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