpath(addr, af, kif, rtableid));
5362 #endif
5363         /*
5364          * Skip check for addresses with embedded interface scope,
5365          * as they would always match anyway.
5366          */
5367         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5368                 return (1);
5369
5370         if (af != AF_INET && af != AF_INET6)
5371                 return (0);
5372
5373         /* Skip checks for ipsec interfaces */
5374         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5375                 return (1);
5376
5377         ifp = NULL;
5378
5379         switch (af) {
5380 #ifdef INET6
5381         case AF_INET6:
5382                 if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5383                         return (0);
5384                 ifp = nh6.nh_ifp;
5385                 break;
5386 #endif
5387 #ifdef INET
5388         case AF_INET:
5389                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5390                         return (0);
5391                 ifp = nh4.nh_ifp;
5392                 break;
5393 #endif
5394         }
5395
5396         /* No interface given, this is a no-route check */
5397         if (kif == NULL)
5398                 return (1);
5399
5400         if (kif->pfik_ifp == NULL)
5401                 return (0);
5402
5403         /* Perform uRPF check if passed input interface */
5404         if (kif->pfik_ifp == ifp)
5405                 return (1);
5406         return (0);
5407 }
5408
5409 #ifdef INET
5410 static void
5411 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5412     struct pf_state *s, struct pf_pdesc *pd)
5413 {
5414         struct mbuf             *m0, *m1;
5415         struct sockaddr_in      dst;
5416         struct ip               *ip;
5417         struct ifnet            *ifp = NULL;
5418         struct pf_addr           naddr;
5419         struct pf_src_node      *sn = NULL;
5420         int                      error = 0;
5421         uint16_t                 ip_len, ip_off;
5422
5423         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5424         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5425             __func__));
5426
5427         if ((pd->pf_mtag == NULL &&
5428             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5429             pd->pf_mtag->routed++ > 3) {
5430                 m0 = *m;
5431                 *m = NULL;
5432                 goto bad_locked;
5433         }
5434
5435         if (r->rt == PF_DUPTO) {
5436                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5437                         if (s)
5438                                 PF_STATE_UNLOCK(s);
5439                         return;
5440                 }
5441         } else {
5442                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5443                         if (s)
5444                                 PF_STATE_UNLOCK(s);
5445                         return;
5446                 }
5447                 m0 = *m;
5448         }
5449
5450         ip = mtod(m0, struct ip *);
5451
5452         bzero(&dst, sizeof(dst));
5453         dst.sin_family = AF_INET;
5454         dst.sin_len = sizeof(dst);
5455         dst.sin_addr = ip->ip_dst;
5456
5457         if (TAILQ_EMPTY(&r->rpool.list)) {
5458                 DPFPRINTF(PF_DEBUG_URGENT,
5459                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5460                 goto bad_locked;
5461         }
5462         if (s == NULL) {
5463                 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5464                     &naddr, NULL, &sn);
5465                 if (!PF_AZERO(&naddr, AF_INET))
5466                         dst.sin_addr.s_addr = naddr.v4.s_addr;
5467                 ifp = r->rpool.cur->kif ?
5468                     r->rpool.cur->kif->pfik_ifp : NULL;
5469         } else {
5470                 if (!PF_AZERO(&s->rt_addr, AF_INET))
5471                         dst.sin_addr.s_addr =
5472                             s->rt_addr.v4.s_addr;
5473                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5474                 PF_STATE_UNLOCK(s);
5475         }
5476         if (ifp == NULL)
5477                 goto bad;
5478
5479         if (oifp != ifp) {
5480                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5481                         goto bad;
5482                 else if (m0 == NULL)
5483                         goto done;
5484                 if (m0->m_len < sizeof(struct ip)) {
5485                         DPFPRINTF(PF_DEBUG_URGENT,
5486                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5487                         goto bad;
5488                 }
5489                 ip = mtod(m0, struct ip *);
5490         }
5491
5492         if (ifp->if_flags & IFF_LOOPBACK)
5493                 m0->m_flags |= M_SKIP_FIREWALL;
5494
5495         ip_len = ntohs(ip->ip_len);
5496         ip_off = ntohs(ip->ip_off);
5497
5498         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5499         m0->m_pkthdr.csum_flags |= CSUM_IP;
5500         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5501                 in_delayed_cksum(m0);
5502                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5503         }
5504 #ifdef SCTP
5505         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5506                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5507                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5508         }
5509 #endif
5510
5511         /*
5512          * If small enough for interface, or the interface will take
5513          * care of the fragmentation for us, we can just send directly.
5514          */
5515         if (ip_len <= ifp->if_mtu ||
5516             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5517                 ip->ip_sum = 0;
5518                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5519                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5520                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5521                 }
5522                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5523                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5524                 goto done;
5525         }
5526
5527         /* Balk when DF bit is set or the interface didn't support TSO. */
5528         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5529                 error = EMSGSIZE;
5530                 KMOD_IPSTAT_INC(ips_cantfrag);
5531                 if (r->rt != PF_DUPTO) {
5532                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5533                             ifp->if_mtu);
5534                         goto done;
5535                 } else
5536                         goto bad;
5537         }
5538
5539         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5540         if (error)
5541                 goto bad;
5542
5543         for (; m0; m0 = m1) {
5544                 m1 = m0->m_nextpkt;
5545                 m0->m_nextpkt = NULL;
5546                 if (error == 0) {
5547                         m_clrprotoflags(m0);
5548                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5549                 } else
5550                         m_freem(m0);
5551         }
5552
5553         if (error == 0)
5554                 KMOD_IPSTAT_INC(ips_fragmented);
5555
5556 done:
5557         if (r->rt != PF_DUPTO)
5558                 *m = NULL;
5559         return;
5560
5561 bad_locked:
5562         if (s)
5563                 PF_STATE_UNLOCK(s);
5564 bad:
5565         m_freem(m0);
5566         goto done;
5567 }
5568 #endif /* INET */
5569
5570 #ifdef INET6
5571 static void
5572 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5573     struct pf_state *s, struct pf_pdesc *pd)
5574 {
5575         struct mbuf             *m0;
5576         struct sockaddr_in6     dst;
5577         struct ip6_hdr          *ip6;
5578         struct ifnet            *ifp = NULL;
5579         struct pf_addr           naddr;
5580         struct pf_src_node      *sn = NULL;
5581
5582         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5583         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5584             __func__));
5585
5586         if ((pd->pf_mtag == NULL &&
5587             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5588             pd->pf_mtag->routed++ > 3) {
5589                 m0 = *m;
5590                 *m = NULL;
5591                 goto bad_locked;
5592         }
5593
5594         if (r->rt == PF_DUPTO) {
5595                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5596                         if (s)
5597                                 PF_STATE_UNLOCK(s);
5598                         return;
5599                 }
5600         } else {
5601                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5602                         if (s)
5603                                 PF_STATE_UNLOCK(s);
5604                         return;
5605                 }
5606                 m0 = *m;
5607         }
5608
5609         ip6 = mtod(m0, struct ip6_hdr *);
5610
5611         bzero(&dst, sizeof(dst));
5612         dst.sin6_family = AF_INET6;
5613         dst.sin6_len = sizeof(dst);
5614         dst.sin6_addr = ip6->ip6_dst;
5615
5616         if (TAILQ_EMPTY(&r->rpool.list)) {
5617                 DPFPRINTF(PF_DEBUG_URGENT,
5618                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5619                 goto bad_locked;
5620         }
5621         if (s == NULL) {
5622                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5623                     &naddr, NULL, &sn);
5624                 if (!PF_AZERO(&naddr, AF_INET6))
5625                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5626                             &naddr, AF_INET6);
5627                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5628         } else {
5629                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5630                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5631                             &s->rt_addr, AF_INET6);
5632                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5633         }
5634
5635         if (s)
5636                 PF_STATE_UNLOCK(s);
5637
5638         if (ifp == NULL)
5639                 goto bad;
5640
5641         if (oifp != ifp) {
5642                 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS)
5643                         goto bad;
5644                 else if (m0 == NULL)
5645                         goto done;
5646                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5647                         DPFPRINTF(PF_DEBUG_URGENT,
5648                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5649                             __func__));
5650                         goto bad;
5651                 }
5652                 ip6 = mtod(m0, struct ip6_hdr *);
5653         }
5654
5655         if (ifp->if_flags & IFF_LOOPBACK)
5656                 m0->m_flags |= M_SKIP_FIREWALL;
5657
5658         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5659             ~ifp->if_hwassist) {
5660                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5661                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5662                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5663         }
5664
5665         /*
5666          * If the packet is too large for the outgoing interface,
5667          * send back an icmp6 error.
5668          */
5669         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5670                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5671         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5672                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5673         else {
5674                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5675                 if (r->rt != PF_DUPTO)
5676                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5677                 else
5678                         goto bad;
5679         }
5680
5681 done:
5682         if (r->rt != PF_DUPTO)
5683                 *m = NULL;
5684         return;
5685
5686 bad_locked:
5687         if (s)
5688                 PF_STATE_UNLOCK(s);
5689 bad:
5690         m_freem(m0);
5691         goto done;
5692 }
5693 #endif /* INET6 */
5694
5695 /*
5696  * FreeBSD supports cksum offloads for the following drivers.
5697  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5698  *   ti(4), txp(4), xl(4)
5699  *
5700  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5701  *  network driver performed cksum including pseudo header, need to verify
5702  *   csum_data
5703  * CSUM_DATA_VALID :
5704  *  network driver performed cksum, needs to additional pseudo header
5705  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5706  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5707  *
5708  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5709  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5710  * TCP/UDP layer.
5711  * Also, set csum_data to 0xffff to force cksum validation.
5712  */
5713 static int
5714 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5715 {
5716         u_int16_t sum = 0;
5717         int hw_assist = 0;
5718         struct ip *ip;
5719
5720         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5721                 return (1);
5722         if (m->m_pkthdr.len < off + len)
5723                 return (1);
5724
5725         switch (p) {
5726         case IPPROTO_TCP:
5727                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5728                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5729                                 sum = m->m_pkthdr.csum_data;
5730                         } else {
5731                                 ip = mtod(m, struct ip *);
5732                                 sum = in_pseudo(ip->ip_src.s_addr,
5733                                 ip->ip_dst.s_addr, htonl((u_short)len +
5734                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5735                         }
5736                         sum ^= 0xffff;
5737                         ++hw_assist;
5738                 }
5739                 break;
5740         case IPPROTO_UDP:
5741                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5742                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5743                                 sum = m->m_pkthdr.csum_data;
5744                         } else {
5745                                 ip = mtod(m, struct ip *);
5746                                 sum = in_pseudo(ip->ip_src.s_addr,
5747                                 ip->ip_dst.s_addr, htonl((u_short)len +
5748                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5749                         }
5750                         sum ^= 0xffff;
5751                         ++hw_assist;
5752                 }
5753                 break;
5754         case IPPROTO_ICMP:
5755 #ifdef INET6
5756         case IPPROTO_ICMPV6:
5757 #endif /* INET6 */
5758                 break;
5759         default:
5760                 return (1);
5761         }
5762
5763         if (!hw_assist) {
5764                 switch (af) {
5765                 case AF_INET:
5766                         if (p == IPPROTO_ICMP) {
5767                                 if (m->m_len < off)
5768                                         return (1);
5769                                 m->m_data += off;
5770                                 m->m_len -= off;
5771                                 sum = in_cksum(m, len);
5772                                 m->m_data -= off;
5773                                 m->m_len += off;
5774                         } else {
5775                                 if (m->m_len < sizeof(struct ip))
5776                                         return (1);
5777                                 sum = in4_cksum(m, p, off, len);
5778                         }
5779                         break;
5780 #ifdef INET6
5781                 case AF_INET6:
5782                         if (m->m_len < sizeof(struct ip6_hdr))
5783                                 return (1);
5784                         sum = in6_cksum(m, p, off, len);
5785                         break;
5786 #endif /* INET6 */
5787                 default:
5788                         return (1);
5789                 }
5790         }
5791         if (sum) {
5792                 switch (p) {
5793                 case IPPROTO_TCP:
5794                     {
5795                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5796                         break;
5797                     }
5798                 case IPPROTO_UDP:
5799                     {
5800                         KMOD_UDPSTAT_INC(udps_badsum);
5801                         break;
5802                     }
5803 #ifdef INET
5804                 case IPPROTO_ICMP:
5805                     {
5806                         KMOD_ICMPSTAT_INC(icps_checksum);
5807                         break;
5808                     }
5809 #endif
5810 #ifdef INET6
5811                 case IPPROTO_ICMPV6:
5812                     {
5813                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5814                         break;
5815                     }
5816 #endif /* INET6 */
5817                 }
5818                 return (1);
5819         } else {
5820                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5821                         m->m_pkthdr.csum_flags |=
5822                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5823                         m->m_pkthdr.csum_data = 0xffff;
5824                 }
5825         }
5826         return (0);
5827 }
5828
5829
5830 #ifdef INET
5831 int
5832 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5833 {
5834         struct pfi_kif          *kif;
5835         u_short                  action, reason = 0, log = 0;
5836         struct mbuf             *m = *m0;
5837         struct ip               *h = NULL;
5838         struct m_tag            *ipfwtag;
5839         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5840         struct pf_state         *s = NULL;
5841         struct pf_ruleset       *ruleset = NULL;
5842         struct pf_pdesc          pd;
5843         int                      off, dirndx, pqid = 0;
5844
5845         M_ASSERTPKTHDR(m);
5846
5847         if (!V_pf_status.running)
5848                 return (PF_PASS);
5849
5850         memset(&pd, 0, sizeof(pd));
5851
5852         kif = (struct pfi_kif *)ifp->if_pf_kif;
5853
5854         if (kif == NULL) {
5855                 DPFPRINTF(PF_DEBUG_URGENT,
5856                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5857                 return (PF_DROP);
5858         }
5859         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5860                 return (PF_PASS);
5861
5862         if (m->m_flags & M_SKIP_FIREWALL)
5863                 return (PF_PASS);
5864
5865         pd.pf_mtag = pf_find_mtag(m);
5866
5867         PF_RULES_RLOCK();
5868
5869         if (ip_divert_ptr != NULL &&
5870             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5871                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5872                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5873                         if (pd.pf_mtag == NULL &&
5874                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5875                                 action = PF_DROP;
5876                                 goto done;
5877                         }
5878                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5879                         m_tag_delete(m, ipfwtag);
5880                 }
5881                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5882                         m->m_flags |= M_FASTFWD_OURS;
5883                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5884                 }
5885         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5886                 /* We do IP header normalization and packet reassembly here */
5887                 action = PF_DROP;
5888                 goto done;
5889         }
5890         m = *m0;        /* pf_normalize messes with m0 */
5891         h = mtod(m, struct ip *);
5892
5893         off = h->ip_hl << 2;
5894         if (off < (int)sizeof(struct ip)) {
5895                 action = PF_DROP;
5896                 REASON_SET(&reason, PFRES_SHORT);
5897                 log = 1;
5898                 goto done;
5899         }
5900
5901         pd.src = (struct pf_addr *)&h->ip_src;
5902         pd.dst = (struct pf_addr *)&h->ip_dst;
5903         pd.sport = pd.dport = NULL;
5904         pd.ip_sum = &h->ip_sum;
5905         pd.proto_sum = NULL;
5906         pd.proto = h->ip_p;
5907         pd.dir = dir;
5908         pd.sidx = (dir == PF_IN) ? 0 : 1;
5909         pd.didx = (dir == PF_IN) ? 1 : 0;
5910         pd.af = AF_INET;
5911         pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
5912         pd.tot_len = ntohs(h->ip_len);
5913
5914         /* handle fragments that didn't get reassembled by normalization */
5915         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5916                 action = pf_test_fragment(&r, dir, kif, m, h,
5917                     &pd, &a, &ruleset);
5918                 goto done;
5919         }
5920
5921         switch (h->ip_p) {
5922
5923         case IPPROTO_TCP: {
5924                 struct tcphdr   th;
5925
5926                 pd.hdr.tcp = &th;
5927                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5928                     &action, &reason, AF_INET)) {
5929                         log = action != PF_PASS;
5930                         goto done;
5931                 }
5932                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5933                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5934                         pqid = 1;
5935                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5936                 if (action == PF_DROP)
5937                         goto done;
5938                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5939                     &reason);
5940                 if (action == PF_PASS) {
5941                         if (pfsync_update_state_ptr != NULL)
5942                                 pfsync_update_state_ptr(s);
5943                         r = s->rule.ptr;
5944                         a = s->anchor.ptr;
5945                         log = s->log;
5946                 } else if (s == NULL)
5947                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5948                             &a, &ruleset, inp);
5949                 break;
5950         }
5951
5952         case IPPROTO_UDP: {
5953                 struct udphdr   uh;
5954
5955                 pd.hdr.udp = &uh;
5956                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5957                     &action, &reason, AF_INET)) {
5958                         log = action != PF_PASS;
5959                         goto done;
5960                 }
5961                 if (uh.uh_dport == 0 ||
5962                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5963                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5964                         action = PF_DROP;
5965                         REASON_SET(&reason, PFRES_SHORT);
5966                         goto done;
5967                 }
5968                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5969                 if (action == PF_PASS) {
5970                         if (pfsync_update_state_ptr != NULL)
5971                                 pfsync_update_state_ptr(s);
5972                         r = s->rule.ptr;
5973                         a = s->anchor.ptr;
5974                         log = s->log;
5975                 } else if (s == NULL)
5976                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5977                             &a, &ruleset, inp);
5978                 break;
5979         }
5980
5981         case IPPROTO_ICMP: {
5982                 struct icmp     ih;
5983
5984                 pd.hdr.icmp = &ih;
5985                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5986                     &action, &reason, AF_INET)) {
5987                         log = action != PF_PASS;
5988                         goto done;
5989                 }
5990                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5991                     &reason);
5992                 if (action == PF_PASS) {
5993                         if (pfsync_update_state_ptr != NULL)
5994                                 pfsync_update_state_ptr(s);
5995                         r = s->rule.ptr;
5996                         a = s->anchor.ptr;
5997                         log = s->log;
5998                 } else if (s == NULL)
5999                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6000                             &a, &ruleset, inp);
6001                 break;
6002         }
6003
6004 #ifdef INET6
6005         case IPPROTO_ICMPV6: {
6006                 action = PF_DROP;
6007                 DPFPRINTF(PF_DEBUG_MISC,
6008                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6009                 goto done;
6010         }
6011 #endif
6012
6013         default:
6014                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6015                 if (action == PF_PASS) {
6016                         if (pfsync_update_state_ptr != NULL)
6017                                 pfsync_update_state_ptr(s);
6018                         r = s->rule.ptr;
6019                         a = s->anchor.ptr;
6020                         log = s->log;
6021                 } else if (s == NULL)
6022                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6023                             &a, &ruleset, inp);
6024                 break;
6025         }
6026
6027 done:
6028         PF_RULES_RUNLOCK();
6029         if (action == PF_PASS && h->ip_hl > 5 &&
6030             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6031                 action = PF_DROP;
6032                 REASON_SET(&reason, PFRES_IPOPTIONS);
6033                 log = r->log;
6034                 DPFPRINTF(PF_DEBUG_MISC,
6035                     ("pf: dropping packet with ip options\n"));
6036         }
6037
6038         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6039                 action = PF_DROP;
6040                 REASON_SET(&reason, PFRES_MEMORY);
6041         }
6042         if (r->rtableid >= 0)
6043                 M_SETFIB(m, r->rtableid);
6044
6045         if (r->scrub_flags & PFSTATE_SETPRIO) {
6046                 if (pd.tos & IPTOS_LOWDELAY)
6047                         pqid = 1;
6048                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6049                         action = PF_DROP;
6050                         REASON_SET(&reason, PFRES_MEMORY);
6051                         log = 1;
6052                         DPFPRINTF(PF_DEBUG_MISC,
6053                             ("pf: failed to allocate 802.1q mtag\n"));
6054                 }
6055         }
6056
6057 #ifdef ALTQ
6058         if (action == PF_PASS && r->qid) {
6059                 if (pd.pf_mtag == NULL &&
6060                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6061                         action = PF_DROP;
6062                         REASON_SET(&reason, PFRES_MEMORY);
6063                 } else {
6064                         if (s != NULL)
6065                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6066                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6067                                 pd.pf_mtag->qid = r->pqid;
6068                         else
6069                                 pd.pf_mtag->qid = r->qid;
6070                         /* Add hints for ecn. */
6071                         pd.pf_mtag->hdr = h;
6072                 }
6073
6074         }
6075 #endif /* ALTQ */
6076
6077         /*
6078          * connections redirected to loopback should not match sockets
6079          * bound specifically to loopback due to security implications,
6080          * see tcp_input() and in_pcblookup_listen().
6081          */
6082         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6083             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6084             (s->nat_rule.ptr->action == PF_RDR ||
6085             s->nat_rule.ptr->action == PF_BINAT) &&
6086             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6087                 m->m_flags |= M_SKIP_FIREWALL;
6088
6089         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6090             !PACKET_LOOPED(&pd)) {
6091
6092                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6093                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6094                 if (ipfwtag != NULL) {
6095                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6096                             ntohs(r->divert.port);
6097                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6098
6099                         if (s)
6100                                 PF_STATE_UNLOCK(s);
6101
6102                         m_tag_prepend(m, ipfwtag);
6103                         if (m->m_flags & M_FASTFWD_OURS) {
6104                                 if (pd.pf_mtag == NULL &&
6105                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6106                                         action = PF_DROP;
6107                                         REASON_SET(&reason, PFRES_MEMORY);
6108                                         log = 1;
6109                                         DPFPRINTF(PF_DEBUG_MISC,
6110                                             ("pf: failed to allocate tag\n"));
6111                                 } else {
6112                                         pd.pf_mtag->flags |=
6113                                             PF_FASTFWD_OURS_PRESENT;
6114                                         m->m_flags &= ~M_FASTFWD_OURS;
6115                                 }
6116                         }
6117                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6118                         *m0 = NULL;
6119
6120                         return (action);
6121                 } else {
6122                         /* XXX: ipfw has the same behaviour! */
6123                         action = PF_DROP;
6124                         REASON_SET(&reason, PFRES_MEMORY);
6125                         log = 1;
6126                         DPFPRINTF(PF_DEBUG_MISC,
6127                             ("pf: failed to allocate divert tag\n"));
6128                 }
6129         }
6130
6131         if (log) {
6132                 struct pf_rule *lr;
6133
6134                 if (s != NULL && s->nat_rule.ptr != NULL &&
6135                     s->nat_rule.ptr->log & PF_LOG_ALL)
6136                         lr = s->nat_rule.ptr;
6137                 else
6138                         lr = r;
6139                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6140                     (s == NULL));
6141         }
6142
6143         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6144         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6145
6146         if (action == PF_PASS || r->action == PF_DROP) {
6147                 dirndx = (dir == PF_OUT);
6148                 r->packets[dirndx]++;
6149                 r->bytes[dirndx] += pd.tot_len;
6150                 if (a != NULL) {
6151                         a->packets[dirndx]++;
6152                         a->bytes[dirndx] += pd.tot_len;
6153                 }
6154                 if (s != NULL) {
6155                         if (s->nat_rule.ptr != NULL) {
6156                                 s->nat_rule.ptr->packets[dirndx]++;
6157                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6158                         }
6159                         if (s->src_node != NULL) {
6160                                 s->src_node->packets[dirndx]++;
6161                                 s->src_node->bytes[dirndx] += pd.tot_len;
6162                         }
6163                         if (s->nat_src_node != NULL) {
6164                                 s->nat_src_node->packets[dirndx]++;
6165                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6166                         }
6167                         dirndx = (dir == s->direction) ? 0 : 1;
6168                         s->packets[dirndx]++;
6169                         s->bytes[dirndx] += pd.tot_len;
6170                 }
6171                 tr = r;
6172                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6173                 if (nr != NULL && r == &V_pf_default_rule)
6174                         tr = nr;
6175                 if (tr->src.addr.type == PF_ADDR_TABLE)
6176                         pfr_update_stats(tr->src.addr.p.tbl,
6177                             (s == NULL) ? pd.src :
6178                             &s->key[(s->direction == PF_IN)]->
6179                                 addr[(s->direction == PF_OUT)],
6180                             pd.af, pd.tot_len, dir == PF_OUT,
6181                             r->action == PF_PASS, tr->src.neg);
6182                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6183                         pfr_update_stats(tr->dst.addr.p.tbl,
6184                             (s == NULL) ? pd.dst :
6185                             &s->key[(s->direction == PF_IN)]->
6186                                 addr[(s->direction == PF_IN)],
6187                             pd.af, pd.tot_len, dir == PF_OUT,
6188                             r->action == PF_PASS, tr->dst.neg);
6189         }
6190
6191         switch (action) {
6192         case PF_SYNPROXY_DROP:
6193                 m_freem(*m0);
6194         case PF_DEFER:
6195                 *m0 = NULL;
6196                 action = PF_PASS;
6197                 break;
6198         case PF_DROP:
6199                 m_freem(*m0);
6200                 *m0 = NULL;
6201                 break;
6202         default:
6203                 /* pf_route() returns unlocked. */
6204                 if (r->rt) {
6205                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6206                         return (action);
6207                 }
6208                 break;
6209         }
6210         if (s)
6211                 PF_STATE_UNLOCK(s);
6212
6213         return (action);
6214 }
6215 #endif /* INET */
6216
6217 #ifdef INET6
6218 int
6219 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6220 {
6221         struct pfi_kif          *kif;
6222         u_short                  action, reason = 0, log = 0;
6223         struct mbuf             *m = *m0, *n = NULL;
6224         struct m_tag            *mtag;
6225         struct ip6_hdr          *h = NULL;
6226         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6227         struct pf_state         *s = NULL;
6228         struct pf_ruleset       *ruleset = NULL;
6229         struct pf_pdesc          pd;
6230         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6231         int                      fwdir = dir;
6232
6233         M_ASSERTPKTHDR(m);
6234
6235         /* Detect packet forwarding.
6236          * If the input interface is different from the output interface we're
6237          * forwarding.
6238          * We do need to be careful about bridges. If the
6239          * net.link.bridge.pfil_bridge sysctl is set we can be filtering on a
6240          * bridge, so if the input interface is a bridge member and the output
6241          * interface is its bridge or a member of the same bridge we're not
6242          * actually forwarding but bridging.
6243          */
6244         if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif &&
6245             (m->m_pkthdr.rcvif->if_bridge == NULL ||
6246             (m->m_pkthdr.rcvif->if_bridge != ifp->if_softc &&
6247             m->m_pkthdr.rcvif->if_bridge != ifp->if_bridge)))
6248                 fwdir = PF_FWD;
6249
6250         if (dir == PF_FWD)
6251                 dir = PF_OUT;
6252
6253         if (!V_pf_status.running)
6254                 return (PF_PASS);
6255
6256         memset(&pd, 0, sizeof(pd));
6257         pd.pf_mtag = pf_find_mtag(m);
6258
6259         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6260                 return (PF_PASS);
6261
6262         kif = (struct pfi_kif *)ifp->if_pf_kif;
6263         if (kif == NULL) {
6264                 DPFPRINTF(PF_DEBUG_URGENT,
6265                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6266                 return (PF_DROP);
6267         }
6268         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6269                 return (PF_PASS);
6270
6271         if (m->m_flags & M_SKIP_FIREWALL)
6272                 return (PF_PASS);
6273
6274         PF_RULES_RLOCK();
6275
6276         /* We do IP header normalization and packet reassembly here */
6277         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6278                 action = PF_DROP;
6279                 goto done;
6280         }
6281         m = *m0;        /* pf_normalize messes with m0 */
6282         h = mtod(m, struct ip6_hdr *);
6283
6284 #if 1
6285         /*
6286          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6287          * will do something bad, so drop the packet for now.
6288          */
6289         if (htons(h->ip6_plen) == 0) {
6290                 action = PF_DROP;
6291                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6292                 goto done;
6293         }
6294 #endif
6295
6296         pd.src = (struct pf_addr *)&h->ip6_src;
6297         pd.dst = (struct pf_addr *)&h->ip6_dst;
6298         pd.sport = pd.dport = NULL;
6299         pd.ip_sum = NULL;
6300         pd.proto_sum = NULL;
6301         pd.dir = dir;
6302         pd.sidx = (dir == PF_IN) ? 0 : 1;
6303         pd.didx = (dir == PF_IN) ? 1 : 0;
6304         pd.af = AF_INET6;
6305         pd.tos = 0;
6306         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6307
6308         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6309         pd.proto = h->ip6_nxt;
6310         do {
6311                 switch (pd.proto) {
6312                 case IPPROTO_FRAGMENT:
6313                         action = pf_test_fragment(&r, dir, kif, m, h,
6314                             &pd, &a, &ruleset);
6315                         if (action == PF_DROP)
6316                                 REASON_SET(&reason, PFRES_FRAG);
6317                         goto done;
6318                 case IPPROTO_ROUTING: {
6319                         struct ip6_rthdr rthdr;
6320
6321                         if (rh_cnt++) {
6322                                 DPFPRINTF(PF_DEBUG_MISC,
6323                                     ("pf: IPv6 more than one rthdr\n"));
6324                                 action = PF_DROP;
6325                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6326                                 log = 1;
6327                                 goto done;
6328                         }
6329                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6330                             &reason, pd.af)) {
6331                                 DPFPRINTF(PF_DEBUG_MISC,
6332                                     ("pf: IPv6 short rthdr\n"));
6333                                 action = PF_DROP;
6334                                 REASON_SET(&reason, PFRES_SHORT);
6335                                 log = 1;
6336                                 goto done;
6337                         }
6338                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6339                                 DPFPRINTF(PF_DEBUG_MISC,
6340                                     ("pf: IPv6 rthdr0\n"));
6341                                 action = PF_DROP;
6342                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6343                                 log = 1;
6344                                 goto done;
6345                         }
6346                         /* FALLTHROUGH */
6347                 }
6348                 case IPPROTO_AH:
6349                 case IPPROTO_HOPOPTS:
6350                 case IPPROTO_DSTOPTS: {
6351                         /* get next header and header length */
6352                         struct ip6_ext  opt6;
6353
6354                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6355                             NULL, &reason, pd.af)) {
6356                                 DPFPRINTF(PF_DEBUG_MISC,
6357                                     ("pf: IPv6 short opt\n"));
6358                                 action = PF_DROP;
6359                                 log = 1;
6360                                 goto done;
6361                         }
6362                         if (pd.proto == IPPROTO_AH)
6363                                 off += (opt6.ip6e_len + 2) * 4;
6364                         else
6365                                 off += (opt6.ip6e_len + 1) * 8;
6366                         pd.proto = opt6.ip6e_nxt;
6367                         /* goto the next header */
6368                         break;
6369                 }
6370                 default:
6371                         terminal++;
6372                         break;
6373                 }
6374         } while (!terminal);
6375
6376         /* if there's no routing header, use unmodified mbuf for checksumming */
6377         if (!n)
6378                 n = m;
6379
6380         switch (pd.proto) {
6381
6382         case IPPROTO_TCP: {
6383                 struct tcphdr   th;
6384
6385                 pd.hdr.tcp = &th;
6386                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6387                     &action, &reason, AF_INET6)) {
6388                         log = action != PF_PASS;
6389                         goto done;
6390                 }
6391                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6392                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6393                 if (action == PF_DROP)
6394                         goto done;
6395                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6396                     &reason);
6397                 if (action == PF_PASS) {
6398                         if (pfsync_update_state_ptr != NULL)
6399                                 pfsync_update_state_ptr(s);
6400                         r = s->rule.ptr;
6401                         a = s->anchor.ptr;
6402                         log = s->log;
6403                 } else if (s == NULL)
6404                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6405                             &a, &ruleset, inp);
6406                 break;
6407         }
6408
6409         case IPPROTO_UDP: {
6410                 struct udphdr   uh;
6411
6412                 pd.hdr.udp = &uh;
6413                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6414                     &action, &reason, AF_INET6)) {
6415                         log = action != PF_PASS;
6416                         goto done;
6417                 }
6418                 if (uh.uh_dport == 0 ||
6419                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6420                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6421                         action = PF_DROP;
6422                         REASON_SET(&reason, PFRES_SHORT);
6423                         goto done;
6424                 }
6425                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6426                 if (action == PF_PASS) {
6427                         if (pfsync_update_state_ptr != NULL)
6428                                 pfsync_update_state_ptr(s);
6429                         r = s->rule.ptr;
6430                         a = s->anchor.ptr;
6431                         log = s->log;
6432                 } else if (s == NULL)
6433                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6434                             &a, &ruleset, inp);
6435                 break;
6436         }
6437
6438         case IPPROTO_ICMP: {
6439                 action = PF_DROP;
6440                 DPFPRINTF(PF_DEBUG_MISC,
6441                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6442                 goto done;
6443         }
6444
6445         case IPPROTO_ICMPV6: {
6446                 struct icmp6_hdr        ih;
6447
6448                 pd.hdr.icmp6 = &ih;
6449                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6450                     &action, &reason, AF_INET6)) {
6451                         log = action != PF_PASS;
6452                         goto done;
6453                 }
6454                 action = pf_test_state_icmp(&s, dir, kif,
6455                     m, off, h, &pd, &reason);
6456                 if (action == PF_PASS) {
6457                         if (pfsync_update_state_ptr != NULL)
6458                                 pfsync_update_state_ptr(s);
6459                         r = s->rule.ptr;
6460                         a = s->anchor.ptr;
6461                         log = s->log;
6462                 } else if (s == NULL)
6463                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6464                             &a, &ruleset, inp);
6465                 break;
6466         }
6467
6468         default:
6469                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6470                 if (action == PF_PASS) {
6471                         if (pfsync_update_state_ptr != NULL)
6472                                 pfsync_update_state_ptr(s);
6473                         r = s->rule.ptr;
6474                         a = s->anchor.ptr;
6475                         log = s->log;
6476                 } else if (s == NULL)
6477                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6478                             &a, &ruleset, inp);
6479                 break;
6480         }
6481
6482 done:
6483         PF_RULES_RUNLOCK();
6484         if (n != m) {
6485                 m_freem(n);
6486                 n = NULL;
6487         }
6488
6489         /* handle dangerous IPv6 extension headers. */
6490         if (action == PF_PASS && rh_cnt &&
6491             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6492                 action = PF_DROP;
6493                 REASON_SET(&reason, PFRES_IPOPTIONS);
6494                 log = r->log;
6495                 DPFPRINTF(PF_DEBUG_MISC,
6496                     ("pf: dropping packet with dangerous v6 headers\n"));
6497         }
6498
6499         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6500                 action = PF_DROP;
6501                 REASON_SET(&reason, PFRES_MEMORY);
6502         }
6503         if (r->rtableid >= 0)
6504                 M_SETFIB(m, r->rtableid);
6505
6506         if (r->scrub_flags & PFSTATE_SETPRIO) {
6507                 if (pd.tos & IPTOS_LOWDELAY)
6508                         pqid = 1;
6509                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6510                         action = PF_DROP;
6511                         REASON_SET(&reason, PFRES_MEMORY);
6512                         log = 1;
6513                         DPFPRINTF(PF_DEBUG_MISC,
6514                             ("pf: failed to allocate 802.1q mtag\n"));
6515                 }
6516         }
6517
6518 #ifdef ALTQ
6519         if (action == PF_PASS && r->qid) {
6520                 if (pd.pf_mtag == NULL &&
6521                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6522                         action = PF_DROP;
6523                         REASON_SET(&reason, PFRES_MEMORY);
6524                 } else {
6525                         if (s != NULL)
6526                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6527                         if (pd.tos & IPTOS_LOWDELAY)
6528                                 pd.pf_mtag->qid = r->pqid;
6529                         else
6530                                 pd.pf_mtag->qid = r->qid;
6531                         /* Add hints for ecn. */
6532                         pd.pf_mtag->hdr = h;
6533                 }
6534         }
6535 #endif /* ALTQ */
6536
6537         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6538             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6539             (s->nat_rule.ptr->action == PF_RDR ||
6540             s->nat_rule.ptr->action == PF_BINAT) &&
6541             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6542                 m->m_flags |= M_SKIP_FIREWALL;
6543
6544         /* XXX: Anybody working on it?! */
6545         if (r->divert.port)
6546                 printf("pf: divert(9) is not supported for IPv6\n");
6547
6548         if (log) {
6549                 struct pf_rule *lr;
6550
6551                 if (s != NULL && s->nat_rule.ptr != NULL &&
6552                     s->nat_rule.ptr->log & PF_LOG_ALL)
6553                         lr = s->nat_rule.ptr;
6554                 else
6555                         lr = r;
6556                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6557                     &pd, (s == NULL));
6558         }
6559
6560         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6561         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6562
6563         if (action == PF_PASS || r->action == PF_DROP) {
6564                 dirndx = (dir == PF_OUT);
6565                 r->packets[dirndx]++;
6566                 r->bytes[dirndx] += pd.tot_len;
6567                 if (a != NULL) {
6568                         a->packets[dirndx]++;
6569                         a->bytes[dirndx] += pd.tot_len;
6570                 }
6571                 if (s != NULL) {
6572                         if (s->nat_rule.ptr != NULL) {
6573                                 s->nat_rule.ptr->packets[dirndx]++;
6574                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6575                         }
6576                         if (s->src_node != NULL) {
6577                                 s->src_node->packets[dirndx]++;
6578                                 s->src_node->bytes[dirndx] += pd.tot_len;
6579                         }
6580                         if (s->nat_src_node != NULL) {
6581                                 s->nat_src_node->packets[dirndx]++;
6582                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6583                         }
6584                         dirndx = (dir == s->direction) ? 0 : 1;
6585                         s->packets[dirndx]++;
6586                         s->bytes[dirndx] += pd.tot_len;
6587                 }
6588                 tr = r;
6589                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6590                 if (nr != NULL && r == &V_pf_default_rule)
6591                         tr = nr;
6592                 if (tr->src.addr.type == PF_ADDR_TABLE)
6593                         pfr_update_stats(tr->src.addr.p.tbl,
6594                             (s == NULL) ? pd.src :
6595                             &s->key[(s->direction == PF_IN)]->addr[0],
6596                             pd.af, pd.tot_len, dir == PF_OUT,
6597                             r->action == PF_PASS, tr->src.neg);
6598                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6599                         pfr_update_stats(tr->dst.addr.p.tbl,
6600                             (s == NULL) ? pd.dst :
6601                             &s->key[(s->direction == PF_IN)]->addr[1],
6602                             pd.af, pd.tot_len, dir == PF_OUT,
6603                             r->action == PF_PASS, tr->dst.neg);
6604         }
6605
6606         switch (action) {
6607         case PF_SYNPROXY_DROP:
6608                 m_freem(*m0);
6609         case PF_DEFER:
6610                 *m0 = NULL;
6611                 action = PF_PASS;
6612                 break;
6613         case PF_DROP:
6614                 m_freem(*m0);
6615                 *m0 = NULL;
6616                 break;
6617         default:
6618                 /* pf_route6() returns unlocked. */
6619                 if (r->rt) {
6620                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6621                         return (action);
6622                 }
6623                 break;
6624         }
6625
6626         if (s)
6627                 PF_STATE_UNLOCK(s);
6628
6629         /* If reassembled packet passed, create new fragments. */
6630         if (action == PF_PASS && *m0 && fwdir == PF_FWD &&
6631             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6632                 action = pf_refragment6(ifp, m0, mtag);
6633
6634         return (action);
6635 }
6636 #endif /* INET6 */