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/pfil.h>
  72 #include <net/pfvar.h>
  73 #include <net/if_pflog.h>
  74 #include <net/if_pfsync.h>
  75
  76 #include <netinet/in_pcb.h>
  77 #include <netinet/in_var.h>
  78 #include <netinet/in_fib.h>
  79 #include <netinet/ip.h>
  80 #include <netinet/ip_fw.h>
  81 #include <netinet/ip_icmp.h>
  82 #include <netinet/icmp_var.h>
  83 #include <netinet/ip_var.h>
  84 #include <netinet/tcp.h>
  85 #include <netinet/tcp_fsm.h>
  86 #include <netinet/tcp_seq.h>
  87 #include <netinet/tcp_timer.h>
  88 #include <netinet/tcp_var.h>
  89 #include <netinet/udp.h>
  90 #include <netinet/udp_var.h>
  91
  92 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  93
  94 #ifdef INET6
  95 #include <netinet/ip6.h>
  96 #include <netinet/icmp6.h>
  97 #include <netinet6/nd6.h>
  98 #include <netinet6/ip6_var.h>
  99 #include <netinet6/in6_pcb.h>
 100 #include <netinet6/in6_fib.h>
 101 #include <netinet6/scope6_var.h>
 102 #endif /* INET6 */
 103
 104 #include <machine/in_cksum.h>
 105 #include <security/mac/mac_framework.h>
 106
 107 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 108
 109 /*
 110  * Global variables
 111  */
 112
 113 /* state tables */
 114 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 115 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 116 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 117 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 118 VNET_DEFINE(struct pf_kstatus,           pf_status);
 119
 120 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 121 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 122 VNET_DEFINE(int,                         altqs_inactive_open);
 123 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 124
 125 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 126 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 127 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 128 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 129 VNET_DEFINE(int,                         pf_tcp_secret_init);
 130 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 131 VNET_DEFINE(int,                         pf_tcp_iss_off);
 132 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 133 VNET_DECLARE(int,                        pf_vnet_active);
 134 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 135
 136 /*
 137  * Queue for pf_intr() sends.
 138  */
 139 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 140 struct pf_send_entry {
 141         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 142         struct mbuf                     *pfse_m;
 143         enum {
 144                 PFSE_IP,
 145                 PFSE_IP6,
 146                 PFSE_ICMP,
 147                 PFSE_ICMP6,
 148         }                               pfse_type;
 149         struct {
 150                 int             type;
 151                 int             code;
 152                 int             mtu;
 153         } icmpopts;
 154 };
 155
 156 STAILQ_HEAD(pf_send_head, pf_send_entry);
 157 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 158 #define V_pf_sendqueue  VNET(pf_sendqueue)
 159
 160 static struct mtx pf_sendqueue_mtx;
 161 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 162 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 163 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 164
 165 /*
 166  * Queue for pf_overload_task() tasks.
 167  */
 168 struct pf_overload_entry {
 169         SLIST_ENTRY(pf_overload_entry)  next;
 170         struct pf_addr                  addr;
 171         sa_family_t                     af;
 172         uint8_t                         dir;
 173         struct pf_rule                  *rule;
 174 };
 175
 176 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 177 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 178 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 179 static VNET_DEFINE(struct task, pf_overloadtask);
 180 #define V_pf_overloadtask       VNET(pf_overloadtask)
 181
 182 static struct mtx pf_overloadqueue_mtx;
 183 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 184     "pf overload/flush queue", MTX_DEF);
 185 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 186 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 187
 188 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 189 struct mtx pf_unlnkdrules_mtx;
 190 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 191     MTX_DEF);
 192
 193 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 194 #define V_pf_sources_z  VNET(pf_sources_z)
 195 uma_zone_t              pf_mtag_z;
 196 VNET_DEFINE(uma_zone_t,  pf_state_z);
 197 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 198
 199 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 200 #define PFID_CPUBITS    8
 201 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 202 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 203 #define PFID_MAXID      (~PFID_CPUMASK)
 204 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 205
 206 static void              pf_src_tree_remove_state(struct pf_state *);
 207 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 208                             u_int32_t);
 209 static void              pf_add_threshold(struct pf_threshold *);
 210 static int               pf_check_threshold(struct pf_threshold *);
 211
 212 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 213                             u_int16_t *, u_int16_t *, struct pf_addr *,
 214                             u_int16_t, u_int8_t, sa_family_t);
 215 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 216                             struct tcphdr *, struct pf_state_peer *);
 217 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 218                             struct pf_addr *, struct pf_addr *, u_int16_t,
 219                             u_int16_t *, u_int16_t *, u_int16_t *,
 220                             u_int16_t *, u_int8_t, sa_family_t);
 221 static void              pf_send_tcp(struct mbuf *,
 222                             const struct pf_rule *, sa_family_t,
 223                             const struct pf_addr *, const struct pf_addr *,
 224                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 225                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 226                             u_int16_t, struct ifnet *);
 227 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 228                             sa_family_t, struct pf_rule *);
 229 static void              pf_detach_state(struct pf_state *);
 230 static int               pf_state_key_attach(struct pf_state_key *,
 231                             struct pf_state_key *, struct pf_state *);
 232 static void              pf_state_key_detach(struct pf_state *, int);
 233 static int               pf_state_key_ctor(void *, int, void *, int);
 234 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 235 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 236                             int, struct pfi_kif *, struct mbuf *, int,
 237                             struct pf_pdesc *, struct pf_rule **,
 238                             struct pf_ruleset **, struct inpcb *);
 239 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 240                             struct pf_rule *, struct pf_pdesc *,
 241                             struct pf_src_node *, struct pf_state_key *,
 242                             struct pf_state_key *, struct mbuf *, int,
 243                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 244                             struct pf_state **, int, u_int16_t, u_int16_t,
 245                             int);
 246 static int               pf_test_fragment(struct pf_rule **, int,
 247                             struct pfi_kif *, struct mbuf *, void *,
 248                             struct pf_pdesc *, struct pf_rule **,
 249                             struct pf_ruleset **);
 250 static int               pf_tcp_track_full(struct pf_state_peer *,
 251                             struct pf_state_peer *, struct pf_state **,
 252                             struct pfi_kif *, struct mbuf *, int,
 253                             struct pf_pdesc *, u_short *, int *);
 254 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 255                             struct pf_state_peer *, struct pf_state **,
 256                             struct pf_pdesc *, u_short *);
 257 static int               pf_test_state_tcp(struct pf_state **, int,
 258                             struct pfi_kif *, struct mbuf *, int,
 259                             void *, struct pf_pdesc *, u_short *);
 260 static int               pf_test_state_udp(struct pf_state **, int,
 261                             struct pfi_kif *, struct mbuf *, int,
 262                             void *, struct pf_pdesc *);
 263 static int               pf_test_state_icmp(struct pf_state **, int,
 264                             struct pfi_kif *, struct mbuf *, int,
 265                             void *, struct pf_pdesc *, u_short *);
 266 static int               pf_test_state_other(struct pf_state **, int,
 267                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 268 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 269                             sa_family_t);
 270 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 271                             sa_family_t);
 272 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 273                                 int, u_int16_t);
 274 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 275                             u_int8_t, sa_family_t);
 276 static void              pf_print_state_parts(struct pf_state *,
 277                             struct pf_state_key *, struct pf_state_key *);
 278 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 279                             struct pf_addr_wrap *);
 280 static struct pf_state  *pf_find_state(struct pfi_kif *,
 281                             struct pf_state_key_cmp *, u_int);
 282 static int               pf_src_connlimit(struct pf_state **);
 283 static void              pf_overload_task(void *v, int pending);
 284 static int               pf_insert_src_node(struct pf_src_node **,
 285                             struct pf_rule *, struct pf_addr *, sa_family_t);
 286 static u_int             pf_purge_expired_states(u_int, int);
 287 static void              pf_purge_unlinked_rules(void);
 288 static int               pf_mtag_uminit(void *, int, int);
 289 static void              pf_mtag_free(struct m_tag *);
 290 #ifdef INET
 291 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 292                             struct ifnet *, struct pf_state *,
 293                             struct pf_pdesc *);
 294 #endif /* INET */
 295 #ifdef INET6
 296 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 297                             struct pf_addr *, u_int8_t);
 298 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 299                             struct ifnet *, struct pf_state *,
 300                             struct pf_pdesc *);
 301 #endif /* INET6 */
 302
 303 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 304
 305 extern int pf_end_threads;
 306
 307 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 308
 309 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 310                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 311
 312 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 313         do {                                                            \
 314                 (s) = pf_find_state((i), (k), (d));                     \
 315                 if ((s) == NULL)                                        \
 316                         return (PF_DROP);                               \
 317                 if (PACKET_LOOPED(pd))                                  \
 318                         return (PF_PASS);                               \
 319                 if ((d) == PF_OUT &&                                    \
 320                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 321                     (s)->rule.ptr->direction == PF_OUT) ||              \
 322                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 323                     (s)->rule.ptr->direction == PF_IN)) &&              \
 324                     (s)->rt_kif != NULL &&                              \
 325                     (s)->rt_kif != (i))                                 \
 326                         return (PF_PASS);                               \
 327         } while (0)
 328
 329 #define BOUND_IFACE(r, k) \
 330         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 331
 332 #define STATE_INC_COUNTERS(s)                                           \
 333         do {                                                            \
 334                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 335                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 336                 if (s->anchor.ptr != NULL) {                            \
 337                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 338                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 339                 }                                                       \
 340                 if (s->nat_rule.ptr != NULL) {                          \
 341                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 342                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 343                 }                                                       \
 344         } while (0)
 345
 346 #define STATE_DEC_COUNTERS(s)                                           \
 347         do {                                                            \
 348                 if (s->nat_rule.ptr != NULL)                            \
 349                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 350                 if (s->anchor.ptr != NULL)                              \
 351                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 352                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 353         } while (0)
 354
 355 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 356 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 357 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 358 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 359
 360 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 361
 362 u_long  pf_hashmask;
 363 u_long  pf_srchashmask;
 364 static u_long   pf_hashsize;
 365 static u_long   pf_srchashsize;
 366
 367 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 368     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 369 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 370     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 371
 372 VNET_DEFINE(void *, pf_swi_cookie);
 373
 374 VNET_DEFINE(uint32_t, pf_hashseed);
 375 #define V_pf_hashseed   VNET(pf_hashseed)
 376
 377 int
 378 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 379 {
 380
 381         switch (af) {
 382 #ifdef INET
 383         case AF_INET:
 384                 if (a->addr32[0] > b->addr32[0])
 385                         return (1);
 386                 if (a->addr32[0] < b->addr32[0])
 387                         return (-1);
 388                 break;
 389 #endif /* INET */
 390 #ifdef INET6
 391         case AF_INET6:
 392                 if (a->addr32[3] > b->addr32[3])
 393                         return (1);
 394                 if (a->addr32[3] < b->addr32[3])
 395                         return (-1);
 396                 if (a->addr32[2] > b->addr32[2])
 397                         return (1);
 398                 if (a->addr32[2] < b->addr32[2])
 399                         return (-1);
 400                 if (a->addr32[1] > b->addr32[1])
 401                         return (1);
 402                 if (a->addr32[1] < b->addr32[1])
 403                         return (-1);
 404                 if (a->addr32[0] > b->addr32[0])
 405                         return (1);
 406                 if (a->addr32[0] < b->addr32[0])
 407                         return (-1);
 408                 break;
 409 #endif /* INET6 */
 410         default:
 411                 panic("%s: unknown address family %u", __func__, af);
 412         }
 413         return (0);
 414 }
 415
 416 static __inline uint32_t
 417 pf_hashkey(struct pf_state_key *sk)
 418 {
 419         uint32_t h;
 420
 421         h = murmur3_32_hash32((uint32_t *)sk,
 422             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 423             V_pf_hashseed);
 424
 425         return (h & pf_hashmask);
 426 }
 427
 428 static __inline uint32_t
 429 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 430 {
 431         uint32_t h;
 432
 433         switch (af) {
 434         case AF_INET:
 435                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 436                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 437                 break;
 438         case AF_INET6:
 439                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 440                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 441                 break;
 442         default:
 443                 panic("%s: unknown address family %u", __func__, af);
 444         }
 445
 446         return (h & pf_srchashmask);
 447 }
 448
 449 #ifdef ALTQ
 450 static int
 451 pf_state_hash(struct pf_state *s)
 452 {
 453         u_int32_t hv = (intptr_t)s / sizeof(*s);
 454
 455         hv ^= crc32(&s->src, sizeof(s->src));
 456         hv ^= crc32(&s->dst, sizeof(s->dst));
 457         if (hv == 0)
 458                 hv = 1;
 459         return (hv);
 460 }
 461 #endif
 462
 463 #ifdef INET6
 464 void
 465 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 466 {
 467         switch (af) {
 468 #ifdef INET
 469         case AF_INET:
 470                 dst->addr32[0] = src->addr32[0];
 471                 break;
 472 #endif /* INET */
 473         case AF_INET6:
 474                 dst->addr32[0] = src->addr32[0];
 475                 dst->addr32[1] = src->addr32[1];
 476                 dst->addr32[2] = src->addr32[2];
 477                 dst->addr32[3] = src->addr32[3];
 478                 break;
 479         }
 480 }
 481 #endif /* INET6 */
 482
 483 static void
 484 pf_init_threshold(struct pf_threshold *threshold,
 485     u_int32_t limit, u_int32_t seconds)
 486 {
 487         threshold->limit = limit * PF_THRESHOLD_MULT;
 488         threshold->seconds = seconds;
 489         threshold->count = 0;
 490         threshold->last = time_uptime;
 491 }
 492
 493 static void
 494 pf_add_threshold(struct pf_threshold *threshold)
 495 {
 496         u_int32_t t = time_uptime, diff = t - threshold->last;
 497
 498         if (diff >= threshold->seconds)
 499                 threshold->count = 0;
 500         else
 501                 threshold->count -= threshold->count * diff /
 502                     threshold->seconds;
 503         threshold->count += PF_THRESHOLD_MULT;
 504         threshold->last = t;
 505 }
 506
 507 static int
 508 pf_check_threshold(struct pf_threshold *threshold)
 509 {
 510         return (threshold->count > threshold->limit);
 511 }
 512
 513 static int
 514 pf_src_connlimit(struct pf_state **state)
 515 {
 516         struct pf_overload_entry *pfoe;
 517         int bad = 0;
 518
 519         PF_STATE_LOCK_ASSERT(*state);
 520
 521         (*state)->src_node->conn++;
 522         (*state)->src.tcp_est = 1;
 523         pf_add_threshold(&(*state)->src_node->conn_rate);
 524
 525         if ((*state)->rule.ptr->max_src_conn &&
 526             (*state)->rule.ptr->max_src_conn <
 527             (*state)->src_node->conn) {
 528                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 529                 bad++;
 530         }
 531
 532         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 533             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 534                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 535                 bad++;
 536         }
 537
 538         if (!bad)
 539                 return (0);
 540
 541         /* Kill this state. */
 542         (*state)->timeout = PFTM_PURGE;
 543         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 544
 545         if ((*state)->rule.ptr->overload_tbl == NULL)
 546                 return (1);
 547
 548         /* Schedule overloading and flushing task. */
 549         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 550         if (pfoe == NULL)
 551                 return (1);     /* too bad :( */
 552
 553         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 554         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 555         pfoe->rule = (*state)->rule.ptr;
 556         pfoe->dir = (*state)->direction;
 557         PF_OVERLOADQ_LOCK();
 558         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 559         PF_OVERLOADQ_UNLOCK();
 560         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 561
 562         return (1);
 563 }
 564
 565 static void
 566 pf_overload_task(void *v, int pending)
 567 {
 568         struct pf_overload_head queue;
 569         struct pfr_addr p;
 570         struct pf_overload_entry *pfoe, *pfoe1;
 571         uint32_t killed = 0;
 572
 573         CURVNET_SET((struct vnet *)v);
 574
 575         PF_OVERLOADQ_LOCK();
 576         queue = V_pf_overloadqueue;
 577         SLIST_INIT(&V_pf_overloadqueue);
 578         PF_OVERLOADQ_UNLOCK();
 579
 580         bzero(&p, sizeof(p));
 581         SLIST_FOREACH(pfoe, &queue, next) {
 582                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 583                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 584                         printf("%s: blocking address ", __func__);
 585                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 586                         printf("\n");
 587                 }
 588
 589                 p.pfra_af = pfoe->af;
 590                 switch (pfoe->af) {
 591 #ifdef INET
 592                 case AF_INET:
 593                         p.pfra_net = 32;
 594                         p.pfra_ip4addr = pfoe->addr.v4;
 595                         break;
 596 #endif
 597 #ifdef INET6
 598                 case AF_INET6:
 599                         p.pfra_net = 128;
 600                         p.pfra_ip6addr = pfoe->addr.v6;
 601                         break;
 602 #endif
 603                 }
 604
 605                 PF_RULES_WLOCK();
 606                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 607                 PF_RULES_WUNLOCK();
 608         }
 609
 610         /*
 611          * Remove those entries, that don't need flushing.
 612          */
 613         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 614                 if (pfoe->rule->flush == 0) {
 615                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 616                         free(pfoe, M_PFTEMP);
 617                 } else
 618                         counter_u64_add(
 619                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 620
 621         /* If nothing to flush, return. */
 622         if (SLIST_EMPTY(&queue)) {
 623                 CURVNET_RESTORE();
 624                 return;
 625         }
 626
 627         for (int i = 0; i <= pf_hashmask; i++) {
 628                 struct pf_idhash *ih = &V_pf_idhash[i];
 629                 struct pf_state_key *sk;
 630                 struct pf_state *s;
 631
 632                 PF_HASHROW_LOCK(ih);
 633                 LIST_FOREACH(s, &ih->states, entry) {
 634                     sk = s->key[PF_SK_WIRE];
 635                     SLIST_FOREACH(pfoe, &queue, next)
 636                         if (sk->af == pfoe->af &&
 637                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 638                             pfoe->rule == s->rule.ptr) &&
 639                             ((pfoe->dir == PF_OUT &&
 640                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 641                             (pfoe->dir == PF_IN &&
 642                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 643                                 s->timeout = PFTM_PURGE;
 644                                 s->src.state = s->dst.state = TCPS_CLOSED;
 645                                 killed++;
 646                         }
 647                 }
 648                 PF_HASHROW_UNLOCK(ih);
 649         }
 650         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 651                 free(pfoe, M_PFTEMP);
 652         if (V_pf_status.debug >= PF_DEBUG_MISC)
 653                 printf("%s: %u states killed", __func__, killed);
 654
 655         CURVNET_RESTORE();
 656 }
 657
 658 /*
 659  * Can return locked on failure, so that we can consistently
 660  * allocate and insert a new one.
 661  */
 662 struct pf_src_node *
 663 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 664         int returnlocked)
 665 {
 666         struct pf_srchash *sh;
 667         struct pf_src_node *n;
 668
 669         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 670
 671         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 672         PF_HASHROW_LOCK(sh);
 673         LIST_FOREACH(n, &sh->nodes, entry)
 674                 if (n->rule.ptr == rule && n->af == af &&
 675                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 676                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 677                         break;
 678         if (n != NULL) {
 679                 n->states++;
 680                 PF_HASHROW_UNLOCK(sh);
 681         } else if (returnlocked == 0)
 682                 PF_HASHROW_UNLOCK(sh);
 683
 684         return (n);
 685 }
 686
 687 static int
 688 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 689     struct pf_addr *src, sa_family_t af)
 690 {
 691
 692         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 693             rule->rpool.opts & PF_POOL_STICKYADDR),
 694             ("%s for non-tracking rule %p", __func__, rule));
 695
 696         if (*sn == NULL)
 697                 *sn = pf_find_src_node(src, rule, af, 1);
 698
 699         if (*sn == NULL) {
 700                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 701
 702                 PF_HASHROW_ASSERT(sh);
 703
 704                 if (!rule->max_src_nodes ||
 705                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 706                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 707                 else
 708                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 709                             1);
 710                 if ((*sn) == NULL) {
 711                         PF_HASHROW_UNLOCK(sh);
 712                         return (-1);
 713                 }
 714
 715                 pf_init_threshold(&(*sn)->conn_rate,
 716                     rule->max_src_conn_rate.limit,
 717                     rule->max_src_conn_rate.seconds);
 718
 719                 (*sn)->af = af;
 720                 (*sn)->rule.ptr = rule;
 721                 PF_ACPY(&(*sn)->addr, src, af);
 722                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 723                 (*sn)->creation = time_uptime;
 724                 (*sn)->ruletype = rule->action;
 725                 (*sn)->states = 1;
 726                 if ((*sn)->rule.ptr != NULL)
 727                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 728                 PF_HASHROW_UNLOCK(sh);
 729                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 730         } else {
 731                 if (rule->max_src_states &&
 732                     (*sn)->states >= rule->max_src_states) {
 733                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 734                             1);
 735                         return (-1);
 736                 }
 737         }
 738         return (0);
 739 }
 740
 741 void
 742 pf_unlink_src_node(struct pf_src_node *src)
 743 {
 744
 745         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 746         LIST_REMOVE(src, entry);
 747         if (src->rule.ptr)
 748                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 749 }
 750
 751 u_int
 752 pf_free_src_nodes(struct pf_src_node_list *head)
 753 {
 754         struct pf_src_node *sn, *tmp;
 755         u_int count = 0;
 756
 757         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 758                 uma_zfree(V_pf_sources_z, sn);
 759                 count++;
 760         }
 761
 762         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 763
 764         return (count);
 765 }
 766
 767 void
 768 pf_mtag_initialize()
 769 {
 770
 771         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 772             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 773             UMA_ALIGN_PTR, 0);
 774 }
 775
 776 /* Per-vnet data storage structures initialization. */
 777 void
 778 pf_initialize()
 779 {
 780         struct pf_keyhash       *kh;
 781         struct pf_idhash        *ih;
 782         struct pf_srchash       *sh;
 783         u_int i;
 784
 785         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 786                 pf_hashsize = PF_HASHSIZ;
 787         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 788                 pf_srchashsize = PF_SRCHASHSIZ;
 789
 790         V_pf_hashseed = arc4random();
 791
 792         /* States and state keys storage. */
 793         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 794             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 795         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 796         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 797         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 798
 799         V_pf_state_key_z = uma_zcreate("pf state keys",
 800             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 801             UMA_ALIGN_PTR, 0);
 802
 803         V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
 804             M_PFHASH, M_NOWAIT | M_ZERO);
 805         V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
 806             M_PFHASH, M_NOWAIT | M_ZERO);
 807         if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
 808                 printf("pf: Unable to allocate memory for "
 809                     "state_hashsize %lu.\n", pf_hashsize);
 810
 811                 free(V_pf_keyhash, M_PFHASH);
 812                 free(V_pf_idhash, M_PFHASH);
 813
 814                 pf_hashsize = PF_HASHSIZ;
 815                 V_pf_keyhash = mallocarray(pf_hashsize,
 816                     sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
 817                 V_pf_idhash = mallocarray(pf_hashsize,
 818                     sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
 819         }
 820
 821         pf_hashmask = pf_hashsize - 1;
 822         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 823             i++, kh++, ih++) {
 824                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 825                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 826         }
 827
 828         /* Source nodes. */
 829         V_pf_sources_z = uma_zcreate("pf source nodes",
 830             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 831             0);
 832         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 833         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 834         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 835
 836         V_pf_srchash = mallocarray(pf_srchashsize,
 837             sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
 838         if (V_pf_srchash == NULL) {
 839                 printf("pf: Unable to allocate memory for "
 840                     "source_hashsize %lu.\n", pf_srchashsize);
 841
 842                 pf_srchashsize = PF_SRCHASHSIZ;
 843                 V_pf_srchash = mallocarray(pf_srchashsize,
 844                     sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
 845         }
 846
 847         pf_srchashmask = pf_srchashsize - 1;
 848         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 849                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 850
 851         /* ALTQ */
 852         TAILQ_INIT(&V_pf_altqs[0]);
 853         TAILQ_INIT(&V_pf_altqs[1]);
 854         TAILQ_INIT(&V_pf_pabuf);
 855         V_pf_altqs_active = &V_pf_altqs[0];
 856         V_pf_altqs_inactive = &V_pf_altqs[1];
 857
 858         /* Send & overload+flush queues. */
 859         STAILQ_INIT(&V_pf_sendqueue);
 860         SLIST_INIT(&V_pf_overloadqueue);
 861         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 862
 863         /* Unlinked, but may be referenced rules. */
 864         TAILQ_INIT(&V_pf_unlinked_rules);
 865 }
 866
 867 void
 868 pf_mtag_cleanup()
 869 {
 870
 871         uma_zdestroy(pf_mtag_z);
 872 }
 873
 874 void
 875 pf_cleanup()
 876 {
 877         struct pf_keyhash       *kh;
 878         struct pf_idhash        *ih;
 879         struct pf_srchash       *sh;
 880         struct pf_send_entry    *pfse, *next;
 881         u_int i;
 882
 883         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 884             i++, kh++, ih++) {
 885                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 886                     __func__));
 887                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 888                     __func__));
 889                 mtx_destroy(&kh->lock);
 890                 mtx_destroy(&ih->lock);
 891         }
 892         free(V_pf_keyhash, M_PFHASH);
 893         free(V_pf_idhash, M_PFHASH);
 894
 895         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 896                 KASSERT(LIST_EMPTY(&sh->nodes),
 897                     ("%s: source node hash not empty", __func__));
 898                 mtx_destroy(&sh->lock);
 899         }
 900         free(V_pf_srchash, M_PFHASH);
 901
 902         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 903                 m_freem(pfse->pfse_m);
 904                 free(pfse, M_PFTEMP);
 905         }
 906
 907         uma_zdestroy(V_pf_sources_z);
 908         uma_zdestroy(V_pf_state_z);
 909         uma_zdestroy(V_pf_state_key_z);
 910 }
 911
 912 static int
 913 pf_mtag_uminit(void *mem, int size, int how)
 914 {
 915         struct m_tag *t;
 916
 917         t = (struct m_tag *)mem;
 918         t->m_tag_cookie = MTAG_ABI_COMPAT;
 919         t->m_tag_id = PACKET_TAG_PF;
 920         t->m_tag_len = sizeof(struct pf_mtag);
 921         t->m_tag_free = pf_mtag_free;
 922
 923         return (0);
 924 }
 925
 926 static void
 927 pf_mtag_free(struct m_tag *t)
 928 {
 929
 930         uma_zfree(pf_mtag_z, t);
 931 }
 932
 933 struct pf_mtag *
 934 pf_get_mtag(struct mbuf *m)
 935 {
 936         struct m_tag *mtag;
 937
 938         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 939                 return ((struct pf_mtag *)(mtag + 1));
 940
 941         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 942         if (mtag == NULL)
 943                 return (NULL);
 944         bzero(mtag + 1, sizeof(struct pf_mtag));
 945         m_tag_prepend(m, mtag);
 946
 947         return ((struct pf_mtag *)(mtag + 1));
 948 }
 949
 950 static int
 951 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 952     struct pf_state *s)
 953 {
 954         struct pf_keyhash       *khs, *khw, *kh;
 955         struct pf_state_key     *sk, *cur;
 956         struct pf_state         *si, *olds = NULL;
 957         int idx;
 958
 959         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 960         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 961         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 962
 963         /*
 964          * We need to lock hash slots of both keys. To avoid deadlock
 965          * we always lock the slot with lower address first. Unlock order
 966          * isn't important.
 967          *
 968          * We also need to lock ID hash slot before dropping key
 969          * locks. On success we return with ID hash slot locked.
 970          */
 971
 972         if (skw == sks) {
 973                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 974                 PF_HASHROW_LOCK(khs);
 975         } else {
 976                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 977                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 978                 if (khs == khw) {
 979                         PF_HASHROW_LOCK(khs);
 980                 } else if (khs < khw) {
 981                         PF_HASHROW_LOCK(khs);
 982                         PF_HASHROW_LOCK(khw);
 983                 } else {
 984                         PF_HASHROW_LOCK(khw);
 985                         PF_HASHROW_LOCK(khs);
 986                 }
 987         }
 988
 989 #define KEYS_UNLOCK()   do {                    \
 990         if (khs != khw) {                       \
 991                 PF_HASHROW_UNLOCK(khs);         \
 992                 PF_HASHROW_UNLOCK(khw);         \
 993         } else                                  \
 994                 PF_HASHROW_UNLOCK(khs);         \
 995 } while (0)
 996
 997         /*
 998          * First run: start with wire key.
 999          */
1000         sk = skw;
1001         kh = khw;
1002         idx = PF_SK_WIRE;
1003
1004 keyattach:
1005         LIST_FOREACH(cur, &kh->keys, entry)
1006                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1007                         break;
1008
1009         if (cur != NULL) {
1010                 /* Key exists. Check for same kif, if none, add to key. */
1011                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1012                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1013
1014                         PF_HASHROW_LOCK(ih);
1015                         if (si->kif == s->kif &&
1016                             si->direction == s->direction) {
1017                                 if (sk->proto == IPPROTO_TCP &&
1018                                     si->src.state >= TCPS_FIN_WAIT_2 &&
1019                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1020                                         /*
1021                                          * New state matches an old >FIN_WAIT_2
1022                                          * state. We can't drop key hash locks,
1023                                          * thus we can't unlink it properly.
1024                                          *
1025                                          * As a workaround we drop it into
1026                                          * TCPS_CLOSED state, schedule purge
1027                                          * ASAP and push it into the very end
1028                                          * of the slot TAILQ, so that it won't
1029                                          * conflict with our new state.
1030                                          */
1031                                         si->src.state = si->dst.state =
1032                                             TCPS_CLOSED;
1033                                         si->timeout = PFTM_PURGE;
1034                                         olds = si;
1035                                 } else {
1036                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1037                                                 printf("pf: %s key attach "
1038                                                     "failed on %s: ",
1039                                                     (idx == PF_SK_WIRE) ?
1040                                                     "wire" : "stack",
1041                                                     s->kif->pfik_name);
1042                                                 pf_print_state_parts(s,
1043                                                     (idx == PF_SK_WIRE) ?
1044                                                     sk : NULL,
1045                                                     (idx == PF_SK_STACK) ?
1046                                                     sk : NULL);
1047                                                 printf(", existing: ");
1048                                                 pf_print_state_parts(si,
1049                                                     (idx == PF_SK_WIRE) ?
1050                                                     sk : NULL,
1051                                                     (idx == PF_SK_STACK) ?
1052                                                     sk : NULL);
1053                                                 printf("\n");
1054                                         }
1055                                         PF_HASHROW_UNLOCK(ih);
1056                                         KEYS_UNLOCK();
1057                                         uma_zfree(V_pf_state_key_z, sk);
1058                                         if (idx == PF_SK_STACK)
1059                                                 pf_detach_state(s);
1060                                         return (EEXIST); /* collision! */
1061                                 }
1062                         }
1063                         PF_HASHROW_UNLOCK(ih);
1064                 }
1065                 uma_zfree(V_pf_state_key_z, sk);
1066                 s->key[idx] = cur;
1067         } else {
1068                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1069                 s->key[idx] = sk;
1070         }
1071
1072 stateattach:
1073         /* List is sorted, if-bound states before floating. */
1074         if (s->kif == V_pfi_all)
1075                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1076         else
1077                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1078
1079         if (olds) {
1080                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1081                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1082                     key_list[idx]);
1083                 olds = NULL;
1084         }
1085
1086         /*
1087          * Attach done. See how should we (or should not?)
1088          * attach a second key.
1089          */
1090         if (sks == skw) {
1091                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1092                 idx = PF_SK_STACK;
1093                 sks = NULL;
1094                 goto stateattach;
1095         } else if (sks != NULL) {
1096                 /*
1097                  * Continue attaching with stack key.
1098                  */
1099                 sk = sks;
1100                 kh = khs;
1101                 idx = PF_SK_STACK;
1102                 sks = NULL;
1103                 goto keyattach;
1104         }
1105
1106         PF_STATE_LOCK(s);
1107         KEYS_UNLOCK();
1108
1109         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1110             ("%s failure", __func__));
1111
1112         return (0);
1113 #undef  KEYS_UNLOCK
1114 }
1115
1116 static void
1117 pf_detach_state(struct pf_state *s)
1118 {
1119         struct pf_state_key *sks = s->key[PF_SK_STACK];
1120         struct pf_keyhash *kh;
1121
1122         if (sks != NULL) {
1123                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1124                 PF_HASHROW_LOCK(kh);
1125                 if (s->key[PF_SK_STACK] != NULL)
1126                         pf_state_key_detach(s, PF_SK_STACK);
1127                 /*
1128                  * If both point to same key, then we are done.
1129                  */
1130                 if (sks == s->key[PF_SK_WIRE]) {
1131                         pf_state_key_detach(s, PF_SK_WIRE);
1132                         PF_HASHROW_UNLOCK(kh);
1133                         return;
1134                 }
1135                 PF_HASHROW_UNLOCK(kh);
1136         }
1137
1138         if (s->key[PF_SK_WIRE] != NULL) {
1139                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1140                 PF_HASHROW_LOCK(kh);
1141                 if (s->key[PF_SK_WIRE] != NULL)
1142                         pf_state_key_detach(s, PF_SK_WIRE);
1143                 PF_HASHROW_UNLOCK(kh);
1144         }
1145 }
1146
1147 static void
1148 pf_state_key_detach(struct pf_state *s, int idx)
1149 {
1150         struct pf_state_key *sk = s->key[idx];
1151 #ifdef INVARIANTS
1152         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1153
1154         PF_HASHROW_ASSERT(kh);
1155 #endif
1156         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1157         s->key[idx] = NULL;
1158
1159         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1160                 LIST_REMOVE(sk, entry);
1161                 uma_zfree(V_pf_state_key_z, sk);
1162         }
1163 }
1164
1165 static int
1166 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1167 {
1168         struct pf_state_key *sk = mem;
1169
1170         bzero(sk, sizeof(struct pf_state_key_cmp));
1171         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1172         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1173
1174         return (0);
1175 }
1176
1177 struct pf_state_key *
1178 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1179         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1180 {
1181         struct pf_state_key *sk;
1182
1183         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1184         if (sk == NULL)
1185                 return (NULL);
1186
1187         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1188         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1189         sk->port[pd->sidx] = sport;
1190         sk->port[pd->didx] = dport;
1191         sk->proto = pd->proto;
1192         sk->af = pd->af;
1193
1194         return (sk);
1195 }
1196
1197 struct pf_state_key *
1198 pf_state_key_clone(struct pf_state_key *orig)
1199 {
1200         struct pf_state_key *sk;
1201
1202         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1203         if (sk == NULL)
1204                 return (NULL);
1205
1206         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1207
1208         return (sk);
1209 }
1210
1211 int
1212 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1213     struct pf_state_key *sks, struct pf_state *s)
1214 {
1215         struct pf_idhash *ih;
1216         struct pf_state *cur;
1217         int error;
1218
1219         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1220             ("%s: sks not pristine", __func__));
1221         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1222             ("%s: skw not pristine", __func__));
1223         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1224
1225         s->kif = kif;
1226
1227         if (s->id == 0 && s->creatorid == 0) {
1228                 /* XXX: should be atomic, but probability of collision low */
1229                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1230                         V_pf_stateid[curcpu] = 1;
1231                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1232                 s->id = htobe64(s->id);
1233                 s->creatorid = V_pf_status.hostid;
1234         }
1235
1236         /* Returns with ID locked on success. */
1237         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1238                 return (error);
1239
1240         ih = &V_pf_idhash[PF_IDHASH(s)];
1241         PF_HASHROW_ASSERT(ih);
1242         LIST_FOREACH(cur, &ih->states, entry)
1243                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1244                         break;
1245
1246         if (cur != NULL) {
1247                 PF_HASHROW_UNLOCK(ih);
1248                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1249                         printf("pf: state ID collision: "
1250                             "id: %016llx creatorid: %08x\n",
1251                             (unsigned long long)be64toh(s->id),
1252                             ntohl(s->creatorid));
1253                 }
1254                 pf_detach_state(s);
1255                 return (EEXIST);
1256         }
1257         LIST_INSERT_HEAD(&ih->states, s, entry);
1258         /* One for keys, one for ID hash. */
1259         refcount_init(&s->refs, 2);
1260
1261         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1262         if (pfsync_insert_state_ptr != NULL)
1263                 pfsync_insert_state_ptr(s);
1264
1265         /* Returns locked. */
1266         return (0);
1267 }
1268
1269 /*
1270  * Find state by ID: returns with locked row on success.
1271  */
1272 struct pf_state *
1273 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1274 {
1275         struct pf_idhash *ih;
1276         struct pf_state *s;
1277
1278         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1279
1280         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1281
1282         PF_HASHROW_LOCK(ih);
1283         LIST_FOREACH(s, &ih->states, entry)
1284                 if (s->id == id && s->creatorid == creatorid)
1285                         break;
1286
1287         if (s == NULL)
1288                 PF_HASHROW_UNLOCK(ih);
1289
1290         return (s);
1291 }
1292
1293 /*
1294  * Find state by key.
1295  * Returns with ID hash slot locked on success.
1296  */
1297 static struct pf_state *
1298 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1299 {
1300         struct pf_keyhash       *kh;
1301         struct pf_state_key     *sk;
1302         struct pf_state         *s;
1303         int idx;
1304
1305         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1306
1307         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1308
1309         PF_HASHROW_LOCK(kh);
1310         LIST_FOREACH(sk, &kh->keys, entry)
1311                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1312                         break;
1313         if (sk == NULL) {
1314                 PF_HASHROW_UNLOCK(kh);
1315                 return (NULL);
1316         }
1317
1318         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1319
1320         /* List is sorted, if-bound states before floating ones. */
1321         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1322                 if (s->kif == V_pfi_all || s->kif == kif) {
1323                         PF_STATE_LOCK(s);
1324                         PF_HASHROW_UNLOCK(kh);
1325                         if (s->timeout >= PFTM_MAX) {
1326                                 /*
1327                                  * State is either being processed by
1328                                  * pf_unlink_state() in an other thread, or
1329                                  * is scheduled for immediate expiry.
1330                                  */
1331                                 PF_STATE_UNLOCK(s);
1332                                 return (NULL);
1333                         }
1334                         return (s);
1335                 }
1336         PF_HASHROW_UNLOCK(kh);
1337
1338         return (NULL);
1339 }
1340
1341 struct pf_state *
1342 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1343 {
1344         struct pf_keyhash       *kh;
1345         struct pf_state_key     *sk;
1346         struct pf_state         *s, *ret = NULL;
1347         int                      idx, inout = 0;
1348
1349         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1350
1351         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1352
1353         PF_HASHROW_LOCK(kh);
1354         LIST_FOREACH(sk, &kh->keys, entry)
1355                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1356                         break;
1357         if (sk == NULL) {
1358                 PF_HASHROW_UNLOCK(kh);
1359                 return (NULL);
1360         }
1361         switch (dir) {
1362         case PF_IN:
1363                 idx = PF_SK_WIRE;
1364                 break;
1365         case PF_OUT:
1366                 idx = PF_SK_STACK;
1367                 break;
1368         case PF_INOUT:
1369                 idx = PF_SK_WIRE;
1370                 inout = 1;
1371                 break;
1372         default:
1373                 panic("%s: dir %u", __func__, dir);
1374         }
1375 second_run:
1376         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1377                 if (more == NULL) {
1378                         PF_HASHROW_UNLOCK(kh);
1379                         return (s);
1380                 }
1381
1382                 if (ret)
1383                         (*more)++;
1384                 else
1385                         ret = s;
1386         }
1387         if (inout == 1) {
1388                 inout = 0;
1389                 idx = PF_SK_STACK;
1390                 goto second_run;
1391         }
1392         PF_HASHROW_UNLOCK(kh);
1393
1394         return (ret);
1395 }
1396
1397 /* END state table stuff */
1398
1399 static void
1400 pf_send(struct pf_send_entry *pfse)
1401 {
1402
1403         PF_SENDQ_LOCK();
1404         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1405         PF_SENDQ_UNLOCK();
1406         swi_sched(V_pf_swi_cookie, 0);
1407 }
1408
1409 void
1410 pf_intr(void *v)
1411 {
1412         struct pf_send_head queue;
1413         struct pf_send_entry *pfse, *next;
1414
1415         CURVNET_SET((struct vnet *)v);
1416
1417         PF_SENDQ_LOCK();
1418         queue = V_pf_sendqueue;
1419         STAILQ_INIT(&V_pf_sendqueue);
1420         PF_SENDQ_UNLOCK();
1421
1422         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1423                 switch (pfse->pfse_type) {
1424 #ifdef INET
1425                 case PFSE_IP:
1426                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1427                         break;
1428                 case PFSE_ICMP:
1429                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1430                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1431                         break;
1432 #endif /* INET */
1433 #ifdef INET6
1434                 case PFSE_IP6:
1435                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1436                             NULL);
1437                         break;
1438                 case PFSE_ICMP6:
1439                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1440                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1441                         break;
1442 #endif /* INET6 */
1443                 default:
1444                         panic("%s: unknown type", __func__);
1445                 }
1446                 free(pfse, M_PFTEMP);
1447         }
1448         CURVNET_RESTORE();
1449 }
1450
1451 void
1452 pf_purge_thread(void *unused __unused)
1453 {
1454         VNET_ITERATOR_DECL(vnet_iter);
1455         u_int idx = 0;
1456
1457         for (;;) {
1458                 PF_RULES_RLOCK();
1459                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1460                 PF_RULES_RUNLOCK();
1461
1462                 VNET_LIST_RLOCK();
1463                 VNET_FOREACH(vnet_iter) {
1464                         CURVNET_SET(vnet_iter);
1465
1466                 if (pf_end_threads) {
1467                         pf_end_threads++;
1468                         wakeup(pf_purge_thread);
1469                         kproc_exit(0);
1470                 }
1471
1472                 /* Wait while V_pf_default_rule.timeout is initialized. */
1473                 if (V_pf_vnet_active == 0) {
1474                         CURVNET_RESTORE();
1475                         continue;
1476                 }
1477
1478                 /* Process 1/interval fraction of the state table every run. */
1479                 idx = pf_purge_expired_states(idx, pf_hashmask /
1480                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1481
1482                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1483                 if (idx == 0) {
1484                         /*
1485                          * Order is important:
1486                          * - states and src nodes reference rules
1487                          * - states and rules reference kifs
1488                          */
1489                         pf_purge_expired_fragments();
1490                         pf_purge_expired_src_nodes();
1491                         pf_purge_unlinked_rules();
1492                         pfi_kif_purge();
1493                 }
1494                 CURVNET_RESTORE();
1495                 }
1496                 VNET_LIST_RUNLOCK();
1497         }
1498         /* not reached */
1499 }
1500
1501 void
1502 pf_unload_vnet_purge(void)
1503 {
1504
1505         /*
1506          * To cleanse up all kifs and rules we need
1507          * two runs: first one clears reference flags,
1508          * then pf_purge_expired_states() doesn't
1509          * raise them, and then second run frees.
1510          */
1511         pf_purge_unlinked_rules();
1512         pfi_kif_purge();
1513
1514         /*
1515          * Now purge everything.
1516          */
1517         pf_purge_expired_states(0, pf_hashmask);
1518         pf_purge_expired_fragments();
1519         pf_purge_expired_src_nodes();
1520
1521         /*
1522          * Now all kifs & rules should be unreferenced,
1523          * thus should be successfully freed.
1524          */
1525         pf_purge_unlinked_rules();
1526         pfi_kif_purge();
1527 }
1528
1529
1530 u_int32_t
1531 pf_state_expires(const struct pf_state *state)
1532 {
1533         u_int32_t       timeout;
1534         u_int32_t       start;
1535         u_int32_t       end;
1536         u_int32_t       states;
1537
1538         /* handle all PFTM_* > PFTM_MAX here */
1539         if (state->timeout == PFTM_PURGE)
1540                 return (time_uptime);
1541         KASSERT(state->timeout != PFTM_UNLINKED,
1542             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1543         KASSERT((state->timeout < PFTM_MAX),
1544             ("pf_state_expires: timeout > PFTM_MAX"));
1545         timeout = state->rule.ptr->timeout[state->timeout];
1546         if (!timeout)
1547                 timeout = V_pf_default_rule.timeout[state->timeout];
1548         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1549         if (start) {
1550                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1551                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1552         } else {
1553                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1554                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1555                 states = V_pf_status.states;
1556         }
1557         if (end && states > start && start < end) {
1558                 if (states < end)
1559                         return (state->expire + timeout * (end - states) /
1560                             (end - start));
1561                 else
1562                         return (time_uptime);
1563         }
1564         return (state->expire + timeout);
1565 }
1566
1567 void
1568 pf_purge_expired_src_nodes()
1569 {
1570         struct pf_src_node_list  freelist;
1571         struct pf_srchash       *sh;
1572         struct pf_src_node      *cur, *next;
1573         int i;
1574
1575         LIST_INIT(&freelist);
1576         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1577             PF_HASHROW_LOCK(sh);
1578             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1579                 if (cur->states == 0 && cur->expire <= time_uptime) {
1580                         pf_unlink_src_node(cur);
1581                         LIST_INSERT_HEAD(&freelist, cur, entry);
1582                 } else if (cur->rule.ptr != NULL)
1583                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1584             PF_HASHROW_UNLOCK(sh);
1585         }
1586
1587         pf_free_src_nodes(&freelist);
1588
1589         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1590 }
1591
1592 static void
1593 pf_src_tree_remove_state(struct pf_state *s)
1594 {
1595         struct pf_src_node *sn;
1596         struct pf_srchash *sh;
1597         uint32_t timeout;
1598
1599         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1600             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1601             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1602
1603         if (s->src_node != NULL) {
1604                 sn = s->src_node;
1605                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1606                 PF_HASHROW_LOCK(sh);
1607                 if (s->src.tcp_est)
1608                         --sn->conn;
1609                 if (--sn->states == 0)
1610                         sn->expire = time_uptime + timeout;
1611                 PF_HASHROW_UNLOCK(sh);
1612         }
1613         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1614                 sn = s->nat_src_node;
1615                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1616                 PF_HASHROW_LOCK(sh);
1617                 if (--sn->states == 0)
1618                         sn->expire = time_uptime + timeout;
1619                 PF_HASHROW_UNLOCK(sh);
1620         }
1621         s->src_node = s->nat_src_node = NULL;
1622 }
1623
1624 /*
1625  * Unlink and potentilly free a state. Function may be
1626  * called with ID hash row locked, but always returns
1627  * unlocked, since it needs to go through key hash locking.
1628  */
1629 int
1630 pf_unlink_state(struct pf_state *s, u_int flags)
1631 {
1632         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1633
1634         if ((flags & PF_ENTER_LOCKED) == 0)
1635                 PF_HASHROW_LOCK(ih);
1636         else
1637                 PF_HASHROW_ASSERT(ih);
1638
1639         if (s->timeout == PFTM_UNLINKED) {
1640                 /*
1641                  * State is being processed
1642                  * by pf_unlink_state() in
1643                  * an other thread.
1644                  */
1645                 PF_HASHROW_UNLOCK(ih);
1646                 return (0);     /* XXXGL: undefined actually */
1647         }
1648
1649         if (s->src.state == PF_TCPS_PROXY_DST) {
1650                 /* XXX wire key the right one? */
1651                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1652                     &s->key[PF_SK_WIRE]->addr[1],
1653                     &s->key[PF_SK_WIRE]->addr[0],
1654                     s->key[PF_SK_WIRE]->port[1],
1655                     s->key[PF_SK_WIRE]->port[0],
1656                     s->src.seqhi, s->src.seqlo + 1,
1657                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1658         }
1659
1660         LIST_REMOVE(s, entry);
1661         pf_src_tree_remove_state(s);
1662
1663         if (pfsync_delete_state_ptr != NULL)
1664                 pfsync_delete_state_ptr(s);
1665
1666         STATE_DEC_COUNTERS(s);
1667
1668         s->timeout = PFTM_UNLINKED;
1669
1670         PF_HASHROW_UNLOCK(ih);
1671
1672         pf_detach_state(s);
1673         refcount_release(&s->refs);
1674
1675         return (pf_release_state(s));
1676 }
1677
1678 void
1679 pf_free_state(struct pf_state *cur)
1680 {
1681
1682         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1683         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1684             cur->timeout));
1685
1686         pf_normalize_tcp_cleanup(cur);
1687         uma_zfree(V_pf_state_z, cur);
1688         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1689 }
1690
1691 /*
1692  * Called only from pf_purge_thread(), thus serialized.
1693  */
1694 static u_int
1695 pf_purge_expired_states(u_int i, int maxcheck)
1696 {
1697         struct pf_idhash *ih;
1698         struct pf_state *s;
1699
1700         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1701
1702         /*
1703          * Go through hash and unlink states that expire now.
1704          */
1705         while (maxcheck > 0) {
1706
1707                 ih = &V_pf_idhash[i];
1708 relock:
1709                 PF_HASHROW_LOCK(ih);
1710                 LIST_FOREACH(s, &ih->states, entry) {
1711                         if (pf_state_expires(s) <= time_uptime) {
1712                                 V_pf_status.states -=
1713                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1714                                 goto relock;
1715                         }
1716                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1717                         if (s->nat_rule.ptr != NULL)
1718                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1719                         if (s->anchor.ptr != NULL)
1720                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1721                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1722                         if (s->rt_kif)
1723                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1724                 }
1725                 PF_HASHROW_UNLOCK(ih);
1726
1727                 /* Return when we hit end of hash. */
1728                 if (++i > pf_hashmask) {
1729                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1730                         return (0);
1731                 }
1732
1733                 maxcheck--;
1734         }
1735
1736         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1737
1738         return (i);
1739 }
1740
1741 static void
1742 pf_purge_unlinked_rules()
1743 {
1744         struct pf_rulequeue tmpq;
1745         struct pf_rule *r, *r1;
1746
1747         /*
1748          * If we have overloading task pending, then we'd
1749          * better skip purging this time. There is a tiny
1750          * probability that overloading task references
1751          * an already unlinked rule.
1752          */
1753         PF_OVERLOADQ_LOCK();
1754         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1755                 PF_OVERLOADQ_UNLOCK();
1756                 return;
1757         }
1758         PF_OVERLOADQ_UNLOCK();
1759
1760         /*
1761          * Do naive mark-and-sweep garbage collecting of old rules.
1762          * Reference flag is raised by pf_purge_expired_states()
1763          * and pf_purge_expired_src_nodes().
1764          *
1765          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1766          * use a temporary queue.
1767          */
1768         TAILQ_INIT(&tmpq);
1769         PF_UNLNKDRULES_LOCK();
1770         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1771                 if (!(r->rule_flag & PFRULE_REFS)) {
1772                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1773                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1774                 } else
1775                         r->rule_flag &= ~PFRULE_REFS;
1776         }
1777         PF_UNLNKDRULES_UNLOCK();
1778
1779         if (!TAILQ_EMPTY(&tmpq)) {
1780                 PF_RULES_WLOCK();
1781                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1782                         TAILQ_REMOVE(&tmpq, r, entries);
1783                         pf_free_rule(r);
1784                 }
1785                 PF_RULES_WUNLOCK();
1786         }
1787 }
1788
1789 void
1790 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1791 {
1792         switch (af) {
1793 #ifdef INET
1794         case AF_INET: {
1795                 u_int32_t a = ntohl(addr->addr32[0]);
1796                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1797                     (a>>8)&255, a&255);
1798                 if (p) {
1799                         p = ntohs(p);
1800                         printf(":%u", p);
1801                 }
1802                 break;
1803         }
1804 #endif /* INET */
1805 #ifdef INET6
1806         case AF_INET6: {
1807                 u_int16_t b;
1808                 u_int8_t i, curstart, curend, maxstart, maxend;
1809                 curstart = curend = maxstart = maxend = 255;
1810                 for (i = 0; i < 8; i++) {
1811                         if (!addr->addr16[i]) {
1812                                 if (curstart == 255)
1813                                         curstart = i;
1814                                 curend = i;
1815                         } else {
1816                                 if ((curend - curstart) >
1817                                     (maxend - maxstart)) {
1818                                         maxstart = curstart;
1819                                         maxend = curend;
1820                                 }
1821                                 curstart = curend = 255;
1822                         }
1823                 }
1824                 if ((curend - curstart) >
1825                     (maxend - maxstart)) {
1826                         maxstart = curstart;
1827                         maxend = curend;
1828                 }
1829                 for (i = 0; i < 8; i++) {
1830                         if (i >= maxstart && i <= maxend) {
1831                                 if (i == 0)
1832                                         printf(":");
1833                                 if (i == maxend)
1834                                         printf(":");
1835                         } else {
1836                                 b = ntohs(addr->addr16[i]);
1837                                 printf("%x", b);
1838                                 if (i < 7)
1839                                         printf(":");
1840                         }
1841                 }
1842                 if (p) {
1843                         p = ntohs(p);
1844                         printf("[%u]", p);
1845                 }
1846                 break;
1847         }
1848 #endif /* INET6 */
1849         }
1850 }
1851
1852 void
1853 pf_print_state(struct pf_state *s)
1854 {
1855         pf_print_state_parts(s, NULL, NULL);
1856 }
1857
1858 static void
1859 pf_print_state_parts(struct pf_state *s,
1860     struct pf_state_key *skwp, struct pf_state_key *sksp)
1861 {
1862         struct pf_state_key *skw, *sks;
1863         u_int8_t proto, dir;
1864
1865         /* Do our best to fill these, but they're skipped if NULL */
1866         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1867         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1868         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1869         dir = s ? s->direction : 0;
1870
1871         switch (proto) {
1872         case IPPROTO_IPV4:
1873                 printf("IPv4");
1874                 break;
1875         case IPPROTO_IPV6:
1876                 printf("IPv6");
1877                 break;
1878         case IPPROTO_TCP:
1879                 printf("TCP");
1880                 break;
1881         case IPPROTO_UDP:
1882                 printf("UDP");
1883                 break;
1884         case IPPROTO_ICMP:
1885                 printf("ICMP");
1886                 break;
1887         case IPPROTO_ICMPV6:
1888                 printf("ICMPv6");
1889                 break;
1890         default:
1891                 printf("%u", proto);
1892                 break;
1893         }
1894         switch (dir) {
1895         case PF_IN:
1896                 printf(" in");
1897                 break;
1898         case PF_OUT:
1899                 printf(" out");
1900                 break;
1901         }
1902         if (skw) {
1903                 printf(" wire: ");
1904                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1905                 printf(" ");
1906                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1907         }
1908         if (sks) {
1909                 printf(" stack: ");
1910                 if (sks != skw) {
1911                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1912                         printf(" ");
1913                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1914                 } else
1915                         printf("-");
1916         }
1917         if (s) {
1918                 if (proto == IPPROTO_TCP) {
1919                         printf(" [lo=%u high=%u win=%u modulator=%u",
1920                             s->src.seqlo, s->src.seqhi,
1921                             s->src.max_win, s->src.seqdiff);
1922                         if (s->src.wscale && s->dst.wscale)
1923                                 printf(" wscale=%u",
1924                                     s->src.wscale & PF_WSCALE_MASK);
1925                         printf("]");
1926                         printf(" [lo=%u high=%u win=%u modulator=%u",
1927                             s->dst.seqlo, s->dst.seqhi,
1928                             s->dst.max_win, s->dst.seqdiff);
1929                         if (s->src.wscale && s->dst.wscale)
1930                                 printf(" wscale=%u",
1931                                 s->dst.wscale & PF_WSCALE_MASK);
1932                         printf("]");
1933                 }
1934                 printf(" %u:%u", s->src.state, s->dst.state);
1935         }
1936 }
1937
1938 void
1939 pf_print_flags(u_int8_t f)
1940 {
1941         if (f)
1942                 printf(" ");
1943         if (f & TH_FIN)
1944                 printf("F");
1945         if (f & TH_SYN)
1946                 printf("S");
1947         if (f & TH_RST)
1948                 printf("R");
1949         if (f & TH_PUSH)
1950                 printf("P");
1951         if (f & TH_ACK)
1952                 printf("A");
1953         if (f & TH_URG)
1954                 printf("U");
1955         if (f & TH_ECE)
1956                 printf("E");
1957         if (f & TH_CWR)
1958                 printf("W");
1959 }
1960
1961 #define PF_SET_SKIP_STEPS(i)                                    \
1962         do {                                                    \
1963                 while (head[i] != cur) {                        \
1964                         head[i]->skip[i].ptr = cur;             \
1965                         head[i] = TAILQ_NEXT(head[i], entries); \
1966                 }                                               \
1967         } while (0)
1968
1969 void
1970 pf_calc_skip_steps(struct pf_rulequeue *rules)
1971 {
1972         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1973         int i;
1974
1975         cur = TAILQ_FIRST(rules);
1976         prev = cur;
1977         for (i = 0; i < PF_SKIP_COUNT; ++i)
1978                 head[i] = cur;
1979         while (cur != NULL) {
1980
1981                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1982                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1983                 if (cur->direction != prev->direction)
1984                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1985                 if (cur->af != prev->af)
1986                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1987                 if (cur->proto != prev->proto)
1988                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1989                 if (cur->src.neg != prev->src.neg ||
1990                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1991                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1992                 if (cur->src.port[0] != prev->src.port[0] ||
1993                     cur->src.port[1] != prev->src.port[1] ||
1994                     cur->src.port_op != prev->src.port_op)
1995                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1996                 if (cur->dst.neg != prev->dst.neg ||
1997                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1998                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1999                 if (cur->dst.port[0] != prev->dst.port[0] ||
2000                     cur->dst.port[1] != prev->dst.port[1] ||
2001                     cur->dst.port_op != prev->dst.port_op)
2002                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2003
2004                 prev = cur;
2005                 cur = TAILQ_NEXT(cur, entries);
2006         }
2007         for (i = 0; i < PF_SKIP_COUNT; ++i)
2008                 PF_SET_SKIP_STEPS(i);
2009 }
2010
2011 static int
2012 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2013 {
2014         if (aw1->type != aw2->type)
2015                 return (1);
2016         switch (aw1->type) {
2017         case PF_ADDR_ADDRMASK:
2018         case PF_ADDR_RANGE:
2019                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2020                         return (1);
2021                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2022                         return (1);
2023                 return (0);
2024         case PF_ADDR_DYNIFTL:
2025                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2026         case PF_ADDR_NOROUTE:
2027         case PF_ADDR_URPFFAILED:
2028                 return (0);
2029         case PF_ADDR_TABLE:
2030                 return (aw1->p.tbl != aw2->p.tbl);
2031         default:
2032                 printf("invalid address type: %d\n", aw1->type);
2033                 return (1);
2034         }
2035 }
2036
2037 /**
2038  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2039  * header isn't always a full checksum. In some cases (i.e. output) it's a
2040  * pseudo-header checksum, which is a partial checksum over src/dst IP
2041  * addresses, protocol number and length.
2042  *
2043  * That means we have the following cases:
2044  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2045  *      checksums, we need to update the checksum whenever we change anything.
2046  *  * Output (i.e. the checksum is a pseudo-header checksum):
2047  *      x The field being updated is src/dst address or affects the length of
2048  *      the packet. We need to update the pseudo-header checksum (note that this
2049  *      checksum is not ones' complement).
2050  *      x Some other field is being modified (e.g. src/dst port numbers): We
2051  *      don't have to update anything.
2052  **/
2053 u_int16_t
2054 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2055 {
2056         u_int32_t       l;
2057
2058         if (udp && !cksum)
2059                 return (0x0000);
2060         l = cksum + old - new;
2061         l = (l >> 16) + (l & 65535);
2062         l = l & 65535;
2063         if (udp && !l)
2064                 return (0xFFFF);
2065         return (l);
2066 }
2067
2068 u_int16_t
2069 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2070         u_int16_t new, u_int8_t udp)
2071 {
2072         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2073                 return (cksum);
2074
2075         return (pf_cksum_fixup(cksum, old, new, udp));
2076 }
2077
2078 static void
2079 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2080         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2081         sa_family_t af)
2082 {
2083         struct pf_addr  ao;
2084         u_int16_t       po = *p;
2085
2086         PF_ACPY(&ao, a, af);
2087         PF_ACPY(a, an, af);
2088
2089         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2090                 *pc = ~*pc;
2091
2092         *p = pn;
2093
2094         switch (af) {
2095 #ifdef INET
2096         case AF_INET:
2097                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2098                     ao.addr16[0], an->addr16[0], 0),
2099                     ao.addr16[1], an->addr16[1], 0);
2100                 *p = pn;
2101
2102                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2103                     ao.addr16[0], an->addr16[0], u),
2104                     ao.addr16[1], an->addr16[1], u);
2105
2106                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2107                 break;
2108 #endif /* INET */
2109 #ifdef INET6
2110         case AF_INET6:
2111                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2112                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2113                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2114                     ao.addr16[0], an->addr16[0], u),
2115                     ao.addr16[1], an->addr16[1], u),
2116                     ao.addr16[2], an->addr16[2], u),
2117                     ao.addr16[3], an->addr16[3], u),
2118                     ao.addr16[4], an->addr16[4], u),
2119                     ao.addr16[5], an->addr16[5], u),
2120                     ao.addr16[6], an->addr16[6], u),
2121                     ao.addr16[7], an->addr16[7], u);
2122
2123                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2124                 break;
2125 #endif /* INET6 */
2126         }
2127
2128         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2129             CSUM_DELAY_DATA_IPV6)) {
2130                 *pc = ~*pc;
2131                 if (! *pc)
2132                         *pc = 0xffff;
2133         }
2134 }
2135
2136 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2137 void
2138 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2139 {
2140         u_int32_t       ao;
2141
2142         memcpy(&ao, a, sizeof(ao));
2143         memcpy(a, &an, sizeof(u_int32_t));
2144         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2145             ao % 65536, an % 65536, u);
2146 }
2147
2148 void
2149 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2150 {
2151         u_int32_t       ao;
2152
2153         memcpy(&ao, a, sizeof(ao));
2154         memcpy(a, &an, sizeof(u_int32_t));
2155
2156         *c = pf_proto_cksum_fixup(m,
2157             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2158             ao % 65536, an % 65536, udp);
2159 }
2160
2161 #ifdef INET6
2162 static void
2163 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2164 {
2165         struct pf_addr  ao;
2166
2167         PF_ACPY(&ao, a, AF_INET6);
2168         PF_ACPY(a, an, AF_INET6);
2169
2170         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2171             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2172             pf_cksum_fixup(pf_cksum_fixup(*c,
2173             ao.addr16[0], an->addr16[0], u),
2174             ao.addr16[1], an->addr16[1], u),
2175             ao.addr16[2], an->addr16[2], u),
2176             ao.addr16[3], an->addr16[3], u),
2177             ao.addr16[4], an->addr16[4], u),
2178             ao.addr16[5], an->addr16[5], u),
2179             ao.addr16[6], an->addr16[6], u),
2180             ao.addr16[7], an->addr16[7], u);
2181 }
2182 #endif /* INET6 */
2183
2184 static void
2185 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2186     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2187     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2188 {
2189         struct pf_addr  oia, ooa;
2190
2191         PF_ACPY(&oia, ia, af);
2192         if (oa)
2193                 PF_ACPY(&ooa, oa, af);
2194
2195         /* Change inner protocol port, fix inner protocol checksum. */
2196         if (ip != NULL) {
2197                 u_int16_t       oip = *ip;
2198                 u_int32_t       opc;
2199
2200                 if (pc != NULL)
2201                         opc = *pc;
2202                 *ip = np;
2203                 if (pc != NULL)
2204                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2205                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2206                 if (pc != NULL)
2207                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2208         }
2209         /* Change inner ip address, fix inner ip and icmp checksums. */
2210         PF_ACPY(ia, na, af);
2211         switch (af) {
2212 #ifdef INET
2213         case AF_INET: {
2214                 u_int32_t        oh2c = *h2c;
2215
2216                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2217                     oia.addr16[0], ia->addr16[0], 0),
2218                     oia.addr16[1], ia->addr16[1], 0);
2219                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2220                     oia.addr16[0], ia->addr16[0], 0),
2221                     oia.addr16[1], ia->addr16[1], 0);
2222                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2223                 break;
2224         }
2225 #endif /* INET */
2226 #ifdef INET6
2227         case AF_INET6:
2228                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2229                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2230                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2231                     oia.addr16[0], ia->addr16[0], u),
2232                     oia.addr16[1], ia->addr16[1], u),
2233                     oia.addr16[2], ia->addr16[2], u),
2234                     oia.addr16[3], ia->addr16[3], u),
2235                     oia.addr16[4], ia->addr16[4], u),
2236                     oia.addr16[5], ia->addr16[5], u),
2237                     oia.addr16[6], ia->addr16[6], u),
2238                     oia.addr16[7], ia->addr16[7], u);
2239                 break;
2240 #endif /* INET6 */
2241         }
2242         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2243         if (oa) {
2244                 PF_ACPY(oa, na, af);
2245                 switch (af) {
2246 #ifdef INET
2247                 case AF_INET:
2248                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2249                             ooa.addr16[0], oa->addr16[0], 0),
2250                             ooa.addr16[1], oa->addr16[1], 0);
2251                         break;
2252 #endif /* INET */
2253 #ifdef INET6
2254                 case AF_INET6:
2255                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2256                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2257                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2258                             ooa.addr16[0], oa->addr16[0], u),
2259                             ooa.addr16[1], oa->addr16[1], u),
2260                             ooa.addr16[2], oa->addr16[2], u),
2261                             ooa.addr16[3], oa->addr16[3], u),
2262                             ooa.addr16[4], oa->addr16[4], u),
2263                             ooa.addr16[5], oa->addr16[5], u),
2264                             ooa.addr16[6], oa->addr16[6], u),
2265                             ooa.addr16[7], oa->addr16[7], u);
2266                         break;
2267 #endif /* INET6 */
2268                 }
2269         }
2270 }
2271
2272
2273 /*
2274  * Need to modulate the sequence numbers in the TCP SACK option
2275  * (credits to Krzysztof Pfaff for report and patch)
2276  */
2277 static int
2278 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2279     struct tcphdr *th, struct pf_state_peer *dst)
2280 {
2281         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2282         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2283         int copyback = 0, i, olen;
2284         struct sackblk sack;
2285
2286 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2287         if (hlen < TCPOLEN_SACKLEN ||
2288             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2289                 return 0;
2290
2291         while (hlen >= TCPOLEN_SACKLEN) {
2292                 olen = opt[1];
2293                 switch (*opt) {
2294                 case TCPOPT_EOL:        /* FALLTHROUGH */
2295                 case TCPOPT_NOP:
2296                         opt++;
2297                         hlen--;
2298                         break;
2299                 case TCPOPT_SACK:
2300                         if (olen > hlen)
2301                                 olen = hlen;
2302                         if (olen >= TCPOLEN_SACKLEN) {
2303                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2304                                     i += TCPOLEN_SACK) {
2305                                         memcpy(&sack, &opt[i], sizeof(sack));
2306                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2307                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2308                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2309                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2310                                         memcpy(&opt[i], &sack, sizeof(sack));
2311                                 }
2312                                 copyback = 1;
2313                         }
2314                         /* FALLTHROUGH */
2315                 default:
2316                         if (olen < 2)
2317                                 olen = 2;
2318                         hlen -= olen;
2319                         opt += olen;
2320                 }
2321         }
2322
2323         if (copyback)
2324                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2325         return (copyback);
2326 }
2327
2328 static void
2329 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2330     const struct pf_addr *saddr, const struct pf_addr *daddr,
2331     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2332     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2333     u_int16_t rtag, struct ifnet *ifp)
2334 {
2335         struct pf_send_entry *pfse;
2336         struct mbuf     *m;
2337         int              len, tlen;
2338 #ifdef INET
2339         struct ip       *h = NULL;
2340 #endif /* INET */
2341 #ifdef INET6
2342         struct ip6_hdr  *h6 = NULL;
2343 #endif /* INET6 */
2344         struct tcphdr   *th;
2345         char            *opt;
2346         struct pf_mtag  *pf_mtag;
2347
2348         len = 0;
2349         th = NULL;
2350
2351         /* maximum segment size tcp option */
2352         tlen = sizeof(struct tcphdr);
2353         if (mss)
2354                 tlen += 4;
2355
2356         switch (af) {
2357 #ifdef INET
2358         case AF_INET:
2359                 len = sizeof(struct ip) + tlen;
2360                 break;
2361 #endif /* INET */
2362 #ifdef INET6
2363         case AF_INET6:
2364                 len = sizeof(struct ip6_hdr) + tlen;
2365                 break;
2366 #endif /* INET6 */
2367         default:
2368                 panic("%s: unsupported af %d", __func__, af);
2369         }
2370
2371         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2372         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2373         if (pfse == NULL)
2374                 return;
2375         m = m_gethdr(M_NOWAIT, MT_DATA);
2376         if (m == NULL) {
2377                 free(pfse, M_PFTEMP);
2378                 return;
2379         }
2380 #ifdef MAC
2381         mac_netinet_firewall_send(m);
2382 #endif
2383         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2384                 free(pfse, M_PFTEMP);
2385                 m_freem(m);
2386                 return;
2387         }
2388         if (tag)
2389                 m->m_flags |= M_SKIP_FIREWALL;
2390         pf_mtag->tag = rtag;
2391
2392         if (r != NULL && r->rtableid >= 0)
2393                 M_SETFIB(m, r->rtableid);
2394
2395 #ifdef ALTQ
2396         if (r != NULL && r->qid) {
2397                 pf_mtag->qid = r->qid;
2398
2399                 /* add hints for ecn */
2400                 pf_mtag->hdr = mtod(m, struct ip *);
2401         }
2402 #endif /* ALTQ */
2403         m->m_data += max_linkhdr;
2404         m->m_pkthdr.len = m->m_len = len;
2405         m->m_pkthdr.rcvif = NULL;
2406         bzero(m->m_data, len);
2407         switch (af) {
2408 #ifdef INET
2409         case AF_INET:
2410                 h = mtod(m, struct ip *);
2411
2412                 /* IP header fields included in the TCP checksum */
2413                 h->ip_p = IPPROTO_TCP;
2414                 h->ip_len = htons(tlen);
2415                 h->ip_src.s_addr = saddr->v4.s_addr;
2416                 h->ip_dst.s_addr = daddr->v4.s_addr;
2417
2418                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2419                 break;
2420 #endif /* INET */
2421 #ifdef INET6
2422         case AF_INET6:
2423                 h6 = mtod(m, struct ip6_hdr *);
2424
2425                 /* IP header fields included in the TCP checksum */
2426                 h6->ip6_nxt = IPPROTO_TCP;
2427                 h6->ip6_plen = htons(tlen);
2428                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2429                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2430
2431                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2432                 break;
2433 #endif /* INET6 */
2434         }
2435
2436         /* TCP header */
2437         th->th_sport = sport;
2438         th->th_dport = dport;
2439         th->th_seq = htonl(seq);
2440         th->th_ack = htonl(ack);
2441         th->th_off = tlen >> 2;
2442         th->th_flags = flags;
2443         th->th_win = htons(win);
2444
2445         if (mss) {
2446                 opt = (char *)(th + 1);
2447                 opt[0] = TCPOPT_MAXSEG;
2448                 opt[1] = 4;
2449                 HTONS(mss);
2450                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2451         }
2452
2453         switch (af) {
2454 #ifdef INET
2455         case AF_INET:
2456                 /* TCP checksum */
2457                 th->th_sum = in_cksum(m, len);
2458
2459                 /* Finish the IP header */
2460                 h->ip_v = 4;
2461                 h->ip_hl = sizeof(*h) >> 2;
2462                 h->ip_tos = IPTOS_LOWDELAY;
2463                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2464                 h->ip_len = htons(len);
2465                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2466                 h->ip_sum = 0;
2467
2468                 pfse->pfse_type = PFSE_IP;
2469                 break;
2470 #endif /* INET */
2471 #ifdef INET6
2472         case AF_INET6:
2473                 /* TCP checksum */
2474                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2475                     sizeof(struct ip6_hdr), tlen);
2476
2477                 h6->ip6_vfc |= IPV6_VERSION;
2478                 h6->ip6_hlim = IPV6_DEFHLIM;
2479
2480                 pfse->pfse_type = PFSE_IP6;
2481                 break;
2482 #endif /* INET6 */
2483         }
2484         pfse->pfse_m = m;
2485         pf_send(pfse);
2486 }
2487
2488 static int
2489 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2490 {
2491         struct m_tag *mtag;
2492
2493         KASSERT(prio <= PF_PRIO_MAX,
2494             ("%s with invalid pcp", __func__));
2495
2496         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2497         if (mtag == NULL) {
2498                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2499                     sizeof(uint8_t), M_NOWAIT);
2500                 if (mtag == NULL)
2501                         return (ENOMEM);
2502                 m_tag_prepend(m, mtag);
2503         }
2504
2505         *(uint8_t *)(mtag + 1) = prio;
2506         return (0);
2507 }
2508
2509 static int
2510 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2511 {
2512         struct m_tag *mtag;
2513         u_int8_t mpcp;
2514
2515         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2516         if (mtag == NULL)
2517                 return (0);
2518
2519         if (prio == PF_PRIO_ZERO)
2520                 prio = 0;
2521
2522         mpcp = *(uint8_t *)(mtag + 1);
2523
2524         return (mpcp == prio);
2525 }
2526
2527 static void
2528 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2529     struct pf_rule *r)
2530 {
2531         struct pf_send_entry *pfse;
2532         struct mbuf *m0;
2533         struct pf_mtag *pf_mtag;
2534
2535         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2536         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2537         if (pfse == NULL)
2538                 return;
2539
2540         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2541                 free(pfse, M_PFTEMP);
2542                 return;
2543         }
2544
2545         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2546                 free(pfse, M_PFTEMP);
2547                 return;
2548         }
2549         /* XXX: revisit */
2550         m0->m_flags |= M_SKIP_FIREWALL;
2551
2552         if (r->rtableid >= 0)
2553                 M_SETFIB(m0, r->rtableid);
2554
2555 #ifdef ALTQ
2556         if (r->qid) {
2557                 pf_mtag->qid = r->qid;
2558                 /* add hints for ecn */
2559                 pf_mtag->hdr = mtod(m0, struct ip *);
2560         }
2561 #endif /* ALTQ */
2562
2563         switch (af) {
2564 #ifdef INET
2565         case AF_INET:
2566                 pfse->pfse_type = PFSE_ICMP;
2567                 break;
2568 #endif /* INET */
2569 #ifdef INET6
2570         case AF_INET6:
2571                 pfse->pfse_type = PFSE_ICMP6;
2572                 break;
2573 #endif /* INET6 */
2574         }
2575         pfse->pfse_m = m0;
2576         pfse->icmpopts.type = type;
2577         pfse->icmpopts.code = code;
2578         pf_send(pfse);
2579 }
2580
2581 /*
2582  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2583  * If n is 0, they match if they are equal. If n is != 0, they match if they
2584  * are different.
2585  */
2586 int
2587 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2588     struct pf_addr *b, sa_family_t af)
2589 {
2590         int     match = 0;
2591
2592         switch (af) {
2593 #ifdef INET
2594         case AF_INET:
2595                 if ((a->addr32[0] & m->addr32[0]) ==
2596                     (b->addr32[0] & m->addr32[0]))
2597                         match++;
2598                 break;
2599 #endif /* INET */
2600 #ifdef INET6
2601         case AF_INET6:
2602                 if (((a->addr32[0] & m->addr32[0]) ==
2603                      (b->addr32[0] & m->addr32[0])) &&
2604                     ((a->addr32[1] & m->addr32[1]) ==
2605                      (b->addr32[1] & m->addr32[1])) &&
2606                     ((a->addr32[2] & m->addr32[2]) ==
2607                      (b->addr32[2] & m->addr32[2])) &&
2608                     ((a->addr32[3] & m->addr32[3]) ==
2609                      (b->addr32[3] & m->addr32[3])))
2610                         match++;
2611                 break;
2612 #endif /* INET6 */
2613         }
2614         if (match) {
2615                 if (n)
2616                         return (0);
2617                 else
2618                         return (1);
2619         } else {
2620                 if (n)
2621                         return (1);
2622                 else
2623                         return (0);
2624         }
2625 }
2626
2627 /*
2628  * Return 1 if b <= a <= e, otherwise return 0.
2629  */
2630 int
2631 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2632     struct pf_addr *a, sa_family_t af)
2633 {
2634         switch (af) {
2635 #ifdef INET
2636         case AF_INET:
2637                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2638                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2639                         return (0);
2640                 break;
2641 #endif /* INET */
2642 #ifdef INET6
2643         case AF_INET6: {
2644                 int     i;
2645
2646                 /* check a >= b */
2647                 for (i = 0; i < 4; ++i)
2648                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2649                                 break;
2650                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2651                                 return (0);
2652                 /* check a <= e */
2653                 for (i = 0; i < 4; ++i)
2654                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2655                                 break;
2656                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2657                                 return (0);
2658                 break;
2659         }
2660 #endif /* INET6 */
2661         }
2662         return (1);
2663 }
2664
2665 static int
2666 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2667 {
2668         switch (op) {
2669         case PF_OP_IRG:
2670                 return ((p > a1) && (p < a2));
2671         case PF_OP_XRG:
2672                 return ((p < a1) || (p > a2));
2673         case PF_OP_RRG:
2674                 return ((p >= a1) && (p <= a2));
2675         case PF_OP_EQ:
2676                 return (p == a1);
2677         case PF_OP_NE:
2678                 return (p != a1);
2679         case PF_OP_LT:
2680                 return (p < a1);
2681         case PF_OP_LE:
2682                 return (p <= a1);
2683         case PF_OP_GT:
2684                 return (p > a1);
2685         case PF_OP_GE:
2686                 return (p >= a1);
2687         }
2688         return (0); /* never reached */
2689 }
2690
2691 int
2692 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2693 {
2694         NTOHS(a1);
2695         NTOHS(a2);
2696         NTOHS(p);
2697         return (pf_match(op, a1, a2, p));
2698 }
2699
2700 static int
2701 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2702 {
2703         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2704                 return (0);
2705         return (pf_match(op, a1, a2, u));
2706 }
2707
2708 static int
2709 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2710 {
2711         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2712                 return (0);
2713         return (pf_match(op, a1, a2, g));
2714 }
2715
2716 int
2717 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2718 {
2719         if (*tag == -1)
2720                 *tag = mtag;
2721
2722         return ((!r->match_tag_not && r->match_tag == *tag) ||
2723             (r->match_tag_not && r->match_tag != *tag));
2724 }
2725
2726 int
2727 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2728 {
2729
2730         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2731
2732         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2733                 return (ENOMEM);
2734
2735         pd->pf_mtag->tag = tag;
2736
2737         return (0);
2738 }
2739
2740 #define PF_ANCHOR_STACKSIZE     32
2741 struct pf_anchor_stackframe {
2742         struct pf_ruleset       *rs;
2743         struct pf_rule          *r;     /* XXX: + match bit */
2744         struct pf_anchor        *child;
2745 };
2746
2747 /*
2748  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2749  */
2750 #define PF_ANCHORSTACK_MATCH    0x00000001
2751 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2752
2753 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2754 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2755                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2756 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2757                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2758 } while (0)
2759
2760 void
2761 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2762     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2763     int *match)
2764 {
2765         struct pf_anchor_stackframe     *f;
2766
2767         PF_RULES_RASSERT();
2768
2769         if (match)
2770                 *match = 0;
2771         if (*depth >= PF_ANCHOR_STACKSIZE) {
2772                 printf("%s: anchor stack overflow on %s\n",
2773                     __func__, (*r)->anchor->name);
2774                 *r = TAILQ_NEXT(*r, entries);
2775                 return;
2776         } else if (*depth == 0 && a != NULL)
2777                 *a = *r;
2778         f = stack + (*depth)++;
2779         f->rs = *rs;
2780         f->r = *r;
2781         if ((*r)->anchor_wildcard) {
2782                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2783
2784                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2785                         *r = NULL;
2786                         return;
2787                 }
2788                 *rs = &f->child->ruleset;
2789         } else {
2790                 f->child = NULL;
2791                 *rs = &(*r)->anchor->ruleset;
2792         }
2793         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2794 }
2795
2796 int
2797 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2798     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2799     int *match)
2800 {
2801         struct pf_anchor_stackframe     *f;
2802         struct pf_rule *fr;
2803         int quick = 0;
2804
2805         PF_RULES_RASSERT();
2806
2807         do {
2808                 if (*depth <= 0)
2809                         break;
2810                 f = stack + *depth - 1;
2811                 fr = PF_ANCHOR_RULE(f);
2812                 if (f->child != NULL) {
2813                         struct pf_anchor_node *parent;
2814
2815                         /*
2816                          * This block traverses through
2817                          * a wildcard anchor.
2818                          */
2819                         parent = &fr->anchor->children;
2820                         if (match != NULL && *match) {
2821                                 /*
2822                                  * If any of "*" matched, then
2823                                  * "foo/ *" matched, mark frame
2824                                  * appropriately.
2825                                  */
2826                                 PF_ANCHOR_SET_MATCH(f);
2827                                 *match = 0;
2828                         }
2829                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2830                         if (f->child != NULL) {
2831                                 *rs = &f->child->ruleset;
2832                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2833                                 if (*r == NULL)
2834                                         continue;
2835                                 else
2836                                         break;
2837                         }
2838                 }
2839                 (*depth)--;
2840                 if (*depth == 0 && a != NULL)
2841                         *a = NULL;
2842                 *rs = f->rs;
2843                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2844                         quick = fr->quick;
2845                 *r = TAILQ_NEXT(fr, entries);
2846         } while (*r == NULL);
2847
2848         return (quick);
2849 }
2850
2851 #ifdef INET6
2852 void
2853 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2854     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2855 {
2856         switch (af) {
2857 #ifdef INET
2858         case AF_INET:
2859                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2860                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2861                 break;
2862 #endif /* INET */
2863         case AF_INET6:
2864                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2865                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2866                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2867                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2868                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2869                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2870                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2871                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2872                 break;
2873         }
2874 }
2875
2876 void
2877 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2878 {
2879         switch (af) {
2880 #ifdef INET
2881         case AF_INET:
2882                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2883                 break;
2884 #endif /* INET */
2885         case AF_INET6:
2886                 if (addr->addr32[3] == 0xffffffff) {
2887                         addr->addr32[3] = 0;
2888                         if (addr->addr32[2] == 0xffffffff) {
2889                                 addr->addr32[2] = 0;
2890                                 if (addr->addr32[1] == 0xffffffff) {
2891                                         addr->addr32[1] = 0;
2892                                         addr->addr32[0] =
2893                                             htonl(ntohl(addr->addr32[0]) + 1);
2894                                 } else
2895                                         addr->addr32[1] =
2896                                             htonl(ntohl(addr->addr32[1]) + 1);
2897                         } else
2898                                 addr->addr32[2] =
2899                                     htonl(ntohl(addr->addr32[2]) + 1);
2900                 } else
2901                         addr->addr32[3] =
2902                             htonl(ntohl(addr->addr32[3]) + 1);
2903                 break;
2904         }
2905 }
2906 #endif /* INET6 */
2907
2908 int
2909 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2910 {
2911         struct pf_addr          *saddr, *daddr;
2912         u_int16_t                sport, dport;
2913         struct inpcbinfo        *pi;
2914         struct inpcb            *inp;
2915
2916         pd->lookup.uid = UID_MAX;
2917         pd->lookup.gid = GID_MAX;
2918
2919         switch (pd->proto) {
2920         case IPPROTO_TCP:
2921                 if (pd->hdr.tcp == NULL)
2922                         return (-1);
2923                 sport = pd->hdr.tcp->th_sport;
2924                 dport = pd->hdr.tcp->th_dport;
2925                 pi = &V_tcbinfo;
2926                 break;
2927         case IPPROTO_UDP:
2928                 if (pd->hdr.udp == NULL)
2929                         return (-1);
2930                 sport = pd->hdr.udp->uh_sport;
2931                 dport = pd->hdr.udp->uh_dport;
2932                 pi = &V_udbinfo;
2933                 break;
2934         default:
2935                 return (-1);
2936         }
2937         if (direction == PF_IN) {
2938                 saddr = pd->src;
2939                 daddr = pd->dst;
2940         } else {
2941                 u_int16_t       p;
2942
2943                 p = sport;
2944                 sport = dport;
2945                 dport = p;
2946                 saddr = pd->dst;
2947                 daddr = pd->src;
2948         }
2949         switch (pd->af) {
2950 #ifdef INET
2951         case AF_INET:
2952                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2953                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2954                 if (inp == NULL) {
2955                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2956                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2957                            INPLOOKUP_RLOCKPCB, NULL, m);
2958                         if (inp == NULL)
2959                                 return (-1);
2960                 }
2961                 break;
2962 #endif /* INET */
2963 #ifdef INET6
2964         case AF_INET6:
2965                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2966                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2967                 if (inp == NULL) {
2968                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2969                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2970                             INPLOOKUP_RLOCKPCB, NULL, m);
2971                         if (inp == NULL)
2972                                 return (-1);
2973                 }
2974                 break;
2975 #endif /* INET6 */
2976
2977         default:
2978                 return (-1);
2979         }
2980         INP_RLOCK_ASSERT(inp);
2981         pd->lookup.uid = inp->inp_cred->cr_uid;
2982         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2983         INP_RUNLOCK(inp);
2984
2985         return (1);
2986 }
2987
2988 static u_int8_t
2989 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2990 {
2991         int              hlen;
2992         u_int8_t         hdr[60];
2993         u_int8_t        *opt, optlen;
2994         u_int8_t         wscale = 0;
2995
2996         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2997         if (hlen <= sizeof(struct tcphdr))
2998                 return (0);
2999         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3000                 return (0);
3001         opt = hdr + sizeof(struct tcphdr);
3002         hlen -= sizeof(struct tcphdr);
3003         while (hlen >= 3) {
3004                 switch (*opt) {
3005                 case TCPOPT_EOL:
3006                 case TCPOPT_NOP:
3007                         ++opt;
3008                         --hlen;
3009                         break;
3010                 case TCPOPT_WINDOW:
3011                         wscale = opt[2];
3012                         if (wscale > TCP_MAX_WINSHIFT)
3013                                 wscale = TCP_MAX_WINSHIFT;
3014                         wscale |= PF_WSCALE_FLAG;
3015                         /* FALLTHROUGH */
3016                 default:
3017                         optlen = opt[1];
3018                         if (optlen < 2)
3019                                 optlen = 2;
3020                         hlen -= optlen;
3021                         opt += optlen;
3022                         break;
3023                 }
3024         }
3025         return (wscale);
3026 }
3027
3028 static u_int16_t
3029 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3030 {
3031         int              hlen;
3032         u_int8_t         hdr[60];
3033         u_int8_t        *opt, optlen;
3034         u_int16_t        mss = V_tcp_mssdflt;
3035
3036         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3037         if (hlen <= sizeof(struct tcphdr))
3038                 return (0);
3039         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3040                 return (0);
3041         opt = hdr + sizeof(struct tcphdr);
3042         hlen -= sizeof(struct tcphdr);
3043         while (hlen >= TCPOLEN_MAXSEG) {
3044                 switch (*opt) {
3045                 case TCPOPT_EOL:
3046                 case TCPOPT_NOP:
3047                         ++opt;
3048                         --hlen;
3049                         break;
3050                 case TCPOPT_MAXSEG:
3051                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3052                         NTOHS(mss);
3053                         /* FALLTHROUGH */
3054                 default:
3055                         optlen = opt[1];
3056                         if (optlen < 2)
3057                                 optlen = 2;
3058                         hlen -= optlen;
3059                         opt += optlen;
3060                         break;
3061                 }
3062         }
3063         return (mss);
3064 }
3065
3066 static u_int16_t
3067 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3068 {
3069 #ifdef INET
3070         struct nhop4_basic      nh4;
3071 #endif /* INET */
3072 #ifdef INET6
3073         struct nhop6_basic      nh6;
3074         struct in6_addr         dst6;
3075         uint32_t                scopeid;
3076 #endif /* INET6 */
3077         int                      hlen = 0;
3078         uint16_t                 mss = 0;
3079
3080         switch (af) {
3081 #ifdef INET
3082         case AF_INET:
3083                 hlen = sizeof(struct ip);
3084                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3085                         mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3086                 break;
3087 #endif /* INET */
3088 #ifdef INET6
3089         case AF_INET6:
3090                 hlen = sizeof(struct ip6_hdr);
3091                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3092                 if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3093                         mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3094                 break;
3095 #endif /* INET6 */
3096         }
3097
3098         mss = max(V_tcp_mssdflt, mss);
3099         mss = min(mss, offer);
3100         mss = max(mss, 64);             /* sanity - at least max opt space */
3101         return (mss);
3102 }
3103
3104 static u_int32_t
3105 pf_tcp_iss(struct pf_pdesc *pd)
3106 {
3107         MD5_CTX ctx;
3108         u_int32_t digest[4];
3109
3110         if (V_pf_tcp_secret_init == 0) {
3111                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3112                 MD5Init(&V_pf_tcp_secret_ctx);
3113                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3114                     sizeof(V_pf_tcp_secret));
3115                 V_pf_tcp_secret_init = 1;
3116         }
3117
3118         ctx = V_pf_tcp_secret_ctx;
3119
3120         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3121         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3122         if (pd->af == AF_INET6) {
3123                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3124                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3125         } else {
3126                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3127                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3128         }
3129         MD5Final((u_char *)digest, &ctx);
3130         V_pf_tcp_iss_off += 4096;
3131 #define ISN_RANDOM_INCREMENT (4096 - 1)
3132         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3133             V_pf_tcp_iss_off);
3134 #undef  ISN_RANDOM_INCREMENT
3135 }
3136
3137 static int
3138 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3139     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3140     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3141 {
3142         struct pf_rule          *nr = NULL;
3143         struct pf_addr          * const saddr = pd->src;
3144         struct pf_addr          * const daddr = pd->dst;
3145         sa_family_t              af = pd->af;
3146         struct pf_rule          *r, *a = NULL;
3147         struct pf_ruleset       *ruleset = NULL;
3148         struct pf_src_node      *nsn = NULL;
3149         struct tcphdr           *th = pd->hdr.tcp;
3150         struct pf_state_key     *sk = NULL, *nk = NULL;
3151         u_short                  reason;
3152         int                      rewrite = 0, hdrlen = 0;
3153         int                      tag = -1, rtableid = -1;
3154         int                      asd = 0;
3155         int                      match = 0;
3156         int                      state_icmp = 0;
3157         u_int16_t                sport = 0, dport = 0;
3158         u_int16_t                bproto_sum = 0, bip_sum = 0;
3159         u_int8_t                 icmptype = 0, icmpcode = 0;
3160         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3161
3162         PF_RULES_RASSERT();
3163
3164         if (inp != NULL) {
3165                 INP_LOCK_ASSERT(inp);
3166                 pd->lookup.uid = inp->inp_cred->cr_uid;
3167                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3168                 pd->lookup.done = 1;
3169         }
3170
3171         switch (pd->proto) {
3172         case IPPROTO_TCP:
3173                 sport = th->th_sport;
3174                 dport = th->th_dport;
3175                 hdrlen = sizeof(*th);
3176                 break;
3177         case IPPROTO_UDP:
3178                 sport = pd->hdr.udp->uh_sport;
3179                 dport = pd->hdr.udp->uh_dport;
3180                 hdrlen = sizeof(*pd->hdr.udp);
3181                 break;
3182 #ifdef INET
3183         case IPPROTO_ICMP:
3184                 if (pd->af != AF_INET)
3185                         break;
3186                 sport = dport = pd->hdr.icmp->icmp_id;
3187                 hdrlen = sizeof(*pd->hdr.icmp);
3188                 icmptype = pd->hdr.icmp->icmp_type;
3189                 icmpcode = pd->hdr.icmp->icmp_code;
3190
3191                 if (icmptype == ICMP_UNREACH ||
3192                     icmptype == ICMP_SOURCEQUENCH ||
3193                     icmptype == ICMP_REDIRECT ||
3194                     icmptype == ICMP_TIMXCEED ||
3195                     icmptype == ICMP_PARAMPROB)
3196                         state_icmp++;
3197                 break;
3198 #endif /* INET */
3199 #ifdef INET6
3200         case IPPROTO_ICMPV6:
3201                 if (af != AF_INET6)
3202                         break;
3203                 sport = dport = pd->hdr.icmp6->icmp6_id;
3204                 hdrlen = sizeof(*pd->hdr.icmp6);
3205                 icmptype = pd->hdr.icmp6->icmp6_type;
3206                 icmpcode = pd->hdr.icmp6->icmp6_code;
3207
3208                 if (icmptype == ICMP6_DST_UNREACH ||
3209                     icmptype == ICMP6_PACKET_TOO_BIG ||
3210                     icmptype == ICMP6_TIME_EXCEEDED ||
3211                     icmptype == ICMP6_PARAM_PROB)
3212                         state_icmp++;
3213                 break;
3214 #endif /* INET6 */
3215         default:
3216                 sport = dport = hdrlen = 0;
3217                 break;
3218         }
3219
3220         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3221
3222         /* check packet for BINAT/NAT/RDR */
3223         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3224             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3225                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3226                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3227
3228                 if (pd->ip_sum)
3229                         bip_sum = *pd->ip_sum;
3230
3231                 switch (pd->proto) {
3232                 case IPPROTO_TCP:
3233                         bproto_sum = th->th_sum;
3234                         pd->proto_sum = &th->th_sum;
3235
3236                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3237                             nk->port[pd->sidx] != sport) {
3238                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3239                                     &th->th_sum, &nk->addr[pd->sidx],
3240                                     nk->port[pd->sidx], 0, af);
3241                                 pd->sport = &th->th_sport;
3242                                 sport = th->th_sport;
3243                         }
3244
3245                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3246                             nk->port[pd->didx] != dport) {
3247                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3248                                     &th->th_sum, &nk->addr[pd->didx],
3249                                     nk->port[pd->didx], 0, af);
3250                                 dport = th->th_dport;
3251                                 pd->dport = &th->th_dport;
3252                         }
3253                         rewrite++;
3254                         break;
3255                 case IPPROTO_UDP:
3256                         bproto_sum = pd->hdr.udp->uh_sum;
3257                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3258
3259                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3260                             nk->port[pd->sidx] != sport) {
3261                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3262                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3263                                     &nk->addr[pd->sidx],
3264                                     nk->port[pd->sidx], 1, af);
3265                                 sport = pd->hdr.udp->uh_sport;
3266                                 pd->sport = &pd->hdr.udp->uh_sport;
3267                         }
3268
3269                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3270                             nk->port[pd->didx] != dport) {
3271                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3272                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3273                                     &nk->addr[pd->didx],
3274                                     nk->port[pd->didx], 1, af);
3275                                 dport = pd->hdr.udp->uh_dport;
3276                                 pd->dport = &pd->hdr.udp->uh_dport;
3277                         }
3278                         rewrite++;
3279                         break;
3280 #ifdef INET
3281                 case IPPROTO_ICMP:
3282                         nk->port[0] = nk->port[1];
3283                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3284                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3285                                     nk->addr[pd->sidx].v4.s_addr, 0);
3286
3287                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3288                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3289                                     nk->addr[pd->didx].v4.s_addr, 0);
3290
3291                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3292                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3293                                     pd->hdr.icmp->icmp_cksum, sport,
3294                                     nk->port[1], 0);
3295                                 pd->hdr.icmp->icmp_id = nk->port[1];
3296                                 pd->sport = &pd->hdr.icmp->icmp_id;
3297                         }
3298                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3299                         break;
3300 #endif /* INET */
3301 #ifdef INET6
3302                 case IPPROTO_ICMPV6:
3303                         nk->port[0] = nk->port[1];
3304                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3305                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3306                                     &nk->addr[pd->sidx], 0);
3307
3308                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3309                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3310                                     &nk->addr[pd->didx], 0);
3311                         rewrite++;
3312                         break;
3313 #endif /* INET */
3314                 default:
3315                         switch (af) {
3316 #ifdef INET
3317                         case AF_INET:
3318                                 if (PF_ANEQ(saddr,
3319                                     &nk->addr[pd->sidx], AF_INET))
3320                                         pf_change_a(&saddr->v4.s_addr,
3321                                             pd->ip_sum,
3322                                             nk->addr[pd->sidx].v4.s_addr, 0);
3323
3324                                 if (PF_ANEQ(daddr,
3325                                     &nk->addr[pd->didx], AF_INET))
3326                                         pf_change_a(&daddr->v4.s_addr,
3327                                             pd->ip_sum,
3328                                             nk->addr[pd->didx].v4.s_addr, 0);
3329                                 break;
3330 #endif /* INET */
3331 #ifdef INET6
3332                         case AF_INET6:
3333                                 if (PF_ANEQ(saddr,
3334                                     &nk->addr[pd->sidx], AF_INET6))
3335                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3336
3337                                 if (PF_ANEQ(daddr,
3338                                     &nk->addr[pd->didx], AF_INET6))
3339                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3340                                 break;
3341 #endif /* INET */
3342                         }
3343                         break;
3344                 }
3345                 if (nr->natpass)
3346                         r = NULL;
3347                 pd->nat_rule = nr;
3348         }
3349
3350         while (r != NULL) {
3351                 r->evaluations++;
3352                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3353                         r = r->skip[PF_SKIP_IFP].ptr;
3354                 else if (r->direction && r->direction != direction)
3355                         r = r->skip[PF_SKIP_DIR].ptr;
3356                 else if (r->af && r->af != af)
3357                         r = r->skip[PF_SKIP_AF].ptr;
3358                 else if (r->proto && r->proto != pd->proto)
3359                         r = r->skip[PF_SKIP_PROTO].ptr;
3360                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3361                     r->src.neg, kif, M_GETFIB(m)))
3362                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3363                 /* tcp/udp only. port_op always 0 in other cases */
3364                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3365                     r->src.port[0], r->src.port[1], sport))
3366                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3367                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3368                     r->dst.neg, NULL, M_GETFIB(m)))
3369                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3370                 /* tcp/udp only. port_op always 0 in other cases */
3371                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3372                     r->dst.port[0], r->dst.port[1], dport))
3373                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3374                 /* icmp only. type always 0 in other cases */
3375                 else if (r->type && r->type != icmptype + 1)
3376                         r = TAILQ_NEXT(r, entries);
3377                 /* icmp only. type always 0 in other cases */
3378                 else if (r->code && r->code != icmpcode + 1)
3379                         r = TAILQ_NEXT(r, entries);
3380                 else if (r->tos && !(r->tos == pd->tos))
3381                         r = TAILQ_NEXT(r, entries);
3382                 else if (r->rule_flag & PFRULE_FRAGMENT)
3383                         r = TAILQ_NEXT(r, entries);
3384                 else if (pd->proto == IPPROTO_TCP &&
3385                     (r->flagset & th->th_flags) != r->flags)
3386                         r = TAILQ_NEXT(r, entries);
3387                 /* tcp/udp only. uid.op always 0 in other cases */
3388                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3389                     pf_socket_lookup(direction, pd, m), 1)) &&
3390                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3391                     pd->lookup.uid))
3392                         r = TAILQ_NEXT(r, entries);
3393                 /* tcp/udp only. gid.op always 0 in other cases */
3394                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3395                     pf_socket_lookup(direction, pd, m), 1)) &&
3396                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3397                     pd->lookup.gid))
3398                         r = TAILQ_NEXT(r, entries);
3399                 else if (r->prio &&
3400                     !pf_match_ieee8021q_pcp(r->prio, m))
3401                         r = TAILQ_NEXT(r, entries);
3402                 else if (r->prob &&
3403                     r->prob <= arc4random())
3404                         r = TAILQ_NEXT(r, entries);
3405                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3406                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3407                         r = TAILQ_NEXT(r, entries);
3408                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3409                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3410                     pf_osfp_fingerprint(pd, m, off, th),
3411                     r->os_fingerprint)))
3412                         r = TAILQ_NEXT(r, entries);
3413                 else {
3414                         if (r->tag)
3415                                 tag = r->tag;
3416                         if (r->rtableid >= 0)
3417                                 rtableid = r->rtableid;
3418                         if (r->anchor == NULL) {
3419                                 match = 1;
3420                                 *rm = r;
3421                                 *am = a;
3422                                 *rsm = ruleset;
3423                                 if ((*rm)->quick)
3424                                         break;
3425                                 r = TAILQ_NEXT(r, entries);
3426                         } else
3427                                 pf_step_into_anchor(anchor_stack, &asd,
3428                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3429                                     &match);
3430                 }
3431                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3432                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3433                         break;
3434         }
3435         r = *rm;
3436         a = *am;
3437         ruleset = *rsm;
3438
3439         REASON_SET(&reason, PFRES_MATCH);
3440
3441         if (r->log || (nr != NULL && nr->log)) {
3442                 if (rewrite)
3443                         m_copyback(m, off, hdrlen, pd->hdr.any);
3444                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3445                     ruleset, pd, 1);
3446         }
3447
3448         if ((r->action == PF_DROP) &&
3449             ((r->rule_flag & PFRULE_RETURNRST) ||
3450             (r->rule_flag & PFRULE_RETURNICMP) ||
3451             (r->rule_flag & PFRULE_RETURN))) {
3452                 /* undo NAT changes, if they have taken place */
3453                 if (nr != NULL) {
3454                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3455                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3456                         if (pd->sport)
3457                                 *pd->sport = sk->port[pd->sidx];
3458                         if (pd->dport)
3459                                 *pd->dport = sk->port[pd->didx];
3460                         if (pd->proto_sum)
3461                                 *pd->proto_sum = bproto_sum;
3462                         if (pd->ip_sum)
3463                                 *pd->ip_sum = bip_sum;
3464                         m_copyback(m, off, hdrlen, pd->hdr.any);
3465                 }
3466                 if (pd->proto == IPPROTO_TCP &&
3467                     ((r->rule_flag & PFRULE_RETURNRST) ||
3468                     (r->rule_flag & PFRULE_RETURN)) &&
3469                     !(th->th_flags & TH_RST)) {
3470                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3471                         int              len = 0;
3472 #ifdef INET
3473                         struct ip       *h4;
3474 #endif
3475 #ifdef INET6
3476                         struct ip6_hdr  *h6;
3477 #endif
3478
3479                         switch (af) {
3480 #ifdef INET
3481                         case AF_INET:
3482                                 h4 = mtod(m, struct ip *);
3483                                 len = ntohs(h4->ip_len) - off;
3484                                 break;
3485 #endif
3486 #ifdef INET6
3487                         case AF_INET6:
3488                                 h6 = mtod(m, struct ip6_hdr *);
3489                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3490                                 break;
3491 #endif
3492                         }
3493
3494                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3495                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3496                         else {
3497                                 if (th->th_flags & TH_SYN)
3498                                         ack++;
3499                                 if (th->th_flags & TH_FIN)
3500                                         ack++;
3501                                 pf_send_tcp(m, r, af, pd->dst,
3502                                     pd->src, th->th_dport, th->th_sport,
3503                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3504                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3505                         }
3506                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3507                     r->return_icmp)
3508                         pf_send_icmp(m, r->return_icmp >> 8,
3509                             r->return_icmp & 255, af, r);
3510                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3511                     r->return_icmp6)
3512                         pf_send_icmp(m, r->return_icmp6 >> 8,
3513                             r->return_icmp6 & 255, af, r);
3514         }
3515
3516         if (r->action == PF_DROP)
3517                 goto cleanup;
3518
3519         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3520                 REASON_SET(&reason, PFRES_MEMORY);
3521                 goto cleanup;
3522         }
3523         if (rtableid >= 0)
3524                 M_SETFIB(m, rtableid);
3525
3526         if (!state_icmp && (r->keep_state || nr != NULL ||
3527             (pd->flags & PFDESC_TCP_NORM))) {
3528                 int action;
3529                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3530                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3531                     hdrlen);
3532                 if (action != PF_PASS)
3533                         return (action);
3534         } else {
3535                 if (sk != NULL)
3536                         uma_zfree(V_pf_state_key_z, sk);
3537                 if (nk != NULL)
3538                         uma_zfree(V_pf_state_key_z, nk);
3539         }
3540
3541         /* copy back packet headers if we performed NAT operations */
3542         if (rewrite)
3543                 m_copyback(m, off, hdrlen, pd->hdr.any);
3544
3545         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3546             direction == PF_OUT &&
3547             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3548                 /*
3549                  * We want the state created, but we dont
3550                  * want to send this in case a partner
3551                  * firewall has to know about it to allow
3552                  * replies through it.
3553                  */
3554                 return (PF_DEFER);
3555
3556         return (PF_PASS);
3557
3558 cleanup:
3559         if (sk != NULL)
3560                 uma_zfree(V_pf_state_key_z, sk);
3561         if (nk != NULL)
3562                 uma_zfree(V_pf_state_key_z, nk);
3563         return (PF_DROP);
3564 }
3565
3566 static int
3567 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3568     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3569     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3570     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3571     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3572 {
3573         struct pf_state         *s = NULL;
3574         struct pf_src_node      *sn = NULL;
3575         struct tcphdr           *th = pd->hdr.tcp;
3576         u_int16_t                mss = V_tcp_mssdflt;
3577         u_short                  reason;
3578
3579         /* check maximums */
3580         if (r->max_states &&
3581             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3582                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3583                 REASON_SET(&reason, PFRES_MAXSTATES);
3584                 goto csfailed;
3585         }
3586         /* src node for filter rule */
3587         if ((r->rule_flag & PFRULE_SRCTRACK ||
3588             r->rpool.opts & PF_POOL_STICKYADDR) &&
3589             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3590                 REASON_SET(&reason, PFRES_SRCLIMIT);
3591                 goto csfailed;
3592         }
3593         /* src node for translation rule */
3594         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3595             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3596                 REASON_SET(&reason, PFRES_SRCLIMIT);
3597                 goto csfailed;
3598         }
3599         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3600         if (s == NULL) {
3601                 REASON_SET(&reason, PFRES_MEMORY);
3602                 goto csfailed;
3603         }
3604         s->rule.ptr = r;
3605         s->nat_rule.ptr = nr;
3606         s->anchor.ptr = a;
3607         STATE_INC_COUNTERS(s);
3608         if (r->allow_opts)
3609                 s->state_flags |= PFSTATE_ALLOWOPTS;
3610         if (r->rule_flag & PFRULE_STATESLOPPY)
3611                 s->state_flags |= PFSTATE_SLOPPY;
3612         s->log = r->log & PF_LOG_ALL;
3613         s->sync_state = PFSYNC_S_NONE;
3614         if (nr != NULL)
3615                 s->log |= nr->log & PF_LOG_ALL;
3616         switch (pd->proto) {
3617         case IPPROTO_TCP:
3618                 s->src.seqlo = ntohl(th->th_seq);
3619                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3620                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3621                     r->keep_state == PF_STATE_MODULATE) {
3622                         /* Generate sequence number modulator */
3623                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3624                             0)
3625                                 s->src.seqdiff = 1;
3626                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3627                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3628                         *rewrite = 1;
3629                 } else
3630                         s->src.seqdiff = 0;
3631                 if (th->th_flags & TH_SYN) {
3632                         s->src.seqhi++;
3633                         s->src.wscale = pf_get_wscale(m, off,
3634                             th->th_off, pd->af);
3635                 }
3636                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3637                 if (s->src.wscale & PF_WSCALE_MASK) {
3638                         /* Remove scale factor from initial window */
3639                         int win = s->src.max_win;
3640                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3641                         s->src.max_win = (win - 1) >>
3642                             (s->src.wscale & PF_WSCALE_MASK);
3643                 }
3644                 if (th->th_flags & TH_FIN)
3645                         s->src.seqhi++;
3646                 s->dst.seqhi = 1;
3647                 s->dst.max_win = 1;
3648                 s->src.state = TCPS_SYN_SENT;
3649                 s->dst.state = TCPS_CLOSED;
3650                 s->timeout = PFTM_TCP_FIRST_PACKET;
3651                 break;
3652         case IPPROTO_UDP:
3653                 s->src.state = PFUDPS_SINGLE;
3654                 s->dst.state = PFUDPS_NO_TRAFFIC;
3655                 s->timeout = PFTM_UDP_FIRST_PACKET;
3656                 break;
3657         case IPPROTO_ICMP:
3658 #ifdef INET6
3659         case IPPROTO_ICMPV6:
3660 #endif
3661                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3662                 break;
3663         default:
3664                 s->src.state = PFOTHERS_SINGLE;
3665                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3666                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3667         }
3668
3669         if (r->rt && r->rt != PF_FASTROUTE) {
3670                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3671                         REASON_SET(&reason, PFRES_MAPFAILED);
3672                         pf_src_tree_remove_state(s);
3673                         STATE_DEC_COUNTERS(s);
3674                         uma_zfree(V_pf_state_z, s);
3675                         goto csfailed;
3676                 }
3677                 s->rt_kif = r->rpool.cur->kif;
3678         }
3679
3680         s->creation = time_uptime;
3681         s->expire = time_uptime;
3682
3683         if (sn != NULL)
3684                 s->src_node = sn;
3685         if (nsn != NULL) {
3686                 /* XXX We only modify one side for now. */
3687                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3688                 s->nat_src_node = nsn;
3689         }
3690         if (pd->proto == IPPROTO_TCP) {
3691                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3692                     off, pd, th, &s->src, &s->dst)) {
3693                         REASON_SET(&reason, PFRES_MEMORY);
3694                         pf_src_tree_remove_state(s);
3695                         STATE_DEC_COUNTERS(s);
3696                         uma_zfree(V_pf_state_z, s);
3697                         return (PF_DROP);
3698                 }
3699                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3700                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3701                     &s->src, &s->dst, rewrite)) {
3702                         /* This really shouldn't happen!!! */
3703                         DPFPRINTF(PF_DEBUG_URGENT,
3704                             ("pf_normalize_tcp_stateful failed on first pkt"));
3705                         pf_normalize_tcp_cleanup(s);
3706                         pf_src_tree_remove_state(s);
3707                         STATE_DEC_COUNTERS(s);
3708                         uma_zfree(V_pf_state_z, s);
3709                         return (PF_DROP);
3710                 }
3711         }
3712         s->direction = pd->dir;
3713
3714         /*
3715          * sk/nk could already been setup by pf_get_translation().
3716          */
3717         if (nr == NULL) {
3718                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3719                     __func__, nr, sk, nk));
3720                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3721                 if (sk == NULL)
3722                         goto csfailed;
3723                 nk = sk;
3724         } else
3725                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3726                     __func__, nr, sk, nk));
3727
3728         /* Swap sk/nk for PF_OUT. */
3729         if (pf_state_insert(BOUND_IFACE(r, kif),
3730             (pd->dir == PF_IN) ? sk : nk,
3731             (pd->dir == PF_IN) ? nk : sk, s)) {
3732                 if (pd->proto == IPPROTO_TCP)
3733                         pf_normalize_tcp_cleanup(s);
3734                 REASON_SET(&reason, PFRES_STATEINS);
3735                 pf_src_tree_remove_state(s);
3736                 STATE_DEC_COUNTERS(s);
3737                 uma_zfree(V_pf_state_z, s);
3738                 return (PF_DROP);
3739         } else
3740                 *sm = s;
3741
3742         if (tag > 0)
3743                 s->tag = tag;
3744         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3745             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3746                 s->src.state = PF_TCPS_PROXY_SRC;
3747                 /* undo NAT changes, if they have taken place */
3748                 if (nr != NULL) {
3749                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3750                         if (pd->dir == PF_OUT)
3751                                 skt = s->key[PF_SK_STACK];
3752                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3753                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3754                         if (pd->sport)
3755                                 *pd->sport = skt->port[pd->sidx];
3756                         if (pd->dport)
3757                                 *pd->dport = skt->port[pd->didx];
3758                         if (pd->proto_sum)
3759                                 *pd->proto_sum = bproto_sum;
3760                         if (pd->ip_sum)
3761                                 *pd->ip_sum = bip_sum;
3762                         m_copyback(m, off, hdrlen, pd->hdr.any);
3763                 }
3764                 s->src.seqhi = htonl(arc4random());
3765                 /* Find mss option */
3766                 int rtid = M_GETFIB(m);
3767                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3768                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3769                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3770                 s->src.mss = mss;
3771                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3772                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3773                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3774                 REASON_SET(&reason, PFRES_SYNPROXY);
3775                 return (PF_SYNPROXY_DROP);
3776         }
3777
3778         return (PF_PASS);
3779
3780 csfailed:
3781         if (sk != NULL)
3782                 uma_zfree(V_pf_state_key_z, sk);
3783         if (nk != NULL)
3784                 uma_zfree(V_pf_state_key_z, nk);
3785
3786         if (sn != NULL) {
3787                 struct pf_srchash *sh;
3788
3789                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3790                 PF_HASHROW_LOCK(sh);
3791                 if (--sn->states == 0 && sn->expire == 0) {
3792                         pf_unlink_src_node(sn);
3793                         uma_zfree(V_pf_sources_z, sn);
3794                         counter_u64_add(
3795                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3796                 }
3797                 PF_HASHROW_UNLOCK(sh);
3798         }
3799
3800         if (nsn != sn && nsn != NULL) {
3801                 struct pf_srchash *sh;
3802
3803                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3804                 PF_HASHROW_LOCK(sh);
3805                 if (--nsn->states == 0 && nsn->expire == 0) {
3806                         pf_unlink_src_node(nsn);
3807                         uma_zfree(V_pf_sources_z, nsn);
3808                         counter_u64_add(
3809                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3810                 }
3811                 PF_HASHROW_UNLOCK(sh);
3812         }
3813
3814         return (PF_DROP);
3815 }
3816
3817 static int
3818 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3819     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3820     struct pf_ruleset **rsm)
3821 {
3822         struct pf_rule          *r, *a = NULL;
3823         struct pf_ruleset       *ruleset = NULL;
3824         sa_family_t              af = pd->af;
3825         u_short                  reason;
3826         int                      tag = -1;
3827         int                      asd = 0;
3828         int                      match = 0;
3829         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3830
3831         PF_RULES_RASSERT();
3832
3833         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3834         while (r != NULL) {
3835                 r->evaluations++;
3836                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3837                         r = r->skip[PF_SKIP_IFP].ptr;
3838                 else if (r->direction && r->direction != direction)
3839                         r = r->skip[PF_SKIP_DIR].ptr;
3840                 else if (r->af && r->af != af)
3841                         r = r->skip[PF_SKIP_AF].ptr;
3842                 else if (r->proto && r->proto != pd->proto)
3843                         r = r->skip[PF_SKIP_PROTO].ptr;
3844                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3845                     r->src.neg, kif, M_GETFIB(m)))
3846                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3847                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3848                     r->dst.neg, NULL, M_GETFIB(m)))
3849                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3850                 else if (r->tos && !(r->tos == pd->tos))
3851                         r = TAILQ_NEXT(r, entries);
3852                 else if (r->os_fingerprint != PF_OSFP_ANY)
3853                         r = TAILQ_NEXT(r, entries);
3854                 else if (pd->proto == IPPROTO_UDP &&
3855                     (r->src.port_op || r->dst.port_op))
3856                         r = TAILQ_NEXT(r, entries);
3857                 else if (pd->proto == IPPROTO_TCP &&
3858                     (r->src.port_op || r->dst.port_op || r->flagset))
3859                         r = TAILQ_NEXT(r, entries);
3860                 else if ((pd->proto == IPPROTO_ICMP ||
3861                     pd->proto == IPPROTO_ICMPV6) &&
3862                     (r->type || r->code))
3863                         r = TAILQ_NEXT(r, entries);
3864                 else if (r->prio &&
3865                     !pf_match_ieee8021q_pcp(r->prio, m))
3866                         r = TAILQ_NEXT(r, entries);
3867                 else if (r->prob && r->prob <=
3868                     (arc4random() % (UINT_MAX - 1) + 1))
3869                         r = TAILQ_NEXT(r, entries);
3870                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3871                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3872                         r = TAILQ_NEXT(r, entries);
3873                 else {
3874                         if (r->anchor == NULL) {
3875                                 match = 1;
3876                                 *rm = r;
3877                                 *am = a;
3878                                 *rsm = ruleset;
3879                                 if ((*rm)->quick)
3880                                         break;
3881                                 r = TAILQ_NEXT(r, entries);
3882                         } else
3883                                 pf_step_into_anchor(anchor_stack, &asd,
3884                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3885                                     &match);
3886                 }
3887                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3888                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3889                         break;
3890         }
3891         r = *rm;
3892         a = *am;
3893         ruleset = *rsm;
3894
3895         REASON_SET(&reason, PFRES_MATCH);
3896
3897         if (r->log)
3898                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3899                     1);
3900
3901         if (r->action != PF_PASS)
3902                 return (PF_DROP);
3903
3904         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3905                 REASON_SET(&reason, PFRES_MEMORY);
3906                 return (PF_DROP);
3907         }
3908
3909         return (PF_PASS);
3910 }
3911
3912 static int
3913 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3914         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3915         struct pf_pdesc *pd, u_short *reason, int *copyback)
3916 {
3917         struct tcphdr           *th = pd->hdr.tcp;
3918         u_int16_t                win = ntohs(th->th_win);
3919         u_int32_t                ack, end, seq, orig_seq;
3920         u_int8_t                 sws, dws;
3921         int                      ackskew;
3922
3923         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3924                 sws = src->wscale & PF_WSCALE_MASK;
3925                 dws = dst->wscale & PF_WSCALE_MASK;
3926         } else
3927                 sws = dws = 0;
3928
3929         /*
3930          * Sequence tracking algorithm from Guido van Rooij's paper:
3931          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3932          *      tcp_filtering.ps
3933          */
3934
3935         orig_seq = seq = ntohl(th->th_seq);
3936         if (src->seqlo == 0) {
3937                 /* First packet from this end. Set its state */
3938
3939                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3940                     src->scrub == NULL) {
3941                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3942                                 REASON_SET(reason, PFRES_MEMORY);
3943                                 return (PF_DROP);
3944                         }
3945                 }
3946
3947                 /* Deferred generation of sequence number modulator */
3948                 if (dst->seqdiff && !src->seqdiff) {
3949                         /* use random iss for the TCP server */
3950                         while ((src->seqdiff = arc4random() - seq) == 0)
3951                                 ;
3952                         ack = ntohl(th->th_ack) - dst->seqdiff;
3953                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3954                             src->seqdiff), 0);
3955                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3956                         *copyback = 1;
3957                 } else {
3958                         ack = ntohl(th->th_ack);
3959                 }
3960
3961                 end = seq + pd->p_len;
3962                 if (th->th_flags & TH_SYN) {
3963                         end++;
3964                         if (dst->wscale & PF_WSCALE_FLAG) {
3965                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3966                                     pd->af);
3967                                 if (src->wscale & PF_WSCALE_FLAG) {
3968                                         /* Remove scale factor from initial
3969                                          * window */
3970                                         sws = src->wscale & PF_WSCALE_MASK;
3971                                         win = ((u_int32_t)win + (1 << sws) - 1)
3972                                             >> sws;
3973                                         dws = dst->wscale & PF_WSCALE_MASK;
3974                                 } else {
3975                                         /* fixup other window */
3976                                         dst->max_win <<= dst->wscale &
3977                                             PF_WSCALE_MASK;
3978                                         /* in case of a retrans SYN|ACK */
3979                                         dst->wscale = 0;
3980                                 }
3981                         }
3982                 }
3983                 if (th->th_flags & TH_FIN)
3984                         end++;
3985
3986                 src->seqlo = seq;
3987                 if (src->state < TCPS_SYN_SENT)
3988                         src->state = TCPS_SYN_SENT;
3989
3990                 /*
3991                  * May need to slide the window (seqhi may have been set by
3992                  * the crappy stack check or if we picked up the connection
3993                  * after establishment)
3994                  */
3995                 if (src->seqhi == 1 ||
3996                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3997                         src->seqhi = end + MAX(1, dst->max_win << dws);
3998                 if (win > src->max_win)
3999                         src->max_win = win;
4000
4001         } else {
4002                 ack = ntohl(th->th_ack) - dst->seqdiff;
4003                 if (src->seqdiff) {
4004                         /* Modulate sequence numbers */
4005                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4006                             src->seqdiff), 0);
4007                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4008                         *copyback = 1;
4009                 }
4010                 end = seq + pd->p_len;
4011                 if (th->th_flags & TH_SYN)
4012                         end++;
4013                 if (th->th_flags & TH_FIN)
4014                         end++;
4015         }
4016
4017         if ((th->th_flags & TH_ACK) == 0) {
4018                 /* Let it pass through the ack skew check */
4019                 ack = dst->seqlo;
4020         } else if ((ack == 0 &&
4021             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4022             /* broken tcp stacks do not set ack */
4023             (dst->state < TCPS_SYN_SENT)) {
4024                 /*
4025                  * Many stacks (ours included) will set the ACK number in an
4026                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4027                  */
4028                 ack = dst->seqlo;
4029         }
4030
4031         if (seq == end) {
4032                 /* Ease sequencing restrictions on no data packets */
4033                 seq = src->seqlo;
4034                 end = seq;
4035         }
4036
4037         ackskew = dst->seqlo - ack;
4038
4039
4040         /*
4041          * Need to demodulate the sequence numbers in any TCP SACK options
4042          * (Selective ACK). We could optionally validate the SACK values
4043          * against the current ACK window, either forwards or backwards, but
4044          * I'm not confident that SACK has been implemented properly
4045          * everywhere. It wouldn't surprise me if several stacks accidentally
4046          * SACK too far backwards of previously ACKed data. There really aren't
4047          * any security implications of bad SACKing unless the target stack
4048          * doesn't validate the option length correctly. Someone trying to
4049          * spoof into a TCP connection won't bother blindly sending SACK
4050          * options anyway.
4051          */
4052         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4053                 if (pf_modulate_sack(m, off, pd, th, dst))
4054                         *copyback = 1;
4055         }
4056
4057
4058 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4059         if (SEQ_GEQ(src->seqhi, end) &&
4060             /* Last octet inside other's window space */
4061             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4062             /* Retrans: not more than one window back */
4063             (ackskew >= -MAXACKWINDOW) &&
4064             /* Acking not more than one reassembled fragment backwards */
4065             (ackskew <= (MAXACKWINDOW << sws)) &&
4066             /* Acking not more than one window forward */
4067             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4068             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4069             (pd->flags & PFDESC_IP_REAS) == 0)) {
4070             /* Require an exact/+1 sequence match on resets when possible */
4071
4072                 if (dst->scrub || src->scrub) {
4073                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4074                             *state, src, dst, copyback))
4075                                 return (PF_DROP);
4076                 }
4077
4078                 /* update max window */
4079                 if (src->max_win < win)
4080                         src->max_win = win;
4081                 /* synchronize sequencing */
4082                 if (SEQ_GT(end, src->seqlo))
4083                         src->seqlo = end;
4084                 /* slide the window of what the other end can send */
4085                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4086                         dst->seqhi = ack + MAX((win << sws), 1);
4087
4088
4089                 /* update states */
4090                 if (th->th_flags & TH_SYN)
4091                         if (src->state < TCPS_SYN_SENT)
4092                                 src->state = TCPS_SYN_SENT;
4093                 if (th->th_flags & TH_FIN)
4094                         if (src->state < TCPS_CLOSING)
4095                                 src->state = TCPS_CLOSING;
4096                 if (th->th_flags & TH_ACK) {
4097                         if (dst->state == TCPS_SYN_SENT) {
4098                                 dst->state = TCPS_ESTABLISHED;
4099                                 if (src->state == TCPS_ESTABLISHED &&
4100                                     (*state)->src_node != NULL &&
4101                                     pf_src_connlimit(state)) {
4102                                         REASON_SET(reason, PFRES_SRCLIMIT);
4103                                         return (PF_DROP);
4104                                 }
4105                         } else if (dst->state == TCPS_CLOSING)
4106                                 dst->state = TCPS_FIN_WAIT_2;
4107                 }
4108                 if (th->th_flags & TH_RST)
4109                         src->state = dst->state = TCPS_TIME_WAIT;
4110
4111                 /* update expire time */
4112                 (*state)->expire = time_uptime;
4113                 if (src->state >= TCPS_FIN_WAIT_2 &&
4114                     dst->state >= TCPS_FIN_WAIT_2)
4115                         (*state)->timeout = PFTM_TCP_CLOSED;
4116                 else if (src->state >= TCPS_CLOSING &&
4117                     dst->state >= TCPS_CLOSING)
4118                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4119                 else if (src->state < TCPS_ESTABLISHED ||
4120                     dst->state < TCPS_ESTABLISHED)
4121                         (*state)->timeout = PFTM_TCP_OPENING;
4122                 else if (src->state >= TCPS_CLOSING ||
4123                     dst->state >= TCPS_CLOSING)
4124                         (*state)->timeout = PFTM_TCP_CLOSING;
4125                 else
4126                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4127
4128                 /* Fall through to PASS packet */
4129
4130         } else if ((dst->state < TCPS_SYN_SENT ||
4131                 dst->state >= TCPS_FIN_WAIT_2 ||
4132                 src->state >= TCPS_FIN_WAIT_2) &&
4133             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4134             /* Within a window forward of the originating packet */
4135             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4136             /* Within a window backward of the originating packet */
4137
4138                 /*
4139                  * This currently handles three situations:
4140                  *  1) Stupid stacks will shotgun SYNs before their peer
4141                  *     replies.
4142                  *  2) When PF catches an already established stream (the
4143                  *     firewall rebooted, the state table was flushed, routes
4144                  *     changed...)
4145                  *  3) Packets get funky immediately after the connection
4146                  *     closes (this should catch Solaris spurious ACK|FINs
4147                  *     that web servers like to spew after a close)
4148                  *
4149                  * This must be a little more careful than the above code
4150                  * since packet floods will also be caught here. We don't
4151                  * update the TTL here to mitigate the damage of a packet
4152                  * flood and so the same code can handle awkward establishment
4153                  * and a loosened connection close.
4154                  * In the establishment case, a correct peer response will
4155                  * validate the connection, go through the normal state code
4156                  * and keep updating the state TTL.
4157                  */
4158
4159                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4160                         printf("pf: loose state match: ");
4161                         pf_print_state(*state);
4162                         pf_print_flags(th->th_flags);
4163                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4164                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4165                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4166                             (unsigned long long)(*state)->packets[1],
4167                             pd->dir == PF_IN ? "in" : "out",
4168                             pd->dir == (*state)->direction ? "fwd" : "rev");
4169                 }
4170
4171                 if (dst->scrub || src->scrub) {
4172                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4173                             *state, src, dst, copyback))
4174                                 return (PF_DROP);
4175                 }
4176
4177                 /* update max window */
4178                 if (src->max_win < win)
4179                         src->max_win = win;
4180                 /* synchronize sequencing */
4181                 if (SEQ_GT(end, src->seqlo))
4182                         src->seqlo = end;
4183                 /* slide the window of what the other end can send */
4184                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4185                         dst->seqhi = ack + MAX((win << sws), 1);
4186
4187                 /*
4188                  * Cannot set dst->seqhi here since this could be a shotgunned
4189                  * SYN and not an already established connection.
4190                  */
4191
4192                 if (th->th_flags & TH_FIN)
4193                         if (src->state < TCPS_CLOSING)
4194                                 src->state = TCPS_CLOSING;
4195                 if (th->th_flags & TH_RST)
4196                         src->state = dst->state = TCPS_TIME_WAIT;
4197
4198                 /* Fall through to PASS packet */
4199
4200         } else {
4201                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4202                     (*state)->src.state == TCPS_SYN_SENT) {
4203                         /* Send RST for state mismatches during handshake */
4204                         if (!(th->th_flags & TH_RST))
4205                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4206                                     pd->dst, pd->src, th->th_dport,
4207                                     th->th_sport, ntohl(th->th_ack), 0,
4208                                     TH_RST, 0, 0,
4209                                     (*state)->rule.ptr->return_ttl, 1, 0,
4210                                     kif->pfik_ifp);
4211                         src->seqlo = 0;
4212                         src->seqhi = 1;
4213                         src->max_win = 1;
4214                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4215                         printf("pf: BAD state: ");
4216                         pf_print_state(*state);
4217                         pf_print_flags(th->th_flags);
4218                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4219                             "pkts=%llu:%llu dir=%s,%s\n",
4220                             seq, orig_seq, ack, pd->p_len, ackskew,
4221                             (unsigned long long)(*state)->packets[0],
4222                             (unsigned long long)(*state)->packets[1],
4223                             pd->dir == PF_IN ? "in" : "out",
4224                             pd->dir == (*state)->direction ? "fwd" : "rev");
4225                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4226                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4227                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4228                             ' ': '2',
4229                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4230                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4231                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4232                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4233                 }
4234                 REASON_SET(reason, PFRES_BADSTATE);
4235                 return (PF_DROP);
4236         }
4237
4238         return (PF_PASS);
4239 }
4240
4241 static int
4242 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4243         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4244 {
4245         struct tcphdr           *th = pd->hdr.tcp;
4246
4247         if (th->th_flags & TH_SYN)
4248                 if (src->state < TCPS_SYN_SENT)
4249                         src->state = TCPS_SYN_SENT;
4250         if (th->th_flags & TH_FIN)
4251                 if (src->state < TCPS_CLOSING)
4252                         src->state = TCPS_CLOSING;
4253         if (th->th_flags & TH_ACK) {
4254                 if (dst->state == TCPS_SYN_SENT) {
4255                         dst->state = TCPS_ESTABLISHED;
4256                         if (src->state == TCPS_ESTABLISHED &&
4257                             (*state)->src_node != NULL &&
4258                             pf_src_connlimit(state)) {
4259                                 REASON_SET(reason, PFRES_SRCLIMIT);
4260                                 return (PF_DROP);
4261                         }
4262                 } else if (dst->state == TCPS_CLOSING) {
4263                         dst->state = TCPS_FIN_WAIT_2;
4264                 } else if (src->state == TCPS_SYN_SENT &&
4265                     dst->state < TCPS_SYN_SENT) {
4266                         /*
4267                          * Handle a special sloppy case where we only see one
4268                          * half of the connection. If there is a ACK after
4269                          * the initial SYN without ever seeing a packet from
4270                          * the destination, set the connection to established.
4271                          */
4272                         dst->state = src->state = TCPS_ESTABLISHED;
4273                         if ((*state)->src_node != NULL &&
4274                             pf_src_connlimit(state)) {
4275                                 REASON_SET(reason, PFRES_SRCLIMIT);
4276                                 return (PF_DROP);
4277                         }
4278                 } else if (src->state == TCPS_CLOSING &&
4279                     dst->state == TCPS_ESTABLISHED &&
4280                     dst->seqlo == 0) {
4281                         /*
4282                          * Handle the closing of half connections where we
4283                          * don't see the full bidirectional FIN/ACK+ACK
4284                          * handshake.
4285                          */
4286                         dst->state = TCPS_CLOSING;
4287                 }
4288         }
4289         if (th->th_flags & TH_RST)
4290                 src->state = dst->state = TCPS_TIME_WAIT;
4291
4292         /* update expire time */
4293         (*state)->expire = time_uptime;
4294         if (src->state >= TCPS_FIN_WAIT_2 &&
4295             dst->state >= TCPS_FIN_WAIT_2)
4296                 (*state)->timeout = PFTM_TCP_CLOSED;
4297         else if (src->state >= TCPS_CLOSING &&
4298             dst->state >= TCPS_CLOSING)
4299                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4300         else if (src->state < TCPS_ESTABLISHED ||
4301             dst->state < TCPS_ESTABLISHED)
4302                 (*state)->timeout = PFTM_TCP_OPENING;
4303         else if (src->state >= TCPS_CLOSING ||
4304             dst->state >= TCPS_CLOSING)
4305                 (*state)->timeout = PFTM_TCP_CLOSING;
4306         else
4307                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4308
4309         return (PF_PASS);
4310 }
4311
4312 static int
4313 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4314     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4315     u_short *reason)
4316 {
4317         struct pf_state_key_cmp  key;
4318         struct tcphdr           *th = pd->hdr.tcp;
4319         int                      copyback = 0;
4320         struct pf_state_peer    *src, *dst;
4321         struct pf_state_key     *sk;
4322
4323         bzero(&key, sizeof(key));
4324         key.af = pd->af;
4325         key.proto = IPPROTO_TCP;
4326         if (direction == PF_IN) {       /* wire side, straight */
4327                 PF_ACPY(&key.addr[0], pd->src, key.af);
4328                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4329                 key.port[0] = th->th_sport;
4330                 key.port[1] = th->th_dport;
4331         } else {                        /* stack side, reverse */
4332                 PF_ACPY(&key.addr[1], pd->src, key.af);
4333                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4334                 key.port[1] = th->th_sport;
4335                 key.port[0] = th->th_dport;
4336         }
4337
4338         STATE_LOOKUP(kif, &key, direction, *state, pd);
4339
4340         if (direction == (*state)->direction) {
4341                 src = &(*state)->src;
4342                 dst = &(*state)->dst;
4343         } else {
4344                 src = &(*state)->dst;
4345                 dst = &(*state)->src;
4346         }
4347
4348         sk = (*state)->key[pd->didx];
4349
4350         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4351                 if (direction != (*state)->direction) {
4352                         REASON_SET(reason, PFRES_SYNPROXY);
4353                         return (PF_SYNPROXY_DROP);
4354                 }
4355                 if (th->th_flags & TH_SYN) {
4356                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4357                                 REASON_SET(reason, PFRES_SYNPROXY);
4358                                 return (PF_DROP);
4359                         }
4360                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4361                             pd->src, th->th_dport, th->th_sport,
4362                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4363                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4364                         REASON_SET(reason, PFRES_SYNPROXY);
4365                         return (PF_SYNPROXY_DROP);
4366                 } else if (!(th->th_flags & TH_ACK) ||
4367                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4368                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4369                         REASON_SET(reason, PFRES_SYNPROXY);
4370                         return (PF_DROP);
4371                 } else if ((*state)->src_node != NULL &&
4372                     pf_src_connlimit(state)) {
4373                         REASON_SET(reason, PFRES_SRCLIMIT);
4374                         return (PF_DROP);
4375                 } else
4376                         (*state)->src.state = PF_TCPS_PROXY_DST;
4377         }
4378         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4379                 if (direction == (*state)->direction) {
4380                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4381                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4382                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4383                                 REASON_SET(reason, PFRES_SYNPROXY);
4384                                 return (PF_DROP);
4385                         }
4386                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4387                         if ((*state)->dst.seqhi == 1)
4388                                 (*state)->dst.seqhi = htonl(arc4random());
4389                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4390                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4391                             sk->port[pd->sidx], sk->port[pd->didx],
4392                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4393                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4394                         REASON_SET(reason, PFRES_SYNPROXY);
4395                         return (PF_SYNPROXY_DROP);
4396                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4397                     (TH_SYN|TH_ACK)) ||
4398                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4399                         REASON_SET(reason, PFRES_SYNPROXY);
4400                         return (PF_DROP);
4401                 } else {
4402                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4403                         (*state)->dst.seqlo = ntohl(th->th_seq);
4404                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4405                             pd->src, th->th_dport, th->th_sport,
4406                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4407                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4408                             (*state)->tag, NULL);
4409                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4410                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4411                             sk->port[pd->sidx], sk->port[pd->didx],
4412                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4413                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4414                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4415                             (*state)->src.seqlo;
4416                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4417                             (*state)->dst.seqlo;
4418                         (*state)->src.seqhi = (*state)->src.seqlo +
4419                             (*state)->dst.max_win;
4420                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4421                             (*state)->src.max_win;
4422                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4423                         (*state)->src.state = (*state)->dst.state =
4424                             TCPS_ESTABLISHED;
4425                         REASON_SET(reason, PFRES_SYNPROXY);
4426                         return (PF_SYNPROXY_DROP);
4427                 }
4428         }
4429
4430         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4431             dst->state >= TCPS_FIN_WAIT_2 &&
4432             src->state >= TCPS_FIN_WAIT_2) {
4433                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4434                         printf("pf: state reuse ");
4435                         pf_print_state(*state);
4436                         pf_print_flags(th->th_flags);
4437                         printf("\n");
4438                 }
4439                 /* XXX make sure it's the same direction ?? */
4440                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4441                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4442                 *state = NULL;
4443                 return (PF_DROP);
4444         }
4445
4446         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4447                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4448                         return (PF_DROP);
4449         } else {
4450                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4451                     &copyback) == PF_DROP)
4452                         return (PF_DROP);
4453         }
4454
4455         /* translate source/destination address, if necessary */
4456         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4457                 struct pf_state_key *nk = (*state)->key[pd->didx];
4458
4459                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4460                     nk->port[pd->sidx] != th->th_sport)
4461                         pf_change_ap(m, pd->src, &th->th_sport,
4462                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4463                             nk->port[pd->sidx], 0, pd->af);
4464
4465                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4466                     nk->port[pd->didx] != th->th_dport)
4467                         pf_change_ap(m, pd->dst, &th->th_dport,
4468                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4469                             nk->port[pd->didx], 0, pd->af);
4470                 copyback = 1;
4471         }
4472
4473         /* Copyback sequence modulation or stateful scrub changes if needed */
4474         if (copyback)
4475                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4476
4477         return (PF_PASS);
4478 }
4479
4480 static int
4481 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4482     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4483 {
4484         struct pf_state_peer    *src, *dst;
4485         struct pf_state_key_cmp  key;
4486         struct udphdr           *uh = pd->hdr.udp;
4487
4488         bzero(&key, sizeof(key));
4489         key.af = pd->af;
4490         key.proto = IPPROTO_UDP;
4491         if (direction == PF_IN) {       /* wire side, straight */
4492                 PF_ACPY(&key.addr[0], pd->src, key.af);
4493                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4494                 key.port[0] = uh->uh_sport;
4495                 key.port[1] = uh->uh_dport;
4496         } else {                        /* stack side, reverse */
4497                 PF_ACPY(&key.addr[1], pd->src, key.af);
4498                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4499                 key.port[1] = uh->uh_sport;
4500                 key.port[0] = uh->uh_dport;
4501         }
4502
4503         STATE_LOOKUP(kif, &key, direction, *state, pd);
4504
4505         if (direction == (*state)->direction) {
4506                 src = &(*state)->src;
4507                 dst = &(*state)->dst;
4508         } else {
4509                 src = &(*state)->dst;
4510                 dst = &(*state)->src;
4511         }
4512
4513         /* update states */
4514         if (src->state < PFUDPS_SINGLE)
4515                 src->state = PFUDPS_SINGLE;
4516         if (dst->state == PFUDPS_SINGLE)
4517                 dst->state = PFUDPS_MULTIPLE;
4518
4519         /* update expire time */
4520         (*state)->expire = time_uptime;
4521         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4522                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4523         else
4524                 (*state)->timeout = PFTM_UDP_SINGLE;
4525
4526         /* translate source/destination address, if necessary */
4527         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4528                 struct pf_state_key *nk = (*state)->key[pd->didx];
4529
4530                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4531                     nk->port[pd->sidx] != uh->uh_sport)
4532                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4533                             &uh->uh_sum, &nk->addr[pd->sidx],
4534                             nk->port[pd->sidx], 1, pd->af);
4535
4536                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4537                     nk->port[pd->didx] != uh->uh_dport)
4538                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4539                             &uh->uh_sum, &nk->addr[pd->didx],
4540                             nk->port[pd->didx], 1, pd->af);
4541                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4542         }
4543
4544         return (PF_PASS);
4545 }
4546
4547 static int
4548 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4549     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4550 {
4551         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4552         u_int16_t        icmpid = 0, *icmpsum;
4553         u_int8_t         icmptype, icmpcode;
4554         int              state_icmp = 0;
4555         struct pf_state_key_cmp key;
4556
4557         bzero(&key, sizeof(key));
4558         switch (pd->proto) {
4559 #ifdef INET
4560         case IPPROTO_ICMP:
4561                 icmptype = pd->hdr.icmp->icmp_type;
4562                 icmpcode = pd->hdr.icmp->icmp_code;
4563                 icmpid = pd->hdr.icmp->icmp_id;
4564                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4565
4566                 if (icmptype == ICMP_UNREACH ||
4567                     icmptype == ICMP_SOURCEQUENCH ||
4568                     icmptype == ICMP_REDIRECT ||
4569                     icmptype == ICMP_TIMXCEED ||
4570                     icmptype == ICMP_PARAMPROB)
4571                         state_icmp++;
4572                 break;
4573 #endif /* INET */
4574 #ifdef INET6
4575         case IPPROTO_ICMPV6:
4576                 icmptype = pd->hdr.icmp6->icmp6_type;
4577                 icmpcode = pd->hdr.icmp6->icmp6_code;
4578                 icmpid = pd->hdr.icmp6->icmp6_id;
4579                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4580
4581                 if (icmptype == ICMP6_DST_UNREACH ||
4582                     icmptype == ICMP6_PACKET_TOO_BIG ||
4583                     icmptype == ICMP6_TIME_EXCEEDED ||
4584                     icmptype == ICMP6_PARAM_PROB)
4585                         state_icmp++;
4586                 break;
4587 #endif /* INET6 */
4588         }
4589
4590         if (!state_icmp) {
4591
4592                 /*
4593                  * ICMP query/reply message not related to a TCP/UDP packet.
4594                  * Search for an ICMP state.
4595                  */
4596                 key.af = pd->af;
4597                 key.proto = pd->proto;
4598                 key.port[0] = key.port[1] = icmpid;
4599                 if (direction == PF_IN) {       /* wire side, straight */
4600                         PF_ACPY(&key.addr[0], pd->src, key.af);
4601                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4602                 } else {                        /* stack side, reverse */
4603                         PF_ACPY(&key.addr[1], pd->src, key.af);
4604                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4605                 }
4606
4607                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4608
4609                 (*state)->expire = time_uptime;
4610                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4611
4612                 /* translate source/destination address, if necessary */
4613                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4614                         struct pf_state_key *nk = (*state)->key[pd->didx];
4615
4616                         switch (pd->af) {
4617 #ifdef INET
4618                         case AF_INET:
4619                                 if (PF_ANEQ(pd->src,
4620                                     &nk->addr[pd->sidx], AF_INET))
4621                                         pf_change_a(&saddr->v4.s_addr,
4622                                             pd->ip_sum,
4623                                             nk->addr[pd->sidx].v4.s_addr, 0);
4624
4625                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4626                                     AF_INET))
4627                                         pf_change_a(&daddr->v4.s_addr,
4628                                             pd->ip_sum,
4629                                             nk->addr[pd->didx].v4.s_addr, 0);
4630
4631                                 if (nk->port[0] !=
4632                                     pd->hdr.icmp->icmp_id) {
4633                                         pd->hdr.icmp->icmp_cksum =
4634                                             pf_cksum_fixup(
4635                                             pd->hdr.icmp->icmp_cksum, icmpid,
4636                                             nk->port[pd->sidx], 0);
4637                                         pd->hdr.icmp->icmp_id =
4638                                             nk->port[pd->sidx];
4639                                 }
4640
4641                                 m_copyback(m, off, ICMP_MINLEN,
4642                                     (caddr_t )pd->hdr.icmp);
4643                                 break;
4644 #endif /* INET */
4645 #ifdef INET6
4646                         case AF_INET6:
4647                                 if (PF_ANEQ(pd->src,
4648                                     &nk->addr[pd->sidx], AF_INET6))
4649                                         pf_change_a6(saddr,
4650                                             &pd->hdr.icmp6->icmp6_cksum,
4651                                             &nk->addr[pd->sidx], 0);
4652
4653                                 if (PF_ANEQ(pd->dst,
4654                                     &nk->addr[pd->didx], AF_INET6))
4655                                         pf_change_a6(daddr,
4656                                             &pd->hdr.icmp6->icmp6_cksum,
4657                                             &nk->addr[pd->didx], 0);
4658
4659                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4660                                     (caddr_t )pd->hdr.icmp6);
4661                                 break;
4662 #endif /* INET6 */
4663                         }
4664                 }
4665                 return (PF_PASS);
4666
4667         } else {
4668                 /*
4669                  * ICMP error message in response to a TCP/UDP packet.
4670                  * Extract the inner TCP/UDP header and search for that state.
4671                  */
4672
4673                 struct pf_pdesc pd2;
4674                 bzero(&pd2, sizeof pd2);
4675 #ifdef INET
4676                 struct ip       h2;
4677 #endif /* INET */
4678 #ifdef INET6
4679                 struct ip6_hdr  h2_6;
4680                 int             terminal = 0;
4681 #endif /* INET6 */
4682                 int             ipoff2 = 0;
4683                 int             off2 = 0;
4684
4685                 pd2.af = pd->af;
4686                 /* Payload packet is from the opposite direction. */
4687                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4688                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4689                 switch (pd->af) {
4690 #ifdef INET
4691                 case AF_INET:
4692                         /* offset of h2 in mbuf chain */
4693                         ipoff2 = off + ICMP_MINLEN;
4694
4695                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4696                             NULL, reason, pd2.af)) {
4697                                 DPFPRINTF(PF_DEBUG_MISC,
4698                                     ("pf: ICMP error message too short "
4699                                     "(ip)\n"));
4700                                 return (PF_DROP);
4701                         }
4702                         /*
4703                          * ICMP error messages don't refer to non-first
4704                          * fragments
4705                          */
4706                         if (h2.ip_off & htons(IP_OFFMASK)) {
4707                                 REASON_SET(reason, PFRES_FRAG);
4708                                 return (PF_DROP);
4709                         }
4710
4711                         /* offset of protocol header that follows h2 */
4712                         off2 = ipoff2 + (h2.ip_hl << 2);
4713
4714                         pd2.proto = h2.ip_p;
4715                         pd2.src = (struct pf_addr *)&h2.ip_src;
4716                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4717                         pd2.ip_sum = &h2.ip_sum;
4718                         break;
4719 #endif /* INET */
4720 #ifdef INET6
4721                 case AF_INET6:
4722                         ipoff2 = off + sizeof(struct icmp6_hdr);
4723
4724                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4725                             NULL, reason, pd2.af)) {
4726                                 DPFPRINTF(PF_DEBUG_MISC,
4727                                     ("pf: ICMP error message too short "
4728                                     "(ip6)\n"));
4729                                 return (PF_DROP);
4730                         }
4731                         pd2.proto = h2_6.ip6_nxt;
4732                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4733                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4734                         pd2.ip_sum = NULL;
4735                         off2 = ipoff2 + sizeof(h2_6);
4736                         do {
4737                                 switch (pd2.proto) {
4738                                 case IPPROTO_FRAGMENT:
4739                                         /*
4740                                          * ICMPv6 error messages for
4741                                          * non-first fragments
4742                                          */
4743                                         REASON_SET(reason, PFRES_FRAG);
4744                                         return (PF_DROP);
4745                                 case IPPROTO_AH:
4746                                 case IPPROTO_HOPOPTS:
4747                                 case IPPROTO_ROUTING:
4748                                 case IPPROTO_DSTOPTS: {
4749                                         /* get next header and header length */
4750                                         struct ip6_ext opt6;
4751
4752                                         if (!pf_pull_hdr(m, off2, &opt6,
4753                                             sizeof(opt6), NULL, reason,
4754                                             pd2.af)) {
4755                                                 DPFPRINTF(PF_DEBUG_MISC,
4756                                                     ("pf: ICMPv6 short opt\n"));
4757                                                 return (PF_DROP);
4758                                         }
4759                                         if (pd2.proto == IPPROTO_AH)
4760                                                 off2 += (opt6.ip6e_len + 2) * 4;
4761                                         else
4762                                                 off2 += (opt6.ip6e_len + 1) * 8;
4763                                         pd2.proto = opt6.ip6e_nxt;
4764                                         /* goto the next header */
4765                                         break;
4766                                 }
4767                                 default:
4768                                         terminal++;
4769                                         break;
4770                                 }
4771                         } while (!terminal);
4772                         break;
4773 #endif /* INET6 */
4774                 }
4775
4776                 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
4777                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
4778                                 printf("pf: BAD ICMP %d:%d outer dst: ",
4779                                     icmptype, icmpcode);
4780                                 pf_print_host(pd->src, 0, pd->af);
4781                                 printf(" -> ");
4782                                 pf_print_host(pd->dst, 0, pd->af);
4783                                 printf(" inner src: ");
4784                                 pf_print_host(pd2.src, 0, pd2.af);
4785                                 printf(" -> ");
4786                                 pf_print_host(pd2.dst, 0, pd2.af);
4787                                 printf("\n");
4788                         }
4789                         REASON_SET(reason, PFRES_BADSTATE);
4790                         return (PF_DROP);
4791                 }
4792
4793                 switch (pd2.proto) {
4794                 case IPPROTO_TCP: {
4795                         struct tcphdr            th;
4796                         u_int32_t                seq;
4797                         struct pf_state_peer    *src, *dst;
4798                         u_int8_t                 dws;
4799                         int                      copyback = 0;
4800
4801                         /*
4802                          * Only the first 8 bytes of the TCP header can be
4803                          * expected. Don't access any TCP header fields after
4804                          * th_seq, an ackskew test is not possible.
4805                          */
4806                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4807                             pd2.af)) {
4808                                 DPFPRINTF(PF_DEBUG_MISC,
4809                                     ("pf: ICMP error message too short "
4810                                     "(tcp)\n"));
4811                                 return (PF_DROP);
4812                         }
4813
4814                         key.af = pd2.af;
4815                         key.proto = IPPROTO_TCP;
4816                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4817                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4818                         key.port[pd2.sidx] = th.th_sport;
4819                         key.port[pd2.didx] = th.th_dport;
4820
4821                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4822
4823                         if (direction == (*state)->direction) {
4824                                 src = &(*state)->dst;
4825                                 dst = &(*state)->src;
4826                         } else {
4827                                 src = &(*state)->src;
4828                                 dst = &(*state)->dst;
4829                         }
4830
4831                         if (src->wscale && dst->wscale)
4832                                 dws = dst->wscale & PF_WSCALE_MASK;
4833                         else
4834                                 dws = 0;
4835
4836                         /* Demodulate sequence number */
4837                         seq = ntohl(th.th_seq) - src->seqdiff;
4838                         if (src->seqdiff) {
4839                                 pf_change_a(&th.th_seq, icmpsum,
4840                                     htonl(seq), 0);
4841                                 copyback = 1;
4842                         }
4843
4844                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4845                             (!SEQ_GEQ(src->seqhi, seq) ||
4846                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4847                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4848                                         printf("pf: BAD ICMP %d:%d ",
4849                                             icmptype, icmpcode);
4850                                         pf_print_host(pd->src, 0, pd->af);
4851                                         printf(" -> ");
4852                                         pf_print_host(pd->dst, 0, pd->af);
4853                                         printf(" state: ");
4854                                         pf_print_state(*state);
4855                                         printf(" seq=%u\n", seq);
4856                                 }
4857                                 REASON_SET(reason, PFRES_BADSTATE);
4858                                 return (PF_DROP);
4859                         } else {
4860                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4861                                         printf("pf: OK ICMP %d:%d ",
4862                                             icmptype, icmpcode);
4863                                         pf_print_host(pd->src, 0, pd->af);
4864                                         printf(" -> ");
4865                                         pf_print_host(pd->dst, 0, pd->af);
4866                                         printf(" state: ");
4867                                         pf_print_state(*state);
4868                                         printf(" seq=%u\n", seq);
4869                                 }
4870                         }
4871
4872                         /* translate source/destination address, if necessary */
4873                         if ((*state)->key[PF_SK_WIRE] !=
4874                             (*state)->key[PF_SK_STACK]) {
4875                                 struct pf_state_key *nk =
4876                                     (*state)->key[pd->didx];
4877
4878                                 if (PF_ANEQ(pd2.src,
4879                                     &nk->addr[pd2.sidx], pd2.af) ||
4880                                     nk->port[pd2.sidx] != th.th_sport)
4881                                         pf_change_icmp(pd2.src, &th.th_sport,
4882                                             daddr, &nk->addr[pd2.sidx],
4883                                             nk->port[pd2.sidx], NULL,
4884                                             pd2.ip_sum, icmpsum,
4885                                             pd->ip_sum, 0, pd2.af);
4886
4887                                 if (PF_ANEQ(pd2.dst,
4888                                     &nk->addr[pd2.didx], pd2.af) ||
4889                                     nk->port[pd2.didx] != th.th_dport)
4890                                         pf_change_icmp(pd2.dst, &th.th_dport,
4891                                             saddr, &nk->addr[pd2.didx],
4892                                             nk->port[pd2.didx], NULL,
4893                                             pd2.ip_sum, icmpsum,
4894                                             pd->ip_sum, 0, pd2.af);
4895                                 copyback = 1;
4896                         }
4897
4898                         if (copyback) {
4899                                 switch (pd2.af) {
4900 #ifdef INET
4901                                 case AF_INET:
4902                                         m_copyback(m, off, ICMP_MINLEN,
4903                                             (caddr_t )pd->hdr.icmp);
4904                                         m_copyback(m, ipoff2, sizeof(h2),
4905                                             (caddr_t )&h2);
4906                                         break;
4907 #endif /* INET */
4908 #ifdef INET6
4909                                 case AF_INET6:
4910                                         m_copyback(m, off,
4911                                             sizeof(struct icmp6_hdr),
4912                                             (caddr_t )pd->hdr.icmp6);
4913                                         m_copyback(m, ipoff2, sizeof(h2_6),
4914                                             (caddr_t )&h2_6);
4915                                         break;
4916 #endif /* INET6 */
4917                                 }
4918                                 m_copyback(m, off2, 8, (caddr_t)&th);
4919                         }
4920
4921                         return (PF_PASS);
4922                         break;
4923                 }
4924                 case IPPROTO_UDP: {
4925                         struct udphdr           uh;
4926
4927                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4928                             NULL, reason, pd2.af)) {
4929                                 DPFPRINTF(PF_DEBUG_MISC,
4930                                     ("pf: ICMP error message too short "
4931                                     "(udp)\n"));
4932                                 return (PF_DROP);
4933                         }
4934
4935                         key.af = pd2.af;
4936                         key.proto = IPPROTO_UDP;
4937                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4938                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4939                         key.port[pd2.sidx] = uh.uh_sport;
4940                         key.port[pd2.didx] = uh.uh_dport;
4941
4942                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4943
4944                         /* translate source/destination address, if necessary */
4945                         if ((*state)->key[PF_SK_WIRE] !=
4946                             (*state)->key[PF_SK_STACK]) {
4947                                 struct pf_state_key *nk =
4948                                     (*state)->key[pd->didx];
4949
4950                                 if (PF_ANEQ(pd2.src,
4951                                     &nk->addr[pd2.sidx], pd2.af) ||
4952                                     nk->port[pd2.sidx] != uh.uh_sport)
4953                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4954                                             daddr, &nk->addr[pd2.sidx],
4955                                             nk->port[pd2.sidx], &uh.uh_sum,
4956                                             pd2.ip_sum, icmpsum,
4957                                             pd->ip_sum, 1, pd2.af);
4958
4959                                 if (PF_ANEQ(pd2.dst,
4960                                     &nk->addr[pd2.didx], pd2.af) ||
4961                                     nk->port[pd2.didx] != uh.uh_dport)
4962                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4963                                             saddr, &nk->addr[pd2.didx],
4964                                             nk->port[pd2.didx], &uh.uh_sum,
4965                                             pd2.ip_sum, icmpsum,
4966                                             pd->ip_sum, 1, pd2.af);
4967
4968                                 switch (pd2.af) {
4969 #ifdef INET
4970                                 case AF_INET:
4971                                         m_copyback(m, off, ICMP_MINLEN,
4972                                             (caddr_t )pd->hdr.icmp);
4973                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4974                                         break;
4975 #endif /* INET */
4976 #ifdef INET6
4977                                 case AF_INET6:
4978                                         m_copyback(m, off,
4979                                             sizeof(struct icmp6_hdr),
4980                                             (caddr_t )pd->hdr.icmp6);
4981                                         m_copyback(m, ipoff2, sizeof(h2_6),
4982                                             (caddr_t )&h2_6);
4983                                         break;
4984 #endif /* INET6 */
4985                                 }
4986                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4987                         }
4988                         return (PF_PASS);
4989                         break;
4990                 }
4991 #ifdef INET
4992                 case IPPROTO_ICMP: {
4993                         struct icmp             iih;
4994
4995                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4996                             NULL, reason, pd2.af)) {
4997                                 DPFPRINTF(PF_DEBUG_MISC,
4998                                     ("pf: ICMP error message too short i"
4999                                     "(icmp)\n"));
5000                                 return (PF_DROP);
5001                         }
5002
5003                         key.af = pd2.af;
5004                         key.proto = IPPROTO_ICMP;
5005                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5006                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5007                         key.port[0] = key.port[1] = iih.icmp_id;
5008
5009                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5010
5011                         /* translate source/destination address, if necessary */
5012                         if ((*state)->key[PF_SK_WIRE] !=
5013                             (*state)->key[PF_SK_STACK]) {
5014                                 struct pf_state_key *nk =
5015                                     (*state)->key[pd->didx];
5016
5017                                 if (PF_ANEQ(pd2.src,
5018                                     &nk->addr[pd2.sidx], pd2.af) ||
5019                                     nk->port[pd2.sidx] != iih.icmp_id)
5020                                         pf_change_icmp(pd2.src, &iih.icmp_id,
5021                                             daddr, &nk->addr[pd2.sidx],
5022                                             nk->port[pd2.sidx], NULL,
5023                                             pd2.ip_sum, icmpsum,
5024                                             pd->ip_sum, 0, AF_INET);
5025
5026                                 if (PF_ANEQ(pd2.dst,
5027                                     &nk->addr[pd2.didx], pd2.af) ||
5028                                     nk->port[pd2.didx] != iih.icmp_id)
5029                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
5030                                             saddr, &nk->addr[pd2.didx],
5031                                             nk->port[pd2.didx], NULL,
5032                                             pd2.ip_sum, icmpsum,
5033                                             pd->ip_sum, 0, AF_INET);
5034
5035                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5036                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5037                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5038                         }
5039                         return (PF_PASS);
5040                         break;
5041                 }
5042 #endif /* INET */
5043 #ifdef INET6
5044                 case IPPROTO_ICMPV6: {
5045                         struct icmp6_hdr        iih;
5046
5047                         if (!pf_pull_hdr(m, off2, &iih,
5048                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5049                                 DPFPRINTF(PF_DEBUG_MISC,
5050                                     ("pf: ICMP error message too short "
5051                                     "(icmp6)\n"));
5052                                 return (PF_DROP);
5053                         }
5054
5055                         key.af = pd2.af;
5056                         key.proto = IPPROTO_ICMPV6;
5057                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5058                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5059                         key.port[0] = key.port[1] = iih.icmp6_id;
5060
5061                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5062
5063                         /* translate source/destination address, if necessary */
5064                         if ((*state)->key[PF_SK_WIRE] !=
5065                             (*state)->key[PF_SK_STACK]) {
5066                                 struct pf_state_key *nk =
5067                                     (*state)->key[pd->didx];
5068
5069                                 if (PF_ANEQ(pd2.src,
5070                                     &nk->addr[pd2.sidx], pd2.af) ||
5071                                     nk->port[pd2.sidx] != iih.icmp6_id)
5072                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5073                                             daddr, &nk->addr[pd2.sidx],
5074                                             nk->port[pd2.sidx], NULL,
5075                                             pd2.ip_sum, icmpsum,
5076                                             pd->ip_sum, 0, AF_INET6);
5077
5078                                 if (PF_ANEQ(pd2.dst,
5079                                     &nk->addr[pd2.didx], pd2.af) ||
5080                                     nk->port[pd2.didx] != iih.icmp6_id)
5081                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5082                                             saddr, &nk->addr[pd2.didx],
5083                                             nk->port[pd2.didx], NULL,
5084                                             pd2.ip_sum, icmpsum,
5085                                             pd->ip_sum, 0, AF_INET6);
5086
5087                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5088                                     (caddr_t)pd->hdr.icmp6);
5089                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5090                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5091                                     (caddr_t)&iih);
5092                         }
5093                         return (PF_PASS);
5094                         break;
5095                 }
5096 #endif /* INET6 */
5097                 default: {
5098                         key.af = pd2.af;
5099                         key.proto = pd2.proto;
5100                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5101                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5102                         key.port[0] = key.port[1] = 0;
5103
5104                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5105
5106                         /* translate source/destination address, if necessary */
5107                         if ((*state)->key[PF_SK_WIRE] !=
5108                             (*state)->key[PF_SK_STACK]) {
5109                                 struct pf_state_key *nk =
5110                                     (*state)->key[pd->didx];
5111
5112                                 if (PF_ANEQ(pd2.src,
5113                                     &nk->addr[pd2.sidx], pd2.af))
5114                                         pf_change_icmp(pd2.src, NULL, daddr,
5115                                             &nk->addr[pd2.sidx], 0, NULL,
5116                                             pd2.ip_sum, icmpsum,
5117                                             pd->ip_sum, 0, pd2.af);
5118
5119                                 if (PF_ANEQ(pd2.dst,
5120                                     &nk->addr[pd2.didx], pd2.af))
5121                                         pf_change_icmp(pd2.dst, NULL, saddr,
5122                                             &nk->addr[pd2.didx], 0, NULL,
5123                                             pd2.ip_sum, icmpsum,
5124                                             pd->ip_sum, 0, pd2.af);
5125
5126                                 switch (pd2.af) {
5127 #ifdef INET
5128                                 case AF_INET:
5129                                         m_copyback(m, off, ICMP_MINLEN,
5130                                             (caddr_t)pd->hdr.icmp);
5131                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5132                                         break;
5133 #endif /* INET */
5134 #ifdef INET6
5135                                 case AF_INET6:
5136                                         m_copyback(m, off,
5137                                             sizeof(struct icmp6_hdr),
5138                                             (caddr_t )pd->hdr.icmp6);
5139                                         m_copyback(m, ipoff2, sizeof(h2_6),
5140                                             (caddr_t )&h2_6);
5141                                         break;
5142 #endif /* INET6 */
5143                                 }
5144                         }
5145                         return (PF_PASS);
5146                         break;
5147                 }
5148                 }
5149         }
5150 }
5151
5152 static int
5153 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5154     struct mbuf *m, struct pf_pdesc *pd)
5155 {
5156         struct pf_state_peer    *src, *dst;
5157         struct pf_state_key_cmp  key;
5158
5159         bzero(&key, sizeof(key));
5160         key.af = pd->af;
5161         key.proto = pd->proto;
5162         if (direction == PF_IN) {
5163                 PF_ACPY(&key.addr[0], pd->src, key.af);
5164                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5165                 key.port[0] = key.port[1] = 0;
5166         } else {
5167                 PF_ACPY(&key.addr[1], pd->src, key.af);
5168                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5169                 key.port[1] = key.port[0] = 0;
5170         }
5171
5172         STATE_LOOKUP(kif, &key, direction, *state, pd);
5173
5174         if (direction == (*state)->direction) {
5175                 src = &(*state)->src;
5176                 dst = &(*state)->dst;
5177         } else {
5178                 src = &(*state)->dst;
5179                 dst = &(*state)->src;
5180         }
5181
5182         /* update states */
5183         if (src->state < PFOTHERS_SINGLE)
5184                 src->state = PFOTHERS_SINGLE;
5185         if (dst->state == PFOTHERS_SINGLE)
5186                 dst->state = PFOTHERS_MULTIPLE;
5187
5188         /* update expire time */
5189         (*state)->expire = time_uptime;
5190         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5191                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5192         else
5193                 (*state)->timeout = PFTM_OTHER_SINGLE;
5194
5195         /* translate source/destination address, if necessary */
5196         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5197                 struct pf_state_key *nk = (*state)->key[pd->didx];
5198
5199                 KASSERT(nk, ("%s: nk is null", __func__));
5200                 KASSERT(pd, ("%s: pd is null", __func__));
5201                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5202                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5203                 switch (pd->af) {
5204 #ifdef INET
5205                 case AF_INET:
5206                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5207                                 pf_change_a(&pd->src->v4.s_addr,
5208                                     pd->ip_sum,
5209                                     nk->addr[pd->sidx].v4.s_addr,
5210                                     0);
5211
5212
5213                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5214                                 pf_change_a(&pd->dst->v4.s_addr,
5215                                     pd->ip_sum,
5216                                     nk->addr[pd->didx].v4.s_addr,
5217                                     0);
5218
5219                                 break;
5220 #endif /* INET */
5221 #ifdef INET6
5222                 case AF_INET6:
5223                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5224                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5225
5226                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5227                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5228 #endif /* INET6 */
5229                 }
5230         }
5231         return (PF_PASS);
5232 }
5233
5234 /*
5235  * ipoff and off are measured from the start of the mbuf chain.
5236  * h must be at "ipoff" on the mbuf chain.
5237  */
5238 void *
5239 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5240     u_short *actionp, u_short *reasonp, sa_family_t af)
5241 {
5242         switch (af) {
5243 #ifdef INET
5244         case AF_INET: {
5245                 struct ip       *h = mtod(m, struct ip *);
5246                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5247
5248                 if (fragoff) {
5249                         if (fragoff >= len)
5250                                 ACTION_SET(actionp, PF_PASS);
5251                         else {
5252                                 ACTION_SET(actionp, PF_DROP);
5253                                 REASON_SET(reasonp, PFRES_FRAG);
5254                         }
5255                         return (NULL);
5256                 }
5257                 if (m->m_pkthdr.len < off + len ||
5258                     ntohs(h->ip_len) < off + len) {
5259                         ACTION_SET(actionp, PF_DROP);
5260                         REASON_SET(reasonp, PFRES_SHORT);
5261                         return (NULL);
5262                 }
5263                 break;
5264         }
5265 #endif /* INET */
5266 #ifdef INET6
5267         case AF_INET6: {
5268                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5269
5270                 if (m->m_pkthdr.len < off + len ||
5271                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5272                     (unsigned)(off + len)) {
5273                         ACTION_SET(actionp, PF_DROP);
5274                         REASON_SET(reasonp, PFRES_SHORT);
5275                         return (NULL);
5276                 }
5277                 break;
5278         }
5279 #endif /* INET6 */
5280         }
5281         m_copydata(m, off, len, p);
5282         return (p);
5283 }
5284
5285 #ifdef RADIX_MPATH
5286 static int
5287 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5288     int rtableid)
5289 {
5290         struct radix_node_head  *rnh;
5291         struct sockaddr_in      *dst;
5292         int                      ret = 1;
5293         int                      check_mpath;
5294 #ifdef INET6
5295         struct sockaddr_in6     *dst6;
5296         struct route_in6         ro;
5297 #else
5298         struct route             ro;
5299 #endif
5300         struct radix_node       *rn;
5301         struct rtentry          *rt;
5302         struct ifnet            *ifp;
5303
5304         check_mpath = 0;
5305         /* XXX: stick to table 0 for now */
5306         rnh = rt_tables_get_rnh(0, af);
5307         if (rnh != NULL && rn_mpath_capable(rnh))
5308                 check_mpath = 1;
5309         bzero(&ro, sizeof(ro));
5310         switch (af) {
5311         case AF_INET:
5312                 dst = satosin(&ro.ro_dst);
5313                 dst->sin_family = AF_INET;
5314                 dst->sin_len = sizeof(*dst);
5315                 dst->sin_addr = addr->v4;
5316                 break;
5317 #ifdef INET6
5318         case AF_INET6:
5319                 /*
5320                  * Skip check for addresses with embedded interface scope,
5321                  * as they would always match anyway.
5322                  */
5323                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5324                         goto out;
5325                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5326                 dst6->sin6_family = AF_INET6;
5327                 dst6->sin6_len = sizeof(*dst6);
5328                 dst6->sin6_addr = addr->v6;
5329                 break;
5330 #endif /* INET6 */
5331         default:
5332                 return (0);
5333         }
5334
5335         /* Skip checks for ipsec interfaces */
5336         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5337                 goto out;
5338
5339         switch (af) {
5340 #ifdef INET6
5341         case AF_INET6:
5342                 in6_rtalloc_ign(&ro, 0, rtableid);
5343                 break;
5344 #endif
5345 #ifdef INET
5346         case AF_INET:
5347                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5348                 break;
5349 #endif
5350         }
5351
5352         if (ro.ro_rt != NULL) {
5353                 /* No interface given, this is a no-route check */
5354                 if (kif == NULL)
5355                         goto out;
5356
5357                 if (kif->pfik_ifp == NULL) {
5358                         ret = 0;
5359                         goto out;
5360                 }
5361
5362                 /* Perform uRPF check if passed input interface */
5363                 ret = 0;
5364                 rn = (struct radix_node *)ro.ro_rt;
5365                 do {
5366                         rt = (struct rtentry *)rn;
5367                         ifp = rt->rt_ifp;
5368
5369                         if (kif->pfik_ifp == ifp)
5370                                 ret = 1;
5371                         rn = rn_mpath_next(rn);
5372                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5373         } else
5374                 ret = 0;
5375 out:
5376         if (ro.ro_rt != NULL)
5377                 RTFREE(ro.ro_rt);
5378         return (ret);
5379 }
5380 #endif
5381
5382 int
5383 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5384     int rtableid)
5385 {
5386 #ifdef INET
5387         struct nhop4_basic      nh4;
5388 #endif
5389 #ifdef INET6
5390         struct nhop6_basic      nh6;
5391 #endif
5392         struct ifnet            *ifp;
5393 #ifdef RADIX_MPATH
5394         struct radix_node_head  *rnh;
5395
5396         /* XXX: stick to table 0 for now */
5397         rnh = rt_tables_get_rnh(0, af);
5398         if (rnh != NULL && rn_mpath_capable(rnh))
5399                 return (pf_routable_oldmpath(addr, af, kif, rtableid));
5400 #endif
5401         /*
5402          * Skip check for addresses with embedded interface scope,
5403          * as they would always match anyway.
5404          */
5405         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5406                 return (1);
5407
5408         if (af != AF_INET && af != AF_INET6)
5409                 return (0);
5410
5411         /* Skip checks for ipsec interfaces */
5412         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5413                 return (1);
5414
5415         ifp = NULL;
5416
5417         switch (af) {
5418 #ifdef INET6
5419         case AF_INET6:
5420                 if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5421                         return (0);
5422                 ifp = nh6.nh_ifp;
5423                 break;
5424 #endif
5425 #ifdef INET
5426         case AF_INET:
5427                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5428                         return (0);
5429                 ifp = nh4.nh_ifp;
5430                 break;
5431 #endif
5432         }
5433
5434         /* No interface given, this is a no-route check */
5435         if (kif == NULL)
5436                 return (1);
5437
5438         if (kif->pfik_ifp == NULL)
5439                 return (0);
5440
5441         /* Perform uRPF check if passed input interface */
5442         if (kif->pfik_ifp == ifp)
5443                 return (1);
5444         return (0);
5445 }
5446
5447 #ifdef INET
5448 static void
5449 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5450     struct pf_state *s, struct pf_pdesc *pd)
5451 {
5452         struct mbuf             *m0, *m1;
5453         struct sockaddr_in      dst;
5454         struct ip               *ip;
5455         struct ifnet            *ifp = NULL;
5456         struct pf_addr           naddr;
5457         struct pf_src_node      *sn = NULL;
5458         int                      error = 0;
5459         uint16_t                 ip_len, ip_off;
5460
5461         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5462         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5463             __func__));
5464
5465         if ((pd->pf_mtag == NULL &&
5466             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5467             pd->pf_mtag->routed++ > 3) {
5468                 m0 = *m;
5469                 *m = NULL;
5470                 goto bad_locked;
5471         }
5472
5473         if (r->rt == PF_DUPTO) {
5474                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5475                         if (s)
5476                                 PF_STATE_UNLOCK(s);
5477                         return;
5478                 }
5479         } else {
5480                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5481                         if (s)
5482                                 PF_STATE_UNLOCK(s);
5483                         return;
5484                 }
5485                 m0 = *m;
5486         }
5487
5488         ip = mtod(m0, struct ip *);
5489
5490         bzero(&dst, sizeof(dst));
5491         dst.sin_family = AF_INET;
5492         dst.sin_len = sizeof(dst);
5493         dst.sin_addr = ip->ip_dst;
5494
5495         if (r->rt == PF_FASTROUTE) {
5496                 struct nhop4_basic nh4;
5497
5498                 if (s)
5499                         PF_STATE_UNLOCK(s);
5500
5501                 if (fib4_lookup_nh_basic(M_GETFIB(m0), ip->ip_dst, 0,
5502                     m0->m_pkthdr.flowid, &nh4) != 0) {
5503                         KMOD_IPSTAT_INC(ips_noroute);
5504                         error = EHOSTUNREACH;
5505                         goto bad;
5506                 }
5507
5508                 ifp = nh4.nh_ifp;
5509                 dst.sin_addr = nh4.nh_addr;
5510         } else {
5511                 if (TAILQ_EMPTY(&r->rpool.list)) {
5512                         DPFPRINTF(PF_DEBUG_URGENT,
5513                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5514                         goto bad_locked;
5515                 }
5516                 if (s == NULL) {
5517                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5518                             &naddr, NULL, &sn);
5519                         if (!PF_AZERO(&naddr, AF_INET))
5520                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5521                         ifp = r->rpool.cur->kif ?
5522                             r->rpool.cur->kif->pfik_ifp : NULL;
5523                 } else {
5524                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5525                                 dst.sin_addr.s_addr =
5526                                     s->rt_addr.v4.s_addr;
5527                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5528                         PF_STATE_UNLOCK(s);
5529                 }
5530         }
5531         if (ifp == NULL)
5532                 goto bad;
5533
5534         if (oifp != ifp) {
5535                 if (pf_test(PF_OUT, 0, ifp, &m0, NULL) != PF_PASS)
5536                         goto bad;
5537                 else if (m0 == NULL)
5538                         goto done;
5539                 if (m0->m_len < sizeof(struct ip)) {
5540                         DPFPRINTF(PF_DEBUG_URGENT,
5541                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5542                         goto bad;
5543                 }
5544                 ip = mtod(m0, struct ip *);
5545         }
5546
5547         if (ifp->if_flags & IFF_LOOPBACK)
5548                 m0->m_flags |= M_SKIP_FIREWALL;
5549
5550         ip_len = ntohs(ip->ip_len);
5551         ip_off = ntohs(ip->ip_off);
5552
5553         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5554         m0->m_pkthdr.csum_flags |= CSUM_IP;
5555         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5556                 in_delayed_cksum(m0);
5557                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5558         }
5559 #ifdef SCTP
5560         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5561                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5562                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5563         }
5564 #endif
5565
5566         /*
5567          * If small enough for interface, or the interface will take
5568          * care of the fragmentation for us, we can just send directly.
5569          */
5570         if (ip_len <= ifp->if_mtu ||
5571             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5572                 ip->ip_sum = 0;
5573                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5574                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5575                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5576                 }
5577                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5578                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5579                 goto done;
5580         }
5581
5582         /* Balk when DF bit is set or the interface didn't support TSO. */
5583         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5584                 error = EMSGSIZE;
5585                 KMOD_IPSTAT_INC(ips_cantfrag);
5586                 if (r->rt != PF_DUPTO) {
5587                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5588                             ifp->if_mtu);
5589                         goto done;
5590                 } else
5591                         goto bad;
5592         }
5593
5594         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5595         if (error)
5596                 goto bad;
5597
5598         for (; m0; m0 = m1) {
5599                 m1 = m0->m_nextpkt;
5600                 m0->m_nextpkt = NULL;
5601                 if (error == 0) {
5602                         m_clrprotoflags(m0);
5603                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5604                 } else
5605                         m_freem(m0);
5606         }
5607
5608         if (error == 0)
5609                 KMOD_IPSTAT_INC(ips_fragmented);
5610
5611 done:
5612         if (r->rt != PF_DUPTO)
5613                 *m = NULL;
5614         return;
5615
5616 bad_locked:
5617         if (s)
5618                 PF_STATE_UNLOCK(s);
5619 bad:
5620         m_freem(m0);
5621         goto done;
5622 }
5623 #endif /* INET */
5624
5625 #ifdef INET6
5626 static void
5627 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5628     struct pf_state *s, struct pf_pdesc *pd)
5629 {
5630         struct mbuf             *m0;
5631         struct sockaddr_in6     dst;
5632         struct ip6_hdr          *ip6;
5633         struct ifnet            *ifp = NULL;
5634         struct pf_addr           naddr;
5635         struct pf_src_node      *sn = NULL;
5636
5637         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5638         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5639             __func__));
5640
5641         if ((pd->pf_mtag == NULL &&
5642             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5643             pd->pf_mtag->routed++ > 3) {
5644                 m0 = *m;
5645                 *m = NULL;
5646                 goto bad_locked;
5647         }
5648
5649         if (r->rt == PF_DUPTO) {
5650                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5651                         if (s)
5652                                 PF_STATE_UNLOCK(s);
5653                         return;
5654                 }
5655         } else {
5656                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5657                         if (s)
5658                                 PF_STATE_UNLOCK(s);
5659                         return;
5660                 }
5661                 m0 = *m;
5662         }
5663
5664         ip6 = mtod(m0, struct ip6_hdr *);
5665
5666         bzero(&dst, sizeof(dst));
5667         dst.sin6_family = AF_INET6;
5668         dst.sin6_len = sizeof(dst);
5669         dst.sin6_addr = ip6->ip6_dst;
5670
5671         /* Cheat. XXX why only in the v6 case??? */
5672         if (r->rt == PF_FASTROUTE) {
5673                 if (s)
5674                         PF_STATE_UNLOCK(s);
5675                 m0->m_flags |= M_SKIP_FIREWALL;
5676                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5677                 *m = NULL;
5678                 return;
5679         }
5680
5681         if (TAILQ_EMPTY(&r->rpool.list)) {
5682                 DPFPRINTF(PF_DEBUG_URGENT,
5683                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5684                 goto bad_locked;
5685         }
5686         if (s == NULL) {
5687                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5688                     &naddr, NULL, &sn);
5689                 if (!PF_AZERO(&naddr, AF_INET6))
5690                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5691                             &naddr, AF_INET6);
5692                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5693         } else {
5694                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5695                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5696                             &s->rt_addr, AF_INET6);
5697                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5698         }
5699
5700         if (s)
5701                 PF_STATE_UNLOCK(s);
5702
5703         if (ifp == NULL)
5704                 goto bad;
5705
5706         if (oifp != ifp) {
5707                 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, NULL) != PF_PASS)
5708                         goto bad;
5709                 else if (m0 == NULL)
5710                         goto done;
5711                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5712                         DPFPRINTF(PF_DEBUG_URGENT,
5713                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5714                             __func__));
5715                         goto bad;
5716                 }
5717                 ip6 = mtod(m0, struct ip6_hdr *);
5718         }
5719
5720         if (ifp->if_flags & IFF_LOOPBACK)
5721                 m0->m_flags |= M_SKIP_FIREWALL;
5722
5723         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5724             ~ifp->if_hwassist) {
5725                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5726                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5727                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5728         }
5729
5730         /*
5731          * If the packet is too large for the outgoing interface,
5732          * send back an icmp6 error.
5733          */
5734         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5735                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5736         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5737                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5738         else {
5739                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5740                 if (r->rt != PF_DUPTO)
5741                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5742                 else
5743                         goto bad;
5744         }
5745
5746 done:
5747         if (r->rt != PF_DUPTO)
5748                 *m = NULL;
5749         return;
5750
5751 bad_locked:
5752         if (s)
5753                 PF_STATE_UNLOCK(s);
5754 bad:
5755         m_freem(m0);
5756         goto done;
5757 }
5758 #endif /* INET6 */
5759
5760 /*
5761  * FreeBSD supports cksum offloads for the following drivers.
5762  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5763  *   ti(4), txp(4), xl(4)
5764  *
5765  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5766  *  network driver performed cksum including pseudo header, need to verify
5767  *   csum_data
5768  * CSUM_DATA_VALID :
5769  *  network driver performed cksum, needs to additional pseudo header
5770  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5771  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5772  *
5773  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5774  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5775  * TCP/UDP layer.
5776  * Also, set csum_data to 0xffff to force cksum validation.
5777  */
5778 static int
5779 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5780 {
5781         u_int16_t sum = 0;
5782         int hw_assist = 0;
5783         struct ip *ip;
5784
5785         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5786                 return (1);
5787         if (m->m_pkthdr.len < off + len)
5788                 return (1);
5789
5790         switch (p) {
5791         case IPPROTO_TCP:
5792                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5793                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5794                                 sum = m->m_pkthdr.csum_data;
5795                         } else {
5796                                 ip = mtod(m, struct ip *);
5797                                 sum = in_pseudo(ip->ip_src.s_addr,
5798                                 ip->ip_dst.s_addr, htonl((u_short)len +
5799                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5800                         }
5801                         sum ^= 0xffff;
5802                         ++hw_assist;
5803                 }
5804                 break;
5805         case IPPROTO_UDP:
5806                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5807                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5808                                 sum = m->m_pkthdr.csum_data;
5809                         } else {
5810                                 ip = mtod(m, struct ip *);
5811                                 sum = in_pseudo(ip->ip_src.s_addr,
5812                                 ip->ip_dst.s_addr, htonl((u_short)len +
5813                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5814                         }
5815                         sum ^= 0xffff;
5816                         ++hw_assist;
5817                 }
5818                 break;
5819         case IPPROTO_ICMP:
5820 #ifdef INET6
5821         case IPPROTO_ICMPV6:
5822 #endif /* INET6 */
5823                 break;
5824         default:
5825                 return (1);
5826         }
5827
5828         if (!hw_assist) {
5829                 switch (af) {
5830                 case AF_INET:
5831                         if (p == IPPROTO_ICMP) {
5832                                 if (m->m_len < off)
5833                                         return (1);
5834                                 m->m_data += off;
5835                                 m->m_len -= off;
5836                                 sum = in_cksum(m, len);
5837                                 m->m_data -= off;
5838                                 m->m_len += off;
5839                         } else {
5840                                 if (m->m_len < sizeof(struct ip))
5841                                         return (1);
5842                                 sum = in4_cksum(m, p, off, len);
5843                         }
5844                         break;
5845 #ifdef INET6
5846                 case AF_INET6:
5847                         if (m->m_len < sizeof(struct ip6_hdr))
5848                                 return (1);
5849                         sum = in6_cksum(m, p, off, len);
5850                         break;
5851 #endif /* INET6 */
5852                 default:
5853                         return (1);
5854                 }
5855         }
5856         if (sum) {
5857                 switch (p) {
5858                 case IPPROTO_TCP:
5859                     {
5860                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5861                         break;
5862                     }
5863                 case IPPROTO_UDP:
5864                     {
5865                         KMOD_UDPSTAT_INC(udps_badsum);
5866                         break;
5867                     }
5868 #ifdef INET
5869                 case IPPROTO_ICMP:
5870                     {
5871                         KMOD_ICMPSTAT_INC(icps_checksum);
5872                         break;
5873                     }
5874 #endif
5875 #ifdef INET6
5876                 case IPPROTO_ICMPV6:
5877                     {
5878                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5879                         break;
5880                     }
5881 #endif /* INET6 */
5882                 }
5883                 return (1);
5884         } else {
5885                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5886                         m->m_pkthdr.csum_flags |=
5887                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5888                         m->m_pkthdr.csum_data = 0xffff;
5889                 }
5890         }
5891         return (0);
5892 }
5893
5894
5895 #ifdef INET
5896 int
5897 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5898 {
5899         struct pfi_kif          *kif;
5900         u_short                  action, reason = 0, log = 0;
5901         struct mbuf             *m = *m0;
5902         struct ip               *h = NULL;
5903         struct m_tag            *ipfwtag;
5904         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5905         struct pf_state         *s = NULL;
5906         struct pf_ruleset       *ruleset = NULL;
5907         struct pf_pdesc          pd;
5908         int                      off, dirndx, pqid = 0;
5909
5910         M_ASSERTPKTHDR(m);
5911
5912         if (!V_pf_status.running)
5913                 return (PF_PASS);
5914
5915         memset(&pd, 0, sizeof(pd));
5916
5917         kif = (struct pfi_kif *)ifp->if_pf_kif;
5918
5919         if (kif == NULL) {
5920                 DPFPRINTF(PF_DEBUG_URGENT,
5921                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5922                 return (PF_DROP);
5923         }
5924         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5925                 return (PF_PASS);
5926
5927         if (m->m_flags & M_SKIP_FIREWALL)
5928                 return (PF_PASS);
5929
5930         pd.pf_mtag = pf_find_mtag(m);
5931
5932         PF_RULES_RLOCK();
5933
5934         if (ip_divert_ptr != NULL &&
5935             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5936                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5937                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5938                         if (pd.pf_mtag == NULL &&
5939                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5940                                 action = PF_DROP;
5941                                 goto done;
5942                         }
5943                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5944                         m_tag_delete(m, ipfwtag);
5945                 }
5946                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5947                         m->m_flags |= M_FASTFWD_OURS;
5948                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5949                 }
5950         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5951                 /* We do IP header normalization and packet reassembly here */
5952                 action = PF_DROP;
5953                 goto done;
5954         }
5955         m = *m0;        /* pf_normalize messes with m0 */
5956         h = mtod(m, struct ip *);
5957
5958         off = h->ip_hl << 2;
5959         if (off < (int)sizeof(struct ip)) {
5960                 action = PF_DROP;
5961                 REASON_SET(&reason, PFRES_SHORT);
5962                 log = 1;
5963                 goto done;
5964         }
5965
5966         pd.src = (struct pf_addr *)&h->ip_src;
5967         pd.dst = (struct pf_addr *)&h->ip_dst;
5968         pd.sport = pd.dport = NULL;
5969         pd.ip_sum = &h->ip_sum;
5970         pd.proto_sum = NULL;
5971         pd.proto = h->ip_p;
5972         pd.dir = dir;
5973         pd.sidx = (dir == PF_IN) ? 0 : 1;
5974         pd.didx = (dir == PF_IN) ? 1 : 0;
5975         pd.af = AF_INET;
5976         pd.tos = h->ip_tos;
5977         pd.tot_len = ntohs(h->ip_len);
5978
5979         /* handle fragments that didn't get reassembled by normalization */
5980         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5981                 action = pf_test_fragment(&r, dir, kif, m, h,
5982                     &pd, &a, &ruleset);
5983                 goto done;
5984         }
5985
5986         switch (h->ip_p) {
5987
5988         case IPPROTO_TCP: {
5989                 struct tcphdr   th;
5990
5991                 pd.hdr.tcp = &th;
5992                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5993                     &action, &reason, AF_INET)) {
5994                         log = action != PF_PASS;
5995                         goto done;
5996                 }
5997                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5998                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5999                         pqid = 1;
6000                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6001                 if (action == PF_DROP)
6002                         goto done;
6003                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6004                     &reason);
6005                 if (action == PF_PASS) {
6006                         if (pfsync_update_state_ptr != NULL)
6007                                 pfsync_update_state_ptr(s);
6008                         r = s->rule.ptr;
6009                         a = s->anchor.ptr;
6010                         log = s->log;
6011                 } else if (s == NULL)
6012                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6013                             &a, &ruleset, inp);
6014                 break;
6015         }
6016
6017         case IPPROTO_UDP: {
6018                 struct udphdr   uh;
6019
6020                 pd.hdr.udp = &uh;
6021                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6022                     &action, &reason, AF_INET)) {
6023                         log = action != PF_PASS;
6024                         goto done;
6025                 }
6026                 if (uh.uh_dport == 0 ||
6027                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6028                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6029                         action = PF_DROP;
6030                         REASON_SET(&reason, PFRES_SHORT);
6031                         goto done;
6032                 }
6033                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6034                 if (action == PF_PASS) {
6035                         if (pfsync_update_state_ptr != NULL)
6036                                 pfsync_update_state_ptr(s);
6037                         r = s->rule.ptr;
6038                         a = s->anchor.ptr;
6039                         log = s->log;
6040                 } else if (s == NULL)
6041                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6042                             &a, &ruleset, inp);
6043                 break;
6044         }
6045
6046         case IPPROTO_ICMP: {
6047                 struct icmp     ih;
6048
6049                 pd.hdr.icmp = &ih;
6050                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6051                     &action, &reason, AF_INET)) {
6052                         log = action != PF_PASS;
6053                         goto done;
6054                 }
6055                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6056                     &reason);
6057                 if (action == PF_PASS) {
6058                         if (pfsync_update_state_ptr != NULL)
6059                                 pfsync_update_state_ptr(s);
6060                         r = s->rule.ptr;
6061                         a = s->anchor.ptr;
6062                         log = s->log;
6063                 } else if (s == NULL)
6064                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6065                             &a, &ruleset, inp);
6066                 break;
6067         }
6068
6069 #ifdef INET6
6070         case IPPROTO_ICMPV6: {
6071                 action = PF_DROP;
6072                 DPFPRINTF(PF_DEBUG_MISC,
6073                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6074                 goto done;
6075         }
6076 #endif
6077
6078         default:
6079                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6080                 if (action == PF_PASS) {
6081                         if (pfsync_update_state_ptr != NULL)
6082                                 pfsync_update_state_ptr(s);
6083                         r = s->rule.ptr;
6084                         a = s->anchor.ptr;
6085                         log = s->log;
6086                 } else if (s == NULL)
6087                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6088                             &a, &ruleset, inp);
6089                 break;
6090         }
6091
6092 done:
6093         PF_RULES_RUNLOCK();
6094         if (action == PF_PASS && h->ip_hl > 5 &&
6095             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6096                 action = PF_DROP;
6097                 REASON_SET(&reason, PFRES_IPOPTIONS);
6098                 log = r->log;
6099                 DPFPRINTF(PF_DEBUG_MISC,
6100                     ("pf: dropping packet with ip options\n"));
6101         }
6102
6103         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6104                 action = PF_DROP;
6105                 REASON_SET(&reason, PFRES_MEMORY);
6106         }
6107         if (r->rtableid >= 0)
6108                 M_SETFIB(m, r->rtableid);
6109
6110         if (r->scrub_flags & PFSTATE_SETPRIO) {
6111                 if (pd.tos & IPTOS_LOWDELAY)
6112                         pqid = 1;
6113                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6114                         action = PF_DROP;
6115                         REASON_SET(&reason, PFRES_MEMORY);
6116                         log = 1;
6117                         DPFPRINTF(PF_DEBUG_MISC,
6118                             ("pf: failed to allocate 802.1q mtag\n"));
6119                 }
6120         }
6121
6122 #ifdef ALTQ
6123         if (action == PF_PASS && r->qid) {
6124                 if (pd.pf_mtag == NULL &&
6125                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6126                         action = PF_DROP;
6127                         REASON_SET(&reason, PFRES_MEMORY);
6128                 } else {
6129                         if (s != NULL)
6130                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6131                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6132                                 pd.pf_mtag->qid = r->pqid;
6133                         else
6134                                 pd.pf_mtag->qid = r->qid;
6135                         /* Add hints for ecn. */
6136                         pd.pf_mtag->hdr = h;
6137                 }
6138
6139         }
6140 #endif /* ALTQ */
6141
6142         /*
6143          * connections redirected to loopback should not match sockets
6144          * bound specifically to loopback due to security implications,
6145          * see tcp_input() and in_pcblookup_listen().
6146          */
6147         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6148             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6149             (s->nat_rule.ptr->action == PF_RDR ||
6150             s->nat_rule.ptr->action == PF_BINAT) &&
6151             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6152                 m->m_flags |= M_SKIP_FIREWALL;
6153
6154         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6155             !PACKET_LOOPED(&pd)) {
6156
6157                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6158                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6159                 if (ipfwtag != NULL) {
6160                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6161                             ntohs(r->divert.port);
6162                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6163
6164                         if (s)
6165                                 PF_STATE_UNLOCK(s);
6166
6167                         m_tag_prepend(m, ipfwtag);
6168                         if (m->m_flags & M_FASTFWD_OURS) {
6169                                 if (pd.pf_mtag == NULL &&
6170                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6171                                         action = PF_DROP;
6172                                         REASON_SET(&reason, PFRES_MEMORY);
6173                                         log = 1;
6174                                         DPFPRINTF(PF_DEBUG_MISC,
6175                                             ("pf: failed to allocate tag\n"));
6176                                 } else {
6177                                         pd.pf_mtag->flags |=
6178                                             PF_FASTFWD_OURS_PRESENT;
6179                                         m->m_flags &= ~M_FASTFWD_OURS;
6180                                 }
6181                         }
6182                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6183                         *m0 = NULL;
6184
6185                         return (action);
6186                 } else {
6187                         /* XXX: ipfw has the same behaviour! */
6188                         action = PF_DROP;
6189                         REASON_SET(&reason, PFRES_MEMORY);
6190                         log = 1;
6191                         DPFPRINTF(PF_DEBUG_MISC,
6192                             ("pf: failed to allocate divert tag\n"));
6193                 }
6194         }
6195
6196         if (log) {
6197                 struct pf_rule *lr;
6198
6199                 if (s != NULL && s->nat_rule.ptr != NULL &&
6200                     s->nat_rule.ptr->log & PF_LOG_ALL)
6201                         lr = s->nat_rule.ptr;
6202                 else
6203                         lr = r;
6204                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6205                     (s == NULL));
6206         }
6207
6208         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6209         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6210
6211         if (action == PF_PASS || r->action == PF_DROP) {
6212                 dirndx = (dir == PF_OUT);
6213                 r->packets[dirndx]++;
6214                 r->bytes[dirndx] += pd.tot_len;
6215                 if (a != NULL) {
6216                         a->packets[dirndx]++;
6217                         a->bytes[dirndx] += pd.tot_len;
6218                 }
6219                 if (s != NULL) {
6220                         if (s->nat_rule.ptr != NULL) {
6221                                 s->nat_rule.ptr->packets[dirndx]++;
6222                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6223                         }
6224                         if (s->src_node != NULL) {
6225                                 s->src_node->packets[dirndx]++;
6226                                 s->src_node->bytes[dirndx] += pd.tot_len;
6227                         }
6228                         if (s->nat_src_node != NULL) {
6229                                 s->nat_src_node->packets[dirndx]++;
6230                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6231                         }
6232                         dirndx = (dir == s->direction) ? 0 : 1;
6233                         s->packets[dirndx]++;
6234                         s->bytes[dirndx] += pd.tot_len;
6235                 }
6236                 tr = r;
6237                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6238                 if (nr != NULL && r == &V_pf_default_rule)
6239                         tr = nr;
6240                 if (tr->src.addr.type == PF_ADDR_TABLE)
6241                         pfr_update_stats(tr->src.addr.p.tbl,
6242                             (s == NULL) ? pd.src :
6243                             &s->key[(s->direction == PF_IN)]->
6244                                 addr[(s->direction == PF_OUT)],
6245                             pd.af, pd.tot_len, dir == PF_OUT,
6246                             r->action == PF_PASS, tr->src.neg);
6247                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6248                         pfr_update_stats(tr->dst.addr.p.tbl,
6249                             (s == NULL) ? pd.dst :
6250                             &s->key[(s->direction == PF_IN)]->
6251                                 addr[(s->direction == PF_IN)],
6252                             pd.af, pd.tot_len, dir == PF_OUT,
6253                             r->action == PF_PASS, tr->dst.neg);
6254         }
6255
6256         switch (action) {
6257         case PF_SYNPROXY_DROP:
6258                 m_freem(*m0);
6259         case PF_DEFER:
6260                 *m0 = NULL;
6261                 action = PF_PASS;
6262                 break;
6263         case PF_DROP:
6264                 m_freem(*m0);
6265                 *m0 = NULL;
6266                 break;
6267         default:
6268                 /* pf_route() returns unlocked. */
6269                 if (r->rt) {
6270                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6271                         return (action);
6272                 }
6273                 break;
6274         }
6275         if (s)
6276                 PF_STATE_UNLOCK(s);
6277
6278         return (action);
6279 }
6280 #endif /* INET */
6281
6282 #ifdef INET6
6283 int
6284 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6285 {
6286         struct pfi_kif          *kif;
6287         u_short                  action, reason = 0, log = 0;
6288         struct mbuf             *m = *m0, *n = NULL;
6289         struct m_tag            *mtag;
6290         struct ip6_hdr          *h = NULL;
6291         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6292         struct pf_state         *s = NULL;
6293         struct pf_ruleset       *ruleset = NULL;
6294         struct pf_pdesc          pd;
6295         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6296
6297         M_ASSERTPKTHDR(m);
6298
6299         if (!V_pf_status.running)
6300                 return (PF_PASS);
6301
6302         memset(&pd, 0, sizeof(pd));
6303         pd.pf_mtag = pf_find_mtag(m);
6304
6305         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6306                 return (PF_PASS);
6307
6308         kif = (struct pfi_kif *)ifp->if_pf_kif;
6309         if (kif == NULL) {
6310                 DPFPRINTF(PF_DEBUG_URGENT,
6311                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6312                 return (PF_DROP);
6313         }
6314         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6315                 return (PF_PASS);
6316
6317         if (m->m_flags & M_SKIP_FIREWALL)
6318                 return (PF_PASS);
6319
6320         PF_RULES_RLOCK();
6321
6322         /* We do IP header normalization and packet reassembly here */
6323         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6324                 action = PF_DROP;
6325                 goto done;
6326         }
6327         m = *m0;        /* pf_normalize messes with m0 */
6328         h = mtod(m, struct ip6_hdr *);
6329
6330 #if 1
6331         /*
6332          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6333          * will do something bad, so drop the packet for now.
6334          */
6335         if (htons(h->ip6_plen) == 0) {
6336                 action = PF_DROP;
6337                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6338                 goto done;
6339         }
6340 #endif
6341
6342         pd.src = (struct pf_addr *)&h->ip6_src;
6343         pd.dst = (struct pf_addr *)&h->ip6_dst;
6344         pd.sport = pd.dport = NULL;
6345         pd.ip_sum = NULL;
6346         pd.proto_sum = NULL;
6347         pd.dir = dir;
6348         pd.sidx = (dir == PF_IN) ? 0 : 1;
6349         pd.didx = (dir == PF_IN) ? 1 : 0;
6350         pd.af = AF_INET6;
6351         pd.tos = 0;
6352         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6353
6354         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6355         pd.proto = h->ip6_nxt;
6356         do {
6357                 switch (pd.proto) {
6358                 case IPPROTO_FRAGMENT:
6359                         action = pf_test_fragment(&r, dir, kif, m, h,
6360                             &pd, &a, &ruleset);
6361                         if (action == PF_DROP)
6362                                 REASON_SET(&reason, PFRES_FRAG);
6363                         goto done;
6364                 case IPPROTO_ROUTING: {
6365                         struct ip6_rthdr rthdr;
6366
6367                         if (rh_cnt++) {
6368                                 DPFPRINTF(PF_DEBUG_MISC,
6369                                     ("pf: IPv6 more than one rthdr\n"));
6370                                 action = PF_DROP;
6371                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6372                                 log = 1;
6373                                 goto done;
6374                         }
6375                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6376                             &reason, pd.af)) {
6377                                 DPFPRINTF(PF_DEBUG_MISC,
6378                                     ("pf: IPv6 short rthdr\n"));
6379                                 action = PF_DROP;
6380                                 REASON_SET(&reason, PFRES_SHORT);
6381                                 log = 1;
6382                                 goto done;
6383                         }
6384                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6385                                 DPFPRINTF(PF_DEBUG_MISC,
6386                                     ("pf: IPv6 rthdr0\n"));
6387                                 action = PF_DROP;
6388                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6389                                 log = 1;
6390                                 goto done;
6391                         }
6392                         /* FALLTHROUGH */
6393                 }
6394                 case IPPROTO_AH:
6395                 case IPPROTO_HOPOPTS:
6396                 case IPPROTO_DSTOPTS: {
6397                         /* get next header and header length */
6398                         struct ip6_ext  opt6;
6399
6400                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6401                             NULL, &reason, pd.af)) {
6402                                 DPFPRINTF(PF_DEBUG_MISC,
6403                                     ("pf: IPv6 short opt\n"));
6404                                 action = PF_DROP;
6405                                 log = 1;
6406                                 goto done;
6407                         }
6408                         if (pd.proto == IPPROTO_AH)
6409                                 off += (opt6.ip6e_len + 2) * 4;
6410                         else
6411                                 off += (opt6.ip6e_len + 1) * 8;
6412                         pd.proto = opt6.ip6e_nxt;
6413                         /* goto the next header */
6414                         break;
6415                 }
6416                 default:
6417                         terminal++;
6418                         break;
6419                 }
6420         } while (!terminal);
6421
6422         /* if there's no routing header, use unmodified mbuf for checksumming */
6423         if (!n)
6424                 n = m;
6425
6426         switch (pd.proto) {
6427
6428         case IPPROTO_TCP: {
6429                 struct tcphdr   th;
6430
6431                 pd.hdr.tcp = &th;
6432                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6433                     &action, &reason, AF_INET6)) {
6434                         log = action != PF_PASS;
6435                         goto done;
6436                 }
6437                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6438                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6439                 if (action == PF_DROP)
6440                         goto done;
6441                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6442                     &reason);
6443                 if (action == PF_PASS) {
6444                         if (pfsync_update_state_ptr != NULL)
6445                                 pfsync_update_state_ptr(s);
6446                         r = s->rule.ptr;
6447                         a = s->anchor.ptr;
6448                         log = s->log;
6449                 } else if (s == NULL)
6450                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6451                             &a, &ruleset, inp);
6452                 break;
6453         }
6454
6455         case IPPROTO_UDP: {
6456                 struct udphdr   uh;
6457
6458                 pd.hdr.udp = &uh;
6459                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6460                     &action, &reason, AF_INET6)) {
6461                         log = action != PF_PASS;
6462                         goto done;
6463                 }
6464                 if (uh.uh_dport == 0 ||
6465                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6466                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6467                         action = PF_DROP;
6468                         REASON_SET(&reason, PFRES_SHORT);
6469                         goto done;
6470                 }
6471                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6472                 if (action == PF_PASS) {
6473                         if (pfsync_update_state_ptr != NULL)
6474                                 pfsync_update_state_ptr(s);
6475                         r = s->rule.ptr;
6476                         a = s->anchor.ptr;
6477                         log = s->log;
6478                 } else if (s == NULL)
6479                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6480                             &a, &ruleset, inp);
6481                 break;
6482         }
6483
6484         case IPPROTO_ICMP: {
6485                 action = PF_DROP;
6486                 DPFPRINTF(PF_DEBUG_MISC,
6487                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6488                 goto done;
6489         }
6490
6491         case IPPROTO_ICMPV6: {
6492                 struct icmp6_hdr        ih;
6493
6494                 pd.hdr.icmp6 = &ih;
6495                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6496                     &action, &reason, AF_INET6)) {
6497                         log = action != PF_PASS;
6498                         goto done;
6499                 }
6500                 action = pf_test_state_icmp(&s, dir, kif,
6501                     m, off, h, &pd, &reason);
6502                 if (action == PF_PASS) {
6503                         if (pfsync_update_state_ptr != NULL)
6504                                 pfsync_update_state_ptr(s);
6505                         r = s->rule.ptr;
6506                         a = s->anchor.ptr;
6507                         log = s->log;
6508                 } else if (s == NULL)
6509                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6510                             &a, &ruleset, inp);
6511                 break;
6512         }
6513
6514         default:
6515                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6516                 if (action == PF_PASS) {
6517                         if (pfsync_update_state_ptr != NULL)
6518                                 pfsync_update_state_ptr(s);
6519                         r = s->rule.ptr;
6520                         a = s->anchor.ptr;
6521                         log = s->log;
6522                 } else if (s == NULL)
6523                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6524                             &a, &ruleset, inp);
6525                 break;
6526         }
6527
6528 done:
6529         PF_RULES_RUNLOCK();
6530         if (n != m) {
6531                 m_freem(n);
6532                 n = NULL;
6533         }
6534
6535         /* handle dangerous IPv6 extension headers. */
6536         if (action == PF_PASS && rh_cnt &&
6537             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6538                 action = PF_DROP;
6539                 REASON_SET(&reason, PFRES_IPOPTIONS);
6540                 log = r->log;
6541                 DPFPRINTF(PF_DEBUG_MISC,
6542                     ("pf: dropping packet with dangerous v6 headers\n"));
6543         }
6544
6545         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6546                 action = PF_DROP;
6547                 REASON_SET(&reason, PFRES_MEMORY);
6548         }
6549         if (r->rtableid >= 0)
6550                 M_SETFIB(m, r->rtableid);
6551
6552         if (r->scrub_flags & PFSTATE_SETPRIO) {
6553                 if (pd.tos & IPTOS_LOWDELAY)
6554                         pqid = 1;
6555                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6556                         action = PF_DROP;
6557                         REASON_SET(&reason, PFRES_MEMORY);
6558                         log = 1;
6559                         DPFPRINTF(PF_DEBUG_MISC,
6560                             ("pf: failed to allocate 802.1q mtag\n"));
6561                 }
6562         }
6563
6564 #ifdef ALTQ
6565         if (action == PF_PASS && r->qid) {
6566                 if (pd.pf_mtag == NULL &&
6567                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6568                         action = PF_DROP;
6569                         REASON_SET(&reason, PFRES_MEMORY);
6570                 } else {
6571                         if (s != NULL)
6572                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6573                         if (pd.tos & IPTOS_LOWDELAY)
6574                                 pd.pf_mtag->qid = r->pqid;
6575                         else
6576                                 pd.pf_mtag->qid = r->qid;
6577                         /* Add hints for ecn. */
6578                         pd.pf_mtag->hdr = h;
6579                 }
6580         }
6581 #endif /* ALTQ */
6582
6583         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6584             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6585             (s->nat_rule.ptr->action == PF_RDR ||
6586             s->nat_rule.ptr->action == PF_BINAT) &&
6587             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6588                 m->m_flags |= M_SKIP_FIREWALL;
6589
6590         /* XXX: Anybody working on it?! */
6591         if (r->divert.port)
6592                 printf("pf: divert(9) is not supported for IPv6\n");
6593
6594         if (log) {
6595                 struct pf_rule *lr;
6596
6597                 if (s != NULL && s->nat_rule.ptr != NULL &&
6598                     s->nat_rule.ptr->log & PF_LOG_ALL)
6599                         lr = s->nat_rule.ptr;
6600                 else
6601                         lr = r;
6602                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6603                     &pd, (s == NULL));
6604         }
6605
6606         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6607         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6608
6609         if (action == PF_PASS || r->action == PF_DROP) {
6610                 dirndx = (dir == PF_OUT);
6611                 r->packets[dirndx]++;
6612                 r->bytes[dirndx] += pd.tot_len;
6613                 if (a != NULL) {
6614                         a->packets[dirndx]++;
6615                         a->bytes[dirndx] += pd.tot_len;
6616                 }
6617                 if (s != NULL) {
6618                         if (s->nat_rule.ptr != NULL) {
6619                                 s->nat_rule.ptr->packets[dirndx]++;
6620                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6621                         }
6622                         if (s->src_node != NULL) {
6623                                 s->src_node->packets[dirndx]++;
6624                                 s->src_node->bytes[dirndx] += pd.tot_len;
6625                         }
6626                         if (s->nat_src_node != NULL) {
6627                                 s->nat_src_node->packets[dirndx]++;
6628                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6629                         }
6630                         dirndx = (dir == s->direction) ? 0 : 1;
6631                         s->packets[dirndx]++;
6632                         s->bytes[dirndx] += pd.tot_len;
6633                 }
6634                 tr = r;
6635                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6636                 if (nr != NULL && r == &V_pf_default_rule)
6637                         tr = nr;
6638                 if (tr->src.addr.type == PF_ADDR_TABLE)
6639                         pfr_update_stats(tr->src.addr.p.tbl,
6640                             (s == NULL) ? pd.src :
6641                             &s->key[(s->direction == PF_IN)]->addr[0],
6642                             pd.af, pd.tot_len, dir == PF_OUT,
6643                             r->action == PF_PASS, tr->src.neg);
6644                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6645                         pfr_update_stats(tr->dst.addr.p.tbl,
6646                             (s == NULL) ? pd.dst :
6647                             &s->key[(s->direction == PF_IN)]->addr[1],
6648                             pd.af, pd.tot_len, dir == PF_OUT,
6649                             r->action == PF_PASS, tr->dst.neg);
6650         }
6651
6652         switch (action) {
6653         case PF_SYNPROXY_DROP:
6654                 m_freem(*m0);
6655         case PF_DEFER:
6656                 *m0 = NULL;
6657                 action = PF_PASS;
6658                 break;
6659         case PF_DROP:
6660                 m_freem(*m0);
6661                 *m0 = NULL;
6662                 break;
6663         default:
6664                 /* pf_route6() returns unlocked. */
6665                 if (r->rt) {
6666                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6667                         return (action);
6668                 }
6669                 break;
6670         }
6671
6672         if (s)
6673                 PF_STATE_UNLOCK(s);
6674
6675         /* If reassembled packet passed, create new fragments. */
6676         if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6677             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6678                 action = pf_refragment6(ifp, m0, mtag);
6679
6680         return (action);
6681 }
6682 #endif /* INET6 */