sys/netpfil/pf/pf.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   3  *
   4  * Copyright (c) 2001 Daniel Hartmeier
   5  * Copyright (c) 2002 - 2008 Henning Brauer
   6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  *    - Redistributions of source code must retain the above copyright
  14  *      notice, this list of conditions and the following disclaimer.
  15  *    - Redistributions in binary form must reproduce the above
  16  *      copyright notice, this list of conditions and the following
  17  *      disclaimer in the documentation and/or other materials provided
  18  *      with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  * Effort sponsored in part by the Defense Advanced Research Projects
  34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  36  *
  37  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_inet.h"
  44 #include "opt_inet6.h"
  45 #include "opt_bpf.h"
  46 #include "opt_pf.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/bus.h>
  50 #include <sys/endian.h>
  51 #include <sys/hash.h>
  52 #include <sys/interrupt.h>
  53 #include <sys/kernel.h>
  54 #include <sys/kthread.h>
  55 #include <sys/limits.h>
  56 #include <sys/mbuf.h>
  57 #include <sys/md5.h>
  58 #include <sys/random.h>
  59 #include <sys/refcount.h>
  60 #include <sys/socket.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/taskqueue.h>
  63 #include <sys/ucred.h>
  64
  65 #include <net/if.h>
  66 #include <net/if_var.h>
  67 #include <net/if_types.h>
  68 #include <net/if_vlan_var.h>
  69 #include <net/route.h>
  70 #include <net/radix_mpath.h>
  71 #include <net/vnet.h>
  72
  73 #include <net/pfvar.h>
  74 #include <net/if_pflog.h>
  75 #include <net/if_pfsync.h>
  76
  77 #include <netinet/in_pcb.h>
  78 #include <netinet/in_var.h>
  79 #include <netinet/in_fib.h>
  80 #include <netinet/ip.h>
  81 #include <netinet/ip_fw.h>
  82 #include <netinet/ip_icmp.h>
  83 #include <netinet/icmp_var.h>
  84 #include <netinet/ip_var.h>
  85 #include <netinet/tcp.h>
  86 #include <netinet/tcp_fsm.h>
  87 #include <netinet/tcp_seq.h>
  88 #include <netinet/tcp_timer.h>
  89 #include <netinet/tcp_var.h>
  90 #include <netinet/udp.h>
  91 #include <netinet/udp_var.h>
  92
  93 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  94
  95 #ifdef INET6
  96 #include <netinet/ip6.h>
  97 #include <netinet/icmp6.h>
  98 #include <netinet6/nd6.h>
  99 #include <netinet6/ip6_var.h>
 100 #include <netinet6/in6_pcb.h>
 101 #include <netinet6/in6_fib.h>
 102 #include <netinet6/scope6_var.h>
 103 #endif /* INET6 */
 104
 105 #include <machine/in_cksum.h>
 106 #include <security/mac/mac_framework.h>
 107
 108 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 109
 110 /*
 111  * Global variables
 112  */
 113
 114 /* state tables */
 115 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 116 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 117 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 118 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 119 VNET_DEFINE(struct pf_kstatus,           pf_status);
 120
 121 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 122 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 123 VNET_DEFINE(int,                         altqs_inactive_open);
 124 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 125
 126 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 127 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 128 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 129 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 130 VNET_DEFINE(int,                         pf_tcp_secret_init);
 131 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 132 VNET_DEFINE(int,                         pf_tcp_iss_off);
 133 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 134 VNET_DECLARE(int,                        pf_vnet_active);
 135 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 136
 137 static VNET_DEFINE(uint32_t, pf_purge_idx);
 138 #define V_pf_purge_idx  VNET(pf_purge_idx)
 139
 140 /*
 141  * Queue for pf_intr() sends.
 142  */
 143 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 144 struct pf_send_entry {
 145         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 146         struct mbuf                     *pfse_m;
 147         enum {
 148                 PFSE_IP,
 149                 PFSE_IP6,
 150                 PFSE_ICMP,
 151                 PFSE_ICMP6,
 152         }                               pfse_type;
 153         struct {
 154                 int             type;
 155                 int             code;
 156                 int             mtu;
 157         } icmpopts;
 158 };
 159
 160 STAILQ_HEAD(pf_send_head, pf_send_entry);
 161 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 162 #define V_pf_sendqueue  VNET(pf_sendqueue)
 163
 164 static struct mtx pf_sendqueue_mtx;
 165 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 166 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 167 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 168
 169 /*
 170  * Queue for pf_overload_task() tasks.
 171  */
 172 struct pf_overload_entry {
 173         SLIST_ENTRY(pf_overload_entry)  next;
 174         struct pf_addr                  addr;
 175         sa_family_t                     af;
 176         uint8_t                         dir;
 177         struct pf_rule                  *rule;
 178 };
 179
 180 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 181 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 182 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 183 static VNET_DEFINE(struct task, pf_overloadtask);
 184 #define V_pf_overloadtask       VNET(pf_overloadtask)
 185
 186 static struct mtx pf_overloadqueue_mtx;
 187 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 188     "pf overload/flush queue", MTX_DEF);
 189 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 190 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 191
 192 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 193 struct mtx pf_unlnkdrules_mtx;
 194 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 195     MTX_DEF);
 196
 197 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 198 #define V_pf_sources_z  VNET(pf_sources_z)
 199 uma_zone_t              pf_mtag_z;
 200 VNET_DEFINE(uma_zone_t,  pf_state_z);
 201 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 202
 203 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 204 #define PFID_CPUBITS    8
 205 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 206 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 207 #define PFID_MAXID      (~PFID_CPUMASK)
 208 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 209
 210 static void              pf_src_tree_remove_state(struct pf_state *);
 211 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 212                             u_int32_t);
 213 static void              pf_add_threshold(struct pf_threshold *);
 214 static int               pf_check_threshold(struct pf_threshold *);
 215
 216 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 217                             u_int16_t *, u_int16_t *, struct pf_addr *,
 218                             u_int16_t, u_int8_t, sa_family_t);
 219 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 220                             struct tcphdr *, struct pf_state_peer *);
 221 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 222                             struct pf_addr *, struct pf_addr *, u_int16_t,
 223                             u_int16_t *, u_int16_t *, u_int16_t *,
 224                             u_int16_t *, u_int8_t, sa_family_t);
 225 static void              pf_send_tcp(struct mbuf *,
 226                             const struct pf_rule *, sa_family_t,
 227                             const struct pf_addr *, const struct pf_addr *,
 228                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 229                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 230                             u_int16_t, struct ifnet *);
 231 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 232                             sa_family_t, struct pf_rule *);
 233 static void              pf_detach_state(struct pf_state *);
 234 static int               pf_state_key_attach(struct pf_state_key *,
 235                             struct pf_state_key *, struct pf_state *);
 236 static void              pf_state_key_detach(struct pf_state *, int);
 237 static int               pf_state_key_ctor(void *, int, void *, int);
 238 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 239 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 240                             int, struct pfi_kif *, struct mbuf *, int,
 241                             struct pf_pdesc *, struct pf_rule **,
 242                             struct pf_ruleset **, struct inpcb *);
 243 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 244                             struct pf_rule *, struct pf_pdesc *,
 245                             struct pf_src_node *, struct pf_state_key *,
 246                             struct pf_state_key *, struct mbuf *, int,
 247                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 248                             struct pf_state **, int, u_int16_t, u_int16_t,
 249                             int);
 250 static int               pf_test_fragment(struct pf_rule **, int,
 251                             struct pfi_kif *, struct mbuf *, void *,
 252                             struct pf_pdesc *, struct pf_rule **,
 253                             struct pf_ruleset **);
 254 static int               pf_tcp_track_full(struct pf_state_peer *,
 255                             struct pf_state_peer *, struct pf_state **,
 256                             struct pfi_kif *, struct mbuf *, int,
 257                             struct pf_pdesc *, u_short *, int *);
 258 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 259                             struct pf_state_peer *, struct pf_state **,
 260                             struct pf_pdesc *, u_short *);
 261 static int               pf_test_state_tcp(struct pf_state **, int,
 262                             struct pfi_kif *, struct mbuf *, int,
 263                             void *, struct pf_pdesc *, u_short *);
 264 static int               pf_test_state_udp(struct pf_state **, int,
 265                             struct pfi_kif *, struct mbuf *, int,
 266                             void *, struct pf_pdesc *);
 267 static int               pf_test_state_icmp(struct pf_state **, int,
 268                             struct pfi_kif *, struct mbuf *, int,
 269                             void *, struct pf_pdesc *, u_short *);
 270 static int               pf_test_state_other(struct pf_state **, int,
 271                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 272 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 273                             sa_family_t);
 274 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 275                             sa_family_t);
 276 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 277                                 int, u_int16_t);
 278 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 279                             u_int8_t, sa_family_t);
 280 static void              pf_print_state_parts(struct pf_state *,
 281                             struct pf_state_key *, struct pf_state_key *);
 282 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 283                             struct pf_addr_wrap *);
 284 static struct pf_state  *pf_find_state(struct pfi_kif *,
 285                             struct pf_state_key_cmp *, u_int);
 286 static int               pf_src_connlimit(struct pf_state **);
 287 static void              pf_overload_task(void *v, int pending);
 288 static int               pf_insert_src_node(struct pf_src_node **,
 289                             struct pf_rule *, struct pf_addr *, sa_family_t);
 290 static u_int             pf_purge_expired_states(u_int, int);
 291 static void              pf_purge_unlinked_rules(void);
 292 static int               pf_mtag_uminit(void *, int, int);
 293 static void              pf_mtag_free(struct m_tag *);
 294 #ifdef INET
 295 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 296                             struct ifnet *, struct pf_state *,
 297                             struct pf_pdesc *);
 298 #endif /* INET */
 299 #ifdef INET6
 300 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 301                             struct pf_addr *, u_int8_t);
 302 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 303                             struct ifnet *, struct pf_state *,
 304                             struct pf_pdesc *);
 305 #endif /* INET6 */
 306
 307 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 308
 309 extern int pf_end_threads;
 310 extern struct proc *pf_purge_proc;
 311
 312 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 313
 314 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 315                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 316
 317 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 318         do {                                                            \
 319                 (s) = pf_find_state((i), (k), (d));                     \
 320                 if ((s) == NULL)                                        \
 321                         return (PF_DROP);                               \
 322                 if (PACKET_LOOPED(pd))                                  \
 323                         return (PF_PASS);                               \
 324                 if ((d) == PF_OUT &&                                    \
 325                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 326                     (s)->rule.ptr->direction == PF_OUT) ||              \
 327                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 328                     (s)->rule.ptr->direction == PF_IN)) &&              \
 329                     (s)->rt_kif != NULL &&                              \
 330                     (s)->rt_kif != (i))                                 \
 331                         return (PF_PASS);                               \
 332         } while (0)
 333
 334 #define BOUND_IFACE(r, k) \
 335         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 336
 337 #define STATE_INC_COUNTERS(s)                                           \
 338         do {                                                            \
 339                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 340                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 341                 if (s->anchor.ptr != NULL) {                            \
 342                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 343                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 344                 }                                                       \
 345                 if (s->nat_rule.ptr != NULL) {                          \
 346                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 347                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 348                 }                                                       \
 349         } while (0)
 350
 351 #define STATE_DEC_COUNTERS(s)                                           \
 352         do {                                                            \
 353                 if (s->nat_rule.ptr != NULL)                            \
 354                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 355                 if (s->anchor.ptr != NULL)                              \
 356                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 357                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 358         } while (0)
 359
 360 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 361 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 362 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 363 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 364
 365 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 366
 367 u_long  pf_hashmask;
 368 u_long  pf_srchashmask;
 369 static u_long   pf_hashsize;
 370 static u_long   pf_srchashsize;
 371
 372 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 373     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 374 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 375     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 376
 377 VNET_DEFINE(void *, pf_swi_cookie);
 378
 379 VNET_DEFINE(uint32_t, pf_hashseed);
 380 #define V_pf_hashseed   VNET(pf_hashseed)
 381
 382 int
 383 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 384 {
 385
 386         switch (af) {
 387 #ifdef INET
 388         case AF_INET:
 389                 if (a->addr32[0] > b->addr32[0])
 390                         return (1);
 391                 if (a->addr32[0] < b->addr32[0])
 392                         return (-1);
 393                 break;
 394 #endif /* INET */
 395 #ifdef INET6
 396         case AF_INET6:
 397                 if (a->addr32[3] > b->addr32[3])
 398                         return (1);
 399                 if (a->addr32[3] < b->addr32[3])
 400                         return (-1);
 401                 if (a->addr32[2] > b->addr32[2])
 402                         return (1);
 403                 if (a->addr32[2] < b->addr32[2])
 404                         return (-1);
 405                 if (a->addr32[1] > b->addr32[1])
 406                         return (1);
 407                 if (a->addr32[1] < b->addr32[1])
 408                         return (-1);
 409                 if (a->addr32[0] > b->addr32[0])
 410                         return (1);
 411                 if (a->addr32[0] < b->addr32[0])
 412                         return (-1);
 413                 break;
 414 #endif /* INET6 */
 415         default:
 416                 panic("%s: unknown address family %u", __func__, af);
 417         }
 418         return (0);
 419 }
 420
 421 static __inline uint32_t
 422 pf_hashkey(struct pf_state_key *sk)
 423 {
 424         uint32_t h;
 425
 426         h = murmur3_32_hash32((uint32_t *)sk,
 427             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 428             V_pf_hashseed);
 429
 430         return (h & pf_hashmask);
 431 }
 432
 433 static __inline uint32_t
 434 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 435 {
 436         uint32_t h;
 437
 438         switch (af) {
 439         case AF_INET:
 440                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 441                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 442                 break;
 443         case AF_INET6:
 444                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 445                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 446                 break;
 447         default:
 448                 panic("%s: unknown address family %u", __func__, af);
 449         }
 450
 451         return (h & pf_srchashmask);
 452 }
 453
 454 #ifdef ALTQ
 455 static int
 456 pf_state_hash(struct pf_state *s)
 457 {
 458         u_int32_t hv = (intptr_t)s / sizeof(*s);
 459
 460         hv ^= crc32(&s->src, sizeof(s->src));
 461         hv ^= crc32(&s->dst, sizeof(s->dst));
 462         if (hv == 0)
 463                 hv = 1;
 464         return (hv);
 465 }
 466 #endif
 467
 468 #ifdef INET6
 469 void
 470 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 471 {
 472         switch (af) {
 473 #ifdef INET
 474         case AF_INET:
 475                 dst->addr32[0] = src->addr32[0];
 476                 break;
 477 #endif /* INET */
 478         case AF_INET6:
 479                 dst->addr32[0] = src->addr32[0];
 480                 dst->addr32[1] = src->addr32[1];
 481                 dst->addr32[2] = src->addr32[2];
 482                 dst->addr32[3] = src->addr32[3];
 483                 break;
 484         }
 485 }
 486 #endif /* INET6 */
 487
 488 static void
 489 pf_init_threshold(struct pf_threshold *threshold,
 490     u_int32_t limit, u_int32_t seconds)
 491 {
 492         threshold->limit = limit * PF_THRESHOLD_MULT;
 493         threshold->seconds = seconds;
 494         threshold->count = 0;
 495         threshold->last = time_uptime;
 496 }
 497
 498 static void
 499 pf_add_threshold(struct pf_threshold *threshold)
 500 {
 501         u_int32_t t = time_uptime, diff = t - threshold->last;
 502
 503         if (diff >= threshold->seconds)
 504                 threshold->count = 0;
 505         else
 506                 threshold->count -= threshold->count * diff /
 507                     threshold->seconds;
 508         threshold->count += PF_THRESHOLD_MULT;
 509         threshold->last = t;
 510 }
 511
 512 static int
 513 pf_check_threshold(struct pf_threshold *threshold)
 514 {
 515         return (threshold->count > threshold->limit);
 516 }
 517
 518 static int
 519 pf_src_connlimit(struct pf_state **state)
 520 {
 521         struct pf_overload_entry *pfoe;
 522         int bad = 0;
 523
 524         PF_STATE_LOCK_ASSERT(*state);
 525
 526         (*state)->src_node->conn++;
 527         (*state)->src.tcp_est = 1;
 528         pf_add_threshold(&(*state)->src_node->conn_rate);
 529
 530         if ((*state)->rule.ptr->max_src_conn &&
 531             (*state)->rule.ptr->max_src_conn <
 532             (*state)->src_node->conn) {
 533                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 534                 bad++;
 535         }
 536
 537         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 538             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 539                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 540                 bad++;
 541         }
 542
 543         if (!bad)
 544                 return (0);
 545
 546         /* Kill this state. */
 547         (*state)->timeout = PFTM_PURGE;
 548         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 549
 550         if ((*state)->rule.ptr->overload_tbl == NULL)
 551                 return (1);
 552
 553         /* Schedule overloading and flushing task. */
 554         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 555         if (pfoe == NULL)
 556                 return (1);     /* too bad :( */
 557
 558         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 559         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 560         pfoe->rule = (*state)->rule.ptr;
 561         pfoe->dir = (*state)->direction;
 562         PF_OVERLOADQ_LOCK();
 563         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 564         PF_OVERLOADQ_UNLOCK();
 565         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 566
 567         return (1);
 568 }
 569
 570 static void
 571 pf_overload_task(void *v, int pending)
 572 {
 573         struct pf_overload_head queue;
 574         struct pfr_addr p;
 575         struct pf_overload_entry *pfoe, *pfoe1;
 576         uint32_t killed = 0;
 577
 578         CURVNET_SET((struct vnet *)v);
 579
 580         PF_OVERLOADQ_LOCK();
 581         queue = V_pf_overloadqueue;
 582         SLIST_INIT(&V_pf_overloadqueue);
 583         PF_OVERLOADQ_UNLOCK();
 584
 585         bzero(&p, sizeof(p));
 586         SLIST_FOREACH(pfoe, &queue, next) {
 587                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 588                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 589                         printf("%s: blocking address ", __func__);
 590                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 591                         printf("\n");
 592                 }
 593
 594                 p.pfra_af = pfoe->af;
 595                 switch (pfoe->af) {
 596 #ifdef INET
 597                 case AF_INET:
 598                         p.pfra_net = 32;
 599                         p.pfra_ip4addr = pfoe->addr.v4;
 600                         break;
 601 #endif
 602 #ifdef INET6
 603                 case AF_INET6:
 604                         p.pfra_net = 128;
 605                         p.pfra_ip6addr = pfoe->addr.v6;
 606                         break;
 607 #endif
 608                 }
 609
 610                 PF_RULES_WLOCK();
 611                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 612                 PF_RULES_WUNLOCK();
 613         }
 614
 615         /*
 616          * Remove those entries, that don't need flushing.
 617          */
 618         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 619                 if (pfoe->rule->flush == 0) {
 620                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 621                         free(pfoe, M_PFTEMP);
 622                 } else
 623                         counter_u64_add(
 624                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 625
 626         /* If nothing to flush, return. */
 627         if (SLIST_EMPTY(&queue)) {
 628                 CURVNET_RESTORE();
 629                 return;
 630         }
 631
 632         for (int i = 0; i <= pf_hashmask; i++) {
 633                 struct pf_idhash *ih = &V_pf_idhash[i];
 634                 struct pf_state_key *sk;
 635                 struct pf_state *s;
 636
 637                 PF_HASHROW_LOCK(ih);
 638                 LIST_FOREACH(s, &ih->states, entry) {
 639                     sk = s->key[PF_SK_WIRE];
 640                     SLIST_FOREACH(pfoe, &queue, next)
 641                         if (sk->af == pfoe->af &&
 642                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 643                             pfoe->rule == s->rule.ptr) &&
 644                             ((pfoe->dir == PF_OUT &&
 645                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 646                             (pfoe->dir == PF_IN &&
 647                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 648                                 s->timeout = PFTM_PURGE;
 649                                 s->src.state = s->dst.state = TCPS_CLOSED;
 650                                 killed++;
 651                         }
 652                 }
 653                 PF_HASHROW_UNLOCK(ih);
 654         }
 655         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 656                 free(pfoe, M_PFTEMP);
 657         if (V_pf_status.debug >= PF_DEBUG_MISC)
 658                 printf("%s: %u states killed", __func__, killed);
 659
 660         CURVNET_RESTORE();
 661 }
 662
 663 /*
 664  * Can return locked on failure, so that we can consistently
 665  * allocate and insert a new one.
 666  */
 667 struct pf_src_node *
 668 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 669         int returnlocked)
 670 {
 671         struct pf_srchash *sh;
 672         struct pf_src_node *n;
 673
 674         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 675
 676         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 677         PF_HASHROW_LOCK(sh);
 678         LIST_FOREACH(n, &sh->nodes, entry)
 679                 if (n->rule.ptr == rule && n->af == af &&
 680                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 681                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 682                         break;
 683         if (n != NULL) {
 684                 n->states++;
 685                 PF_HASHROW_UNLOCK(sh);
 686         } else if (returnlocked == 0)
 687                 PF_HASHROW_UNLOCK(sh);
 688
 689         return (n);
 690 }
 691
 692 static int
 693 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 694     struct pf_addr *src, sa_family_t af)
 695 {
 696
 697         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 698             rule->rpool.opts & PF_POOL_STICKYADDR),
 699             ("%s for non-tracking rule %p", __func__, rule));
 700
 701         if (*sn == NULL)
 702                 *sn = pf_find_src_node(src, rule, af, 1);
 703
 704         if (*sn == NULL) {
 705                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 706
 707                 PF_HASHROW_ASSERT(sh);
 708
 709                 if (!rule->max_src_nodes ||
 710                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 711                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 712                 else
 713                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 714                             1);
 715                 if ((*sn) == NULL) {
 716                         PF_HASHROW_UNLOCK(sh);
 717                         return (-1);
 718                 }
 719
 720                 pf_init_threshold(&(*sn)->conn_rate,
 721                     rule->max_src_conn_rate.limit,
 722                     rule->max_src_conn_rate.seconds);
 723
 724                 (*sn)->af = af;
 725                 (*sn)->rule.ptr = rule;
 726                 PF_ACPY(&(*sn)->addr, src, af);
 727                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 728                 (*sn)->creation = time_uptime;
 729                 (*sn)->ruletype = rule->action;
 730                 (*sn)->states = 1;
 731                 if ((*sn)->rule.ptr != NULL)
 732                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 733                 PF_HASHROW_UNLOCK(sh);
 734                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 735         } else {
 736                 if (rule->max_src_states &&
 737                     (*sn)->states >= rule->max_src_states) {
 738                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 739                             1);
 740                         return (-1);
 741                 }
 742         }
 743         return (0);
 744 }
 745
 746 void
 747 pf_unlink_src_node(struct pf_src_node *src)
 748 {
 749
 750         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 751         LIST_REMOVE(src, entry);
 752         if (src->rule.ptr)
 753                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 754 }
 755
 756 u_int
 757 pf_free_src_nodes(struct pf_src_node_list *head)
 758 {
 759         struct pf_src_node *sn, *tmp;
 760         u_int count = 0;
 761
 762         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 763                 uma_zfree(V_pf_sources_z, sn);
 764                 count++;
 765         }
 766
 767         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 768
 769         return (count);
 770 }
 771
 772 void
 773 pf_mtag_initialize()
 774 {
 775
 776         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 777             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 778             UMA_ALIGN_PTR, 0);
 779 }
 780
 781 /* Per-vnet data storage structures initialization. */
 782 void
 783 pf_initialize()
 784 {
 785         struct pf_keyhash       *kh;
 786         struct pf_idhash        *ih;
 787         struct pf_srchash       *sh;
 788         u_int i;
 789
 790         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 791                 pf_hashsize = PF_HASHSIZ;
 792         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 793                 pf_srchashsize = PF_HASHSIZ / 4;
 794
 795         V_pf_hashseed = arc4random();
 796
 797         /* States and state keys storage. */
 798         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 799             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 800         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 801         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 802         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 803
 804         V_pf_state_key_z = uma_zcreate("pf state keys",
 805             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 806             UMA_ALIGN_PTR, 0);
 807         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 808             M_PFHASH, M_WAITOK | M_ZERO);
 809         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 810             M_PFHASH, M_WAITOK | M_ZERO);
 811         pf_hashmask = pf_hashsize - 1;
 812         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 813             i++, kh++, ih++) {
 814                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 815                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 816         }
 817
 818         /* Source nodes. */
 819         V_pf_sources_z = uma_zcreate("pf source nodes",
 820             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 821             0);
 822         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 823         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 824         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 825         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 826           M_PFHASH, M_WAITOK|M_ZERO);
 827         pf_srchashmask = pf_srchashsize - 1;
 828         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 829                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 830
 831         /* ALTQ */
 832         TAILQ_INIT(&V_pf_altqs[0]);
 833         TAILQ_INIT(&V_pf_altqs[1]);
 834         TAILQ_INIT(&V_pf_pabuf);
 835         V_pf_altqs_active = &V_pf_altqs[0];
 836         V_pf_altqs_inactive = &V_pf_altqs[1];
 837
 838         /* Send & overload+flush queues. */
 839         STAILQ_INIT(&V_pf_sendqueue);
 840         SLIST_INIT(&V_pf_overloadqueue);
 841         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 842
 843         /* Unlinked, but may be referenced rules. */
 844         TAILQ_INIT(&V_pf_unlinked_rules);
 845 }
 846
 847 void
 848 pf_mtag_cleanup()
 849 {
 850
 851         uma_zdestroy(pf_mtag_z);
 852 }
 853
 854 void
 855 pf_cleanup()
 856 {
 857         struct pf_keyhash       *kh;
 858         struct pf_idhash        *ih;
 859         struct pf_srchash       *sh;
 860         struct pf_send_entry    *pfse, *next;
 861         u_int i;
 862
 863         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 864             i++, kh++, ih++) {
 865                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 866                     __func__));
 867                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 868                     __func__));
 869                 mtx_destroy(&kh->lock);
 870                 mtx_destroy(&ih->lock);
 871         }
 872         free(V_pf_keyhash, M_PFHASH);
 873         free(V_pf_idhash, M_PFHASH);
 874
 875         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 876                 KASSERT(LIST_EMPTY(&sh->nodes),
 877                     ("%s: source node hash not empty", __func__));
 878                 mtx_destroy(&sh->lock);
 879         }
 880         free(V_pf_srchash, M_PFHASH);
 881
 882         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 883                 m_freem(pfse->pfse_m);
 884                 free(pfse, M_PFTEMP);
 885         }
 886
 887         uma_zdestroy(V_pf_sources_z);
 888         uma_zdestroy(V_pf_state_z);
 889         uma_zdestroy(V_pf_state_key_z);
 890 }
 891
 892 static int
 893 pf_mtag_uminit(void *mem, int size, int how)
 894 {
 895         struct m_tag *t;
 896
 897         t = (struct m_tag *)mem;
 898         t->m_tag_cookie = MTAG_ABI_COMPAT;
 899         t->m_tag_id = PACKET_TAG_PF;
 900         t->m_tag_len = sizeof(struct pf_mtag);
 901         t->m_tag_free = pf_mtag_free;
 902
 903         return (0);
 904 }
 905
 906 static void
 907 pf_mtag_free(struct m_tag *t)
 908 {
 909
 910         uma_zfree(pf_mtag_z, t);
 911 }
 912
 913 struct pf_mtag *
 914 pf_get_mtag(struct mbuf *m)
 915 {
 916         struct m_tag *mtag;
 917
 918         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 919                 return ((struct pf_mtag *)(mtag + 1));
 920
 921         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 922         if (mtag == NULL)
 923                 return (NULL);
 924         bzero(mtag + 1, sizeof(struct pf_mtag));
 925         m_tag_prepend(m, mtag);
 926
 927         return ((struct pf_mtag *)(mtag + 1));
 928 }
 929
 930 static int
 931 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 932     struct pf_state *s)
 933 {
 934         struct pf_keyhash       *khs, *khw, *kh;
 935         struct pf_state_key     *sk, *cur;
 936         struct pf_state         *si, *olds = NULL;
 937         int idx;
 938
 939         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 940         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 941         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 942
 943         /*
 944          * We need to lock hash slots of both keys. To avoid deadlock
 945          * we always lock the slot with lower address first. Unlock order
 946          * isn't important.
 947          *
 948          * We also need to lock ID hash slot before dropping key
 949          * locks. On success we return with ID hash slot locked.
 950          */
 951
 952         if (skw == sks) {
 953                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 954                 PF_HASHROW_LOCK(khs);
 955         } else {
 956                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 957                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 958                 if (khs == khw) {
 959                         PF_HASHROW_LOCK(khs);
 960                 } else if (khs < khw) {
 961                         PF_HASHROW_LOCK(khs);
 962                         PF_HASHROW_LOCK(khw);
 963                 } else {
 964                         PF_HASHROW_LOCK(khw);
 965                         PF_HASHROW_LOCK(khs);
 966                 }
 967         }
 968
 969 #define KEYS_UNLOCK()   do {                    \
 970         if (khs != khw) {                       \
 971                 PF_HASHROW_UNLOCK(khs);         \
 972                 PF_HASHROW_UNLOCK(khw);         \
 973         } else                                  \
 974                 PF_HASHROW_UNLOCK(khs);         \
 975 } while (0)
 976
 977         /*
 978          * First run: start with wire key.
 979          */
 980         sk = skw;
 981         kh = khw;
 982         idx = PF_SK_WIRE;
 983
 984 keyattach:
 985         LIST_FOREACH(cur, &kh->keys, entry)
 986                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 987                         break;
 988
 989         if (cur != NULL) {
 990                 /* Key exists. Check for same kif, if none, add to key. */
 991                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 992                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 993
 994                         PF_HASHROW_LOCK(ih);
 995                         if (si->kif == s->kif &&
 996                             si->direction == s->direction) {
 997                                 if (sk->proto == IPPROTO_TCP &&
 998                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 999                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1000                                         /*
1001                                          * New state matches an old >FIN_WAIT_2
1002                                          * state. We can't drop key hash locks,
1003                                          * thus we can't unlink it properly.
1004                                          *
1005                                          * As a workaround we drop it into
1006                                          * TCPS_CLOSED state, schedule purge
1007                                          * ASAP and push it into the very end
1008                                          * of the slot TAILQ, so that it won't
1009                                          * conflict with our new state.
1010                                          */
1011                                         si->src.state = si->dst.state =
1012                                             TCPS_CLOSED;
1013                                         si->timeout = PFTM_PURGE;
1014                                         olds = si;
1015                                 } else {
1016                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1017                                                 printf("pf: %s key attach "
1018                                                     "failed on %s: ",
1019                                                     (idx == PF_SK_WIRE) ?
1020                                                     "wire" : "stack",
1021                                                     s->kif->pfik_name);
1022                                                 pf_print_state_parts(s,
1023                                                     (idx == PF_SK_WIRE) ?
1024                                                     sk : NULL,
1025                                                     (idx == PF_SK_STACK) ?
1026                                                     sk : NULL);
1027                                                 printf(", existing: ");
1028                                                 pf_print_state_parts(si,
1029                                                     (idx == PF_SK_WIRE) ?
1030                                                     sk : NULL,
1031                                                     (idx == PF_SK_STACK) ?
1032                                                     sk : NULL);
1033                                                 printf("\n");
1034                                         }
1035                                         PF_HASHROW_UNLOCK(ih);
1036                                         KEYS_UNLOCK();
1037                                         uma_zfree(V_pf_state_key_z, sk);
1038                                         if (idx == PF_SK_STACK)
1039                                                 pf_detach_state(s);
1040                                         return (EEXIST); /* collision! */
1041                                 }
1042                         }
1043                         PF_HASHROW_UNLOCK(ih);
1044                 }
1045                 uma_zfree(V_pf_state_key_z, sk);
1046                 s->key[idx] = cur;
1047         } else {
1048                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1049                 s->key[idx] = sk;
1050         }
1051
1052 stateattach:
1053         /* List is sorted, if-bound states before floating. */
1054         if (s->kif == V_pfi_all)
1055                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1056         else
1057                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1058
1059         if (olds) {
1060                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1061                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1062                     key_list[idx]);
1063                 olds = NULL;
1064         }
1065
1066         /*
1067          * Attach done. See how should we (or should not?)
1068          * attach a second key.
1069          */
1070         if (sks == skw) {
1071                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1072                 idx = PF_SK_STACK;
1073                 sks = NULL;
1074                 goto stateattach;
1075         } else if (sks != NULL) {
1076                 /*
1077                  * Continue attaching with stack key.
1078                  */
1079                 sk = sks;
1080                 kh = khs;
1081                 idx = PF_SK_STACK;
1082                 sks = NULL;
1083                 goto keyattach;
1084         }
1085
1086         PF_STATE_LOCK(s);
1087         KEYS_UNLOCK();
1088
1089         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1090             ("%s failure", __func__));
1091
1092         return (0);
1093 #undef  KEYS_UNLOCK
1094 }
1095
1096 static void
1097 pf_detach_state(struct pf_state *s)
1098 {
1099         struct pf_state_key *sks = s->key[PF_SK_STACK];
1100         struct pf_keyhash *kh;
1101
1102         if (sks != NULL) {
1103                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1104                 PF_HASHROW_LOCK(kh);
1105                 if (s->key[PF_SK_STACK] != NULL)
1106                         pf_state_key_detach(s, PF_SK_STACK);
1107                 /*
1108                  * If both point to same key, then we are done.
1109                  */
1110                 if (sks == s->key[PF_SK_WIRE]) {
1111                         pf_state_key_detach(s, PF_SK_WIRE);
1112                         PF_HASHROW_UNLOCK(kh);
1113                         return;
1114                 }
1115                 PF_HASHROW_UNLOCK(kh);
1116         }
1117
1118         if (s->key[PF_SK_WIRE] != NULL) {
1119                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1120                 PF_HASHROW_LOCK(kh);
1121                 if (s->key[PF_SK_WIRE] != NULL)
1122                         pf_state_key_detach(s, PF_SK_WIRE);
1123                 PF_HASHROW_UNLOCK(kh);
1124         }
1125 }
1126
1127 static void
1128 pf_state_key_detach(struct pf_state *s, int idx)
1129 {
1130         struct pf_state_key *sk = s->key[idx];
1131 #ifdef INVARIANTS
1132         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1133
1134         PF_HASHROW_ASSERT(kh);
1135 #endif
1136         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1137         s->key[idx] = NULL;
1138
1139         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1140                 LIST_REMOVE(sk, entry);
1141                 uma_zfree(V_pf_state_key_z, sk);
1142         }
1143 }
1144
1145 static int
1146 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1147 {
1148         struct pf_state_key *sk = mem;
1149
1150         bzero(sk, sizeof(struct pf_state_key_cmp));
1151         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1152         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1153
1154         return (0);
1155 }
1156
1157 struct pf_state_key *
1158 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1159         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1160 {
1161         struct pf_state_key *sk;
1162
1163         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1164         if (sk == NULL)
1165                 return (NULL);
1166
1167         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1168         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1169         sk->port[pd->sidx] = sport;
1170         sk->port[pd->didx] = dport;
1171         sk->proto = pd->proto;
1172         sk->af = pd->af;
1173
1174         return (sk);
1175 }
1176
1177 struct pf_state_key *
1178 pf_state_key_clone(struct pf_state_key *orig)
1179 {
1180         struct pf_state_key *sk;
1181
1182         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1183         if (sk == NULL)
1184                 return (NULL);
1185
1186         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1187
1188         return (sk);
1189 }
1190
1191 int
1192 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1193     struct pf_state_key *sks, struct pf_state *s)
1194 {
1195         struct pf_idhash *ih;
1196         struct pf_state *cur;
1197         int error;
1198
1199         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1200             ("%s: sks not pristine", __func__));
1201         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1202             ("%s: skw not pristine", __func__));
1203         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1204
1205         s->kif = kif;
1206
1207         if (s->id == 0 && s->creatorid == 0) {
1208                 /* XXX: should be atomic, but probability of collision low */
1209                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1210                         V_pf_stateid[curcpu] = 1;
1211                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1212                 s->id = htobe64(s->id);
1213                 s->creatorid = V_pf_status.hostid;
1214         }
1215
1216         /* Returns with ID locked on success. */
1217         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1218                 return (error);
1219
1220         ih = &V_pf_idhash[PF_IDHASH(s)];
1221         PF_HASHROW_ASSERT(ih);
1222         LIST_FOREACH(cur, &ih->states, entry)
1223                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1224                         break;
1225
1226         if (cur != NULL) {
1227                 PF_HASHROW_UNLOCK(ih);
1228                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1229                         printf("pf: state ID collision: "
1230                             "id: %016llx creatorid: %08x\n",
1231                             (unsigned long long)be64toh(s->id),
1232                             ntohl(s->creatorid));
1233                 }
1234                 pf_detach_state(s);
1235                 return (EEXIST);
1236         }
1237         LIST_INSERT_HEAD(&ih->states, s, entry);
1238         /* One for keys, one for ID hash. */
1239         refcount_init(&s->refs, 2);
1240
1241         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1242         if (pfsync_insert_state_ptr != NULL)
1243                 pfsync_insert_state_ptr(s);
1244
1245         /* Returns locked. */
1246         return (0);
1247 }
1248
1249 /*
1250  * Find state by ID: returns with locked row on success.
1251  */
1252 struct pf_state *
1253 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1254 {
1255         struct pf_idhash *ih;
1256         struct pf_state *s;
1257
1258         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1259
1260         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1261
1262         PF_HASHROW_LOCK(ih);
1263         LIST_FOREACH(s, &ih->states, entry)
1264                 if (s->id == id && s->creatorid == creatorid)
1265                         break;
1266
1267         if (s == NULL)
1268                 PF_HASHROW_UNLOCK(ih);
1269
1270         return (s);
1271 }
1272
1273 /*
1274  * Find state by key.
1275  * Returns with ID hash slot locked on success.
1276  */
1277 static struct pf_state *
1278 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1279 {
1280         struct pf_keyhash       *kh;
1281         struct pf_state_key     *sk;
1282         struct pf_state         *s;
1283         int idx;
1284
1285         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1286
1287         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1288
1289         PF_HASHROW_LOCK(kh);
1290         LIST_FOREACH(sk, &kh->keys, entry)
1291                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1292                         break;
1293         if (sk == NULL) {
1294                 PF_HASHROW_UNLOCK(kh);
1295                 return (NULL);
1296         }
1297
1298         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1299
1300         /* List is sorted, if-bound states before floating ones. */
1301         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1302                 if (s->kif == V_pfi_all || s->kif == kif) {
1303                         PF_STATE_LOCK(s);
1304                         PF_HASHROW_UNLOCK(kh);
1305                         if (s->timeout >= PFTM_MAX) {
1306                                 /*
1307                                  * State is either being processed by
1308                                  * pf_unlink_state() in an other thread, or
1309                                  * is scheduled for immediate expiry.
1310                                  */
1311                                 PF_STATE_UNLOCK(s);
1312                                 return (NULL);
1313                         }
1314                         return (s);
1315                 }
1316         PF_HASHROW_UNLOCK(kh);
1317
1318         return (NULL);
1319 }
1320
1321 struct pf_state *
1322 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1323 {
1324         struct pf_keyhash       *kh;
1325         struct pf_state_key     *sk;
1326         struct pf_state         *s, *ret = NULL;
1327         int                      idx, inout = 0;
1328
1329         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1330
1331         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1332
1333         PF_HASHROW_LOCK(kh);
1334         LIST_FOREACH(sk, &kh->keys, entry)
1335                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1336                         break;
1337         if (sk == NULL) {
1338                 PF_HASHROW_UNLOCK(kh);
1339                 return (NULL);
1340         }
1341         switch (dir) {
1342         case PF_IN:
1343                 idx = PF_SK_WIRE;
1344                 break;
1345         case PF_OUT:
1346                 idx = PF_SK_STACK;
1347                 break;
1348         case PF_INOUT:
1349                 idx = PF_SK_WIRE;
1350                 inout = 1;
1351                 break;
1352         default:
1353                 panic("%s: dir %u", __func__, dir);
1354         }
1355 second_run:
1356         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1357                 if (more == NULL) {
1358                         PF_HASHROW_UNLOCK(kh);
1359                         return (s);
1360                 }
1361
1362                 if (ret)
1363                         (*more)++;
1364                 else
1365                         ret = s;
1366         }
1367         if (inout == 1) {
1368                 inout = 0;
1369                 idx = PF_SK_STACK;
1370                 goto second_run;
1371         }
1372         PF_HASHROW_UNLOCK(kh);
1373
1374         return (ret);
1375 }
1376
1377 /* END state table stuff */
1378
1379 static void
1380 pf_send(struct pf_send_entry *pfse)
1381 {
1382
1383         PF_SENDQ_LOCK();
1384         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1385         PF_SENDQ_UNLOCK();
1386         swi_sched(V_pf_swi_cookie, 0);
1387 }
1388
1389 void
1390 pf_intr(void *v)
1391 {
1392         struct pf_send_head queue;
1393         struct pf_send_entry *pfse, *next;
1394
1395         CURVNET_SET((struct vnet *)v);
1396
1397         PF_SENDQ_LOCK();
1398         queue = V_pf_sendqueue;
1399         STAILQ_INIT(&V_pf_sendqueue);
1400         PF_SENDQ_UNLOCK();
1401
1402         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1403                 switch (pfse->pfse_type) {
1404 #ifdef INET
1405                 case PFSE_IP:
1406                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1407                         break;
1408                 case PFSE_ICMP:
1409                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1410                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1411                         break;
1412 #endif /* INET */
1413 #ifdef INET6
1414                 case PFSE_IP6:
1415                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1416                             NULL);
1417                         break;
1418                 case PFSE_ICMP6:
1419                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1420                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1421                         break;
1422 #endif /* INET6 */
1423                 default:
1424                         panic("%s: unknown type", __func__);
1425                 }
1426                 free(pfse, M_PFTEMP);
1427         }
1428         CURVNET_RESTORE();
1429 }
1430
1431 void
1432 pf_purge_thread(void *unused __unused)
1433 {
1434         VNET_ITERATOR_DECL(vnet_iter);
1435
1436         sx_xlock(&pf_end_lock);
1437         while (pf_end_threads == 0) {
1438                 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1439
1440                 VNET_LIST_RLOCK();
1441                 VNET_FOREACH(vnet_iter) {
1442                         CURVNET_SET(vnet_iter);
1443
1444
1445                         /* Wait until V_pf_default_rule is initialized. */
1446                         if (V_pf_vnet_active == 0) {
1447                                 CURVNET_RESTORE();
1448                                 continue;
1449                         }
1450
1451                         /*
1452                          *  Process 1/interval fraction of the state
1453                          * table every run.
1454                          */
1455                         V_pf_purge_idx =
1456                             pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1457                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1458
1459                         /*
1460                          * Purge other expired types every
1461                          * PFTM_INTERVAL seconds.
1462                          */
1463                         if (V_pf_purge_idx == 0) {
1464                                 /*
1465                                  * Order is important:
1466                                  * - states and src nodes reference rules
1467                                  * - states and rules reference kifs
1468                                  */
1469                                 pf_purge_expired_fragments();
1470                                 pf_purge_expired_src_nodes();
1471                                 pf_purge_unlinked_rules();
1472                                 pfi_kif_purge();
1473                         }
1474                         CURVNET_RESTORE();
1475                 }
1476                 VNET_LIST_RUNLOCK();
1477         }
1478
1479         pf_end_threads++;
1480         sx_xunlock(&pf_end_lock);
1481         kproc_exit(0);
1482 }
1483
1484 void
1485 pf_unload_vnet_purge(void)
1486 {
1487
1488         /*
1489          * To cleanse up all kifs and rules we need
1490          * two runs: first one clears reference flags,
1491          * then pf_purge_expired_states() doesn't
1492          * raise them, and then second run frees.
1493          */
1494         pf_purge_unlinked_rules();
1495         pfi_kif_purge();
1496
1497         /*
1498          * Now purge everything.
1499          */
1500         pf_purge_expired_states(0, pf_hashmask);
1501         pf_purge_fragments(UINT_MAX);
1502         pf_purge_expired_src_nodes();
1503
1504         /*
1505          * Now all kifs & rules should be unreferenced,
1506          * thus should be successfully freed.
1507          */
1508         pf_purge_unlinked_rules();
1509         pfi_kif_purge();
1510 }
1511
1512
1513 u_int32_t
1514 pf_state_expires(const struct pf_state *state)
1515 {
1516         u_int32_t       timeout;
1517         u_int32_t       start;
1518         u_int32_t       end;
1519         u_int32_t       states;
1520
1521         /* handle all PFTM_* > PFTM_MAX here */
1522         if (state->timeout == PFTM_PURGE)
1523                 return (time_uptime);
1524         KASSERT(state->timeout != PFTM_UNLINKED,
1525             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1526         KASSERT((state->timeout < PFTM_MAX),
1527             ("pf_state_expires: timeout > PFTM_MAX"));
1528         timeout = state->rule.ptr->timeout[state->timeout];
1529         if (!timeout)
1530                 timeout = V_pf_default_rule.timeout[state->timeout];
1531         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1532         if (start) {
1533                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1534                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1535         } else {
1536                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1537                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1538                 states = V_pf_status.states;
1539         }
1540         if (end && states > start && start < end) {
1541                 if (states < end)
1542                         return (state->expire + timeout * (end - states) /
1543                             (end - start));
1544                 else
1545                         return (time_uptime);
1546         }
1547         return (state->expire + timeout);
1548 }
1549
1550 void
1551 pf_purge_expired_src_nodes()
1552 {
1553         struct pf_src_node_list  freelist;
1554         struct pf_srchash       *sh;
1555         struct pf_src_node      *cur, *next;
1556         int i;
1557
1558         LIST_INIT(&freelist);
1559         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1560             PF_HASHROW_LOCK(sh);
1561             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1562                 if (cur->states == 0 && cur->expire <= time_uptime) {
1563                         pf_unlink_src_node(cur);
1564                         LIST_INSERT_HEAD(&freelist, cur, entry);
1565                 } else if (cur->rule.ptr != NULL)
1566                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1567             PF_HASHROW_UNLOCK(sh);
1568         }
1569
1570         pf_free_src_nodes(&freelist);
1571
1572         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1573 }
1574
1575 static void
1576 pf_src_tree_remove_state(struct pf_state *s)
1577 {
1578         struct pf_src_node *sn;
1579         struct pf_srchash *sh;
1580         uint32_t timeout;
1581
1582         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1583             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1584             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1585
1586         if (s->src_node != NULL) {
1587                 sn = s->src_node;
1588                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1589                 PF_HASHROW_LOCK(sh);
1590                 if (s->src.tcp_est)
1591                         --sn->conn;
1592                 if (--sn->states == 0)
1593                         sn->expire = time_uptime + timeout;
1594                 PF_HASHROW_UNLOCK(sh);
1595         }
1596         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1597                 sn = s->nat_src_node;
1598                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1599                 PF_HASHROW_LOCK(sh);
1600                 if (--sn->states == 0)
1601                         sn->expire = time_uptime + timeout;
1602                 PF_HASHROW_UNLOCK(sh);
1603         }
1604         s->src_node = s->nat_src_node = NULL;
1605 }
1606
1607 /*
1608  * Unlink and potentilly free a state. Function may be
1609  * called with ID hash row locked, but always returns
1610  * unlocked, since it needs to go through key hash locking.
1611  */
1612 int
1613 pf_unlink_state(struct pf_state *s, u_int flags)
1614 {
1615         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1616
1617         if ((flags & PF_ENTER_LOCKED) == 0)
1618                 PF_HASHROW_LOCK(ih);
1619         else
1620                 PF_HASHROW_ASSERT(ih);
1621
1622         if (s->timeout == PFTM_UNLINKED) {
1623                 /*
1624                  * State is being processed
1625                  * by pf_unlink_state() in
1626                  * an other thread.
1627                  */
1628                 PF_HASHROW_UNLOCK(ih);
1629                 return (0);     /* XXXGL: undefined actually */
1630         }
1631
1632         if (s->src.state == PF_TCPS_PROXY_DST) {
1633                 /* XXX wire key the right one? */
1634                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1635                     &s->key[PF_SK_WIRE]->addr[1],
1636                     &s->key[PF_SK_WIRE]->addr[0],
1637                     s->key[PF_SK_WIRE]->port[1],
1638                     s->key[PF_SK_WIRE]->port[0],
1639                     s->src.seqhi, s->src.seqlo + 1,
1640                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1641         }
1642
1643         LIST_REMOVE(s, entry);
1644         pf_src_tree_remove_state(s);
1645
1646         if (pfsync_delete_state_ptr != NULL)
1647                 pfsync_delete_state_ptr(s);
1648
1649         STATE_DEC_COUNTERS(s);
1650
1651         s->timeout = PFTM_UNLINKED;
1652
1653         PF_HASHROW_UNLOCK(ih);
1654
1655         pf_detach_state(s);
1656         /* pf_state_insert() initialises refs to 2, so we can never release the
1657          * last reference here, only in pf_release_state(). */
1658         (void)refcount_release(&s->refs);
1659
1660         return (pf_release_state(s));
1661 }
1662
1663 void
1664 pf_free_state(struct pf_state *cur)
1665 {
1666
1667         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1668         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1669             cur->timeout));
1670
1671         pf_normalize_tcp_cleanup(cur);
1672         uma_zfree(V_pf_state_z, cur);
1673         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1674 }
1675
1676 /*
1677  * Called only from pf_purge_thread(), thus serialized.
1678  */
1679 static u_int
1680 pf_purge_expired_states(u_int i, int maxcheck)
1681 {
1682         struct pf_idhash *ih;
1683         struct pf_state *s;
1684
1685         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1686
1687         /*
1688          * Go through hash and unlink states that expire now.
1689          */
1690         while (maxcheck > 0) {
1691
1692                 ih = &V_pf_idhash[i];
1693 relock:
1694                 PF_HASHROW_LOCK(ih);
1695                 LIST_FOREACH(s, &ih->states, entry) {
1696                         if (pf_state_expires(s) <= time_uptime) {
1697                                 V_pf_status.states -=
1698                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1699                                 goto relock;
1700                         }
1701                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1702                         if (s->nat_rule.ptr != NULL)
1703                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1704                         if (s->anchor.ptr != NULL)
1705                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1706                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1707                         if (s->rt_kif)
1708                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1709                 }
1710                 PF_HASHROW_UNLOCK(ih);
1711
1712                 /* Return when we hit end of hash. */
1713                 if (++i > pf_hashmask) {
1714                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1715                         return (0);
1716                 }
1717
1718                 maxcheck--;
1719         }
1720
1721         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1722
1723         return (i);
1724 }
1725
1726 static void
1727 pf_purge_unlinked_rules()
1728 {
1729         struct pf_rulequeue tmpq;
1730         struct pf_rule *r, *r1;
1731
1732         /*
1733          * If we have overloading task pending, then we'd
1734          * better skip purging this time. There is a tiny
1735          * probability that overloading task references
1736          * an already unlinked rule.
1737          */
1738         PF_OVERLOADQ_LOCK();
1739         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1740                 PF_OVERLOADQ_UNLOCK();
1741                 return;
1742         }
1743         PF_OVERLOADQ_UNLOCK();
1744
1745         /*
1746          * Do naive mark-and-sweep garbage collecting of old rules.
1747          * Reference flag is raised by pf_purge_expired_states()
1748          * and pf_purge_expired_src_nodes().
1749          *
1750          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1751          * use a temporary queue.
1752          */
1753         TAILQ_INIT(&tmpq);
1754         PF_UNLNKDRULES_LOCK();
1755         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1756                 if (!(r->rule_flag & PFRULE_REFS)) {
1757                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1758                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1759                 } else
1760                         r->rule_flag &= ~PFRULE_REFS;
1761         }
1762         PF_UNLNKDRULES_UNLOCK();
1763
1764         if (!TAILQ_EMPTY(&tmpq)) {
1765                 PF_RULES_WLOCK();
1766                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1767                         TAILQ_REMOVE(&tmpq, r, entries);
1768                         pf_free_rule(r);
1769                 }
1770                 PF_RULES_WUNLOCK();
1771         }
1772 }
1773
1774 void
1775 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1776 {
1777         switch (af) {
1778 #ifdef INET
1779         case AF_INET: {
1780                 u_int32_t a = ntohl(addr->addr32[0]);
1781                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1782                     (a>>8)&255, a&255);
1783                 if (p) {
1784                         p = ntohs(p);
1785                         printf(":%u", p);
1786                 }
1787                 break;
1788         }
1789 #endif /* INET */
1790 #ifdef INET6
1791         case AF_INET6: {
1792                 u_int16_t b;
1793                 u_int8_t i, curstart, curend, maxstart, maxend;
1794                 curstart = curend = maxstart = maxend = 255;
1795                 for (i = 0; i < 8; i++) {
1796                         if (!addr->addr16[i]) {
1797                                 if (curstart == 255)
1798                                         curstart = i;
1799                                 curend = i;
1800                         } else {
1801                                 if ((curend - curstart) >
1802                                     (maxend - maxstart)) {
1803                                         maxstart = curstart;
1804                                         maxend = curend;
1805                                 }
1806                                 curstart = curend = 255;
1807                         }
1808                 }
1809                 if ((curend - curstart) >
1810                     (maxend - maxstart)) {
1811                         maxstart = curstart;
1812                         maxend = curend;
1813                 }
1814                 for (i = 0; i < 8; i++) {
1815                         if (i >= maxstart && i <= maxend) {
1816                                 if (i == 0)
1817                                         printf(":");
1818                                 if (i == maxend)
1819                                         printf(":");
1820                         } else {
1821                                 b = ntohs(addr->addr16[i]);
1822                                 printf("%x", b);
1823                                 if (i < 7)
1824                                         printf(":");
1825                         }
1826                 }
1827                 if (p) {
1828                         p = ntohs(p);
1829                         printf("[%u]", p);
1830                 }
1831                 break;
1832         }
1833 #endif /* INET6 */
1834         }
1835 }
1836
1837 void
1838 pf_print_state(struct pf_state *s)
1839 {
1840         pf_print_state_parts(s, NULL, NULL);
1841 }
1842
1843 static void
1844 pf_print_state_parts(struct pf_state *s,
1845     struct pf_state_key *skwp, struct pf_state_key *sksp)
1846 {
1847         struct pf_state_key *skw, *sks;
1848         u_int8_t proto, dir;
1849
1850         /* Do our best to fill these, but they're skipped if NULL */
1851         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1852         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1853         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1854         dir = s ? s->direction : 0;
1855
1856         switch (proto) {
1857         case IPPROTO_IPV4:
1858                 printf("IPv4");
1859                 break;
1860         case IPPROTO_IPV6:
1861                 printf("IPv6");
1862                 break;
1863         case IPPROTO_TCP:
1864                 printf("TCP");
1865                 break;
1866         case IPPROTO_UDP:
1867                 printf("UDP");
1868                 break;
1869         case IPPROTO_ICMP:
1870                 printf("ICMP");
1871                 break;
1872         case IPPROTO_ICMPV6:
1873                 printf("ICMPv6");
1874                 break;
1875         default:
1876                 printf("%u", proto);
1877                 break;
1878         }
1879         switch (dir) {
1880         case PF_IN:
1881                 printf(" in");
1882                 break;
1883         case PF_OUT:
1884                 printf(" out");
1885                 break;
1886         }
1887         if (skw) {
1888                 printf(" wire: ");
1889                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1890                 printf(" ");
1891                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1892         }
1893         if (sks) {
1894                 printf(" stack: ");
1895                 if (sks != skw) {
1896                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1897                         printf(" ");
1898                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1899                 } else
1900                         printf("-");
1901         }
1902         if (s) {
1903                 if (proto == IPPROTO_TCP) {
1904                         printf(" [lo=%u high=%u win=%u modulator=%u",
1905                             s->src.seqlo, s->src.seqhi,
1906                             s->src.max_win, s->src.seqdiff);
1907                         if (s->src.wscale && s->dst.wscale)
1908                                 printf(" wscale=%u",
1909                                     s->src.wscale & PF_WSCALE_MASK);
1910                         printf("]");
1911                         printf(" [lo=%u high=%u win=%u modulator=%u",
1912                             s->dst.seqlo, s->dst.seqhi,
1913                             s->dst.max_win, s->dst.seqdiff);
1914                         if (s->src.wscale && s->dst.wscale)
1915                                 printf(" wscale=%u",
1916                                 s->dst.wscale & PF_WSCALE_MASK);
1917                         printf("]");
1918                 }
1919                 printf(" %u:%u", s->src.state, s->dst.state);
1920         }
1921 }
1922
1923 void
1924 pf_print_flags(u_int8_t f)
1925 {
1926         if (f)
1927                 printf(" ");
1928         if (f & TH_FIN)
1929                 printf("F");
1930         if (f & TH_SYN)
1931                 printf("S");
1932         if (f & TH_RST)
1933                 printf("R");
1934         if (f & TH_PUSH)
1935                 printf("P");
1936         if (f & TH_ACK)
1937                 printf("A");
1938         if (f & TH_URG)
1939                 printf("U");
1940         if (f & TH_ECE)
1941                 printf("E");
1942         if (f & TH_CWR)
1943                 printf("W");
1944 }
1945
1946 #define PF_SET_SKIP_STEPS(i)                                    \
1947         do {                                                    \
1948                 while (head[i] != cur) {                        \
1949                         head[i]->skip[i].ptr = cur;             \
1950                         head[i] = TAILQ_NEXT(head[i], entries); \
1951                 }                                               \
1952         } while (0)
1953
1954 void
1955 pf_calc_skip_steps(struct pf_rulequeue *rules)
1956 {
1957         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1958         int i;
1959
1960         cur = TAILQ_FIRST(rules);
1961         prev = cur;
1962         for (i = 0; i < PF_SKIP_COUNT; ++i)
1963                 head[i] = cur;
1964         while (cur != NULL) {
1965
1966                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1967                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1968                 if (cur->direction != prev->direction)
1969                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1970                 if (cur->af != prev->af)
1971                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1972                 if (cur->proto != prev->proto)
1973                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1974                 if (cur->src.neg != prev->src.neg ||
1975                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1976                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1977                 if (cur->src.port[0] != prev->src.port[0] ||
1978                     cur->src.port[1] != prev->src.port[1] ||
1979                     cur->src.port_op != prev->src.port_op)
1980                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1981                 if (cur->dst.neg != prev->dst.neg ||
1982                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1983                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1984                 if (cur->dst.port[0] != prev->dst.port[0] ||
1985                     cur->dst.port[1] != prev->dst.port[1] ||
1986                     cur->dst.port_op != prev->dst.port_op)
1987                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1988
1989                 prev = cur;
1990                 cur = TAILQ_NEXT(cur, entries);
1991         }
1992         for (i = 0; i < PF_SKIP_COUNT; ++i)
1993                 PF_SET_SKIP_STEPS(i);
1994 }
1995
1996 static int
1997 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1998 {
1999         if (aw1->type != aw2->type)
2000                 return (1);
2001         switch (aw1->type) {
2002         case PF_ADDR_ADDRMASK:
2003         case PF_ADDR_RANGE:
2004                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2005                         return (1);
2006                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2007                         return (1);
2008                 return (0);
2009         case PF_ADDR_DYNIFTL:
2010                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2011         case PF_ADDR_NOROUTE:
2012         case PF_ADDR_URPFFAILED:
2013                 return (0);
2014         case PF_ADDR_TABLE:
2015                 return (aw1->p.tbl != aw2->p.tbl);
2016         default:
2017                 printf("invalid address type: %d\n", aw1->type);
2018                 return (1);
2019         }
2020 }
2021
2022 /**
2023  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2024  * header isn't always a full checksum. In some cases (i.e. output) it's a
2025  * pseudo-header checksum, which is a partial checksum over src/dst IP
2026  * addresses, protocol number and length.
2027  *
2028  * That means we have the following cases:
2029  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2030  *      checksums, we need to update the checksum whenever we change anything.
2031  *  * Output (i.e. the checksum is a pseudo-header checksum):
2032  *      x The field being updated is src/dst address or affects the length of
2033  *      the packet. We need to update the pseudo-header checksum (note that this
2034  *      checksum is not ones' complement).
2035  *      x Some other field is being modified (e.g. src/dst port numbers): We
2036  *      don't have to update anything.
2037  **/
2038 u_int16_t
2039 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2040 {
2041         u_int32_t       l;
2042
2043         if (udp && !cksum)
2044                 return (0x0000);
2045         l = cksum + old - new;
2046         l = (l >> 16) + (l & 65535);
2047         l = l & 65535;
2048         if (udp && !l)
2049                 return (0xFFFF);
2050         return (l);
2051 }
2052
2053 u_int16_t
2054 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2055         u_int16_t new, u_int8_t udp)
2056 {
2057         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2058                 return (cksum);
2059
2060         return (pf_cksum_fixup(cksum, old, new, udp));
2061 }
2062
2063 static void
2064 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2065         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2066         sa_family_t af)
2067 {
2068         struct pf_addr  ao;
2069         u_int16_t       po = *p;
2070
2071         PF_ACPY(&ao, a, af);
2072         PF_ACPY(a, an, af);
2073
2074         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2075                 *pc = ~*pc;
2076
2077         *p = pn;
2078
2079         switch (af) {
2080 #ifdef INET
2081         case AF_INET:
2082                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2083                     ao.addr16[0], an->addr16[0], 0),
2084                     ao.addr16[1], an->addr16[1], 0);
2085                 *p = pn;
2086
2087                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2088                     ao.addr16[0], an->addr16[0], u),
2089                     ao.addr16[1], an->addr16[1], u);
2090
2091                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2092                 break;
2093 #endif /* INET */
2094 #ifdef INET6
2095         case AF_INET6:
2096                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2097                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2098                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2099                     ao.addr16[0], an->addr16[0], u),
2100                     ao.addr16[1], an->addr16[1], u),
2101                     ao.addr16[2], an->addr16[2], u),
2102                     ao.addr16[3], an->addr16[3], u),
2103                     ao.addr16[4], an->addr16[4], u),
2104                     ao.addr16[5], an->addr16[5], u),
2105                     ao.addr16[6], an->addr16[6], u),
2106                     ao.addr16[7], an->addr16[7], u);
2107
2108                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2109                 break;
2110 #endif /* INET6 */
2111         }
2112
2113         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2114             CSUM_DELAY_DATA_IPV6)) {
2115                 *pc = ~*pc;
2116                 if (! *pc)
2117                         *pc = 0xffff;
2118         }
2119 }
2120
2121 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2122 void
2123 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2124 {
2125         u_int32_t       ao;
2126
2127         memcpy(&ao, a, sizeof(ao));
2128         memcpy(a, &an, sizeof(u_int32_t));
2129         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2130             ao % 65536, an % 65536, u);
2131 }
2132
2133 void
2134 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2135 {
2136         u_int32_t       ao;
2137
2138         memcpy(&ao, a, sizeof(ao));
2139         memcpy(a, &an, sizeof(u_int32_t));
2140
2141         *c = pf_proto_cksum_fixup(m,
2142             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2143             ao % 65536, an % 65536, udp);
2144 }
2145
2146 #ifdef INET6
2147 static void
2148 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2149 {
2150         struct pf_addr  ao;
2151
2152         PF_ACPY(&ao, a, AF_INET6);
2153         PF_ACPY(a, an, AF_INET6);
2154
2155         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2156             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2157             pf_cksum_fixup(pf_cksum_fixup(*c,
2158             ao.addr16[0], an->addr16[0], u),
2159             ao.addr16[1], an->addr16[1], u),
2160             ao.addr16[2], an->addr16[2], u),
2161             ao.addr16[3], an->addr16[3], u),
2162             ao.addr16[4], an->addr16[4], u),
2163             ao.addr16[5], an->addr16[5], u),
2164             ao.addr16[6], an->addr16[6], u),
2165             ao.addr16[7], an->addr16[7], u);
2166 }
2167 #endif /* INET6 */
2168
2169 static void
2170 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2171     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2172     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2173 {
2174         struct pf_addr  oia, ooa;
2175
2176         PF_ACPY(&oia, ia, af);
2177         if (oa)
2178                 PF_ACPY(&ooa, oa, af);
2179
2180         /* Change inner protocol port, fix inner protocol checksum. */
2181         if (ip != NULL) {
2182                 u_int16_t       oip = *ip;
2183                 u_int32_t       opc;
2184
2185                 if (pc != NULL)
2186                         opc = *pc;
2187                 *ip = np;
2188                 if (pc != NULL)
2189                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2190                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2191                 if (pc != NULL)
2192                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2193         }
2194         /* Change inner ip address, fix inner ip and icmp checksums. */
2195         PF_ACPY(ia, na, af);
2196         switch (af) {
2197 #ifdef INET
2198         case AF_INET: {
2199                 u_int32_t        oh2c = *h2c;
2200
2201                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2202                     oia.addr16[0], ia->addr16[0], 0),
2203                     oia.addr16[1], ia->addr16[1], 0);
2204                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2205                     oia.addr16[0], ia->addr16[0], 0),
2206                     oia.addr16[1], ia->addr16[1], 0);
2207                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2208                 break;
2209         }
2210 #endif /* INET */
2211 #ifdef INET6
2212         case AF_INET6:
2213                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2214                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2215                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2216                     oia.addr16[0], ia->addr16[0], u),
2217                     oia.addr16[1], ia->addr16[1], u),
2218                     oia.addr16[2], ia->addr16[2], u),
2219                     oia.addr16[3], ia->addr16[3], u),
2220                     oia.addr16[4], ia->addr16[4], u),
2221                     oia.addr16[5], ia->addr16[5], u),
2222                     oia.addr16[6], ia->addr16[6], u),
2223                     oia.addr16[7], ia->addr16[7], u);
2224                 break;
2225 #endif /* INET6 */
2226         }
2227         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2228         if (oa) {
2229                 PF_ACPY(oa, na, af);
2230                 switch (af) {
2231 #ifdef INET
2232                 case AF_INET:
2233                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2234                             ooa.addr16[0], oa->addr16[0], 0),
2235                             ooa.addr16[1], oa->addr16[1], 0);
2236                         break;
2237 #endif /* INET */
2238 #ifdef INET6
2239                 case AF_INET6:
2240                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2241                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2242                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2243                             ooa.addr16[0], oa->addr16[0], u),
2244                             ooa.addr16[1], oa->addr16[1], u),
2245                             ooa.addr16[2], oa->addr16[2], u),
2246                             ooa.addr16[3], oa->addr16[3], u),
2247                             ooa.addr16[4], oa->addr16[4], u),
2248                             ooa.addr16[5], oa->addr16[5], u),
2249                             ooa.addr16[6], oa->addr16[6], u),
2250                             ooa.addr16[7], oa->addr16[7], u);
2251                         break;
2252 #endif /* INET6 */
2253                 }
2254         }
2255 }
2256
2257
2258 /*
2259  * Need to modulate the sequence numbers in the TCP SACK option
2260  * (credits to Krzysztof Pfaff for report and patch)
2261  */
2262 static int
2263 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2264     struct tcphdr *th, struct pf_state_peer *dst)
2265 {
2266         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2267         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2268         int copyback = 0, i, olen;
2269         struct sackblk sack;
2270
2271 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2272         if (hlen < TCPOLEN_SACKLEN ||
2273             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2274                 return 0;
2275
2276         while (hlen >= TCPOLEN_SACKLEN) {
2277                 olen = opt[1];
2278                 switch (*opt) {
2279                 case TCPOPT_EOL:        /* FALLTHROUGH */
2280                 case TCPOPT_NOP:
2281                         opt++;
2282                         hlen--;
2283                         break;
2284                 case TCPOPT_SACK:
2285                         if (olen > hlen)
2286                                 olen = hlen;
2287                         if (olen >= TCPOLEN_SACKLEN) {
2288                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2289                                     i += TCPOLEN_SACK) {
2290                                         memcpy(&sack, &opt[i], sizeof(sack));
2291                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2292                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2293                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2294                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2295                                         memcpy(&opt[i], &sack, sizeof(sack));
2296                                 }
2297                                 copyback = 1;
2298                         }
2299                         /* FALLTHROUGH */
2300                 default:
2301                         if (olen < 2)
2302                                 olen = 2;
2303                         hlen -= olen;
2304                         opt += olen;
2305                 }
2306         }
2307
2308         if (copyback)
2309                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2310         return (copyback);
2311 }
2312
2313 static void
2314 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2315     const struct pf_addr *saddr, const struct pf_addr *daddr,
2316     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2317     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2318     u_int16_t rtag, struct ifnet *ifp)
2319 {
2320         struct pf_send_entry *pfse;
2321         struct mbuf     *m;
2322         int              len, tlen;
2323 #ifdef INET
2324         struct ip       *h = NULL;
2325 #endif /* INET */
2326 #ifdef INET6
2327         struct ip6_hdr  *h6 = NULL;
2328 #endif /* INET6 */
2329         struct tcphdr   *th;
2330         char            *opt;
2331         struct pf_mtag  *pf_mtag;
2332
2333         len = 0;
2334         th = NULL;
2335
2336         /* maximum segment size tcp option */
2337         tlen = sizeof(struct tcphdr);
2338         if (mss)
2339                 tlen += 4;
2340
2341         switch (af) {
2342 #ifdef INET
2343         case AF_INET:
2344                 len = sizeof(struct ip) + tlen;
2345                 break;
2346 #endif /* INET */
2347 #ifdef INET6
2348         case AF_INET6:
2349                 len = sizeof(struct ip6_hdr) + tlen;
2350                 break;
2351 #endif /* INET6 */
2352         default:
2353                 panic("%s: unsupported af %d", __func__, af);
2354         }
2355
2356         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2357         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2358         if (pfse == NULL)
2359                 return;
2360         m = m_gethdr(M_NOWAIT, MT_DATA);
2361         if (m == NULL) {
2362                 free(pfse, M_PFTEMP);
2363                 return;
2364         }
2365 #ifdef MAC
2366         mac_netinet_firewall_send(m);
2367 #endif
2368         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2369                 free(pfse, M_PFTEMP);
2370                 m_freem(m);
2371                 return;
2372         }
2373         if (tag)
2374                 m->m_flags |= M_SKIP_FIREWALL;
2375         pf_mtag->tag = rtag;
2376
2377         if (r != NULL && r->rtableid >= 0)
2378                 M_SETFIB(m, r->rtableid);
2379
2380 #ifdef ALTQ
2381         if (r != NULL && r->qid) {
2382                 pf_mtag->qid = r->qid;
2383
2384                 /* add hints for ecn */
2385                 pf_mtag->hdr = mtod(m, struct ip *);
2386         }
2387 #endif /* ALTQ */
2388         m->m_data += max_linkhdr;
2389         m->m_pkthdr.len = m->m_len = len;
2390         m->m_pkthdr.rcvif = NULL;
2391         bzero(m->m_data, len);
2392         switch (af) {
2393 #ifdef INET
2394         case AF_INET:
2395                 h = mtod(m, struct ip *);
2396
2397                 /* IP header fields included in the TCP checksum */
2398                 h->ip_p = IPPROTO_TCP;
2399                 h->ip_len = htons(tlen);
2400                 h->ip_src.s_addr = saddr->v4.s_addr;
2401                 h->ip_dst.s_addr = daddr->v4.s_addr;
2402
2403                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2404                 break;
2405 #endif /* INET */
2406 #ifdef INET6
2407         case AF_INET6:
2408                 h6 = mtod(m, struct ip6_hdr *);
2409
2410                 /* IP header fields included in the TCP checksum */
2411                 h6->ip6_nxt = IPPROTO_TCP;
2412                 h6->ip6_plen = htons(tlen);
2413                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2414                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2415
2416                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2417                 break;
2418 #endif /* INET6 */
2419         }
2420
2421         /* TCP header */
2422         th->th_sport = sport;
2423         th->th_dport = dport;
2424         th->th_seq = htonl(seq);
2425         th->th_ack = htonl(ack);
2426         th->th_off = tlen >> 2;
2427         th->th_flags = flags;
2428         th->th_win = htons(win);
2429
2430         if (mss) {
2431                 opt = (char *)(th + 1);
2432                 opt[0] = TCPOPT_MAXSEG;
2433                 opt[1] = 4;
2434                 HTONS(mss);
2435                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2436         }
2437
2438         switch (af) {
2439 #ifdef INET
2440         case AF_INET:
2441                 /* TCP checksum */
2442                 th->th_sum = in_cksum(m, len);
2443
2444                 /* Finish the IP header */
2445                 h->ip_v = 4;
2446                 h->ip_hl = sizeof(*h) >> 2;
2447                 h->ip_tos = IPTOS_LOWDELAY;
2448                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2449                 h->ip_len = htons(len);
2450                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2451                 h->ip_sum = 0;
2452
2453                 pfse->pfse_type = PFSE_IP;
2454                 break;
2455 #endif /* INET */
2456 #ifdef INET6
2457         case AF_INET6:
2458                 /* TCP checksum */
2459                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2460                     sizeof(struct ip6_hdr), tlen);
2461
2462                 h6->ip6_vfc |= IPV6_VERSION;
2463                 h6->ip6_hlim = IPV6_DEFHLIM;
2464
2465                 pfse->pfse_type = PFSE_IP6;
2466                 break;
2467 #endif /* INET6 */
2468         }
2469         pfse->pfse_m = m;
2470         pf_send(pfse);
2471 }
2472
2473 static int
2474 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2475 {
2476         struct m_tag *mtag;
2477
2478         KASSERT(prio <= PF_PRIO_MAX,
2479             ("%s with invalid pcp", __func__));
2480
2481         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2482         if (mtag == NULL) {
2483                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2484                     sizeof(uint8_t), M_NOWAIT);
2485                 if (mtag == NULL)
2486                         return (ENOMEM);
2487                 m_tag_prepend(m, mtag);
2488         }
2489
2490         *(uint8_t *)(mtag + 1) = prio;
2491         return (0);
2492 }
2493
2494 static int
2495 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2496 {
2497         struct m_tag *mtag;
2498         u_int8_t mpcp;
2499
2500         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2501         if (mtag == NULL)
2502                 return (0);
2503
2504         if (prio == PF_PRIO_ZERO)
2505                 prio = 0;
2506
2507         mpcp = *(uint8_t *)(mtag + 1);
2508
2509         return (mpcp == prio);
2510 }
2511
2512 static void
2513 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2514     struct pf_rule *r)
2515 {
2516         struct pf_send_entry *pfse;
2517         struct mbuf *m0;
2518         struct pf_mtag *pf_mtag;
2519
2520         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2521         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2522         if (pfse == NULL)
2523                 return;
2524
2525         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2526                 free(pfse, M_PFTEMP);
2527                 return;
2528         }
2529
2530         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2531                 free(pfse, M_PFTEMP);
2532                 return;
2533         }
2534         /* XXX: revisit */
2535         m0->m_flags |= M_SKIP_FIREWALL;
2536
2537         if (r->rtableid >= 0)
2538                 M_SETFIB(m0, r->rtableid);
2539
2540 #ifdef ALTQ
2541         if (r->qid) {
2542                 pf_mtag->qid = r->qid;
2543                 /* add hints for ecn */
2544                 pf_mtag->hdr = mtod(m0, struct ip *);
2545         }
2546 #endif /* ALTQ */
2547
2548         switch (af) {
2549 #ifdef INET
2550         case AF_INET:
2551                 pfse->pfse_type = PFSE_ICMP;
2552                 break;
2553 #endif /* INET */
2554 #ifdef INET6
2555         case AF_INET6:
2556                 pfse->pfse_type = PFSE_ICMP6;
2557                 break;
2558 #endif /* INET6 */
2559         }
2560         pfse->pfse_m = m0;
2561         pfse->icmpopts.type = type;
2562         pfse->icmpopts.code = code;
2563         pf_send(pfse);
2564 }
2565
2566 /*
2567  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2568  * If n is 0, they match if they are equal. If n is != 0, they match if they
2569  * are different.
2570  */
2571 int
2572 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2573     struct pf_addr *b, sa_family_t af)
2574 {
2575         int     match = 0;
2576
2577         switch (af) {
2578 #ifdef INET
2579         case AF_INET:
2580                 if ((a->addr32[0] & m->addr32[0]) ==
2581                     (b->addr32[0] & m->addr32[0]))
2582                         match++;
2583                 break;
2584 #endif /* INET */
2585 #ifdef INET6
2586         case AF_INET6:
2587                 if (((a->addr32[0] & m->addr32[0]) ==
2588                      (b->addr32[0] & m->addr32[0])) &&
2589                     ((a->addr32[1] & m->addr32[1]) ==
2590                      (b->addr32[1] & m->addr32[1])) &&
2591                     ((a->addr32[2] & m->addr32[2]) ==
2592                      (b->addr32[2] & m->addr32[2])) &&
2593                     ((a->addr32[3] & m->addr32[3]) ==
2594                      (b->addr32[3] & m->addr32[3])))
2595                         match++;
2596                 break;
2597 #endif /* INET6 */
2598         }
2599         if (match) {
2600                 if (n)
2601                         return (0);
2602                 else
2603                         return (1);
2604         } else {
2605                 if (n)
2606                         return (1);
2607                 else
2608                         return (0);
2609         }
2610 }
2611
2612 /*
2613  * Return 1 if b <= a <= e, otherwise return 0.
2614  */
2615 int
2616 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2617     struct pf_addr *a, sa_family_t af)
2618 {
2619         switch (af) {
2620 #ifdef INET
2621         case AF_INET:
2622                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2623                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2624                         return (0);
2625                 break;
2626 #endif /* INET */
2627 #ifdef INET6
2628         case AF_INET6: {
2629                 int     i;
2630
2631                 /* check a >= b */
2632                 for (i = 0; i < 4; ++i)
2633                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2634                                 break;
2635                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2636                                 return (0);
2637                 /* check a <= e */
2638                 for (i = 0; i < 4; ++i)
2639                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2640                                 break;
2641                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2642                                 return (0);
2643                 break;
2644         }
2645 #endif /* INET6 */
2646         }
2647         return (1);
2648 }
2649
2650 static int
2651 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2652 {
2653         switch (op) {
2654         case PF_OP_IRG:
2655                 return ((p > a1) && (p < a2));
2656         case PF_OP_XRG:
2657                 return ((p < a1) || (p > a2));
2658         case PF_OP_RRG:
2659                 return ((p >= a1) && (p <= a2));
2660         case PF_OP_EQ:
2661                 return (p == a1);
2662         case PF_OP_NE:
2663                 return (p != a1);
2664         case PF_OP_LT:
2665                 return (p < a1);
2666         case PF_OP_LE:
2667                 return (p <= a1);
2668         case PF_OP_GT:
2669                 return (p > a1);
2670         case PF_OP_GE:
2671                 return (p >= a1);
2672         }
2673         return (0); /* never reached */
2674 }
2675
2676 int
2677 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2678 {
2679         NTOHS(a1);
2680         NTOHS(a2);
2681         NTOHS(p);
2682         return (pf_match(op, a1, a2, p));
2683 }
2684
2685 static int
2686 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2687 {
2688         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2689                 return (0);
2690         return (pf_match(op, a1, a2, u));
2691 }
2692
2693 static int
2694 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2695 {
2696         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2697                 return (0);
2698         return (pf_match(op, a1, a2, g));
2699 }
2700
2701 int
2702 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2703 {
2704         if (*tag == -1)
2705                 *tag = mtag;
2706
2707         return ((!r->match_tag_not && r->match_tag == *tag) ||
2708             (r->match_tag_not && r->match_tag != *tag));
2709 }
2710
2711 int
2712 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2713 {
2714
2715         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2716
2717         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2718                 return (ENOMEM);
2719
2720         pd->pf_mtag->tag = tag;
2721
2722         return (0);
2723 }
2724
2725 #define PF_ANCHOR_STACKSIZE     32
2726 struct pf_anchor_stackframe {
2727         struct pf_ruleset       *rs;
2728         struct pf_rule          *r;     /* XXX: + match bit */
2729         struct pf_anchor        *child;
2730 };
2731
2732 /*
2733  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2734  */
2735 #define PF_ANCHORSTACK_MATCH    0x00000001
2736 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2737
2738 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2739 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2740                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2741 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2742                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2743 } while (0)
2744
2745 void
2746 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2747     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2748     int *match)
2749 {
2750         struct pf_anchor_stackframe     *f;
2751
2752         PF_RULES_RASSERT();
2753
2754         if (match)
2755                 *match = 0;
2756         if (*depth >= PF_ANCHOR_STACKSIZE) {
2757                 printf("%s: anchor stack overflow on %s\n",
2758                     __func__, (*r)->anchor->name);
2759                 *r = TAILQ_NEXT(*r, entries);
2760                 return;
2761         } else if (*depth == 0 && a != NULL)
2762                 *a = *r;
2763         f = stack + (*depth)++;
2764         f->rs = *rs;
2765         f->r = *r;
2766         if ((*r)->anchor_wildcard) {
2767                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2768
2769                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2770                         *r = NULL;
2771                         return;
2772                 }
2773                 *rs = &f->child->ruleset;
2774         } else {
2775                 f->child = NULL;
2776                 *rs = &(*r)->anchor->ruleset;
2777         }
2778         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2779 }
2780
2781 int
2782 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2783     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2784     int *match)
2785 {
2786         struct pf_anchor_stackframe     *f;
2787         struct pf_rule *fr;
2788         int quick = 0;
2789
2790         PF_RULES_RASSERT();
2791
2792         do {
2793                 if (*depth <= 0)
2794                         break;
2795                 f = stack + *depth - 1;
2796                 fr = PF_ANCHOR_RULE(f);
2797                 if (f->child != NULL) {
2798                         struct pf_anchor_node *parent;
2799
2800                         /*
2801                          * This block traverses through
2802                          * a wildcard anchor.
2803                          */
2804                         parent = &fr->anchor->children;
2805                         if (match != NULL && *match) {
2806                                 /*
2807                                  * If any of "*" matched, then
2808                                  * "foo/ *" matched, mark frame
2809                                  * appropriately.
2810                                  */
2811                                 PF_ANCHOR_SET_MATCH(f);
2812                                 *match = 0;
2813                         }
2814                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2815                         if (f->child != NULL) {
2816                                 *rs = &f->child->ruleset;
2817                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2818                                 if (*r == NULL)
2819                                         continue;
2820                                 else
2821                                         break;
2822                         }
2823                 }
2824                 (*depth)--;
2825                 if (*depth == 0 && a != NULL)
2826                         *a = NULL;
2827                 *rs = f->rs;
2828                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2829                         quick = fr->quick;
2830                 *r = TAILQ_NEXT(fr, entries);
2831         } while (*r == NULL);
2832
2833         return (quick);
2834 }
2835
2836 #ifdef INET6
2837 void
2838 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2839     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2840 {
2841         switch (af) {
2842 #ifdef INET
2843         case AF_INET:
2844                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2845                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2846                 break;
2847 #endif /* INET */
2848         case AF_INET6:
2849                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2850                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2851                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2852                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2853                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2854                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2855                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2856                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2857                 break;
2858         }
2859 }
2860
2861 void
2862 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2863 {
2864         switch (af) {
2865 #ifdef INET
2866         case AF_INET:
2867                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2868                 break;
2869 #endif /* INET */
2870         case AF_INET6:
2871                 if (addr->addr32[3] == 0xffffffff) {
2872                         addr->addr32[3] = 0;
2873                         if (addr->addr32[2] == 0xffffffff) {
2874                                 addr->addr32[2] = 0;
2875                                 if (addr->addr32[1] == 0xffffffff) {
2876                                         addr->addr32[1] = 0;
2877                                         addr->addr32[0] =
2878                                             htonl(ntohl(addr->addr32[0]) + 1);
2879                                 } else
2880                                         addr->addr32[1] =
2881                                             htonl(ntohl(addr->addr32[1]) + 1);
2882                         } else
2883                                 addr->addr32[2] =
2884                                     htonl(ntohl(addr->addr32[2]) + 1);
2885                 } else
2886                         addr->addr32[3] =
2887                             htonl(ntohl(addr->addr32[3]) + 1);
2888                 break;
2889         }
2890 }
2891 #endif /* INET6 */
2892
2893 int
2894 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2895 {
2896         struct pf_addr          *saddr, *daddr;
2897         u_int16_t                sport, dport;
2898         struct inpcbinfo        *pi;
2899         struct inpcb            *inp;
2900
2901         pd->lookup.uid = UID_MAX;
2902         pd->lookup.gid = GID_MAX;
2903
2904         switch (pd->proto) {
2905         case IPPROTO_TCP:
2906                 if (pd->hdr.tcp == NULL)
2907                         return (-1);
2908                 sport = pd->hdr.tcp->th_sport;
2909                 dport = pd->hdr.tcp->th_dport;
2910                 pi = &V_tcbinfo;
2911                 break;
2912         case IPPROTO_UDP:
2913                 if (pd->hdr.udp == NULL)
2914                         return (-1);
2915                 sport = pd->hdr.udp->uh_sport;
2916                 dport = pd->hdr.udp->uh_dport;
2917                 pi = &V_udbinfo;
2918                 break;
2919         default:
2920                 return (-1);
2921         }
2922         if (direction == PF_IN) {
2923                 saddr = pd->src;
2924                 daddr = pd->dst;
2925         } else {
2926                 u_int16_t       p;
2927
2928                 p = sport;
2929                 sport = dport;
2930                 dport = p;
2931                 saddr = pd->dst;
2932                 daddr = pd->src;
2933         }
2934         switch (pd->af) {
2935 #ifdef INET
2936         case AF_INET:
2937                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2938                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2939                 if (inp == NULL) {
2940                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2941                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2942                            INPLOOKUP_RLOCKPCB, NULL, m);
2943                         if (inp == NULL)
2944                                 return (-1);
2945                 }
2946                 break;
2947 #endif /* INET */
2948 #ifdef INET6
2949         case AF_INET6:
2950                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2951                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2952                 if (inp == NULL) {
2953                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2954                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2955                             INPLOOKUP_RLOCKPCB, NULL, m);
2956                         if (inp == NULL)
2957                                 return (-1);
2958                 }
2959                 break;
2960 #endif /* INET6 */
2961
2962         default:
2963                 return (-1);
2964         }
2965         INP_RLOCK_ASSERT(inp);
2966         pd->lookup.uid = inp->inp_cred->cr_uid;
2967         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2968         INP_RUNLOCK(inp);
2969
2970         return (1);
2971 }
2972
2973 static u_int8_t
2974 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2975 {
2976         int              hlen;
2977         u_int8_t         hdr[60];
2978         u_int8_t        *opt, optlen;
2979         u_int8_t         wscale = 0;
2980
2981         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2982         if (hlen <= sizeof(struct tcphdr))
2983                 return (0);
2984         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2985                 return (0);
2986         opt = hdr + sizeof(struct tcphdr);
2987         hlen -= sizeof(struct tcphdr);
2988         while (hlen >= 3) {
2989                 switch (*opt) {
2990                 case TCPOPT_EOL:
2991                 case TCPOPT_NOP:
2992                         ++opt;
2993                         --hlen;
2994                         break;
2995                 case TCPOPT_WINDOW:
2996                         wscale = opt[2];
2997                         if (wscale > TCP_MAX_WINSHIFT)
2998                                 wscale = TCP_MAX_WINSHIFT;
2999                         wscale |= PF_WSCALE_FLAG;
3000                         /* FALLTHROUGH */
3001                 default:
3002                         optlen = opt[1];
3003                         if (optlen < 2)
3004                                 optlen = 2;
3005                         hlen -= optlen;
3006                         opt += optlen;
3007                         break;
3008                 }
3009         }
3010         return (wscale);
3011 }
3012
3013 static u_int16_t
3014 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3015 {
3016         int              hlen;
3017         u_int8_t         hdr[60];
3018         u_int8_t        *opt, optlen;
3019         u_int16_t        mss = V_tcp_mssdflt;
3020
3021         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3022         if (hlen <= sizeof(struct tcphdr))
3023                 return (0);
3024         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3025                 return (0);
3026         opt = hdr + sizeof(struct tcphdr);
3027         hlen -= sizeof(struct tcphdr);
3028         while (hlen >= TCPOLEN_MAXSEG) {
3029                 switch (*opt) {
3030                 case TCPOPT_EOL:
3031                 case TCPOPT_NOP:
3032                         ++opt;
3033                         --hlen;
3034                         break;
3035                 case TCPOPT_MAXSEG:
3036                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3037                         NTOHS(mss);
3038                         /* FALLTHROUGH */
3039                 default:
3040                         optlen = opt[1];
3041                         if (optlen < 2)
3042                                 optlen = 2;
3043                         hlen -= optlen;
3044                         opt += optlen;
3045                         break;
3046                 }
3047         }
3048         return (mss);
3049 }
3050
3051 static u_int16_t
3052 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3053 {
3054 #ifdef INET
3055         struct nhop4_basic      nh4;
3056 #endif /* INET */
3057 #ifdef INET6
3058         struct nhop6_basic      nh6;
3059         struct in6_addr         dst6;
3060         uint32_t                scopeid;
3061 #endif /* INET6 */
3062         int                      hlen = 0;
3063         uint16_t                 mss = 0;
3064
3065         switch (af) {
3066 #ifdef INET
3067         case AF_INET:
3068                 hlen = sizeof(struct ip);
3069                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3070                         mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3071                 break;
3072 #endif /* INET */
3073 #ifdef INET6
3074         case AF_INET6:
3075                 hlen = sizeof(struct ip6_hdr);
3076                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3077                 if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3078                         mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3079                 break;
3080 #endif /* INET6 */
3081         }
3082
3083         mss = max(V_tcp_mssdflt, mss);
3084         mss = min(mss, offer);
3085         mss = max(mss, 64);             /* sanity - at least max opt space */
3086         return (mss);
3087 }
3088
3089 static u_int32_t
3090 pf_tcp_iss(struct pf_pdesc *pd)
3091 {
3092         MD5_CTX ctx;
3093         u_int32_t digest[4];
3094
3095         if (V_pf_tcp_secret_init == 0) {
3096                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3097                 MD5Init(&V_pf_tcp_secret_ctx);
3098                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3099                     sizeof(V_pf_tcp_secret));
3100                 V_pf_tcp_secret_init = 1;
3101         }
3102
3103         ctx = V_pf_tcp_secret_ctx;
3104
3105         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3106         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3107         if (pd->af == AF_INET6) {
3108                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3109                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3110         } else {
3111                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3112                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3113         }
3114         MD5Final((u_char *)digest, &ctx);
3115         V_pf_tcp_iss_off += 4096;
3116 #define ISN_RANDOM_INCREMENT (4096 - 1)
3117         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3118             V_pf_tcp_iss_off);
3119 #undef  ISN_RANDOM_INCREMENT
3120 }
3121
3122 static int
3123 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3124     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3125     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3126 {
3127         struct pf_rule          *nr = NULL;
3128         struct pf_addr          * const saddr = pd->src;
3129         struct pf_addr          * const daddr = pd->dst;
3130         sa_family_t              af = pd->af;
3131         struct pf_rule          *r, *a = NULL;
3132         struct pf_ruleset       *ruleset = NULL;
3133         struct pf_src_node      *nsn = NULL;
3134         struct tcphdr           *th = pd->hdr.tcp;
3135         struct pf_state_key     *sk = NULL, *nk = NULL;
3136         u_short                  reason;
3137         int                      rewrite = 0, hdrlen = 0;
3138         int                      tag = -1, rtableid = -1;
3139         int                      asd = 0;
3140         int                      match = 0;
3141         int                      state_icmp = 0;
3142         u_int16_t                sport = 0, dport = 0;
3143         u_int16_t                bproto_sum = 0, bip_sum = 0;
3144         u_int8_t                 icmptype = 0, icmpcode = 0;
3145         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3146
3147         PF_RULES_RASSERT();
3148
3149         if (inp != NULL) {
3150                 INP_LOCK_ASSERT(inp);
3151                 pd->lookup.uid = inp->inp_cred->cr_uid;
3152                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3153                 pd->lookup.done = 1;
3154         }
3155
3156         switch (pd->proto) {
3157         case IPPROTO_TCP:
3158                 sport = th->th_sport;
3159                 dport = th->th_dport;
3160                 hdrlen = sizeof(*th);
3161                 break;
3162         case IPPROTO_UDP:
3163                 sport = pd->hdr.udp->uh_sport;
3164                 dport = pd->hdr.udp->uh_dport;
3165                 hdrlen = sizeof(*pd->hdr.udp);
3166                 break;
3167 #ifdef INET
3168         case IPPROTO_ICMP:
3169                 if (pd->af != AF_INET)
3170                         break;
3171                 sport = dport = pd->hdr.icmp->icmp_id;
3172                 hdrlen = sizeof(*pd->hdr.icmp);
3173                 icmptype = pd->hdr.icmp->icmp_type;
3174                 icmpcode = pd->hdr.icmp->icmp_code;
3175
3176                 if (icmptype == ICMP_UNREACH ||
3177                     icmptype == ICMP_SOURCEQUENCH ||
3178                     icmptype == ICMP_REDIRECT ||
3179                     icmptype == ICMP_TIMXCEED ||
3180                     icmptype == ICMP_PARAMPROB)
3181                         state_icmp++;
3182                 break;
3183 #endif /* INET */
3184 #ifdef INET6
3185         case IPPROTO_ICMPV6:
3186                 if (af != AF_INET6)
3187                         break;
3188                 sport = dport = pd->hdr.icmp6->icmp6_id;
3189                 hdrlen = sizeof(*pd->hdr.icmp6);
3190                 icmptype = pd->hdr.icmp6->icmp6_type;
3191                 icmpcode = pd->hdr.icmp6->icmp6_code;
3192
3193                 if (icmptype == ICMP6_DST_UNREACH ||
3194                     icmptype == ICMP6_PACKET_TOO_BIG ||
3195                     icmptype == ICMP6_TIME_EXCEEDED ||
3196                     icmptype == ICMP6_PARAM_PROB)
3197                         state_icmp++;
3198                 break;
3199 #endif /* INET6 */
3200         default:
3201                 sport = dport = hdrlen = 0;
3202                 break;
3203         }
3204
3205         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3206
3207         /* check packet for BINAT/NAT/RDR */
3208         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3209             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3210                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3211                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3212
3213                 if (pd->ip_sum)
3214                         bip_sum = *pd->ip_sum;
3215
3216                 switch (pd->proto) {
3217                 case IPPROTO_TCP:
3218                         bproto_sum = th->th_sum;
3219                         pd->proto_sum = &th->th_sum;
3220
3221                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3222                             nk->port[pd->sidx] != sport) {
3223                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3224                                     &th->th_sum, &nk->addr[pd->sidx],
3225                                     nk->port[pd->sidx], 0, af);
3226                                 pd->sport = &th->th_sport;
3227                                 sport = th->th_sport;
3228                         }
3229
3230                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3231                             nk->port[pd->didx] != dport) {
3232                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3233                                     &th->th_sum, &nk->addr[pd->didx],
3234                                     nk->port[pd->didx], 0, af);
3235                                 dport = th->th_dport;
3236                                 pd->dport = &th->th_dport;
3237                         }
3238                         rewrite++;
3239                         break;
3240                 case IPPROTO_UDP:
3241                         bproto_sum = pd->hdr.udp->uh_sum;
3242                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3243
3244                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3245                             nk->port[pd->sidx] != sport) {
3246                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3247                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3248                                     &nk->addr[pd->sidx],
3249                                     nk->port[pd->sidx], 1, af);
3250                                 sport = pd->hdr.udp->uh_sport;
3251                                 pd->sport = &pd->hdr.udp->uh_sport;
3252                         }
3253
3254                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3255                             nk->port[pd->didx] != dport) {
3256                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3257                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3258                                     &nk->addr[pd->didx],
3259                                     nk->port[pd->didx], 1, af);
3260                                 dport = pd->hdr.udp->uh_dport;
3261                                 pd->dport = &pd->hdr.udp->uh_dport;
3262                         }
3263                         rewrite++;
3264                         break;
3265 #ifdef INET
3266                 case IPPROTO_ICMP:
3267                         nk->port[0] = nk->port[1];
3268                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3269                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3270                                     nk->addr[pd->sidx].v4.s_addr, 0);
3271
3272                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3273                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3274                                     nk->addr[pd->didx].v4.s_addr, 0);
3275
3276                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3277                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3278                                     pd->hdr.icmp->icmp_cksum, sport,
3279                                     nk->port[1], 0);
3280                                 pd->hdr.icmp->icmp_id = nk->port[1];
3281                                 pd->sport = &pd->hdr.icmp->icmp_id;
3282                         }
3283                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3284                         break;
3285 #endif /* INET */
3286 #ifdef INET6
3287                 case IPPROTO_ICMPV6:
3288                         nk->port[0] = nk->port[1];
3289                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3290                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3291                                     &nk->addr[pd->sidx], 0);
3292
3293                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3294                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3295                                     &nk->addr[pd->didx], 0);
3296                         rewrite++;
3297                         break;
3298 #endif /* INET */
3299                 default:
3300                         switch (af) {
3301 #ifdef INET
3302                         case AF_INET:
3303                                 if (PF_ANEQ(saddr,
3304                                     &nk->addr[pd->sidx], AF_INET))
3305                                         pf_change_a(&saddr->v4.s_addr,
3306                                             pd->ip_sum,
3307                                             nk->addr[pd->sidx].v4.s_addr, 0);
3308
3309                                 if (PF_ANEQ(daddr,
3310                                     &nk->addr[pd->didx], AF_INET))
3311                                         pf_change_a(&daddr->v4.s_addr,
3312                                             pd->ip_sum,
3313                                             nk->addr[pd->didx].v4.s_addr, 0);
3314                                 break;
3315 #endif /* INET */
3316 #ifdef INET6
3317                         case AF_INET6:
3318                                 if (PF_ANEQ(saddr,
3319                                     &nk->addr[pd->sidx], AF_INET6))
3320                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3321
3322                                 if (PF_ANEQ(daddr,
3323                                     &nk->addr[pd->didx], AF_INET6))
3324                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3325                                 break;
3326 #endif /* INET */
3327                         }
3328                         break;
3329                 }
3330                 if (nr->natpass)
3331                         r = NULL;
3332                 pd->nat_rule = nr;
3333         }
3334
3335         while (r != NULL) {
3336                 r->evaluations++;
3337                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3338                         r = r->skip[PF_SKIP_IFP].ptr;
3339                 else if (r->direction && r->direction != direction)
3340                         r = r->skip[PF_SKIP_DIR].ptr;
3341                 else if (r->af && r->af != af)
3342                         r = r->skip[PF_SKIP_AF].ptr;
3343                 else if (r->proto && r->proto != pd->proto)
3344                         r = r->skip[PF_SKIP_PROTO].ptr;
3345                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3346                     r->src.neg, kif, M_GETFIB(m)))
3347                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3348                 /* tcp/udp only. port_op always 0 in other cases */
3349                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3350                     r->src.port[0], r->src.port[1], sport))
3351                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3352                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3353                     r->dst.neg, NULL, M_GETFIB(m)))
3354                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3355                 /* tcp/udp only. port_op always 0 in other cases */
3356                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3357                     r->dst.port[0], r->dst.port[1], dport))
3358                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3359                 /* icmp only. type always 0 in other cases */
3360                 else if (r->type && r->type != icmptype + 1)
3361                         r = TAILQ_NEXT(r, entries);
3362                 /* icmp only. type always 0 in other cases */
3363                 else if (r->code && r->code != icmpcode + 1)
3364                         r = TAILQ_NEXT(r, entries);
3365                 else if (r->tos && !(r->tos == pd->tos))
3366                         r = TAILQ_NEXT(r, entries);
3367                 else if (r->rule_flag & PFRULE_FRAGMENT)
3368                         r = TAILQ_NEXT(r, entries);
3369                 else if (pd->proto == IPPROTO_TCP &&
3370                     (r->flagset & th->th_flags) != r->flags)
3371                         r = TAILQ_NEXT(r, entries);
3372                 /* tcp/udp only. uid.op always 0 in other cases */
3373                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3374                     pf_socket_lookup(direction, pd, m), 1)) &&
3375                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3376                     pd->lookup.uid))
3377                         r = TAILQ_NEXT(r, entries);
3378                 /* tcp/udp only. gid.op always 0 in other cases */
3379                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3380                     pf_socket_lookup(direction, pd, m), 1)) &&
3381                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3382                     pd->lookup.gid))
3383                         r = TAILQ_NEXT(r, entries);
3384                 else if (r->prio &&
3385                     !pf_match_ieee8021q_pcp(r->prio, m))
3386                         r = TAILQ_NEXT(r, entries);
3387                 else if (r->prob &&
3388                     r->prob <= arc4random())
3389                         r = TAILQ_NEXT(r, entries);
3390                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3391                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3392                         r = TAILQ_NEXT(r, entries);
3393                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3394                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3395                     pf_osfp_fingerprint(pd, m, off, th),
3396                     r->os_fingerprint)))
3397                         r = TAILQ_NEXT(r, entries);
3398                 else {
3399                         if (r->tag)
3400                                 tag = r->tag;
3401                         if (r->rtableid >= 0)
3402                                 rtableid = r->rtableid;
3403                         if (r->anchor == NULL) {
3404                                 match = 1;
3405                                 *rm = r;
3406                                 *am = a;
3407                                 *rsm = ruleset;
3408                                 if ((*rm)->quick)
3409                                         break;
3410                                 r = TAILQ_NEXT(r, entries);
3411                         } else
3412                                 pf_step_into_anchor(anchor_stack, &asd,
3413                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3414                                     &match);
3415                 }
3416                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3417                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3418                         break;
3419         }
3420         r = *rm;
3421         a = *am;
3422         ruleset = *rsm;
3423
3424         REASON_SET(&reason, PFRES_MATCH);
3425
3426         if (r->log || (nr != NULL && nr->log)) {
3427                 if (rewrite)
3428                         m_copyback(m, off, hdrlen, pd->hdr.any);
3429                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3430                     ruleset, pd, 1);
3431         }
3432
3433         if ((r->action == PF_DROP) &&
3434             ((r->rule_flag & PFRULE_RETURNRST) ||
3435             (r->rule_flag & PFRULE_RETURNICMP) ||
3436             (r->rule_flag & PFRULE_RETURN))) {
3437                 /* undo NAT changes, if they have taken place */
3438                 if (nr != NULL) {
3439                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3440                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3441                         if (pd->sport)
3442                                 *pd->sport = sk->port[pd->sidx];
3443                         if (pd->dport)
3444                                 *pd->dport = sk->port[pd->didx];
3445                         if (pd->proto_sum)
3446                                 *pd->proto_sum = bproto_sum;
3447                         if (pd->ip_sum)
3448                                 *pd->ip_sum = bip_sum;
3449                         m_copyback(m, off, hdrlen, pd->hdr.any);
3450                 }
3451                 if (pd->proto == IPPROTO_TCP &&
3452                     ((r->rule_flag & PFRULE_RETURNRST) ||
3453                     (r->rule_flag & PFRULE_RETURN)) &&
3454                     !(th->th_flags & TH_RST)) {
3455                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3456                         int              len = 0;
3457 #ifdef INET
3458                         struct ip       *h4;
3459 #endif
3460 #ifdef INET6
3461                         struct ip6_hdr  *h6;
3462 #endif
3463
3464                         switch (af) {
3465 #ifdef INET
3466                         case AF_INET:
3467                                 h4 = mtod(m, struct ip *);
3468                                 len = ntohs(h4->ip_len) - off;
3469                                 break;
3470 #endif
3471 #ifdef INET6
3472                         case AF_INET6:
3473                                 h6 = mtod(m, struct ip6_hdr *);
3474                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3475                                 break;
3476 #endif
3477                         }
3478
3479                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3480                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3481                         else {
3482                                 if (th->th_flags & TH_SYN)
3483                                         ack++;
3484                                 if (th->th_flags & TH_FIN)
3485                                         ack++;
3486                                 pf_send_tcp(m, r, af, pd->dst,
3487                                     pd->src, th->th_dport, th->th_sport,
3488                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3489                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3490                         }
3491                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3492                     r->return_icmp)
3493                         pf_send_icmp(m, r->return_icmp >> 8,
3494                             r->return_icmp & 255, af, r);
3495                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3496                     r->return_icmp6)
3497                         pf_send_icmp(m, r->return_icmp6 >> 8,
3498                             r->return_icmp6 & 255, af, r);
3499         }
3500
3501         if (r->action == PF_DROP)
3502                 goto cleanup;
3503
3504         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3505                 REASON_SET(&reason, PFRES_MEMORY);
3506                 goto cleanup;
3507         }
3508         if (rtableid >= 0)
3509                 M_SETFIB(m, rtableid);
3510
3511         if (!state_icmp && (r->keep_state || nr != NULL ||
3512             (pd->flags & PFDESC_TCP_NORM))) {
3513                 int action;
3514                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3515                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3516                     hdrlen);
3517                 if (action != PF_PASS)
3518                         return (action);
3519         } else {
3520                 if (sk != NULL)
3521                         uma_zfree(V_pf_state_key_z, sk);
3522                 if (nk != NULL)
3523                         uma_zfree(V_pf_state_key_z, nk);
3524         }
3525
3526         /* copy back packet headers if we performed NAT operations */
3527         if (rewrite)
3528                 m_copyback(m, off, hdrlen, pd->hdr.any);
3529
3530         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3531             direction == PF_OUT &&
3532             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3533                 /*
3534                  * We want the state created, but we dont
3535                  * want to send this in case a partner
3536                  * firewall has to know about it to allow
3537                  * replies through it.
3538                  */
3539                 return (PF_DEFER);
3540
3541         return (PF_PASS);
3542
3543 cleanup:
3544         if (sk != NULL)
3545                 uma_zfree(V_pf_state_key_z, sk);
3546         if (nk != NULL)
3547                 uma_zfree(V_pf_state_key_z, nk);
3548         return (PF_DROP);
3549 }
3550
3551 static int
3552 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3553     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3554     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3555     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3556     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3557 {
3558         struct pf_state         *s = NULL;
3559         struct pf_src_node      *sn = NULL;
3560         struct tcphdr           *th = pd->hdr.tcp;
3561         u_int16_t                mss = V_tcp_mssdflt;
3562         u_short                  reason;
3563
3564         /* check maximums */
3565         if (r->max_states &&
3566             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3567                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3568                 REASON_SET(&reason, PFRES_MAXSTATES);
3569                 goto csfailed;
3570         }
3571         /* src node for filter rule */
3572         if ((r->rule_flag & PFRULE_SRCTRACK ||
3573             r->rpool.opts & PF_POOL_STICKYADDR) &&
3574             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3575                 REASON_SET(&reason, PFRES_SRCLIMIT);
3576                 goto csfailed;
3577         }
3578         /* src node for translation rule */
3579         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3580             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3581                 REASON_SET(&reason, PFRES_SRCLIMIT);
3582                 goto csfailed;
3583         }
3584         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3585         if (s == NULL) {
3586                 REASON_SET(&reason, PFRES_MEMORY);
3587                 goto csfailed;
3588         }
3589         s->rule.ptr = r;
3590         s->nat_rule.ptr = nr;
3591         s->anchor.ptr = a;
3592         STATE_INC_COUNTERS(s);
3593         if (r->allow_opts)
3594                 s->state_flags |= PFSTATE_ALLOWOPTS;
3595         if (r->rule_flag & PFRULE_STATESLOPPY)
3596                 s->state_flags |= PFSTATE_SLOPPY;
3597         s->log = r->log & PF_LOG_ALL;
3598         s->sync_state = PFSYNC_S_NONE;
3599         if (nr != NULL)
3600                 s->log |= nr->log & PF_LOG_ALL;
3601         switch (pd->proto) {
3602         case IPPROTO_TCP:
3603                 s->src.seqlo = ntohl(th->th_seq);
3604                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3605                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3606                     r->keep_state == PF_STATE_MODULATE) {
3607                         /* Generate sequence number modulator */
3608                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3609                             0)
3610                                 s->src.seqdiff = 1;
3611                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3612                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3613                         *rewrite = 1;
3614                 } else
3615                         s->src.seqdiff = 0;
3616                 if (th->th_flags & TH_SYN) {
3617                         s->src.seqhi++;
3618                         s->src.wscale = pf_get_wscale(m, off,
3619                             th->th_off, pd->af);
3620                 }
3621                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3622                 if (s->src.wscale & PF_WSCALE_MASK) {
3623                         /* Remove scale factor from initial window */
3624                         int win = s->src.max_win;
3625                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3626                         s->src.max_win = (win - 1) >>
3627                             (s->src.wscale & PF_WSCALE_MASK);
3628                 }
3629                 if (th->th_flags & TH_FIN)
3630                         s->src.seqhi++;
3631                 s->dst.seqhi = 1;
3632                 s->dst.max_win = 1;
3633                 s->src.state = TCPS_SYN_SENT;
3634                 s->dst.state = TCPS_CLOSED;
3635                 s->timeout = PFTM_TCP_FIRST_PACKET;
3636                 break;
3637         case IPPROTO_UDP:
3638                 s->src.state = PFUDPS_SINGLE;
3639                 s->dst.state = PFUDPS_NO_TRAFFIC;
3640                 s->timeout = PFTM_UDP_FIRST_PACKET;
3641                 break;
3642         case IPPROTO_ICMP:
3643 #ifdef INET6
3644         case IPPROTO_ICMPV6:
3645 #endif
3646                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3647                 break;
3648         default:
3649                 s->src.state = PFOTHERS_SINGLE;
3650                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3651                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3652         }
3653
3654         if (r->rt) {
3655                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3656                         REASON_SET(&reason, PFRES_MAPFAILED);
3657                         pf_src_tree_remove_state(s);
3658                         STATE_DEC_COUNTERS(s);
3659                         uma_zfree(V_pf_state_z, s);
3660                         goto csfailed;
3661                 }
3662                 s->rt_kif = r->rpool.cur->kif;
3663         }
3664
3665         s->creation = time_uptime;
3666         s->expire = time_uptime;
3667
3668         if (sn != NULL)
3669                 s->src_node = sn;
3670         if (nsn != NULL) {
3671                 /* XXX We only modify one side for now. */
3672                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3673                 s->nat_src_node = nsn;
3674         }
3675         if (pd->proto == IPPROTO_TCP) {
3676                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3677                     off, pd, th, &s->src, &s->dst)) {
3678                         REASON_SET(&reason, PFRES_MEMORY);
3679                         pf_src_tree_remove_state(s);
3680                         STATE_DEC_COUNTERS(s);
3681                         uma_zfree(V_pf_state_z, s);
3682                         return (PF_DROP);
3683                 }
3684                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3685                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3686                     &s->src, &s->dst, rewrite)) {
3687                         /* This really shouldn't happen!!! */
3688                         DPFPRINTF(PF_DEBUG_URGENT,
3689                             ("pf_normalize_tcp_stateful failed on first pkt"));
3690                         pf_normalize_tcp_cleanup(s);
3691                         pf_src_tree_remove_state(s);
3692                         STATE_DEC_COUNTERS(s);
3693                         uma_zfree(V_pf_state_z, s);
3694                         return (PF_DROP);
3695                 }
3696         }
3697         s->direction = pd->dir;
3698
3699         /*
3700          * sk/nk could already been setup by pf_get_translation().
3701          */
3702         if (nr == NULL) {
3703                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3704                     __func__, nr, sk, nk));
3705                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3706                 if (sk == NULL)
3707                         goto csfailed;
3708                 nk = sk;
3709         } else
3710                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3711                     __func__, nr, sk, nk));
3712
3713         /* Swap sk/nk for PF_OUT. */
3714         if (pf_state_insert(BOUND_IFACE(r, kif),
3715             (pd->dir == PF_IN) ? sk : nk,
3716             (pd->dir == PF_IN) ? nk : sk, s)) {
3717                 if (pd->proto == IPPROTO_TCP)
3718                         pf_normalize_tcp_cleanup(s);
3719                 REASON_SET(&reason, PFRES_STATEINS);
3720                 pf_src_tree_remove_state(s);
3721                 STATE_DEC_COUNTERS(s);
3722                 uma_zfree(V_pf_state_z, s);
3723                 return (PF_DROP);
3724         } else
3725                 *sm = s;
3726
3727         if (tag > 0)
3728                 s->tag = tag;
3729         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3730             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3731                 s->src.state = PF_TCPS_PROXY_SRC;
3732                 /* undo NAT changes, if they have taken place */
3733                 if (nr != NULL) {
3734                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3735                         if (pd->dir == PF_OUT)
3736                                 skt = s->key[PF_SK_STACK];
3737                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3738                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3739                         if (pd->sport)
3740                                 *pd->sport = skt->port[pd->sidx];
3741                         if (pd->dport)
3742                                 *pd->dport = skt->port[pd->didx];
3743                         if (pd->proto_sum)
3744                                 *pd->proto_sum = bproto_sum;
3745                         if (pd->ip_sum)
3746                                 *pd->ip_sum = bip_sum;
3747                         m_copyback(m, off, hdrlen, pd->hdr.any);
3748                 }
3749                 s->src.seqhi = htonl(arc4random());
3750                 /* Find mss option */
3751                 int rtid = M_GETFIB(m);
3752                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3753                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3754                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3755                 s->src.mss = mss;
3756                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3757                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3758                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3759                 REASON_SET(&reason, PFRES_SYNPROXY);
3760                 return (PF_SYNPROXY_DROP);
3761         }
3762
3763         return (PF_PASS);
3764
3765 csfailed:
3766         if (sk != NULL)
3767                 uma_zfree(V_pf_state_key_z, sk);
3768         if (nk != NULL)
3769                 uma_zfree(V_pf_state_key_z, nk);
3770
3771         if (sn != NULL) {
3772                 struct pf_srchash *sh;
3773
3774                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3775                 PF_HASHROW_LOCK(sh);
3776                 if (--sn->states == 0 && sn->expire == 0) {
3777                         pf_unlink_src_node(sn);
3778                         uma_zfree(V_pf_sources_z, sn);
3779                         counter_u64_add(
3780                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3781                 }
3782                 PF_HASHROW_UNLOCK(sh);
3783         }
3784
3785         if (nsn != sn && nsn != NULL) {
3786                 struct pf_srchash *sh;
3787
3788                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3789                 PF_HASHROW_LOCK(sh);
3790                 if (--nsn->states == 0 && nsn->expire == 0) {
3791                         pf_unlink_src_node(nsn);
3792                         uma_zfree(V_pf_sources_z, nsn);
3793                         counter_u64_add(
3794                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3795                 }
3796                 PF_HASHROW_UNLOCK(sh);
3797         }
3798
3799         return (PF_DROP);
3800 }
3801
3802 static int
3803 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3804     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3805     struct pf_ruleset **rsm)
3806 {
3807         struct pf_rule          *r, *a = NULL;
3808         struct pf_ruleset       *ruleset = NULL;
3809         sa_family_t              af = pd->af;
3810         u_short                  reason;
3811         int                      tag = -1;
3812         int                      asd = 0;
3813         int                      match = 0;
3814         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3815
3816         PF_RULES_RASSERT();
3817
3818         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3819         while (r != NULL) {
3820                 r->evaluations++;
3821                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3822                         r = r->skip[PF_SKIP_IFP].ptr;
3823                 else if (r->direction && r->direction != direction)
3824                         r = r->skip[PF_SKIP_DIR].ptr;
3825                 else if (r->af && r->af != af)
3826                         r = r->skip[PF_SKIP_AF].ptr;
3827                 else if (r->proto && r->proto != pd->proto)
3828                         r = r->skip[PF_SKIP_PROTO].ptr;
3829                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3830                     r->src.neg, kif, M_GETFIB(m)))
3831                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3832                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3833                     r->dst.neg, NULL, M_GETFIB(m)))
3834                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3835                 else if (r->tos && !(r->tos == pd->tos))
3836                         r = TAILQ_NEXT(r, entries);
3837                 else if (r->os_fingerprint != PF_OSFP_ANY)
3838                         r = TAILQ_NEXT(r, entries);
3839                 else if (pd->proto == IPPROTO_UDP &&
3840                     (r->src.port_op || r->dst.port_op))
3841                         r = TAILQ_NEXT(r, entries);
3842                 else if (pd->proto == IPPROTO_TCP &&
3843                     (r->src.port_op || r->dst.port_op || r->flagset))
3844                         r = TAILQ_NEXT(r, entries);
3845                 else if ((pd->proto == IPPROTO_ICMP ||
3846                     pd->proto == IPPROTO_ICMPV6) &&
3847                     (r->type || r->code))
3848                         r = TAILQ_NEXT(r, entries);
3849                 else if (r->prio &&
3850                     !pf_match_ieee8021q_pcp(r->prio, m))
3851                         r = TAILQ_NEXT(r, entries);
3852                 else if (r->prob && r->prob <=
3853                     (arc4random() % (UINT_MAX - 1) + 1))
3854                         r = TAILQ_NEXT(r, entries);
3855                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3856                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3857                         r = TAILQ_NEXT(r, entries);
3858                 else {
3859                         if (r->anchor == NULL) {
3860                                 match = 1;
3861                                 *rm = r;
3862                                 *am = a;
3863                                 *rsm = ruleset;
3864                                 if ((*rm)->quick)
3865                                         break;
3866                                 r = TAILQ_NEXT(r, entries);
3867                         } else
3868                                 pf_step_into_anchor(anchor_stack, &asd,
3869                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3870                                     &match);
3871                 }
3872                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3873                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3874                         break;
3875         }
3876         r = *rm;
3877         a = *am;
3878         ruleset = *rsm;
3879
3880         REASON_SET(&reason, PFRES_MATCH);
3881
3882         if (r->log)
3883                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3884                     1);
3885
3886         if (r->action != PF_PASS)
3887                 return (PF_DROP);
3888
3889         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3890                 REASON_SET(&reason, PFRES_MEMORY);
3891                 return (PF_DROP);
3892         }
3893
3894         return (PF_PASS);
3895 }
3896
3897 static int
3898 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3899         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3900         struct pf_pdesc *pd, u_short *reason, int *copyback)
3901 {
3902         struct tcphdr           *th = pd->hdr.tcp;
3903         u_int16_t                win = ntohs(th->th_win);
3904         u_int32_t                ack, end, seq, orig_seq;
3905         u_int8_t                 sws, dws;
3906         int                      ackskew;
3907
3908         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3909                 sws = src->wscale & PF_WSCALE_MASK;
3910                 dws = dst->wscale & PF_WSCALE_MASK;
3911         } else
3912                 sws = dws = 0;
3913
3914         /*
3915          * Sequence tracking algorithm from Guido van Rooij's paper:
3916          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3917          *      tcp_filtering.ps
3918          */
3919
3920         orig_seq = seq = ntohl(th->th_seq);
3921         if (src->seqlo == 0) {
3922                 /* First packet from this end. Set its state */
3923
3924                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3925                     src->scrub == NULL) {
3926                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3927                                 REASON_SET(reason, PFRES_MEMORY);
3928                                 return (PF_DROP);
3929                         }
3930                 }
3931
3932                 /* Deferred generation of sequence number modulator */
3933                 if (dst->seqdiff && !src->seqdiff) {
3934                         /* use random iss for the TCP server */
3935                         while ((src->seqdiff = arc4random() - seq) == 0)
3936                                 ;
3937                         ack = ntohl(th->th_ack) - dst->seqdiff;
3938                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3939                             src->seqdiff), 0);
3940                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3941                         *copyback = 1;
3942                 } else {
3943                         ack = ntohl(th->th_ack);
3944                 }
3945
3946                 end = seq + pd->p_len;
3947                 if (th->th_flags & TH_SYN) {
3948                         end++;
3949                         if (dst->wscale & PF_WSCALE_FLAG) {
3950                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3951                                     pd->af);
3952                                 if (src->wscale & PF_WSCALE_FLAG) {
3953                                         /* Remove scale factor from initial
3954                                          * window */
3955                                         sws = src->wscale & PF_WSCALE_MASK;
3956                                         win = ((u_int32_t)win + (1 << sws) - 1)
3957                                             >> sws;
3958                                         dws = dst->wscale & PF_WSCALE_MASK;
3959                                 } else {
3960                                         /* fixup other window */
3961                                         dst->max_win <<= dst->wscale &
3962                                             PF_WSCALE_MASK;
3963                                         /* in case of a retrans SYN|ACK */
3964                                         dst->wscale = 0;
3965                                 }
3966                         }
3967                 }
3968                 if (th->th_flags & TH_FIN)
3969                         end++;
3970
3971                 src->seqlo = seq;
3972                 if (src->state < TCPS_SYN_SENT)
3973                         src->state = TCPS_SYN_SENT;
3974
3975                 /*
3976                  * May need to slide the window (seqhi may have been set by
3977                  * the crappy stack check or if we picked up the connection
3978                  * after establishment)
3979                  */
3980                 if (src->seqhi == 1 ||
3981                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3982                         src->seqhi = end + MAX(1, dst->max_win << dws);
3983                 if (win > src->max_win)
3984                         src->max_win = win;
3985
3986         } else {
3987                 ack = ntohl(th->th_ack) - dst->seqdiff;
3988                 if (src->seqdiff) {
3989                         /* Modulate sequence numbers */
3990                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3991                             src->seqdiff), 0);
3992                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3993                         *copyback = 1;
3994                 }
3995                 end = seq + pd->p_len;
3996                 if (th->th_flags & TH_SYN)
3997                         end++;
3998                 if (th->th_flags & TH_FIN)
3999                         end++;
4000         }
4001
4002         if ((th->th_flags & TH_ACK) == 0) {
4003                 /* Let it pass through the ack skew check */
4004                 ack = dst->seqlo;
4005         } else if ((ack == 0 &&
4006             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4007             /* broken tcp stacks do not set ack */
4008             (dst->state < TCPS_SYN_SENT)) {
4009                 /*
4010                  * Many stacks (ours included) will set the ACK number in an
4011                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4012                  */
4013                 ack = dst->seqlo;
4014         }
4015
4016         if (seq == end) {
4017                 /* Ease sequencing restrictions on no data packets */
4018                 seq = src->seqlo;
4019                 end = seq;
4020         }
4021
4022         ackskew = dst->seqlo - ack;
4023
4024
4025         /*
4026          * Need to demodulate the sequence numbers in any TCP SACK options
4027          * (Selective ACK). We could optionally validate the SACK values
4028          * against the current ACK window, either forwards or backwards, but
4029          * I'm not confident that SACK has been implemented properly
4030          * everywhere. It wouldn't surprise me if several stacks accidentally
4031          * SACK too far backwards of previously ACKed data. There really aren't
4032          * any security implications of bad SACKing unless the target stack
4033          * doesn't validate the option length correctly. Someone trying to
4034          * spoof into a TCP connection won't bother blindly sending SACK
4035          * options anyway.
4036          */
4037         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4038                 if (pf_modulate_sack(m, off, pd, th, dst))
4039                         *copyback = 1;
4040         }
4041
4042
4043 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4044         if (SEQ_GEQ(src->seqhi, end) &&
4045             /* Last octet inside other's window space */
4046             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4047             /* Retrans: not more than one window back */
4048             (ackskew >= -MAXACKWINDOW) &&
4049             /* Acking not more than one reassembled fragment backwards */
4050             (ackskew <= (MAXACKWINDOW << sws)) &&
4051             /* Acking not more than one window forward */
4052             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4053             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4054             (pd->flags & PFDESC_IP_REAS) == 0)) {
4055             /* Require an exact/+1 sequence match on resets when possible */
4056
4057                 if (dst->scrub || src->scrub) {
4058                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4059                             *state, src, dst, copyback))
4060                                 return (PF_DROP);
4061                 }
4062
4063                 /* update max window */
4064                 if (src->max_win < win)
4065                         src->max_win = win;
4066                 /* synchronize sequencing */
4067                 if (SEQ_GT(end, src->seqlo))
4068                         src->seqlo = end;
4069                 /* slide the window of what the other end can send */
4070                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4071                         dst->seqhi = ack + MAX((win << sws), 1);
4072
4073
4074                 /* update states */
4075                 if (th->th_flags & TH_SYN)
4076                         if (src->state < TCPS_SYN_SENT)
4077                                 src->state = TCPS_SYN_SENT;
4078                 if (th->th_flags & TH_FIN)
4079                         if (src->state < TCPS_CLOSING)
4080                                 src->state = TCPS_CLOSING;
4081                 if (th->th_flags & TH_ACK) {
4082                         if (dst->state == TCPS_SYN_SENT) {
4083                                 dst->state = TCPS_ESTABLISHED;
4084                                 if (src->state == TCPS_ESTABLISHED &&
4085                                     (*state)->src_node != NULL &&
4086                                     pf_src_connlimit(state)) {
4087                                         REASON_SET(reason, PFRES_SRCLIMIT);
4088                                         return (PF_DROP);
4089                                 }
4090                         } else if (dst->state == TCPS_CLOSING)
4091                                 dst->state = TCPS_FIN_WAIT_2;
4092                 }
4093                 if (th->th_flags & TH_RST)
4094                         src->state = dst->state = TCPS_TIME_WAIT;
4095
4096                 /* update expire time */
4097                 (*state)->expire = time_uptime;
4098                 if (src->state >= TCPS_FIN_WAIT_2 &&
4099                     dst->state >= TCPS_FIN_WAIT_2)
4100                         (*state)->timeout = PFTM_TCP_CLOSED;
4101                 else if (src->state >= TCPS_CLOSING &&
4102                     dst->state >= TCPS_CLOSING)
4103                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4104                 else if (src->state < TCPS_ESTABLISHED ||
4105                     dst->state < TCPS_ESTABLISHED)
4106                         (*state)->timeout = PFTM_TCP_OPENING;
4107                 else if (src->state >= TCPS_CLOSING ||
4108                     dst->state >= TCPS_CLOSING)
4109                         (*state)->timeout = PFTM_TCP_CLOSING;
4110                 else
4111                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4112
4113                 /* Fall through to PASS packet */
4114
4115         } else if ((dst->state < TCPS_SYN_SENT ||
4116                 dst->state >= TCPS_FIN_WAIT_2 ||
4117                 src->state >= TCPS_FIN_WAIT_2) &&
4118             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4119             /* Within a window forward of the originating packet */
4120             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4121             /* Within a window backward of the originating packet */
4122
4123                 /*
4124                  * This currently handles three situations:
4125                  *  1) Stupid stacks will shotgun SYNs before their peer
4126                  *     replies.
4127                  *  2) When PF catches an already established stream (the
4128                  *     firewall rebooted, the state table was flushed, routes
4129                  *     changed...)
4130                  *  3) Packets get funky immediately after the connection
4131                  *     closes (this should catch Solaris spurious ACK|FINs
4132                  *     that web servers like to spew after a close)
4133                  *
4134                  * This must be a little more careful than the above code
4135                  * since packet floods will also be caught here. We don't
4136                  * update the TTL here to mitigate the damage of a packet
4137                  * flood and so the same code can handle awkward establishment
4138                  * and a loosened connection close.
4139                  * In the establishment case, a correct peer response will
4140                  * validate the connection, go through the normal state code
4141                  * and keep updating the state TTL.
4142                  */
4143
4144                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4145                         printf("pf: loose state match: ");
4146                         pf_print_state(*state);
4147                         pf_print_flags(th->th_flags);
4148                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4149                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4150                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4151                             (unsigned long long)(*state)->packets[1],
4152                             pd->dir == PF_IN ? "in" : "out",
4153                             pd->dir == (*state)->direction ? "fwd" : "rev");
4154                 }
4155
4156                 if (dst->scrub || src->scrub) {
4157                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4158                             *state, src, dst, copyback))
4159                                 return (PF_DROP);
4160                 }
4161
4162                 /* update max window */
4163                 if (src->max_win < win)
4164                         src->max_win = win;
4165                 /* synchronize sequencing */
4166                 if (SEQ_GT(end, src->seqlo))
4167                         src->seqlo = end;
4168                 /* slide the window of what the other end can send */
4169                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4170                         dst->seqhi = ack + MAX((win << sws), 1);
4171
4172                 /*
4173                  * Cannot set dst->seqhi here since this could be a shotgunned
4174                  * SYN and not an already established connection.
4175                  */
4176
4177                 if (th->th_flags & TH_FIN)
4178                         if (src->state < TCPS_CLOSING)
4179                                 src->state = TCPS_CLOSING;
4180                 if (th->th_flags & TH_RST)
4181                         src->state = dst->state = TCPS_TIME_WAIT;
4182
4183                 /* Fall through to PASS packet */
4184
4185         } else {
4186                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4187                     (*state)->src.state == TCPS_SYN_SENT) {
4188                         /* Send RST for state mismatches during handshake */
4189                         if (!(th->th_flags & TH_RST))
4190                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4191                                     pd->dst, pd->src, th->th_dport,
4192                                     th->th_sport, ntohl(th->th_ack), 0,
4193                                     TH_RST, 0, 0,
4194                                     (*state)->rule.ptr->return_ttl, 1, 0,
4195                                     kif->pfik_ifp);
4196                         src->seqlo = 0;
4197                         src->seqhi = 1;
4198                         src->max_win = 1;
4199                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4200                         printf("pf: BAD state: ");
4201                         pf_print_state(*state);
4202                         pf_print_flags(th->th_flags);
4203                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4204                             "pkts=%llu:%llu dir=%s,%s\n",
4205                             seq, orig_seq, ack, pd->p_len, ackskew,
4206                             (unsigned long long)(*state)->packets[0],
4207                             (unsigned long long)(*state)->packets[1],
4208                             pd->dir == PF_IN ? "in" : "out",
4209                             pd->dir == (*state)->direction ? "fwd" : "rev");
4210                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4211                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4212                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4213                             ' ': '2',
4214                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4215                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4216                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4217                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4218                 }
4219                 REASON_SET(reason, PFRES_BADSTATE);
4220                 return (PF_DROP);
4221         }
4222
4223         return (PF_PASS);
4224 }
4225
4226 static int
4227 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4228         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4229 {
4230         struct tcphdr           *th = pd->hdr.tcp;
4231
4232         if (th->th_flags & TH_SYN)
4233                 if (src->state < TCPS_SYN_SENT)
4234                         src->state = TCPS_SYN_SENT;
4235         if (th->th_flags & TH_FIN)
4236                 if (src->state < TCPS_CLOSING)
4237                         src->state = TCPS_CLOSING;
4238         if (th->th_flags & TH_ACK) {
4239                 if (dst->state == TCPS_SYN_SENT) {
4240                         dst->state = TCPS_ESTABLISHED;
4241                         if (src->state == TCPS_ESTABLISHED &&
4242                             (*state)->src_node != NULL &&
4243                             pf_src_connlimit(state)) {
4244                                 REASON_SET(reason, PFRES_SRCLIMIT);
4245                                 return (PF_DROP);
4246                         }
4247                 } else if (dst->state == TCPS_CLOSING) {
4248                         dst->state = TCPS_FIN_WAIT_2;
4249                 } else if (src->state == TCPS_SYN_SENT &&
4250                     dst->state < TCPS_SYN_SENT) {
4251                         /*
4252                          * Handle a special sloppy case where we only see one
4253                          * half of the connection. If there is a ACK after
4254                          * the initial SYN without ever seeing a packet from
4255                          * the destination, set the connection to established.
4256                          */
4257                         dst->state = src->state = TCPS_ESTABLISHED;
4258                         if ((*state)->src_node != NULL &&
4259                             pf_src_connlimit(state)) {
4260                                 REASON_SET(reason, PFRES_SRCLIMIT);
4261                                 return (PF_DROP);
4262                         }
4263                 } else if (src->state == TCPS_CLOSING &&
4264                     dst->state == TCPS_ESTABLISHED &&
4265                     dst->seqlo == 0) {
4266                         /*
4267                          * Handle the closing of half connections where we
4268                          * don't see the full bidirectional FIN/ACK+ACK
4269                          * handshake.
4270                          */
4271                         dst->state = TCPS_CLOSING;
4272                 }
4273         }
4274         if (th->th_flags & TH_RST)
4275                 src->state = dst->state = TCPS_TIME_WAIT;
4276
4277         /* update expire time */
4278         (*state)->expire = time_uptime;
4279         if (src->state >= TCPS_FIN_WAIT_2 &&
4280             dst->state >= TCPS_FIN_WAIT_2)
4281                 (*state)->timeout = PFTM_TCP_CLOSED;
4282         else if (src->state >= TCPS_CLOSING &&
4283             dst->state >= TCPS_CLOSING)
4284                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4285         else if (src->state < TCPS_ESTABLISHED ||
4286             dst->state < TCPS_ESTABLISHED)
4287                 (*state)->timeout = PFTM_TCP_OPENING;
4288         else if (src->state >= TCPS_CLOSING ||
4289             dst->state >= TCPS_CLOSING)
4290                 (*state)->timeout = PFTM_TCP_CLOSING;
4291         else
4292                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4293
4294         return (PF_PASS);
4295 }
4296
4297 static int
4298 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4299     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4300     u_short *reason)
4301 {
4302         struct pf_state_key_cmp  key;
4303         struct tcphdr           *th = pd->hdr.tcp;
4304         int                      copyback = 0;
4305         struct pf_state_peer    *src, *dst;
4306         struct pf_state_key     *sk;
4307
4308         bzero(&key, sizeof(key));
4309         key.af = pd->af;
4310         key.proto = IPPROTO_TCP;
4311         if (direction == PF_IN) {       /* wire side, straight */
4312                 PF_ACPY(&key.addr[0], pd->src, key.af);
4313                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4314                 key.port[0] = th->th_sport;
4315                 key.port[1] = th->th_dport;
4316         } else {                        /* stack side, reverse */
4317                 PF_ACPY(&key.addr[1], pd->src, key.af);
4318                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4319                 key.port[1] = th->th_sport;
4320                 key.port[0] = th->th_dport;
4321         }
4322
4323         STATE_LOOKUP(kif, &key, direction, *state, pd);
4324
4325         if (direction == (*state)->direction) {
4326                 src = &(*state)->src;
4327                 dst = &(*state)->dst;
4328         } else {
4329                 src = &(*state)->dst;
4330                 dst = &(*state)->src;
4331         }
4332
4333         sk = (*state)->key[pd->didx];
4334
4335         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4336                 if (direction != (*state)->direction) {
4337                         REASON_SET(reason, PFRES_SYNPROXY);
4338                         return (PF_SYNPROXY_DROP);
4339                 }
4340                 if (th->th_flags & TH_SYN) {
4341                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4342                                 REASON_SET(reason, PFRES_SYNPROXY);
4343                                 return (PF_DROP);
4344                         }
4345                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4346                             pd->src, th->th_dport, th->th_sport,
4347                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4348                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4349                         REASON_SET(reason, PFRES_SYNPROXY);
4350                         return (PF_SYNPROXY_DROP);
4351                 } else if (!(th->th_flags & TH_ACK) ||
4352                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4353                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4354                         REASON_SET(reason, PFRES_SYNPROXY);
4355                         return (PF_DROP);
4356                 } else if ((*state)->src_node != NULL &&
4357                     pf_src_connlimit(state)) {
4358                         REASON_SET(reason, PFRES_SRCLIMIT);
4359                         return (PF_DROP);
4360                 } else
4361                         (*state)->src.state = PF_TCPS_PROXY_DST;
4362         }
4363         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4364                 if (direction == (*state)->direction) {
4365                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4366                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4367                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4368                                 REASON_SET(reason, PFRES_SYNPROXY);
4369                                 return (PF_DROP);
4370                         }
4371                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4372                         if ((*state)->dst.seqhi == 1)
4373                                 (*state)->dst.seqhi = htonl(arc4random());
4374                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4375                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4376                             sk->port[pd->sidx], sk->port[pd->didx],
4377                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4378                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4379                         REASON_SET(reason, PFRES_SYNPROXY);
4380                         return (PF_SYNPROXY_DROP);
4381                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4382                     (TH_SYN|TH_ACK)) ||
4383                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4384                         REASON_SET(reason, PFRES_SYNPROXY);
4385                         return (PF_DROP);
4386                 } else {
4387                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4388                         (*state)->dst.seqlo = ntohl(th->th_seq);
4389                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4390                             pd->src, th->th_dport, th->th_sport,
4391                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4392                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4393                             (*state)->tag, NULL);
4394                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4395                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4396                             sk->port[pd->sidx], sk->port[pd->didx],
4397                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4398                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4399                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4400                             (*state)->src.seqlo;
4401                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4402                             (*state)->dst.seqlo;
4403                         (*state)->src.seqhi = (*state)->src.seqlo +
4404                             (*state)->dst.max_win;
4405                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4406                             (*state)->src.max_win;
4407                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4408                         (*state)->src.state = (*state)->dst.state =
4409                             TCPS_ESTABLISHED;
4410                         REASON_SET(reason, PFRES_SYNPROXY);
4411                         return (PF_SYNPROXY_DROP);
4412                 }
4413         }
4414
4415         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4416             dst->state >= TCPS_FIN_WAIT_2 &&
4417             src->state >= TCPS_FIN_WAIT_2) {
4418                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4419                         printf("pf: state reuse ");
4420                         pf_print_state(*state);
4421                         pf_print_flags(th->th_flags);
4422                         printf("\n");
4423                 }
4424                 /* XXX make sure it's the same direction ?? */
4425                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4426                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4427                 *state = NULL;
4428                 return (PF_DROP);
4429         }
4430
4431         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4432                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4433                         return (PF_DROP);
4434         } else {
4435                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4436                     &copyback) == PF_DROP)
4437                         return (PF_DROP);
4438         }
4439
4440         /* translate source/destination address, if necessary */
4441         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4442                 struct pf_state_key *nk = (*state)->key[pd->didx];
4443
4444                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4445                     nk->port[pd->sidx] != th->th_sport)
4446                         pf_change_ap(m, pd->src, &th->th_sport,
4447                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4448                             nk->port[pd->sidx], 0, pd->af);
4449
4450                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4451                     nk->port[pd->didx] != th->th_dport)
4452                         pf_change_ap(m, pd->dst, &th->th_dport,
4453                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4454                             nk->port[pd->didx], 0, pd->af);
4455                 copyback = 1;
4456         }
4457
4458         /* Copyback sequence modulation or stateful scrub changes if needed */
4459         if (copyback)
4460                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4461
4462         return (PF_PASS);
4463 }
4464
4465 static int
4466 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4467     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4468 {
4469         struct pf_state_peer    *src, *dst;
4470         struct pf_state_key_cmp  key;
4471         struct udphdr           *uh = pd->hdr.udp;
4472
4473         bzero(&key, sizeof(key));
4474         key.af = pd->af;
4475         key.proto = IPPROTO_UDP;
4476         if (direction == PF_IN) {       /* wire side, straight */
4477                 PF_ACPY(&key.addr[0], pd->src, key.af);
4478                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4479                 key.port[0] = uh->uh_sport;
4480                 key.port[1] = uh->uh_dport;
4481         } else {                        /* stack side, reverse */
4482                 PF_ACPY(&key.addr[1], pd->src, key.af);
4483                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4484                 key.port[1] = uh->uh_sport;
4485                 key.port[0] = uh->uh_dport;
4486         }
4487
4488         STATE_LOOKUP(kif, &key, direction, *state, pd);
4489
4490         if (direction == (*state)->direction) {
4491                 src = &(*state)->src;
4492                 dst = &(*state)->dst;
4493         } else {
4494                 src = &(*state)->dst;
4495                 dst = &(*state)->src;
4496         }
4497
4498         /* update states */
4499         if (src->state < PFUDPS_SINGLE)
4500                 src->state = PFUDPS_SINGLE;
4501         if (dst->state == PFUDPS_SINGLE)
4502                 dst->state = PFUDPS_MULTIPLE;
4503
4504         /* update expire time */
4505         (*state)->expire = time_uptime;
4506         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4507                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4508         else
4509                 (*state)->timeout = PFTM_UDP_SINGLE;
4510
4511         /* translate source/destination address, if necessary */
4512         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4513                 struct pf_state_key *nk = (*state)->key[pd->didx];
4514
4515                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4516                     nk->port[pd->sidx] != uh->uh_sport)
4517                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4518                             &uh->uh_sum, &nk->addr[pd->sidx],
4519                             nk->port[pd->sidx], 1, pd->af);
4520
4521                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4522                     nk->port[pd->didx] != uh->uh_dport)
4523                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4524                             &uh->uh_sum, &nk->addr[pd->didx],
4525                             nk->port[pd->didx], 1, pd->af);
4526                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4527         }
4528
4529         return (PF_PASS);
4530 }
4531
4532 static int
4533 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4534     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4535 {
4536         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4537         u_int16_t        icmpid = 0, *icmpsum;
4538         u_int8_t         icmptype;
4539         int              state_icmp = 0;
4540         struct pf_state_key_cmp key;
4541
4542         bzero(&key, sizeof(key));
4543         switch (pd->proto) {
4544 #ifdef INET
4545         case IPPROTO_ICMP:
4546                 icmptype = pd->hdr.icmp->icmp_type;
4547                 icmpid = pd->hdr.icmp->icmp_id;
4548                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4549
4550                 if (icmptype == ICMP_UNREACH ||
4551                     icmptype == ICMP_SOURCEQUENCH ||
4552                     icmptype == ICMP_REDIRECT ||
4553                     icmptype == ICMP_TIMXCEED ||
4554                     icmptype == ICMP_PARAMPROB)
4555                         state_icmp++;
4556                 break;
4557 #endif /* INET */
4558 #ifdef INET6
4559         case IPPROTO_ICMPV6:
4560                 icmptype = pd->hdr.icmp6->icmp6_type;
4561                 icmpid = pd->hdr.icmp6->icmp6_id;
4562                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4563
4564                 if (icmptype == ICMP6_DST_UNREACH ||
4565                     icmptype == ICMP6_PACKET_TOO_BIG ||
4566                     icmptype == ICMP6_TIME_EXCEEDED ||
4567                     icmptype == ICMP6_PARAM_PROB)
4568                         state_icmp++;
4569                 break;
4570 #endif /* INET6 */
4571         }
4572
4573         if (!state_icmp) {
4574
4575                 /*
4576                  * ICMP query/reply message not related to a TCP/UDP packet.
4577                  * Search for an ICMP state.
4578                  */
4579                 key.af = pd->af;
4580                 key.proto = pd->proto;
4581                 key.port[0] = key.port[1] = icmpid;
4582                 if (direction == PF_IN) {       /* wire side, straight */
4583                         PF_ACPY(&key.addr[0], pd->src, key.af);
4584                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4585                 } else {                        /* stack side, reverse */
4586                         PF_ACPY(&key.addr[1], pd->src, key.af);
4587                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4588                 }
4589
4590                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4591
4592                 (*state)->expire = time_uptime;
4593                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4594
4595                 /* translate source/destination address, if necessary */
4596                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4597                         struct pf_state_key *nk = (*state)->key[pd->didx];
4598
4599                         switch (pd->af) {
4600 #ifdef INET
4601                         case AF_INET:
4602                                 if (PF_ANEQ(pd->src,
4603                                     &nk->addr[pd->sidx], AF_INET))
4604                                         pf_change_a(&saddr->v4.s_addr,
4605                                             pd->ip_sum,
4606                                             nk->addr[pd->sidx].v4.s_addr, 0);
4607
4608                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4609                                     AF_INET))
4610                                         pf_change_a(&daddr->v4.s_addr,
4611                                             pd->ip_sum,
4612                                             nk->addr[pd->didx].v4.s_addr, 0);
4613
4614                                 if (nk->port[0] !=
4615                                     pd->hdr.icmp->icmp_id) {
4616                                         pd->hdr.icmp->icmp_cksum =
4617                                             pf_cksum_fixup(
4618                                             pd->hdr.icmp->icmp_cksum, icmpid,
4619                                             nk->port[pd->sidx], 0);
4620                                         pd->hdr.icmp->icmp_id =
4621                                             nk->port[pd->sidx];
4622                                 }
4623
4624                                 m_copyback(m, off, ICMP_MINLEN,
4625                                     (caddr_t )pd->hdr.icmp);
4626                                 break;
4627 #endif /* INET */
4628 #ifdef INET6
4629                         case AF_INET6:
4630                                 if (PF_ANEQ(pd->src,
4631                                     &nk->addr[pd->sidx], AF_INET6))
4632                                         pf_change_a6(saddr,
4633                                             &pd->hdr.icmp6->icmp6_cksum,
4634                                             &nk->addr[pd->sidx], 0);
4635
4636                                 if (PF_ANEQ(pd->dst,
4637                                     &nk->addr[pd->didx], AF_INET6))
4638                                         pf_change_a6(daddr,
4639                                             &pd->hdr.icmp6->icmp6_cksum,
4640                                             &nk->addr[pd->didx], 0);
4641
4642                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4643                                     (caddr_t )pd->hdr.icmp6);
4644                                 break;
4645 #endif /* INET6 */
4646                         }
4647                 }
4648                 return (PF_PASS);
4649
4650         } else {
4651                 /*
4652                  * ICMP error message in response to a TCP/UDP packet.
4653                  * Extract the inner TCP/UDP header and search for that state.
4654                  */
4655
4656                 struct pf_pdesc pd2;
4657                 bzero(&pd2, sizeof pd2);
4658 #ifdef INET
4659                 struct ip       h2;
4660 #endif /* INET */
4661 #ifdef INET6
4662                 struct ip6_hdr  h2_6;
4663                 int             terminal = 0;
4664 #endif /* INET6 */
4665                 int             ipoff2 = 0;
4666                 int             off2 = 0;
4667
4668                 pd2.af = pd->af;
4669                 /* Payload packet is from the opposite direction. */
4670                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4671                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4672                 switch (pd->af) {
4673 #ifdef INET
4674                 case AF_INET:
4675                         /* offset of h2 in mbuf chain */
4676                         ipoff2 = off + ICMP_MINLEN;
4677
4678                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4679                             NULL, reason, pd2.af)) {
4680                                 DPFPRINTF(PF_DEBUG_MISC,
4681                                     ("pf: ICMP error message too short "
4682                                     "(ip)\n"));
4683                                 return (PF_DROP);
4684                         }
4685                         /*
4686                          * ICMP error messages don't refer to non-first
4687                          * fragments
4688                          */
4689                         if (h2.ip_off & htons(IP_OFFMASK)) {
4690                                 REASON_SET(reason, PFRES_FRAG);
4691                                 return (PF_DROP);
4692                         }
4693
4694                         /* offset of protocol header that follows h2 */
4695                         off2 = ipoff2 + (h2.ip_hl << 2);
4696
4697                         pd2.proto = h2.ip_p;
4698                         pd2.src = (struct pf_addr *)&h2.ip_src;
4699                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4700                         pd2.ip_sum = &h2.ip_sum;
4701                         break;
4702 #endif /* INET */
4703 #ifdef INET6
4704                 case AF_INET6:
4705                         ipoff2 = off + sizeof(struct icmp6_hdr);
4706
4707                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4708                             NULL, reason, pd2.af)) {
4709                                 DPFPRINTF(PF_DEBUG_MISC,
4710                                     ("pf: ICMP error message too short "
4711                                     "(ip6)\n"));
4712                                 return (PF_DROP);
4713                         }
4714                         pd2.proto = h2_6.ip6_nxt;
4715                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4716                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4717                         pd2.ip_sum = NULL;
4718                         off2 = ipoff2 + sizeof(h2_6);
4719                         do {
4720                                 switch (pd2.proto) {
4721                                 case IPPROTO_FRAGMENT:
4722                                         /*
4723                                          * ICMPv6 error messages for
4724                                          * non-first fragments
4725                                          */
4726                                         REASON_SET(reason, PFRES_FRAG);
4727                                         return (PF_DROP);
4728                                 case IPPROTO_AH:
4729                                 case IPPROTO_HOPOPTS:
4730                                 case IPPROTO_ROUTING:
4731                                 case IPPROTO_DSTOPTS: {
4732                                         /* get next header and header length */
4733                                         struct ip6_ext opt6;
4734
4735                                         if (!pf_pull_hdr(m, off2, &opt6,
4736                                             sizeof(opt6), NULL, reason,
4737                                             pd2.af)) {
4738                                                 DPFPRINTF(PF_DEBUG_MISC,
4739                                                     ("pf: ICMPv6 short opt\n"));
4740                                                 return (PF_DROP);
4741                                         }
4742                                         if (pd2.proto == IPPROTO_AH)
4743                                                 off2 += (opt6.ip6e_len + 2) * 4;
4744                                         else
4745                                                 off2 += (opt6.ip6e_len + 1) * 8;
4746                                         pd2.proto = opt6.ip6e_nxt;
4747                                         /* goto the next header */
4748                                         break;
4749                                 }
4750                                 default:
4751                                         terminal++;
4752                                         break;
4753                                 }
4754                         } while (!terminal);
4755                         break;
4756 #endif /* INET6 */
4757                 }
4758
4759                 switch (pd2.proto) {
4760                 case IPPROTO_TCP: {
4761                         struct tcphdr            th;
4762                         u_int32_t                seq;
4763                         struct pf_state_peer    *src, *dst;
4764                         u_int8_t                 dws;
4765                         int                      copyback = 0;
4766
4767                         /*
4768                          * Only the first 8 bytes of the TCP header can be
4769                          * expected. Don't access any TCP header fields after
4770                          * th_seq, an ackskew test is not possible.
4771                          */
4772                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4773                             pd2.af)) {
4774                                 DPFPRINTF(PF_DEBUG_MISC,
4775                                     ("pf: ICMP error message too short "
4776                                     "(tcp)\n"));
4777                                 return (PF_DROP);
4778                         }
4779
4780                         key.af = pd2.af;
4781                         key.proto = IPPROTO_TCP;
4782                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4783                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4784                         key.port[pd2.sidx] = th.th_sport;
4785                         key.port[pd2.didx] = th.th_dport;
4786
4787                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4788
4789                         if (direction == (*state)->direction) {
4790                                 src = &(*state)->dst;
4791                                 dst = &(*state)->src;
4792                         } else {
4793                                 src = &(*state)->src;
4794                                 dst = &(*state)->dst;
4795                         }
4796
4797                         if (src->wscale && dst->wscale)
4798                                 dws = dst->wscale & PF_WSCALE_MASK;
4799                         else
4800                                 dws = 0;
4801
4802                         /* Demodulate sequence number */
4803                         seq = ntohl(th.th_seq) - src->seqdiff;
4804                         if (src->seqdiff) {
4805                                 pf_change_a(&th.th_seq, icmpsum,
4806                                     htonl(seq), 0);
4807                                 copyback = 1;
4808                         }
4809
4810                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4811                             (!SEQ_GEQ(src->seqhi, seq) ||
4812                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4813                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4814                                         printf("pf: BAD ICMP %d:%d ",
4815                                             icmptype, pd->hdr.icmp->icmp_code);
4816                                         pf_print_host(pd->src, 0, pd->af);
4817                                         printf(" -> ");
4818                                         pf_print_host(pd->dst, 0, pd->af);
4819                                         printf(" state: ");
4820                                         pf_print_state(*state);
4821                                         printf(" seq=%u\n", seq);
4822                                 }
4823                                 REASON_SET(reason, PFRES_BADSTATE);
4824                                 return (PF_DROP);
4825                         } else {
4826                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4827                                         printf("pf: OK ICMP %d:%d ",
4828                                             icmptype, pd->hdr.icmp->icmp_code);
4829                                         pf_print_host(pd->src, 0, pd->af);
4830                                         printf(" -> ");
4831                                         pf_print_host(pd->dst, 0, pd->af);
4832                                         printf(" state: ");
4833                                         pf_print_state(*state);
4834                                         printf(" seq=%u\n", seq);
4835                                 }
4836                         }
4837
4838                         /* translate source/destination address, if necessary */
4839                         if ((*state)->key[PF_SK_WIRE] !=
4840                             (*state)->key[PF_SK_STACK]) {
4841                                 struct pf_state_key *nk =
4842                                     (*state)->key[pd->didx];
4843
4844                                 if (PF_ANEQ(pd2.src,
4845                                     &nk->addr[pd2.sidx], pd2.af) ||
4846                                     nk->port[pd2.sidx] != th.th_sport)
4847                                         pf_change_icmp(pd2.src, &th.th_sport,
4848                                             daddr, &nk->addr[pd2.sidx],
4849                                             nk->port[pd2.sidx], NULL,
4850                                             pd2.ip_sum, icmpsum,
4851                                             pd->ip_sum, 0, pd2.af);
4852
4853                                 if (PF_ANEQ(pd2.dst,
4854                                     &nk->addr[pd2.didx], pd2.af) ||
4855                                     nk->port[pd2.didx] != th.th_dport)
4856                                         pf_change_icmp(pd2.dst, &th.th_dport,
4857                                             saddr, &nk->addr[pd2.didx],
4858                                             nk->port[pd2.didx], NULL,
4859                                             pd2.ip_sum, icmpsum,
4860                                             pd->ip_sum, 0, pd2.af);
4861                                 copyback = 1;
4862                         }
4863
4864                         if (copyback) {
4865                                 switch (pd2.af) {
4866 #ifdef INET
4867                                 case AF_INET:
4868                                         m_copyback(m, off, ICMP_MINLEN,
4869                                             (caddr_t )pd->hdr.icmp);
4870                                         m_copyback(m, ipoff2, sizeof(h2),
4871                                             (caddr_t )&h2);
4872                                         break;
4873 #endif /* INET */
4874 #ifdef INET6
4875                                 case AF_INET6:
4876                                         m_copyback(m, off,
4877                                             sizeof(struct icmp6_hdr),
4878                                             (caddr_t )pd->hdr.icmp6);
4879                                         m_copyback(m, ipoff2, sizeof(h2_6),
4880                                             (caddr_t )&h2_6);
4881                                         break;
4882 #endif /* INET6 */
4883                                 }
4884                                 m_copyback(m, off2, 8, (caddr_t)&th);
4885                         }
4886
4887                         return (PF_PASS);
4888                         break;
4889                 }
4890                 case IPPROTO_UDP: {
4891                         struct udphdr           uh;
4892
4893                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4894                             NULL, reason, pd2.af)) {
4895                                 DPFPRINTF(PF_DEBUG_MISC,
4896                                     ("pf: ICMP error message too short "
4897                                     "(udp)\n"));
4898                                 return (PF_DROP);
4899                         }
4900
4901                         key.af = pd2.af;
4902                         key.proto = IPPROTO_UDP;
4903                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4904                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4905                         key.port[pd2.sidx] = uh.uh_sport;
4906                         key.port[pd2.didx] = uh.uh_dport;
4907
4908                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4909
4910                         /* translate source/destination address, if necessary */
4911                         if ((*state)->key[PF_SK_WIRE] !=
4912                             (*state)->key[PF_SK_STACK]) {
4913                                 struct pf_state_key *nk =
4914                                     (*state)->key[pd->didx];
4915
4916                                 if (PF_ANEQ(pd2.src,
4917                                     &nk->addr[pd2.sidx], pd2.af) ||
4918                                     nk->port[pd2.sidx] != uh.uh_sport)
4919                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4920                                             daddr, &nk->addr[pd2.sidx],
4921                                             nk->port[pd2.sidx], &uh.uh_sum,
4922                                             pd2.ip_sum, icmpsum,
4923                                             pd->ip_sum, 1, pd2.af);
4924
4925                                 if (PF_ANEQ(pd2.dst,
4926                                     &nk->addr[pd2.didx], pd2.af) ||
4927                                     nk->port[pd2.didx] != uh.uh_dport)
4928                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4929                                             saddr, &nk->addr[pd2.didx],
4930                                             nk->port[pd2.didx], &uh.uh_sum,
4931                                             pd2.ip_sum, icmpsum,
4932                                             pd->ip_sum, 1, pd2.af);
4933
4934                                 switch (pd2.af) {
4935 #ifdef INET
4936                                 case AF_INET:
4937                                         m_copyback(m, off, ICMP_MINLEN,
4938                                             (caddr_t )pd->hdr.icmp);
4939                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4940                                         break;
4941 #endif /* INET */
4942 #ifdef INET6
4943                                 case AF_INET6:
4944                                         m_copyback(m, off,
4945                                             sizeof(struct icmp6_hdr),
4946                                             (caddr_t )pd->hdr.icmp6);
4947                                         m_copyback(m, ipoff2, sizeof(h2_6),
4948                                             (caddr_t )&h2_6);
4949                                         break;
4950 #endif /* INET6 */
4951                                 }
4952                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4953                         }
4954                         return (PF_PASS);
4955                         break;
4956                 }
4957 #ifdef INET
4958                 case IPPROTO_ICMP: {
4959                         struct icmp             iih;
4960
4961                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4962                             NULL, reason, pd2.af)) {
4963                                 DPFPRINTF(PF_DEBUG_MISC,
4964                                     ("pf: ICMP error message too short i"
4965                                     "(icmp)\n"));
4966                                 return (PF_DROP);
4967                         }
4968
4969                         key.af = pd2.af;
4970                         key.proto = IPPROTO_ICMP;
4971                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4972                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4973                         key.port[0] = key.port[1] = iih.icmp_id;
4974
4975                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4976
4977                         /* translate source/destination address, if necessary */
4978                         if ((*state)->key[PF_SK_WIRE] !=
4979                             (*state)->key[PF_SK_STACK]) {
4980                                 struct pf_state_key *nk =
4981                                     (*state)->key[pd->didx];
4982
4983                                 if (PF_ANEQ(pd2.src,
4984                                     &nk->addr[pd2.sidx], pd2.af) ||
4985                                     nk->port[pd2.sidx] != iih.icmp_id)
4986                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4987                                             daddr, &nk->addr[pd2.sidx],
4988                                             nk->port[pd2.sidx], NULL,
4989                                             pd2.ip_sum, icmpsum,
4990                                             pd->ip_sum, 0, AF_INET);
4991
4992                                 if (PF_ANEQ(pd2.dst,
4993                                     &nk->addr[pd2.didx], pd2.af) ||
4994                                     nk->port[pd2.didx] != iih.icmp_id)
4995                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4996                                             saddr, &nk->addr[pd2.didx],
4997                                             nk->port[pd2.didx], NULL,
4998                                             pd2.ip_sum, icmpsum,
4999                                             pd->ip_sum, 0, AF_INET);
5000
5001                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5002                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5003                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5004                         }
5005                         return (PF_PASS);
5006                         break;
5007                 }
5008 #endif /* INET */
5009 #ifdef INET6
5010                 case IPPROTO_ICMPV6: {
5011                         struct icmp6_hdr        iih;
5012
5013                         if (!pf_pull_hdr(m, off2, &iih,
5014                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5015                                 DPFPRINTF(PF_DEBUG_MISC,
5016                                     ("pf: ICMP error message too short "
5017                                     "(icmp6)\n"));
5018                                 return (PF_DROP);
5019                         }
5020
5021                         key.af = pd2.af;
5022                         key.proto = IPPROTO_ICMPV6;
5023                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5024                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5025                         key.port[0] = key.port[1] = iih.icmp6_id;
5026
5027                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5028
5029                         /* translate source/destination address, if necessary */
5030                         if ((*state)->key[PF_SK_WIRE] !=
5031                             (*state)->key[PF_SK_STACK]) {
5032                                 struct pf_state_key *nk =
5033                                     (*state)->key[pd->didx];
5034
5035                                 if (PF_ANEQ(pd2.src,
5036                                     &nk->addr[pd2.sidx], pd2.af) ||
5037                                     nk->port[pd2.sidx] != iih.icmp6_id)
5038                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5039                                             daddr, &nk->addr[pd2.sidx],
5040                                             nk->port[pd2.sidx], NULL,
5041                                             pd2.ip_sum, icmpsum,
5042                                             pd->ip_sum, 0, AF_INET6);
5043
5044                                 if (PF_ANEQ(pd2.dst,
5045                                     &nk->addr[pd2.didx], pd2.af) ||
5046                                     nk->port[pd2.didx] != iih.icmp6_id)
5047                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5048                                             saddr, &nk->addr[pd2.didx],
5049                                             nk->port[pd2.didx], NULL,
5050                                             pd2.ip_sum, icmpsum,
5051                                             pd->ip_sum, 0, AF_INET6);
5052
5053                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5054                                     (caddr_t)pd->hdr.icmp6);
5055                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5056                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5057                                     (caddr_t)&iih);
5058                         }
5059                         return (PF_PASS);
5060                         break;
5061                 }
5062 #endif /* INET6 */
5063                 default: {
5064                         key.af = pd2.af;
5065                         key.proto = pd2.proto;
5066                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5067                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5068                         key.port[0] = key.port[1] = 0;
5069
5070                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5071
5072                         /* translate source/destination address, if necessary */
5073                         if ((*state)->key[PF_SK_WIRE] !=
5074                             (*state)->key[PF_SK_STACK]) {
5075                                 struct pf_state_key *nk =
5076                                     (*state)->key[pd->didx];
5077
5078                                 if (PF_ANEQ(pd2.src,
5079                                     &nk->addr[pd2.sidx], pd2.af))
5080                                         pf_change_icmp(pd2.src, NULL, daddr,
5081                                             &nk->addr[pd2.sidx], 0, NULL,
5082                                             pd2.ip_sum, icmpsum,
5083                                             pd->ip_sum, 0, pd2.af);
5084
5085                                 if (PF_ANEQ(pd2.dst,
5086                                     &nk->addr[pd2.didx], pd2.af))
5087                                         pf_change_icmp(pd2.dst, NULL, saddr,
5088                                             &nk->addr[pd2.didx], 0, NULL,
5089                                             pd2.ip_sum, icmpsum,
5090                                             pd->ip_sum, 0, pd2.af);
5091
5092                                 switch (pd2.af) {
5093 #ifdef INET
5094                                 case AF_INET:
5095                                         m_copyback(m, off, ICMP_MINLEN,
5096                                             (caddr_t)pd->hdr.icmp);
5097                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5098                                         break;
5099 #endif /* INET */
5100 #ifdef INET6
5101                                 case AF_INET6:
5102                                         m_copyback(m, off,
5103                                             sizeof(struct icmp6_hdr),
5104                                             (caddr_t )pd->hdr.icmp6);
5105                                         m_copyback(m, ipoff2, sizeof(h2_6),
5106                                             (caddr_t )&h2_6);
5107                                         break;
5108 #endif /* INET6 */
5109                                 }
5110                         }
5111                         return (PF_PASS);
5112                         break;
5113                 }
5114                 }
5115         }
5116 }
5117
5118 static int
5119 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5120     struct mbuf *m, struct pf_pdesc *pd)
5121 {
5122         struct pf_state_peer    *src, *dst;
5123         struct pf_state_key_cmp  key;
5124
5125         bzero(&key, sizeof(key));
5126         key.af = pd->af;
5127         key.proto = pd->proto;
5128         if (direction == PF_IN) {
5129                 PF_ACPY(&key.addr[0], pd->src, key.af);
5130                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5131                 key.port[0] = key.port[1] = 0;
5132         } else {
5133                 PF_ACPY(&key.addr[1], pd->src, key.af);
5134                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5135                 key.port[1] = key.port[0] = 0;
5136         }
5137
5138         STATE_LOOKUP(kif, &key, direction, *state, pd);
5139
5140         if (direction == (*state)->direction) {
5141                 src = &(*state)->src;
5142                 dst = &(*state)->dst;
5143         } else {
5144                 src = &(*state)->dst;
5145                 dst = &(*state)->src;
5146         }
5147
5148         /* update states */
5149         if (src->state < PFOTHERS_SINGLE)
5150                 src->state = PFOTHERS_SINGLE;
5151         if (dst->state == PFOTHERS_SINGLE)
5152                 dst->state = PFOTHERS_MULTIPLE;
5153
5154         /* update expire time */
5155         (*state)->expire = time_uptime;
5156         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5157                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5158         else
5159                 (*state)->timeout = PFTM_OTHER_SINGLE;
5160
5161         /* translate source/destination address, if necessary */
5162         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5163                 struct pf_state_key *nk = (*state)->key[pd->didx];
5164
5165                 KASSERT(nk, ("%s: nk is null", __func__));
5166                 KASSERT(pd, ("%s: pd is null", __func__));
5167                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5168                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5169                 switch (pd->af) {
5170 #ifdef INET
5171                 case AF_INET:
5172                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5173                                 pf_change_a(&pd->src->v4.s_addr,
5174                                     pd->ip_sum,
5175                                     nk->addr[pd->sidx].v4.s_addr,
5176                                     0);
5177
5178
5179                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5180                                 pf_change_a(&pd->dst->v4.s_addr,
5181                                     pd->ip_sum,
5182                                     nk->addr[pd->didx].v4.s_addr,
5183                                     0);
5184
5185                                 break;
5186 #endif /* INET */
5187 #ifdef INET6
5188                 case AF_INET6:
5189                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5190                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5191
5192                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5193                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5194 #endif /* INET6 */
5195                 }
5196         }
5197         return (PF_PASS);
5198 }
5199
5200 /*
5201  * ipoff and off are measured from the start of the mbuf chain.
5202  * h must be at "ipoff" on the mbuf chain.
5203  */
5204 void *
5205 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5206     u_short *actionp, u_short *reasonp, sa_family_t af)
5207 {
5208         switch (af) {
5209 #ifdef INET
5210         case AF_INET: {
5211                 struct ip       *h = mtod(m, struct ip *);
5212                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5213
5214                 if (fragoff) {
5215                         if (fragoff >= len)
5216                                 ACTION_SET(actionp, PF_PASS);
5217                         else {
5218                                 ACTION_SET(actionp, PF_DROP);
5219                                 REASON_SET(reasonp, PFRES_FRAG);
5220                         }
5221                         return (NULL);
5222                 }
5223                 if (m->m_pkthdr.len < off + len ||
5224                     ntohs(h->ip_len) < off + len) {
5225                         ACTION_SET(actionp, PF_DROP);
5226                         REASON_SET(reasonp, PFRES_SHORT);
5227                         return (NULL);
5228                 }
5229                 break;
5230         }
5231 #endif /* INET */
5232 #ifdef INET6
5233         case AF_INET6: {
5234                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5235
5236                 if (m->m_pkthdr.len < off + len ||
5237                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5238                     (unsigned)(off + len)) {
5239                         ACTION_SET(actionp, PF_DROP);
5240                         REASON_SET(reasonp, PFRES_SHORT);
5241                         return (NULL);
5242                 }
5243                 break;
5244         }
5245 #endif /* INET6 */
5246         }
5247         m_copydata(m, off, len, p);
5248         return (p);
5249 }
5250
5251 #ifdef RADIX_MPATH
5252 static int
5253 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5254     int rtableid)
5255 {
5256         struct radix_node_head  *rnh;
5257         struct sockaddr_in      *dst;
5258         int                      ret = 1;
5259         int                      check_mpath;
5260 #ifdef INET6
5261         struct sockaddr_in6     *dst6;
5262         struct route_in6         ro;
5263 #else
5264         struct route             ro;
5265 #endif
5266         struct radix_node       *rn;
5267         struct rtentry          *rt;
5268         struct ifnet            *ifp;
5269
5270         check_mpath = 0;
5271         /* XXX: stick to table 0 for now */
5272         rnh = rt_tables_get_rnh(0, af);
5273         if (rnh != NULL && rn_mpath_capable(rnh))
5274                 check_mpath = 1;
5275         bzero(&ro, sizeof(ro));
5276         switch (af) {
5277         case AF_INET:
5278                 dst = satosin(&ro.ro_dst);
5279                 dst->sin_family = AF_INET;
5280                 dst->sin_len = sizeof(*dst);
5281                 dst->sin_addr = addr->v4;
5282                 break;
5283 #ifdef INET6
5284         case AF_INET6:
5285                 /*
5286                  * Skip check for addresses with embedded interface scope,
5287                  * as they would always match anyway.
5288                  */
5289                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5290                         goto out;
5291                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5292                 dst6->sin6_family = AF_INET6;
5293                 dst6->sin6_len = sizeof(*dst6);
5294                 dst6->sin6_addr = addr->v6;
5295                 break;
5296 #endif /* INET6 */
5297         default:
5298                 return (0);
5299         }
5300
5301         /* Skip checks for ipsec interfaces */
5302         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5303                 goto out;
5304
5305         switch (af) {
5306 #ifdef INET6
5307         case AF_INET6:
5308                 in6_rtalloc_ign(&ro, 0, rtableid);
5309                 break;
5310 #endif
5311 #ifdef INET
5312         case AF_INET:
5313                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5314                 break;
5315 #endif
5316         }
5317
5318         if (ro.ro_rt != NULL) {
5319                 /* No interface given, this is a no-route check */
5320                 if (kif == NULL)
5321                         goto out;
5322
5323                 if (kif->pfik_ifp == NULL) {
5324                         ret = 0;
5325                         goto out;
5326                 }
5327
5328                 /* Perform uRPF check if passed input interface */
5329                 ret = 0;
5330                 rn = (struct radix_node *)ro.ro_rt;
5331                 do {
5332                         rt = (struct rtentry *)rn;
5333                         ifp = rt->rt_ifp;
5334
5335                         if (kif->pfik_ifp == ifp)
5336                                 ret = 1;
5337                         rn = rn_mpath_next(rn);
5338                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5339         } else
5340                 ret = 0;
5341 out:
5342         if (ro.ro_rt != NULL)
5343                 RTFREE(ro.ro_rt);
5344         return (ret);
5345 }
5346 #endif
5347
5348 int
5349 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5350     int rtableid)
5351 {
5352 #ifdef INET
5353         struct nhop4_basic      nh4;
5354 #endif
5355 #ifdef INET6
5356         struct nhop6_basic      nh6;
5357 #endif
5358         struct ifnet            *ifp;
5359 #ifdef RADIX_MPATH
5360         struct radix_node_head  *rnh;
5361
5362         /* XXX: stick to table 0 for now */
5363         rnh = rt_tables_get_rnh(0, af);
5364         if (rnh != NULL && rn_mpath_capable(rnh))
5365                 return (pf_routable_oldmpath(addr, af, kif, rtableid));
5366 #endif
5367         /*
5368          * Skip check for addresses with embedded interface scope,
5369          * as they would always match anyway.
5370          */
5371         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5372                 return (1);
5373
5374         if (af != AF_INET && af != AF_INET6)
5375                 return (0);
5376
5377         /* Skip checks for ipsec interfaces */
5378         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5379                 return (1);
5380
5381         ifp = NULL;
5382
5383         switch (af) {
5384 #ifdef INET6
5385         case AF_INET6:
5386                 if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5387                         return (0);
5388                 ifp = nh6.nh_ifp;
5389                 break;
5390 #endif
5391 #ifdef INET
5392         case AF_INET:
5393                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5394                         return (0);
5395                 ifp = nh4.nh_ifp;
5396                 break;
5397 #endif
5398         }
5399
5400         /* No interface given, this is a no-route check */
5401         if (kif == NULL)
5402                 return (1);
5403
5404         if (kif->pfik_ifp == NULL)
5405                 return (0);
5406
5407         /* Perform uRPF check if passed input interface */
5408         if (kif->pfik_ifp == ifp)
5409                 return (1);
5410         return (0);
5411 }
5412
5413 #ifdef INET
5414 static void
5415 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5416     struct pf_state *s, struct pf_pdesc *pd)
5417 {
5418         struct mbuf             *m0, *m1;
5419         struct sockaddr_in      dst;
5420         struct ip               *ip;
5421         struct ifnet            *ifp = NULL;
5422         struct pf_addr           naddr;
5423         struct pf_src_node      *sn = NULL;
5424         int                      error = 0;
5425         uint16_t                 ip_len, ip_off;
5426
5427         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5428         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5429             __func__));
5430
5431         if ((pd->pf_mtag == NULL &&
5432             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5433             pd->pf_mtag->routed++ > 3) {
5434                 m0 = *m;
5435                 *m = NULL;
5436                 goto bad_locked;
5437         }
5438
5439         if (r->rt == PF_DUPTO) {
5440                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5441                         if (s)
5442                                 PF_STATE_UNLOCK(s);
5443                         return;
5444                 }
5445         } else {
5446                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5447                         if (s)
5448                                 PF_STATE_UNLOCK(s);
5449                         return;
5450                 }
5451                 m0 = *m;
5452         }
5453
5454         ip = mtod(m0, struct ip *);
5455
5456         bzero(&dst, sizeof(dst));
5457         dst.sin_family = AF_INET;
5458         dst.sin_len = sizeof(dst);
5459         dst.sin_addr = ip->ip_dst;
5460
5461         if (TAILQ_EMPTY(&r->rpool.list)) {
5462                 DPFPRINTF(PF_DEBUG_URGENT,
5463                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5464                 goto bad_locked;
5465         }
5466         if (s == NULL) {
5467                 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5468                     &naddr, NULL, &sn);
5469                 if (!PF_AZERO(&naddr, AF_INET))
5470                         dst.sin_addr.s_addr = naddr.v4.s_addr;
5471                 ifp = r->rpool.cur->kif ?
5472                     r->rpool.cur->kif->pfik_ifp : NULL;
5473         } else {
5474                 if (!PF_AZERO(&s->rt_addr, AF_INET))
5475                         dst.sin_addr.s_addr =
5476                             s->rt_addr.v4.s_addr;
5477                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5478                 PF_STATE_UNLOCK(s);
5479         }
5480         if (ifp == NULL)
5481                 goto bad;
5482
5483         if (oifp != ifp) {
5484                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5485                         goto bad;
5486                 else if (m0 == NULL)
5487                         goto done;
5488                 if (m0->m_len < sizeof(struct ip)) {
5489                         DPFPRINTF(PF_DEBUG_URGENT,
5490                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5491                         goto bad;
5492                 }
5493                 ip = mtod(m0, struct ip *);
5494         }
5495
5496         if (ifp->if_flags & IFF_LOOPBACK)
5497                 m0->m_flags |= M_SKIP_FIREWALL;
5498
5499         ip_len = ntohs(ip->ip_len);
5500         ip_off = ntohs(ip->ip_off);
5501
5502         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5503         m0->m_pkthdr.csum_flags |= CSUM_IP;
5504         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5505                 in_delayed_cksum(m0);
5506                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5507         }
5508 #ifdef SCTP
5509         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5510                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5511                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5512         }
5513 #endif
5514
5515         /*
5516          * If small enough for interface, or the interface will take
5517          * care of the fragmentation for us, we can just send directly.
5518          */
5519         if (ip_len <= ifp->if_mtu ||
5520             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5521                 ip->ip_sum = 0;
5522                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5523                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5524                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5525                 }
5526                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5527                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5528                 goto done;
5529         }
5530
5531         /* Balk when DF bit is set or the interface didn't support TSO. */
5532         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5533                 error = EMSGSIZE;
5534                 KMOD_IPSTAT_INC(ips_cantfrag);
5535                 if (r->rt != PF_DUPTO) {
5536                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5537                             ifp->if_mtu);
5538                         goto done;
5539                 } else
5540                         goto bad;
5541         }
5542
5543         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5544         if (error)
5545                 goto bad;
5546
5547         for (; m0; m0 = m1) {
5548                 m1 = m0->m_nextpkt;
5549                 m0->m_nextpkt = NULL;
5550                 if (error == 0) {
5551                         m_clrprotoflags(m0);
5552                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5553                 } else
5554                         m_freem(m0);
5555         }
5556
5557         if (error == 0)
5558                 KMOD_IPSTAT_INC(ips_fragmented);
5559
5560 done:
5561         if (r->rt != PF_DUPTO)
5562                 *m = NULL;
5563         return;
5564
5565 bad_locked:
5566         if (s)
5567                 PF_STATE_UNLOCK(s);
5568 bad:
5569         m_freem(m0);
5570         goto done;
5571 }
5572 #endif /* INET */
5573
5574 #ifdef INET6
5575 static void
5576 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5577     struct pf_state *s, struct pf_pdesc *pd)
5578 {
5579         struct mbuf             *m0;
5580         struct sockaddr_in6     dst;
5581         struct ip6_hdr          *ip6;
5582         struct ifnet            *ifp = NULL;
5583         struct pf_addr           naddr;
5584         struct pf_src_node      *sn = NULL;
5585
5586         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5587         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5588             __func__));
5589
5590         if ((pd->pf_mtag == NULL &&
5591             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5592             pd->pf_mtag->routed++ > 3) {
5593                 m0 = *m;
5594                 *m = NULL;
5595                 goto bad_locked;
5596         }
5597
5598         if (r->rt == PF_DUPTO) {
5599                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5600                         if (s)
5601                                 PF_STATE_UNLOCK(s);
5602                         return;
5603                 }
5604         } else {
5605                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5606                         if (s)
5607                                 PF_STATE_UNLOCK(s);
5608                         return;
5609                 }
5610                 m0 = *m;
5611         }
5612
5613         ip6 = mtod(m0, struct ip6_hdr *);
5614
5615         bzero(&dst, sizeof(dst));
5616         dst.sin6_family = AF_INET6;
5617         dst.sin6_len = sizeof(dst);
5618         dst.sin6_addr = ip6->ip6_dst;
5619
5620         if (TAILQ_EMPTY(&r->rpool.list)) {
5621                 DPFPRINTF(PF_DEBUG_URGENT,
5622                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5623                 goto bad_locked;
5624         }
5625         if (s == NULL) {
5626                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5627                     &naddr, NULL, &sn);
5628                 if (!PF_AZERO(&naddr, AF_INET6))
5629                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5630                             &naddr, AF_INET6);
5631                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5632         } else {
5633                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5634                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5635                             &s->rt_addr, AF_INET6);
5636                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5637         }
5638
5639         if (s)
5640                 PF_STATE_UNLOCK(s);
5641
5642         if (ifp == NULL)
5643                 goto bad;
5644
5645         if (oifp != ifp) {
5646                 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS)
5647                         goto bad;
5648                 else if (m0 == NULL)
5649                         goto done;
5650                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5651                         DPFPRINTF(PF_DEBUG_URGENT,
5652                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5653                             __func__));
5654                         goto bad;
5655                 }
5656                 ip6 = mtod(m0, struct ip6_hdr *);
5657         }
5658
5659         if (ifp->if_flags & IFF_LOOPBACK)
5660                 m0->m_flags |= M_SKIP_FIREWALL;
5661
5662         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5663             ~ifp->if_hwassist) {
5664                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5665                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5666                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5667         }
5668
5669         /*
5670          * If the packet is too large for the outgoing interface,
5671          * send back an icmp6 error.
5672          */
5673         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5674                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5675         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5676                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5677         else {
5678                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5679                 if (r->rt != PF_DUPTO)
5680                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5681                 else
5682                         goto bad;
5683         }
5684
5685 done:
5686         if (r->rt != PF_DUPTO)
5687                 *m = NULL;
5688         return;
5689
5690 bad_locked:
5691         if (s)
5692                 PF_STATE_UNLOCK(s);
5693 bad:
5694         m_freem(m0);
5695         goto done;
5696 }
5697 #endif /* INET6 */
5698
5699 /*
5700  * FreeBSD supports cksum offloads for the following drivers.
5701  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5702  *   ti(4), txp(4), xl(4)
5703  *
5704  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5705  *  network driver performed cksum including pseudo header, need to verify
5706  *   csum_data
5707  * CSUM_DATA_VALID :
5708  *  network driver performed cksum, needs to additional pseudo header
5709  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5710  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5711  *
5712  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5713  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5714  * TCP/UDP layer.
5715  * Also, set csum_data to 0xffff to force cksum validation.
5716  */
5717 static int
5718 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5719 {
5720         u_int16_t sum = 0;
5721         int hw_assist = 0;
5722         struct ip *ip;
5723
5724         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5725                 return (1);
5726         if (m->m_pkthdr.len < off + len)
5727                 return (1);
5728
5729         switch (p) {
5730         case IPPROTO_TCP:
5731                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5732                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5733                                 sum = m->m_pkthdr.csum_data;
5734                         } else {
5735                                 ip = mtod(m, struct ip *);
5736                                 sum = in_pseudo(ip->ip_src.s_addr,
5737                                 ip->ip_dst.s_addr, htonl((u_short)len +
5738                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5739                         }
5740                         sum ^= 0xffff;
5741                         ++hw_assist;
5742                 }
5743                 break;
5744         case IPPROTO_UDP:
5745                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5746                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5747                                 sum = m->m_pkthdr.csum_data;
5748                         } else {
5749                                 ip = mtod(m, struct ip *);
5750                                 sum = in_pseudo(ip->ip_src.s_addr,
5751                                 ip->ip_dst.s_addr, htonl((u_short)len +
5752                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5753                         }
5754                         sum ^= 0xffff;
5755                         ++hw_assist;
5756                 }
5757                 break;
5758         case IPPROTO_ICMP:
5759 #ifdef INET6
5760         case IPPROTO_ICMPV6:
5761 #endif /* INET6 */
5762                 break;
5763         default:
5764                 return (1);
5765         }
5766
5767         if (!hw_assist) {
5768                 switch (af) {
5769                 case AF_INET:
5770                         if (p == IPPROTO_ICMP) {
5771                                 if (m->m_len < off)
5772                                         return (1);
5773                                 m->m_data += off;
5774                                 m->m_len -= off;
5775                                 sum = in_cksum(m, len);
5776                                 m->m_data -= off;
5777                                 m->m_len += off;
5778                         } else {
5779                                 if (m->m_len < sizeof(struct ip))
5780                                         return (1);
5781                                 sum = in4_cksum(m, p, off, len);
5782                         }
5783                         break;
5784 #ifdef INET6
5785                 case AF_INET6:
5786                         if (m->m_len < sizeof(struct ip6_hdr))
5787                                 return (1);
5788                         sum = in6_cksum(m, p, off, len);
5789                         break;
5790 #endif /* INET6 */
5791                 default:
5792                         return (1);
5793                 }
5794         }
5795         if (sum) {
5796                 switch (p) {
5797                 case IPPROTO_TCP:
5798                     {
5799                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5800                         break;
5801                     }
5802                 case IPPROTO_UDP:
5803                     {
5804                         KMOD_UDPSTAT_INC(udps_badsum);
5805                         break;
5806                     }
5807 #ifdef INET
5808                 case IPPROTO_ICMP:
5809                     {
5810                         KMOD_ICMPSTAT_INC(icps_checksum);
5811                         break;
5812                     }
5813 #endif
5814 #ifdef INET6
5815                 case IPPROTO_ICMPV6:
5816                     {
5817                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5818                         break;
5819                     }
5820 #endif /* INET6 */
5821                 }
5822                 return (1);
5823         } else {
5824                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5825                         m->m_pkthdr.csum_flags |=
5826                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5827                         m->m_pkthdr.csum_data = 0xffff;
5828                 }
5829         }
5830         return (0);
5831 }
5832
5833
5834 #ifdef INET
5835 int
5836 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5837 {
5838         struct pfi_kif          *kif;
5839         u_short                  action, reason = 0, log = 0;
5840         struct mbuf             *m = *m0;
5841         struct ip               *h = NULL;
5842         struct m_tag            *ipfwtag;
5843         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5844         struct pf_state         *s = NULL;
5845         struct pf_ruleset       *ruleset = NULL;
5846         struct pf_pdesc          pd;
5847         int                      off, dirndx, pqid = 0;
5848
5849         M_ASSERTPKTHDR(m);
5850
5851         if (!V_pf_status.running)
5852                 return (PF_PASS);
5853
5854         memset(&pd, 0, sizeof(pd));
5855
5856         kif = (struct pfi_kif *)ifp->if_pf_kif;
5857
5858         if (kif == NULL) {
5859                 DPFPRINTF(PF_DEBUG_URGENT,
5860                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5861                 return (PF_DROP);
5862         }
5863         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5864                 return (PF_PASS);
5865
5866         if (m->m_flags & M_SKIP_FIREWALL)
5867                 return (PF_PASS);
5868
5869         pd.pf_mtag = pf_find_mtag(m);
5870
5871         PF_RULES_RLOCK();
5872
5873         if (ip_divert_ptr != NULL &&
5874             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5875                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5876                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5877                         if (pd.pf_mtag == NULL &&
5878                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5879                                 action = PF_DROP;
5880                                 goto done;
5881                         }
5882                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5883                         m_tag_delete(m, ipfwtag);
5884                 }
5885                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5886                         m->m_flags |= M_FASTFWD_OURS;
5887                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5888                 }
5889         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5890                 /* We do IP header normalization and packet reassembly here */
5891                 action = PF_DROP;
5892                 goto done;
5893         }
5894         m = *m0;        /* pf_normalize messes with m0 */
5895         h = mtod(m, struct ip *);
5896
5897         off = h->ip_hl << 2;
5898         if (off < (int)sizeof(struct ip)) {
5899                 action = PF_DROP;
5900                 REASON_SET(&reason, PFRES_SHORT);
5901                 log = 1;
5902                 goto done;
5903         }
5904
5905         pd.src = (struct pf_addr *)&h->ip_src;
5906         pd.dst = (struct pf_addr *)&h->ip_dst;
5907         pd.sport = pd.dport = NULL;
5908         pd.ip_sum = &h->ip_sum;
5909         pd.proto_sum = NULL;
5910         pd.proto = h->ip_p;
5911         pd.dir = dir;
5912         pd.sidx = (dir == PF_IN) ? 0 : 1;
5913         pd.didx = (dir == PF_IN) ? 1 : 0;
5914         pd.af = AF_INET;
5915         pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
5916         pd.tot_len = ntohs(h->ip_len);
5917
5918         /* handle fragments that didn't get reassembled by normalization */
5919         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5920                 action = pf_test_fragment(&r, dir, kif, m, h,
5921                     &pd, &a, &ruleset);
5922                 goto done;
5923         }
5924
5925         switch (h->ip_p) {
5926
5927         case IPPROTO_TCP: {
5928                 struct tcphdr   th;
5929
5930                 pd.hdr.tcp = &th;
5931                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5932                     &action, &reason, AF_INET)) {
5933                         log = action != PF_PASS;
5934                         goto done;
5935                 }
5936                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5937                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5938                         pqid = 1;
5939                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5940                 if (action == PF_DROP)
5941                         goto done;
5942                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5943                     &reason);
5944                 if (action == PF_PASS) {
5945                         if (pfsync_update_state_ptr != NULL)
5946                                 pfsync_update_state_ptr(s);
5947                         r = s->rule.ptr;
5948                         a = s->anchor.ptr;
5949                         log = s->log;
5950                 } else if (s == NULL)
5951                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5952                             &a, &ruleset, inp);
5953                 break;
5954         }
5955
5956         case IPPROTO_UDP: {
5957                 struct udphdr   uh;
5958
5959                 pd.hdr.udp = &uh;
5960                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5961                     &action, &reason, AF_INET)) {
5962                         log = action != PF_PASS;
5963                         goto done;
5964                 }
5965                 if (uh.uh_dport == 0 ||
5966                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5967                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5968                         action = PF_DROP;
5969                         REASON_SET(&reason, PFRES_SHORT);
5970                         goto done;
5971                 }
5972                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5973                 if (action == PF_PASS) {
5974                         if (pfsync_update_state_ptr != NULL)
5975                                 pfsync_update_state_ptr(s);
5976                         r = s->rule.ptr;
5977                         a = s->anchor.ptr;
5978                         log = s->log;
5979                 } else if (s == NULL)
5980                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5981                             &a, &ruleset, inp);
5982                 break;
5983         }
5984
5985         case IPPROTO_ICMP: {
5986                 struct icmp     ih;
5987
5988                 pd.hdr.icmp = &ih;
5989                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5990                     &action, &reason, AF_INET)) {
5991                         log = action != PF_PASS;
5992                         goto done;
5993                 }
5994                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5995                     &reason);
5996                 if (action == PF_PASS) {
5997                         if (pfsync_update_state_ptr != NULL)
5998                                 pfsync_update_state_ptr(s);
5999                         r = s->rule.ptr;
6000                         a = s->anchor.ptr;
6001                         log = s->log;
6002                 } else if (s == NULL)
6003                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6004                             &a, &ruleset, inp);
6005                 break;
6006         }
6007
6008 #ifdef INET6
6009         case IPPROTO_ICMPV6: {
6010                 action = PF_DROP;
6011                 DPFPRINTF(PF_DEBUG_MISC,
6012                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6013                 goto done;
6014         }
6015 #endif
6016
6017         default:
6018                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6019                 if (action == PF_PASS) {
6020                         if (pfsync_update_state_ptr != NULL)
6021                                 pfsync_update_state_ptr(s);
6022                         r = s->rule.ptr;
6023                         a = s->anchor.ptr;
6024                         log = s->log;
6025                 } else if (s == NULL)
6026                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6027                             &a, &ruleset, inp);
6028                 break;
6029         }
6030
6031 done:
6032         PF_RULES_RUNLOCK();
6033         if (action == PF_PASS && h->ip_hl > 5 &&
6034             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6035                 action = PF_DROP;
6036                 REASON_SET(&reason, PFRES_IPOPTIONS);
6037                 log = r->log;
6038                 DPFPRINTF(PF_DEBUG_MISC,
6039                     ("pf: dropping packet with ip options\n"));
6040         }
6041
6042         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6043                 action = PF_DROP;
6044                 REASON_SET(&reason, PFRES_MEMORY);
6045         }
6046         if (r->rtableid >= 0)
6047                 M_SETFIB(m, r->rtableid);
6048
6049         if (r->scrub_flags & PFSTATE_SETPRIO) {
6050                 if (pd.tos & IPTOS_LOWDELAY)
6051                         pqid = 1;
6052                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6053                         action = PF_DROP;
6054                         REASON_SET(&reason, PFRES_MEMORY);
6055                         log = 1;
6056                         DPFPRINTF(PF_DEBUG_MISC,
6057                             ("pf: failed to allocate 802.1q mtag\n"));
6058                 }
6059         }
6060
6061 #ifdef ALTQ
6062         if (action == PF_PASS && r->qid) {
6063                 if (pd.pf_mtag == NULL &&
6064                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6065                         action = PF_DROP;
6066                         REASON_SET(&reason, PFRES_MEMORY);
6067                 } else {
6068                         if (s != NULL)
6069                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6070                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6071                                 pd.pf_mtag->qid = r->pqid;
6072                         else
6073                                 pd.pf_mtag->qid = r->qid;
6074                         /* Add hints for ecn. */
6075                         pd.pf_mtag->hdr = h;
6076                 }
6077
6078         }
6079 #endif /* ALTQ */
6080
6081         /*
6082          * connections redirected to loopback should not match sockets
6083          * bound specifically to loopback due to security implications,
6084          * see tcp_input() and in_pcblookup_listen().
6085          */
6086         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6087             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6088             (s->nat_rule.ptr->action == PF_RDR ||
6089             s->nat_rule.ptr->action == PF_BINAT) &&
6090             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6091                 m->m_flags |= M_SKIP_FIREWALL;
6092
6093         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6094             !PACKET_LOOPED(&pd)) {
6095
6096                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6097                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6098                 if (ipfwtag != NULL) {
6099                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6100                             ntohs(r->divert.port);
6101                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6102
6103                         if (s)
6104                                 PF_STATE_UNLOCK(s);
6105
6106                         m_tag_prepend(m, ipfwtag);
6107                         if (m->m_flags & M_FASTFWD_OURS) {
6108                                 if (pd.pf_mtag == NULL &&
6109                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6110                                         action = PF_DROP;
6111                                         REASON_SET(&reason, PFRES_MEMORY);
6112                                         log = 1;
6113                                         DPFPRINTF(PF_DEBUG_MISC,
6114                                             ("pf: failed to allocate tag\n"));
6115                                 } else {
6116                                         pd.pf_mtag->flags |=
6117                                             PF_FASTFWD_OURS_PRESENT;
6118                                         m->m_flags &= ~M_FASTFWD_OURS;
6119                                 }
6120                         }
6121                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6122                         *m0 = NULL;
6123
6124                         return (action);
6125                 } else {
6126                         /* XXX: ipfw has the same behaviour! */
6127                         action = PF_DROP;
6128                         REASON_SET(&reason, PFRES_MEMORY);
6129                         log = 1;
6130                         DPFPRINTF(PF_DEBUG_MISC,
6131                             ("pf: failed to allocate divert tag\n"));
6132                 }
6133         }
6134
6135         if (log) {
6136                 struct pf_rule *lr;
6137
6138                 if (s != NULL && s->nat_rule.ptr != NULL &&
6139                     s->nat_rule.ptr->log & PF_LOG_ALL)
6140                         lr = s->nat_rule.ptr;
6141                 else
6142                         lr = r;
6143                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6144                     (s == NULL));
6145         }
6146
6147         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6148         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6149
6150         if (action == PF_PASS || r->action == PF_DROP) {
6151                 dirndx = (dir == PF_OUT);
6152                 r->packets[dirndx]++;
6153                 r->bytes[dirndx] += pd.tot_len;
6154                 if (a != NULL) {
6155                         a->packets[dirndx]++;
6156                         a->bytes[dirndx] += pd.tot_len;
6157                 }
6158                 if (s != NULL) {
6159                         if (s->nat_rule.ptr != NULL) {
6160                                 s->nat_rule.ptr->packets[dirndx]++;
6161                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6162                         }
6163                         if (s->src_node != NULL) {
6164                                 s->src_node->packets[dirndx]++;
6165                                 s->src_node->bytes[dirndx] += pd.tot_len;
6166                         }
6167                         if (s->nat_src_node != NULL) {
6168                                 s->nat_src_node->packets[dirndx]++;
6169                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6170                         }
6171                         dirndx = (dir == s->direction) ? 0 : 1;
6172                         s->packets[dirndx]++;
6173                         s->bytes[dirndx] += pd.tot_len;
6174                 }
6175                 tr = r;
6176                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6177                 if (nr != NULL && r == &V_pf_default_rule)
6178                         tr = nr;
6179                 if (tr->src.addr.type == PF_ADDR_TABLE)
6180                         pfr_update_stats(tr->src.addr.p.tbl,
6181                             (s == NULL) ? pd.src :
6182                             &s->key[(s->direction == PF_IN)]->
6183                                 addr[(s->direction == PF_OUT)],
6184                             pd.af, pd.tot_len, dir == PF_OUT,
6185                             r->action == PF_PASS, tr->src.neg);
6186                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6187                         pfr_update_stats(tr->dst.addr.p.tbl,
6188                             (s == NULL) ? pd.dst :
6189                             &s->key[(s->direction == PF_IN)]->
6190                                 addr[(s->direction == PF_IN)],
6191                             pd.af, pd.tot_len, dir == PF_OUT,
6192                             r->action == PF_PASS, tr->dst.neg);
6193         }
6194
6195         switch (action) {
6196         case PF_SYNPROXY_DROP:
6197                 m_freem(*m0);
6198         case PF_DEFER:
6199                 *m0 = NULL;
6200                 action = PF_PASS;
6201                 break;
6202         case PF_DROP:
6203                 m_freem(*m0);
6204                 *m0 = NULL;
6205                 break;
6206         default:
6207                 /* pf_route() returns unlocked. */
6208                 if (r->rt) {
6209                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6210                         return (action);
6211                 }
6212                 break;
6213         }
6214         if (s)
6215                 PF_STATE_UNLOCK(s);
6216
6217         return (action);
6218 }
6219 #endif /* INET */
6220
6221 #ifdef INET6
6222 int
6223 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6224 {
6225         struct pfi_kif          *kif;
6226         u_short                  action, reason = 0, log = 0;
6227         struct mbuf             *m = *m0, *n = NULL;
6228         struct m_tag            *mtag;
6229         struct ip6_hdr          *h = NULL;
6230         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6231         struct pf_state         *s = NULL;
6232         struct pf_ruleset       *ruleset = NULL;
6233         struct pf_pdesc          pd;
6234         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6235         int                      fwdir = dir;
6236
6237         M_ASSERTPKTHDR(m);
6238
6239         /* Detect packet forwarding.
6240          * If the input interface is different from the output interface we're
6241          * forwarding.
6242          * We do need to be careful about bridges. If the
6243          * net.link.bridge.pfil_bridge sysctl is set we can be filtering on a
6244          * bridge, so if the input interface is a bridge member and the output
6245          * interface is its bridge or a member of the same bridge we're not
6246          * actually forwarding but bridging.
6247          */
6248         if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif &&
6249             (m->m_pkthdr.rcvif->if_bridge == NULL ||
6250             (m->m_pkthdr.rcvif->if_bridge != ifp->if_softc &&
6251             m->m_pkthdr.rcvif->if_bridge != ifp->if_bridge)))
6252                 fwdir = PF_FWD;
6253
6254         if (dir == PF_FWD)
6255                 dir = PF_OUT;
6256
6257         if (!V_pf_status.running)
6258                 return (PF_PASS);
6259
6260         memset(&pd, 0, sizeof(pd));
6261         pd.pf_mtag = pf_find_mtag(m);
6262
6263         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6264                 return (PF_PASS);
6265
6266         kif = (struct pfi_kif *)ifp->if_pf_kif;
6267         if (kif == NULL) {
6268                 DPFPRINTF(PF_DEBUG_URGENT,
6269                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6270                 return (PF_DROP);
6271         }
6272         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6273                 return (PF_PASS);
6274
6275         if (m->m_flags & M_SKIP_FIREWALL)
6276                 return (PF_PASS);
6277
6278         PF_RULES_RLOCK();
6279
6280         /* We do IP header normalization and packet reassembly here */
6281         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6282                 action = PF_DROP;
6283                 goto done;
6284         }
6285         m = *m0;        /* pf_normalize messes with m0 */
6286         h = mtod(m, struct ip6_hdr *);
6287
6288 #if 1
6289         /*
6290          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6291          * will do something bad, so drop the packet for now.
6292          */
6293         if (htons(h->ip6_plen) == 0) {
6294                 action = PF_DROP;
6295                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6296                 goto done;
6297         }
6298 #endif
6299
6300         pd.src = (struct pf_addr *)&h->ip6_src;
6301         pd.dst = (struct pf_addr *)&h->ip6_dst;
6302         pd.sport = pd.dport = NULL;
6303         pd.ip_sum = NULL;
6304         pd.proto_sum = NULL;
6305         pd.dir = dir;
6306         pd.sidx = (dir == PF_IN) ? 0 : 1;
6307         pd.didx = (dir == PF_IN) ? 1 : 0;
6308         pd.af = AF_INET6;
6309         pd.tos = 0;
6310         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6311
6312         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6313         pd.proto = h->ip6_nxt;
6314         do {
6315                 switch (pd.proto) {
6316                 case IPPROTO_FRAGMENT:
6317                         action = pf_test_fragment(&r, dir, kif, m, h,
6318                             &pd, &a, &ruleset);
6319                         if (action == PF_DROP)
6320                                 REASON_SET(&reason, PFRES_FRAG);
6321                         goto done;
6322                 case IPPROTO_ROUTING: {
6323                         struct ip6_rthdr rthdr;
6324
6325                         if (rh_cnt++) {
6326                                 DPFPRINTF(PF_DEBUG_MISC,
6327                                     ("pf: IPv6 more than one rthdr\n"));
6328                                 action = PF_DROP;
6329                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6330                                 log = 1;
6331                                 goto done;
6332                         }
6333                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6334                             &reason, pd.af)) {
6335                                 DPFPRINTF(PF_DEBUG_MISC,
6336                                     ("pf: IPv6 short rthdr\n"));
6337                                 action = PF_DROP;
6338                                 REASON_SET(&reason, PFRES_SHORT);
6339                                 log = 1;
6340                                 goto done;
6341                         }
6342                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6343                                 DPFPRINTF(PF_DEBUG_MISC,
6344                                     ("pf: IPv6 rthdr0\n"));
6345                                 action = PF_DROP;
6346                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6347                                 log = 1;
6348                                 goto done;
6349                         }
6350                         /* FALLTHROUGH */
6351                 }
6352                 case IPPROTO_AH:
6353                 case IPPROTO_HOPOPTS:
6354                 case IPPROTO_DSTOPTS: {
6355                         /* get next header and header length */
6356                         struct ip6_ext  opt6;
6357
6358                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6359                             NULL, &reason, pd.af)) {
6360                                 DPFPRINTF(PF_DEBUG_MISC,
6361                                     ("pf: IPv6 short opt\n"));
6362                                 action = PF_DROP;
6363                                 log = 1;
6364                                 goto done;
6365                         }
6366                         if (pd.proto == IPPROTO_AH)
6367                                 off += (opt6.ip6e_len + 2) * 4;
6368                         else
6369                                 off += (opt6.ip6e_len + 1) * 8;
6370                         pd.proto = opt6.ip6e_nxt;
6371                         /* goto the next header */
6372                         break;
6373                 }
6374                 default:
6375                         terminal++;
6376                         break;
6377                 }
6378         } while (!terminal);
6379
6380         /* if there's no routing header, use unmodified mbuf for checksumming */
6381         if (!n)
6382                 n = m;
6383
6384         switch (pd.proto) {
6385
6386         case IPPROTO_TCP: {
6387                 struct tcphdr   th;
6388
6389                 pd.hdr.tcp = &th;
6390                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6391                     &action, &reason, AF_INET6)) {
6392                         log = action != PF_PASS;
6393                         goto done;
6394                 }
6395                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6396                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6397                 if (action == PF_DROP)
6398                         goto done;
6399                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6400                     &reason);
6401                 if (action == PF_PASS) {
6402                         if (pfsync_update_state_ptr != NULL)
6403                                 pfsync_update_state_ptr(s);
6404                         r = s->rule.ptr;
6405                         a = s->anchor.ptr;
6406                         log = s->log;
6407                 } else if (s == NULL)
6408                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6409                             &a, &ruleset, inp);
6410                 break;
6411         }
6412
6413         case IPPROTO_UDP: {
6414                 struct udphdr   uh;
6415
6416                 pd.hdr.udp = &uh;
6417                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6418                     &action, &reason, AF_INET6)) {
6419                         log = action != PF_PASS;
6420                         goto done;
6421                 }
6422                 if (uh.uh_dport == 0 ||
6423                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6424                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6425                         action = PF_DROP;
6426                         REASON_SET(&reason, PFRES_SHORT);
6427                         goto done;
6428                 }
6429                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6430                 if (action == PF_PASS) {
6431                         if (pfsync_update_state_ptr != NULL)
6432                                 pfsync_update_state_ptr(s);
6433                         r = s->rule.ptr;
6434                         a = s->anchor.ptr;
6435                         log = s->log;
6436                 } else if (s == NULL)
6437                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6438                             &a, &ruleset, inp);
6439                 break;
6440         }
6441
6442         case IPPROTO_ICMP: {
6443                 action = PF_DROP;
6444                 DPFPRINTF(PF_DEBUG_MISC,
6445                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6446                 goto done;
6447         }
6448
6449         case IPPROTO_ICMPV6: {
6450                 struct icmp6_hdr        ih;
6451
6452                 pd.hdr.icmp6 = &ih;
6453                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6454                     &action, &reason, AF_INET6)) {
6455                         log = action != PF_PASS;
6456                         goto done;
6457                 }
6458                 action = pf_test_state_icmp(&s, dir, kif,
6459                     m, off, h, &pd, &reason);
6460                 if (action == PF_PASS) {
6461                         if (pfsync_update_state_ptr != NULL)
6462                                 pfsync_update_state_ptr(s);
6463                         r = s->rule.ptr;
6464                         a = s->anchor.ptr;
6465                         log = s->log;
6466                 } else if (s == NULL)
6467                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6468                             &a, &ruleset, inp);
6469                 break;
6470         }
6471
6472         default:
6473                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6474                 if (action == PF_PASS) {
6475                         if (pfsync_update_state_ptr != NULL)
6476                                 pfsync_update_state_ptr(s);
6477                         r = s->rule.ptr;
6478                         a = s->anchor.ptr;
6479                         log = s->log;
6480                 } else if (s == NULL)
6481                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6482                             &a, &ruleset, inp);
6483                 break;
6484         }
6485
6486 done:
6487         PF_RULES_RUNLOCK();
6488         if (n != m) {
6489                 m_freem(n);
6490                 n = NULL;
6491         }
6492
6493         /* handle dangerous IPv6 extension headers. */
6494         if (action == PF_PASS && rh_cnt &&
6495             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6496                 action = PF_DROP;
6497                 REASON_SET(&reason, PFRES_IPOPTIONS);
6498                 log = r->log;
6499                 DPFPRINTF(PF_DEBUG_MISC,
6500                     ("pf: dropping packet with dangerous v6 headers\n"));
6501         }
6502
6503         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6504                 action = PF_DROP;
6505                 REASON_SET(&reason, PFRES_MEMORY);
6506         }
6507         if (r->rtableid >= 0)
6508                 M_SETFIB(m, r->rtableid);
6509
6510         if (r->scrub_flags & PFSTATE_SETPRIO) {
6511                 if (pd.tos & IPTOS_LOWDELAY)
6512                         pqid = 1;
6513                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6514                         action = PF_DROP;
6515                         REASON_SET(&reason, PFRES_MEMORY);
6516                         log = 1;
6517                         DPFPRINTF(PF_DEBUG_MISC,
6518                             ("pf: failed to allocate 802.1q mtag\n"));
6519                 }
6520         }
6521
6522 #ifdef ALTQ
6523         if (action == PF_PASS && r->qid) {
6524                 if (pd.pf_mtag == NULL &&
6525                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6526                         action = PF_DROP;
6527                         REASON_SET(&reason, PFRES_MEMORY);
6528                 } else {
6529                         if (s != NULL)
6530                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6531                         if (pd.tos & IPTOS_LOWDELAY)
6532                                 pd.pf_mtag->qid = r->pqid;
6533                         else
6534                                 pd.pf_mtag->qid = r->qid;
6535                         /* Add hints for ecn. */
6536                         pd.pf_mtag->hdr = h;
6537                 }
6538         }
6539 #endif /* ALTQ */
6540
6541         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6542             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6543             (s->nat_rule.ptr->action == PF_RDR ||
6544             s->nat_rule.ptr->action == PF_BINAT) &&
6545             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6546                 m->m_flags |= M_SKIP_FIREWALL;
6547
6548         /* XXX: Anybody working on it?! */
6549         if (r->divert.port)
6550                 printf("pf: divert(9) is not supported for IPv6\n");
6551
6552         if (log) {
6553                 struct pf_rule *lr;
6554
6555                 if (s != NULL && s->nat_rule.ptr != NULL &&
6556                     s->nat_rule.ptr->log & PF_LOG_ALL)
6557                         lr = s->nat_rule.ptr;
6558                 else
6559                         lr = r;
6560                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6561                     &pd, (s == NULL));
6562         }
6563
6564         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6565         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6566
6567         if (action == PF_PASS || r->action == PF_DROP) {
6568                 dirndx = (dir == PF_OUT);
6569                 r->packets[dirndx]++;
6570                 r->bytes[dirndx] += pd.tot_len;
6571                 if (a != NULL) {
6572                         a->packets[dirndx]++;
6573                         a->bytes[dirndx] += pd.tot_len;
6574                 }
6575                 if (s != NULL) {
6576                         if (s->nat_rule.ptr != NULL) {
6577                                 s->nat_rule.ptr->packets[dirndx]++;
6578                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6579                         }
6580                         if (s->src_node != NULL) {
6581                                 s->src_node->packets[dirndx]++;
6582                                 s->src_node->bytes[dirndx] += pd.tot_len;
6583                         }
6584                         if (s->nat_src_node != NULL) {
6585                                 s->nat_src_node->packets[dirndx]++;
6586                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6587                         }
6588                         dirndx = (dir == s->direction) ? 0 : 1;
6589                         s->packets[dirndx]++;
6590                         s->bytes[dirndx] += pd.tot_len;
6591                 }
6592                 tr = r;
6593                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6594                 if (nr != NULL && r == &V_pf_default_rule)
6595                         tr = nr;
6596                 if (tr->src.addr.type == PF_ADDR_TABLE)
6597                         pfr_update_stats(tr->src.addr.p.tbl,
6598                             (s == NULL) ? pd.src :
6599                             &s->key[(s->direction == PF_IN)]->addr[0],
6600                             pd.af, pd.tot_len, dir == PF_OUT,
6601                             r->action == PF_PASS, tr->src.neg);
6602                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6603                         pfr_update_stats(tr->dst.addr.p.tbl,
6604                             (s == NULL) ? pd.dst :
6605                             &s->key[(s->direction == PF_IN)]->addr[1],
6606                             pd.af, pd.tot_len, dir == PF_OUT,
6607                             r->action == PF_PASS, tr->dst.neg);
6608         }
6609
6610         switch (action) {
6611         case PF_SYNPROXY_DROP:
6612                 m_freem(*m0);
6613         case PF_DEFER:
6614                 *m0 = NULL;
6615                 action = PF_PASS;
6616                 break;
6617         case PF_DROP:
6618                 m_freem(*m0);
6619                 *m0 = NULL;
6620                 break;
6621         default:
6622                 /* pf_route6() returns unlocked. */
6623                 if (r->rt) {
6624                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6625                         return (action);
6626                 }
6627                 break;
6628         }
6629
6630         if (s)
6631                 PF_STATE_UNLOCK(s);
6632
6633         /* If reassembled packet passed, create new fragments. */
6634         if (action == PF_PASS && *m0 && fwdir == PF_FWD &&
6635             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6636                 action = pf_refragment6(ifp, m0, mtag);
6637
6638         return (action);
6639 }
6640 #endif /* INET6 */