sys/netpfil/pf/pf.c

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