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_types.h>
  65 #include <net/route.h>
  66 #include <net/radix_mpath.h>
  67 #include <net/vnet.h>
  68
  69 #include <net/pfvar.h>
  70 #include <net/if_pflog.h>
  71 #include <net/if_pfsync.h>
  72
  73 #include <netinet/in_pcb.h>
  74 #include <netinet/in_var.h>
  75 #include <netinet/ip.h>
  76 #include <netinet/ip_fw.h>
  77 #include <netinet/ip_icmp.h>
  78 #include <netinet/icmp_var.h>
  79 #include <netinet/ip_var.h>
  80 #include <netinet/tcp.h>
  81 #include <netinet/tcp_fsm.h>
  82 #include <netinet/tcp_seq.h>
  83 #include <netinet/tcp_timer.h>
  84 #include <netinet/tcp_var.h>
  85 #include <netinet/udp.h>
  86 #include <netinet/udp_var.h>
  87
  88 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  89
  90 #ifdef INET6
  91 #include <netinet/ip6.h>
  92 #include <netinet/icmp6.h>
  93 #include <netinet6/nd6.h>
  94 #include <netinet6/ip6_var.h>
  95 #include <netinet6/in6_pcb.h>
  96 #endif /* INET6 */
  97
  98 #include <machine/in_cksum.h>
  99 #include <security/mac/mac_framework.h>
 100
 101 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 102
 103 /*
 104  * Global variables
 105  */
 106
 107 /* state tables */
 108 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 109 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 110 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 111 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 112 VNET_DEFINE(struct pf_kstatus,           pf_status);
 113
 114 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 115 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 116 VNET_DEFINE(int,                         altqs_inactive_open);
 117 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 118
 119 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 120 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 121 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 122 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 123 VNET_DEFINE(int,                         pf_tcp_secret_init);
 124 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 125 VNET_DEFINE(int,                         pf_tcp_iss_off);
 126 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 127
 128 /*
 129  * Queue for pf_intr() sends.
 130  */
 131 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 132 struct pf_send_entry {
 133         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 134         struct mbuf                     *pfse_m;
 135         enum {
 136                 PFSE_IP,
 137                 PFSE_IP6,
 138                 PFSE_ICMP,
 139                 PFSE_ICMP6,
 140         }                               pfse_type;
 141         union {
 142                 struct route            ro;
 143                 struct {
 144                         int             type;
 145                         int             code;
 146                         int             mtu;
 147                 } icmpopts;
 148         } u;
 149 #define pfse_ro         u.ro
 150 #define pfse_icmp_type  u.icmpopts.type
 151 #define pfse_icmp_code  u.icmpopts.code
 152 #define pfse_icmp_mtu   u.icmpopts.mtu
 153 };
 154
 155 STAILQ_HEAD(pf_send_head, pf_send_entry);
 156 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 157 #define V_pf_sendqueue  VNET(pf_sendqueue)
 158
 159 static struct mtx pf_sendqueue_mtx;
 160 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 161 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 162
 163 /*
 164  * Queue for pf_overload_task() tasks.
 165  */
 166 struct pf_overload_entry {
 167         SLIST_ENTRY(pf_overload_entry)  next;
 168         struct pf_addr                  addr;
 169         sa_family_t                     af;
 170         uint8_t                         dir;
 171         struct pf_rule                  *rule;
 172 };
 173
 174 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 175 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 176 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 177 static VNET_DEFINE(struct task, pf_overloadtask);
 178 #define V_pf_overloadtask       VNET(pf_overloadtask)
 179
 180 static struct mtx pf_overloadqueue_mtx;
 181 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 182 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 183
 184 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 185 struct mtx pf_unlnkdrules_mtx;
 186
 187 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 188 #define V_pf_sources_z  VNET(pf_sources_z)
 189 uma_zone_t              pf_mtag_z;
 190 VNET_DEFINE(uma_zone_t,  pf_state_z);
 191 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 192
 193 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 194 #define PFID_CPUBITS    8
 195 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 196 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 197 #define PFID_MAXID      (~PFID_CPUMASK)
 198 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 199
 200 static void              pf_src_tree_remove_state(struct pf_state *);
 201 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 202                             u_int32_t);
 203 static void              pf_add_threshold(struct pf_threshold *);
 204 static int               pf_check_threshold(struct pf_threshold *);
 205
 206 static void              pf_change_ap(struct pf_addr *, u_int16_t *,
 207                             u_int16_t *, u_int16_t *, struct pf_addr *,
 208                             u_int16_t, u_int8_t, sa_family_t);
 209 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 210                             struct tcphdr *, struct pf_state_peer *);
 211 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 212                             struct pf_addr *, struct pf_addr *, u_int16_t,
 213                             u_int16_t *, u_int16_t *, u_int16_t *,
 214                             u_int16_t *, u_int8_t, sa_family_t);
 215 static void              pf_send_tcp(struct mbuf *,
 216                             const struct pf_rule *, sa_family_t,
 217                             const struct pf_addr *, const struct pf_addr *,
 218                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 219                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 220                             u_int16_t, struct ifnet *);
 221 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 222                             sa_family_t, struct pf_rule *);
 223 static void              pf_detach_state(struct pf_state *);
 224 static int               pf_state_key_attach(struct pf_state_key *,
 225                             struct pf_state_key *, struct pf_state *);
 226 static void              pf_state_key_detach(struct pf_state *, int);
 227 static int               pf_state_key_ctor(void *, int, void *, int);
 228 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 229 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 230                             int, struct pfi_kif *, struct mbuf *, int,
 231                             struct pf_pdesc *, struct pf_rule **,
 232                             struct pf_ruleset **, struct inpcb *);
 233 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 234                             struct pf_rule *, struct pf_pdesc *,
 235                             struct pf_src_node *, struct pf_state_key *,
 236                             struct pf_state_key *, struct mbuf *, int,
 237                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 238                             struct pf_state **, int, u_int16_t, u_int16_t,
 239                             int);
 240 static int               pf_test_fragment(struct pf_rule **, int,
 241                             struct pfi_kif *, struct mbuf *, void *,
 242                             struct pf_pdesc *, struct pf_rule **,
 243                             struct pf_ruleset **);
 244 static int               pf_tcp_track_full(struct pf_state_peer *,
 245                             struct pf_state_peer *, struct pf_state **,
 246                             struct pfi_kif *, struct mbuf *, int,
 247                             struct pf_pdesc *, u_short *, int *);
 248 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 249                             struct pf_state_peer *, struct pf_state **,
 250                             struct pf_pdesc *, u_short *);
 251 static int               pf_test_state_tcp(struct pf_state **, int,
 252                             struct pfi_kif *, struct mbuf *, int,
 253                             void *, struct pf_pdesc *, u_short *);
 254 static int               pf_test_state_udp(struct pf_state **, int,
 255                             struct pfi_kif *, struct mbuf *, int,
 256                             void *, struct pf_pdesc *);
 257 static int               pf_test_state_icmp(struct pf_state **, int,
 258                             struct pfi_kif *, struct mbuf *, int,
 259                             void *, struct pf_pdesc *, u_short *);
 260 static int               pf_test_state_other(struct pf_state **, int,
 261                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 262 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 263                             sa_family_t);
 264 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 267                                 int, u_int16_t);
 268 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 269                             u_int8_t, sa_family_t);
 270 static void              pf_print_state_parts(struct pf_state *,
 271                             struct pf_state_key *, struct pf_state_key *);
 272 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 273                             struct pf_addr_wrap *);
 274 static struct pf_state  *pf_find_state(struct pfi_kif *,
 275                             struct pf_state_key_cmp *, u_int);
 276 static int               pf_src_connlimit(struct pf_state **);
 277 static void              pf_overload_task(void *v, int pending);
 278 static int               pf_insert_src_node(struct pf_src_node **,
 279                             struct pf_rule *, struct pf_addr *, sa_family_t);
 280 static u_int             pf_purge_expired_states(u_int, int);
 281 static void              pf_purge_unlinked_rules(void);
 282 static int               pf_mtag_uminit(void *, int, int);
 283 static void              pf_mtag_free(struct m_tag *);
 284 #ifdef INET
 285 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 286                             struct ifnet *, struct pf_state *,
 287                             struct pf_pdesc *);
 288 #endif /* INET */
 289 #ifdef INET6
 290 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 291                             struct pf_addr *, u_int8_t);
 292 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 293                             struct ifnet *, struct pf_state *,
 294                             struct pf_pdesc *);
 295 #endif /* INET6 */
 296
 297 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 298
 299 VNET_DECLARE(int, pf_end_threads);
 300
 301 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 302
 303 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 304                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 305
 306 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 307         do {                                                            \
 308                 (s) = pf_find_state((i), (k), (d));                     \
 309                 if ((s) == NULL)                                        \
 310                         return (PF_DROP);                               \
 311                 if (PACKET_LOOPED(pd))                                  \
 312                         return (PF_PASS);                               \
 313                 if ((d) == PF_OUT &&                                    \
 314                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 315                     (s)->rule.ptr->direction == PF_OUT) ||              \
 316                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 317                     (s)->rule.ptr->direction == PF_IN)) &&              \
 318                     (s)->rt_kif != NULL &&                              \
 319                     (s)->rt_kif != (i))                                 \
 320                         return (PF_PASS);                               \
 321         } while (0)
 322
 323 #define BOUND_IFACE(r, k) \
 324         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 325
 326 #define STATE_INC_COUNTERS(s)                                           \
 327         do {                                                            \
 328                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 329                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 330                 if (s->anchor.ptr != NULL) {                            \
 331                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 332                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 333                 }                                                       \
 334                 if (s->nat_rule.ptr != NULL) {                          \
 335                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 336                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 337                 }                                                       \
 338         } while (0)
 339
 340 #define STATE_DEC_COUNTERS(s)                                           \
 341         do {                                                            \
 342                 if (s->nat_rule.ptr != NULL)                            \
 343                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 344                 if (s->anchor.ptr != NULL)                              \
 345                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 346                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 347         } while (0)
 348
 349 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 350 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 351 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 352 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 353
 354 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 355
 356 u_long  pf_hashmask;
 357 u_long  pf_srchashmask;
 358 static u_long   pf_hashsize;
 359 static u_long   pf_srchashsize;
 360
 361 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 362     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 363 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 364     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 365
 366 VNET_DEFINE(void *, pf_swi_cookie);
 367
 368 VNET_DEFINE(uint32_t, pf_hashseed);
 369 #define V_pf_hashseed   VNET(pf_hashseed)
 370
 371 int
 372 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 373 {
 374
 375         switch (af) {
 376 #ifdef INET
 377         case AF_INET:
 378                 if (a->addr32[0] > b->addr32[0])
 379                         return (1);
 380                 if (a->addr32[0] < b->addr32[0])
 381                         return (-1);
 382                 break;
 383 #endif /* INET */
 384 #ifdef INET6
 385         case AF_INET6:
 386                 if (a->addr32[3] > b->addr32[3])
 387                         return (1);
 388                 if (a->addr32[3] < b->addr32[3])
 389                         return (-1);
 390                 if (a->addr32[2] > b->addr32[2])
 391                         return (1);
 392                 if (a->addr32[2] < b->addr32[2])
 393                         return (-1);
 394                 if (a->addr32[1] > b->addr32[1])
 395                         return (1);
 396                 if (a->addr32[1] < b->addr32[1])
 397                         return (-1);
 398                 if (a->addr32[0] > b->addr32[0])
 399                         return (1);
 400                 if (a->addr32[0] < b->addr32[0])
 401                         return (-1);
 402                 break;
 403 #endif /* INET6 */
 404         default:
 405                 panic("%s: unknown address family %u", __func__, af);
 406         }
 407         return (0);
 408 }
 409
 410 static __inline uint32_t
 411 pf_hashkey(struct pf_state_key *sk)
 412 {
 413         uint32_t h;
 414
 415         h = murmur3_aligned_32((uint32_t *)sk,
 416                                sizeof(struct pf_state_key_cmp),
 417                                V_pf_hashseed);
 418
 419         return (h & pf_hashmask);
 420 }
 421
 422 static __inline uint32_t
 423 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 424 {
 425         uint32_t h;
 426
 427         switch (af) {
 428         case AF_INET:
 429                 h = murmur3_aligned_32((uint32_t *)&addr->v4,
 430                                        sizeof(addr->v4), V_pf_hashseed);
 431                 break;
 432         case AF_INET6:
 433                 h = murmur3_aligned_32((uint32_t *)&addr->v6,
 434                                        sizeof(addr->v6), V_pf_hashseed);
 435                 break;
 436         default:
 437                 panic("%s: unknown address family %u", __func__, af);
 438         }
 439
 440         return (h & pf_srchashmask);
 441 }
 442
 443 #ifdef INET6
 444 void
 445 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 446 {
 447         switch (af) {
 448 #ifdef INET
 449         case AF_INET:
 450                 dst->addr32[0] = src->addr32[0];
 451                 break;
 452 #endif /* INET */
 453         case AF_INET6:
 454                 dst->addr32[0] = src->addr32[0];
 455                 dst->addr32[1] = src->addr32[1];
 456                 dst->addr32[2] = src->addr32[2];
 457                 dst->addr32[3] = src->addr32[3];
 458                 break;
 459         }
 460 }
 461 #endif /* INET6 */
 462
 463 static void
 464 pf_init_threshold(struct pf_threshold *threshold,
 465     u_int32_t limit, u_int32_t seconds)
 466 {
 467         threshold->limit = limit * PF_THRESHOLD_MULT;
 468         threshold->seconds = seconds;
 469         threshold->count = 0;
 470         threshold->last = time_uptime;
 471 }
 472
 473 static void
 474 pf_add_threshold(struct pf_threshold *threshold)
 475 {
 476         u_int32_t t = time_uptime, diff = t - threshold->last;
 477
 478         if (diff >= threshold->seconds)
 479                 threshold->count = 0;
 480         else
 481                 threshold->count -= threshold->count * diff /
 482                     threshold->seconds;
 483         threshold->count += PF_THRESHOLD_MULT;
 484         threshold->last = t;
 485 }
 486
 487 static int
 488 pf_check_threshold(struct pf_threshold *threshold)
 489 {
 490         return (threshold->count > threshold->limit);
 491 }
 492
 493 static int
 494 pf_src_connlimit(struct pf_state **state)
 495 {
 496         struct pf_overload_entry *pfoe;
 497         int bad = 0;
 498
 499         PF_STATE_LOCK_ASSERT(*state);
 500
 501         (*state)->src_node->conn++;
 502         (*state)->src.tcp_est = 1;
 503         pf_add_threshold(&(*state)->src_node->conn_rate);
 504
 505         if ((*state)->rule.ptr->max_src_conn &&
 506             (*state)->rule.ptr->max_src_conn <
 507             (*state)->src_node->conn) {
 508                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 509                 bad++;
 510         }
 511
 512         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 513             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 514                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 515                 bad++;
 516         }
 517
 518         if (!bad)
 519                 return (0);
 520
 521         /* Kill this state. */
 522         (*state)->timeout = PFTM_PURGE;
 523         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 524
 525         if ((*state)->rule.ptr->overload_tbl == NULL)
 526                 return (1);
 527
 528         /* Schedule overloading and flushing task. */
 529         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 530         if (pfoe == NULL)
 531                 return (1);     /* too bad :( */
 532
 533         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 534         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 535         pfoe->rule = (*state)->rule.ptr;
 536         pfoe->dir = (*state)->direction;
 537         PF_OVERLOADQ_LOCK();
 538         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 539         PF_OVERLOADQ_UNLOCK();
 540         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 541
 542         return (1);
 543 }
 544
 545 static void
 546 pf_overload_task(void *v, int pending)
 547 {
 548         struct pf_overload_head queue;
 549         struct pfr_addr p;
 550         struct pf_overload_entry *pfoe, *pfoe1;
 551         uint32_t killed = 0;
 552
 553         CURVNET_SET((struct vnet *)v);
 554
 555         PF_OVERLOADQ_LOCK();
 556         queue = V_pf_overloadqueue;
 557         SLIST_INIT(&V_pf_overloadqueue);
 558         PF_OVERLOADQ_UNLOCK();
 559
 560         bzero(&p, sizeof(p));
 561         SLIST_FOREACH(pfoe, &queue, next) {
 562                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 563                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 564                         printf("%s: blocking address ", __func__);
 565                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 566                         printf("\n");
 567                 }
 568
 569                 p.pfra_af = pfoe->af;
 570                 switch (pfoe->af) {
 571 #ifdef INET
 572                 case AF_INET:
 573                         p.pfra_net = 32;
 574                         p.pfra_ip4addr = pfoe->addr.v4;
 575                         break;
 576 #endif
 577 #ifdef INET6
 578                 case AF_INET6:
 579                         p.pfra_net = 128;
 580                         p.pfra_ip6addr = pfoe->addr.v6;
 581                         break;
 582 #endif
 583                 }
 584
 585                 PF_RULES_WLOCK();
 586                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 587                 PF_RULES_WUNLOCK();
 588         }
 589
 590         /*
 591          * Remove those entries, that don't need flushing.
 592          */
 593         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 594                 if (pfoe->rule->flush == 0) {
 595                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 596                         free(pfoe, M_PFTEMP);
 597                 } else
 598                         counter_u64_add(
 599                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 600
 601         /* If nothing to flush, return. */
 602         if (SLIST_EMPTY(&queue)) {
 603                 CURVNET_RESTORE();
 604                 return;
 605         }
 606
 607         for (int i = 0; i <= pf_hashmask; i++) {
 608                 struct pf_idhash *ih = &V_pf_idhash[i];
 609                 struct pf_state_key *sk;
 610                 struct pf_state *s;
 611
 612                 PF_HASHROW_LOCK(ih);
 613                 LIST_FOREACH(s, &ih->states, entry) {
 614                     sk = s->key[PF_SK_WIRE];
 615                     SLIST_FOREACH(pfoe, &queue, next)
 616                         if (sk->af == pfoe->af &&
 617                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 618                             pfoe->rule == s->rule.ptr) &&
 619                             ((pfoe->dir == PF_OUT &&
 620                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 621                             (pfoe->dir == PF_IN &&
 622                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 623                                 s->timeout = PFTM_PURGE;
 624                                 s->src.state = s->dst.state = TCPS_CLOSED;
 625                                 killed++;
 626                         }
 627                 }
 628                 PF_HASHROW_UNLOCK(ih);
 629         }
 630         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 631                 free(pfoe, M_PFTEMP);
 632         if (V_pf_status.debug >= PF_DEBUG_MISC)
 633                 printf("%s: %u states killed", __func__, killed);
 634
 635         CURVNET_RESTORE();
 636 }
 637
 638 /*
 639  * Can return locked on failure, so that we can consistently
 640  * allocate and insert a new one.
 641  */
 642 struct pf_src_node *
 643 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 644         int returnlocked)
 645 {
 646         struct pf_srchash *sh;
 647         struct pf_src_node *n;
 648
 649         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 650
 651         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 652         PF_HASHROW_LOCK(sh);
 653         LIST_FOREACH(n, &sh->nodes, entry)
 654                 if (n->rule.ptr == rule && n->af == af &&
 655                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 656                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 657                         break;
 658         if (n != NULL) {
 659                 n->states++;
 660                 PF_HASHROW_UNLOCK(sh);
 661         } else if (returnlocked == 0)
 662                 PF_HASHROW_UNLOCK(sh);
 663
 664         return (n);
 665 }
 666
 667 static int
 668 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 669     struct pf_addr *src, sa_family_t af)
 670 {
 671
 672         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 673             rule->rpool.opts & PF_POOL_STICKYADDR),
 674             ("%s for non-tracking rule %p", __func__, rule));
 675
 676         if (*sn == NULL)
 677                 *sn = pf_find_src_node(src, rule, af, 1);
 678
 679         if (*sn == NULL) {
 680                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 681
 682                 PF_HASHROW_ASSERT(sh);
 683
 684                 if (!rule->max_src_nodes ||
 685                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 686                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 687                 else
 688                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 689                             1);
 690                 if ((*sn) == NULL) {
 691                         PF_HASHROW_UNLOCK(sh);
 692                         return (-1);
 693                 }
 694
 695                 pf_init_threshold(&(*sn)->conn_rate,
 696                     rule->max_src_conn_rate.limit,
 697                     rule->max_src_conn_rate.seconds);
 698
 699                 (*sn)->af = af;
 700                 (*sn)->rule.ptr = rule;
 701                 PF_ACPY(&(*sn)->addr, src, af);
 702                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 703                 (*sn)->creation = time_uptime;
 704                 (*sn)->ruletype = rule->action;
 705                 (*sn)->states = 1;
 706                 if ((*sn)->rule.ptr != NULL)
 707                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 708                 PF_HASHROW_UNLOCK(sh);
 709                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 710         } else {
 711                 if (rule->max_src_states &&
 712                     (*sn)->states >= rule->max_src_states) {
 713                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 714                             1);
 715                         return (-1);
 716                 }
 717         }
 718         return (0);
 719 }
 720
 721 void
 722 pf_unlink_src_node(struct pf_src_node *src)
 723 {
 724
 725         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 726         LIST_REMOVE(src, entry);
 727         if (src->rule.ptr)
 728                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 729 }
 730
 731 u_int
 732 pf_free_src_nodes(struct pf_src_node_list *head)
 733 {
 734         struct pf_src_node *sn, *tmp;
 735         u_int count = 0;
 736
 737         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 738                 uma_zfree(V_pf_sources_z, sn);
 739                 count++;
 740         }
 741
 742         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 743
 744         return (count);
 745 }
 746
 747 void
 748 pf_mtag_initialize()
 749 {
 750
 751         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 752             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 753             UMA_ALIGN_PTR, 0);
 754 }
 755
 756 /* Per-vnet data storage structures initialization. */
 757 void
 758 pf_initialize()
 759 {
 760         struct pf_keyhash       *kh;
 761         struct pf_idhash        *ih;
 762         struct pf_srchash       *sh;
 763         u_int i;
 764
 765         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &pf_hashsize);
 766         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 767                 pf_hashsize = PF_HASHSIZ;
 768         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &pf_srchashsize);
 769         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 770                 pf_srchashsize = PF_HASHSIZ / 4;
 771
 772         V_pf_hashseed = arc4random();
 773
 774         /* States and state keys storage. */
 775         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 776             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 777         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 778         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 779         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 780
 781         V_pf_state_key_z = uma_zcreate("pf state keys",
 782             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 783             UMA_ALIGN_PTR, 0);
 784         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 785             M_PFHASH, M_WAITOK | M_ZERO);
 786         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 787             M_PFHASH, M_WAITOK | M_ZERO);
 788         pf_hashmask = pf_hashsize - 1;
 789         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 790             i++, kh++, ih++) {
 791                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 792                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 793         }
 794
 795         /* Source nodes. */
 796         V_pf_sources_z = uma_zcreate("pf source nodes",
 797             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 798             0);
 799         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 800         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 801         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 802         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 803           M_PFHASH, M_WAITOK|M_ZERO);
 804         pf_srchashmask = pf_srchashsize - 1;
 805         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 806                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 807
 808         /* ALTQ */
 809         TAILQ_INIT(&V_pf_altqs[0]);
 810         TAILQ_INIT(&V_pf_altqs[1]);
 811         TAILQ_INIT(&V_pf_pabuf);
 812         V_pf_altqs_active = &V_pf_altqs[0];
 813         V_pf_altqs_inactive = &V_pf_altqs[1];
 814
 815
 816         /* Send & overload+flush queues. */
 817         STAILQ_INIT(&V_pf_sendqueue);
 818         SLIST_INIT(&V_pf_overloadqueue);
 819         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 820         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 821         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 822             MTX_DEF);
 823
 824         /* Unlinked, but may be referenced rules. */
 825         TAILQ_INIT(&V_pf_unlinked_rules);
 826         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 827 }
 828
 829 void
 830 pf_mtag_cleanup()
 831 {
 832
 833         uma_zdestroy(pf_mtag_z);
 834 }
 835
 836 void
 837 pf_cleanup()
 838 {
 839         struct pf_keyhash       *kh;
 840         struct pf_idhash        *ih;
 841         struct pf_srchash       *sh;
 842         struct pf_send_entry    *pfse, *next;
 843         u_int i;
 844
 845         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 846             i++, kh++, ih++) {
 847                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 848                     __func__));
 849                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 850                     __func__));
 851                 mtx_destroy(&kh->lock);
 852                 mtx_destroy(&ih->lock);
 853         }
 854         free(V_pf_keyhash, M_PFHASH);
 855         free(V_pf_idhash, M_PFHASH);
 856
 857         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 858                 KASSERT(LIST_EMPTY(&sh->nodes),
 859                     ("%s: source node hash not empty", __func__));
 860                 mtx_destroy(&sh->lock);
 861         }
 862         free(V_pf_srchash, M_PFHASH);
 863
 864         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 865                 m_freem(pfse->pfse_m);
 866                 free(pfse, M_PFTEMP);
 867         }
 868
 869         mtx_destroy(&pf_sendqueue_mtx);
 870         mtx_destroy(&pf_overloadqueue_mtx);
 871         mtx_destroy(&pf_unlnkdrules_mtx);
 872
 873         uma_zdestroy(V_pf_sources_z);
 874         uma_zdestroy(V_pf_state_z);
 875         uma_zdestroy(V_pf_state_key_z);
 876 }
 877
 878 static int
 879 pf_mtag_uminit(void *mem, int size, int how)
 880 {
 881         struct m_tag *t;
 882
 883         t = (struct m_tag *)mem;
 884         t->m_tag_cookie = MTAG_ABI_COMPAT;
 885         t->m_tag_id = PACKET_TAG_PF;
 886         t->m_tag_len = sizeof(struct pf_mtag);
 887         t->m_tag_free = pf_mtag_free;
 888
 889         return (0);
 890 }
 891
 892 static void
 893 pf_mtag_free(struct m_tag *t)
 894 {
 895
 896         uma_zfree(pf_mtag_z, t);
 897 }
 898
 899 struct pf_mtag *
 900 pf_get_mtag(struct mbuf *m)
 901 {
 902         struct m_tag *mtag;
 903
 904         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 905                 return ((struct pf_mtag *)(mtag + 1));
 906
 907         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 908         if (mtag == NULL)
 909                 return (NULL);
 910         bzero(mtag + 1, sizeof(struct pf_mtag));
 911         m_tag_prepend(m, mtag);
 912
 913         return ((struct pf_mtag *)(mtag + 1));
 914 }
 915
 916 static int
 917 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 918     struct pf_state *s)
 919 {
 920         struct pf_keyhash       *khs, *khw, *kh;
 921         struct pf_state_key     *sk, *cur;
 922         struct pf_state         *si, *olds = NULL;
 923         int idx;
 924
 925         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 926         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 927         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 928
 929         /*
 930          * We need to lock hash slots of both keys. To avoid deadlock
 931          * we always lock the slot with lower address first. Unlock order
 932          * isn't important.
 933          *
 934          * We also need to lock ID hash slot before dropping key
 935          * locks. On success we return with ID hash slot locked.
 936          */
 937
 938         if (skw == sks) {
 939                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 940                 PF_HASHROW_LOCK(khs);
 941         } else {
 942                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 943                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 944                 if (khs == khw) {
 945                         PF_HASHROW_LOCK(khs);
 946                 } else if (khs < khw) {
 947                         PF_HASHROW_LOCK(khs);
 948                         PF_HASHROW_LOCK(khw);
 949                 } else {
 950                         PF_HASHROW_LOCK(khw);
 951                         PF_HASHROW_LOCK(khs);
 952                 }
 953         }
 954
 955 #define KEYS_UNLOCK()   do {                    \
 956         if (khs != khw) {                       \
 957                 PF_HASHROW_UNLOCK(khs);         \
 958                 PF_HASHROW_UNLOCK(khw);         \
 959         } else                                  \
 960                 PF_HASHROW_UNLOCK(khs);         \
 961 } while (0)
 962
 963         /*
 964          * First run: start with wire key.
 965          */
 966         sk = skw;
 967         kh = khw;
 968         idx = PF_SK_WIRE;
 969
 970 keyattach:
 971         LIST_FOREACH(cur, &kh->keys, entry)
 972                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 973                         break;
 974
 975         if (cur != NULL) {
 976                 /* Key exists. Check for same kif, if none, add to key. */
 977                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 978                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 979
 980                         PF_HASHROW_LOCK(ih);
 981                         if (si->kif == s->kif &&
 982                             si->direction == s->direction) {
 983                                 if (sk->proto == IPPROTO_TCP &&
 984                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 985                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 986                                         /*
 987                                          * New state matches an old >FIN_WAIT_2
 988                                          * state. We can't drop key hash locks,
 989                                          * thus we can't unlink it properly.
 990                                          *
 991                                          * As a workaround we drop it into
 992                                          * TCPS_CLOSED state, schedule purge
 993                                          * ASAP and push it into the very end
 994                                          * of the slot TAILQ, so that it won't
 995                                          * conflict with our new state.
 996                                          */
 997                                         si->src.state = si->dst.state =
 998                                             TCPS_CLOSED;
 999                                         si->timeout = PFTM_PURGE;
1000                                         olds = si;
1001                                 } else {
1002                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1003                                                 printf("pf: %s key attach "
1004                                                     "failed on %s: ",
1005                                                     (idx == PF_SK_WIRE) ?
1006                                                     "wire" : "stack",
1007                                                     s->kif->pfik_name);
1008                                                 pf_print_state_parts(s,
1009                                                     (idx == PF_SK_WIRE) ?
1010                                                     sk : NULL,
1011                                                     (idx == PF_SK_STACK) ?
1012                                                     sk : NULL);
1013                                                 printf(", existing: ");
1014                                                 pf_print_state_parts(si,
1015                                                     (idx == PF_SK_WIRE) ?
1016                                                     sk : NULL,
1017                                                     (idx == PF_SK_STACK) ?
1018                                                     sk : NULL);
1019                                                 printf("\n");
1020                                         }
1021                                         PF_HASHROW_UNLOCK(ih);
1022                                         KEYS_UNLOCK();
1023                                         uma_zfree(V_pf_state_key_z, sk);
1024                                         if (idx == PF_SK_STACK)
1025                                                 pf_detach_state(s);
1026                                         return (EEXIST); /* collision! */
1027                                 }
1028                         }
1029                         PF_HASHROW_UNLOCK(ih);
1030                 }
1031                 uma_zfree(V_pf_state_key_z, sk);
1032                 s->key[idx] = cur;
1033         } else {
1034                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1035                 s->key[idx] = sk;
1036         }
1037
1038 stateattach:
1039         /* List is sorted, if-bound states before floating. */
1040         if (s->kif == V_pfi_all)
1041                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1042         else
1043                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1044
1045         if (olds) {
1046                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1047                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1048                     key_list[idx]);
1049                 olds = NULL;
1050         }
1051
1052         /*
1053          * Attach done. See how should we (or should not?)
1054          * attach a second key.
1055          */
1056         if (sks == skw) {
1057                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1058                 idx = PF_SK_STACK;
1059                 sks = NULL;
1060                 goto stateattach;
1061         } else if (sks != NULL) {
1062                 /*
1063                  * Continue attaching with stack key.
1064                  */
1065                 sk = sks;
1066                 kh = khs;
1067                 idx = PF_SK_STACK;
1068                 sks = NULL;
1069                 goto keyattach;
1070         }
1071
1072         PF_STATE_LOCK(s);
1073         KEYS_UNLOCK();
1074
1075         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1076             ("%s failure", __func__));
1077
1078         return (0);
1079 #undef  KEYS_UNLOCK
1080 }
1081
1082 static void
1083 pf_detach_state(struct pf_state *s)
1084 {
1085         struct pf_state_key *sks = s->key[PF_SK_STACK];
1086         struct pf_keyhash *kh;
1087
1088         if (sks != NULL) {
1089                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1090                 PF_HASHROW_LOCK(kh);
1091                 if (s->key[PF_SK_STACK] != NULL)
1092                         pf_state_key_detach(s, PF_SK_STACK);
1093                 /*
1094                  * If both point to same key, then we are done.
1095                  */
1096                 if (sks == s->key[PF_SK_WIRE]) {
1097                         pf_state_key_detach(s, PF_SK_WIRE);
1098                         PF_HASHROW_UNLOCK(kh);
1099                         return;
1100                 }
1101                 PF_HASHROW_UNLOCK(kh);
1102         }
1103
1104         if (s->key[PF_SK_WIRE] != NULL) {
1105                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1106                 PF_HASHROW_LOCK(kh);
1107                 if (s->key[PF_SK_WIRE] != NULL)
1108                         pf_state_key_detach(s, PF_SK_WIRE);
1109                 PF_HASHROW_UNLOCK(kh);
1110         }
1111 }
1112
1113 static void
1114 pf_state_key_detach(struct pf_state *s, int idx)
1115 {
1116         struct pf_state_key *sk = s->key[idx];
1117 #ifdef INVARIANTS
1118         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1119
1120         PF_HASHROW_ASSERT(kh);
1121 #endif
1122         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1123         s->key[idx] = NULL;
1124
1125         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1126                 LIST_REMOVE(sk, entry);
1127                 uma_zfree(V_pf_state_key_z, sk);
1128         }
1129 }
1130
1131 static int
1132 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1133 {
1134         struct pf_state_key *sk = mem;
1135
1136         bzero(sk, sizeof(struct pf_state_key_cmp));
1137         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1138         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1139
1140         return (0);
1141 }
1142
1143 struct pf_state_key *
1144 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1145         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1146 {
1147         struct pf_state_key *sk;
1148
1149         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1150         if (sk == NULL)
1151                 return (NULL);
1152
1153         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1154         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1155         sk->port[pd->sidx] = sport;
1156         sk->port[pd->didx] = dport;
1157         sk->proto = pd->proto;
1158         sk->af = pd->af;
1159
1160         return (sk);
1161 }
1162
1163 struct pf_state_key *
1164 pf_state_key_clone(struct pf_state_key *orig)
1165 {
1166         struct pf_state_key *sk;
1167
1168         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1169         if (sk == NULL)
1170                 return (NULL);
1171
1172         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1173
1174         return (sk);
1175 }
1176
1177 int
1178 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1179     struct pf_state_key *sks, struct pf_state *s)
1180 {
1181         struct pf_idhash *ih;
1182         struct pf_state *cur;
1183         int error;
1184
1185         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1186             ("%s: sks not pristine", __func__));
1187         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1188             ("%s: skw not pristine", __func__));
1189         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1190
1191         s->kif = kif;
1192
1193         if (s->id == 0 && s->creatorid == 0) {
1194                 /* XXX: should be atomic, but probability of collision low */
1195                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1196                         V_pf_stateid[curcpu] = 1;
1197                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1198                 s->id = htobe64(s->id);
1199                 s->creatorid = V_pf_status.hostid;
1200         }
1201
1202         /* Returns with ID locked on success. */
1203         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1204                 return (error);
1205
1206         ih = &V_pf_idhash[PF_IDHASH(s)];
1207         PF_HASHROW_ASSERT(ih);
1208         LIST_FOREACH(cur, &ih->states, entry)
1209                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1210                         break;
1211
1212         if (cur != NULL) {
1213                 PF_HASHROW_UNLOCK(ih);
1214                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1215                         printf("pf: state ID collision: "
1216                             "id: %016llx creatorid: %08x\n",
1217                             (unsigned long long)be64toh(s->id),
1218                             ntohl(s->creatorid));
1219                 }
1220                 pf_detach_state(s);
1221                 return (EEXIST);
1222         }
1223         LIST_INSERT_HEAD(&ih->states, s, entry);
1224         /* One for keys, one for ID hash. */
1225         refcount_init(&s->refs, 2);
1226
1227         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1228         if (pfsync_insert_state_ptr != NULL)
1229                 pfsync_insert_state_ptr(s);
1230
1231         /* Returns locked. */
1232         return (0);
1233 }
1234
1235 /*
1236  * Find state by ID: returns with locked row on success.
1237  */
1238 struct pf_state *
1239 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1240 {
1241         struct pf_idhash *ih;
1242         struct pf_state *s;
1243
1244         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1245
1246         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1247
1248         PF_HASHROW_LOCK(ih);
1249         LIST_FOREACH(s, &ih->states, entry)
1250                 if (s->id == id && s->creatorid == creatorid)
1251                         break;
1252
1253         if (s == NULL)
1254                 PF_HASHROW_UNLOCK(ih);
1255
1256         return (s);
1257 }
1258
1259 /*
1260  * Find state by key.
1261  * Returns with ID hash slot locked on success.
1262  */
1263 static struct pf_state *
1264 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1265 {
1266         struct pf_keyhash       *kh;
1267         struct pf_state_key     *sk;
1268         struct pf_state         *s;
1269         int idx;
1270
1271         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1272
1273         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1274
1275         PF_HASHROW_LOCK(kh);
1276         LIST_FOREACH(sk, &kh->keys, entry)
1277                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1278                         break;
1279         if (sk == NULL) {
1280                 PF_HASHROW_UNLOCK(kh);
1281                 return (NULL);
1282         }
1283
1284         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1285
1286         /* List is sorted, if-bound states before floating ones. */
1287         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1288                 if (s->kif == V_pfi_all || s->kif == kif) {
1289                         PF_STATE_LOCK(s);
1290                         PF_HASHROW_UNLOCK(kh);
1291                         if (s->timeout >= PFTM_MAX) {
1292                                 /*
1293                                  * State is either being processed by
1294                                  * pf_unlink_state() in an other thread, or
1295                                  * is scheduled for immediate expiry.
1296                                  */
1297                                 PF_STATE_UNLOCK(s);
1298                                 return (NULL);
1299                         }
1300                         return (s);
1301                 }
1302         PF_HASHROW_UNLOCK(kh);
1303
1304         return (NULL);
1305 }
1306
1307 struct pf_state *
1308 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1309 {
1310         struct pf_keyhash       *kh;
1311         struct pf_state_key     *sk;
1312         struct pf_state         *s, *ret = NULL;
1313         int                      idx, inout = 0;
1314
1315         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1316
1317         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1318
1319         PF_HASHROW_LOCK(kh);
1320         LIST_FOREACH(sk, &kh->keys, entry)
1321                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1322                         break;
1323         if (sk == NULL) {
1324                 PF_HASHROW_UNLOCK(kh);
1325                 return (NULL);
1326         }
1327         switch (dir) {
1328         case PF_IN:
1329                 idx = PF_SK_WIRE;
1330                 break;
1331         case PF_OUT:
1332                 idx = PF_SK_STACK;
1333                 break;
1334         case PF_INOUT:
1335                 idx = PF_SK_WIRE;
1336                 inout = 1;
1337                 break;
1338         default:
1339                 panic("%s: dir %u", __func__, dir);
1340         }
1341 second_run:
1342         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1343                 if (more == NULL) {
1344                         PF_HASHROW_UNLOCK(kh);
1345                         return (s);
1346                 }
1347
1348                 if (ret)
1349                         (*more)++;
1350                 else
1351                         ret = s;
1352         }
1353         if (inout == 1) {
1354                 inout = 0;
1355                 idx = PF_SK_STACK;
1356                 goto second_run;
1357         }
1358         PF_HASHROW_UNLOCK(kh);
1359
1360         return (ret);
1361 }
1362
1363 /* END state table stuff */
1364
1365 static void
1366 pf_send(struct pf_send_entry *pfse)
1367 {
1368
1369         PF_SENDQ_LOCK();
1370         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1371         PF_SENDQ_UNLOCK();
1372         swi_sched(V_pf_swi_cookie, 0);
1373 }
1374
1375 void
1376 pf_intr(void *v)
1377 {
1378         struct pf_send_head queue;
1379         struct pf_send_entry *pfse, *next;
1380
1381         CURVNET_SET((struct vnet *)v);
1382
1383         PF_SENDQ_LOCK();
1384         queue = V_pf_sendqueue;
1385         STAILQ_INIT(&V_pf_sendqueue);
1386         PF_SENDQ_UNLOCK();
1387
1388         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1389                 switch (pfse->pfse_type) {
1390 #ifdef INET
1391                 case PFSE_IP:
1392                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1393                         break;
1394                 case PFSE_ICMP:
1395                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1396                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1397                         break;
1398 #endif /* INET */
1399 #ifdef INET6
1400                 case PFSE_IP6:
1401                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1402                             NULL);
1403                         break;
1404                 case PFSE_ICMP6:
1405                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1406                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1407                         break;
1408 #endif /* INET6 */
1409                 default:
1410                         panic("%s: unknown type", __func__);
1411                 }
1412                 free(pfse, M_PFTEMP);
1413         }
1414         CURVNET_RESTORE();
1415 }
1416
1417 void
1418 pf_purge_thread(void *v)
1419 {
1420         u_int idx = 0;
1421
1422         CURVNET_SET((struct vnet *)v);
1423
1424         for (;;) {
1425                 PF_RULES_RLOCK();
1426                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1427
1428                 if (V_pf_end_threads) {
1429                         /*
1430                          * To cleanse up all kifs and rules we need
1431                          * two runs: first one clears reference flags,
1432                          * then pf_purge_expired_states() doesn't
1433                          * raise them, and then second run frees.
1434                          */
1435                         PF_RULES_RUNLOCK();
1436                         pf_purge_unlinked_rules();
1437                         pfi_kif_purge();
1438
1439                         /*
1440                          * Now purge everything.
1441                          */
1442                         pf_purge_expired_states(0, pf_hashmask);
1443                         pf_purge_expired_fragments();
1444                         pf_purge_expired_src_nodes();
1445
1446                         /*
1447                          * Now all kifs & rules should be unreferenced,
1448                          * thus should be successfully freed.
1449                          */
1450                         pf_purge_unlinked_rules();
1451                         pfi_kif_purge();
1452
1453                         /*
1454                          * Announce success and exit.
1455                          */
1456                         PF_RULES_RLOCK();
1457                         V_pf_end_threads++;
1458                         PF_RULES_RUNLOCK();
1459                         wakeup(pf_purge_thread);
1460                         kproc_exit(0);
1461                 }
1462                 PF_RULES_RUNLOCK();
1463
1464                 /* Process 1/interval fraction of the state table every run. */
1465                 idx = pf_purge_expired_states(idx, pf_hashmask /
1466                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1467
1468                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1469                 if (idx == 0) {
1470                         /*
1471                          * Order is important:
1472                          * - states and src nodes reference rules
1473                          * - states and rules reference kifs
1474                          */
1475                         pf_purge_expired_fragments();
1476                         pf_purge_expired_src_nodes();
1477                         pf_purge_unlinked_rules();
1478                         pfi_kif_purge();
1479                 }
1480         }
1481         /* not reached */
1482         CURVNET_RESTORE();
1483 }
1484
1485 u_int32_t
1486 pf_state_expires(const struct pf_state *state)
1487 {
1488         u_int32_t       timeout;
1489         u_int32_t       start;
1490         u_int32_t       end;
1491         u_int32_t       states;
1492
1493         /* handle all PFTM_* > PFTM_MAX here */
1494         if (state->timeout == PFTM_PURGE)
1495                 return (time_uptime);
1496         KASSERT(state->timeout != PFTM_UNLINKED,
1497             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1498         KASSERT((state->timeout < PFTM_MAX),
1499             ("pf_state_expires: timeout > PFTM_MAX"));
1500         timeout = state->rule.ptr->timeout[state->timeout];
1501         if (!timeout)
1502                 timeout = V_pf_default_rule.timeout[state->timeout];
1503         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1504         if (start) {
1505                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1506                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1507         } else {
1508                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1509                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1510                 states = V_pf_status.states;
1511         }
1512         if (end && states > start && start < end) {
1513                 if (states < end)
1514                         return (state->expire + timeout * (end - states) /
1515                             (end - start));
1516                 else
1517                         return (time_uptime);
1518         }
1519         return (state->expire + timeout);
1520 }
1521
1522 void
1523 pf_purge_expired_src_nodes()
1524 {
1525         struct pf_src_node_list  freelist;
1526         struct pf_srchash       *sh;
1527         struct pf_src_node      *cur, *next;
1528         int i;
1529
1530         LIST_INIT(&freelist);
1531         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1532             PF_HASHROW_LOCK(sh);
1533             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1534                 if (cur->states == 0 && cur->expire <= time_uptime) {
1535                         pf_unlink_src_node(cur);
1536                         LIST_INSERT_HEAD(&freelist, cur, entry);
1537                 } else if (cur->rule.ptr != NULL)
1538                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1539             PF_HASHROW_UNLOCK(sh);
1540         }
1541
1542         pf_free_src_nodes(&freelist);
1543
1544         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1545 }
1546
1547 static void
1548 pf_src_tree_remove_state(struct pf_state *s)
1549 {
1550         struct pf_src_node *sn;
1551         struct pf_srchash *sh;
1552         uint32_t timeout;
1553
1554         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1555             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1556             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1557
1558         if (s->src_node != NULL) {
1559                 sn = s->src_node;
1560                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1561                 PF_HASHROW_LOCK(sh);
1562                 if (s->src.tcp_est)
1563                         --sn->conn;
1564                 if (--sn->states == 0)
1565                         sn->expire = time_uptime + timeout;
1566                 PF_HASHROW_UNLOCK(sh);
1567         }
1568         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1569                 sn = s->nat_src_node;
1570                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1571                 PF_HASHROW_LOCK(sh);
1572                 if (--sn->states == 0)
1573                         sn->expire = time_uptime + timeout;
1574                 PF_HASHROW_UNLOCK(sh);
1575         }
1576         s->src_node = s->nat_src_node = NULL;
1577 }
1578
1579 /*
1580  * Unlink and potentilly free a state. Function may be
1581  * called with ID hash row locked, but always returns
1582  * unlocked, since it needs to go through key hash locking.
1583  */
1584 int
1585 pf_unlink_state(struct pf_state *s, u_int flags)
1586 {
1587         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1588
1589         if ((flags & PF_ENTER_LOCKED) == 0)
1590                 PF_HASHROW_LOCK(ih);
1591         else
1592                 PF_HASHROW_ASSERT(ih);
1593
1594         if (s->timeout == PFTM_UNLINKED) {
1595                 /*
1596                  * State is being processed
1597                  * by pf_unlink_state() in
1598                  * an other thread.
1599                  */
1600                 PF_HASHROW_UNLOCK(ih);
1601                 return (0);     /* XXXGL: undefined actually */
1602         }
1603
1604         if (s->src.state == PF_TCPS_PROXY_DST) {
1605                 /* XXX wire key the right one? */
1606                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1607                     &s->key[PF_SK_WIRE]->addr[1],
1608                     &s->key[PF_SK_WIRE]->addr[0],
1609                     s->key[PF_SK_WIRE]->port[1],
1610                     s->key[PF_SK_WIRE]->port[0],
1611                     s->src.seqhi, s->src.seqlo + 1,
1612                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1613         }
1614
1615         LIST_REMOVE(s, entry);
1616         pf_src_tree_remove_state(s);
1617
1618         if (pfsync_delete_state_ptr != NULL)
1619                 pfsync_delete_state_ptr(s);
1620
1621         STATE_DEC_COUNTERS(s);
1622
1623         s->timeout = PFTM_UNLINKED;
1624
1625         PF_HASHROW_UNLOCK(ih);
1626
1627         pf_detach_state(s);
1628         refcount_release(&s->refs);
1629
1630         return (pf_release_state(s));
1631 }
1632
1633 void
1634 pf_free_state(struct pf_state *cur)
1635 {
1636
1637         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1638         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1639             cur->timeout));
1640
1641         pf_normalize_tcp_cleanup(cur);
1642         uma_zfree(V_pf_state_z, cur);
1643         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1644 }
1645
1646 /*
1647  * Called only from pf_purge_thread(), thus serialized.
1648  */
1649 static u_int
1650 pf_purge_expired_states(u_int i, int maxcheck)
1651 {
1652         struct pf_idhash *ih;
1653         struct pf_state *s;
1654
1655         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1656
1657         /*
1658          * Go through hash and unlink states that expire now.
1659          */
1660         while (maxcheck > 0) {
1661
1662                 ih = &V_pf_idhash[i];
1663 relock:
1664                 PF_HASHROW_LOCK(ih);
1665                 LIST_FOREACH(s, &ih->states, entry) {
1666                         if (pf_state_expires(s) <= time_uptime) {
1667                                 V_pf_status.states -=
1668                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1669                                 goto relock;
1670                         }
1671                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1672                         if (s->nat_rule.ptr != NULL)
1673                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1674                         if (s->anchor.ptr != NULL)
1675                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1676                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1677                         if (s->rt_kif)
1678                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1679                 }
1680                 PF_HASHROW_UNLOCK(ih);
1681
1682                 /* Return when we hit end of hash. */
1683                 if (++i > pf_hashmask) {
1684                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1685                         return (0);
1686                 }
1687
1688                 maxcheck--;
1689         }
1690
1691         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1692
1693         return (i);
1694 }
1695
1696 static void
1697 pf_purge_unlinked_rules()
1698 {
1699         struct pf_rulequeue tmpq;
1700         struct pf_rule *r, *r1;
1701
1702         /*
1703          * If we have overloading task pending, then we'd
1704          * better skip purging this time. There is a tiny
1705          * probability that overloading task references
1706          * an already unlinked rule.
1707          */
1708         PF_OVERLOADQ_LOCK();
1709         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1710                 PF_OVERLOADQ_UNLOCK();
1711                 return;
1712         }
1713         PF_OVERLOADQ_UNLOCK();
1714
1715         /*
1716          * Do naive mark-and-sweep garbage collecting of old rules.
1717          * Reference flag is raised by pf_purge_expired_states()
1718          * and pf_purge_expired_src_nodes().
1719          *
1720          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1721          * use a temporary queue.
1722          */
1723         TAILQ_INIT(&tmpq);
1724         PF_UNLNKDRULES_LOCK();
1725         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1726                 if (!(r->rule_flag & PFRULE_REFS)) {
1727                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1728                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1729                 } else
1730                         r->rule_flag &= ~PFRULE_REFS;
1731         }
1732         PF_UNLNKDRULES_UNLOCK();
1733
1734         if (!TAILQ_EMPTY(&tmpq)) {
1735                 PF_RULES_WLOCK();
1736                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1737                         TAILQ_REMOVE(&tmpq, r, entries);
1738                         pf_free_rule(r);
1739                 }
1740                 PF_RULES_WUNLOCK();
1741         }
1742 }
1743
1744 void
1745 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1746 {
1747         switch (af) {
1748 #ifdef INET
1749         case AF_INET: {
1750                 u_int32_t a = ntohl(addr->addr32[0]);
1751                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1752                     (a>>8)&255, a&255);
1753                 if (p) {
1754                         p = ntohs(p);
1755                         printf(":%u", p);
1756                 }
1757                 break;
1758         }
1759 #endif /* INET */
1760 #ifdef INET6
1761         case AF_INET6: {
1762                 u_int16_t b;
1763                 u_int8_t i, curstart, curend, maxstart, maxend;
1764                 curstart = curend = maxstart = maxend = 255;
1765                 for (i = 0; i < 8; i++) {
1766                         if (!addr->addr16[i]) {
1767                                 if (curstart == 255)
1768                                         curstart = i;
1769                                 curend = i;
1770                         } else {
1771                                 if ((curend - curstart) >
1772                                     (maxend - maxstart)) {
1773                                         maxstart = curstart;
1774                                         maxend = curend;
1775                                 }
1776                                 curstart = curend = 255;
1777                         }
1778                 }
1779                 if ((curend - curstart) >
1780                     (maxend - maxstart)) {
1781                         maxstart = curstart;
1782                         maxend = curend;
1783                 }
1784                 for (i = 0; i < 8; i++) {
1785                         if (i >= maxstart && i <= maxend) {
1786                                 if (i == 0)
1787                                         printf(":");
1788                                 if (i == maxend)
1789                                         printf(":");
1790                         } else {
1791                                 b = ntohs(addr->addr16[i]);
1792                                 printf("%x", b);
1793                                 if (i < 7)
1794                                         printf(":");
1795                         }
1796                 }
1797                 if (p) {
1798                         p = ntohs(p);
1799                         printf("[%u]", p);
1800                 }
1801                 break;
1802         }
1803 #endif /* INET6 */
1804         }
1805 }
1806
1807 void
1808 pf_print_state(struct pf_state *s)
1809 {
1810         pf_print_state_parts(s, NULL, NULL);
1811 }
1812
1813 static void
1814 pf_print_state_parts(struct pf_state *s,
1815     struct pf_state_key *skwp, struct pf_state_key *sksp)
1816 {
1817         struct pf_state_key *skw, *sks;
1818         u_int8_t proto, dir;
1819
1820         /* Do our best to fill these, but they're skipped if NULL */
1821         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1822         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1823         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1824         dir = s ? s->direction : 0;
1825
1826         switch (proto) {
1827         case IPPROTO_IPV4:
1828                 printf("IPv4");
1829                 break;
1830         case IPPROTO_IPV6:
1831                 printf("IPv6");
1832                 break;
1833         case IPPROTO_TCP:
1834                 printf("TCP");
1835                 break;
1836         case IPPROTO_UDP:
1837                 printf("UDP");
1838                 break;
1839         case IPPROTO_ICMP:
1840                 printf("ICMP");
1841                 break;
1842         case IPPROTO_ICMPV6:
1843                 printf("ICMPv6");
1844                 break;
1845         default:
1846                 printf("%u", skw->proto);
1847                 break;
1848         }
1849         switch (dir) {
1850         case PF_IN:
1851                 printf(" in");
1852                 break;
1853         case PF_OUT:
1854                 printf(" out");
1855                 break;
1856         }
1857         if (skw) {
1858                 printf(" wire: ");
1859                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1860                 printf(" ");
1861                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1862         }
1863         if (sks) {
1864                 printf(" stack: ");
1865                 if (sks != skw) {
1866                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1867                         printf(" ");
1868                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1869                 } else
1870                         printf("-");
1871         }
1872         if (s) {
1873                 if (proto == IPPROTO_TCP) {
1874                         printf(" [lo=%u high=%u win=%u modulator=%u",
1875                             s->src.seqlo, s->src.seqhi,
1876                             s->src.max_win, s->src.seqdiff);
1877                         if (s->src.wscale && s->dst.wscale)
1878                                 printf(" wscale=%u",
1879                                     s->src.wscale & PF_WSCALE_MASK);
1880                         printf("]");
1881                         printf(" [lo=%u high=%u win=%u modulator=%u",
1882                             s->dst.seqlo, s->dst.seqhi,
1883                             s->dst.max_win, s->dst.seqdiff);
1884                         if (s->src.wscale && s->dst.wscale)
1885                                 printf(" wscale=%u",
1886                                 s->dst.wscale & PF_WSCALE_MASK);
1887                         printf("]");
1888                 }
1889                 printf(" %u:%u", s->src.state, s->dst.state);
1890         }
1891 }
1892
1893 void
1894 pf_print_flags(u_int8_t f)
1895 {
1896         if (f)
1897                 printf(" ");
1898         if (f & TH_FIN)
1899                 printf("F");
1900         if (f & TH_SYN)
1901                 printf("S");
1902         if (f & TH_RST)
1903                 printf("R");
1904         if (f & TH_PUSH)
1905                 printf("P");
1906         if (f & TH_ACK)
1907                 printf("A");
1908         if (f & TH_URG)
1909                 printf("U");
1910         if (f & TH_ECE)
1911                 printf("E");
1912         if (f & TH_CWR)
1913                 printf("W");
1914 }
1915
1916 #define PF_SET_SKIP_STEPS(i)                                    \
1917         do {                                                    \
1918                 while (head[i] != cur) {                        \
1919                         head[i]->skip[i].ptr = cur;             \
1920                         head[i] = TAILQ_NEXT(head[i], entries); \
1921                 }                                               \
1922         } while (0)
1923
1924 void
1925 pf_calc_skip_steps(struct pf_rulequeue *rules)
1926 {
1927         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1928         int i;
1929
1930         cur = TAILQ_FIRST(rules);
1931         prev = cur;
1932         for (i = 0; i < PF_SKIP_COUNT; ++i)
1933                 head[i] = cur;
1934         while (cur != NULL) {
1935
1936                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1937                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1938                 if (cur->direction != prev->direction)
1939                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1940                 if (cur->af != prev->af)
1941                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1942                 if (cur->proto != prev->proto)
1943                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1944                 if (cur->src.neg != prev->src.neg ||
1945                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1946                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1947                 if (cur->src.port[0] != prev->src.port[0] ||
1948                     cur->src.port[1] != prev->src.port[1] ||
1949                     cur->src.port_op != prev->src.port_op)
1950                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1951                 if (cur->dst.neg != prev->dst.neg ||
1952                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1953                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1954                 if (cur->dst.port[0] != prev->dst.port[0] ||
1955                     cur->dst.port[1] != prev->dst.port[1] ||
1956                     cur->dst.port_op != prev->dst.port_op)
1957                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1958
1959                 prev = cur;
1960                 cur = TAILQ_NEXT(cur, entries);
1961         }
1962         for (i = 0; i < PF_SKIP_COUNT; ++i)
1963                 PF_SET_SKIP_STEPS(i);
1964 }
1965
1966 static int
1967 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1968 {
1969         if (aw1->type != aw2->type)
1970                 return (1);
1971         switch (aw1->type) {
1972         case PF_ADDR_ADDRMASK:
1973         case PF_ADDR_RANGE:
1974                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1975                         return (1);
1976                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1977                         return (1);
1978                 return (0);
1979         case PF_ADDR_DYNIFTL:
1980                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1981         case PF_ADDR_NOROUTE:
1982         case PF_ADDR_URPFFAILED:
1983                 return (0);
1984         case PF_ADDR_TABLE:
1985                 return (aw1->p.tbl != aw2->p.tbl);
1986         default:
1987                 printf("invalid address type: %d\n", aw1->type);
1988                 return (1);
1989         }
1990 }
1991
1992 u_int16_t
1993 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1994 {
1995         u_int32_t       l;
1996
1997         if (udp && !cksum)
1998                 return (0x0000);
1999         l = cksum + old - new;
2000         l = (l >> 16) + (l & 65535);
2001         l = l & 65535;
2002         if (udp && !l)
2003                 return (0xFFFF);
2004         return (l);
2005 }
2006
2007 static void
2008 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
2009     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
2010 {
2011         struct pf_addr  ao;
2012         u_int16_t       po = *p;
2013
2014         PF_ACPY(&ao, a, af);
2015         PF_ACPY(a, an, af);
2016
2017         *p = pn;
2018
2019         switch (af) {
2020 #ifdef INET
2021         case AF_INET:
2022                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2023                     ao.addr16[0], an->addr16[0], 0),
2024                     ao.addr16[1], an->addr16[1], 0);
2025                 *p = pn;
2026                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2027                     ao.addr16[0], an->addr16[0], u),
2028                     ao.addr16[1], an->addr16[1], u),
2029                     po, pn, u);
2030                 break;
2031 #endif /* INET */
2032 #ifdef INET6
2033         case AF_INET6:
2034                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2035                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2036                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2037                     ao.addr16[0], an->addr16[0], u),
2038                     ao.addr16[1], an->addr16[1], u),
2039                     ao.addr16[2], an->addr16[2], u),
2040                     ao.addr16[3], an->addr16[3], u),
2041                     ao.addr16[4], an->addr16[4], u),
2042                     ao.addr16[5], an->addr16[5], u),
2043                     ao.addr16[6], an->addr16[6], u),
2044                     ao.addr16[7], an->addr16[7], u),
2045                     po, pn, u);
2046                 break;
2047 #endif /* INET6 */
2048         }
2049 }
2050
2051
2052 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2053 void
2054 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2055 {
2056         u_int32_t       ao;
2057
2058         memcpy(&ao, a, sizeof(ao));
2059         memcpy(a, &an, sizeof(u_int32_t));
2060         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2061             ao % 65536, an % 65536, u);
2062 }
2063
2064 #ifdef INET6
2065 static void
2066 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2067 {
2068         struct pf_addr  ao;
2069
2070         PF_ACPY(&ao, a, AF_INET6);
2071         PF_ACPY(a, an, AF_INET6);
2072
2073         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2074             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2075             pf_cksum_fixup(pf_cksum_fixup(*c,
2076             ao.addr16[0], an->addr16[0], u),
2077             ao.addr16[1], an->addr16[1], u),
2078             ao.addr16[2], an->addr16[2], u),
2079             ao.addr16[3], an->addr16[3], u),
2080             ao.addr16[4], an->addr16[4], u),
2081             ao.addr16[5], an->addr16[5], u),
2082             ao.addr16[6], an->addr16[6], u),
2083             ao.addr16[7], an->addr16[7], u);
2084 }
2085 #endif /* INET6 */
2086
2087 static void
2088 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2089     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2090     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2091 {
2092         struct pf_addr  oia, ooa;
2093
2094         PF_ACPY(&oia, ia, af);
2095         if (oa)
2096                 PF_ACPY(&ooa, oa, af);
2097
2098         /* Change inner protocol port, fix inner protocol checksum. */
2099         if (ip != NULL) {
2100                 u_int16_t       oip = *ip;
2101                 u_int32_t       opc;
2102
2103                 if (pc != NULL)
2104                         opc = *pc;
2105                 *ip = np;
2106                 if (pc != NULL)
2107                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2108                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2109                 if (pc != NULL)
2110                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2111         }
2112         /* Change inner ip address, fix inner ip and icmp checksums. */
2113         PF_ACPY(ia, na, af);
2114         switch (af) {
2115 #ifdef INET
2116         case AF_INET: {
2117                 u_int32_t        oh2c = *h2c;
2118
2119                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2120                     oia.addr16[0], ia->addr16[0], 0),
2121                     oia.addr16[1], ia->addr16[1], 0);
2122                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2123                     oia.addr16[0], ia->addr16[0], 0),
2124                     oia.addr16[1], ia->addr16[1], 0);
2125                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2126                 break;
2127         }
2128 #endif /* INET */
2129 #ifdef INET6
2130         case AF_INET6:
2131                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2132                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2133                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2134                     oia.addr16[0], ia->addr16[0], u),
2135                     oia.addr16[1], ia->addr16[1], u),
2136                     oia.addr16[2], ia->addr16[2], u),
2137                     oia.addr16[3], ia->addr16[3], u),
2138                     oia.addr16[4], ia->addr16[4], u),
2139                     oia.addr16[5], ia->addr16[5], u),
2140                     oia.addr16[6], ia->addr16[6], u),
2141                     oia.addr16[7], ia->addr16[7], u);
2142                 break;
2143 #endif /* INET6 */
2144         }
2145         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2146         if (oa) {
2147                 PF_ACPY(oa, na, af);
2148                 switch (af) {
2149 #ifdef INET
2150                 case AF_INET:
2151                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2152                             ooa.addr16[0], oa->addr16[0], 0),
2153                             ooa.addr16[1], oa->addr16[1], 0);
2154                         break;
2155 #endif /* INET */
2156 #ifdef INET6
2157                 case AF_INET6:
2158                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2159                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2160                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2161                             ooa.addr16[0], oa->addr16[0], u),
2162                             ooa.addr16[1], oa->addr16[1], u),
2163                             ooa.addr16[2], oa->addr16[2], u),
2164                             ooa.addr16[3], oa->addr16[3], u),
2165                             ooa.addr16[4], oa->addr16[4], u),
2166                             ooa.addr16[5], oa->addr16[5], u),
2167                             ooa.addr16[6], oa->addr16[6], u),
2168                             ooa.addr16[7], oa->addr16[7], u);
2169                         break;
2170 #endif /* INET6 */
2171                 }
2172         }
2173 }
2174
2175
2176 /*
2177  * Need to modulate the sequence numbers in the TCP SACK option
2178  * (credits to Krzysztof Pfaff for report and patch)
2179  */
2180 static int
2181 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2182     struct tcphdr *th, struct pf_state_peer *dst)
2183 {
2184         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2185         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2186         int copyback = 0, i, olen;
2187         struct sackblk sack;
2188
2189 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2190         if (hlen < TCPOLEN_SACKLEN ||
2191             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2192                 return 0;
2193
2194         while (hlen >= TCPOLEN_SACKLEN) {
2195                 olen = opt[1];
2196                 switch (*opt) {
2197                 case TCPOPT_EOL:        /* FALLTHROUGH */
2198                 case TCPOPT_NOP:
2199                         opt++;
2200                         hlen--;
2201                         break;
2202                 case TCPOPT_SACK:
2203                         if (olen > hlen)
2204                                 olen = hlen;
2205                         if (olen >= TCPOLEN_SACKLEN) {
2206                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2207                                     i += TCPOLEN_SACK) {
2208                                         memcpy(&sack, &opt[i], sizeof(sack));
2209                                         pf_change_a(&sack.start, &th->th_sum,
2210                                             htonl(ntohl(sack.start) -
2211                                             dst->seqdiff), 0);
2212                                         pf_change_a(&sack.end, &th->th_sum,
2213                                             htonl(ntohl(sack.end) -
2214                                             dst->seqdiff), 0);
2215                                         memcpy(&opt[i], &sack, sizeof(sack));
2216                                 }
2217                                 copyback = 1;
2218                         }
2219                         /* FALLTHROUGH */
2220                 default:
2221                         if (olen < 2)
2222                                 olen = 2;
2223                         hlen -= olen;
2224                         opt += olen;
2225                 }
2226         }
2227
2228         if (copyback)
2229                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2230         return (copyback);
2231 }
2232
2233 static void
2234 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2235     const struct pf_addr *saddr, const struct pf_addr *daddr,
2236     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2237     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2238     u_int16_t rtag, struct ifnet *ifp)
2239 {
2240         struct pf_send_entry *pfse;
2241         struct mbuf     *m;
2242         int              len, tlen;
2243 #ifdef INET
2244         struct ip       *h = NULL;
2245 #endif /* INET */
2246 #ifdef INET6
2247         struct ip6_hdr  *h6 = NULL;
2248 #endif /* INET6 */
2249         struct tcphdr   *th;
2250         char            *opt;
2251         struct pf_mtag  *pf_mtag;
2252
2253         len = 0;
2254         th = NULL;
2255
2256         /* maximum segment size tcp option */
2257         tlen = sizeof(struct tcphdr);
2258         if (mss)
2259                 tlen += 4;
2260
2261         switch (af) {
2262 #ifdef INET
2263         case AF_INET:
2264                 len = sizeof(struct ip) + tlen;
2265                 break;
2266 #endif /* INET */
2267 #ifdef INET6
2268         case AF_INET6:
2269                 len = sizeof(struct ip6_hdr) + tlen;
2270                 break;
2271 #endif /* INET6 */
2272         default:
2273                 panic("%s: unsupported af %d", __func__, af);
2274         }
2275
2276         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2277         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2278         if (pfse == NULL)
2279                 return;
2280         m = m_gethdr(M_NOWAIT, MT_DATA);
2281         if (m == NULL) {
2282                 free(pfse, M_PFTEMP);
2283                 return;
2284         }
2285 #ifdef MAC
2286         mac_netinet_firewall_send(m);
2287 #endif
2288         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2289                 free(pfse, M_PFTEMP);
2290                 m_freem(m);
2291                 return;
2292         }
2293         if (tag)
2294                 m->m_flags |= M_SKIP_FIREWALL;
2295         pf_mtag->tag = rtag;
2296
2297         if (r != NULL && r->rtableid >= 0)
2298                 M_SETFIB(m, r->rtableid);
2299
2300 #ifdef ALTQ
2301         if (r != NULL && r->qid) {
2302                 pf_mtag->qid = r->qid;
2303
2304                 /* add hints for ecn */
2305                 pf_mtag->hdr = mtod(m, struct ip *);
2306         }
2307 #endif /* ALTQ */
2308         m->m_data += max_linkhdr;
2309         m->m_pkthdr.len = m->m_len = len;
2310         m->m_pkthdr.rcvif = NULL;
2311         bzero(m->m_data, len);
2312         switch (af) {
2313 #ifdef INET
2314         case AF_INET:
2315                 h = mtod(m, struct ip *);
2316
2317                 /* IP header fields included in the TCP checksum */
2318                 h->ip_p = IPPROTO_TCP;
2319                 h->ip_len = htons(tlen);
2320                 h->ip_src.s_addr = saddr->v4.s_addr;
2321                 h->ip_dst.s_addr = daddr->v4.s_addr;
2322
2323                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2324                 break;
2325 #endif /* INET */
2326 #ifdef INET6
2327         case AF_INET6:
2328                 h6 = mtod(m, struct ip6_hdr *);
2329
2330                 /* IP header fields included in the TCP checksum */
2331                 h6->ip6_nxt = IPPROTO_TCP;
2332                 h6->ip6_plen = htons(tlen);
2333                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2334                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2335
2336                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2337                 break;
2338 #endif /* INET6 */
2339         }
2340
2341         /* TCP header */
2342         th->th_sport = sport;
2343         th->th_dport = dport;
2344         th->th_seq = htonl(seq);
2345         th->th_ack = htonl(ack);
2346         th->th_off = tlen >> 2;
2347         th->th_flags = flags;
2348         th->th_win = htons(win);
2349
2350         if (mss) {
2351                 opt = (char *)(th + 1);
2352                 opt[0] = TCPOPT_MAXSEG;
2353                 opt[1] = 4;
2354                 HTONS(mss);
2355                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2356         }
2357
2358         switch (af) {
2359 #ifdef INET
2360         case AF_INET:
2361                 /* TCP checksum */
2362                 th->th_sum = in_cksum(m, len);
2363
2364                 /* Finish the IP header */
2365                 h->ip_v = 4;
2366                 h->ip_hl = sizeof(*h) >> 2;
2367                 h->ip_tos = IPTOS_LOWDELAY;
2368                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2369                 h->ip_len = htons(len);
2370                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2371                 h->ip_sum = 0;
2372
2373                 pfse->pfse_type = PFSE_IP;
2374                 break;
2375 #endif /* INET */
2376 #ifdef INET6
2377         case AF_INET6:
2378                 /* TCP checksum */
2379                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2380                     sizeof(struct ip6_hdr), tlen);
2381
2382                 h6->ip6_vfc |= IPV6_VERSION;
2383                 h6->ip6_hlim = IPV6_DEFHLIM;
2384
2385                 pfse->pfse_type = PFSE_IP6;
2386                 break;
2387 #endif /* INET6 */
2388         }
2389         pfse->pfse_m = m;
2390         pf_send(pfse);
2391 }
2392
2393 static void
2394 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2395     struct pf_rule *r)
2396 {
2397         struct pf_send_entry *pfse;
2398         struct mbuf *m0;
2399         struct pf_mtag *pf_mtag;
2400
2401         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2402         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2403         if (pfse == NULL)
2404                 return;
2405
2406         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2407                 free(pfse, M_PFTEMP);
2408                 return;
2409         }
2410
2411         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2412                 free(pfse, M_PFTEMP);
2413                 return;
2414         }
2415         /* XXX: revisit */
2416         m0->m_flags |= M_SKIP_FIREWALL;
2417
2418         if (r->rtableid >= 0)
2419                 M_SETFIB(m0, r->rtableid);
2420
2421 #ifdef ALTQ
2422         if (r->qid) {
2423                 pf_mtag->qid = r->qid;
2424                 /* add hints for ecn */
2425                 pf_mtag->hdr = mtod(m0, struct ip *);
2426         }
2427 #endif /* ALTQ */
2428
2429         switch (af) {
2430 #ifdef INET
2431         case AF_INET:
2432                 pfse->pfse_type = PFSE_ICMP;
2433                 break;
2434 #endif /* INET */
2435 #ifdef INET6
2436         case AF_INET6:
2437                 pfse->pfse_type = PFSE_ICMP6;
2438                 break;
2439 #endif /* INET6 */
2440         }
2441         pfse->pfse_m = m0;
2442         pfse->pfse_icmp_type = type;
2443         pfse->pfse_icmp_code = code;
2444         pf_send(pfse);
2445 }
2446
2447 /*
2448  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2449  * If n is 0, they match if they are equal. If n is != 0, they match if they
2450  * are different.
2451  */
2452 int
2453 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2454     struct pf_addr *b, sa_family_t af)
2455 {
2456         int     match = 0;
2457
2458         switch (af) {
2459 #ifdef INET
2460         case AF_INET:
2461                 if ((a->addr32[0] & m->addr32[0]) ==
2462                     (b->addr32[0] & m->addr32[0]))
2463                         match++;
2464                 break;
2465 #endif /* INET */
2466 #ifdef INET6
2467         case AF_INET6:
2468                 if (((a->addr32[0] & m->addr32[0]) ==
2469                      (b->addr32[0] & m->addr32[0])) &&
2470                     ((a->addr32[1] & m->addr32[1]) ==
2471                      (b->addr32[1] & m->addr32[1])) &&
2472                     ((a->addr32[2] & m->addr32[2]) ==
2473                      (b->addr32[2] & m->addr32[2])) &&
2474                     ((a->addr32[3] & m->addr32[3]) ==
2475                      (b->addr32[3] & m->addr32[3])))
2476                         match++;
2477                 break;
2478 #endif /* INET6 */
2479         }
2480         if (match) {
2481                 if (n)
2482                         return (0);
2483                 else
2484                         return (1);
2485         } else {
2486                 if (n)
2487                         return (1);
2488                 else
2489                         return (0);
2490         }
2491 }
2492
2493 /*
2494  * Return 1 if b <= a <= e, otherwise return 0.
2495  */
2496 int
2497 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2498     struct pf_addr *a, sa_family_t af)
2499 {
2500         switch (af) {
2501 #ifdef INET
2502         case AF_INET:
2503                 if ((a->addr32[0] < b->addr32[0]) ||
2504                     (a->addr32[0] > e->addr32[0]))
2505                         return (0);
2506                 break;
2507 #endif /* INET */
2508 #ifdef INET6
2509         case AF_INET6: {
2510                 int     i;
2511
2512                 /* check a >= b */
2513                 for (i = 0; i < 4; ++i)
2514                         if (a->addr32[i] > b->addr32[i])
2515                                 break;
2516                         else if (a->addr32[i] < b->addr32[i])
2517                                 return (0);
2518                 /* check a <= e */
2519                 for (i = 0; i < 4; ++i)
2520                         if (a->addr32[i] < e->addr32[i])
2521                                 break;
2522                         else if (a->addr32[i] > e->addr32[i])
2523                                 return (0);
2524                 break;
2525         }
2526 #endif /* INET6 */
2527         }
2528         return (1);
2529 }
2530
2531 static int
2532 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2533 {
2534         switch (op) {
2535         case PF_OP_IRG:
2536                 return ((p > a1) && (p < a2));
2537         case PF_OP_XRG:
2538                 return ((p < a1) || (p > a2));
2539         case PF_OP_RRG:
2540                 return ((p >= a1) && (p <= a2));
2541         case PF_OP_EQ:
2542                 return (p == a1);
2543         case PF_OP_NE:
2544                 return (p != a1);
2545         case PF_OP_LT:
2546                 return (p < a1);
2547         case PF_OP_LE:
2548                 return (p <= a1);
2549         case PF_OP_GT:
2550                 return (p > a1);
2551         case PF_OP_GE:
2552                 return (p >= a1);
2553         }
2554         return (0); /* never reached */
2555 }
2556
2557 int
2558 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2559 {
2560         NTOHS(a1);
2561         NTOHS(a2);
2562         NTOHS(p);
2563         return (pf_match(op, a1, a2, p));
2564 }
2565
2566 static int
2567 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2568 {
2569         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2570                 return (0);
2571         return (pf_match(op, a1, a2, u));
2572 }
2573
2574 static int
2575 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2576 {
2577         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2578                 return (0);
2579         return (pf_match(op, a1, a2, g));
2580 }
2581
2582 int
2583 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2584 {
2585         if (*tag == -1)
2586                 *tag = mtag;
2587
2588         return ((!r->match_tag_not && r->match_tag == *tag) ||
2589             (r->match_tag_not && r->match_tag != *tag));
2590 }
2591
2592 int
2593 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2594 {
2595
2596         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2597
2598         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2599                 return (ENOMEM);
2600
2601         pd->pf_mtag->tag = tag;
2602
2603         return (0);
2604 }
2605
2606 #define PF_ANCHOR_STACKSIZE     32
2607 struct pf_anchor_stackframe {
2608         struct pf_ruleset       *rs;
2609         struct pf_rule          *r;     /* XXX: + match bit */
2610         struct pf_anchor        *child;
2611 };
2612
2613 /*
2614  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2615  */
2616 #define PF_ANCHORSTACK_MATCH    0x00000001
2617 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2618
2619 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2620 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2621                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2622 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2623                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2624 } while (0)
2625
2626 void
2627 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2628     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2629     int *match)
2630 {
2631         struct pf_anchor_stackframe     *f;
2632
2633         PF_RULES_RASSERT();
2634
2635         if (match)
2636                 *match = 0;
2637         if (*depth >= PF_ANCHOR_STACKSIZE) {
2638                 printf("%s: anchor stack overflow on %s\n",
2639                     __func__, (*r)->anchor->name);
2640                 *r = TAILQ_NEXT(*r, entries);
2641                 return;
2642         } else if (*depth == 0 && a != NULL)
2643                 *a = *r;
2644         f = stack + (*depth)++;
2645         f->rs = *rs;
2646         f->r = *r;
2647         if ((*r)->anchor_wildcard) {
2648                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2649
2650                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2651                         *r = NULL;
2652                         return;
2653                 }
2654                 *rs = &f->child->ruleset;
2655         } else {
2656                 f->child = NULL;
2657                 *rs = &(*r)->anchor->ruleset;
2658         }
2659         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2660 }
2661
2662 int
2663 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2664     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2665     int *match)
2666 {
2667         struct pf_anchor_stackframe     *f;
2668         struct pf_rule *fr;
2669         int quick = 0;
2670
2671         PF_RULES_RASSERT();
2672
2673         do {
2674                 if (*depth <= 0)
2675                         break;
2676                 f = stack + *depth - 1;
2677                 fr = PF_ANCHOR_RULE(f);
2678                 if (f->child != NULL) {
2679                         struct pf_anchor_node *parent;
2680
2681                         /*
2682                          * This block traverses through
2683                          * a wildcard anchor.
2684                          */
2685                         parent = &fr->anchor->children;
2686                         if (match != NULL && *match) {
2687                                 /*
2688                                  * If any of "*" matched, then
2689                                  * "foo/ *" matched, mark frame
2690                                  * appropriately.
2691                                  */
2692                                 PF_ANCHOR_SET_MATCH(f);
2693                                 *match = 0;
2694                         }
2695                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2696                         if (f->child != NULL) {
2697                                 *rs = &f->child->ruleset;
2698                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2699                                 if (*r == NULL)
2700                                         continue;
2701                                 else
2702                                         break;
2703                         }
2704                 }
2705                 (*depth)--;
2706                 if (*depth == 0 && a != NULL)
2707                         *a = NULL;
2708                 *rs = f->rs;
2709                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2710                         quick = fr->quick;
2711                 *r = TAILQ_NEXT(fr, entries);
2712         } while (*r == NULL);
2713
2714         return (quick);
2715 }
2716
2717 #ifdef INET6
2718 void
2719 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2720     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2721 {
2722         switch (af) {
2723 #ifdef INET
2724         case AF_INET:
2725                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2726                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2727                 break;
2728 #endif /* INET */
2729         case AF_INET6:
2730                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2731                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2732                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2733                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2734                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2735                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2736                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2737                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2738                 break;
2739         }
2740 }
2741
2742 void
2743 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2744 {
2745         switch (af) {
2746 #ifdef INET
2747         case AF_INET:
2748                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2749                 break;
2750 #endif /* INET */
2751         case AF_INET6:
2752                 if (addr->addr32[3] == 0xffffffff) {
2753                         addr->addr32[3] = 0;
2754                         if (addr->addr32[2] == 0xffffffff) {
2755                                 addr->addr32[2] = 0;
2756                                 if (addr->addr32[1] == 0xffffffff) {
2757                                         addr->addr32[1] = 0;
2758                                         addr->addr32[0] =
2759                                             htonl(ntohl(addr->addr32[0]) + 1);
2760                                 } else
2761                                         addr->addr32[1] =
2762                                             htonl(ntohl(addr->addr32[1]) + 1);
2763                         } else
2764                                 addr->addr32[2] =
2765                                     htonl(ntohl(addr->addr32[2]) + 1);
2766                 } else
2767                         addr->addr32[3] =
2768                             htonl(ntohl(addr->addr32[3]) + 1);
2769                 break;
2770         }
2771 }
2772 #endif /* INET6 */
2773
2774 int
2775 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2776 {
2777         struct pf_addr          *saddr, *daddr;
2778         u_int16_t                sport, dport;
2779         struct inpcbinfo        *pi;
2780         struct inpcb            *inp;
2781
2782         pd->lookup.uid = UID_MAX;
2783         pd->lookup.gid = GID_MAX;
2784
2785         switch (pd->proto) {
2786         case IPPROTO_TCP:
2787                 if (pd->hdr.tcp == NULL)
2788                         return (-1);
2789                 sport = pd->hdr.tcp->th_sport;
2790                 dport = pd->hdr.tcp->th_dport;
2791                 pi = &V_tcbinfo;
2792                 break;
2793         case IPPROTO_UDP:
2794                 if (pd->hdr.udp == NULL)
2795                         return (-1);
2796                 sport = pd->hdr.udp->uh_sport;
2797                 dport = pd->hdr.udp->uh_dport;
2798                 pi = &V_udbinfo;
2799                 break;
2800         default:
2801                 return (-1);
2802         }
2803         if (direction == PF_IN) {
2804                 saddr = pd->src;
2805                 daddr = pd->dst;
2806         } else {
2807                 u_int16_t       p;
2808
2809                 p = sport;
2810                 sport = dport;
2811                 dport = p;
2812                 saddr = pd->dst;
2813                 daddr = pd->src;
2814         }
2815         switch (pd->af) {
2816 #ifdef INET
2817         case AF_INET:
2818                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2819                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2820                 if (inp == NULL) {
2821                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2822                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2823                            INPLOOKUP_RLOCKPCB, NULL, m);
2824                         if (inp == NULL)
2825                                 return (-1);
2826                 }
2827                 break;
2828 #endif /* INET */
2829 #ifdef INET6
2830         case AF_INET6:
2831                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2832                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2833                 if (inp == NULL) {
2834                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2835                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2836                             INPLOOKUP_RLOCKPCB, NULL, m);
2837                         if (inp == NULL)
2838                                 return (-1);
2839                 }
2840                 break;
2841 #endif /* INET6 */
2842
2843         default:
2844                 return (-1);
2845         }
2846         INP_RLOCK_ASSERT(inp);
2847         pd->lookup.uid = inp->inp_cred->cr_uid;
2848         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2849         INP_RUNLOCK(inp);
2850
2851         return (1);
2852 }
2853
2854 static u_int8_t
2855 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2856 {
2857         int              hlen;
2858         u_int8_t         hdr[60];
2859         u_int8_t        *opt, optlen;
2860         u_int8_t         wscale = 0;
2861
2862         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2863         if (hlen <= sizeof(struct tcphdr))
2864                 return (0);
2865         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2866                 return (0);
2867         opt = hdr + sizeof(struct tcphdr);
2868         hlen -= sizeof(struct tcphdr);
2869         while (hlen >= 3) {
2870                 switch (*opt) {
2871                 case TCPOPT_EOL:
2872                 case TCPOPT_NOP:
2873                         ++opt;
2874                         --hlen;
2875                         break;
2876                 case TCPOPT_WINDOW:
2877                         wscale = opt[2];
2878                         if (wscale > TCP_MAX_WINSHIFT)
2879                                 wscale = TCP_MAX_WINSHIFT;
2880                         wscale |= PF_WSCALE_FLAG;
2881                         /* FALLTHROUGH */
2882                 default:
2883                         optlen = opt[1];
2884                         if (optlen < 2)
2885                                 optlen = 2;
2886                         hlen -= optlen;
2887                         opt += optlen;
2888                         break;
2889                 }
2890         }
2891         return (wscale);
2892 }
2893
2894 static u_int16_t
2895 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2896 {
2897         int              hlen;
2898         u_int8_t         hdr[60];
2899         u_int8_t        *opt, optlen;
2900         u_int16_t        mss = V_tcp_mssdflt;
2901
2902         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2903         if (hlen <= sizeof(struct tcphdr))
2904                 return (0);
2905         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2906                 return (0);
2907         opt = hdr + sizeof(struct tcphdr);
2908         hlen -= sizeof(struct tcphdr);
2909         while (hlen >= TCPOLEN_MAXSEG) {
2910                 switch (*opt) {
2911                 case TCPOPT_EOL:
2912                 case TCPOPT_NOP:
2913                         ++opt;
2914                         --hlen;
2915                         break;
2916                 case TCPOPT_MAXSEG:
2917                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2918                         NTOHS(mss);
2919                         /* FALLTHROUGH */
2920                 default:
2921                         optlen = opt[1];
2922                         if (optlen < 2)
2923                                 optlen = 2;
2924                         hlen -= optlen;
2925                         opt += optlen;
2926                         break;
2927                 }
2928         }
2929         return (mss);
2930 }
2931
2932 static u_int16_t
2933 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2934 {
2935 #ifdef INET
2936         struct sockaddr_in      *dst;
2937         struct route             ro;
2938 #endif /* INET */
2939 #ifdef INET6
2940         struct sockaddr_in6     *dst6;
2941         struct route_in6         ro6;
2942 #endif /* INET6 */
2943         struct rtentry          *rt = NULL;
2944         int                      hlen = 0;
2945         u_int16_t                mss = V_tcp_mssdflt;
2946
2947         switch (af) {
2948 #ifdef INET
2949         case AF_INET:
2950                 hlen = sizeof(struct ip);
2951                 bzero(&ro, sizeof(ro));
2952                 dst = (struct sockaddr_in *)&ro.ro_dst;
2953                 dst->sin_family = AF_INET;
2954                 dst->sin_len = sizeof(*dst);
2955                 dst->sin_addr = addr->v4;
2956                 in_rtalloc_ign(&ro, 0, rtableid);
2957                 rt = ro.ro_rt;
2958                 break;
2959 #endif /* INET */
2960 #ifdef INET6
2961         case AF_INET6:
2962                 hlen = sizeof(struct ip6_hdr);
2963                 bzero(&ro6, sizeof(ro6));
2964                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2965                 dst6->sin6_family = AF_INET6;
2966                 dst6->sin6_len = sizeof(*dst6);
2967                 dst6->sin6_addr = addr->v6;
2968                 in6_rtalloc_ign(&ro6, 0, rtableid);
2969                 rt = ro6.ro_rt;
2970                 break;
2971 #endif /* INET6 */
2972         }
2973
2974         if (rt && rt->rt_ifp) {
2975                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2976                 mss = max(V_tcp_mssdflt, mss);
2977                 RTFREE(rt);
2978         }
2979         mss = min(mss, offer);
2980         mss = max(mss, 64);             /* sanity - at least max opt space */
2981         return (mss);
2982 }
2983
2984 static u_int32_t
2985 pf_tcp_iss(struct pf_pdesc *pd)
2986 {
2987         MD5_CTX ctx;
2988         u_int32_t digest[4];
2989
2990         if (V_pf_tcp_secret_init == 0) {
2991                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2992                 MD5Init(&V_pf_tcp_secret_ctx);
2993                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2994                     sizeof(V_pf_tcp_secret));
2995                 V_pf_tcp_secret_init = 1;
2996         }
2997
2998         ctx = V_pf_tcp_secret_ctx;
2999
3000         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3001         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3002         if (pd->af == AF_INET6) {
3003                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3004                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3005         } else {
3006                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3007                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3008         }
3009         MD5Final((u_char *)digest, &ctx);
3010         V_pf_tcp_iss_off += 4096;
3011 #define ISN_RANDOM_INCREMENT (4096 - 1)
3012         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3013             V_pf_tcp_iss_off);
3014 #undef  ISN_RANDOM_INCREMENT
3015 }
3016
3017 static int
3018 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3019     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3020     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3021 {
3022         struct pf_rule          *nr = NULL;
3023         struct pf_addr          * const saddr = pd->src;
3024         struct pf_addr          * const daddr = pd->dst;
3025         sa_family_t              af = pd->af;
3026         struct pf_rule          *r, *a = NULL;
3027         struct pf_ruleset       *ruleset = NULL;
3028         struct pf_src_node      *nsn = NULL;
3029         struct tcphdr           *th = pd->hdr.tcp;
3030         struct pf_state_key     *sk = NULL, *nk = NULL;
3031         u_short                  reason;
3032         int                      rewrite = 0, hdrlen = 0;
3033         int                      tag = -1, rtableid = -1;
3034         int                      asd = 0;
3035         int                      match = 0;
3036         int                      state_icmp = 0;
3037         u_int16_t                sport = 0, dport = 0;
3038         u_int16_t                bproto_sum = 0, bip_sum = 0;
3039         u_int8_t                 icmptype = 0, icmpcode = 0;
3040         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3041
3042         PF_RULES_RASSERT();
3043
3044         if (inp != NULL) {
3045                 INP_LOCK_ASSERT(inp);
3046                 pd->lookup.uid = inp->inp_cred->cr_uid;
3047                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3048                 pd->lookup.done = 1;
3049         }
3050
3051         switch (pd->proto) {
3052         case IPPROTO_TCP:
3053                 sport = th->th_sport;
3054                 dport = th->th_dport;
3055                 hdrlen = sizeof(*th);
3056                 break;
3057         case IPPROTO_UDP:
3058                 sport = pd->hdr.udp->uh_sport;
3059                 dport = pd->hdr.udp->uh_dport;
3060                 hdrlen = sizeof(*pd->hdr.udp);
3061                 break;
3062 #ifdef INET
3063         case IPPROTO_ICMP:
3064                 if (pd->af != AF_INET)
3065                         break;
3066                 sport = dport = pd->hdr.icmp->icmp_id;
3067                 hdrlen = sizeof(*pd->hdr.icmp);
3068                 icmptype = pd->hdr.icmp->icmp_type;
3069                 icmpcode = pd->hdr.icmp->icmp_code;
3070
3071                 if (icmptype == ICMP_UNREACH ||
3072                     icmptype == ICMP_SOURCEQUENCH ||
3073                     icmptype == ICMP_REDIRECT ||
3074                     icmptype == ICMP_TIMXCEED ||
3075                     icmptype == ICMP_PARAMPROB)
3076                         state_icmp++;
3077                 break;
3078 #endif /* INET */
3079 #ifdef INET6
3080         case IPPROTO_ICMPV6:
3081                 if (af != AF_INET6)
3082                         break;
3083                 sport = dport = pd->hdr.icmp6->icmp6_id;
3084                 hdrlen = sizeof(*pd->hdr.icmp6);
3085                 icmptype = pd->hdr.icmp6->icmp6_type;
3086                 icmpcode = pd->hdr.icmp6->icmp6_code;
3087
3088                 if (icmptype == ICMP6_DST_UNREACH ||
3089                     icmptype == ICMP6_PACKET_TOO_BIG ||
3090                     icmptype == ICMP6_TIME_EXCEEDED ||
3091                     icmptype == ICMP6_PARAM_PROB)
3092                         state_icmp++;
3093                 break;
3094 #endif /* INET6 */
3095         default:
3096                 sport = dport = hdrlen = 0;
3097                 break;
3098         }
3099
3100         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3101
3102         /* check packet for BINAT/NAT/RDR */
3103         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3104             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3105                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3106                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3107
3108                 if (pd->ip_sum)
3109                         bip_sum = *pd->ip_sum;
3110
3111                 switch (pd->proto) {
3112                 case IPPROTO_TCP:
3113                         bproto_sum = th->th_sum;
3114                         pd->proto_sum = &th->th_sum;
3115
3116                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3117                             nk->port[pd->sidx] != sport) {
3118                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3119                                     &th->th_sum, &nk->addr[pd->sidx],
3120                                     nk->port[pd->sidx], 0, af);
3121                                 pd->sport = &th->th_sport;
3122                                 sport = th->th_sport;
3123                         }
3124
3125                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3126                             nk->port[pd->didx] != dport) {
3127                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3128                                     &th->th_sum, &nk->addr[pd->didx],
3129                                     nk->port[pd->didx], 0, af);
3130                                 dport = th->th_dport;
3131                                 pd->dport = &th->th_dport;
3132                         }
3133                         rewrite++;
3134                         break;
3135                 case IPPROTO_UDP:
3136                         bproto_sum = pd->hdr.udp->uh_sum;
3137                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3138
3139                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3140                             nk->port[pd->sidx] != sport) {
3141                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3142                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3143                                     &nk->addr[pd->sidx],
3144                                     nk->port[pd->sidx], 1, af);
3145                                 sport = pd->hdr.udp->uh_sport;
3146                                 pd->sport = &pd->hdr.udp->uh_sport;
3147                         }
3148
3149                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3150                             nk->port[pd->didx] != dport) {
3151                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3152                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3153                                     &nk->addr[pd->didx],
3154                                     nk->port[pd->didx], 1, af);
3155                                 dport = pd->hdr.udp->uh_dport;
3156                                 pd->dport = &pd->hdr.udp->uh_dport;
3157                         }
3158                         rewrite++;
3159                         break;
3160 #ifdef INET
3161                 case IPPROTO_ICMP:
3162                         nk->port[0] = nk->port[1];
3163                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3164                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3165                                     nk->addr[pd->sidx].v4.s_addr, 0);
3166
3167                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3168                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3169                                     nk->addr[pd->didx].v4.s_addr, 0);
3170
3171                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3172                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3173                                     pd->hdr.icmp->icmp_cksum, sport,
3174                                     nk->port[1], 0);
3175                                 pd->hdr.icmp->icmp_id = nk->port[1];
3176                                 pd->sport = &pd->hdr.icmp->icmp_id;
3177                         }
3178                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3179                         break;
3180 #endif /* INET */
3181 #ifdef INET6
3182                 case IPPROTO_ICMPV6:
3183                         nk->port[0] = nk->port[1];
3184                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3185                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3186                                     &nk->addr[pd->sidx], 0);
3187
3188                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3189                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3190                                     &nk->addr[pd->didx], 0);
3191                         rewrite++;
3192                         break;
3193 #endif /* INET */
3194                 default:
3195                         switch (af) {
3196 #ifdef INET
3197                         case AF_INET:
3198                                 if (PF_ANEQ(saddr,
3199                                     &nk->addr[pd->sidx], AF_INET))
3200                                         pf_change_a(&saddr->v4.s_addr,
3201                                             pd->ip_sum,
3202                                             nk->addr[pd->sidx].v4.s_addr, 0);
3203
3204                                 if (PF_ANEQ(daddr,
3205                                     &nk->addr[pd->didx], AF_INET))
3206                                         pf_change_a(&daddr->v4.s_addr,
3207                                             pd->ip_sum,
3208                                             nk->addr[pd->didx].v4.s_addr, 0);
3209                                 break;
3210 #endif /* INET */
3211 #ifdef INET6
3212                         case AF_INET6:
3213                                 if (PF_ANEQ(saddr,
3214                                     &nk->addr[pd->sidx], AF_INET6))
3215                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3216
3217                                 if (PF_ANEQ(daddr,
3218                                     &nk->addr[pd->didx], AF_INET6))
3219                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3220                                 break;
3221 #endif /* INET */
3222                         }
3223                         break;
3224                 }
3225                 if (nr->natpass)
3226                         r = NULL;
3227                 pd->nat_rule = nr;
3228         }
3229
3230         while (r != NULL) {
3231                 r->evaluations++;
3232                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3233                         r = r->skip[PF_SKIP_IFP].ptr;
3234                 else if (r->direction && r->direction != direction)
3235                         r = r->skip[PF_SKIP_DIR].ptr;
3236                 else if (r->af && r->af != af)
3237                         r = r->skip[PF_SKIP_AF].ptr;
3238                 else if (r->proto && r->proto != pd->proto)
3239                         r = r->skip[PF_SKIP_PROTO].ptr;
3240                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3241                     r->src.neg, kif, M_GETFIB(m)))
3242                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3243                 /* tcp/udp only. port_op always 0 in other cases */
3244                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3245                     r->src.port[0], r->src.port[1], sport))
3246                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3247                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3248                     r->dst.neg, NULL, M_GETFIB(m)))
3249                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3250                 /* tcp/udp only. port_op always 0 in other cases */
3251                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3252                     r->dst.port[0], r->dst.port[1], dport))
3253                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3254                 /* icmp only. type always 0 in other cases */
3255                 else if (r->type && r->type != icmptype + 1)
3256                         r = TAILQ_NEXT(r, entries);
3257                 /* icmp only. type always 0 in other cases */
3258                 else if (r->code && r->code != icmpcode + 1)
3259                         r = TAILQ_NEXT(r, entries);
3260                 else if (r->tos && !(r->tos == pd->tos))
3261                         r = TAILQ_NEXT(r, entries);
3262                 else if (r->rule_flag & PFRULE_FRAGMENT)
3263                         r = TAILQ_NEXT(r, entries);
3264                 else if (pd->proto == IPPROTO_TCP &&
3265                     (r->flagset & th->th_flags) != r->flags)
3266                         r = TAILQ_NEXT(r, entries);
3267                 /* tcp/udp only. uid.op always 0 in other cases */
3268                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3269                     pf_socket_lookup(direction, pd, m), 1)) &&
3270                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3271                     pd->lookup.uid))
3272                         r = TAILQ_NEXT(r, entries);
3273                 /* tcp/udp only. gid.op always 0 in other cases */
3274                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3275                     pf_socket_lookup(direction, pd, m), 1)) &&
3276                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3277                     pd->lookup.gid))
3278                         r = TAILQ_NEXT(r, entries);
3279                 else if (r->prob &&
3280                     r->prob <= arc4random())
3281                         r = TAILQ_NEXT(r, entries);
3282                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3283                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3284                         r = TAILQ_NEXT(r, entries);
3285                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3286                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3287                     pf_osfp_fingerprint(pd, m, off, th),
3288                     r->os_fingerprint)))
3289                         r = TAILQ_NEXT(r, entries);
3290                 else {
3291                         if (r->tag)
3292                                 tag = r->tag;
3293                         if (r->rtableid >= 0)
3294                                 rtableid = r->rtableid;
3295                         if (r->anchor == NULL) {
3296                                 match = 1;
3297                                 *rm = r;
3298                                 *am = a;
3299                                 *rsm = ruleset;
3300                                 if ((*rm)->quick)
3301                                         break;
3302                                 r = TAILQ_NEXT(r, entries);
3303                         } else
3304                                 pf_step_into_anchor(anchor_stack, &asd,
3305                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3306                                     &match);
3307                 }
3308                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3309                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3310                         break;
3311         }
3312         r = *rm;
3313         a = *am;
3314         ruleset = *rsm;
3315
3316         REASON_SET(&reason, PFRES_MATCH);
3317
3318         if (r->log || (nr != NULL && nr->log)) {
3319                 if (rewrite)
3320                         m_copyback(m, off, hdrlen, pd->hdr.any);
3321                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3322                     ruleset, pd, 1);
3323         }
3324
3325         if ((r->action == PF_DROP) &&
3326             ((r->rule_flag & PFRULE_RETURNRST) ||
3327             (r->rule_flag & PFRULE_RETURNICMP) ||
3328             (r->rule_flag & PFRULE_RETURN))) {
3329                 /* undo NAT changes, if they have taken place */
3330                 if (nr != NULL) {
3331                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3332                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3333                         if (pd->sport)
3334                                 *pd->sport = sk->port[pd->sidx];
3335                         if (pd->dport)
3336                                 *pd->dport = sk->port[pd->didx];
3337                         if (pd->proto_sum)
3338                                 *pd->proto_sum = bproto_sum;
3339                         if (pd->ip_sum)
3340                                 *pd->ip_sum = bip_sum;
3341                         m_copyback(m, off, hdrlen, pd->hdr.any);
3342                 }
3343                 if (pd->proto == IPPROTO_TCP &&
3344                     ((r->rule_flag & PFRULE_RETURNRST) ||
3345                     (r->rule_flag & PFRULE_RETURN)) &&
3346                     !(th->th_flags & TH_RST)) {
3347                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3348                         int              len = 0;
3349 #ifdef INET
3350                         struct ip       *h4;
3351 #endif
3352 #ifdef INET6
3353                         struct ip6_hdr  *h6;
3354 #endif
3355
3356                         switch (af) {
3357 #ifdef INET
3358                         case AF_INET:
3359                                 h4 = mtod(m, struct ip *);
3360                                 len = ntohs(h4->ip_len) - off;
3361                                 break;
3362 #endif
3363 #ifdef INET6
3364                         case AF_INET6:
3365                                 h6 = mtod(m, struct ip6_hdr *);
3366                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3367                                 break;
3368 #endif
3369                         }
3370
3371                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3372                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3373                         else {
3374                                 if (th->th_flags & TH_SYN)
3375                                         ack++;
3376                                 if (th->th_flags & TH_FIN)
3377                                         ack++;
3378                                 pf_send_tcp(m, r, af, pd->dst,
3379                                     pd->src, th->th_dport, th->th_sport,
3380                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3381                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3382                         }
3383                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3384                     r->return_icmp)
3385                         pf_send_icmp(m, r->return_icmp >> 8,
3386                             r->return_icmp & 255, af, r);
3387                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3388                     r->return_icmp6)
3389                         pf_send_icmp(m, r->return_icmp6 >> 8,
3390                             r->return_icmp6 & 255, af, r);
3391         }
3392
3393         if (r->action == PF_DROP)
3394                 goto cleanup;
3395
3396         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3397                 REASON_SET(&reason, PFRES_MEMORY);
3398                 goto cleanup;
3399         }
3400         if (rtableid >= 0)
3401                 M_SETFIB(m, rtableid);
3402
3403         if (!state_icmp && (r->keep_state || nr != NULL ||
3404             (pd->flags & PFDESC_TCP_NORM))) {
3405                 int action;
3406                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3407                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3408                     hdrlen);
3409                 if (action != PF_PASS)
3410                         return (action);
3411         } else {
3412                 if (sk != NULL)
3413                         uma_zfree(V_pf_state_key_z, sk);
3414                 if (nk != NULL)
3415                         uma_zfree(V_pf_state_key_z, nk);
3416         }
3417
3418         /* copy back packet headers if we performed NAT operations */
3419         if (rewrite)
3420                 m_copyback(m, off, hdrlen, pd->hdr.any);
3421
3422         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3423             direction == PF_OUT &&
3424             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3425                 /*
3426                  * We want the state created, but we dont
3427                  * want to send this in case a partner
3428                  * firewall has to know about it to allow
3429                  * replies through it.
3430                  */
3431                 return (PF_DEFER);
3432
3433         return (PF_PASS);
3434
3435 cleanup:
3436         if (sk != NULL)
3437                 uma_zfree(V_pf_state_key_z, sk);
3438         if (nk != NULL)
3439                 uma_zfree(V_pf_state_key_z, nk);
3440         return (PF_DROP);
3441 }
3442
3443 static int
3444 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3445     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3446     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3447     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3448     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3449 {
3450         struct pf_state         *s = NULL;
3451         struct pf_src_node      *sn = NULL;
3452         struct tcphdr           *th = pd->hdr.tcp;
3453         u_int16_t                mss = V_tcp_mssdflt;
3454         u_short                  reason;
3455
3456         /* check maximums */
3457         if (r->max_states &&
3458             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3459                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3460                 REASON_SET(&reason, PFRES_MAXSTATES);
3461                 return (PF_DROP);
3462         }
3463         /* src node for filter rule */
3464         if ((r->rule_flag & PFRULE_SRCTRACK ||
3465             r->rpool.opts & PF_POOL_STICKYADDR) &&
3466             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3467                 REASON_SET(&reason, PFRES_SRCLIMIT);
3468                 goto csfailed;
3469         }
3470         /* src node for translation rule */
3471         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3472             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3473                 REASON_SET(&reason, PFRES_SRCLIMIT);
3474                 goto csfailed;
3475         }
3476         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3477         if (s == NULL) {
3478                 REASON_SET(&reason, PFRES_MEMORY);
3479                 goto csfailed;
3480         }
3481         s->rule.ptr = r;
3482         s->nat_rule.ptr = nr;
3483         s->anchor.ptr = a;
3484         STATE_INC_COUNTERS(s);
3485         if (r->allow_opts)
3486                 s->state_flags |= PFSTATE_ALLOWOPTS;
3487         if (r->rule_flag & PFRULE_STATESLOPPY)
3488                 s->state_flags |= PFSTATE_SLOPPY;
3489         s->log = r->log & PF_LOG_ALL;
3490         s->sync_state = PFSYNC_S_NONE;
3491         if (nr != NULL)
3492                 s->log |= nr->log & PF_LOG_ALL;
3493         switch (pd->proto) {
3494         case IPPROTO_TCP:
3495                 s->src.seqlo = ntohl(th->th_seq);
3496                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3497                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3498                     r->keep_state == PF_STATE_MODULATE) {
3499                         /* Generate sequence number modulator */
3500                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3501                             0)
3502                                 s->src.seqdiff = 1;
3503                         pf_change_a(&th->th_seq, &th->th_sum,
3504                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3505                         *rewrite = 1;
3506                 } else
3507                         s->src.seqdiff = 0;
3508                 if (th->th_flags & TH_SYN) {
3509                         s->src.seqhi++;
3510                         s->src.wscale = pf_get_wscale(m, off,
3511                             th->th_off, pd->af);
3512                 }
3513                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3514                 if (s->src.wscale & PF_WSCALE_MASK) {
3515                         /* Remove scale factor from initial window */
3516                         int win = s->src.max_win;
3517                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3518                         s->src.max_win = (win - 1) >>
3519                             (s->src.wscale & PF_WSCALE_MASK);
3520                 }
3521                 if (th->th_flags & TH_FIN)
3522                         s->src.seqhi++;
3523                 s->dst.seqhi = 1;
3524                 s->dst.max_win = 1;
3525                 s->src.state = TCPS_SYN_SENT;
3526                 s->dst.state = TCPS_CLOSED;
3527                 s->timeout = PFTM_TCP_FIRST_PACKET;
3528                 break;
3529         case IPPROTO_UDP:
3530                 s->src.state = PFUDPS_SINGLE;
3531                 s->dst.state = PFUDPS_NO_TRAFFIC;
3532                 s->timeout = PFTM_UDP_FIRST_PACKET;
3533                 break;
3534         case IPPROTO_ICMP:
3535 #ifdef INET6
3536         case IPPROTO_ICMPV6:
3537 #endif
3538                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3539                 break;
3540         default:
3541                 s->src.state = PFOTHERS_SINGLE;
3542                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3543                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3544         }
3545
3546         if (r->rt && r->rt != PF_FASTROUTE) {
3547                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3548                         REASON_SET(&reason, PFRES_BADSTATE);
3549                         pf_src_tree_remove_state(s);
3550                         STATE_DEC_COUNTERS(s);
3551                         uma_zfree(V_pf_state_z, s);
3552                         goto csfailed;
3553                 }
3554                 s->rt_kif = r->rpool.cur->kif;
3555         }
3556
3557         s->creation = time_uptime;
3558         s->expire = time_uptime;
3559
3560         if (sn != NULL)
3561                 s->src_node = sn;
3562         if (nsn != NULL) {
3563                 /* XXX We only modify one side for now. */
3564                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3565                 s->nat_src_node = nsn;
3566         }
3567         if (pd->proto == IPPROTO_TCP) {
3568                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3569                     off, pd, th, &s->src, &s->dst)) {
3570                         REASON_SET(&reason, PFRES_MEMORY);
3571                         pf_src_tree_remove_state(s);
3572                         STATE_DEC_COUNTERS(s);
3573                         uma_zfree(V_pf_state_z, s);
3574                         return (PF_DROP);
3575                 }
3576                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3577                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3578                     &s->src, &s->dst, rewrite)) {
3579                         /* This really shouldn't happen!!! */
3580                         DPFPRINTF(PF_DEBUG_URGENT,
3581                             ("pf_normalize_tcp_stateful failed on first pkt"));
3582                         pf_normalize_tcp_cleanup(s);
3583                         pf_src_tree_remove_state(s);
3584                         STATE_DEC_COUNTERS(s);
3585                         uma_zfree(V_pf_state_z, s);
3586                         return (PF_DROP);
3587                 }
3588         }
3589         s->direction = pd->dir;
3590
3591         /*
3592          * sk/nk could already been setup by pf_get_translation().
3593          */
3594         if (nr == NULL) {
3595                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3596                     __func__, nr, sk, nk));
3597                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3598                 if (sk == NULL)
3599                         goto csfailed;
3600                 nk = sk;
3601         } else
3602                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3603                     __func__, nr, sk, nk));
3604
3605         /* Swap sk/nk for PF_OUT. */
3606         if (pf_state_insert(BOUND_IFACE(r, kif),
3607             (pd->dir == PF_IN) ? sk : nk,
3608             (pd->dir == PF_IN) ? nk : sk, s)) {
3609                 if (pd->proto == IPPROTO_TCP)
3610                         pf_normalize_tcp_cleanup(s);
3611                 REASON_SET(&reason, PFRES_STATEINS);
3612                 pf_src_tree_remove_state(s);
3613                 STATE_DEC_COUNTERS(s);
3614                 uma_zfree(V_pf_state_z, s);
3615                 return (PF_DROP);
3616         } else
3617                 *sm = s;
3618
3619         if (tag > 0)
3620                 s->tag = tag;
3621         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3622             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3623                 s->src.state = PF_TCPS_PROXY_SRC;
3624                 /* undo NAT changes, if they have taken place */
3625                 if (nr != NULL) {
3626                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3627                         if (pd->dir == PF_OUT)
3628                                 skt = s->key[PF_SK_STACK];
3629                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3630                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3631                         if (pd->sport)
3632                                 *pd->sport = skt->port[pd->sidx];
3633                         if (pd->dport)
3634                                 *pd->dport = skt->port[pd->didx];
3635                         if (pd->proto_sum)
3636                                 *pd->proto_sum = bproto_sum;
3637                         if (pd->ip_sum)
3638                                 *pd->ip_sum = bip_sum;
3639                         m_copyback(m, off, hdrlen, pd->hdr.any);
3640                 }
3641                 s->src.seqhi = htonl(arc4random());
3642                 /* Find mss option */
3643                 int rtid = M_GETFIB(m);
3644                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3645                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3646                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3647                 s->src.mss = mss;
3648                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3649                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3650                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3651                 REASON_SET(&reason, PFRES_SYNPROXY);
3652                 return (PF_SYNPROXY_DROP);
3653         }
3654
3655         return (PF_PASS);
3656
3657 csfailed:
3658         if (sk != NULL)
3659                 uma_zfree(V_pf_state_key_z, sk);
3660         if (nk != NULL)
3661                 uma_zfree(V_pf_state_key_z, nk);
3662
3663         if (sn != NULL) {
3664                 struct pf_srchash *sh;
3665
3666                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3667                 PF_HASHROW_LOCK(sh);
3668                 if (--sn->states == 0 && sn->expire == 0) {
3669                         pf_unlink_src_node(sn);
3670                         uma_zfree(V_pf_sources_z, sn);
3671                         counter_u64_add(
3672                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3673                 }
3674                 PF_HASHROW_UNLOCK(sh);
3675         }
3676
3677         if (nsn != sn && nsn != NULL) {
3678                 struct pf_srchash *sh;
3679
3680                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3681                 PF_HASHROW_LOCK(sh);
3682                 if (--nsn->states == 0 && nsn->expire == 0) {
3683                         pf_unlink_src_node(nsn);
3684                         uma_zfree(V_pf_sources_z, nsn);
3685                         counter_u64_add(
3686                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3687                 }
3688                 PF_HASHROW_UNLOCK(sh);
3689         }
3690
3691         return (PF_DROP);
3692 }
3693
3694 static int
3695 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3696     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3697     struct pf_ruleset **rsm)
3698 {
3699         struct pf_rule          *r, *a = NULL;
3700         struct pf_ruleset       *ruleset = NULL;
3701         sa_family_t              af = pd->af;
3702         u_short                  reason;
3703         int                      tag = -1;
3704         int                      asd = 0;
3705         int                      match = 0;
3706         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3707
3708         PF_RULES_RASSERT();
3709
3710         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3711         while (r != NULL) {
3712                 r->evaluations++;
3713                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3714                         r = r->skip[PF_SKIP_IFP].ptr;
3715                 else if (r->direction && r->direction != direction)
3716                         r = r->skip[PF_SKIP_DIR].ptr;
3717                 else if (r->af && r->af != af)
3718                         r = r->skip[PF_SKIP_AF].ptr;
3719                 else if (r->proto && r->proto != pd->proto)
3720                         r = r->skip[PF_SKIP_PROTO].ptr;
3721                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3722                     r->src.neg, kif, M_GETFIB(m)))
3723                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3724                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3725                     r->dst.neg, NULL, M_GETFIB(m)))
3726                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3727                 else if (r->tos && !(r->tos == pd->tos))
3728                         r = TAILQ_NEXT(r, entries);
3729                 else if (r->os_fingerprint != PF_OSFP_ANY)
3730                         r = TAILQ_NEXT(r, entries);
3731                 else if (pd->proto == IPPROTO_UDP &&
3732                     (r->src.port_op || r->dst.port_op))
3733                         r = TAILQ_NEXT(r, entries);
3734                 else if (pd->proto == IPPROTO_TCP &&
3735                     (r->src.port_op || r->dst.port_op || r->flagset))
3736                         r = TAILQ_NEXT(r, entries);
3737                 else if ((pd->proto == IPPROTO_ICMP ||
3738                     pd->proto == IPPROTO_ICMPV6) &&
3739                     (r->type || r->code))
3740                         r = TAILQ_NEXT(r, entries);
3741                 else if (r->prob && r->prob <=
3742                     (arc4random() % (UINT_MAX - 1) + 1))
3743                         r = TAILQ_NEXT(r, entries);
3744                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3745                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3746                         r = TAILQ_NEXT(r, entries);
3747                 else {
3748                         if (r->anchor == NULL) {
3749                                 match = 1;
3750                                 *rm = r;
3751                                 *am = a;
3752                                 *rsm = ruleset;
3753                                 if ((*rm)->quick)
3754                                         break;
3755                                 r = TAILQ_NEXT(r, entries);
3756                         } else
3757                                 pf_step_into_anchor(anchor_stack, &asd,
3758                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3759                                     &match);
3760                 }
3761                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3762                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3763                         break;
3764         }
3765         r = *rm;
3766         a = *am;
3767         ruleset = *rsm;
3768
3769         REASON_SET(&reason, PFRES_MATCH);
3770
3771         if (r->log)
3772                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3773                     1);
3774
3775         if (r->action != PF_PASS)
3776                 return (PF_DROP);
3777
3778         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3779                 REASON_SET(&reason, PFRES_MEMORY);
3780                 return (PF_DROP);
3781         }
3782
3783         return (PF_PASS);
3784 }
3785
3786 static int
3787 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3788         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3789         struct pf_pdesc *pd, u_short *reason, int *copyback)
3790 {
3791         struct tcphdr           *th = pd->hdr.tcp;
3792         u_int16_t                win = ntohs(th->th_win);
3793         u_int32_t                ack, end, seq, orig_seq;
3794         u_int8_t                 sws, dws;
3795         int                      ackskew;
3796
3797         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3798                 sws = src->wscale & PF_WSCALE_MASK;
3799                 dws = dst->wscale & PF_WSCALE_MASK;
3800         } else
3801                 sws = dws = 0;
3802
3803         /*
3804          * Sequence tracking algorithm from Guido van Rooij's paper:
3805          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3806          *      tcp_filtering.ps
3807          */
3808
3809         orig_seq = seq = ntohl(th->th_seq);
3810         if (src->seqlo == 0) {
3811                 /* First packet from this end. Set its state */
3812
3813                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3814                     src->scrub == NULL) {
3815                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3816                                 REASON_SET(reason, PFRES_MEMORY);
3817                                 return (PF_DROP);
3818                         }
3819                 }
3820
3821                 /* Deferred generation of sequence number modulator */
3822                 if (dst->seqdiff && !src->seqdiff) {
3823                         /* use random iss for the TCP server */
3824                         while ((src->seqdiff = arc4random() - seq) == 0)
3825                                 ;
3826                         ack = ntohl(th->th_ack) - dst->seqdiff;
3827                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3828                             src->seqdiff), 0);
3829                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3830                         *copyback = 1;
3831                 } else {
3832                         ack = ntohl(th->th_ack);
3833                 }
3834
3835                 end = seq + pd->p_len;
3836                 if (th->th_flags & TH_SYN) {
3837                         end++;
3838                         if (dst->wscale & PF_WSCALE_FLAG) {
3839                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3840                                     pd->af);
3841                                 if (src->wscale & PF_WSCALE_FLAG) {
3842                                         /* Remove scale factor from initial
3843                                          * window */
3844                                         sws = src->wscale & PF_WSCALE_MASK;
3845                                         win = ((u_int32_t)win + (1 << sws) - 1)
3846                                             >> sws;
3847                                         dws = dst->wscale & PF_WSCALE_MASK;
3848                                 } else {
3849                                         /* fixup other window */
3850                                         dst->max_win <<= dst->wscale &
3851                                             PF_WSCALE_MASK;
3852                                         /* in case of a retrans SYN|ACK */
3853                                         dst->wscale = 0;
3854                                 }
3855                         }
3856                 }
3857                 if (th->th_flags & TH_FIN)
3858                         end++;
3859
3860                 src->seqlo = seq;
3861                 if (src->state < TCPS_SYN_SENT)
3862                         src->state = TCPS_SYN_SENT;
3863
3864                 /*
3865                  * May need to slide the window (seqhi may have been set by
3866                  * the crappy stack check or if we picked up the connection
3867                  * after establishment)
3868                  */
3869                 if (src->seqhi == 1 ||
3870                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3871                         src->seqhi = end + MAX(1, dst->max_win << dws);
3872                 if (win > src->max_win)
3873                         src->max_win = win;
3874
3875         } else {
3876                 ack = ntohl(th->th_ack) - dst->seqdiff;
3877                 if (src->seqdiff) {
3878                         /* Modulate sequence numbers */
3879                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3880                             src->seqdiff), 0);
3881                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3882                         *copyback = 1;
3883                 }
3884                 end = seq + pd->p_len;
3885                 if (th->th_flags & TH_SYN)
3886                         end++;
3887                 if (th->th_flags & TH_FIN)
3888                         end++;
3889         }
3890
3891         if ((th->th_flags & TH_ACK) == 0) {
3892                 /* Let it pass through the ack skew check */
3893                 ack = dst->seqlo;
3894         } else if ((ack == 0 &&
3895             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3896             /* broken tcp stacks do not set ack */
3897             (dst->state < TCPS_SYN_SENT)) {
3898                 /*
3899                  * Many stacks (ours included) will set the ACK number in an
3900                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3901                  */
3902                 ack = dst->seqlo;
3903         }
3904
3905         if (seq == end) {
3906                 /* Ease sequencing restrictions on no data packets */
3907                 seq = src->seqlo;
3908                 end = seq;
3909         }
3910
3911         ackskew = dst->seqlo - ack;
3912
3913
3914         /*
3915          * Need to demodulate the sequence numbers in any TCP SACK options
3916          * (Selective ACK). We could optionally validate the SACK values
3917          * against the current ACK window, either forwards or backwards, but
3918          * I'm not confident that SACK has been implemented properly
3919          * everywhere. It wouldn't surprise me if several stacks accidently
3920          * SACK too far backwards of previously ACKed data. There really aren't
3921          * any security implications of bad SACKing unless the target stack
3922          * doesn't validate the option length correctly. Someone trying to
3923          * spoof into a TCP connection won't bother blindly sending SACK
3924          * options anyway.
3925          */
3926         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3927                 if (pf_modulate_sack(m, off, pd, th, dst))
3928                         *copyback = 1;
3929         }
3930
3931
3932 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3933         if (SEQ_GEQ(src->seqhi, end) &&
3934             /* Last octet inside other's window space */
3935             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3936             /* Retrans: not more than one window back */
3937             (ackskew >= -MAXACKWINDOW) &&
3938             /* Acking not more than one reassembled fragment backwards */
3939             (ackskew <= (MAXACKWINDOW << sws)) &&
3940             /* Acking not more than one window forward */
3941             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3942             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3943             (pd->flags & PFDESC_IP_REAS) == 0)) {
3944             /* Require an exact/+1 sequence match on resets when possible */
3945
3946                 if (dst->scrub || src->scrub) {
3947                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3948                             *state, src, dst, copyback))
3949                                 return (PF_DROP);
3950                 }
3951
3952                 /* update max window */
3953                 if (src->max_win < win)
3954                         src->max_win = win;
3955                 /* synchronize sequencing */
3956                 if (SEQ_GT(end, src->seqlo))
3957                         src->seqlo = end;
3958                 /* slide the window of what the other end can send */
3959                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3960                         dst->seqhi = ack + MAX((win << sws), 1);
3961
3962
3963                 /* update states */
3964                 if (th->th_flags & TH_SYN)
3965                         if (src->state < TCPS_SYN_SENT)
3966                                 src->state = TCPS_SYN_SENT;
3967                 if (th->th_flags & TH_FIN)
3968                         if (src->state < TCPS_CLOSING)
3969                                 src->state = TCPS_CLOSING;
3970                 if (th->th_flags & TH_ACK) {
3971                         if (dst->state == TCPS_SYN_SENT) {
3972                                 dst->state = TCPS_ESTABLISHED;
3973                                 if (src->state == TCPS_ESTABLISHED &&
3974                                     (*state)->src_node != NULL &&
3975                                     pf_src_connlimit(state)) {
3976                                         REASON_SET(reason, PFRES_SRCLIMIT);
3977                                         return (PF_DROP);
3978                                 }
3979                         } else if (dst->state == TCPS_CLOSING)
3980                                 dst->state = TCPS_FIN_WAIT_2;
3981                 }
3982                 if (th->th_flags & TH_RST)
3983                         src->state = dst->state = TCPS_TIME_WAIT;
3984
3985                 /* update expire time */
3986                 (*state)->expire = time_uptime;
3987                 if (src->state >= TCPS_FIN_WAIT_2 &&
3988                     dst->state >= TCPS_FIN_WAIT_2)
3989                         (*state)->timeout = PFTM_TCP_CLOSED;
3990                 else if (src->state >= TCPS_CLOSING &&
3991                     dst->state >= TCPS_CLOSING)
3992                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
3993                 else if (src->state < TCPS_ESTABLISHED ||
3994                     dst->state < TCPS_ESTABLISHED)
3995                         (*state)->timeout = PFTM_TCP_OPENING;
3996                 else if (src->state >= TCPS_CLOSING ||
3997                     dst->state >= TCPS_CLOSING)
3998                         (*state)->timeout = PFTM_TCP_CLOSING;
3999                 else
4000                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4001
4002                 /* Fall through to PASS packet */
4003
4004         } else if ((dst->state < TCPS_SYN_SENT ||
4005                 dst->state >= TCPS_FIN_WAIT_2 ||
4006                 src->state >= TCPS_FIN_WAIT_2) &&
4007             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4008             /* Within a window forward of the originating packet */
4009             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4010             /* Within a window backward of the originating packet */
4011
4012                 /*
4013                  * This currently handles three situations:
4014                  *  1) Stupid stacks will shotgun SYNs before their peer
4015                  *     replies.
4016                  *  2) When PF catches an already established stream (the
4017                  *     firewall rebooted, the state table was flushed, routes
4018                  *     changed...)
4019                  *  3) Packets get funky immediately after the connection
4020                  *     closes (this should catch Solaris spurious ACK|FINs
4021                  *     that web servers like to spew after a close)
4022                  *
4023                  * This must be a little more careful than the above code
4024                  * since packet floods will also be caught here. We don't
4025                  * update the TTL here to mitigate the damage of a packet
4026                  * flood and so the same code can handle awkward establishment
4027                  * and a loosened connection close.
4028                  * In the establishment case, a correct peer response will
4029                  * validate the connection, go through the normal state code
4030                  * and keep updating the state TTL.
4031                  */
4032
4033                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4034                         printf("pf: loose state match: ");
4035                         pf_print_state(*state);
4036                         pf_print_flags(th->th_flags);
4037                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4038                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4039                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4040                             (unsigned long long)(*state)->packets[1],
4041                             pd->dir == PF_IN ? "in" : "out",
4042                             pd->dir == (*state)->direction ? "fwd" : "rev");
4043                 }
4044
4045                 if (dst->scrub || src->scrub) {
4046                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4047                             *state, src, dst, copyback))
4048                                 return (PF_DROP);
4049                 }
4050
4051                 /* update max window */
4052                 if (src->max_win < win)
4053                         src->max_win = win;
4054                 /* synchronize sequencing */
4055                 if (SEQ_GT(end, src->seqlo))
4056                         src->seqlo = end;
4057                 /* slide the window of what the other end can send */
4058                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4059                         dst->seqhi = ack + MAX((win << sws), 1);
4060
4061                 /*
4062                  * Cannot set dst->seqhi here since this could be a shotgunned
4063                  * SYN and not an already established connection.
4064                  */
4065
4066                 if (th->th_flags & TH_FIN)
4067                         if (src->state < TCPS_CLOSING)
4068                                 src->state = TCPS_CLOSING;
4069                 if (th->th_flags & TH_RST)
4070                         src->state = dst->state = TCPS_TIME_WAIT;
4071
4072                 /* Fall through to PASS packet */
4073
4074         } else {
4075                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4076                     (*state)->src.state == TCPS_SYN_SENT) {
4077                         /* Send RST for state mismatches during handshake */
4078                         if (!(th->th_flags & TH_RST))
4079                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4080                                     pd->dst, pd->src, th->th_dport,
4081                                     th->th_sport, ntohl(th->th_ack), 0,
4082                                     TH_RST, 0, 0,
4083                                     (*state)->rule.ptr->return_ttl, 1, 0,
4084                                     kif->pfik_ifp);
4085                         src->seqlo = 0;
4086                         src->seqhi = 1;
4087                         src->max_win = 1;
4088                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4089                         printf("pf: BAD state: ");
4090                         pf_print_state(*state);
4091                         pf_print_flags(th->th_flags);
4092                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4093                             "pkts=%llu:%llu dir=%s,%s\n",
4094                             seq, orig_seq, ack, pd->p_len, ackskew,
4095                             (unsigned long long)(*state)->packets[0],
4096                             (unsigned long long)(*state)->packets[1],
4097                             pd->dir == PF_IN ? "in" : "out",
4098                             pd->dir == (*state)->direction ? "fwd" : "rev");
4099                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4100                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4101                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4102                             ' ': '2',
4103                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4104                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4105                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4106                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4107                 }
4108                 REASON_SET(reason, PFRES_BADSTATE);
4109                 return (PF_DROP);
4110         }
4111
4112         return (PF_PASS);
4113 }
4114
4115 static int
4116 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4117         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4118 {
4119         struct tcphdr           *th = pd->hdr.tcp;
4120
4121         if (th->th_flags & TH_SYN)
4122                 if (src->state < TCPS_SYN_SENT)
4123                         src->state = TCPS_SYN_SENT;
4124         if (th->th_flags & TH_FIN)
4125                 if (src->state < TCPS_CLOSING)
4126                         src->state = TCPS_CLOSING;
4127         if (th->th_flags & TH_ACK) {
4128                 if (dst->state == TCPS_SYN_SENT) {
4129                         dst->state = TCPS_ESTABLISHED;
4130                         if (src->state == TCPS_ESTABLISHED &&
4131                             (*state)->src_node != NULL &&
4132                             pf_src_connlimit(state)) {
4133                                 REASON_SET(reason, PFRES_SRCLIMIT);
4134                                 return (PF_DROP);
4135                         }
4136                 } else if (dst->state == TCPS_CLOSING) {
4137                         dst->state = TCPS_FIN_WAIT_2;
4138                 } else if (src->state == TCPS_SYN_SENT &&
4139                     dst->state < TCPS_SYN_SENT) {
4140                         /*
4141                          * Handle a special sloppy case where we only see one
4142                          * half of the connection. If there is a ACK after
4143                          * the initial SYN without ever seeing a packet from
4144                          * the destination, set the connection to established.
4145                          */
4146                         dst->state = src->state = TCPS_ESTABLISHED;
4147                         if ((*state)->src_node != NULL &&
4148                             pf_src_connlimit(state)) {
4149                                 REASON_SET(reason, PFRES_SRCLIMIT);
4150                                 return (PF_DROP);
4151                         }
4152                 } else if (src->state == TCPS_CLOSING &&
4153                     dst->state == TCPS_ESTABLISHED &&
4154                     dst->seqlo == 0) {
4155                         /*
4156                          * Handle the closing of half connections where we
4157                          * don't see the full bidirectional FIN/ACK+ACK
4158                          * handshake.
4159                          */
4160                         dst->state = TCPS_CLOSING;
4161                 }
4162         }
4163         if (th->th_flags & TH_RST)
4164                 src->state = dst->state = TCPS_TIME_WAIT;
4165
4166         /* update expire time */
4167         (*state)->expire = time_uptime;
4168         if (src->state >= TCPS_FIN_WAIT_2 &&
4169             dst->state >= TCPS_FIN_WAIT_2)
4170                 (*state)->timeout = PFTM_TCP_CLOSED;
4171         else if (src->state >= TCPS_CLOSING &&
4172             dst->state >= TCPS_CLOSING)
4173                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4174         else if (src->state < TCPS_ESTABLISHED ||
4175             dst->state < TCPS_ESTABLISHED)
4176                 (*state)->timeout = PFTM_TCP_OPENING;
4177         else if (src->state >= TCPS_CLOSING ||
4178             dst->state >= TCPS_CLOSING)
4179                 (*state)->timeout = PFTM_TCP_CLOSING;
4180         else
4181                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4182
4183         return (PF_PASS);
4184 }
4185
4186 static int
4187 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4188     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4189     u_short *reason)
4190 {
4191         struct pf_state_key_cmp  key;
4192         struct tcphdr           *th = pd->hdr.tcp;
4193         int                      copyback = 0;
4194         struct pf_state_peer    *src, *dst;
4195         struct pf_state_key     *sk;
4196
4197         bzero(&key, sizeof(key));
4198         key.af = pd->af;
4199         key.proto = IPPROTO_TCP;
4200         if (direction == PF_IN) {       /* wire side, straight */
4201                 PF_ACPY(&key.addr[0], pd->src, key.af);
4202                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4203                 key.port[0] = th->th_sport;
4204                 key.port[1] = th->th_dport;
4205         } else {                        /* stack side, reverse */
4206                 PF_ACPY(&key.addr[1], pd->src, key.af);
4207                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4208                 key.port[1] = th->th_sport;
4209                 key.port[0] = th->th_dport;
4210         }
4211
4212         STATE_LOOKUP(kif, &key, direction, *state, pd);
4213
4214         if (direction == (*state)->direction) {
4215                 src = &(*state)->src;
4216                 dst = &(*state)->dst;
4217         } else {
4218                 src = &(*state)->dst;
4219                 dst = &(*state)->src;
4220         }
4221
4222         sk = (*state)->key[pd->didx];
4223
4224         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4225                 if (direction != (*state)->direction) {
4226                         REASON_SET(reason, PFRES_SYNPROXY);
4227                         return (PF_SYNPROXY_DROP);
4228                 }
4229                 if (th->th_flags & TH_SYN) {
4230                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4231                                 REASON_SET(reason, PFRES_SYNPROXY);
4232                                 return (PF_DROP);
4233                         }
4234                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4235                             pd->src, th->th_dport, th->th_sport,
4236                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4237                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4238                         REASON_SET(reason, PFRES_SYNPROXY);
4239                         return (PF_SYNPROXY_DROP);
4240                 } else if (!(th->th_flags & TH_ACK) ||
4241                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4242                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4243                         REASON_SET(reason, PFRES_SYNPROXY);
4244                         return (PF_DROP);
4245                 } else if ((*state)->src_node != NULL &&
4246                     pf_src_connlimit(state)) {
4247                         REASON_SET(reason, PFRES_SRCLIMIT);
4248                         return (PF_DROP);
4249                 } else
4250                         (*state)->src.state = PF_TCPS_PROXY_DST;
4251         }
4252         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4253                 if (direction == (*state)->direction) {
4254                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4255                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4256                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4257                                 REASON_SET(reason, PFRES_SYNPROXY);
4258                                 return (PF_DROP);
4259                         }
4260                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4261                         if ((*state)->dst.seqhi == 1)
4262                                 (*state)->dst.seqhi = htonl(arc4random());
4263                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4264                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4265                             sk->port[pd->sidx], sk->port[pd->didx],
4266                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4267                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4268                         REASON_SET(reason, PFRES_SYNPROXY);
4269                         return (PF_SYNPROXY_DROP);
4270                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4271                     (TH_SYN|TH_ACK)) ||
4272                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4273                         REASON_SET(reason, PFRES_SYNPROXY);
4274                         return (PF_DROP);
4275                 } else {
4276                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4277                         (*state)->dst.seqlo = ntohl(th->th_seq);
4278                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4279                             pd->src, th->th_dport, th->th_sport,
4280                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4281                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4282                             (*state)->tag, NULL);
4283                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4284                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4285                             sk->port[pd->sidx], sk->port[pd->didx],
4286                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4287                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4288                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4289                             (*state)->src.seqlo;
4290                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4291                             (*state)->dst.seqlo;
4292                         (*state)->src.seqhi = (*state)->src.seqlo +
4293                             (*state)->dst.max_win;
4294                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4295                             (*state)->src.max_win;
4296                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4297                         (*state)->src.state = (*state)->dst.state =
4298                             TCPS_ESTABLISHED;
4299                         REASON_SET(reason, PFRES_SYNPROXY);
4300                         return (PF_SYNPROXY_DROP);
4301                 }
4302         }
4303
4304         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4305             dst->state >= TCPS_FIN_WAIT_2 &&
4306             src->state >= TCPS_FIN_WAIT_2) {
4307                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4308                         printf("pf: state reuse ");
4309                         pf_print_state(*state);
4310                         pf_print_flags(th->th_flags);
4311                         printf("\n");
4312                 }
4313                 /* XXX make sure it's the same direction ?? */
4314                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4315                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4316                 *state = NULL;
4317                 return (PF_DROP);
4318         }
4319
4320         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4321                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4322                         return (PF_DROP);
4323         } else {
4324                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4325                     &copyback) == PF_DROP)
4326                         return (PF_DROP);
4327         }
4328
4329         /* translate source/destination address, if necessary */
4330         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4331                 struct pf_state_key *nk = (*state)->key[pd->didx];
4332
4333                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4334                     nk->port[pd->sidx] != th->th_sport)
4335                         pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4336                             &th->th_sum, &nk->addr[pd->sidx],
4337                             nk->port[pd->sidx], 0, pd->af);
4338
4339                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4340                     nk->port[pd->didx] != th->th_dport)
4341                         pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4342                             &th->th_sum, &nk->addr[pd->didx],
4343                             nk->port[pd->didx], 0, pd->af);
4344                 copyback = 1;
4345         }
4346
4347         /* Copyback sequence modulation or stateful scrub changes if needed */
4348         if (copyback)
4349                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4350
4351         return (PF_PASS);
4352 }
4353
4354 static int
4355 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4356     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4357 {
4358         struct pf_state_peer    *src, *dst;
4359         struct pf_state_key_cmp  key;
4360         struct udphdr           *uh = pd->hdr.udp;
4361
4362         bzero(&key, sizeof(key));
4363         key.af = pd->af;
4364         key.proto = IPPROTO_UDP;
4365         if (direction == PF_IN) {       /* wire side, straight */
4366                 PF_ACPY(&key.addr[0], pd->src, key.af);
4367                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4368                 key.port[0] = uh->uh_sport;
4369                 key.port[1] = uh->uh_dport;
4370         } else {                        /* stack side, reverse */
4371                 PF_ACPY(&key.addr[1], pd->src, key.af);
4372                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4373                 key.port[1] = uh->uh_sport;
4374                 key.port[0] = uh->uh_dport;
4375         }
4376
4377         STATE_LOOKUP(kif, &key, direction, *state, pd);
4378
4379         if (direction == (*state)->direction) {
4380                 src = &(*state)->src;
4381                 dst = &(*state)->dst;
4382         } else {
4383                 src = &(*state)->dst;
4384                 dst = &(*state)->src;
4385         }
4386
4387         /* update states */
4388         if (src->state < PFUDPS_SINGLE)
4389                 src->state = PFUDPS_SINGLE;
4390         if (dst->state == PFUDPS_SINGLE)
4391                 dst->state = PFUDPS_MULTIPLE;
4392
4393         /* update expire time */
4394         (*state)->expire = time_uptime;
4395         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4396                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4397         else
4398                 (*state)->timeout = PFTM_UDP_SINGLE;
4399
4400         /* translate source/destination address, if necessary */
4401         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4402                 struct pf_state_key *nk = (*state)->key[pd->didx];
4403
4404                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4405                     nk->port[pd->sidx] != uh->uh_sport)
4406                         pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4407                             &uh->uh_sum, &nk->addr[pd->sidx],
4408                             nk->port[pd->sidx], 1, pd->af);
4409
4410                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4411                     nk->port[pd->didx] != uh->uh_dport)
4412                         pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4413                             &uh->uh_sum, &nk->addr[pd->didx],
4414                             nk->port[pd->didx], 1, pd->af);
4415                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4416         }
4417
4418         return (PF_PASS);
4419 }
4420
4421 static int
4422 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4423     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4424 {
4425         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4426         u_int16_t        icmpid = 0, *icmpsum;
4427         u_int8_t         icmptype;
4428         int              state_icmp = 0;
4429         struct pf_state_key_cmp key;
4430
4431         bzero(&key, sizeof(key));
4432         switch (pd->proto) {
4433 #ifdef INET
4434         case IPPROTO_ICMP:
4435                 icmptype = pd->hdr.icmp->icmp_type;
4436                 icmpid = pd->hdr.icmp->icmp_id;
4437                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4438
4439                 if (icmptype == ICMP_UNREACH ||
4440                     icmptype == ICMP_SOURCEQUENCH ||
4441                     icmptype == ICMP_REDIRECT ||
4442                     icmptype == ICMP_TIMXCEED ||
4443                     icmptype == ICMP_PARAMPROB)
4444                         state_icmp++;
4445                 break;
4446 #endif /* INET */
4447 #ifdef INET6
4448         case IPPROTO_ICMPV6:
4449                 icmptype = pd->hdr.icmp6->icmp6_type;
4450                 icmpid = pd->hdr.icmp6->icmp6_id;
4451                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4452
4453                 if (icmptype == ICMP6_DST_UNREACH ||
4454                     icmptype == ICMP6_PACKET_TOO_BIG ||
4455                     icmptype == ICMP6_TIME_EXCEEDED ||
4456                     icmptype == ICMP6_PARAM_PROB)
4457                         state_icmp++;
4458                 break;
4459 #endif /* INET6 */
4460         }
4461
4462         if (!state_icmp) {
4463
4464                 /*
4465                  * ICMP query/reply message not related to a TCP/UDP packet.
4466                  * Search for an ICMP state.
4467                  */
4468                 key.af = pd->af;
4469                 key.proto = pd->proto;
4470                 key.port[0] = key.port[1] = icmpid;
4471                 if (direction == PF_IN) {       /* wire side, straight */
4472                         PF_ACPY(&key.addr[0], pd->src, key.af);
4473                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4474                 } else {                        /* stack side, reverse */
4475                         PF_ACPY(&key.addr[1], pd->src, key.af);
4476                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4477                 }
4478
4479                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4480
4481                 (*state)->expire = time_uptime;
4482                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4483
4484                 /* translate source/destination address, if necessary */
4485                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4486                         struct pf_state_key *nk = (*state)->key[pd->didx];
4487
4488                         switch (pd->af) {
4489 #ifdef INET
4490                         case AF_INET:
4491                                 if (PF_ANEQ(pd->src,
4492                                     &nk->addr[pd->sidx], AF_INET))
4493                                         pf_change_a(&saddr->v4.s_addr,
4494                                             pd->ip_sum,
4495                                             nk->addr[pd->sidx].v4.s_addr, 0);
4496
4497                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4498                                     AF_INET))
4499                                         pf_change_a(&daddr->v4.s_addr,
4500                                             pd->ip_sum,
4501                                             nk->addr[pd->didx].v4.s_addr, 0);
4502
4503                                 if (nk->port[0] !=
4504                                     pd->hdr.icmp->icmp_id) {
4505                                         pd->hdr.icmp->icmp_cksum =
4506                                             pf_cksum_fixup(
4507                                             pd->hdr.icmp->icmp_cksum, icmpid,
4508                                             nk->port[pd->sidx], 0);
4509                                         pd->hdr.icmp->icmp_id =
4510                                             nk->port[pd->sidx];
4511                                 }
4512
4513                                 m_copyback(m, off, ICMP_MINLEN,
4514                                     (caddr_t )pd->hdr.icmp);
4515                                 break;
4516 #endif /* INET */
4517 #ifdef INET6
4518                         case AF_INET6:
4519                                 if (PF_ANEQ(pd->src,
4520                                     &nk->addr[pd->sidx], AF_INET6))
4521                                         pf_change_a6(saddr,
4522                                             &pd->hdr.icmp6->icmp6_cksum,
4523                                             &nk->addr[pd->sidx], 0);
4524
4525                                 if (PF_ANEQ(pd->dst,
4526                                     &nk->addr[pd->didx], AF_INET6))
4527                                         pf_change_a6(daddr,
4528                                             &pd->hdr.icmp6->icmp6_cksum,
4529                                             &nk->addr[pd->didx], 0);
4530
4531                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4532                                     (caddr_t )pd->hdr.icmp6);
4533                                 break;
4534 #endif /* INET6 */
4535                         }
4536                 }
4537                 return (PF_PASS);
4538
4539         } else {
4540                 /*
4541                  * ICMP error message in response to a TCP/UDP packet.
4542                  * Extract the inner TCP/UDP header and search for that state.
4543                  */
4544
4545                 struct pf_pdesc pd2;
4546                 bzero(&pd2, sizeof pd2);
4547 #ifdef INET
4548                 struct ip       h2;
4549 #endif /* INET */
4550 #ifdef INET6
4551                 struct ip6_hdr  h2_6;
4552                 int             terminal = 0;
4553 #endif /* INET6 */
4554                 int             ipoff2 = 0;
4555                 int             off2 = 0;
4556
4557                 pd2.af = pd->af;
4558                 /* Payload packet is from the opposite direction. */
4559                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4560                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4561                 switch (pd->af) {
4562 #ifdef INET
4563                 case AF_INET:
4564                         /* offset of h2 in mbuf chain */
4565                         ipoff2 = off + ICMP_MINLEN;
4566
4567                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4568                             NULL, reason, pd2.af)) {
4569                                 DPFPRINTF(PF_DEBUG_MISC,
4570                                     ("pf: ICMP error message too short "
4571                                     "(ip)\n"));
4572                                 return (PF_DROP);
4573                         }
4574                         /*
4575                          * ICMP error messages don't refer to non-first
4576                          * fragments
4577                          */
4578                         if (h2.ip_off & htons(IP_OFFMASK)) {
4579                                 REASON_SET(reason, PFRES_FRAG);
4580                                 return (PF_DROP);
4581                         }
4582
4583                         /* offset of protocol header that follows h2 */
4584                         off2 = ipoff2 + (h2.ip_hl << 2);
4585
4586                         pd2.proto = h2.ip_p;
4587                         pd2.src = (struct pf_addr *)&h2.ip_src;
4588                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4589                         pd2.ip_sum = &h2.ip_sum;
4590                         break;
4591 #endif /* INET */
4592 #ifdef INET6
4593                 case AF_INET6:
4594                         ipoff2 = off + sizeof(struct icmp6_hdr);
4595
4596                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4597                             NULL, reason, pd2.af)) {
4598                                 DPFPRINTF(PF_DEBUG_MISC,
4599                                     ("pf: ICMP error message too short "
4600                                     "(ip6)\n"));
4601                                 return (PF_DROP);
4602                         }
4603                         pd2.proto = h2_6.ip6_nxt;
4604                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4605                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4606                         pd2.ip_sum = NULL;
4607                         off2 = ipoff2 + sizeof(h2_6);
4608                         do {
4609                                 switch (pd2.proto) {
4610                                 case IPPROTO_FRAGMENT:
4611                                         /*
4612                                          * ICMPv6 error messages for
4613                                          * non-first fragments
4614                                          */
4615                                         REASON_SET(reason, PFRES_FRAG);
4616                                         return (PF_DROP);
4617                                 case IPPROTO_AH:
4618                                 case IPPROTO_HOPOPTS:
4619                                 case IPPROTO_ROUTING:
4620                                 case IPPROTO_DSTOPTS: {
4621                                         /* get next header and header length */
4622                                         struct ip6_ext opt6;
4623
4624                                         if (!pf_pull_hdr(m, off2, &opt6,
4625                                             sizeof(opt6), NULL, reason,
4626                                             pd2.af)) {
4627                                                 DPFPRINTF(PF_DEBUG_MISC,
4628                                                     ("pf: ICMPv6 short opt\n"));
4629                                                 return (PF_DROP);
4630                                         }
4631                                         if (pd2.proto == IPPROTO_AH)
4632                                                 off2 += (opt6.ip6e_len + 2) * 4;
4633                                         else
4634                                                 off2 += (opt6.ip6e_len + 1) * 8;
4635                                         pd2.proto = opt6.ip6e_nxt;
4636                                         /* goto the next header */
4637                                         break;
4638                                 }
4639                                 default:
4640                                         terminal++;
4641                                         break;
4642                                 }
4643                         } while (!terminal);
4644                         break;
4645 #endif /* INET6 */
4646                 }
4647
4648                 switch (pd2.proto) {
4649                 case IPPROTO_TCP: {
4650                         struct tcphdr            th;
4651                         u_int32_t                seq;
4652                         struct pf_state_peer    *src, *dst;
4653                         u_int8_t                 dws;
4654                         int                      copyback = 0;
4655
4656                         /*
4657                          * Only the first 8 bytes of the TCP header can be
4658                          * expected. Don't access any TCP header fields after
4659                          * th_seq, an ackskew test is not possible.
4660                          */
4661                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4662                             pd2.af)) {
4663                                 DPFPRINTF(PF_DEBUG_MISC,
4664                                     ("pf: ICMP error message too short "
4665                                     "(tcp)\n"));
4666                                 return (PF_DROP);
4667                         }
4668
4669                         key.af = pd2.af;
4670                         key.proto = IPPROTO_TCP;
4671                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4672                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4673                         key.port[pd2.sidx] = th.th_sport;
4674                         key.port[pd2.didx] = th.th_dport;
4675
4676                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4677
4678                         if (direction == (*state)->direction) {
4679                                 src = &(*state)->dst;
4680                                 dst = &(*state)->src;
4681                         } else {
4682                                 src = &(*state)->src;
4683                                 dst = &(*state)->dst;
4684                         }
4685
4686                         if (src->wscale && dst->wscale)
4687                                 dws = dst->wscale & PF_WSCALE_MASK;
4688                         else
4689                                 dws = 0;
4690
4691                         /* Demodulate sequence number */
4692                         seq = ntohl(th.th_seq) - src->seqdiff;
4693                         if (src->seqdiff) {
4694                                 pf_change_a(&th.th_seq, icmpsum,
4695                                     htonl(seq), 0);
4696                                 copyback = 1;
4697                         }
4698
4699                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4700                             (!SEQ_GEQ(src->seqhi, seq) ||
4701                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4702                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4703                                         printf("pf: BAD ICMP %d:%d ",
4704                                             icmptype, pd->hdr.icmp->icmp_code);
4705                                         pf_print_host(pd->src, 0, pd->af);
4706                                         printf(" -> ");
4707                                         pf_print_host(pd->dst, 0, pd->af);
4708                                         printf(" state: ");
4709                                         pf_print_state(*state);
4710                                         printf(" seq=%u\n", seq);
4711                                 }
4712                                 REASON_SET(reason, PFRES_BADSTATE);
4713                                 return (PF_DROP);
4714                         } else {
4715                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4716                                         printf("pf: OK ICMP %d:%d ",
4717                                             icmptype, pd->hdr.icmp->icmp_code);
4718                                         pf_print_host(pd->src, 0, pd->af);
4719                                         printf(" -> ");
4720                                         pf_print_host(pd->dst, 0, pd->af);
4721                                         printf(" state: ");
4722                                         pf_print_state(*state);
4723                                         printf(" seq=%u\n", seq);
4724                                 }
4725                         }
4726
4727                         /* translate source/destination address, if necessary */
4728                         if ((*state)->key[PF_SK_WIRE] !=
4729                             (*state)->key[PF_SK_STACK]) {
4730                                 struct pf_state_key *nk =
4731                                     (*state)->key[pd->didx];
4732
4733                                 if (PF_ANEQ(pd2.src,
4734                                     &nk->addr[pd2.sidx], pd2.af) ||
4735                                     nk->port[pd2.sidx] != th.th_sport)
4736                                         pf_change_icmp(pd2.src, &th.th_sport,
4737                                             daddr, &nk->addr[pd2.sidx],
4738                                             nk->port[pd2.sidx], NULL,
4739                                             pd2.ip_sum, icmpsum,
4740                                             pd->ip_sum, 0, pd2.af);
4741
4742                                 if (PF_ANEQ(pd2.dst,
4743                                     &nk->addr[pd2.didx], pd2.af) ||
4744                                     nk->port[pd2.didx] != th.th_dport)
4745                                         pf_change_icmp(pd2.dst, &th.th_dport,
4746                                             NULL, /* XXX Inbound NAT? */
4747                                             &nk->addr[pd2.didx],
4748                                             nk->port[pd2.didx], NULL,
4749                                             pd2.ip_sum, icmpsum,
4750                                             pd->ip_sum, 0, pd2.af);
4751                                 copyback = 1;
4752                         }
4753
4754                         if (copyback) {
4755                                 switch (pd2.af) {
4756 #ifdef INET
4757                                 case AF_INET:
4758                                         m_copyback(m, off, ICMP_MINLEN,
4759                                             (caddr_t )pd->hdr.icmp);
4760                                         m_copyback(m, ipoff2, sizeof(h2),
4761                                             (caddr_t )&h2);
4762                                         break;
4763 #endif /* INET */
4764 #ifdef INET6
4765                                 case AF_INET6:
4766                                         m_copyback(m, off,
4767                                             sizeof(struct icmp6_hdr),
4768                                             (caddr_t )pd->hdr.icmp6);
4769                                         m_copyback(m, ipoff2, sizeof(h2_6),
4770                                             (caddr_t )&h2_6);
4771                                         break;
4772 #endif /* INET6 */
4773                                 }
4774                                 m_copyback(m, off2, 8, (caddr_t)&th);
4775                         }
4776
4777                         return (PF_PASS);
4778                         break;
4779                 }
4780                 case IPPROTO_UDP: {
4781                         struct udphdr           uh;
4782
4783                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4784                             NULL, reason, pd2.af)) {
4785                                 DPFPRINTF(PF_DEBUG_MISC,
4786                                     ("pf: ICMP error message too short "
4787                                     "(udp)\n"));
4788                                 return (PF_DROP);
4789                         }
4790
4791                         key.af = pd2.af;
4792                         key.proto = IPPROTO_UDP;
4793                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4794                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4795                         key.port[pd2.sidx] = uh.uh_sport;
4796                         key.port[pd2.didx] = uh.uh_dport;
4797
4798                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4799
4800                         /* translate source/destination address, if necessary */
4801                         if ((*state)->key[PF_SK_WIRE] !=
4802                             (*state)->key[PF_SK_STACK]) {
4803                                 struct pf_state_key *nk =
4804                                     (*state)->key[pd->didx];
4805
4806                                 if (PF_ANEQ(pd2.src,
4807                                     &nk->addr[pd2.sidx], pd2.af) ||
4808                                     nk->port[pd2.sidx] != uh.uh_sport)
4809                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4810                                             daddr, &nk->addr[pd2.sidx],
4811                                             nk->port[pd2.sidx], &uh.uh_sum,
4812                                             pd2.ip_sum, icmpsum,
4813                                             pd->ip_sum, 1, pd2.af);
4814
4815                                 if (PF_ANEQ(pd2.dst,
4816                                     &nk->addr[pd2.didx], pd2.af) ||
4817                                     nk->port[pd2.didx] != uh.uh_dport)
4818                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4819                                             NULL, /* XXX Inbound NAT? */
4820                                             &nk->addr[pd2.didx],
4821                                             nk->port[pd2.didx], &uh.uh_sum,
4822                                             pd2.ip_sum, icmpsum,
4823                                             pd->ip_sum, 1, pd2.af);
4824
4825                                 switch (pd2.af) {
4826 #ifdef INET
4827                                 case AF_INET:
4828                                         m_copyback(m, off, ICMP_MINLEN,
4829                                             (caddr_t )pd->hdr.icmp);
4830                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4831                                         break;
4832 #endif /* INET */
4833 #ifdef INET6
4834                                 case AF_INET6:
4835                                         m_copyback(m, off,
4836                                             sizeof(struct icmp6_hdr),
4837                                             (caddr_t )pd->hdr.icmp6);
4838                                         m_copyback(m, ipoff2, sizeof(h2_6),
4839                                             (caddr_t )&h2_6);
4840                                         break;
4841 #endif /* INET6 */
4842                                 }
4843                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4844                         }
4845                         return (PF_PASS);
4846                         break;
4847                 }
4848 #ifdef INET
4849                 case IPPROTO_ICMP: {
4850                         struct icmp             iih;
4851
4852                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4853                             NULL, reason, pd2.af)) {
4854                                 DPFPRINTF(PF_DEBUG_MISC,
4855                                     ("pf: ICMP error message too short i"
4856                                     "(icmp)\n"));
4857                                 return (PF_DROP);
4858                         }
4859
4860                         key.af = pd2.af;
4861                         key.proto = IPPROTO_ICMP;
4862                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4863                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4864                         key.port[0] = key.port[1] = iih.icmp_id;
4865
4866                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4867
4868                         /* translate source/destination address, if necessary */
4869                         if ((*state)->key[PF_SK_WIRE] !=
4870                             (*state)->key[PF_SK_STACK]) {
4871                                 struct pf_state_key *nk =
4872                                     (*state)->key[pd->didx];
4873
4874                                 if (PF_ANEQ(pd2.src,
4875                                     &nk->addr[pd2.sidx], pd2.af) ||
4876                                     nk->port[pd2.sidx] != iih.icmp_id)
4877                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4878                                             daddr, &nk->addr[pd2.sidx],
4879                                             nk->port[pd2.sidx], NULL,
4880                                             pd2.ip_sum, icmpsum,
4881                                             pd->ip_sum, 0, AF_INET);
4882
4883                                 if (PF_ANEQ(pd2.dst,
4884                                     &nk->addr[pd2.didx], pd2.af) ||
4885                                     nk->port[pd2.didx] != iih.icmp_id)
4886                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4887                                             NULL, /* XXX Inbound NAT? */
4888                                             &nk->addr[pd2.didx],
4889                                             nk->port[pd2.didx], NULL,
4890                                             pd2.ip_sum, icmpsum,
4891                                             pd->ip_sum, 0, AF_INET);
4892
4893                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4894                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4895                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4896                         }
4897                         return (PF_PASS);
4898                         break;
4899                 }
4900 #endif /* INET */
4901 #ifdef INET6
4902                 case IPPROTO_ICMPV6: {
4903                         struct icmp6_hdr        iih;
4904
4905                         if (!pf_pull_hdr(m, off2, &iih,
4906                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4907                                 DPFPRINTF(PF_DEBUG_MISC,
4908                                     ("pf: ICMP error message too short "
4909                                     "(icmp6)\n"));
4910                                 return (PF_DROP);
4911                         }
4912
4913                         key.af = pd2.af;
4914                         key.proto = IPPROTO_ICMPV6;
4915                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4916                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4917                         key.port[0] = key.port[1] = iih.icmp6_id;
4918
4919                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4920
4921                         /* translate source/destination address, if necessary */
4922                         if ((*state)->key[PF_SK_WIRE] !=
4923                             (*state)->key[PF_SK_STACK]) {
4924                                 struct pf_state_key *nk =
4925                                     (*state)->key[pd->didx];
4926
4927                                 if (PF_ANEQ(pd2.src,
4928                                     &nk->addr[pd2.sidx], pd2.af) ||
4929                                     nk->port[pd2.sidx] != iih.icmp6_id)
4930                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4931                                             daddr, &nk->addr[pd2.sidx],
4932                                             nk->port[pd2.sidx], NULL,
4933                                             pd2.ip_sum, icmpsum,
4934                                             pd->ip_sum, 0, AF_INET6);
4935
4936                                 if (PF_ANEQ(pd2.dst,
4937                                     &nk->addr[pd2.didx], pd2.af) ||
4938                                     nk->port[pd2.didx] != iih.icmp6_id)
4939                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
4940                                             NULL, /* XXX Inbound NAT? */
4941                                             &nk->addr[pd2.didx],
4942                                             nk->port[pd2.didx], NULL,
4943                                             pd2.ip_sum, icmpsum,
4944                                             pd->ip_sum, 0, AF_INET6);
4945
4946                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4947                                     (caddr_t)pd->hdr.icmp6);
4948                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4949                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
4950                                     (caddr_t)&iih);
4951                         }
4952                         return (PF_PASS);
4953                         break;
4954                 }
4955 #endif /* INET6 */
4956                 default: {
4957                         key.af = pd2.af;
4958                         key.proto = pd2.proto;
4959                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4960                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4961                         key.port[0] = key.port[1] = 0;
4962
4963                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4964
4965                         /* translate source/destination address, if necessary */
4966                         if ((*state)->key[PF_SK_WIRE] !=
4967                             (*state)->key[PF_SK_STACK]) {
4968                                 struct pf_state_key *nk =
4969                                     (*state)->key[pd->didx];
4970
4971                                 if (PF_ANEQ(pd2.src,
4972                                     &nk->addr[pd2.sidx], pd2.af))
4973                                         pf_change_icmp(pd2.src, NULL, daddr,
4974                                             &nk->addr[pd2.sidx], 0, NULL,
4975                                             pd2.ip_sum, icmpsum,
4976                                             pd->ip_sum, 0, pd2.af);
4977
4978                                 if (PF_ANEQ(pd2.dst,
4979                                     &nk->addr[pd2.didx], pd2.af))
4980                                         pf_change_icmp(pd2.src, NULL,
4981                                             NULL, /* XXX Inbound NAT? */
4982                                             &nk->addr[pd2.didx], 0, NULL,
4983                                             pd2.ip_sum, icmpsum,
4984                                             pd->ip_sum, 0, pd2.af);
4985
4986                                 switch (pd2.af) {
4987 #ifdef INET
4988                                 case AF_INET:
4989                                         m_copyback(m, off, ICMP_MINLEN,
4990                                             (caddr_t)pd->hdr.icmp);
4991                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4992                                         break;
4993 #endif /* INET */
4994 #ifdef INET6
4995                                 case AF_INET6:
4996                                         m_copyback(m, off,
4997                                             sizeof(struct icmp6_hdr),
4998                                             (caddr_t )pd->hdr.icmp6);
4999                                         m_copyback(m, ipoff2, sizeof(h2_6),
5000                                             (caddr_t )&h2_6);
5001                                         break;
5002 #endif /* INET6 */
5003                                 }
5004                         }
5005                         return (PF_PASS);
5006                         break;
5007                 }
5008                 }
5009         }
5010 }
5011
5012 static int
5013 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5014     struct mbuf *m, struct pf_pdesc *pd)
5015 {
5016         struct pf_state_peer    *src, *dst;
5017         struct pf_state_key_cmp  key;
5018
5019         bzero(&key, sizeof(key));
5020         key.af = pd->af;
5021         key.proto = pd->proto;
5022         if (direction == PF_IN) {
5023                 PF_ACPY(&key.addr[0], pd->src, key.af);
5024                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5025                 key.port[0] = key.port[1] = 0;
5026         } else {
5027                 PF_ACPY(&key.addr[1], pd->src, key.af);
5028                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5029                 key.port[1] = key.port[0] = 0;
5030         }
5031
5032         STATE_LOOKUP(kif, &key, direction, *state, pd);
5033
5034         if (direction == (*state)->direction) {
5035                 src = &(*state)->src;
5036                 dst = &(*state)->dst;
5037         } else {
5038                 src = &(*state)->dst;
5039                 dst = &(*state)->src;
5040         }
5041
5042         /* update states */
5043         if (src->state < PFOTHERS_SINGLE)
5044                 src->state = PFOTHERS_SINGLE;
5045         if (dst->state == PFOTHERS_SINGLE)
5046                 dst->state = PFOTHERS_MULTIPLE;
5047
5048         /* update expire time */
5049         (*state)->expire = time_uptime;
5050         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5051                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5052         else
5053                 (*state)->timeout = PFTM_OTHER_SINGLE;
5054
5055         /* translate source/destination address, if necessary */
5056         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5057                 struct pf_state_key *nk = (*state)->key[pd->didx];
5058
5059                 KASSERT(nk, ("%s: nk is null", __func__));
5060                 KASSERT(pd, ("%s: pd is null", __func__));
5061                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5062                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5063                 switch (pd->af) {
5064 #ifdef INET
5065                 case AF_INET:
5066                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5067                                 pf_change_a(&pd->src->v4.s_addr,
5068                                     pd->ip_sum,
5069                                     nk->addr[pd->sidx].v4.s_addr,
5070                                     0);
5071
5072
5073                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5074                                 pf_change_a(&pd->dst->v4.s_addr,
5075                                     pd->ip_sum,
5076                                     nk->addr[pd->didx].v4.s_addr,
5077                                     0);
5078
5079                                 break;
5080 #endif /* INET */
5081 #ifdef INET6
5082                 case AF_INET6:
5083                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5084                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5085
5086                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5087                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5088 #endif /* INET6 */
5089                 }
5090         }
5091         return (PF_PASS);
5092 }
5093
5094 /*
5095  * ipoff and off are measured from the start of the mbuf chain.
5096  * h must be at "ipoff" on the mbuf chain.
5097  */
5098 void *
5099 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5100     u_short *actionp, u_short *reasonp, sa_family_t af)
5101 {
5102         switch (af) {
5103 #ifdef INET
5104         case AF_INET: {
5105                 struct ip       *h = mtod(m, struct ip *);
5106                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5107
5108                 if (fragoff) {
5109                         if (fragoff >= len)
5110                                 ACTION_SET(actionp, PF_PASS);
5111                         else {
5112                                 ACTION_SET(actionp, PF_DROP);
5113                                 REASON_SET(reasonp, PFRES_FRAG);
5114                         }
5115                         return (NULL);
5116                 }
5117                 if (m->m_pkthdr.len < off + len ||
5118                     ntohs(h->ip_len) < off + len) {
5119                         ACTION_SET(actionp, PF_DROP);
5120                         REASON_SET(reasonp, PFRES_SHORT);
5121                         return (NULL);
5122                 }
5123                 break;
5124         }
5125 #endif /* INET */
5126 #ifdef INET6
5127         case AF_INET6: {
5128                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5129
5130                 if (m->m_pkthdr.len < off + len ||
5131                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5132                     (unsigned)(off + len)) {
5133                         ACTION_SET(actionp, PF_DROP);
5134                         REASON_SET(reasonp, PFRES_SHORT);
5135                         return (NULL);
5136                 }
5137                 break;
5138         }
5139 #endif /* INET6 */
5140         }
5141         m_copydata(m, off, len, p);
5142         return (p);
5143 }
5144
5145 int
5146 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5147     int rtableid)
5148 {
5149 #ifdef RADIX_MPATH
5150         struct radix_node_head  *rnh;
5151 #endif
5152         struct sockaddr_in      *dst;
5153         int                      ret = 1;
5154         int                      check_mpath;
5155 #ifdef INET6
5156         struct sockaddr_in6     *dst6;
5157         struct route_in6         ro;
5158 #else
5159         struct route             ro;
5160 #endif
5161         struct radix_node       *rn;
5162         struct rtentry          *rt;
5163         struct ifnet            *ifp;
5164
5165         check_mpath = 0;
5166 #ifdef RADIX_MPATH
5167         /* XXX: stick to table 0 for now */
5168         rnh = rt_tables_get_rnh(0, af);
5169         if (rnh != NULL && rn_mpath_capable(rnh))
5170                 check_mpath = 1;
5171 #endif
5172         bzero(&ro, sizeof(ro));
5173         switch (af) {
5174         case AF_INET:
5175                 dst = satosin(&ro.ro_dst);
5176                 dst->sin_family = AF_INET;
5177                 dst->sin_len = sizeof(*dst);
5178                 dst->sin_addr = addr->v4;
5179                 break;
5180 #ifdef INET6
5181         case AF_INET6:
5182                 /*
5183                  * Skip check for addresses with embedded interface scope,
5184                  * as they would always match anyway.
5185                  */
5186                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5187                         goto out;
5188                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5189                 dst6->sin6_family = AF_INET6;
5190                 dst6->sin6_len = sizeof(*dst6);
5191                 dst6->sin6_addr = addr->v6;
5192                 break;
5193 #endif /* INET6 */
5194         default:
5195                 return (0);
5196         }
5197
5198         /* Skip checks for ipsec interfaces */
5199         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5200                 goto out;
5201
5202         switch (af) {
5203 #ifdef INET6
5204         case AF_INET6:
5205                 in6_rtalloc_ign(&ro, 0, rtableid);
5206                 break;
5207 #endif
5208 #ifdef INET
5209         case AF_INET:
5210                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5211                 break;
5212 #endif
5213         default:
5214                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5215                 break;
5216         }
5217
5218         if (ro.ro_rt != NULL) {
5219                 /* No interface given, this is a no-route check */
5220                 if (kif == NULL)
5221                         goto out;
5222
5223                 if (kif->pfik_ifp == NULL) {
5224                         ret = 0;
5225                         goto out;
5226                 }
5227
5228                 /* Perform uRPF check if passed input interface */
5229                 ret = 0;
5230                 rn = (struct radix_node *)ro.ro_rt;
5231                 do {
5232                         rt = (struct rtentry *)rn;
5233                         ifp = rt->rt_ifp;
5234
5235                         if (kif->pfik_ifp == ifp)
5236                                 ret = 1;
5237 #ifdef RADIX_MPATH
5238                         rn = rn_mpath_next(rn);
5239 #endif
5240                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5241         } else
5242                 ret = 0;
5243 out:
5244         if (ro.ro_rt != NULL)
5245                 RTFREE(ro.ro_rt);
5246         return (ret);
5247 }
5248
5249 #ifdef INET
5250 static void
5251 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5252     struct pf_state *s, struct pf_pdesc *pd)
5253 {
5254         struct mbuf             *m0, *m1;
5255         struct sockaddr_in      dst;
5256         struct ip               *ip;
5257         struct ifnet            *ifp = NULL;
5258         struct pf_addr           naddr;
5259         struct pf_src_node      *sn = NULL;
5260         int                      error = 0;
5261         uint16_t                 ip_len, ip_off;
5262
5263         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5264         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5265             __func__));
5266
5267         if ((pd->pf_mtag == NULL &&
5268             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5269             pd->pf_mtag->routed++ > 3) {
5270                 m0 = *m;
5271                 *m = NULL;
5272                 goto bad_locked;
5273         }
5274
5275         if (r->rt == PF_DUPTO) {
5276                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5277                         if (s)
5278                                 PF_STATE_UNLOCK(s);
5279                         return;
5280                 }
5281         } else {
5282                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5283                         if (s)
5284                                 PF_STATE_UNLOCK(s);
5285                         return;
5286                 }
5287                 m0 = *m;
5288         }
5289
5290         ip = mtod(m0, struct ip *);
5291
5292         bzero(&dst, sizeof(dst));
5293         dst.sin_family = AF_INET;
5294         dst.sin_len = sizeof(dst);
5295         dst.sin_addr = ip->ip_dst;
5296
5297         if (r->rt == PF_FASTROUTE) {
5298                 struct rtentry *rt;
5299
5300                 if (s)
5301                         PF_STATE_UNLOCK(s);
5302                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5303                 if (rt == NULL) {
5304                         KMOD_IPSTAT_INC(ips_noroute);
5305                         error = EHOSTUNREACH;
5306                         goto bad;
5307                 }
5308
5309                 ifp = rt->rt_ifp;
5310                 counter_u64_add(rt->rt_pksent, 1);
5311
5312                 if (rt->rt_flags & RTF_GATEWAY)
5313                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5314                 RTFREE_LOCKED(rt);
5315         } else {
5316                 if (TAILQ_EMPTY(&r->rpool.list)) {
5317                         DPFPRINTF(PF_DEBUG_URGENT,
5318                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5319                         goto bad_locked;
5320                 }
5321                 if (s == NULL) {
5322                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5323                             &naddr, NULL, &sn);
5324                         if (!PF_AZERO(&naddr, AF_INET))
5325                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5326                         ifp = r->rpool.cur->kif ?
5327                             r->rpool.cur->kif->pfik_ifp : NULL;
5328                 } else {
5329                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5330                                 dst.sin_addr.s_addr =
5331                                     s->rt_addr.v4.s_addr;
5332                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5333                         PF_STATE_UNLOCK(s);
5334                 }
5335         }
5336         if (ifp == NULL)
5337                 goto bad;
5338
5339         if (oifp != ifp) {
5340                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5341                         goto bad;
5342                 else if (m0 == NULL)
5343                         goto done;
5344                 if (m0->m_len < sizeof(struct ip)) {
5345                         DPFPRINTF(PF_DEBUG_URGENT,
5346                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5347                         goto bad;
5348                 }
5349                 ip = mtod(m0, struct ip *);
5350         }
5351
5352         if (ifp->if_flags & IFF_LOOPBACK)
5353                 m0->m_flags |= M_SKIP_FIREWALL;
5354
5355         ip_len = ntohs(ip->ip_len);
5356         ip_off = ntohs(ip->ip_off);
5357
5358         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5359         m0->m_pkthdr.csum_flags |= CSUM_IP;
5360         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5361                 in_delayed_cksum(m0);
5362                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5363         }
5364 #ifdef SCTP
5365         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5366                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5367                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5368         }
5369 #endif
5370
5371         /*
5372          * If small enough for interface, or the interface will take
5373          * care of the fragmentation for us, we can just send directly.
5374          */
5375         if (ip_len <= ifp->if_mtu ||
5376             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5377             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5378                 ip->ip_sum = 0;
5379                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5380                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5381                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5382                 }
5383                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5384                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5385                 goto done;
5386         }
5387
5388         /* Balk when DF bit is set or the interface didn't support TSO. */
5389         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5390                 error = EMSGSIZE;
5391                 KMOD_IPSTAT_INC(ips_cantfrag);
5392                 if (r->rt != PF_DUPTO) {
5393                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5394                             ifp->if_mtu);
5395                         goto done;
5396                 } else
5397                         goto bad;
5398         }
5399
5400         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5401         if (error)
5402                 goto bad;
5403
5404         for (; m0; m0 = m1) {
5405                 m1 = m0->m_nextpkt;
5406                 m0->m_nextpkt = NULL;
5407                 if (error == 0) {
5408                         m_clrprotoflags(m0);
5409                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5410                 } else
5411                         m_freem(m0);
5412         }
5413
5414         if (error == 0)
5415                 KMOD_IPSTAT_INC(ips_fragmented);
5416
5417 done:
5418         if (r->rt != PF_DUPTO)
5419                 *m = NULL;
5420         return;
5421
5422 bad_locked:
5423         if (s)
5424                 PF_STATE_UNLOCK(s);
5425 bad:
5426         m_freem(m0);
5427         goto done;
5428 }
5429 #endif /* INET */
5430
5431 #ifdef INET6
5432 static void
5433 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5434     struct pf_state *s, struct pf_pdesc *pd)
5435 {
5436         struct mbuf             *m0;
5437         struct sockaddr_in6     dst;
5438         struct ip6_hdr          *ip6;
5439         struct ifnet            *ifp = NULL;
5440         struct pf_addr           naddr;
5441         struct pf_src_node      *sn = NULL;
5442
5443         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5444         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5445             __func__));
5446
5447         if ((pd->pf_mtag == NULL &&
5448             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5449             pd->pf_mtag->routed++ > 3) {
5450                 m0 = *m;
5451                 *m = NULL;
5452                 goto bad_locked;
5453         }
5454
5455         if (r->rt == PF_DUPTO) {
5456                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5457                         if (s)
5458                                 PF_STATE_UNLOCK(s);
5459                         return;
5460                 }
5461         } else {
5462                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5463                         if (s)
5464                                 PF_STATE_UNLOCK(s);
5465                         return;
5466                 }
5467                 m0 = *m;
5468         }
5469
5470         ip6 = mtod(m0, struct ip6_hdr *);
5471
5472         bzero(&dst, sizeof(dst));
5473         dst.sin6_family = AF_INET6;
5474         dst.sin6_len = sizeof(dst);
5475         dst.sin6_addr = ip6->ip6_dst;
5476
5477         /* Cheat. XXX why only in the v6 case??? */
5478         if (r->rt == PF_FASTROUTE) {
5479                 if (s)
5480                         PF_STATE_UNLOCK(s);
5481                 m0->m_flags |= M_SKIP_FIREWALL;
5482                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5483                 *m = NULL;
5484                 return;
5485         }
5486
5487         if (TAILQ_EMPTY(&r->rpool.list)) {
5488                 DPFPRINTF(PF_DEBUG_URGENT,
5489                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5490                 goto bad_locked;
5491         }
5492         if (s == NULL) {
5493                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5494                     &naddr, NULL, &sn);
5495                 if (!PF_AZERO(&naddr, AF_INET6))
5496                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5497                             &naddr, AF_INET6);
5498                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5499         } else {
5500                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5501                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5502                             &s->rt_addr, AF_INET6);
5503                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5504         }
5505
5506         if (s)
5507                 PF_STATE_UNLOCK(s);
5508
5509         if (ifp == NULL)
5510                 goto bad;
5511
5512         if (oifp != ifp) {
5513                 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS)
5514                         goto bad;
5515                 else if (m0 == NULL)
5516                         goto done;
5517                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5518                         DPFPRINTF(PF_DEBUG_URGENT,
5519                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5520                             __func__));
5521                         goto bad;
5522                 }
5523                 ip6 = mtod(m0, struct ip6_hdr *);
5524         }
5525
5526         if (ifp->if_flags & IFF_LOOPBACK)
5527                 m0->m_flags |= M_SKIP_FIREWALL;
5528
5529         /*
5530          * If the packet is too large for the outgoing interface,
5531          * send back an icmp6 error.
5532          */
5533         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5534                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5535         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5536                 nd6_output(ifp, ifp, m0, &dst, NULL);
5537         else {
5538                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5539                 if (r->rt != PF_DUPTO)
5540                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5541                 else
5542                         goto bad;
5543         }
5544
5545 done:
5546         if (r->rt != PF_DUPTO)
5547                 *m = NULL;
5548         return;
5549
5550 bad_locked:
5551         if (s)
5552                 PF_STATE_UNLOCK(s);
5553 bad:
5554         m_freem(m0);
5555         goto done;
5556 }
5557 #endif /* INET6 */
5558
5559 /*
5560  * FreeBSD supports cksum offloads for the following drivers.
5561  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5562  *   ti(4), txp(4), xl(4)
5563  *
5564  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5565  *  network driver performed cksum including pseudo header, need to verify
5566  *   csum_data
5567  * CSUM_DATA_VALID :
5568  *  network driver performed cksum, needs to additional pseudo header
5569  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5570  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5571  *
5572  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5573  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5574  * TCP/UDP layer.
5575  * Also, set csum_data to 0xffff to force cksum validation.
5576  */
5577 static int
5578 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5579 {
5580         u_int16_t sum = 0;
5581         int hw_assist = 0;
5582         struct ip *ip;
5583
5584         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5585                 return (1);
5586         if (m->m_pkthdr.len < off + len)
5587                 return (1);
5588
5589         switch (p) {
5590         case IPPROTO_TCP:
5591                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5592                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5593                                 sum = m->m_pkthdr.csum_data;
5594                         } else {
5595                                 ip = mtod(m, struct ip *);
5596                                 sum = in_pseudo(ip->ip_src.s_addr,
5597                                 ip->ip_dst.s_addr, htonl((u_short)len +
5598                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5599                         }
5600                         sum ^= 0xffff;
5601                         ++hw_assist;
5602                 }
5603                 break;
5604         case IPPROTO_UDP:
5605                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5606                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5607                                 sum = m->m_pkthdr.csum_data;
5608                         } else {
5609                                 ip = mtod(m, struct ip *);
5610                                 sum = in_pseudo(ip->ip_src.s_addr,
5611                                 ip->ip_dst.s_addr, htonl((u_short)len +
5612                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5613                         }
5614                         sum ^= 0xffff;
5615                         ++hw_assist;
5616                 }
5617                 break;
5618         case IPPROTO_ICMP:
5619 #ifdef INET6
5620         case IPPROTO_ICMPV6:
5621 #endif /* INET6 */
5622                 break;
5623         default:
5624                 return (1);
5625         }
5626
5627         if (!hw_assist) {
5628                 switch (af) {
5629                 case AF_INET:
5630                         if (p == IPPROTO_ICMP) {
5631                                 if (m->m_len < off)
5632                                         return (1);
5633                                 m->m_data += off;
5634                                 m->m_len -= off;
5635                                 sum = in_cksum(m, len);
5636                                 m->m_data -= off;
5637                                 m->m_len += off;
5638                         } else {
5639                                 if (m->m_len < sizeof(struct ip))
5640                                         return (1);
5641                                 sum = in4_cksum(m, p, off, len);
5642                         }
5643                         break;
5644 #ifdef INET6
5645                 case AF_INET6:
5646                         if (m->m_len < sizeof(struct ip6_hdr))
5647                                 return (1);
5648                         sum = in6_cksum(m, p, off, len);
5649                         break;
5650 #endif /* INET6 */
5651                 default:
5652                         return (1);
5653                 }
5654         }
5655         if (sum) {
5656                 switch (p) {
5657                 case IPPROTO_TCP:
5658                     {
5659                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5660                         break;
5661                     }
5662                 case IPPROTO_UDP:
5663                     {
5664                         KMOD_UDPSTAT_INC(udps_badsum);
5665                         break;
5666                     }
5667 #ifdef INET
5668                 case IPPROTO_ICMP:
5669                     {
5670                         KMOD_ICMPSTAT_INC(icps_checksum);
5671                         break;
5672                     }
5673 #endif
5674 #ifdef INET6
5675                 case IPPROTO_ICMPV6:
5676                     {
5677                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5678                         break;
5679                     }
5680 #endif /* INET6 */
5681                 }
5682                 return (1);
5683         } else {
5684                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5685                         m->m_pkthdr.csum_flags |=
5686                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5687                         m->m_pkthdr.csum_data = 0xffff;
5688                 }
5689         }
5690         return (0);
5691 }
5692
5693
5694 #ifdef INET
5695 int
5696 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5697 {
5698         struct pfi_kif          *kif;
5699         u_short                  action, reason = 0, log = 0;
5700         struct mbuf             *m = *m0;
5701         struct ip               *h = NULL;
5702         struct m_tag            *ipfwtag;
5703         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5704         struct pf_state         *s = NULL;
5705         struct pf_ruleset       *ruleset = NULL;
5706         struct pf_pdesc          pd;
5707         int                      off, dirndx, pqid = 0;
5708
5709         M_ASSERTPKTHDR(m);
5710
5711         if (!V_pf_status.running)
5712                 return (PF_PASS);
5713
5714         memset(&pd, 0, sizeof(pd));
5715
5716         kif = (struct pfi_kif *)ifp->if_pf_kif;
5717
5718         if (kif == NULL) {
5719                 DPFPRINTF(PF_DEBUG_URGENT,
5720                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5721                 return (PF_DROP);
5722         }
5723         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5724                 return (PF_PASS);
5725
5726         if (m->m_flags & M_SKIP_FIREWALL)
5727                 return (PF_PASS);
5728
5729         pd.pf_mtag = pf_find_mtag(m);
5730
5731         PF_RULES_RLOCK();
5732
5733         if (ip_divert_ptr != NULL &&
5734             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5735                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5736                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5737                         if (pd.pf_mtag == NULL &&
5738                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5739                                 action = PF_DROP;
5740                                 goto done;
5741                         }
5742                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5743                         m_tag_delete(m, ipfwtag);
5744                 }
5745                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5746                         m->m_flags |= M_FASTFWD_OURS;
5747                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5748                 }
5749         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5750                 /* We do IP header normalization and packet reassembly here */
5751                 action = PF_DROP;
5752                 goto done;
5753         }
5754         m = *m0;        /* pf_normalize messes with m0 */
5755         h = mtod(m, struct ip *);
5756
5757         off = h->ip_hl << 2;
5758         if (off < (int)sizeof(struct ip)) {
5759                 action = PF_DROP;
5760                 REASON_SET(&reason, PFRES_SHORT);
5761                 log = 1;
5762                 goto done;
5763         }
5764
5765         pd.src = (struct pf_addr *)&h->ip_src;
5766         pd.dst = (struct pf_addr *)&h->ip_dst;
5767         pd.sport = pd.dport = NULL;
5768         pd.ip_sum = &h->ip_sum;
5769         pd.proto_sum = NULL;
5770         pd.proto = h->ip_p;
5771         pd.dir = dir;
5772         pd.sidx = (dir == PF_IN) ? 0 : 1;
5773         pd.didx = (dir == PF_IN) ? 1 : 0;
5774         pd.af = AF_INET;
5775         pd.tos = h->ip_tos;
5776         pd.tot_len = ntohs(h->ip_len);
5777
5778         /* handle fragments that didn't get reassembled by normalization */
5779         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5780                 action = pf_test_fragment(&r, dir, kif, m, h,
5781                     &pd, &a, &ruleset);
5782                 goto done;
5783         }
5784
5785         switch (h->ip_p) {
5786
5787         case IPPROTO_TCP: {
5788                 struct tcphdr   th;
5789
5790                 pd.hdr.tcp = &th;
5791                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5792                     &action, &reason, AF_INET)) {
5793                         log = action != PF_PASS;
5794                         goto done;
5795                 }
5796                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5797                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5798                         pqid = 1;
5799                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5800                 if (action == PF_DROP)
5801                         goto done;
5802                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5803                     &reason);
5804                 if (action == PF_PASS) {
5805                         if (pfsync_update_state_ptr != NULL)
5806                                 pfsync_update_state_ptr(s);
5807                         r = s->rule.ptr;
5808                         a = s->anchor.ptr;
5809                         log = s->log;
5810                 } else if (s == NULL)
5811                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5812                             &a, &ruleset, inp);
5813                 break;
5814         }
5815
5816         case IPPROTO_UDP: {
5817                 struct udphdr   uh;
5818
5819                 pd.hdr.udp = &uh;
5820                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5821                     &action, &reason, AF_INET)) {
5822                         log = action != PF_PASS;
5823                         goto done;
5824                 }
5825                 if (uh.uh_dport == 0 ||
5826                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5827                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5828                         action = PF_DROP;
5829                         REASON_SET(&reason, PFRES_SHORT);
5830                         goto done;
5831                 }
5832                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5833                 if (action == PF_PASS) {
5834                         if (pfsync_update_state_ptr != NULL)
5835                                 pfsync_update_state_ptr(s);
5836                         r = s->rule.ptr;
5837                         a = s->anchor.ptr;
5838                         log = s->log;
5839                 } else if (s == NULL)
5840                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5841                             &a, &ruleset, inp);
5842                 break;
5843         }
5844
5845         case IPPROTO_ICMP: {
5846                 struct icmp     ih;
5847
5848                 pd.hdr.icmp = &ih;
5849                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5850                     &action, &reason, AF_INET)) {
5851                         log = action != PF_PASS;
5852                         goto done;
5853                 }
5854                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5855                     &reason);
5856                 if (action == PF_PASS) {
5857                         if (pfsync_update_state_ptr != NULL)
5858                                 pfsync_update_state_ptr(s);
5859                         r = s->rule.ptr;
5860                         a = s->anchor.ptr;
5861                         log = s->log;
5862                 } else if (s == NULL)
5863                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5864                             &a, &ruleset, inp);
5865                 break;
5866         }
5867
5868 #ifdef INET6
5869         case IPPROTO_ICMPV6: {
5870                 action = PF_DROP;
5871                 DPFPRINTF(PF_DEBUG_MISC,
5872                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5873                 goto done;
5874         }
5875 #endif
5876
5877         default:
5878                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5879                 if (action == PF_PASS) {
5880                         if (pfsync_update_state_ptr != NULL)
5881                                 pfsync_update_state_ptr(s);
5882                         r = s->rule.ptr;
5883                         a = s->anchor.ptr;
5884                         log = s->log;
5885                 } else if (s == NULL)
5886                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5887                             &a, &ruleset, inp);
5888                 break;
5889         }
5890
5891 done:
5892         PF_RULES_RUNLOCK();
5893         if (action == PF_PASS && h->ip_hl > 5 &&
5894             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5895                 action = PF_DROP;
5896                 REASON_SET(&reason, PFRES_IPOPTIONS);
5897                 log = 1;
5898                 DPFPRINTF(PF_DEBUG_MISC,
5899                     ("pf: dropping packet with ip options\n"));
5900         }
5901
5902         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5903                 action = PF_DROP;
5904                 REASON_SET(&reason, PFRES_MEMORY);
5905         }
5906         if (r->rtableid >= 0)
5907                 M_SETFIB(m, r->rtableid);
5908
5909 #ifdef ALTQ
5910         if (action == PF_PASS && r->qid) {
5911                 if (pd.pf_mtag == NULL &&
5912                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5913                         action = PF_DROP;
5914                         REASON_SET(&reason, PFRES_MEMORY);
5915                 } else {
5916                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
5917                                 pd.pf_mtag->qid = r->pqid;
5918                         else
5919                                 pd.pf_mtag->qid = r->qid;
5920                         /* Add hints for ecn. */
5921                         pd.pf_mtag->hdr = h;
5922                 }
5923
5924         }
5925 #endif /* ALTQ */
5926
5927         /*
5928          * connections redirected to loopback should not match sockets
5929          * bound specifically to loopback due to security implications,
5930          * see tcp_input() and in_pcblookup_listen().
5931          */
5932         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5933             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5934             (s->nat_rule.ptr->action == PF_RDR ||
5935             s->nat_rule.ptr->action == PF_BINAT) &&
5936             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5937                 m->m_flags |= M_SKIP_FIREWALL;
5938
5939         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5940             !PACKET_LOOPED(&pd)) {
5941
5942                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5943                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5944                 if (ipfwtag != NULL) {
5945                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5946                             ntohs(r->divert.port);
5947                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5948
5949                         if (s)
5950                                 PF_STATE_UNLOCK(s);
5951
5952                         m_tag_prepend(m, ipfwtag);
5953                         if (m->m_flags & M_FASTFWD_OURS) {
5954                                 if (pd.pf_mtag == NULL &&
5955                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5956                                         action = PF_DROP;
5957                                         REASON_SET(&reason, PFRES_MEMORY);
5958                                         log = 1;
5959                                         DPFPRINTF(PF_DEBUG_MISC,
5960                                             ("pf: failed to allocate tag\n"));
5961                                 } else {
5962                                         pd.pf_mtag->flags |=
5963                                             PF_FASTFWD_OURS_PRESENT;
5964                                         m->m_flags &= ~M_FASTFWD_OURS;
5965                                 }
5966                         }
5967                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
5968                         *m0 = NULL;
5969
5970                         return (action);
5971                 } else {
5972                         /* XXX: ipfw has the same behaviour! */
5973                         action = PF_DROP;
5974                         REASON_SET(&reason, PFRES_MEMORY);
5975                         log = 1;
5976                         DPFPRINTF(PF_DEBUG_MISC,
5977                             ("pf: failed to allocate divert tag\n"));
5978                 }
5979         }
5980
5981         if (log) {
5982                 struct pf_rule *lr;
5983
5984                 if (s != NULL && s->nat_rule.ptr != NULL &&
5985                     s->nat_rule.ptr->log & PF_LOG_ALL)
5986                         lr = s->nat_rule.ptr;
5987                 else
5988                         lr = r;
5989                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5990                     (s == NULL));
5991         }
5992
5993         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5994         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5995
5996         if (action == PF_PASS || r->action == PF_DROP) {
5997                 dirndx = (dir == PF_OUT);
5998                 r->packets[dirndx]++;
5999                 r->bytes[dirndx] += pd.tot_len;
6000                 if (a != NULL) {
6001                         a->packets[dirndx]++;
6002                         a->bytes[dirndx] += pd.tot_len;
6003                 }
6004                 if (s != NULL) {
6005                         if (s->nat_rule.ptr != NULL) {
6006                                 s->nat_rule.ptr->packets[dirndx]++;
6007                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6008                         }
6009                         if (s->src_node != NULL) {
6010                                 s->src_node->packets[dirndx]++;
6011                                 s->src_node->bytes[dirndx] += pd.tot_len;
6012                         }
6013                         if (s->nat_src_node != NULL) {
6014                                 s->nat_src_node->packets[dirndx]++;
6015                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6016                         }
6017                         dirndx = (dir == s->direction) ? 0 : 1;
6018                         s->packets[dirndx]++;
6019                         s->bytes[dirndx] += pd.tot_len;
6020                 }
6021                 tr = r;
6022                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6023                 if (nr != NULL && r == &V_pf_default_rule)
6024                         tr = nr;
6025                 if (tr->src.addr.type == PF_ADDR_TABLE)
6026                         pfr_update_stats(tr->src.addr.p.tbl,
6027                             (s == NULL) ? pd.src :
6028                             &s->key[(s->direction == PF_IN)]->
6029                                 addr[(s->direction == PF_OUT)],
6030                             pd.af, pd.tot_len, dir == PF_OUT,
6031                             r->action == PF_PASS, tr->src.neg);
6032                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6033                         pfr_update_stats(tr->dst.addr.p.tbl,
6034                             (s == NULL) ? pd.dst :
6035                             &s->key[(s->direction == PF_IN)]->
6036                                 addr[(s->direction == PF_IN)],
6037                             pd.af, pd.tot_len, dir == PF_OUT,
6038                             r->action == PF_PASS, tr->dst.neg);
6039         }
6040
6041         switch (action) {
6042         case PF_SYNPROXY_DROP:
6043                 m_freem(*m0);
6044         case PF_DEFER:
6045                 *m0 = NULL;
6046                 action = PF_PASS;
6047                 break;
6048         case PF_DROP:
6049                 m_freem(*m0);
6050                 *m0 = NULL;
6051                 break;
6052         default:
6053                 /* pf_route() returns unlocked. */
6054                 if (r->rt) {
6055                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6056                         return (action);
6057                 }
6058                 break;
6059         }
6060         if (s)
6061                 PF_STATE_UNLOCK(s);
6062
6063         return (action);
6064 }
6065 #endif /* INET */
6066
6067 #ifdef INET6
6068 int
6069 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6070 {
6071         struct pfi_kif          *kif;
6072         u_short                  action, reason = 0, log = 0;
6073         struct mbuf             *m = *m0, *n = NULL;
6074         struct m_tag            *mtag;
6075         struct ip6_hdr          *h = NULL;
6076         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6077         struct pf_state         *s = NULL;
6078         struct pf_ruleset       *ruleset = NULL;
6079         struct pf_pdesc          pd;
6080         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6081         int                      fwdir = dir;
6082
6083         M_ASSERTPKTHDR(m);
6084
6085         if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif)
6086                 fwdir = PF_FWD;
6087
6088         if (!V_pf_status.running)
6089                 return (PF_PASS);
6090
6091         memset(&pd, 0, sizeof(pd));
6092         pd.pf_mtag = pf_find_mtag(m);
6093
6094         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6095                 return (PF_PASS);
6096
6097         kif = (struct pfi_kif *)ifp->if_pf_kif;
6098         if (kif == NULL) {
6099                 DPFPRINTF(PF_DEBUG_URGENT,
6100                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6101                 return (PF_DROP);
6102         }
6103         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6104                 return (PF_PASS);
6105
6106         if (m->m_flags & M_SKIP_FIREWALL)
6107                 return (PF_PASS);
6108
6109         PF_RULES_RLOCK();
6110
6111         /* We do IP header normalization and packet reassembly here */
6112         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6113                 action = PF_DROP;
6114                 goto done;
6115         }
6116         m = *m0;        /* pf_normalize messes with m0 */
6117         h = mtod(m, struct ip6_hdr *);
6118
6119 #if 1
6120         /*
6121          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6122          * will do something bad, so drop the packet for now.
6123          */
6124         if (htons(h->ip6_plen) == 0) {
6125                 action = PF_DROP;
6126                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6127                 goto done;
6128         }
6129 #endif
6130
6131         pd.src = (struct pf_addr *)&h->ip6_src;
6132         pd.dst = (struct pf_addr *)&h->ip6_dst;
6133         pd.sport = pd.dport = NULL;
6134         pd.ip_sum = NULL;
6135         pd.proto_sum = NULL;
6136         pd.dir = dir;
6137         pd.sidx = (dir == PF_IN) ? 0 : 1;
6138         pd.didx = (dir == PF_IN) ? 1 : 0;
6139         pd.af = AF_INET6;
6140         pd.tos = 0;
6141         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6142
6143         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6144         pd.proto = h->ip6_nxt;
6145         do {
6146                 switch (pd.proto) {
6147                 case IPPROTO_FRAGMENT:
6148                         action = pf_test_fragment(&r, dir, kif, m, h,
6149                             &pd, &a, &ruleset);
6150                         if (action == PF_DROP)
6151                                 REASON_SET(&reason, PFRES_FRAG);
6152                         goto done;
6153                 case IPPROTO_ROUTING: {
6154                         struct ip6_rthdr rthdr;
6155
6156                         if (rh_cnt++) {
6157                                 DPFPRINTF(PF_DEBUG_MISC,
6158                                     ("pf: IPv6 more than one rthdr\n"));
6159                                 action = PF_DROP;
6160                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6161                                 log = 1;
6162                                 goto done;
6163                         }
6164                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6165                             &reason, pd.af)) {
6166                                 DPFPRINTF(PF_DEBUG_MISC,
6167                                     ("pf: IPv6 short rthdr\n"));
6168                                 action = PF_DROP;
6169                                 REASON_SET(&reason, PFRES_SHORT);
6170                                 log = 1;
6171                                 goto done;
6172                         }
6173                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6174                                 DPFPRINTF(PF_DEBUG_MISC,
6175                                     ("pf: IPv6 rthdr0\n"));
6176                                 action = PF_DROP;
6177                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6178                                 log = 1;
6179                                 goto done;
6180                         }
6181                         /* FALLTHROUGH */
6182                 }
6183                 case IPPROTO_AH:
6184                 case IPPROTO_HOPOPTS:
6185                 case IPPROTO_DSTOPTS: {
6186                         /* get next header and header length */
6187                         struct ip6_ext  opt6;
6188
6189                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6190                             NULL, &reason, pd.af)) {
6191                                 DPFPRINTF(PF_DEBUG_MISC,
6192                                     ("pf: IPv6 short opt\n"));
6193                                 action = PF_DROP;
6194                                 log = 1;
6195                                 goto done;
6196                         }
6197                         if (pd.proto == IPPROTO_AH)
6198                                 off += (opt6.ip6e_len + 2) * 4;
6199                         else
6200                                 off += (opt6.ip6e_len + 1) * 8;
6201                         pd.proto = opt6.ip6e_nxt;
6202                         /* goto the next header */
6203                         break;
6204                 }
6205                 default:
6206                         terminal++;
6207                         break;
6208                 }
6209         } while (!terminal);
6210
6211         /* if there's no routing header, use unmodified mbuf for checksumming */
6212         if (!n)
6213                 n = m;
6214
6215         switch (pd.proto) {
6216
6217         case IPPROTO_TCP: {
6218                 struct tcphdr   th;
6219
6220                 pd.hdr.tcp = &th;
6221                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6222                     &action, &reason, AF_INET6)) {
6223                         log = action != PF_PASS;
6224                         goto done;
6225                 }
6226                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6227                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6228                 if (action == PF_DROP)
6229                         goto done;
6230                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6231                     &reason);
6232                 if (action == PF_PASS) {
6233                         if (pfsync_update_state_ptr != NULL)
6234                                 pfsync_update_state_ptr(s);
6235                         r = s->rule.ptr;
6236                         a = s->anchor.ptr;
6237                         log = s->log;
6238                 } else if (s == NULL)
6239                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6240                             &a, &ruleset, inp);
6241                 break;
6242         }
6243
6244         case IPPROTO_UDP: {
6245                 struct udphdr   uh;
6246
6247                 pd.hdr.udp = &uh;
6248                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6249                     &action, &reason, AF_INET6)) {
6250                         log = action != PF_PASS;
6251                         goto done;
6252                 }
6253                 if (uh.uh_dport == 0 ||
6254                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6255                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6256                         action = PF_DROP;
6257                         REASON_SET(&reason, PFRES_SHORT);
6258                         goto done;
6259                 }
6260                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6261                 if (action == PF_PASS) {
6262                         if (pfsync_update_state_ptr != NULL)
6263                                 pfsync_update_state_ptr(s);
6264                         r = s->rule.ptr;
6265                         a = s->anchor.ptr;
6266                         log = s->log;
6267                 } else if (s == NULL)
6268                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6269                             &a, &ruleset, inp);
6270                 break;
6271         }
6272
6273         case IPPROTO_ICMP: {
6274                 action = PF_DROP;
6275                 DPFPRINTF(PF_DEBUG_MISC,
6276                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6277                 goto done;
6278         }
6279
6280         case IPPROTO_ICMPV6: {
6281                 struct icmp6_hdr        ih;
6282
6283                 pd.hdr.icmp6 = &ih;
6284                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6285                     &action, &reason, AF_INET6)) {
6286                         log = action != PF_PASS;
6287                         goto done;
6288                 }
6289                 action = pf_test_state_icmp(&s, dir, kif,
6290                     m, off, h, &pd, &reason);
6291                 if (action == PF_PASS) {
6292                         if (pfsync_update_state_ptr != NULL)
6293                                 pfsync_update_state_ptr(s);
6294                         r = s->rule.ptr;
6295                         a = s->anchor.ptr;
6296                         log = s->log;
6297                 } else if (s == NULL)
6298                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6299                             &a, &ruleset, inp);
6300                 break;
6301         }
6302
6303         default:
6304                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6305                 if (action == PF_PASS) {
6306                         if (pfsync_update_state_ptr != NULL)
6307                                 pfsync_update_state_ptr(s);
6308                         r = s->rule.ptr;
6309                         a = s->anchor.ptr;
6310                         log = s->log;
6311                 } else if (s == NULL)
6312                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6313                             &a, &ruleset, inp);
6314                 break;
6315         }
6316
6317 done:
6318         PF_RULES_RUNLOCK();
6319         if (n != m) {
6320                 m_freem(n);
6321                 n = NULL;
6322         }
6323
6324         /* handle dangerous IPv6 extension headers. */
6325         if (action == PF_PASS && rh_cnt &&
6326             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6327                 action = PF_DROP;
6328                 REASON_SET(&reason, PFRES_IPOPTIONS);
6329                 log = 1;
6330                 DPFPRINTF(PF_DEBUG_MISC,
6331                     ("pf: dropping packet with dangerous v6 headers\n"));
6332         }
6333
6334         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6335                 action = PF_DROP;
6336                 REASON_SET(&reason, PFRES_MEMORY);
6337         }
6338         if (r->rtableid >= 0)
6339                 M_SETFIB(m, r->rtableid);
6340
6341 #ifdef ALTQ
6342         if (action == PF_PASS && r->qid) {
6343                 if (pd.pf_mtag == NULL &&
6344                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6345                         action = PF_DROP;
6346                         REASON_SET(&reason, PFRES_MEMORY);
6347                 } else {
6348                         if (pd.tos & IPTOS_LOWDELAY)
6349                                 pd.pf_mtag->qid = r->pqid;
6350                         else
6351                                 pd.pf_mtag->qid = r->qid;
6352                         /* Add hints for ecn. */
6353                         pd.pf_mtag->hdr = h;
6354                 }
6355         }
6356 #endif /* ALTQ */
6357
6358         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6359             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6360             (s->nat_rule.ptr->action == PF_RDR ||
6361             s->nat_rule.ptr->action == PF_BINAT) &&
6362             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6363                 m->m_flags |= M_SKIP_FIREWALL;
6364
6365         /* XXX: Anybody working on it?! */
6366         if (r->divert.port)
6367                 printf("pf: divert(9) is not supported for IPv6\n");
6368
6369         if (log) {
6370                 struct pf_rule *lr;
6371
6372                 if (s != NULL && s->nat_rule.ptr != NULL &&
6373                     s->nat_rule.ptr->log & PF_LOG_ALL)
6374                         lr = s->nat_rule.ptr;
6375                 else
6376                         lr = r;
6377                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6378                     &pd, (s == NULL));
6379         }
6380
6381         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6382         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6383
6384         if (action == PF_PASS || r->action == PF_DROP) {
6385                 dirndx = (dir == PF_OUT);
6386                 r->packets[dirndx]++;
6387                 r->bytes[dirndx] += pd.tot_len;
6388                 if (a != NULL) {
6389                         a->packets[dirndx]++;
6390                         a->bytes[dirndx] += pd.tot_len;
6391                 }
6392                 if (s != NULL) {
6393                         if (s->nat_rule.ptr != NULL) {
6394                                 s->nat_rule.ptr->packets[dirndx]++;
6395                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6396                         }
6397                         if (s->src_node != NULL) {
6398                                 s->src_node->packets[dirndx]++;
6399                                 s->src_node->bytes[dirndx] += pd.tot_len;
6400                         }
6401                         if (s->nat_src_node != NULL) {
6402                                 s->nat_src_node->packets[dirndx]++;
6403                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6404                         }
6405                         dirndx = (dir == s->direction) ? 0 : 1;
6406                         s->packets[dirndx]++;
6407                         s->bytes[dirndx] += pd.tot_len;
6408                 }
6409                 tr = r;
6410                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6411                 if (nr != NULL && r == &V_pf_default_rule)
6412                         tr = nr;
6413                 if (tr->src.addr.type == PF_ADDR_TABLE)
6414                         pfr_update_stats(tr->src.addr.p.tbl,
6415                             (s == NULL) ? pd.src :
6416                             &s->key[(s->direction == PF_IN)]->addr[0],
6417                             pd.af, pd.tot_len, dir == PF_OUT,
6418                             r->action == PF_PASS, tr->src.neg);
6419                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6420                         pfr_update_stats(tr->dst.addr.p.tbl,
6421                             (s == NULL) ? pd.dst :
6422                             &s->key[(s->direction == PF_IN)]->addr[1],
6423                             pd.af, pd.tot_len, dir == PF_OUT,
6424                             r->action == PF_PASS, tr->dst.neg);
6425         }
6426
6427         switch (action) {
6428         case PF_SYNPROXY_DROP:
6429                 m_freem(*m0);
6430         case PF_DEFER:
6431                 *m0 = NULL;
6432                 action = PF_PASS;
6433                 break;
6434         case PF_DROP:
6435                 m_freem(*m0);
6436                 *m0 = NULL;
6437                 break;
6438         default:
6439                 /* pf_route6() returns unlocked. */
6440                 if (r->rt) {
6441                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6442                         return (action);
6443                 }
6444                 break;
6445         }
6446
6447         if (s)
6448                 PF_STATE_UNLOCK(s);
6449
6450         /* If reassembled packet passed, create new fragments. */
6451         if (action == PF_PASS && *m0 && fwdir == PF_FWD &&
6452             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6453                 action = pf_refragment6(ifp, m0, mtag);
6454
6455         return (action);
6456 }
6457 #endif /* INET6 */