sys/netpfil/pf/pf.c

   1 /*-
   2  * Copyright (c) 2001 Daniel Hartmeier
   3  * Copyright (c) 2002 - 2008 Henning Brauer
   4  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  *    - Redistributions of source code must retain the above copyright
  12  *      notice, this list of conditions and the following disclaimer.
  13  *    - Redistributions in binary form must reproduce the above
  14  *      copyright notice, this list of conditions and the following
  15  *      disclaimer in the documentation and/or other materials provided
  16  *      with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  21  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  22  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  26  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  *
  31  * Effort sponsored in part by the Defense Advanced Research Projects
  32  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  33  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  34  *
  35  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43 #include "opt_bpf.h"
  44 #include "opt_pf.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/bus.h>
  48 #include <sys/endian.h>
  49 #include <sys/hash.h>
  50 #include <sys/interrupt.h>
  51 #include <sys/kernel.h>
  52 #include <sys/kthread.h>
  53 #include <sys/limits.h>
  54 #include <sys/mbuf.h>
  55 #include <sys/md5.h>
  56 #include <sys/random.h>
  57 #include <sys/refcount.h>
  58 #include <sys/socket.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/taskqueue.h>
  61 #include <sys/ucred.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_types.h>
  65 #include <net/route.h>
  66 #include <net/radix_mpath.h>
  67 #include <net/vnet.h>
  68
  69 #include <net/pfvar.h>
  70 #include <net/if_pflog.h>
  71 #include <net/if_pfsync.h>
  72
  73 #include <netinet/in_pcb.h>
  74 #include <netinet/in_var.h>
  75 #include <netinet/ip.h>
  76 #include <netinet/ip_fw.h>
  77 #include <netinet/ip_icmp.h>
  78 #include <netinet/icmp_var.h>
  79 #include <netinet/ip_var.h>
  80 #include <netinet/tcp.h>
  81 #include <netinet/tcp_fsm.h>
  82 #include <netinet/tcp_seq.h>
  83 #include <netinet/tcp_timer.h>
  84 #include <netinet/tcp_var.h>
  85 #include <netinet/udp.h>
  86 #include <netinet/udp_var.h>
  87
  88 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  89
  90 #ifdef INET6
  91 #include <netinet/ip6.h>
  92 #include <netinet/icmp6.h>
  93 #include <netinet6/nd6.h>
  94 #include <netinet6/ip6_var.h>
  95 #include <netinet6/in6_pcb.h>
  96 #endif /* INET6 */
  97
  98 #include <machine/in_cksum.h>
  99 #include <security/mac/mac_framework.h>
 100
 101 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 102
 103 /*
 104  * Global variables
 105  */
 106
 107 /* state tables */
 108 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 109 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 110 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 111 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 112 VNET_DEFINE(struct pf_kstatus,           pf_status);
 113
 114 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 115 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 116 VNET_DEFINE(int,                         altqs_inactive_open);
 117 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 118
 119 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 120 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 121 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 122 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 123 VNET_DEFINE(int,                         pf_tcp_secret_init);
 124 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 125 VNET_DEFINE(int,                         pf_tcp_iss_off);
 126 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 127
 128 /*
 129  * Queue for pf_intr() sends.
 130  */
 131 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 132 struct pf_send_entry {
 133         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 134         struct mbuf                     *pfse_m;
 135         enum {
 136                 PFSE_IP,
 137                 PFSE_IP6,
 138                 PFSE_ICMP,
 139                 PFSE_ICMP6,
 140         }                               pfse_type;
 141         union {
 142                 struct route            ro;
 143                 struct {
 144                         int             type;
 145                         int             code;
 146                         int             mtu;
 147                 } icmpopts;
 148         } u;
 149 #define pfse_ro         u.ro
 150 #define pfse_icmp_type  u.icmpopts.type
 151 #define pfse_icmp_code  u.icmpopts.code
 152 #define pfse_icmp_mtu   u.icmpopts.mtu
 153 };
 154
 155 STAILQ_HEAD(pf_send_head, pf_send_entry);
 156 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 157 #define V_pf_sendqueue  VNET(pf_sendqueue)
 158
 159 static struct mtx pf_sendqueue_mtx;
 160 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 161 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 162
 163 /*
 164  * Queue for pf_overload_task() tasks.
 165  */
 166 struct pf_overload_entry {
 167         SLIST_ENTRY(pf_overload_entry)  next;
 168         struct pf_addr                  addr;
 169         sa_family_t                     af;
 170         uint8_t                         dir;
 171         struct pf_rule                  *rule;
 172 };
 173
 174 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 175 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 176 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 177 static VNET_DEFINE(struct task, pf_overloadtask);
 178 #define V_pf_overloadtask       VNET(pf_overloadtask)
 179
 180 static struct mtx pf_overloadqueue_mtx;
 181 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 182 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 183
 184 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 185 struct mtx pf_unlnkdrules_mtx;
 186
 187 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 188 #define V_pf_sources_z  VNET(pf_sources_z)
 189 uma_zone_t              pf_mtag_z;
 190 VNET_DEFINE(uma_zone_t,  pf_state_z);
 191 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 192
 193 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 194 #define PFID_CPUBITS    8
 195 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 196 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 197 #define PFID_MAXID      (~PFID_CPUMASK)
 198 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 199
 200 static void              pf_src_tree_remove_state(struct pf_state *);
 201 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 202                             u_int32_t);
 203 static void              pf_add_threshold(struct pf_threshold *);
 204 static int               pf_check_threshold(struct pf_threshold *);
 205
 206 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 207                             u_int16_t *, u_int16_t *, struct pf_addr *,
 208                             u_int16_t, u_int8_t, sa_family_t);
 209 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 210                             struct tcphdr *, struct pf_state_peer *);
 211 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 212                             struct pf_addr *, struct pf_addr *, u_int16_t,
 213                             u_int16_t *, u_int16_t *, u_int16_t *,
 214                             u_int16_t *, u_int8_t, sa_family_t);
 215 static void              pf_send_tcp(struct mbuf *,
 216                             const struct pf_rule *, sa_family_t,
 217                             const struct pf_addr *, const struct pf_addr *,
 218                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 219                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 220                             u_int16_t, struct ifnet *);
 221 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 222                             sa_family_t, struct pf_rule *);
 223 static void              pf_detach_state(struct pf_state *);
 224 static int               pf_state_key_attach(struct pf_state_key *,
 225                             struct pf_state_key *, struct pf_state *);
 226 static void              pf_state_key_detach(struct pf_state *, int);
 227 static int               pf_state_key_ctor(void *, int, void *, int);
 228 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 229 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 230                             int, struct pfi_kif *, struct mbuf *, int,
 231                             struct pf_pdesc *, struct pf_rule **,
 232                             struct pf_ruleset **, struct inpcb *);
 233 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 234                             struct pf_rule *, struct pf_pdesc *,
 235                             struct pf_src_node *, struct pf_state_key *,
 236                             struct pf_state_key *, struct mbuf *, int,
 237                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 238                             struct pf_state **, int, u_int16_t, u_int16_t,
 239                             int);
 240 static int               pf_test_fragment(struct pf_rule **, int,
 241                             struct pfi_kif *, struct mbuf *, void *,
 242                             struct pf_pdesc *, struct pf_rule **,
 243                             struct pf_ruleset **);
 244 static int               pf_tcp_track_full(struct pf_state_peer *,
 245                             struct pf_state_peer *, struct pf_state **,
 246                             struct pfi_kif *, struct mbuf *, int,
 247                             struct pf_pdesc *, u_short *, int *);
 248 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 249                             struct pf_state_peer *, struct pf_state **,
 250                             struct pf_pdesc *, u_short *);
 251 static int               pf_test_state_tcp(struct pf_state **, int,
 252                             struct pfi_kif *, struct mbuf *, int,
 253                             void *, struct pf_pdesc *, u_short *);
 254 static int               pf_test_state_udp(struct pf_state **, int,
 255                             struct pfi_kif *, struct mbuf *, int,
 256                             void *, struct pf_pdesc *);
 257 static int               pf_test_state_icmp(struct pf_state **, int,
 258                             struct pfi_kif *, struct mbuf *, int,
 259                             void *, struct pf_pdesc *, u_short *);
 260 static int               pf_test_state_other(struct pf_state **, int,
 261                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 262 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 263                             sa_family_t);
 264 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 267                                 int, u_int16_t);
 268 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 269                             u_int8_t, sa_family_t);
 270 static void              pf_print_state_parts(struct pf_state *,
 271                             struct pf_state_key *, struct pf_state_key *);
 272 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 273                             struct pf_addr_wrap *);
 274 static struct pf_state  *pf_find_state(struct pfi_kif *,
 275                             struct pf_state_key_cmp *, u_int);
 276 static int               pf_src_connlimit(struct pf_state **);
 277 static void              pf_overload_task(void *v, int pending);
 278 static int               pf_insert_src_node(struct pf_src_node **,
 279                             struct pf_rule *, struct pf_addr *, sa_family_t);
 280 static u_int             pf_purge_expired_states(u_int, int);
 281 static void              pf_purge_unlinked_rules(void);
 282 static int               pf_mtag_uminit(void *, int, int);
 283 static void              pf_mtag_free(struct m_tag *);
 284 #ifdef INET
 285 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 286                             struct ifnet *, struct pf_state *,
 287                             struct pf_pdesc *);
 288 #endif /* INET */
 289 #ifdef INET6
 290 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 291                             struct pf_addr *, u_int8_t);
 292 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 293                             struct ifnet *, struct pf_state *,
 294                             struct pf_pdesc *);
 295 #endif /* INET6 */
 296
 297 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 298
 299 VNET_DECLARE(int, pf_end_threads);
 300
 301 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 302
 303 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 304                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 305
 306 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 307         do {                                                            \
 308                 (s) = pf_find_state((i), (k), (d));                     \
 309                 if ((s) == NULL)                                        \
 310                         return (PF_DROP);                               \
 311                 if (PACKET_LOOPED(pd))                                  \
 312                         return (PF_PASS);                               \
 313                 if ((d) == PF_OUT &&                                    \
 314                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 315                     (s)->rule.ptr->direction == PF_OUT) ||              \
 316                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 317                     (s)->rule.ptr->direction == PF_IN)) &&              \
 318                     (s)->rt_kif != NULL &&                              \
 319                     (s)->rt_kif != (i))                                 \
 320                         return (PF_PASS);                               \
 321         } while (0)
 322
 323 #define BOUND_IFACE(r, k) \
 324         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 325
 326 #define STATE_INC_COUNTERS(s)                                           \
 327         do {                                                            \
 328                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 329                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 330                 if (s->anchor.ptr != NULL) {                            \
 331                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 332                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 333                 }                                                       \
 334                 if (s->nat_rule.ptr != NULL) {                          \
 335                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 336                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 337                 }                                                       \
 338         } while (0)
 339
 340 #define STATE_DEC_COUNTERS(s)                                           \
 341         do {                                                            \
 342                 if (s->nat_rule.ptr != NULL)                            \
 343                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 344                 if (s->anchor.ptr != NULL)                              \
 345                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 346                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 347         } while (0)
 348
 349 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 350 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 351 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 352 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 353
 354 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 355
 356 u_long  pf_hashmask;
 357 u_long  pf_srchashmask;
 358 static u_long   pf_hashsize;
 359 static u_long   pf_srchashsize;
 360
 361 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 362     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 363 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 364     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 365
 366 VNET_DEFINE(void *, pf_swi_cookie);
 367
 368 VNET_DEFINE(uint32_t, pf_hashseed);
 369 #define V_pf_hashseed   VNET(pf_hashseed)
 370
 371 int
 372 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 373 {
 374
 375         switch (af) {
 376 #ifdef INET
 377         case AF_INET:
 378                 if (a->addr32[0] > b->addr32[0])
 379                         return (1);
 380                 if (a->addr32[0] < b->addr32[0])
 381                         return (-1);
 382                 break;
 383 #endif /* INET */
 384 #ifdef INET6
 385         case AF_INET6:
 386                 if (a->addr32[3] > b->addr32[3])
 387                         return (1);
 388                 if (a->addr32[3] < b->addr32[3])
 389                         return (-1);
 390                 if (a->addr32[2] > b->addr32[2])
 391                         return (1);
 392                 if (a->addr32[2] < b->addr32[2])
 393                         return (-1);
 394                 if (a->addr32[1] > b->addr32[1])
 395                         return (1);
 396                 if (a->addr32[1] < b->addr32[1])
 397                         return (-1);
 398                 if (a->addr32[0] > b->addr32[0])
 399                         return (1);
 400                 if (a->addr32[0] < b->addr32[0])
 401                         return (-1);
 402                 break;
 403 #endif /* INET6 */
 404         default:
 405                 panic("%s: unknown address family %u", __func__, af);
 406         }
 407         return (0);
 408 }
 409
 410 static __inline uint32_t
 411 pf_hashkey(struct pf_state_key *sk)
 412 {
 413         uint32_t h;
 414
 415         h = murmur3_aligned_32((uint32_t *)sk,
 416                                sizeof(struct pf_state_key_cmp),
 417                                V_pf_hashseed);
 418
 419         return (h & pf_hashmask);
 420 }
 421
 422 static __inline uint32_t
 423 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 424 {
 425         uint32_t h;
 426
 427         switch (af) {
 428         case AF_INET:
 429                 h = murmur3_aligned_32((uint32_t *)&addr->v4,
 430                                        sizeof(addr->v4), V_pf_hashseed);
 431                 break;
 432         case AF_INET6:
 433                 h = murmur3_aligned_32((uint32_t *)&addr->v6,
 434                                        sizeof(addr->v6), V_pf_hashseed);
 435                 break;
 436         default:
 437                 panic("%s: unknown address family %u", __func__, af);
 438         }
 439
 440         return (h & pf_srchashmask);
 441 }
 442
 443 #ifdef ALTQ
 444 static int
 445 pf_state_hash(struct pf_state *s)
 446 {
 447         u_int32_t hv = (intptr_t)s / sizeof(*s);
 448
 449         hv ^= crc32(&s->src, sizeof(s->src));
 450         hv ^= crc32(&s->dst, sizeof(s->dst));
 451         if (hv == 0)
 452                 hv = 1;
 453         return (hv);
 454 }
 455 #endif
 456
 457 #ifdef INET6
 458 void
 459 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 460 {
 461         switch (af) {
 462 #ifdef INET
 463         case AF_INET:
 464                 dst->addr32[0] = src->addr32[0];
 465                 break;
 466 #endif /* INET */
 467         case AF_INET6:
 468                 dst->addr32[0] = src->addr32[0];
 469                 dst->addr32[1] = src->addr32[1];
 470                 dst->addr32[2] = src->addr32[2];
 471                 dst->addr32[3] = src->addr32[3];
 472                 break;
 473         }
 474 }
 475 #endif /* INET6 */
 476
 477 static void
 478 pf_init_threshold(struct pf_threshold *threshold,
 479     u_int32_t limit, u_int32_t seconds)
 480 {
 481         threshold->limit = limit * PF_THRESHOLD_MULT;
 482         threshold->seconds = seconds;
 483         threshold->count = 0;
 484         threshold->last = time_uptime;
 485 }
 486
 487 static void
 488 pf_add_threshold(struct pf_threshold *threshold)
 489 {
 490         u_int32_t t = time_uptime, diff = t - threshold->last;
 491
 492         if (diff >= threshold->seconds)
 493                 threshold->count = 0;
 494         else
 495                 threshold->count -= threshold->count * diff /
 496                     threshold->seconds;
 497         threshold->count += PF_THRESHOLD_MULT;
 498         threshold->last = t;
 499 }
 500
 501 static int
 502 pf_check_threshold(struct pf_threshold *threshold)
 503 {
 504         return (threshold->count > threshold->limit);
 505 }
 506
 507 static int
 508 pf_src_connlimit(struct pf_state **state)
 509 {
 510         struct pf_overload_entry *pfoe;
 511         int bad = 0;
 512
 513         PF_STATE_LOCK_ASSERT(*state);
 514
 515         (*state)->src_node->conn++;
 516         (*state)->src.tcp_est = 1;
 517         pf_add_threshold(&(*state)->src_node->conn_rate);
 518
 519         if ((*state)->rule.ptr->max_src_conn &&
 520             (*state)->rule.ptr->max_src_conn <
 521             (*state)->src_node->conn) {
 522                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 523                 bad++;
 524         }
 525
 526         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 527             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 528                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 529                 bad++;
 530         }
 531
 532         if (!bad)
 533                 return (0);
 534
 535         /* Kill this state. */
 536         (*state)->timeout = PFTM_PURGE;
 537         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 538
 539         if ((*state)->rule.ptr->overload_tbl == NULL)
 540                 return (1);
 541
 542         /* Schedule overloading and flushing task. */
 543         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 544         if (pfoe == NULL)
 545                 return (1);     /* too bad :( */
 546
 547         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 548         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 549         pfoe->rule = (*state)->rule.ptr;
 550         pfoe->dir = (*state)->direction;
 551         PF_OVERLOADQ_LOCK();
 552         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 553         PF_OVERLOADQ_UNLOCK();
 554         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 555
 556         return (1);
 557 }
 558
 559 static void
 560 pf_overload_task(void *v, int pending)
 561 {
 562         struct pf_overload_head queue;
 563         struct pfr_addr p;
 564         struct pf_overload_entry *pfoe, *pfoe1;
 565         uint32_t killed = 0;
 566
 567         CURVNET_SET((struct vnet *)v);
 568
 569         PF_OVERLOADQ_LOCK();
 570         queue = V_pf_overloadqueue;
 571         SLIST_INIT(&V_pf_overloadqueue);
 572         PF_OVERLOADQ_UNLOCK();
 573
 574         bzero(&p, sizeof(p));
 575         SLIST_FOREACH(pfoe, &queue, next) {
 576                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 577                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 578                         printf("%s: blocking address ", __func__);
 579                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 580                         printf("\n");
 581                 }
 582
 583                 p.pfra_af = pfoe->af;
 584                 switch (pfoe->af) {
 585 #ifdef INET
 586                 case AF_INET:
 587                         p.pfra_net = 32;
 588                         p.pfra_ip4addr = pfoe->addr.v4;
 589                         break;
 590 #endif
 591 #ifdef INET6
 592                 case AF_INET6:
 593                         p.pfra_net = 128;
 594                         p.pfra_ip6addr = pfoe->addr.v6;
 595                         break;
 596 #endif
 597                 }
 598
 599                 PF_RULES_WLOCK();
 600                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 601                 PF_RULES_WUNLOCK();
 602         }
 603
 604         /*
 605          * Remove those entries, that don't need flushing.
 606          */
 607         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 608                 if (pfoe->rule->flush == 0) {
 609                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 610                         free(pfoe, M_PFTEMP);
 611                 } else
 612                         counter_u64_add(
 613                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 614
 615         /* If nothing to flush, return. */
 616         if (SLIST_EMPTY(&queue)) {
 617                 CURVNET_RESTORE();
 618                 return;
 619         }
 620
 621         for (int i = 0; i <= pf_hashmask; i++) {
 622                 struct pf_idhash *ih = &V_pf_idhash[i];
 623                 struct pf_state_key *sk;
 624                 struct pf_state *s;
 625
 626                 PF_HASHROW_LOCK(ih);
 627                 LIST_FOREACH(s, &ih->states, entry) {
 628                     sk = s->key[PF_SK_WIRE];
 629                     SLIST_FOREACH(pfoe, &queue, next)
 630                         if (sk->af == pfoe->af &&
 631                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 632                             pfoe->rule == s->rule.ptr) &&
 633                             ((pfoe->dir == PF_OUT &&
 634                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 635                             (pfoe->dir == PF_IN &&
 636                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 637                                 s->timeout = PFTM_PURGE;
 638                                 s->src.state = s->dst.state = TCPS_CLOSED;
 639                                 killed++;
 640                         }
 641                 }
 642                 PF_HASHROW_UNLOCK(ih);
 643         }
 644         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 645                 free(pfoe, M_PFTEMP);
 646         if (V_pf_status.debug >= PF_DEBUG_MISC)
 647                 printf("%s: %u states killed", __func__, killed);
 648
 649         CURVNET_RESTORE();
 650 }
 651
 652 /*
 653  * Can return locked on failure, so that we can consistently
 654  * allocate and insert a new one.
 655  */
 656 struct pf_src_node *
 657 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 658         int returnlocked)
 659 {
 660         struct pf_srchash *sh;
 661         struct pf_src_node *n;
 662
 663         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 664
 665         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 666         PF_HASHROW_LOCK(sh);
 667         LIST_FOREACH(n, &sh->nodes, entry)
 668                 if (n->rule.ptr == rule && n->af == af &&
 669                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 670                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 671                         break;
 672         if (n != NULL) {
 673                 n->states++;
 674                 PF_HASHROW_UNLOCK(sh);
 675         } else if (returnlocked == 0)
 676                 PF_HASHROW_UNLOCK(sh);
 677
 678         return (n);
 679 }
 680
 681 static int
 682 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 683     struct pf_addr *src, sa_family_t af)
 684 {
 685
 686         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 687             rule->rpool.opts & PF_POOL_STICKYADDR),
 688             ("%s for non-tracking rule %p", __func__, rule));
 689
 690         if (*sn == NULL)
 691                 *sn = pf_find_src_node(src, rule, af, 1);
 692
 693         if (*sn == NULL) {
 694                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 695
 696                 PF_HASHROW_ASSERT(sh);
 697
 698                 if (!rule->max_src_nodes ||
 699                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 700                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 701                 else
 702                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 703                             1);
 704                 if ((*sn) == NULL) {
 705                         PF_HASHROW_UNLOCK(sh);
 706                         return (-1);
 707                 }
 708
 709                 pf_init_threshold(&(*sn)->conn_rate,
 710                     rule->max_src_conn_rate.limit,
 711                     rule->max_src_conn_rate.seconds);
 712
 713                 (*sn)->af = af;
 714                 (*sn)->rule.ptr = rule;
 715                 PF_ACPY(&(*sn)->addr, src, af);
 716                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 717                 (*sn)->creation = time_uptime;
 718                 (*sn)->ruletype = rule->action;
 719                 (*sn)->states = 1;
 720                 if ((*sn)->rule.ptr != NULL)
 721                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 722                 PF_HASHROW_UNLOCK(sh);
 723                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 724         } else {
 725                 if (rule->max_src_states &&
 726                     (*sn)->states >= rule->max_src_states) {
 727                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 728                             1);
 729                         return (-1);
 730                 }
 731         }
 732         return (0);
 733 }
 734
 735 void
 736 pf_unlink_src_node(struct pf_src_node *src)
 737 {
 738
 739         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 740         LIST_REMOVE(src, entry);
 741         if (src->rule.ptr)
 742                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 743 }
 744
 745 u_int
 746 pf_free_src_nodes(struct pf_src_node_list *head)
 747 {
 748         struct pf_src_node *sn, *tmp;
 749         u_int count = 0;
 750
 751         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 752                 uma_zfree(V_pf_sources_z, sn);
 753                 count++;
 754         }
 755
 756         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 757
 758         return (count);
 759 }
 760
 761 void
 762 pf_mtag_initialize()
 763 {
 764
 765         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 766             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 767             UMA_ALIGN_PTR, 0);
 768 }
 769
 770 /* Per-vnet data storage structures initialization. */
 771 void
 772 pf_initialize()
 773 {
 774         struct pf_keyhash       *kh;
 775         struct pf_idhash        *ih;
 776         struct pf_srchash       *sh;
 777         u_int i;
 778
 779         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &pf_hashsize);
 780         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 781                 pf_hashsize = PF_HASHSIZ;
 782         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &pf_srchashsize);
 783         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 784                 pf_srchashsize = PF_HASHSIZ / 4;
 785
 786         V_pf_hashseed = arc4random();
 787
 788         /* States and state keys storage. */
 789         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 790             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 791         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 792         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 793         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 794
 795         V_pf_state_key_z = uma_zcreate("pf state keys",
 796             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 797             UMA_ALIGN_PTR, 0);
 798         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 799             M_PFHASH, M_WAITOK | M_ZERO);
 800         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 801             M_PFHASH, M_WAITOK | M_ZERO);
 802         pf_hashmask = pf_hashsize - 1;
 803         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 804             i++, kh++, ih++) {
 805                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 806                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 807         }
 808
 809         /* Source nodes. */
 810         V_pf_sources_z = uma_zcreate("pf source nodes",
 811             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 812             0);
 813         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 814         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 815         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 816         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 817           M_PFHASH, M_WAITOK|M_ZERO);
 818         pf_srchashmask = pf_srchashsize - 1;
 819         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 820                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 821
 822         /* ALTQ */
 823         TAILQ_INIT(&V_pf_altqs[0]);
 824         TAILQ_INIT(&V_pf_altqs[1]);
 825         TAILQ_INIT(&V_pf_pabuf);
 826         V_pf_altqs_active = &V_pf_altqs[0];
 827         V_pf_altqs_inactive = &V_pf_altqs[1];
 828
 829
 830         /* Send & overload+flush queues. */
 831         STAILQ_INIT(&V_pf_sendqueue);
 832         SLIST_INIT(&V_pf_overloadqueue);
 833         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 834         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 835         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 836             MTX_DEF);
 837
 838         /* Unlinked, but may be referenced rules. */
 839         TAILQ_INIT(&V_pf_unlinked_rules);
 840         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 841 }
 842
 843 void
 844 pf_mtag_cleanup()
 845 {
 846
 847         uma_zdestroy(pf_mtag_z);
 848 }
 849
 850 void
 851 pf_cleanup()
 852 {
 853         struct pf_keyhash       *kh;
 854         struct pf_idhash        *ih;
 855         struct pf_srchash       *sh;
 856         struct pf_send_entry    *pfse, *next;
 857         u_int i;
 858
 859         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 860             i++, kh++, ih++) {
 861                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 862                     __func__));
 863                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 864                     __func__));
 865                 mtx_destroy(&kh->lock);
 866                 mtx_destroy(&ih->lock);
 867         }
 868         free(V_pf_keyhash, M_PFHASH);
 869         free(V_pf_idhash, M_PFHASH);
 870
 871         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 872                 KASSERT(LIST_EMPTY(&sh->nodes),
 873                     ("%s: source node hash not empty", __func__));
 874                 mtx_destroy(&sh->lock);
 875         }
 876         free(V_pf_srchash, M_PFHASH);
 877
 878         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 879                 m_freem(pfse->pfse_m);
 880                 free(pfse, M_PFTEMP);
 881         }
 882
 883         mtx_destroy(&pf_sendqueue_mtx);
 884         mtx_destroy(&pf_overloadqueue_mtx);
 885         mtx_destroy(&pf_unlnkdrules_mtx);
 886
 887         uma_zdestroy(V_pf_sources_z);
 888         uma_zdestroy(V_pf_state_z);
 889         uma_zdestroy(V_pf_state_key_z);
 890 }
 891
 892 static int
 893 pf_mtag_uminit(void *mem, int size, int how)
 894 {
 895         struct m_tag *t;
 896
 897         t = (struct m_tag *)mem;
 898         t->m_tag_cookie = MTAG_ABI_COMPAT;
 899         t->m_tag_id = PACKET_TAG_PF;
 900         t->m_tag_len = sizeof(struct pf_mtag);
 901         t->m_tag_free = pf_mtag_free;
 902
 903         return (0);
 904 }
 905
 906 static void
 907 pf_mtag_free(struct m_tag *t)
 908 {
 909
 910         uma_zfree(pf_mtag_z, t);
 911 }
 912
 913 struct pf_mtag *
 914 pf_get_mtag(struct mbuf *m)
 915 {
 916         struct m_tag *mtag;
 917
 918         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 919                 return ((struct pf_mtag *)(mtag + 1));
 920
 921         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 922         if (mtag == NULL)
 923                 return (NULL);
 924         bzero(mtag + 1, sizeof(struct pf_mtag));
 925         m_tag_prepend(m, mtag);
 926
 927         return ((struct pf_mtag *)(mtag + 1));
 928 }
 929
 930 static int
 931 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 932     struct pf_state *s)
 933 {
 934         struct pf_keyhash       *khs, *khw, *kh;
 935         struct pf_state_key     *sk, *cur;
 936         struct pf_state         *si, *olds = NULL;
 937         int idx;
 938
 939         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 940         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 941         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 942
 943         /*
 944          * We need to lock hash slots of both keys. To avoid deadlock
 945          * we always lock the slot with lower address first. Unlock order
 946          * isn't important.
 947          *
 948          * We also need to lock ID hash slot before dropping key
 949          * locks. On success we return with ID hash slot locked.
 950          */
 951
 952         if (skw == sks) {
 953                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 954                 PF_HASHROW_LOCK(khs);
 955         } else {
 956                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 957                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 958                 if (khs == khw) {
 959                         PF_HASHROW_LOCK(khs);
 960                 } else if (khs < khw) {
 961                         PF_HASHROW_LOCK(khs);
 962                         PF_HASHROW_LOCK(khw);
 963                 } else {
 964                         PF_HASHROW_LOCK(khw);
 965                         PF_HASHROW_LOCK(khs);
 966                 }
 967         }
 968
 969 #define KEYS_UNLOCK()   do {                    \
 970         if (khs != khw) {                       \
 971                 PF_HASHROW_UNLOCK(khs);         \
 972                 PF_HASHROW_UNLOCK(khw);         \
 973         } else                                  \
 974                 PF_HASHROW_UNLOCK(khs);         \
 975 } while (0)
 976
 977         /*
 978          * First run: start with wire key.
 979          */
 980         sk = skw;
 981         kh = khw;
 982         idx = PF_SK_WIRE;
 983
 984 keyattach:
 985         LIST_FOREACH(cur, &kh->keys, entry)
 986                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 987                         break;
 988
 989         if (cur != NULL) {
 990                 /* Key exists. Check for same kif, if none, add to key. */
 991                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 992                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 993
 994                         PF_HASHROW_LOCK(ih);
 995                         if (si->kif == s->kif &&
 996                             si->direction == s->direction) {
 997                                 if (sk->proto == IPPROTO_TCP &&
 998                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 999                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1000                                         /*
1001                                          * New state matches an old >FIN_WAIT_2
1002                                          * state. We can't drop key hash locks,
1003                                          * thus we can't unlink it properly.
1004                                          *
1005                                          * As a workaround we drop it into
1006                                          * TCPS_CLOSED state, schedule purge
1007                                          * ASAP and push it into the very end
1008                                          * of the slot TAILQ, so that it won't
1009                                          * conflict with our new state.
1010                                          */
1011                                         si->src.state = si->dst.state =
1012                                             TCPS_CLOSED;
1013                                         si->timeout = PFTM_PURGE;
1014                                         olds = si;
1015                                 } else {
1016                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1017                                                 printf("pf: %s key attach "
1018                                                     "failed on %s: ",
1019                                                     (idx == PF_SK_WIRE) ?
1020                                                     "wire" : "stack",
1021                                                     s->kif->pfik_name);
1022                                                 pf_print_state_parts(s,
1023                                                     (idx == PF_SK_WIRE) ?
1024                                                     sk : NULL,
1025                                                     (idx == PF_SK_STACK) ?
1026                                                     sk : NULL);
1027                                                 printf(", existing: ");
1028                                                 pf_print_state_parts(si,
1029                                                     (idx == PF_SK_WIRE) ?
1030                                                     sk : NULL,
1031                                                     (idx == PF_SK_STACK) ?
1032                                                     sk : NULL);
1033                                                 printf("\n");
1034                                         }
1035                                         PF_HASHROW_UNLOCK(ih);
1036                                         KEYS_UNLOCK();
1037                                         uma_zfree(V_pf_state_key_z, sk);
1038                                         if (idx == PF_SK_STACK)
1039                                                 pf_detach_state(s);
1040                                         return (EEXIST); /* collision! */
1041                                 }
1042                         }
1043                         PF_HASHROW_UNLOCK(ih);
1044                 }
1045                 uma_zfree(V_pf_state_key_z, sk);
1046                 s->key[idx] = cur;
1047         } else {
1048                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1049                 s->key[idx] = sk;
1050         }
1051
1052 stateattach:
1053         /* List is sorted, if-bound states before floating. */
1054         if (s->kif == V_pfi_all)
1055                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1056         else
1057                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1058
1059         if (olds) {
1060                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1061                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1062                     key_list[idx]);
1063                 olds = NULL;
1064         }
1065
1066         /*
1067          * Attach done. See how should we (or should not?)
1068          * attach a second key.
1069          */
1070         if (sks == skw) {
1071                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1072                 idx = PF_SK_STACK;
1073                 sks = NULL;
1074                 goto stateattach;
1075         } else if (sks != NULL) {
1076                 /*
1077                  * Continue attaching with stack key.
1078                  */
1079                 sk = sks;
1080                 kh = khs;
1081                 idx = PF_SK_STACK;
1082                 sks = NULL;
1083                 goto keyattach;
1084         }
1085
1086         PF_STATE_LOCK(s);
1087         KEYS_UNLOCK();
1088
1089         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1090             ("%s failure", __func__));
1091
1092         return (0);
1093 #undef  KEYS_UNLOCK
1094 }
1095
1096 static void
1097 pf_detach_state(struct pf_state *s)
1098 {
1099         struct pf_state_key *sks = s->key[PF_SK_STACK];
1100         struct pf_keyhash *kh;
1101
1102         if (sks != NULL) {
1103                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1104                 PF_HASHROW_LOCK(kh);
1105                 if (s->key[PF_SK_STACK] != NULL)
1106                         pf_state_key_detach(s, PF_SK_STACK);
1107                 /*
1108                  * If both point to same key, then we are done.
1109                  */
1110                 if (sks == s->key[PF_SK_WIRE]) {
1111                         pf_state_key_detach(s, PF_SK_WIRE);
1112                         PF_HASHROW_UNLOCK(kh);
1113                         return;
1114                 }
1115                 PF_HASHROW_UNLOCK(kh);
1116         }
1117
1118         if (s->key[PF_SK_WIRE] != NULL) {
1119                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1120                 PF_HASHROW_LOCK(kh);
1121                 if (s->key[PF_SK_WIRE] != NULL)
1122                         pf_state_key_detach(s, PF_SK_WIRE);
1123                 PF_HASHROW_UNLOCK(kh);
1124         }
1125 }
1126
1127 static void
1128 pf_state_key_detach(struct pf_state *s, int idx)
1129 {
1130         struct pf_state_key *sk = s->key[idx];
1131 #ifdef INVARIANTS
1132         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1133
1134         PF_HASHROW_ASSERT(kh);
1135 #endif
1136         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1137         s->key[idx] = NULL;
1138
1139         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1140                 LIST_REMOVE(sk, entry);
1141                 uma_zfree(V_pf_state_key_z, sk);
1142         }
1143 }
1144
1145 static int
1146 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1147 {
1148         struct pf_state_key *sk = mem;
1149
1150         bzero(sk, sizeof(struct pf_state_key_cmp));
1151         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1152         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1153
1154         return (0);
1155 }
1156
1157 struct pf_state_key *
1158 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1159         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1160 {
1161         struct pf_state_key *sk;
1162
1163         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1164         if (sk == NULL)
1165                 return (NULL);
1166
1167         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1168         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1169         sk->port[pd->sidx] = sport;
1170         sk->port[pd->didx] = dport;
1171         sk->proto = pd->proto;
1172         sk->af = pd->af;
1173
1174         return (sk);
1175 }
1176
1177 struct pf_state_key *
1178 pf_state_key_clone(struct pf_state_key *orig)
1179 {
1180         struct pf_state_key *sk;
1181
1182         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1183         if (sk == NULL)
1184                 return (NULL);
1185
1186         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1187
1188         return (sk);
1189 }
1190
1191 int
1192 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1193     struct pf_state_key *sks, struct pf_state *s)
1194 {
1195         struct pf_idhash *ih;
1196         struct pf_state *cur;
1197         int error;
1198
1199         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1200             ("%s: sks not pristine", __func__));
1201         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1202             ("%s: skw not pristine", __func__));
1203         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1204
1205         s->kif = kif;
1206
1207         if (s->id == 0 && s->creatorid == 0) {
1208                 /* XXX: should be atomic, but probability of collision low */
1209                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1210                         V_pf_stateid[curcpu] = 1;
1211                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1212                 s->id = htobe64(s->id);
1213                 s->creatorid = V_pf_status.hostid;
1214         }
1215
1216         /* Returns with ID locked on success. */
1217         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1218                 return (error);
1219
1220         ih = &V_pf_idhash[PF_IDHASH(s)];
1221         PF_HASHROW_ASSERT(ih);
1222         LIST_FOREACH(cur, &ih->states, entry)
1223                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1224                         break;
1225
1226         if (cur != NULL) {
1227                 PF_HASHROW_UNLOCK(ih);
1228                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1229                         printf("pf: state ID collision: "
1230                             "id: %016llx creatorid: %08x\n",
1231                             (unsigned long long)be64toh(s->id),
1232                             ntohl(s->creatorid));
1233                 }
1234                 pf_detach_state(s);
1235                 return (EEXIST);
1236         }
1237         LIST_INSERT_HEAD(&ih->states, s, entry);
1238         /* One for keys, one for ID hash. */
1239         refcount_init(&s->refs, 2);
1240
1241         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1242         if (pfsync_insert_state_ptr != NULL)
1243                 pfsync_insert_state_ptr(s);
1244
1245         /* Returns locked. */
1246         return (0);
1247 }
1248
1249 /*
1250  * Find state by ID: returns with locked row on success.
1251  */
1252 struct pf_state *
1253 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1254 {
1255         struct pf_idhash *ih;
1256         struct pf_state *s;
1257
1258         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1259
1260         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1261
1262         PF_HASHROW_LOCK(ih);
1263         LIST_FOREACH(s, &ih->states, entry)
1264                 if (s->id == id && s->creatorid == creatorid)
1265                         break;
1266
1267         if (s == NULL)
1268                 PF_HASHROW_UNLOCK(ih);
1269
1270         return (s);
1271 }
1272
1273 /*
1274  * Find state by key.
1275  * Returns with ID hash slot locked on success.
1276  */
1277 static struct pf_state *
1278 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1279 {
1280         struct pf_keyhash       *kh;
1281         struct pf_state_key     *sk;
1282         struct pf_state         *s;
1283         int idx;
1284
1285         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1286
1287         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1288
1289         PF_HASHROW_LOCK(kh);
1290         LIST_FOREACH(sk, &kh->keys, entry)
1291                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1292                         break;
1293         if (sk == NULL) {
1294                 PF_HASHROW_UNLOCK(kh);
1295                 return (NULL);
1296         }
1297
1298         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1299
1300         /* List is sorted, if-bound states before floating ones. */
1301         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1302                 if (s->kif == V_pfi_all || s->kif == kif) {
1303                         PF_STATE_LOCK(s);
1304                         PF_HASHROW_UNLOCK(kh);
1305                         if (s->timeout >= PFTM_MAX) {
1306                                 /*
1307                                  * State is either being processed by
1308                                  * pf_unlink_state() in an other thread, or
1309                                  * is scheduled for immediate expiry.
1310                                  */
1311                                 PF_STATE_UNLOCK(s);
1312                                 return (NULL);
1313                         }
1314                         return (s);
1315                 }
1316         PF_HASHROW_UNLOCK(kh);
1317
1318         return (NULL);
1319 }
1320
1321 struct pf_state *
1322 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1323 {
1324         struct pf_keyhash       *kh;
1325         struct pf_state_key     *sk;
1326         struct pf_state         *s, *ret = NULL;
1327         int                      idx, inout = 0;
1328
1329         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1330
1331         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1332
1333         PF_HASHROW_LOCK(kh);
1334         LIST_FOREACH(sk, &kh->keys, entry)
1335                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1336                         break;
1337         if (sk == NULL) {
1338                 PF_HASHROW_UNLOCK(kh);
1339                 return (NULL);
1340         }
1341         switch (dir) {
1342         case PF_IN:
1343                 idx = PF_SK_WIRE;
1344                 break;
1345         case PF_OUT:
1346                 idx = PF_SK_STACK;
1347                 break;
1348         case PF_INOUT:
1349                 idx = PF_SK_WIRE;
1350                 inout = 1;
1351                 break;
1352         default:
1353                 panic("%s: dir %u", __func__, dir);
1354         }
1355 second_run:
1356         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1357                 if (more == NULL) {
1358                         PF_HASHROW_UNLOCK(kh);
1359                         return (s);
1360                 }
1361
1362                 if (ret)
1363                         (*more)++;
1364                 else
1365                         ret = s;
1366         }
1367         if (inout == 1) {
1368                 inout = 0;
1369                 idx = PF_SK_STACK;
1370                 goto second_run;
1371         }
1372         PF_HASHROW_UNLOCK(kh);
1373
1374         return (ret);
1375 }
1376
1377 /* END state table stuff */
1378
1379 static void
1380 pf_send(struct pf_send_entry *pfse)
1381 {
1382
1383         PF_SENDQ_LOCK();
1384         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1385         PF_SENDQ_UNLOCK();
1386         swi_sched(V_pf_swi_cookie, 0);
1387 }
1388
1389 void
1390 pf_intr(void *v)
1391 {
1392         struct pf_send_head queue;
1393         struct pf_send_entry *pfse, *next;
1394
1395         CURVNET_SET((struct vnet *)v);
1396
1397         PF_SENDQ_LOCK();
1398         queue = V_pf_sendqueue;
1399         STAILQ_INIT(&V_pf_sendqueue);
1400         PF_SENDQ_UNLOCK();
1401
1402         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1403                 switch (pfse->pfse_type) {
1404 #ifdef INET
1405                 case PFSE_IP:
1406                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1407                         break;
1408                 case PFSE_ICMP:
1409                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1410                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1411                         break;
1412 #endif /* INET */
1413 #ifdef INET6
1414                 case PFSE_IP6:
1415                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1416                             NULL);
1417                         break;
1418                 case PFSE_ICMP6:
1419                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1420                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1421                         break;
1422 #endif /* INET6 */
1423                 default:
1424                         panic("%s: unknown type", __func__);
1425                 }
1426                 free(pfse, M_PFTEMP);
1427         }
1428         CURVNET_RESTORE();
1429 }
1430
1431 void
1432 pf_purge_thread(void *v)
1433 {
1434         u_int idx = 0;
1435
1436         CURVNET_SET((struct vnet *)v);
1437
1438         for (;;) {
1439                 PF_RULES_RLOCK();
1440                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1441
1442                 if (V_pf_end_threads) {
1443                         /*
1444                          * To cleanse up all kifs and rules we need
1445                          * two runs: first one clears reference flags,
1446                          * then pf_purge_expired_states() doesn't
1447                          * raise them, and then second run frees.
1448                          */
1449                         PF_RULES_RUNLOCK();
1450                         pf_purge_unlinked_rules();
1451                         pfi_kif_purge();
1452
1453                         /*
1454                          * Now purge everything.
1455                          */
1456                         pf_purge_expired_states(0, pf_hashmask);
1457                         pf_purge_expired_fragments();
1458                         pf_purge_expired_src_nodes();
1459
1460                         /*
1461                          * Now all kifs & rules should be unreferenced,
1462                          * thus should be successfully freed.
1463                          */
1464                         pf_purge_unlinked_rules();
1465                         pfi_kif_purge();
1466
1467                         /*
1468                          * Announce success and exit.
1469                          */
1470                         PF_RULES_RLOCK();
1471                         V_pf_end_threads++;
1472                         PF_RULES_RUNLOCK();
1473                         wakeup(pf_purge_thread);
1474                         kproc_exit(0);
1475                 }
1476                 PF_RULES_RUNLOCK();
1477
1478                 /* Process 1/interval fraction of the state table every run. */
1479                 idx = pf_purge_expired_states(idx, pf_hashmask /
1480                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1481
1482                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1483                 if (idx == 0) {
1484                         /*
1485                          * Order is important:
1486                          * - states and src nodes reference rules
1487                          * - states and rules reference kifs
1488                          */
1489                         pf_purge_expired_fragments();
1490                         pf_purge_expired_src_nodes();
1491                         pf_purge_unlinked_rules();
1492                         pfi_kif_purge();
1493                 }
1494         }
1495         /* not reached */
1496         CURVNET_RESTORE();
1497 }
1498
1499 u_int32_t
1500 pf_state_expires(const struct pf_state *state)
1501 {
1502         u_int32_t       timeout;
1503         u_int32_t       start;
1504         u_int32_t       end;
1505         u_int32_t       states;
1506
1507         /* handle all PFTM_* > PFTM_MAX here */
1508         if (state->timeout == PFTM_PURGE)
1509                 return (time_uptime);
1510         KASSERT(state->timeout != PFTM_UNLINKED,
1511             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1512         KASSERT((state->timeout < PFTM_MAX),
1513             ("pf_state_expires: timeout > PFTM_MAX"));
1514         timeout = state->rule.ptr->timeout[state->timeout];
1515         if (!timeout)
1516                 timeout = V_pf_default_rule.timeout[state->timeout];
1517         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1518         if (start) {
1519                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1520                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1521         } else {
1522                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1523                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1524                 states = V_pf_status.states;
1525         }
1526         if (end && states > start && start < end) {
1527                 if (states < end)
1528                         return (state->expire + timeout * (end - states) /
1529                             (end - start));
1530                 else
1531                         return (time_uptime);
1532         }
1533         return (state->expire + timeout);
1534 }
1535
1536 void
1537 pf_purge_expired_src_nodes()
1538 {
1539         struct pf_src_node_list  freelist;
1540         struct pf_srchash       *sh;
1541         struct pf_src_node      *cur, *next;
1542         int i;
1543
1544         LIST_INIT(&freelist);
1545         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1546             PF_HASHROW_LOCK(sh);
1547             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1548                 if (cur->states == 0 && cur->expire <= time_uptime) {
1549                         pf_unlink_src_node(cur);
1550                         LIST_INSERT_HEAD(&freelist, cur, entry);
1551                 } else if (cur->rule.ptr != NULL)
1552                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1553             PF_HASHROW_UNLOCK(sh);
1554         }
1555
1556         pf_free_src_nodes(&freelist);
1557
1558         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1559 }
1560
1561 static void
1562 pf_src_tree_remove_state(struct pf_state *s)
1563 {
1564         struct pf_src_node *sn;
1565         struct pf_srchash *sh;
1566         uint32_t timeout;
1567
1568         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1569             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1570             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1571
1572         if (s->src_node != NULL) {
1573                 sn = s->src_node;
1574                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1575                 PF_HASHROW_LOCK(sh);
1576                 if (s->src.tcp_est)
1577                         --sn->conn;
1578                 if (--sn->states == 0)
1579                         sn->expire = time_uptime + timeout;
1580                 PF_HASHROW_UNLOCK(sh);
1581         }
1582         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1583                 sn = s->nat_src_node;
1584                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1585                 PF_HASHROW_LOCK(sh);
1586                 if (--sn->states == 0)
1587                         sn->expire = time_uptime + timeout;
1588                 PF_HASHROW_UNLOCK(sh);
1589         }
1590         s->src_node = s->nat_src_node = NULL;
1591 }
1592
1593 /*
1594  * Unlink and potentilly free a state. Function may be
1595  * called with ID hash row locked, but always returns
1596  * unlocked, since it needs to go through key hash locking.
1597  */
1598 int
1599 pf_unlink_state(struct pf_state *s, u_int flags)
1600 {
1601         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1602
1603         if ((flags & PF_ENTER_LOCKED) == 0)
1604                 PF_HASHROW_LOCK(ih);
1605         else
1606                 PF_HASHROW_ASSERT(ih);
1607
1608         if (s->timeout == PFTM_UNLINKED) {
1609                 /*
1610                  * State is being processed
1611                  * by pf_unlink_state() in
1612                  * an other thread.
1613                  */
1614                 PF_HASHROW_UNLOCK(ih);
1615                 return (0);     /* XXXGL: undefined actually */
1616         }
1617
1618         if (s->src.state == PF_TCPS_PROXY_DST) {
1619                 /* XXX wire key the right one? */
1620                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1621                     &s->key[PF_SK_WIRE]->addr[1],
1622                     &s->key[PF_SK_WIRE]->addr[0],
1623                     s->key[PF_SK_WIRE]->port[1],
1624                     s->key[PF_SK_WIRE]->port[0],
1625                     s->src.seqhi, s->src.seqlo + 1,
1626                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1627         }
1628
1629         LIST_REMOVE(s, entry);
1630         pf_src_tree_remove_state(s);
1631
1632         if (pfsync_delete_state_ptr != NULL)
1633                 pfsync_delete_state_ptr(s);
1634
1635         STATE_DEC_COUNTERS(s);
1636
1637         s->timeout = PFTM_UNLINKED;
1638
1639         PF_HASHROW_UNLOCK(ih);
1640
1641         pf_detach_state(s);
1642         refcount_release(&s->refs);
1643
1644         return (pf_release_state(s));
1645 }
1646
1647 void
1648 pf_free_state(struct pf_state *cur)
1649 {
1650
1651         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1652         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1653             cur->timeout));
1654
1655         pf_normalize_tcp_cleanup(cur);
1656         uma_zfree(V_pf_state_z, cur);
1657         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1658 }
1659
1660 /*
1661  * Called only from pf_purge_thread(), thus serialized.
1662  */
1663 static u_int
1664 pf_purge_expired_states(u_int i, int maxcheck)
1665 {
1666         struct pf_idhash *ih;
1667         struct pf_state *s;
1668
1669         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1670
1671         /*
1672          * Go through hash and unlink states that expire now.
1673          */
1674         while (maxcheck > 0) {
1675
1676                 ih = &V_pf_idhash[i];
1677 relock:
1678                 PF_HASHROW_LOCK(ih);
1679                 LIST_FOREACH(s, &ih->states, entry) {
1680                         if (pf_state_expires(s) <= time_uptime) {
1681                                 V_pf_status.states -=
1682                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1683                                 goto relock;
1684                         }
1685                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1686                         if (s->nat_rule.ptr != NULL)
1687                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1688                         if (s->anchor.ptr != NULL)
1689                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1690                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1691                         if (s->rt_kif)
1692                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1693                 }
1694                 PF_HASHROW_UNLOCK(ih);
1695
1696                 /* Return when we hit end of hash. */
1697                 if (++i > pf_hashmask) {
1698                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1699                         return (0);
1700                 }
1701
1702                 maxcheck--;
1703         }
1704
1705         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1706
1707         return (i);
1708 }
1709
1710 static void
1711 pf_purge_unlinked_rules()
1712 {
1713         struct pf_rulequeue tmpq;
1714         struct pf_rule *r, *r1;
1715
1716         /*
1717          * If we have overloading task pending, then we'd
1718          * better skip purging this time. There is a tiny
1719          * probability that overloading task references
1720          * an already unlinked rule.
1721          */
1722         PF_OVERLOADQ_LOCK();
1723         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1724                 PF_OVERLOADQ_UNLOCK();
1725                 return;
1726         }
1727         PF_OVERLOADQ_UNLOCK();
1728
1729         /*
1730          * Do naive mark-and-sweep garbage collecting of old rules.
1731          * Reference flag is raised by pf_purge_expired_states()
1732          * and pf_purge_expired_src_nodes().
1733          *
1734          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1735          * use a temporary queue.
1736          */
1737         TAILQ_INIT(&tmpq);
1738         PF_UNLNKDRULES_LOCK();
1739         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1740                 if (!(r->rule_flag & PFRULE_REFS)) {
1741                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1742                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1743                 } else
1744                         r->rule_flag &= ~PFRULE_REFS;
1745         }
1746         PF_UNLNKDRULES_UNLOCK();
1747
1748         if (!TAILQ_EMPTY(&tmpq)) {
1749                 PF_RULES_WLOCK();
1750                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1751                         TAILQ_REMOVE(&tmpq, r, entries);
1752                         pf_free_rule(r);
1753                 }
1754                 PF_RULES_WUNLOCK();
1755         }
1756 }
1757
1758 void
1759 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1760 {
1761         switch (af) {
1762 #ifdef INET
1763         case AF_INET: {
1764                 u_int32_t a = ntohl(addr->addr32[0]);
1765                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1766                     (a>>8)&255, a&255);
1767                 if (p) {
1768                         p = ntohs(p);
1769                         printf(":%u", p);
1770                 }
1771                 break;
1772         }
1773 #endif /* INET */
1774 #ifdef INET6
1775         case AF_INET6: {
1776                 u_int16_t b;
1777                 u_int8_t i, curstart, curend, maxstart, maxend;
1778                 curstart = curend = maxstart = maxend = 255;
1779                 for (i = 0; i < 8; i++) {
1780                         if (!addr->addr16[i]) {
1781                                 if (curstart == 255)
1782                                         curstart = i;
1783                                 curend = i;
1784                         } else {
1785                                 if ((curend - curstart) >
1786                                     (maxend - maxstart)) {
1787                                         maxstart = curstart;
1788                                         maxend = curend;
1789                                 }
1790                                 curstart = curend = 255;
1791                         }
1792                 }
1793                 if ((curend - curstart) >
1794                     (maxend - maxstart)) {
1795                         maxstart = curstart;
1796                         maxend = curend;
1797                 }
1798                 for (i = 0; i < 8; i++) {
1799                         if (i >= maxstart && i <= maxend) {
1800                                 if (i == 0)
1801                                         printf(":");
1802                                 if (i == maxend)
1803                                         printf(":");
1804                         } else {
1805                                 b = ntohs(addr->addr16[i]);
1806                                 printf("%x", b);
1807                                 if (i < 7)
1808                                         printf(":");
1809                         }
1810                 }
1811                 if (p) {
1812                         p = ntohs(p);
1813                         printf("[%u]", p);
1814                 }
1815                 break;
1816         }
1817 #endif /* INET6 */
1818         }
1819 }
1820
1821 void
1822 pf_print_state(struct pf_state *s)
1823 {
1824         pf_print_state_parts(s, NULL, NULL);
1825 }
1826
1827 static void
1828 pf_print_state_parts(struct pf_state *s,
1829     struct pf_state_key *skwp, struct pf_state_key *sksp)
1830 {
1831         struct pf_state_key *skw, *sks;
1832         u_int8_t proto, dir;
1833
1834         /* Do our best to fill these, but they're skipped if NULL */
1835         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1836         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1837         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1838         dir = s ? s->direction : 0;
1839
1840         switch (proto) {
1841         case IPPROTO_IPV4:
1842                 printf("IPv4");
1843                 break;
1844         case IPPROTO_IPV6:
1845                 printf("IPv6");
1846                 break;
1847         case IPPROTO_TCP:
1848                 printf("TCP");
1849                 break;
1850         case IPPROTO_UDP:
1851                 printf("UDP");
1852                 break;
1853         case IPPROTO_ICMP:
1854                 printf("ICMP");
1855                 break;
1856         case IPPROTO_ICMPV6:
1857                 printf("ICMPv6");
1858                 break;
1859         default:
1860                 printf("%u", skw->proto);
1861                 break;
1862         }
1863         switch (dir) {
1864         case PF_IN:
1865                 printf(" in");
1866                 break;
1867         case PF_OUT:
1868                 printf(" out");
1869                 break;
1870         }
1871         if (skw) {
1872                 printf(" wire: ");
1873                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1874                 printf(" ");
1875                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1876         }
1877         if (sks) {
1878                 printf(" stack: ");
1879                 if (sks != skw) {
1880                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1881                         printf(" ");
1882                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1883                 } else
1884                         printf("-");
1885         }
1886         if (s) {
1887                 if (proto == IPPROTO_TCP) {
1888                         printf(" [lo=%u high=%u win=%u modulator=%u",
1889                             s->src.seqlo, s->src.seqhi,
1890                             s->src.max_win, s->src.seqdiff);
1891                         if (s->src.wscale && s->dst.wscale)
1892                                 printf(" wscale=%u",
1893                                     s->src.wscale & PF_WSCALE_MASK);
1894                         printf("]");
1895                         printf(" [lo=%u high=%u win=%u modulator=%u",
1896                             s->dst.seqlo, s->dst.seqhi,
1897                             s->dst.max_win, s->dst.seqdiff);
1898                         if (s->src.wscale && s->dst.wscale)
1899                                 printf(" wscale=%u",
1900                                 s->dst.wscale & PF_WSCALE_MASK);
1901                         printf("]");
1902                 }
1903                 printf(" %u:%u", s->src.state, s->dst.state);
1904         }
1905 }
1906
1907 void
1908 pf_print_flags(u_int8_t f)
1909 {
1910         if (f)
1911                 printf(" ");
1912         if (f & TH_FIN)
1913                 printf("F");
1914         if (f & TH_SYN)
1915                 printf("S");
1916         if (f & TH_RST)
1917                 printf("R");
1918         if (f & TH_PUSH)
1919                 printf("P");
1920         if (f & TH_ACK)
1921                 printf("A");
1922         if (f & TH_URG)
1923                 printf("U");
1924         if (f & TH_ECE)
1925                 printf("E");
1926         if (f & TH_CWR)
1927                 printf("W");
1928 }
1929
1930 #define PF_SET_SKIP_STEPS(i)                                    \
1931         do {                                                    \
1932                 while (head[i] != cur) {                        \
1933                         head[i]->skip[i].ptr = cur;             \
1934                         head[i] = TAILQ_NEXT(head[i], entries); \
1935                 }                                               \
1936         } while (0)
1937
1938 void
1939 pf_calc_skip_steps(struct pf_rulequeue *rules)
1940 {
1941         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1942         int i;
1943
1944         cur = TAILQ_FIRST(rules);
1945         prev = cur;
1946         for (i = 0; i < PF_SKIP_COUNT; ++i)
1947                 head[i] = cur;
1948         while (cur != NULL) {
1949
1950                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1951                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1952                 if (cur->direction != prev->direction)
1953                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1954                 if (cur->af != prev->af)
1955                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1956                 if (cur->proto != prev->proto)
1957                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1958                 if (cur->src.neg != prev->src.neg ||
1959                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1960                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1961                 if (cur->src.port[0] != prev->src.port[0] ||
1962                     cur->src.port[1] != prev->src.port[1] ||
1963                     cur->src.port_op != prev->src.port_op)
1964                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1965                 if (cur->dst.neg != prev->dst.neg ||
1966                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1967                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1968                 if (cur->dst.port[0] != prev->dst.port[0] ||
1969                     cur->dst.port[1] != prev->dst.port[1] ||
1970                     cur->dst.port_op != prev->dst.port_op)
1971                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1972
1973                 prev = cur;
1974                 cur = TAILQ_NEXT(cur, entries);
1975         }
1976         for (i = 0; i < PF_SKIP_COUNT; ++i)
1977                 PF_SET_SKIP_STEPS(i);
1978 }
1979
1980 static int
1981 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1982 {
1983         if (aw1->type != aw2->type)
1984                 return (1);
1985         switch (aw1->type) {
1986         case PF_ADDR_ADDRMASK:
1987         case PF_ADDR_RANGE:
1988                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
1989                         return (1);
1990                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
1991                         return (1);
1992                 return (0);
1993         case PF_ADDR_DYNIFTL:
1994                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1995         case PF_ADDR_NOROUTE:
1996         case PF_ADDR_URPFFAILED:
1997                 return (0);
1998         case PF_ADDR_TABLE:
1999                 return (aw1->p.tbl != aw2->p.tbl);
2000         default:
2001                 printf("invalid address type: %d\n", aw1->type);
2002                 return (1);
2003         }
2004 }
2005
2006 /**
2007  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2008  * header isn't always a full checksum. In some cases (i.e. output) it's a
2009  * pseudo-header checksum, which is a partial checksum over src/dst IP
2010  * addresses, protocol number and length.
2011  *
2012  * That means we have the following cases:
2013  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2014  *      checksums, we need to update the checksum whenever we change anything.
2015  *  * Output (i.e. the checksum is a pseudo-header checksum):
2016  *      x The field being updated is src/dst address or affects the length of
2017  *      the packet. We need to update the pseudo-header checksum (note that this
2018  *      checksum is not ones' complement).
2019  *      x Some other field is being modified (e.g. src/dst port numbers): We
2020  *      don't have to update anything.
2021  **/
2022 u_int16_t
2023 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2024 {
2025         u_int32_t       l;
2026
2027         if (udp && !cksum)
2028                 return (0x0000);
2029         l = cksum + old - new;
2030         l = (l >> 16) + (l & 65535);
2031         l = l & 65535;
2032         if (udp && !l)
2033                 return (0xFFFF);
2034         return (l);
2035 }
2036
2037 u_int16_t
2038 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2039         u_int16_t new, u_int8_t udp)
2040 {
2041         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2042                 return (cksum);
2043
2044         return (pf_cksum_fixup(cksum, old, new, udp));
2045 }
2046
2047 static void
2048 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2049         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2050         sa_family_t af)
2051 {
2052         struct pf_addr  ao;
2053         u_int16_t       po = *p;
2054
2055         PF_ACPY(&ao, a, af);
2056         PF_ACPY(a, an, af);
2057
2058         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2059                 *pc = ~*pc;
2060
2061         *p = pn;
2062
2063         switch (af) {
2064 #ifdef INET
2065         case AF_INET:
2066                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2067                     ao.addr16[0], an->addr16[0], 0),
2068                     ao.addr16[1], an->addr16[1], 0);
2069                 *p = pn;
2070
2071                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2072                     ao.addr16[0], an->addr16[0], u),
2073                     ao.addr16[1], an->addr16[1], u);
2074
2075                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2076                 break;
2077 #endif /* INET */
2078 #ifdef INET6
2079         case AF_INET6:
2080                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2081                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2082                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2083                     ao.addr16[0], an->addr16[0], u),
2084                     ao.addr16[1], an->addr16[1], u),
2085                     ao.addr16[2], an->addr16[2], u),
2086                     ao.addr16[3], an->addr16[3], u),
2087                     ao.addr16[4], an->addr16[4], u),
2088                     ao.addr16[5], an->addr16[5], u),
2089                     ao.addr16[6], an->addr16[6], u),
2090                     ao.addr16[7], an->addr16[7], u);
2091
2092                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2093                 break;
2094 #endif /* INET6 */
2095         }
2096
2097         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2098             CSUM_DELAY_DATA_IPV6)) {
2099                 *pc = ~*pc;
2100                 if (! *pc)
2101                         *pc = 0xffff;
2102         }
2103 }
2104
2105 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2106 void
2107 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2108 {
2109         u_int32_t       ao;
2110
2111         memcpy(&ao, a, sizeof(ao));
2112         memcpy(a, &an, sizeof(u_int32_t));
2113         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2114             ao % 65536, an % 65536, u);
2115 }
2116
2117 void
2118 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2119 {
2120         u_int32_t       ao;
2121
2122         memcpy(&ao, a, sizeof(ao));
2123         memcpy(a, &an, sizeof(u_int32_t));
2124
2125         *c = pf_proto_cksum_fixup(m,
2126             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2127             ao % 65536, an % 65536, udp);
2128 }
2129
2130 #ifdef INET6
2131 static void
2132 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2133 {
2134         struct pf_addr  ao;
2135
2136         PF_ACPY(&ao, a, AF_INET6);
2137         PF_ACPY(a, an, AF_INET6);
2138
2139         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2140             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2141             pf_cksum_fixup(pf_cksum_fixup(*c,
2142             ao.addr16[0], an->addr16[0], u),
2143             ao.addr16[1], an->addr16[1], u),
2144             ao.addr16[2], an->addr16[2], u),
2145             ao.addr16[3], an->addr16[3], u),
2146             ao.addr16[4], an->addr16[4], u),
2147             ao.addr16[5], an->addr16[5], u),
2148             ao.addr16[6], an->addr16[6], u),
2149             ao.addr16[7], an->addr16[7], u);
2150 }
2151 #endif /* INET6 */
2152
2153 static void
2154 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2155     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2156     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2157 {
2158         struct pf_addr  oia, ooa;
2159
2160         PF_ACPY(&oia, ia, af);
2161         if (oa)
2162                 PF_ACPY(&ooa, oa, af);
2163
2164         /* Change inner protocol port, fix inner protocol checksum. */
2165         if (ip != NULL) {
2166                 u_int16_t       oip = *ip;
2167                 u_int32_t       opc;
2168
2169                 if (pc != NULL)
2170                         opc = *pc;
2171                 *ip = np;
2172                 if (pc != NULL)
2173                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2174                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2175                 if (pc != NULL)
2176                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2177         }
2178         /* Change inner ip address, fix inner ip and icmp checksums. */
2179         PF_ACPY(ia, na, af);
2180         switch (af) {
2181 #ifdef INET
2182         case AF_INET: {
2183                 u_int32_t        oh2c = *h2c;
2184
2185                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2186                     oia.addr16[0], ia->addr16[0], 0),
2187                     oia.addr16[1], ia->addr16[1], 0);
2188                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2189                     oia.addr16[0], ia->addr16[0], 0),
2190                     oia.addr16[1], ia->addr16[1], 0);
2191                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2192                 break;
2193         }
2194 #endif /* INET */
2195 #ifdef INET6
2196         case AF_INET6:
2197                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2198                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2199                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2200                     oia.addr16[0], ia->addr16[0], u),
2201                     oia.addr16[1], ia->addr16[1], u),
2202                     oia.addr16[2], ia->addr16[2], u),
2203                     oia.addr16[3], ia->addr16[3], u),
2204                     oia.addr16[4], ia->addr16[4], u),
2205                     oia.addr16[5], ia->addr16[5], u),
2206                     oia.addr16[6], ia->addr16[6], u),
2207                     oia.addr16[7], ia->addr16[7], u);
2208                 break;
2209 #endif /* INET6 */
2210         }
2211         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2212         if (oa) {
2213                 PF_ACPY(oa, na, af);
2214                 switch (af) {
2215 #ifdef INET
2216                 case AF_INET:
2217                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2218                             ooa.addr16[0], oa->addr16[0], 0),
2219                             ooa.addr16[1], oa->addr16[1], 0);
2220                         break;
2221 #endif /* INET */
2222 #ifdef INET6
2223                 case AF_INET6:
2224                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2225                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2226                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2227                             ooa.addr16[0], oa->addr16[0], u),
2228                             ooa.addr16[1], oa->addr16[1], u),
2229                             ooa.addr16[2], oa->addr16[2], u),
2230                             ooa.addr16[3], oa->addr16[3], u),
2231                             ooa.addr16[4], oa->addr16[4], u),
2232                             ooa.addr16[5], oa->addr16[5], u),
2233                             ooa.addr16[6], oa->addr16[6], u),
2234                             ooa.addr16[7], oa->addr16[7], u);
2235                         break;
2236 #endif /* INET6 */
2237                 }
2238         }
2239 }
2240
2241
2242 /*
2243  * Need to modulate the sequence numbers in the TCP SACK option
2244  * (credits to Krzysztof Pfaff for report and patch)
2245  */
2246 static int
2247 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2248     struct tcphdr *th, struct pf_state_peer *dst)
2249 {
2250         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2251         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2252         int copyback = 0, i, olen;
2253         struct sackblk sack;
2254
2255 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2256         if (hlen < TCPOLEN_SACKLEN ||
2257             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2258                 return 0;
2259
2260         while (hlen >= TCPOLEN_SACKLEN) {
2261                 olen = opt[1];
2262                 switch (*opt) {
2263                 case TCPOPT_EOL:        /* FALLTHROUGH */
2264                 case TCPOPT_NOP:
2265                         opt++;
2266                         hlen--;
2267                         break;
2268                 case TCPOPT_SACK:
2269                         if (olen > hlen)
2270                                 olen = hlen;
2271                         if (olen >= TCPOLEN_SACKLEN) {
2272                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2273                                     i += TCPOLEN_SACK) {
2274                                         memcpy(&sack, &opt[i], sizeof(sack));
2275                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2276                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2277                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2278                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2279                                         memcpy(&opt[i], &sack, sizeof(sack));
2280                                 }
2281                                 copyback = 1;
2282                         }
2283                         /* FALLTHROUGH */
2284                 default:
2285                         if (olen < 2)
2286                                 olen = 2;
2287                         hlen -= olen;
2288                         opt += olen;
2289                 }
2290         }
2291
2292         if (copyback)
2293                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2294         return (copyback);
2295 }
2296
2297 static void
2298 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2299     const struct pf_addr *saddr, const struct pf_addr *daddr,
2300     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2301     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2302     u_int16_t rtag, struct ifnet *ifp)
2303 {
2304         struct pf_send_entry *pfse;
2305         struct mbuf     *m;
2306         int              len, tlen;
2307 #ifdef INET
2308         struct ip       *h = NULL;
2309 #endif /* INET */
2310 #ifdef INET6
2311         struct ip6_hdr  *h6 = NULL;
2312 #endif /* INET6 */
2313         struct tcphdr   *th;
2314         char            *opt;
2315         struct pf_mtag  *pf_mtag;
2316
2317         len = 0;
2318         th = NULL;
2319
2320         /* maximum segment size tcp option */
2321         tlen = sizeof(struct tcphdr);
2322         if (mss)
2323                 tlen += 4;
2324
2325         switch (af) {
2326 #ifdef INET
2327         case AF_INET:
2328                 len = sizeof(struct ip) + tlen;
2329                 break;
2330 #endif /* INET */
2331 #ifdef INET6
2332         case AF_INET6:
2333                 len = sizeof(struct ip6_hdr) + tlen;
2334                 break;
2335 #endif /* INET6 */
2336         default:
2337                 panic("%s: unsupported af %d", __func__, af);
2338         }
2339
2340         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2341         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2342         if (pfse == NULL)
2343                 return;
2344         m = m_gethdr(M_NOWAIT, MT_DATA);
2345         if (m == NULL) {
2346                 free(pfse, M_PFTEMP);
2347                 return;
2348         }
2349 #ifdef MAC
2350         mac_netinet_firewall_send(m);
2351 #endif
2352         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2353                 free(pfse, M_PFTEMP);
2354                 m_freem(m);
2355                 return;
2356         }
2357         if (tag)
2358                 m->m_flags |= M_SKIP_FIREWALL;
2359         pf_mtag->tag = rtag;
2360
2361         if (r != NULL && r->rtableid >= 0)
2362                 M_SETFIB(m, r->rtableid);
2363
2364 #ifdef ALTQ
2365         if (r != NULL && r->qid) {
2366                 pf_mtag->qid = r->qid;
2367
2368                 /* add hints for ecn */
2369                 pf_mtag->hdr = mtod(m, struct ip *);
2370         }
2371 #endif /* ALTQ */
2372         m->m_data += max_linkhdr;
2373         m->m_pkthdr.len = m->m_len = len;
2374         m->m_pkthdr.rcvif = NULL;
2375         bzero(m->m_data, len);
2376         switch (af) {
2377 #ifdef INET
2378         case AF_INET:
2379                 h = mtod(m, struct ip *);
2380
2381                 /* IP header fields included in the TCP checksum */
2382                 h->ip_p = IPPROTO_TCP;
2383                 h->ip_len = htons(tlen);
2384                 h->ip_src.s_addr = saddr->v4.s_addr;
2385                 h->ip_dst.s_addr = daddr->v4.s_addr;
2386
2387                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2388                 break;
2389 #endif /* INET */
2390 #ifdef INET6
2391         case AF_INET6:
2392                 h6 = mtod(m, struct ip6_hdr *);
2393
2394                 /* IP header fields included in the TCP checksum */
2395                 h6->ip6_nxt = IPPROTO_TCP;
2396                 h6->ip6_plen = htons(tlen);
2397                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2398                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2399
2400                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2401                 break;
2402 #endif /* INET6 */
2403         }
2404
2405         /* TCP header */
2406         th->th_sport = sport;
2407         th->th_dport = dport;
2408         th->th_seq = htonl(seq);
2409         th->th_ack = htonl(ack);
2410         th->th_off = tlen >> 2;
2411         th->th_flags = flags;
2412         th->th_win = htons(win);
2413
2414         if (mss) {
2415                 opt = (char *)(th + 1);
2416                 opt[0] = TCPOPT_MAXSEG;
2417                 opt[1] = 4;
2418                 HTONS(mss);
2419                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2420         }
2421
2422         switch (af) {
2423 #ifdef INET
2424         case AF_INET:
2425                 /* TCP checksum */
2426                 th->th_sum = in_cksum(m, len);
2427
2428                 /* Finish the IP header */
2429                 h->ip_v = 4;
2430                 h->ip_hl = sizeof(*h) >> 2;
2431                 h->ip_tos = IPTOS_LOWDELAY;
2432                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2433                 h->ip_len = htons(len);
2434                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2435                 h->ip_sum = 0;
2436
2437                 pfse->pfse_type = PFSE_IP;
2438                 break;
2439 #endif /* INET */
2440 #ifdef INET6
2441         case AF_INET6:
2442                 /* TCP checksum */
2443                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2444                     sizeof(struct ip6_hdr), tlen);
2445
2446                 h6->ip6_vfc |= IPV6_VERSION;
2447                 h6->ip6_hlim = IPV6_DEFHLIM;
2448
2449                 pfse->pfse_type = PFSE_IP6;
2450                 break;
2451 #endif /* INET6 */
2452         }
2453         pfse->pfse_m = m;
2454         pf_send(pfse);
2455 }
2456
2457 static void
2458 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2459     struct pf_rule *r)
2460 {
2461         struct pf_send_entry *pfse;
2462         struct mbuf *m0;
2463         struct pf_mtag *pf_mtag;
2464
2465         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2466         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2467         if (pfse == NULL)
2468                 return;
2469
2470         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2471                 free(pfse, M_PFTEMP);
2472                 return;
2473         }
2474
2475         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2476                 free(pfse, M_PFTEMP);
2477                 return;
2478         }
2479         /* XXX: revisit */
2480         m0->m_flags |= M_SKIP_FIREWALL;
2481
2482         if (r->rtableid >= 0)
2483                 M_SETFIB(m0, r->rtableid);
2484
2485 #ifdef ALTQ
2486         if (r->qid) {
2487                 pf_mtag->qid = r->qid;
2488                 /* add hints for ecn */
2489                 pf_mtag->hdr = mtod(m0, struct ip *);
2490         }
2491 #endif /* ALTQ */
2492
2493         switch (af) {
2494 #ifdef INET
2495         case AF_INET:
2496                 pfse->pfse_type = PFSE_ICMP;
2497                 break;
2498 #endif /* INET */
2499 #ifdef INET6
2500         case AF_INET6:
2501                 pfse->pfse_type = PFSE_ICMP6;
2502                 break;
2503 #endif /* INET6 */
2504         }
2505         pfse->pfse_m = m0;
2506         pfse->pfse_icmp_type = type;
2507         pfse->pfse_icmp_code = code;
2508         pf_send(pfse);
2509 }
2510
2511 /*
2512  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2513  * If n is 0, they match if they are equal. If n is != 0, they match if they
2514  * are different.
2515  */
2516 int
2517 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2518     struct pf_addr *b, sa_family_t af)
2519 {
2520         int     match = 0;
2521
2522         switch (af) {
2523 #ifdef INET
2524         case AF_INET:
2525                 if ((a->addr32[0] & m->addr32[0]) ==
2526                     (b->addr32[0] & m->addr32[0]))
2527                         match++;
2528                 break;
2529 #endif /* INET */
2530 #ifdef INET6
2531         case AF_INET6:
2532                 if (((a->addr32[0] & m->addr32[0]) ==
2533                      (b->addr32[0] & m->addr32[0])) &&
2534                     ((a->addr32[1] & m->addr32[1]) ==
2535                      (b->addr32[1] & m->addr32[1])) &&
2536                     ((a->addr32[2] & m->addr32[2]) ==
2537                      (b->addr32[2] & m->addr32[2])) &&
2538                     ((a->addr32[3] & m->addr32[3]) ==
2539                      (b->addr32[3] & m->addr32[3])))
2540                         match++;
2541                 break;
2542 #endif /* INET6 */
2543         }
2544         if (match) {
2545                 if (n)
2546                         return (0);
2547                 else
2548                         return (1);
2549         } else {
2550                 if (n)
2551                         return (1);
2552                 else
2553                         return (0);
2554         }
2555 }
2556
2557 /*
2558  * Return 1 if b <= a <= e, otherwise return 0.
2559  */
2560 int
2561 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2562     struct pf_addr *a, sa_family_t af)
2563 {
2564         switch (af) {
2565 #ifdef INET
2566         case AF_INET:
2567                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2568                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2569                         return (0);
2570                 break;
2571 #endif /* INET */
2572 #ifdef INET6
2573         case AF_INET6: {
2574                 int     i;
2575
2576                 /* check a >= b */
2577                 for (i = 0; i < 4; ++i)
2578                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2579                                 break;
2580                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2581                                 return (0);
2582                 /* check a <= e */
2583                 for (i = 0; i < 4; ++i)
2584                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2585                                 break;
2586                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2587                                 return (0);
2588                 break;
2589         }
2590 #endif /* INET6 */
2591         }
2592         return (1);
2593 }
2594
2595 static int
2596 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2597 {
2598         switch (op) {
2599         case PF_OP_IRG:
2600                 return ((p > a1) && (p < a2));
2601         case PF_OP_XRG:
2602                 return ((p < a1) || (p > a2));
2603         case PF_OP_RRG:
2604                 return ((p >= a1) && (p <= a2));
2605         case PF_OP_EQ:
2606                 return (p == a1);
2607         case PF_OP_NE:
2608                 return (p != a1);
2609         case PF_OP_LT:
2610                 return (p < a1);
2611         case PF_OP_LE:
2612                 return (p <= a1);
2613         case PF_OP_GT:
2614                 return (p > a1);
2615         case PF_OP_GE:
2616                 return (p >= a1);
2617         }
2618         return (0); /* never reached */
2619 }
2620
2621 int
2622 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2623 {
2624         NTOHS(a1);
2625         NTOHS(a2);
2626         NTOHS(p);
2627         return (pf_match(op, a1, a2, p));
2628 }
2629
2630 static int
2631 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2632 {
2633         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2634                 return (0);
2635         return (pf_match(op, a1, a2, u));
2636 }
2637
2638 static int
2639 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2640 {
2641         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2642                 return (0);
2643         return (pf_match(op, a1, a2, g));
2644 }
2645
2646 int
2647 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2648 {
2649         if (*tag == -1)
2650                 *tag = mtag;
2651
2652         return ((!r->match_tag_not && r->match_tag == *tag) ||
2653             (r->match_tag_not && r->match_tag != *tag));
2654 }
2655
2656 int
2657 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2658 {
2659
2660         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2661
2662         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2663                 return (ENOMEM);
2664
2665         pd->pf_mtag->tag = tag;
2666
2667         return (0);
2668 }
2669
2670 #define PF_ANCHOR_STACKSIZE     32
2671 struct pf_anchor_stackframe {
2672         struct pf_ruleset       *rs;
2673         struct pf_rule          *r;     /* XXX: + match bit */
2674         struct pf_anchor        *child;
2675 };
2676
2677 /*
2678  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2679  */
2680 #define PF_ANCHORSTACK_MATCH    0x00000001
2681 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2682
2683 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2684 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2685                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2686 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2687                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2688 } while (0)
2689
2690 void
2691 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2692     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2693     int *match)
2694 {
2695         struct pf_anchor_stackframe     *f;
2696
2697         PF_RULES_RASSERT();
2698
2699         if (match)
2700                 *match = 0;
2701         if (*depth >= PF_ANCHOR_STACKSIZE) {
2702                 printf("%s: anchor stack overflow on %s\n",
2703                     __func__, (*r)->anchor->name);
2704                 *r = TAILQ_NEXT(*r, entries);
2705                 return;
2706         } else if (*depth == 0 && a != NULL)
2707                 *a = *r;
2708         f = stack + (*depth)++;
2709         f->rs = *rs;
2710         f->r = *r;
2711         if ((*r)->anchor_wildcard) {
2712                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2713
2714                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2715                         *r = NULL;
2716                         return;
2717                 }
2718                 *rs = &f->child->ruleset;
2719         } else {
2720                 f->child = NULL;
2721                 *rs = &(*r)->anchor->ruleset;
2722         }
2723         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2724 }
2725
2726 int
2727 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2728     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2729     int *match)
2730 {
2731         struct pf_anchor_stackframe     *f;
2732         struct pf_rule *fr;
2733         int quick = 0;
2734
2735         PF_RULES_RASSERT();
2736
2737         do {
2738                 if (*depth <= 0)
2739                         break;
2740                 f = stack + *depth - 1;
2741                 fr = PF_ANCHOR_RULE(f);
2742                 if (f->child != NULL) {
2743                         struct pf_anchor_node *parent;
2744
2745                         /*
2746                          * This block traverses through
2747                          * a wildcard anchor.
2748                          */
2749                         parent = &fr->anchor->children;
2750                         if (match != NULL && *match) {
2751                                 /*
2752                                  * If any of "*" matched, then
2753                                  * "foo/ *" matched, mark frame
2754                                  * appropriately.
2755                                  */
2756                                 PF_ANCHOR_SET_MATCH(f);
2757                                 *match = 0;
2758                         }
2759                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2760                         if (f->child != NULL) {
2761                                 *rs = &f->child->ruleset;
2762                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2763                                 if (*r == NULL)
2764                                         continue;
2765                                 else
2766                                         break;
2767                         }
2768                 }
2769                 (*depth)--;
2770                 if (*depth == 0 && a != NULL)
2771                         *a = NULL;
2772                 *rs = f->rs;
2773                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2774                         quick = fr->quick;
2775                 *r = TAILQ_NEXT(fr, entries);
2776         } while (*r == NULL);
2777
2778         return (quick);
2779 }
2780
2781 #ifdef INET6
2782 void
2783 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2784     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2785 {
2786         switch (af) {
2787 #ifdef INET
2788         case AF_INET:
2789                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2790                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2791                 break;
2792 #endif /* INET */
2793         case AF_INET6:
2794                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2795                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2796                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2797                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2798                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2799                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2800                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2801                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2802                 break;
2803         }
2804 }
2805
2806 void
2807 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2808 {
2809         switch (af) {
2810 #ifdef INET
2811         case AF_INET:
2812                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2813                 break;
2814 #endif /* INET */
2815         case AF_INET6:
2816                 if (addr->addr32[3] == 0xffffffff) {
2817                         addr->addr32[3] = 0;
2818                         if (addr->addr32[2] == 0xffffffff) {
2819                                 addr->addr32[2] = 0;
2820                                 if (addr->addr32[1] == 0xffffffff) {
2821                                         addr->addr32[1] = 0;
2822                                         addr->addr32[0] =
2823                                             htonl(ntohl(addr->addr32[0]) + 1);
2824                                 } else
2825                                         addr->addr32[1] =
2826                                             htonl(ntohl(addr->addr32[1]) + 1);
2827                         } else
2828                                 addr->addr32[2] =
2829                                     htonl(ntohl(addr->addr32[2]) + 1);
2830                 } else
2831                         addr->addr32[3] =
2832                             htonl(ntohl(addr->addr32[3]) + 1);
2833                 break;
2834         }
2835 }
2836 #endif /* INET6 */
2837
2838 int
2839 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2840 {
2841         struct pf_addr          *saddr, *daddr;
2842         u_int16_t                sport, dport;
2843         struct inpcbinfo        *pi;
2844         struct inpcb            *inp;
2845
2846         pd->lookup.uid = UID_MAX;
2847         pd->lookup.gid = GID_MAX;
2848
2849         switch (pd->proto) {
2850         case IPPROTO_TCP:
2851                 if (pd->hdr.tcp == NULL)
2852                         return (-1);
2853                 sport = pd->hdr.tcp->th_sport;
2854                 dport = pd->hdr.tcp->th_dport;
2855                 pi = &V_tcbinfo;
2856                 break;
2857         case IPPROTO_UDP:
2858                 if (pd->hdr.udp == NULL)
2859                         return (-1);
2860                 sport = pd->hdr.udp->uh_sport;
2861                 dport = pd->hdr.udp->uh_dport;
2862                 pi = &V_udbinfo;
2863                 break;
2864         default:
2865                 return (-1);
2866         }
2867         if (direction == PF_IN) {
2868                 saddr = pd->src;
2869                 daddr = pd->dst;
2870         } else {
2871                 u_int16_t       p;
2872
2873                 p = sport;
2874                 sport = dport;
2875                 dport = p;
2876                 saddr = pd->dst;
2877                 daddr = pd->src;
2878         }
2879         switch (pd->af) {
2880 #ifdef INET
2881         case AF_INET:
2882                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2883                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2884                 if (inp == NULL) {
2885                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2886                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2887                            INPLOOKUP_RLOCKPCB, NULL, m);
2888                         if (inp == NULL)
2889                                 return (-1);
2890                 }
2891                 break;
2892 #endif /* INET */
2893 #ifdef INET6
2894         case AF_INET6:
2895                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2896                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2897                 if (inp == NULL) {
2898                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2899                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2900                             INPLOOKUP_RLOCKPCB, NULL, m);
2901                         if (inp == NULL)
2902                                 return (-1);
2903                 }
2904                 break;
2905 #endif /* INET6 */
2906
2907         default:
2908                 return (-1);
2909         }
2910         INP_RLOCK_ASSERT(inp);
2911         pd->lookup.uid = inp->inp_cred->cr_uid;
2912         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2913         INP_RUNLOCK(inp);
2914
2915         return (1);
2916 }
2917
2918 static u_int8_t
2919 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2920 {
2921         int              hlen;
2922         u_int8_t         hdr[60];
2923         u_int8_t        *opt, optlen;
2924         u_int8_t         wscale = 0;
2925
2926         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2927         if (hlen <= sizeof(struct tcphdr))
2928                 return (0);
2929         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2930                 return (0);
2931         opt = hdr + sizeof(struct tcphdr);
2932         hlen -= sizeof(struct tcphdr);
2933         while (hlen >= 3) {
2934                 switch (*opt) {
2935                 case TCPOPT_EOL:
2936                 case TCPOPT_NOP:
2937                         ++opt;
2938                         --hlen;
2939                         break;
2940                 case TCPOPT_WINDOW:
2941                         wscale = opt[2];
2942                         if (wscale > TCP_MAX_WINSHIFT)
2943                                 wscale = TCP_MAX_WINSHIFT;
2944                         wscale |= PF_WSCALE_FLAG;
2945                         /* FALLTHROUGH */
2946                 default:
2947                         optlen = opt[1];
2948                         if (optlen < 2)
2949                                 optlen = 2;
2950                         hlen -= optlen;
2951                         opt += optlen;
2952                         break;
2953                 }
2954         }
2955         return (wscale);
2956 }
2957
2958 static u_int16_t
2959 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2960 {
2961         int              hlen;
2962         u_int8_t         hdr[60];
2963         u_int8_t        *opt, optlen;
2964         u_int16_t        mss = V_tcp_mssdflt;
2965
2966         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2967         if (hlen <= sizeof(struct tcphdr))
2968                 return (0);
2969         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2970                 return (0);
2971         opt = hdr + sizeof(struct tcphdr);
2972         hlen -= sizeof(struct tcphdr);
2973         while (hlen >= TCPOLEN_MAXSEG) {
2974                 switch (*opt) {
2975                 case TCPOPT_EOL:
2976                 case TCPOPT_NOP:
2977                         ++opt;
2978                         --hlen;
2979                         break;
2980                 case TCPOPT_MAXSEG:
2981                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2982                         NTOHS(mss);
2983                         /* FALLTHROUGH */
2984                 default:
2985                         optlen = opt[1];
2986                         if (optlen < 2)
2987                                 optlen = 2;
2988                         hlen -= optlen;
2989                         opt += optlen;
2990                         break;
2991                 }
2992         }
2993         return (mss);
2994 }
2995
2996 static u_int16_t
2997 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2998 {
2999 #ifdef INET
3000         struct sockaddr_in      *dst;
3001         struct route             ro;
3002 #endif /* INET */
3003 #ifdef INET6
3004         struct sockaddr_in6     *dst6;
3005         struct route_in6         ro6;
3006 #endif /* INET6 */
3007         struct rtentry          *rt = NULL;
3008         int                      hlen = 0;
3009         u_int16_t                mss = V_tcp_mssdflt;
3010
3011         switch (af) {
3012 #ifdef INET
3013         case AF_INET:
3014                 hlen = sizeof(struct ip);
3015                 bzero(&ro, sizeof(ro));
3016                 dst = (struct sockaddr_in *)&ro.ro_dst;
3017                 dst->sin_family = AF_INET;
3018                 dst->sin_len = sizeof(*dst);
3019                 dst->sin_addr = addr->v4;
3020                 in_rtalloc_ign(&ro, 0, rtableid);
3021                 rt = ro.ro_rt;
3022                 break;
3023 #endif /* INET */
3024 #ifdef INET6
3025         case AF_INET6:
3026                 hlen = sizeof(struct ip6_hdr);
3027                 bzero(&ro6, sizeof(ro6));
3028                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
3029                 dst6->sin6_family = AF_INET6;
3030                 dst6->sin6_len = sizeof(*dst6);
3031                 dst6->sin6_addr = addr->v6;
3032                 in6_rtalloc_ign(&ro6, 0, rtableid);
3033                 rt = ro6.ro_rt;
3034                 break;
3035 #endif /* INET6 */
3036         }
3037
3038         if (rt && rt->rt_ifp) {
3039                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3040                 mss = max(V_tcp_mssdflt, mss);
3041                 RTFREE(rt);
3042         }
3043         mss = min(mss, offer);
3044         mss = max(mss, 64);             /* sanity - at least max opt space */
3045         return (mss);
3046 }
3047
3048 static u_int32_t
3049 pf_tcp_iss(struct pf_pdesc *pd)
3050 {
3051         MD5_CTX ctx;
3052         u_int32_t digest[4];
3053
3054         if (V_pf_tcp_secret_init == 0) {
3055                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3056                 MD5Init(&V_pf_tcp_secret_ctx);
3057                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3058                     sizeof(V_pf_tcp_secret));
3059                 V_pf_tcp_secret_init = 1;
3060         }
3061
3062         ctx = V_pf_tcp_secret_ctx;
3063
3064         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3065         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3066         if (pd->af == AF_INET6) {
3067                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3068                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3069         } else {
3070                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3071                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3072         }
3073         MD5Final((u_char *)digest, &ctx);
3074         V_pf_tcp_iss_off += 4096;
3075 #define ISN_RANDOM_INCREMENT (4096 - 1)
3076         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3077             V_pf_tcp_iss_off);
3078 #undef  ISN_RANDOM_INCREMENT
3079 }
3080
3081 static int
3082 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3083     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3084     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3085 {
3086         struct pf_rule          *nr = NULL;
3087         struct pf_addr          * const saddr = pd->src;
3088         struct pf_addr          * const daddr = pd->dst;
3089         sa_family_t              af = pd->af;
3090         struct pf_rule          *r, *a = NULL;
3091         struct pf_ruleset       *ruleset = NULL;
3092         struct pf_src_node      *nsn = NULL;
3093         struct tcphdr           *th = pd->hdr.tcp;
3094         struct pf_state_key     *sk = NULL, *nk = NULL;
3095         u_short                  reason;
3096         int                      rewrite = 0, hdrlen = 0;
3097         int                      tag = -1, rtableid = -1;
3098         int                      asd = 0;
3099         int                      match = 0;
3100         int                      state_icmp = 0;
3101         u_int16_t                sport = 0, dport = 0;
3102         u_int16_t                bproto_sum = 0, bip_sum = 0;
3103         u_int8_t                 icmptype = 0, icmpcode = 0;
3104         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3105
3106         PF_RULES_RASSERT();
3107
3108         if (inp != NULL) {
3109                 INP_LOCK_ASSERT(inp);
3110                 pd->lookup.uid = inp->inp_cred->cr_uid;
3111                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3112                 pd->lookup.done = 1;
3113         }
3114
3115         switch (pd->proto) {
3116         case IPPROTO_TCP:
3117                 sport = th->th_sport;
3118                 dport = th->th_dport;
3119                 hdrlen = sizeof(*th);
3120                 break;
3121         case IPPROTO_UDP:
3122                 sport = pd->hdr.udp->uh_sport;
3123                 dport = pd->hdr.udp->uh_dport;
3124                 hdrlen = sizeof(*pd->hdr.udp);
3125                 break;
3126 #ifdef INET
3127         case IPPROTO_ICMP:
3128                 if (pd->af != AF_INET)
3129                         break;
3130                 sport = dport = pd->hdr.icmp->icmp_id;
3131                 hdrlen = sizeof(*pd->hdr.icmp);
3132                 icmptype = pd->hdr.icmp->icmp_type;
3133                 icmpcode = pd->hdr.icmp->icmp_code;
3134
3135                 if (icmptype == ICMP_UNREACH ||
3136                     icmptype == ICMP_SOURCEQUENCH ||
3137                     icmptype == ICMP_REDIRECT ||
3138                     icmptype == ICMP_TIMXCEED ||
3139                     icmptype == ICMP_PARAMPROB)
3140                         state_icmp++;
3141                 break;
3142 #endif /* INET */
3143 #ifdef INET6
3144         case IPPROTO_ICMPV6:
3145                 if (af != AF_INET6)
3146                         break;
3147                 sport = dport = pd->hdr.icmp6->icmp6_id;
3148                 hdrlen = sizeof(*pd->hdr.icmp6);
3149                 icmptype = pd->hdr.icmp6->icmp6_type;
3150                 icmpcode = pd->hdr.icmp6->icmp6_code;
3151
3152                 if (icmptype == ICMP6_DST_UNREACH ||
3153                     icmptype == ICMP6_PACKET_TOO_BIG ||
3154                     icmptype == ICMP6_TIME_EXCEEDED ||
3155                     icmptype == ICMP6_PARAM_PROB)
3156                         state_icmp++;
3157                 break;
3158 #endif /* INET6 */
3159         default:
3160                 sport = dport = hdrlen = 0;
3161                 break;
3162         }
3163
3164         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3165
3166         /* check packet for BINAT/NAT/RDR */
3167         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3168             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3169                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3170                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3171
3172                 if (pd->ip_sum)
3173                         bip_sum = *pd->ip_sum;
3174
3175                 switch (pd->proto) {
3176                 case IPPROTO_TCP:
3177                         bproto_sum = th->th_sum;
3178                         pd->proto_sum = &th->th_sum;
3179
3180                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3181                             nk->port[pd->sidx] != sport) {
3182                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3183                                     &th->th_sum, &nk->addr[pd->sidx],
3184                                     nk->port[pd->sidx], 0, af);
3185                                 pd->sport = &th->th_sport;
3186                                 sport = th->th_sport;
3187                         }
3188
3189                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3190                             nk->port[pd->didx] != dport) {
3191                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3192                                     &th->th_sum, &nk->addr[pd->didx],
3193                                     nk->port[pd->didx], 0, af);
3194                                 dport = th->th_dport;
3195                                 pd->dport = &th->th_dport;
3196                         }
3197                         rewrite++;
3198                         break;
3199                 case IPPROTO_UDP:
3200                         bproto_sum = pd->hdr.udp->uh_sum;
3201                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3202
3203                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3204                             nk->port[pd->sidx] != sport) {
3205                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3206                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3207                                     &nk->addr[pd->sidx],
3208                                     nk->port[pd->sidx], 1, af);
3209                                 sport = pd->hdr.udp->uh_sport;
3210                                 pd->sport = &pd->hdr.udp->uh_sport;
3211                         }
3212
3213                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3214                             nk->port[pd->didx] != dport) {
3215                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3216                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3217                                     &nk->addr[pd->didx],
3218                                     nk->port[pd->didx], 1, af);
3219                                 dport = pd->hdr.udp->uh_dport;
3220                                 pd->dport = &pd->hdr.udp->uh_dport;
3221                         }
3222                         rewrite++;
3223                         break;
3224 #ifdef INET
3225                 case IPPROTO_ICMP:
3226                         nk->port[0] = nk->port[1];
3227                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3228                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3229                                     nk->addr[pd->sidx].v4.s_addr, 0);
3230
3231                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3232                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3233                                     nk->addr[pd->didx].v4.s_addr, 0);
3234
3235                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3236                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3237                                     pd->hdr.icmp->icmp_cksum, sport,
3238                                     nk->port[1], 0);
3239                                 pd->hdr.icmp->icmp_id = nk->port[1];
3240                                 pd->sport = &pd->hdr.icmp->icmp_id;
3241                         }
3242                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3243                         break;
3244 #endif /* INET */
3245 #ifdef INET6
3246                 case IPPROTO_ICMPV6:
3247                         nk->port[0] = nk->port[1];
3248                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3249                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3250                                     &nk->addr[pd->sidx], 0);
3251
3252                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3253                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3254                                     &nk->addr[pd->didx], 0);
3255                         rewrite++;
3256                         break;
3257 #endif /* INET */
3258                 default:
3259                         switch (af) {
3260 #ifdef INET
3261                         case AF_INET:
3262                                 if (PF_ANEQ(saddr,
3263                                     &nk->addr[pd->sidx], AF_INET))
3264                                         pf_change_a(&saddr->v4.s_addr,
3265                                             pd->ip_sum,
3266                                             nk->addr[pd->sidx].v4.s_addr, 0);
3267
3268                                 if (PF_ANEQ(daddr,
3269                                     &nk->addr[pd->didx], AF_INET))
3270                                         pf_change_a(&daddr->v4.s_addr,
3271                                             pd->ip_sum,
3272                                             nk->addr[pd->didx].v4.s_addr, 0);
3273                                 break;
3274 #endif /* INET */
3275 #ifdef INET6
3276                         case AF_INET6:
3277                                 if (PF_ANEQ(saddr,
3278                                     &nk->addr[pd->sidx], AF_INET6))
3279                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3280
3281                                 if (PF_ANEQ(daddr,
3282                                     &nk->addr[pd->didx], AF_INET6))
3283                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3284                                 break;
3285 #endif /* INET */
3286                         }
3287                         break;
3288                 }
3289                 if (nr->natpass)
3290                         r = NULL;
3291                 pd->nat_rule = nr;
3292         }
3293
3294         while (r != NULL) {
3295                 r->evaluations++;
3296                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3297                         r = r->skip[PF_SKIP_IFP].ptr;
3298                 else if (r->direction && r->direction != direction)
3299                         r = r->skip[PF_SKIP_DIR].ptr;
3300                 else if (r->af && r->af != af)
3301                         r = r->skip[PF_SKIP_AF].ptr;
3302                 else if (r->proto && r->proto != pd->proto)
3303                         r = r->skip[PF_SKIP_PROTO].ptr;
3304                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3305                     r->src.neg, kif, M_GETFIB(m)))
3306                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3307                 /* tcp/udp only. port_op always 0 in other cases */
3308                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3309                     r->src.port[0], r->src.port[1], sport))
3310                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3311                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3312                     r->dst.neg, NULL, M_GETFIB(m)))
3313                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3314                 /* tcp/udp only. port_op always 0 in other cases */
3315                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3316                     r->dst.port[0], r->dst.port[1], dport))
3317                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3318                 /* icmp only. type always 0 in other cases */
3319                 else if (r->type && r->type != icmptype + 1)
3320                         r = TAILQ_NEXT(r, entries);
3321                 /* icmp only. type always 0 in other cases */
3322                 else if (r->code && r->code != icmpcode + 1)
3323                         r = TAILQ_NEXT(r, entries);
3324                 else if (r->tos && !(r->tos == pd->tos))
3325                         r = TAILQ_NEXT(r, entries);
3326                 else if (r->rule_flag & PFRULE_FRAGMENT)
3327                         r = TAILQ_NEXT(r, entries);
3328                 else if (pd->proto == IPPROTO_TCP &&
3329                     (r->flagset & th->th_flags) != r->flags)
3330                         r = TAILQ_NEXT(r, entries);
3331                 /* tcp/udp only. uid.op always 0 in other cases */
3332                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3333                     pf_socket_lookup(direction, pd, m), 1)) &&
3334                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3335                     pd->lookup.uid))
3336                         r = TAILQ_NEXT(r, entries);
3337                 /* tcp/udp only. gid.op always 0 in other cases */
3338                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3339                     pf_socket_lookup(direction, pd, m), 1)) &&
3340                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3341                     pd->lookup.gid))
3342                         r = TAILQ_NEXT(r, entries);
3343                 else if (r->prob &&
3344                     r->prob <= arc4random())
3345                         r = TAILQ_NEXT(r, entries);
3346                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3347                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3348                         r = TAILQ_NEXT(r, entries);
3349                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3350                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3351                     pf_osfp_fingerprint(pd, m, off, th),
3352                     r->os_fingerprint)))
3353                         r = TAILQ_NEXT(r, entries);
3354                 else {
3355                         if (r->tag)
3356                                 tag = r->tag;
3357                         if (r->rtableid >= 0)
3358                                 rtableid = r->rtableid;
3359                         if (r->anchor == NULL) {
3360                                 match = 1;
3361                                 *rm = r;
3362                                 *am = a;
3363                                 *rsm = ruleset;
3364                                 if ((*rm)->quick)
3365                                         break;
3366                                 r = TAILQ_NEXT(r, entries);
3367                         } else
3368                                 pf_step_into_anchor(anchor_stack, &asd,
3369                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3370                                     &match);
3371                 }
3372                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3373                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3374                         break;
3375         }
3376         r = *rm;
3377         a = *am;
3378         ruleset = *rsm;
3379
3380         REASON_SET(&reason, PFRES_MATCH);
3381
3382         if (r->log || (nr != NULL && nr->log)) {
3383                 if (rewrite)
3384                         m_copyback(m, off, hdrlen, pd->hdr.any);
3385                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3386                     ruleset, pd, 1);
3387         }
3388
3389         if ((r->action == PF_DROP) &&
3390             ((r->rule_flag & PFRULE_RETURNRST) ||
3391             (r->rule_flag & PFRULE_RETURNICMP) ||
3392             (r->rule_flag & PFRULE_RETURN))) {
3393                 /* undo NAT changes, if they have taken place */
3394                 if (nr != NULL) {
3395                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3396                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3397                         if (pd->sport)
3398                                 *pd->sport = sk->port[pd->sidx];
3399                         if (pd->dport)
3400                                 *pd->dport = sk->port[pd->didx];
3401                         if (pd->proto_sum)
3402                                 *pd->proto_sum = bproto_sum;
3403                         if (pd->ip_sum)
3404                                 *pd->ip_sum = bip_sum;
3405                         m_copyback(m, off, hdrlen, pd->hdr.any);
3406                 }
3407                 if (pd->proto == IPPROTO_TCP &&
3408                     ((r->rule_flag & PFRULE_RETURNRST) ||
3409                     (r->rule_flag & PFRULE_RETURN)) &&
3410                     !(th->th_flags & TH_RST)) {
3411                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3412                         int              len = 0;
3413 #ifdef INET
3414                         struct ip       *h4;
3415 #endif
3416 #ifdef INET6
3417                         struct ip6_hdr  *h6;
3418 #endif
3419
3420                         switch (af) {
3421 #ifdef INET
3422                         case AF_INET:
3423                                 h4 = mtod(m, struct ip *);
3424                                 len = ntohs(h4->ip_len) - off;
3425                                 break;
3426 #endif
3427 #ifdef INET6
3428                         case AF_INET6:
3429                                 h6 = mtod(m, struct ip6_hdr *);
3430                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3431                                 break;
3432 #endif
3433                         }
3434
3435                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3436                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3437                         else {
3438                                 if (th->th_flags & TH_SYN)
3439                                         ack++;
3440                                 if (th->th_flags & TH_FIN)
3441                                         ack++;
3442                                 pf_send_tcp(m, r, af, pd->dst,
3443                                     pd->src, th->th_dport, th->th_sport,
3444                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3445                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3446                         }
3447                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3448                     r->return_icmp)
3449                         pf_send_icmp(m, r->return_icmp >> 8,
3450                             r->return_icmp & 255, af, r);
3451                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3452                     r->return_icmp6)
3453                         pf_send_icmp(m, r->return_icmp6 >> 8,
3454                             r->return_icmp6 & 255, af, r);
3455         }
3456
3457         if (r->action == PF_DROP)
3458                 goto cleanup;
3459
3460         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3461                 REASON_SET(&reason, PFRES_MEMORY);
3462                 goto cleanup;
3463         }
3464         if (rtableid >= 0)
3465                 M_SETFIB(m, rtableid);
3466
3467         if (!state_icmp && (r->keep_state || nr != NULL ||
3468             (pd->flags & PFDESC_TCP_NORM))) {
3469                 int action;
3470                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3471                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3472                     hdrlen);
3473                 if (action != PF_PASS)
3474                         return (action);
3475         } else {
3476                 if (sk != NULL)
3477                         uma_zfree(V_pf_state_key_z, sk);
3478                 if (nk != NULL)
3479                         uma_zfree(V_pf_state_key_z, nk);
3480         }
3481
3482         /* copy back packet headers if we performed NAT operations */
3483         if (rewrite)
3484                 m_copyback(m, off, hdrlen, pd->hdr.any);
3485
3486         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3487             direction == PF_OUT &&
3488             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3489                 /*
3490                  * We want the state created, but we dont
3491                  * want to send this in case a partner
3492                  * firewall has to know about it to allow
3493                  * replies through it.
3494                  */
3495                 return (PF_DEFER);
3496
3497         return (PF_PASS);
3498
3499 cleanup:
3500         if (sk != NULL)
3501                 uma_zfree(V_pf_state_key_z, sk);
3502         if (nk != NULL)
3503                 uma_zfree(V_pf_state_key_z, nk);
3504         return (PF_DROP);
3505 }
3506
3507 static int
3508 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3509     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3510     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3511     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3512     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3513 {
3514         struct pf_state         *s = NULL;
3515         struct pf_src_node      *sn = NULL;
3516         struct tcphdr           *th = pd->hdr.tcp;
3517         u_int16_t                mss = V_tcp_mssdflt;
3518         u_short                  reason;
3519
3520         /* check maximums */
3521         if (r->max_states &&
3522             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3523                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3524                 REASON_SET(&reason, PFRES_MAXSTATES);
3525                 return (PF_DROP);
3526         }
3527         /* src node for filter rule */
3528         if ((r->rule_flag & PFRULE_SRCTRACK ||
3529             r->rpool.opts & PF_POOL_STICKYADDR) &&
3530             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3531                 REASON_SET(&reason, PFRES_SRCLIMIT);
3532                 goto csfailed;
3533         }
3534         /* src node for translation rule */
3535         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3536             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3537                 REASON_SET(&reason, PFRES_SRCLIMIT);
3538                 goto csfailed;
3539         }
3540         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3541         if (s == NULL) {
3542                 REASON_SET(&reason, PFRES_MEMORY);
3543                 goto csfailed;
3544         }
3545         s->rule.ptr = r;
3546         s->nat_rule.ptr = nr;
3547         s->anchor.ptr = a;
3548         STATE_INC_COUNTERS(s);
3549         if (r->allow_opts)
3550                 s->state_flags |= PFSTATE_ALLOWOPTS;
3551         if (r->rule_flag & PFRULE_STATESLOPPY)
3552                 s->state_flags |= PFSTATE_SLOPPY;
3553         s->log = r->log & PF_LOG_ALL;
3554         s->sync_state = PFSYNC_S_NONE;
3555         if (nr != NULL)
3556                 s->log |= nr->log & PF_LOG_ALL;
3557         switch (pd->proto) {
3558         case IPPROTO_TCP:
3559                 s->src.seqlo = ntohl(th->th_seq);
3560                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3561                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3562                     r->keep_state == PF_STATE_MODULATE) {
3563                         /* Generate sequence number modulator */
3564                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3565                             0)
3566                                 s->src.seqdiff = 1;
3567                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3568                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3569                         *rewrite = 1;
3570                 } else
3571                         s->src.seqdiff = 0;
3572                 if (th->th_flags & TH_SYN) {
3573                         s->src.seqhi++;
3574                         s->src.wscale = pf_get_wscale(m, off,
3575                             th->th_off, pd->af);
3576                 }
3577                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3578                 if (s->src.wscale & PF_WSCALE_MASK) {
3579                         /* Remove scale factor from initial window */
3580                         int win = s->src.max_win;
3581                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3582                         s->src.max_win = (win - 1) >>
3583                             (s->src.wscale & PF_WSCALE_MASK);
3584                 }
3585                 if (th->th_flags & TH_FIN)
3586                         s->src.seqhi++;
3587                 s->dst.seqhi = 1;
3588                 s->dst.max_win = 1;
3589                 s->src.state = TCPS_SYN_SENT;
3590                 s->dst.state = TCPS_CLOSED;
3591                 s->timeout = PFTM_TCP_FIRST_PACKET;
3592                 break;
3593         case IPPROTO_UDP:
3594                 s->src.state = PFUDPS_SINGLE;
3595                 s->dst.state = PFUDPS_NO_TRAFFIC;
3596                 s->timeout = PFTM_UDP_FIRST_PACKET;
3597                 break;
3598         case IPPROTO_ICMP:
3599 #ifdef INET6
3600         case IPPROTO_ICMPV6:
3601 #endif
3602                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3603                 break;
3604         default:
3605                 s->src.state = PFOTHERS_SINGLE;
3606                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3607                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3608         }
3609
3610         if (r->rt && r->rt != PF_FASTROUTE) {
3611                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3612                         REASON_SET(&reason, PFRES_BADSTATE);
3613                         pf_src_tree_remove_state(s);
3614                         STATE_DEC_COUNTERS(s);
3615                         uma_zfree(V_pf_state_z, s);
3616                         goto csfailed;
3617                 }
3618                 s->rt_kif = r->rpool.cur->kif;
3619         }
3620
3621         s->creation = time_uptime;
3622         s->expire = time_uptime;
3623
3624         if (sn != NULL)
3625                 s->src_node = sn;
3626         if (nsn != NULL) {
3627                 /* XXX We only modify one side for now. */
3628                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3629                 s->nat_src_node = nsn;
3630         }
3631         if (pd->proto == IPPROTO_TCP) {
3632                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3633                     off, pd, th, &s->src, &s->dst)) {
3634                         REASON_SET(&reason, PFRES_MEMORY);
3635                         pf_src_tree_remove_state(s);
3636                         STATE_DEC_COUNTERS(s);
3637                         uma_zfree(V_pf_state_z, s);
3638                         return (PF_DROP);
3639                 }
3640                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3641                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3642                     &s->src, &s->dst, rewrite)) {
3643                         /* This really shouldn't happen!!! */
3644                         DPFPRINTF(PF_DEBUG_URGENT,
3645                             ("pf_normalize_tcp_stateful failed on first pkt"));
3646                         pf_normalize_tcp_cleanup(s);
3647                         pf_src_tree_remove_state(s);
3648                         STATE_DEC_COUNTERS(s);
3649                         uma_zfree(V_pf_state_z, s);
3650                         return (PF_DROP);
3651                 }
3652         }
3653         s->direction = pd->dir;
3654
3655         /*
3656          * sk/nk could already been setup by pf_get_translation().
3657          */
3658         if (nr == NULL) {
3659                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3660                     __func__, nr, sk, nk));
3661                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3662                 if (sk == NULL)
3663                         goto csfailed;
3664                 nk = sk;
3665         } else
3666                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3667                     __func__, nr, sk, nk));
3668
3669         /* Swap sk/nk for PF_OUT. */
3670         if (pf_state_insert(BOUND_IFACE(r, kif),
3671             (pd->dir == PF_IN) ? sk : nk,
3672             (pd->dir == PF_IN) ? nk : sk, s)) {
3673                 if (pd->proto == IPPROTO_TCP)
3674                         pf_normalize_tcp_cleanup(s);
3675                 REASON_SET(&reason, PFRES_STATEINS);
3676                 pf_src_tree_remove_state(s);
3677                 STATE_DEC_COUNTERS(s);
3678                 uma_zfree(V_pf_state_z, s);
3679                 return (PF_DROP);
3680         } else
3681                 *sm = s;
3682
3683         if (tag > 0)
3684                 s->tag = tag;
3685         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3686             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3687                 s->src.state = PF_TCPS_PROXY_SRC;
3688                 /* undo NAT changes, if they have taken place */
3689                 if (nr != NULL) {
3690                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3691                         if (pd->dir == PF_OUT)
3692                                 skt = s->key[PF_SK_STACK];
3693                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3694                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3695                         if (pd->sport)
3696                                 *pd->sport = skt->port[pd->sidx];
3697                         if (pd->dport)
3698                                 *pd->dport = skt->port[pd->didx];
3699                         if (pd->proto_sum)
3700                                 *pd->proto_sum = bproto_sum;
3701                         if (pd->ip_sum)
3702                                 *pd->ip_sum = bip_sum;
3703                         m_copyback(m, off, hdrlen, pd->hdr.any);
3704                 }
3705                 s->src.seqhi = htonl(arc4random());
3706                 /* Find mss option */
3707                 int rtid = M_GETFIB(m);
3708                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3709                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3710                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3711                 s->src.mss = mss;
3712                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3713                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3714                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3715                 REASON_SET(&reason, PFRES_SYNPROXY);
3716                 return (PF_SYNPROXY_DROP);
3717         }
3718
3719         return (PF_PASS);
3720
3721 csfailed:
3722         if (sk != NULL)
3723                 uma_zfree(V_pf_state_key_z, sk);
3724         if (nk != NULL)
3725                 uma_zfree(V_pf_state_key_z, nk);
3726
3727         if (sn != NULL) {
3728                 struct pf_srchash *sh;
3729
3730                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3731                 PF_HASHROW_LOCK(sh);
3732                 if (--sn->states == 0 && sn->expire == 0) {
3733                         pf_unlink_src_node(sn);
3734                         uma_zfree(V_pf_sources_z, sn);
3735                         counter_u64_add(
3736                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3737                 }
3738                 PF_HASHROW_UNLOCK(sh);
3739         }
3740
3741         if (nsn != sn && nsn != NULL) {
3742                 struct pf_srchash *sh;
3743
3744                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3745                 PF_HASHROW_LOCK(sh);
3746                 if (--nsn->states == 0 && nsn->expire == 0) {
3747                         pf_unlink_src_node(nsn);
3748                         uma_zfree(V_pf_sources_z, nsn);
3749                         counter_u64_add(
3750                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3751                 }
3752                 PF_HASHROW_UNLOCK(sh);
3753         }
3754
3755         return (PF_DROP);
3756 }
3757
3758 static int
3759 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3760     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3761     struct pf_ruleset **rsm)
3762 {
3763         struct pf_rule          *r, *a = NULL;
3764         struct pf_ruleset       *ruleset = NULL;
3765         sa_family_t              af = pd->af;
3766         u_short                  reason;
3767         int                      tag = -1;
3768         int                      asd = 0;
3769         int                      match = 0;
3770         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3771
3772         PF_RULES_RASSERT();
3773
3774         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3775         while (r != NULL) {
3776                 r->evaluations++;
3777                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3778                         r = r->skip[PF_SKIP_IFP].ptr;
3779                 else if (r->direction && r->direction != direction)
3780                         r = r->skip[PF_SKIP_DIR].ptr;
3781                 else if (r->af && r->af != af)
3782                         r = r->skip[PF_SKIP_AF].ptr;
3783                 else if (r->proto && r->proto != pd->proto)
3784                         r = r->skip[PF_SKIP_PROTO].ptr;
3785                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3786                     r->src.neg, kif, M_GETFIB(m)))
3787                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3788                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3789                     r->dst.neg, NULL, M_GETFIB(m)))
3790                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3791                 else if (r->tos && !(r->tos == pd->tos))
3792                         r = TAILQ_NEXT(r, entries);
3793                 else if (r->os_fingerprint != PF_OSFP_ANY)
3794                         r = TAILQ_NEXT(r, entries);
3795                 else if (pd->proto == IPPROTO_UDP &&
3796                     (r->src.port_op || r->dst.port_op))
3797                         r = TAILQ_NEXT(r, entries);
3798                 else if (pd->proto == IPPROTO_TCP &&
3799                     (r->src.port_op || r->dst.port_op || r->flagset))
3800                         r = TAILQ_NEXT(r, entries);
3801                 else if ((pd->proto == IPPROTO_ICMP ||
3802                     pd->proto == IPPROTO_ICMPV6) &&
3803                     (r->type || r->code))
3804                         r = TAILQ_NEXT(r, entries);
3805                 else if (r->prob && r->prob <=
3806                     (arc4random() % (UINT_MAX - 1) + 1))
3807                         r = TAILQ_NEXT(r, entries);
3808                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3809                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3810                         r = TAILQ_NEXT(r, entries);
3811                 else {
3812                         if (r->anchor == NULL) {
3813                                 match = 1;
3814                                 *rm = r;
3815                                 *am = a;
3816                                 *rsm = ruleset;
3817                                 if ((*rm)->quick)
3818                                         break;
3819                                 r = TAILQ_NEXT(r, entries);
3820                         } else
3821                                 pf_step_into_anchor(anchor_stack, &asd,
3822                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3823                                     &match);
3824                 }
3825                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3826                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3827                         break;
3828         }
3829         r = *rm;
3830         a = *am;
3831         ruleset = *rsm;
3832
3833         REASON_SET(&reason, PFRES_MATCH);
3834
3835         if (r->log)
3836                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3837                     1);
3838
3839         if (r->action != PF_PASS)
3840                 return (PF_DROP);
3841
3842         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3843                 REASON_SET(&reason, PFRES_MEMORY);
3844                 return (PF_DROP);
3845         }
3846
3847         return (PF_PASS);
3848 }
3849
3850 static int
3851 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3852         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3853         struct pf_pdesc *pd, u_short *reason, int *copyback)
3854 {
3855         struct tcphdr           *th = pd->hdr.tcp;
3856         u_int16_t                win = ntohs(th->th_win);
3857         u_int32_t                ack, end, seq, orig_seq;
3858         u_int8_t                 sws, dws;
3859         int                      ackskew;
3860
3861         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3862                 sws = src->wscale & PF_WSCALE_MASK;
3863                 dws = dst->wscale & PF_WSCALE_MASK;
3864         } else
3865                 sws = dws = 0;
3866
3867         /*
3868          * Sequence tracking algorithm from Guido van Rooij's paper:
3869          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3870          *      tcp_filtering.ps
3871          */
3872
3873         orig_seq = seq = ntohl(th->th_seq);
3874         if (src->seqlo == 0) {
3875                 /* First packet from this end. Set its state */
3876
3877                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3878                     src->scrub == NULL) {
3879                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3880                                 REASON_SET(reason, PFRES_MEMORY);
3881                                 return (PF_DROP);
3882                         }
3883                 }
3884
3885                 /* Deferred generation of sequence number modulator */
3886                 if (dst->seqdiff && !src->seqdiff) {
3887                         /* use random iss for the TCP server */
3888                         while ((src->seqdiff = arc4random() - seq) == 0)
3889                                 ;
3890                         ack = ntohl(th->th_ack) - dst->seqdiff;
3891                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3892                             src->seqdiff), 0);
3893                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3894                         *copyback = 1;
3895                 } else {
3896                         ack = ntohl(th->th_ack);
3897                 }
3898
3899                 end = seq + pd->p_len;
3900                 if (th->th_flags & TH_SYN) {
3901                         end++;
3902                         if (dst->wscale & PF_WSCALE_FLAG) {
3903                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3904                                     pd->af);
3905                                 if (src->wscale & PF_WSCALE_FLAG) {
3906                                         /* Remove scale factor from initial
3907                                          * window */
3908                                         sws = src->wscale & PF_WSCALE_MASK;
3909                                         win = ((u_int32_t)win + (1 << sws) - 1)
3910                                             >> sws;
3911                                         dws = dst->wscale & PF_WSCALE_MASK;
3912                                 } else {
3913                                         /* fixup other window */
3914                                         dst->max_win <<= dst->wscale &
3915                                             PF_WSCALE_MASK;
3916                                         /* in case of a retrans SYN|ACK */
3917                                         dst->wscale = 0;
3918                                 }
3919                         }
3920                 }
3921                 if (th->th_flags & TH_FIN)
3922                         end++;
3923
3924                 src->seqlo = seq;
3925                 if (src->state < TCPS_SYN_SENT)
3926                         src->state = TCPS_SYN_SENT;
3927
3928                 /*
3929                  * May need to slide the window (seqhi may have been set by
3930                  * the crappy stack check or if we picked up the connection
3931                  * after establishment)
3932                  */
3933                 if (src->seqhi == 1 ||
3934                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3935                         src->seqhi = end + MAX(1, dst->max_win << dws);
3936                 if (win > src->max_win)
3937                         src->max_win = win;
3938
3939         } else {
3940                 ack = ntohl(th->th_ack) - dst->seqdiff;
3941                 if (src->seqdiff) {
3942                         /* Modulate sequence numbers */
3943                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3944                             src->seqdiff), 0);
3945                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3946                         *copyback = 1;
3947                 }
3948                 end = seq + pd->p_len;
3949                 if (th->th_flags & TH_SYN)
3950                         end++;
3951                 if (th->th_flags & TH_FIN)
3952                         end++;
3953         }
3954
3955         if ((th->th_flags & TH_ACK) == 0) {
3956                 /* Let it pass through the ack skew check */
3957                 ack = dst->seqlo;
3958         } else if ((ack == 0 &&
3959             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3960             /* broken tcp stacks do not set ack */
3961             (dst->state < TCPS_SYN_SENT)) {
3962                 /*
3963                  * Many stacks (ours included) will set the ACK number in an
3964                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3965                  */
3966                 ack = dst->seqlo;
3967         }
3968
3969         if (seq == end) {
3970                 /* Ease sequencing restrictions on no data packets */
3971                 seq = src->seqlo;
3972                 end = seq;
3973         }
3974
3975         ackskew = dst->seqlo - ack;
3976
3977
3978         /*
3979          * Need to demodulate the sequence numbers in any TCP SACK options
3980          * (Selective ACK). We could optionally validate the SACK values
3981          * against the current ACK window, either forwards or backwards, but
3982          * I'm not confident that SACK has been implemented properly
3983          * everywhere. It wouldn't surprise me if several stacks accidently
3984          * SACK too far backwards of previously ACKed data. There really aren't
3985          * any security implications of bad SACKing unless the target stack
3986          * doesn't validate the option length correctly. Someone trying to
3987          * spoof into a TCP connection won't bother blindly sending SACK
3988          * options anyway.
3989          */
3990         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3991                 if (pf_modulate_sack(m, off, pd, th, dst))
3992                         *copyback = 1;
3993         }
3994
3995
3996 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3997         if (SEQ_GEQ(src->seqhi, end) &&
3998             /* Last octet inside other's window space */
3999             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4000             /* Retrans: not more than one window back */
4001             (ackskew >= -MAXACKWINDOW) &&
4002             /* Acking not more than one reassembled fragment backwards */
4003             (ackskew <= (MAXACKWINDOW << sws)) &&
4004             /* Acking not more than one window forward */
4005             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4006             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4007             (pd->flags & PFDESC_IP_REAS) == 0)) {
4008             /* Require an exact/+1 sequence match on resets when possible */
4009
4010                 if (dst->scrub || src->scrub) {
4011                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4012                             *state, src, dst, copyback))
4013                                 return (PF_DROP);
4014                 }
4015
4016                 /* update max window */
4017                 if (src->max_win < win)
4018                         src->max_win = win;
4019                 /* synchronize sequencing */
4020                 if (SEQ_GT(end, src->seqlo))
4021                         src->seqlo = end;
4022                 /* slide the window of what the other end can send */
4023                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4024                         dst->seqhi = ack + MAX((win << sws), 1);
4025
4026
4027                 /* update states */
4028                 if (th->th_flags & TH_SYN)
4029                         if (src->state < TCPS_SYN_SENT)
4030                                 src->state = TCPS_SYN_SENT;
4031                 if (th->th_flags & TH_FIN)
4032                         if (src->state < TCPS_CLOSING)
4033                                 src->state = TCPS_CLOSING;
4034                 if (th->th_flags & TH_ACK) {
4035                         if (dst->state == TCPS_SYN_SENT) {
4036                                 dst->state = TCPS_ESTABLISHED;
4037                                 if (src->state == TCPS_ESTABLISHED &&
4038                                     (*state)->src_node != NULL &&
4039                                     pf_src_connlimit(state)) {
4040                                         REASON_SET(reason, PFRES_SRCLIMIT);
4041                                         return (PF_DROP);
4042                                 }
4043                         } else if (dst->state == TCPS_CLOSING)
4044                                 dst->state = TCPS_FIN_WAIT_2;
4045                 }
4046                 if (th->th_flags & TH_RST)
4047                         src->state = dst->state = TCPS_TIME_WAIT;
4048
4049                 /* update expire time */
4050                 (*state)->expire = time_uptime;
4051                 if (src->state >= TCPS_FIN_WAIT_2 &&
4052                     dst->state >= TCPS_FIN_WAIT_2)
4053                         (*state)->timeout = PFTM_TCP_CLOSED;
4054                 else if (src->state >= TCPS_CLOSING &&
4055                     dst->state >= TCPS_CLOSING)
4056                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4057                 else if (src->state < TCPS_ESTABLISHED ||
4058                     dst->state < TCPS_ESTABLISHED)
4059                         (*state)->timeout = PFTM_TCP_OPENING;
4060                 else if (src->state >= TCPS_CLOSING ||
4061                     dst->state >= TCPS_CLOSING)
4062                         (*state)->timeout = PFTM_TCP_CLOSING;
4063                 else
4064                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4065
4066                 /* Fall through to PASS packet */
4067
4068         } else if ((dst->state < TCPS_SYN_SENT ||
4069                 dst->state >= TCPS_FIN_WAIT_2 ||
4070                 src->state >= TCPS_FIN_WAIT_2) &&
4071             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4072             /* Within a window forward of the originating packet */
4073             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4074             /* Within a window backward of the originating packet */
4075
4076                 /*
4077                  * This currently handles three situations:
4078                  *  1) Stupid stacks will shotgun SYNs before their peer
4079                  *     replies.
4080                  *  2) When PF catches an already established stream (the
4081                  *     firewall rebooted, the state table was flushed, routes
4082                  *     changed...)
4083                  *  3) Packets get funky immediately after the connection
4084                  *     closes (this should catch Solaris spurious ACK|FINs
4085                  *     that web servers like to spew after a close)
4086                  *
4087                  * This must be a little more careful than the above code
4088                  * since packet floods will also be caught here. We don't
4089                  * update the TTL here to mitigate the damage of a packet
4090                  * flood and so the same code can handle awkward establishment
4091                  * and a loosened connection close.
4092                  * In the establishment case, a correct peer response will
4093                  * validate the connection, go through the normal state code
4094                  * and keep updating the state TTL.
4095                  */
4096
4097                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4098                         printf("pf: loose state match: ");
4099                         pf_print_state(*state);
4100                         pf_print_flags(th->th_flags);
4101                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4102                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4103                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4104                             (unsigned long long)(*state)->packets[1],
4105                             pd->dir == PF_IN ? "in" : "out",
4106                             pd->dir == (*state)->direction ? "fwd" : "rev");
4107                 }
4108
4109                 if (dst->scrub || src->scrub) {
4110                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4111                             *state, src, dst, copyback))
4112                                 return (PF_DROP);
4113                 }
4114
4115                 /* update max window */
4116                 if (src->max_win < win)
4117                         src->max_win = win;
4118                 /* synchronize sequencing */
4119                 if (SEQ_GT(end, src->seqlo))
4120                         src->seqlo = end;
4121                 /* slide the window of what the other end can send */
4122                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4123                         dst->seqhi = ack + MAX((win << sws), 1);
4124
4125                 /*
4126                  * Cannot set dst->seqhi here since this could be a shotgunned
4127                  * SYN and not an already established connection.
4128                  */
4129
4130                 if (th->th_flags & TH_FIN)
4131                         if (src->state < TCPS_CLOSING)
4132                                 src->state = TCPS_CLOSING;
4133                 if (th->th_flags & TH_RST)
4134                         src->state = dst->state = TCPS_TIME_WAIT;
4135
4136                 /* Fall through to PASS packet */
4137
4138         } else {
4139                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4140                     (*state)->src.state == TCPS_SYN_SENT) {
4141                         /* Send RST for state mismatches during handshake */
4142                         if (!(th->th_flags & TH_RST))
4143                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4144                                     pd->dst, pd->src, th->th_dport,
4145                                     th->th_sport, ntohl(th->th_ack), 0,
4146                                     TH_RST, 0, 0,
4147                                     (*state)->rule.ptr->return_ttl, 1, 0,
4148                                     kif->pfik_ifp);
4149                         src->seqlo = 0;
4150                         src->seqhi = 1;
4151                         src->max_win = 1;
4152                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4153                         printf("pf: BAD state: ");
4154                         pf_print_state(*state);
4155                         pf_print_flags(th->th_flags);
4156                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4157                             "pkts=%llu:%llu dir=%s,%s\n",
4158                             seq, orig_seq, ack, pd->p_len, ackskew,
4159                             (unsigned long long)(*state)->packets[0],
4160                             (unsigned long long)(*state)->packets[1],
4161                             pd->dir == PF_IN ? "in" : "out",
4162                             pd->dir == (*state)->direction ? "fwd" : "rev");
4163                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4164                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4165                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4166                             ' ': '2',
4167                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4168                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4169                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4170                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4171                 }
4172                 REASON_SET(reason, PFRES_BADSTATE);
4173                 return (PF_DROP);
4174         }
4175
4176         return (PF_PASS);
4177 }
4178
4179 static int
4180 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4181         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4182 {
4183         struct tcphdr           *th = pd->hdr.tcp;
4184
4185         if (th->th_flags & TH_SYN)
4186                 if (src->state < TCPS_SYN_SENT)
4187                         src->state = TCPS_SYN_SENT;
4188         if (th->th_flags & TH_FIN)
4189                 if (src->state < TCPS_CLOSING)
4190                         src->state = TCPS_CLOSING;
4191         if (th->th_flags & TH_ACK) {
4192                 if (dst->state == TCPS_SYN_SENT) {
4193                         dst->state = TCPS_ESTABLISHED;
4194                         if (src->state == TCPS_ESTABLISHED &&
4195                             (*state)->src_node != NULL &&
4196                             pf_src_connlimit(state)) {
4197                                 REASON_SET(reason, PFRES_SRCLIMIT);
4198                                 return (PF_DROP);
4199                         }
4200                 } else if (dst->state == TCPS_CLOSING) {
4201                         dst->state = TCPS_FIN_WAIT_2;
4202                 } else if (src->state == TCPS_SYN_SENT &&
4203                     dst->state < TCPS_SYN_SENT) {
4204                         /*
4205                          * Handle a special sloppy case where we only see one
4206                          * half of the connection. If there is a ACK after
4207                          * the initial SYN without ever seeing a packet from
4208                          * the destination, set the connection to established.
4209                          */
4210                         dst->state = src->state = TCPS_ESTABLISHED;
4211                         if ((*state)->src_node != NULL &&
4212                             pf_src_connlimit(state)) {
4213                                 REASON_SET(reason, PFRES_SRCLIMIT);
4214                                 return (PF_DROP);
4215                         }
4216                 } else if (src->state == TCPS_CLOSING &&
4217                     dst->state == TCPS_ESTABLISHED &&
4218                     dst->seqlo == 0) {
4219                         /*
4220                          * Handle the closing of half connections where we
4221                          * don't see the full bidirectional FIN/ACK+ACK
4222                          * handshake.
4223                          */
4224                         dst->state = TCPS_CLOSING;
4225                 }
4226         }
4227         if (th->th_flags & TH_RST)
4228                 src->state = dst->state = TCPS_TIME_WAIT;
4229
4230         /* update expire time */
4231         (*state)->expire = time_uptime;
4232         if (src->state >= TCPS_FIN_WAIT_2 &&
4233             dst->state >= TCPS_FIN_WAIT_2)
4234                 (*state)->timeout = PFTM_TCP_CLOSED;
4235         else if (src->state >= TCPS_CLOSING &&
4236             dst->state >= TCPS_CLOSING)
4237                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4238         else if (src->state < TCPS_ESTABLISHED ||
4239             dst->state < TCPS_ESTABLISHED)
4240                 (*state)->timeout = PFTM_TCP_OPENING;
4241         else if (src->state >= TCPS_CLOSING ||
4242             dst->state >= TCPS_CLOSING)
4243                 (*state)->timeout = PFTM_TCP_CLOSING;
4244         else
4245                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4246
4247         return (PF_PASS);
4248 }
4249
4250 static int
4251 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4252     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4253     u_short *reason)
4254 {
4255         struct pf_state_key_cmp  key;
4256         struct tcphdr           *th = pd->hdr.tcp;
4257         int                      copyback = 0;
4258         struct pf_state_peer    *src, *dst;
4259         struct pf_state_key     *sk;
4260
4261         bzero(&key, sizeof(key));
4262         key.af = pd->af;
4263         key.proto = IPPROTO_TCP;
4264         if (direction == PF_IN) {       /* wire side, straight */
4265                 PF_ACPY(&key.addr[0], pd->src, key.af);
4266                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4267                 key.port[0] = th->th_sport;
4268                 key.port[1] = th->th_dport;
4269         } else {                        /* stack side, reverse */
4270                 PF_ACPY(&key.addr[1], pd->src, key.af);
4271                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4272                 key.port[1] = th->th_sport;
4273                 key.port[0] = th->th_dport;
4274         }
4275
4276         STATE_LOOKUP(kif, &key, direction, *state, pd);
4277
4278         if (direction == (*state)->direction) {
4279                 src = &(*state)->src;
4280                 dst = &(*state)->dst;
4281         } else {
4282                 src = &(*state)->dst;
4283                 dst = &(*state)->src;
4284         }
4285
4286         sk = (*state)->key[pd->didx];
4287
4288         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4289                 if (direction != (*state)->direction) {
4290                         REASON_SET(reason, PFRES_SYNPROXY);
4291                         return (PF_SYNPROXY_DROP);
4292                 }
4293                 if (th->th_flags & TH_SYN) {
4294                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4295                                 REASON_SET(reason, PFRES_SYNPROXY);
4296                                 return (PF_DROP);
4297                         }
4298                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4299                             pd->src, th->th_dport, th->th_sport,
4300                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4301                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4302                         REASON_SET(reason, PFRES_SYNPROXY);
4303                         return (PF_SYNPROXY_DROP);
4304                 } else if (!(th->th_flags & TH_ACK) ||
4305                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4306                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4307                         REASON_SET(reason, PFRES_SYNPROXY);
4308                         return (PF_DROP);
4309                 } else if ((*state)->src_node != NULL &&
4310                     pf_src_connlimit(state)) {
4311                         REASON_SET(reason, PFRES_SRCLIMIT);
4312                         return (PF_DROP);
4313                 } else
4314                         (*state)->src.state = PF_TCPS_PROXY_DST;
4315         }
4316         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4317                 if (direction == (*state)->direction) {
4318                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4319                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4320                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4321                                 REASON_SET(reason, PFRES_SYNPROXY);
4322                                 return (PF_DROP);
4323                         }
4324                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4325                         if ((*state)->dst.seqhi == 1)
4326                                 (*state)->dst.seqhi = htonl(arc4random());
4327                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4328                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4329                             sk->port[pd->sidx], sk->port[pd->didx],
4330                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4331                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4332                         REASON_SET(reason, PFRES_SYNPROXY);
4333                         return (PF_SYNPROXY_DROP);
4334                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4335                     (TH_SYN|TH_ACK)) ||
4336                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4337                         REASON_SET(reason, PFRES_SYNPROXY);
4338                         return (PF_DROP);
4339                 } else {
4340                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4341                         (*state)->dst.seqlo = ntohl(th->th_seq);
4342                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4343                             pd->src, th->th_dport, th->th_sport,
4344                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4345                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4346                             (*state)->tag, NULL);
4347                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4348                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4349                             sk->port[pd->sidx], sk->port[pd->didx],
4350                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4351                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4352                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4353                             (*state)->src.seqlo;
4354                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4355                             (*state)->dst.seqlo;
4356                         (*state)->src.seqhi = (*state)->src.seqlo +
4357                             (*state)->dst.max_win;
4358                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4359                             (*state)->src.max_win;
4360                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4361                         (*state)->src.state = (*state)->dst.state =
4362                             TCPS_ESTABLISHED;
4363                         REASON_SET(reason, PFRES_SYNPROXY);
4364                         return (PF_SYNPROXY_DROP);
4365                 }
4366         }
4367
4368         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4369             dst->state >= TCPS_FIN_WAIT_2 &&
4370             src->state >= TCPS_FIN_WAIT_2) {
4371                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4372                         printf("pf: state reuse ");
4373                         pf_print_state(*state);
4374                         pf_print_flags(th->th_flags);
4375                         printf("\n");
4376                 }
4377                 /* XXX make sure it's the same direction ?? */
4378                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4379                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4380                 *state = NULL;
4381                 return (PF_DROP);
4382         }
4383
4384         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4385                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4386                         return (PF_DROP);
4387         } else {
4388                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4389                     &copyback) == PF_DROP)
4390                         return (PF_DROP);
4391         }
4392
4393         /* translate source/destination address, if necessary */
4394         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4395                 struct pf_state_key *nk = (*state)->key[pd->didx];
4396
4397                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4398                     nk->port[pd->sidx] != th->th_sport)
4399                         pf_change_ap(m, pd->src, &th->th_sport,
4400                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4401                             nk->port[pd->sidx], 0, pd->af);
4402
4403                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4404                     nk->port[pd->didx] != th->th_dport)
4405                         pf_change_ap(m, pd->dst, &th->th_dport,
4406                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4407                             nk->port[pd->didx], 0, pd->af);
4408                 copyback = 1;
4409         }
4410
4411         /* Copyback sequence modulation or stateful scrub changes if needed */
4412         if (copyback)
4413                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4414
4415         return (PF_PASS);
4416 }
4417
4418 static int
4419 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4420     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4421 {
4422         struct pf_state_peer    *src, *dst;
4423         struct pf_state_key_cmp  key;
4424         struct udphdr           *uh = pd->hdr.udp;
4425
4426         bzero(&key, sizeof(key));
4427         key.af = pd->af;
4428         key.proto = IPPROTO_UDP;
4429         if (direction == PF_IN) {       /* wire side, straight */
4430                 PF_ACPY(&key.addr[0], pd->src, key.af);
4431                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4432                 key.port[0] = uh->uh_sport;
4433                 key.port[1] = uh->uh_dport;
4434         } else {                        /* stack side, reverse */
4435                 PF_ACPY(&key.addr[1], pd->src, key.af);
4436                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4437                 key.port[1] = uh->uh_sport;
4438                 key.port[0] = uh->uh_dport;
4439         }
4440
4441         STATE_LOOKUP(kif, &key, direction, *state, pd);
4442
4443         if (direction == (*state)->direction) {
4444                 src = &(*state)->src;
4445                 dst = &(*state)->dst;
4446         } else {
4447                 src = &(*state)->dst;
4448                 dst = &(*state)->src;
4449         }
4450
4451         /* update states */
4452         if (src->state < PFUDPS_SINGLE)
4453                 src->state = PFUDPS_SINGLE;
4454         if (dst->state == PFUDPS_SINGLE)
4455                 dst->state = PFUDPS_MULTIPLE;
4456
4457         /* update expire time */
4458         (*state)->expire = time_uptime;
4459         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4460                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4461         else
4462                 (*state)->timeout = PFTM_UDP_SINGLE;
4463
4464         /* translate source/destination address, if necessary */
4465         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4466                 struct pf_state_key *nk = (*state)->key[pd->didx];
4467
4468                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4469                     nk->port[pd->sidx] != uh->uh_sport)
4470                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4471                             &uh->uh_sum, &nk->addr[pd->sidx],
4472                             nk->port[pd->sidx], 1, pd->af);
4473
4474                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4475                     nk->port[pd->didx] != uh->uh_dport)
4476                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4477                             &uh->uh_sum, &nk->addr[pd->didx],
4478                             nk->port[pd->didx], 1, pd->af);
4479                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4480         }
4481
4482         return (PF_PASS);
4483 }
4484
4485 static int
4486 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4487     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4488 {
4489         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4490         u_int16_t        icmpid = 0, *icmpsum;
4491         u_int8_t         icmptype;
4492         int              state_icmp = 0;
4493         struct pf_state_key_cmp key;
4494
4495         bzero(&key, sizeof(key));
4496         switch (pd->proto) {
4497 #ifdef INET
4498         case IPPROTO_ICMP:
4499                 icmptype = pd->hdr.icmp->icmp_type;
4500                 icmpid = pd->hdr.icmp->icmp_id;
4501                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4502
4503                 if (icmptype == ICMP_UNREACH ||
4504                     icmptype == ICMP_SOURCEQUENCH ||
4505                     icmptype == ICMP_REDIRECT ||
4506                     icmptype == ICMP_TIMXCEED ||
4507                     icmptype == ICMP_PARAMPROB)
4508                         state_icmp++;
4509                 break;
4510 #endif /* INET */
4511 #ifdef INET6
4512         case IPPROTO_ICMPV6:
4513                 icmptype = pd->hdr.icmp6->icmp6_type;
4514                 icmpid = pd->hdr.icmp6->icmp6_id;
4515                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4516
4517                 if (icmptype == ICMP6_DST_UNREACH ||
4518                     icmptype == ICMP6_PACKET_TOO_BIG ||
4519                     icmptype == ICMP6_TIME_EXCEEDED ||
4520                     icmptype == ICMP6_PARAM_PROB)
4521                         state_icmp++;
4522                 break;
4523 #endif /* INET6 */
4524         }
4525
4526         if (!state_icmp) {
4527
4528                 /*
4529                  * ICMP query/reply message not related to a TCP/UDP packet.
4530                  * Search for an ICMP state.
4531                  */
4532                 key.af = pd->af;
4533                 key.proto = pd->proto;
4534                 key.port[0] = key.port[1] = icmpid;
4535                 if (direction == PF_IN) {       /* wire side, straight */
4536                         PF_ACPY(&key.addr[0], pd->src, key.af);
4537                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4538                 } else {                        /* stack side, reverse */
4539                         PF_ACPY(&key.addr[1], pd->src, key.af);
4540                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4541                 }
4542
4543                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4544
4545                 (*state)->expire = time_uptime;
4546                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4547
4548                 /* translate source/destination address, if necessary */
4549                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4550                         struct pf_state_key *nk = (*state)->key[pd->didx];
4551
4552                         switch (pd->af) {
4553 #ifdef INET
4554                         case AF_INET:
4555                                 if (PF_ANEQ(pd->src,
4556                                     &nk->addr[pd->sidx], AF_INET))
4557                                         pf_change_a(&saddr->v4.s_addr,
4558                                             pd->ip_sum,
4559                                             nk->addr[pd->sidx].v4.s_addr, 0);
4560
4561                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4562                                     AF_INET))
4563                                         pf_change_a(&daddr->v4.s_addr,
4564                                             pd->ip_sum,
4565                                             nk->addr[pd->didx].v4.s_addr, 0);
4566
4567                                 if (nk->port[0] !=
4568                                     pd->hdr.icmp->icmp_id) {
4569                                         pd->hdr.icmp->icmp_cksum =
4570                                             pf_cksum_fixup(
4571                                             pd->hdr.icmp->icmp_cksum, icmpid,
4572                                             nk->port[pd->sidx], 0);
4573                                         pd->hdr.icmp->icmp_id =
4574                                             nk->port[pd->sidx];
4575                                 }
4576
4577                                 m_copyback(m, off, ICMP_MINLEN,
4578                                     (caddr_t )pd->hdr.icmp);
4579                                 break;
4580 #endif /* INET */
4581 #ifdef INET6
4582                         case AF_INET6:
4583                                 if (PF_ANEQ(pd->src,
4584                                     &nk->addr[pd->sidx], AF_INET6))
4585                                         pf_change_a6(saddr,
4586                                             &pd->hdr.icmp6->icmp6_cksum,
4587                                             &nk->addr[pd->sidx], 0);
4588
4589                                 if (PF_ANEQ(pd->dst,
4590                                     &nk->addr[pd->didx], AF_INET6))
4591                                         pf_change_a6(daddr,
4592                                             &pd->hdr.icmp6->icmp6_cksum,
4593                                             &nk->addr[pd->didx], 0);
4594
4595                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4596                                     (caddr_t )pd->hdr.icmp6);
4597                                 break;
4598 #endif /* INET6 */
4599                         }
4600                 }
4601                 return (PF_PASS);
4602
4603         } else {
4604                 /*
4605                  * ICMP error message in response to a TCP/UDP packet.
4606                  * Extract the inner TCP/UDP header and search for that state.
4607                  */
4608
4609                 struct pf_pdesc pd2;
4610                 bzero(&pd2, sizeof pd2);
4611 #ifdef INET
4612                 struct ip       h2;
4613 #endif /* INET */
4614 #ifdef INET6
4615                 struct ip6_hdr  h2_6;
4616                 int             terminal = 0;
4617 #endif /* INET6 */
4618                 int             ipoff2 = 0;
4619                 int             off2 = 0;
4620
4621                 pd2.af = pd->af;
4622                 /* Payload packet is from the opposite direction. */
4623                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4624                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4625                 switch (pd->af) {
4626 #ifdef INET
4627                 case AF_INET:
4628                         /* offset of h2 in mbuf chain */
4629                         ipoff2 = off + ICMP_MINLEN;
4630
4631                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4632                             NULL, reason, pd2.af)) {
4633                                 DPFPRINTF(PF_DEBUG_MISC,
4634                                     ("pf: ICMP error message too short "
4635                                     "(ip)\n"));
4636                                 return (PF_DROP);
4637                         }
4638                         /*
4639                          * ICMP error messages don't refer to non-first
4640                          * fragments
4641                          */
4642                         if (h2.ip_off & htons(IP_OFFMASK)) {
4643                                 REASON_SET(reason, PFRES_FRAG);
4644                                 return (PF_DROP);
4645                         }
4646
4647                         /* offset of protocol header that follows h2 */
4648                         off2 = ipoff2 + (h2.ip_hl << 2);
4649
4650                         pd2.proto = h2.ip_p;
4651                         pd2.src = (struct pf_addr *)&h2.ip_src;
4652                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4653                         pd2.ip_sum = &h2.ip_sum;
4654                         break;
4655 #endif /* INET */
4656 #ifdef INET6
4657                 case AF_INET6:
4658                         ipoff2 = off + sizeof(struct icmp6_hdr);
4659
4660                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4661                             NULL, reason, pd2.af)) {
4662                                 DPFPRINTF(PF_DEBUG_MISC,
4663                                     ("pf: ICMP error message too short "
4664                                     "(ip6)\n"));
4665                                 return (PF_DROP);
4666                         }
4667                         pd2.proto = h2_6.ip6_nxt;
4668                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4669                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4670                         pd2.ip_sum = NULL;
4671                         off2 = ipoff2 + sizeof(h2_6);
4672                         do {
4673                                 switch (pd2.proto) {
4674                                 case IPPROTO_FRAGMENT:
4675                                         /*
4676                                          * ICMPv6 error messages for
4677                                          * non-first fragments
4678                                          */
4679                                         REASON_SET(reason, PFRES_FRAG);
4680                                         return (PF_DROP);
4681                                 case IPPROTO_AH:
4682                                 case IPPROTO_HOPOPTS:
4683                                 case IPPROTO_ROUTING:
4684                                 case IPPROTO_DSTOPTS: {
4685                                         /* get next header and header length */
4686                                         struct ip6_ext opt6;
4687
4688                                         if (!pf_pull_hdr(m, off2, &opt6,
4689                                             sizeof(opt6), NULL, reason,
4690                                             pd2.af)) {
4691                                                 DPFPRINTF(PF_DEBUG_MISC,
4692                                                     ("pf: ICMPv6 short opt\n"));
4693                                                 return (PF_DROP);
4694                                         }
4695                                         if (pd2.proto == IPPROTO_AH)
4696                                                 off2 += (opt6.ip6e_len + 2) * 4;
4697                                         else
4698                                                 off2 += (opt6.ip6e_len + 1) * 8;
4699                                         pd2.proto = opt6.ip6e_nxt;
4700                                         /* goto the next header */
4701                                         break;
4702                                 }
4703                                 default:
4704                                         terminal++;
4705                                         break;
4706                                 }
4707                         } while (!terminal);
4708                         break;
4709 #endif /* INET6 */
4710                 }
4711
4712                 switch (pd2.proto) {
4713                 case IPPROTO_TCP: {
4714                         struct tcphdr            th;
4715                         u_int32_t                seq;
4716                         struct pf_state_peer    *src, *dst;
4717                         u_int8_t                 dws;
4718                         int                      copyback = 0;
4719
4720                         /*
4721                          * Only the first 8 bytes of the TCP header can be
4722                          * expected. Don't access any TCP header fields after
4723                          * th_seq, an ackskew test is not possible.
4724                          */
4725                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4726                             pd2.af)) {
4727                                 DPFPRINTF(PF_DEBUG_MISC,
4728                                     ("pf: ICMP error message too short "
4729                                     "(tcp)\n"));
4730                                 return (PF_DROP);
4731                         }
4732
4733                         key.af = pd2.af;
4734                         key.proto = IPPROTO_TCP;
4735                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4736                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4737                         key.port[pd2.sidx] = th.th_sport;
4738                         key.port[pd2.didx] = th.th_dport;
4739
4740                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4741
4742                         if (direction == (*state)->direction) {
4743                                 src = &(*state)->dst;
4744                                 dst = &(*state)->src;
4745                         } else {
4746                                 src = &(*state)->src;
4747                                 dst = &(*state)->dst;
4748                         }
4749
4750                         if (src->wscale && dst->wscale)
4751                                 dws = dst->wscale & PF_WSCALE_MASK;
4752                         else
4753                                 dws = 0;
4754
4755                         /* Demodulate sequence number */
4756                         seq = ntohl(th.th_seq) - src->seqdiff;
4757                         if (src->seqdiff) {
4758                                 pf_change_a(&th.th_seq, icmpsum,
4759                                     htonl(seq), 0);
4760                                 copyback = 1;
4761                         }
4762
4763                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4764                             (!SEQ_GEQ(src->seqhi, seq) ||
4765                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4766                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4767                                         printf("pf: BAD ICMP %d:%d ",
4768                                             icmptype, pd->hdr.icmp->icmp_code);
4769                                         pf_print_host(pd->src, 0, pd->af);
4770                                         printf(" -> ");
4771                                         pf_print_host(pd->dst, 0, pd->af);
4772                                         printf(" state: ");
4773                                         pf_print_state(*state);
4774                                         printf(" seq=%u\n", seq);
4775                                 }
4776                                 REASON_SET(reason, PFRES_BADSTATE);
4777                                 return (PF_DROP);
4778                         } else {
4779                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4780                                         printf("pf: OK ICMP %d:%d ",
4781                                             icmptype, pd->hdr.icmp->icmp_code);
4782                                         pf_print_host(pd->src, 0, pd->af);
4783                                         printf(" -> ");
4784                                         pf_print_host(pd->dst, 0, pd->af);
4785                                         printf(" state: ");
4786                                         pf_print_state(*state);
4787                                         printf(" seq=%u\n", seq);
4788                                 }
4789                         }
4790
4791                         /* translate source/destination address, if necessary */
4792                         if ((*state)->key[PF_SK_WIRE] !=
4793                             (*state)->key[PF_SK_STACK]) {
4794                                 struct pf_state_key *nk =
4795                                     (*state)->key[pd->didx];
4796
4797                                 if (PF_ANEQ(pd2.src,
4798                                     &nk->addr[pd2.sidx], pd2.af) ||
4799                                     nk->port[pd2.sidx] != th.th_sport)
4800                                         pf_change_icmp(pd2.src, &th.th_sport,
4801                                             daddr, &nk->addr[pd2.sidx],
4802                                             nk->port[pd2.sidx], NULL,
4803                                             pd2.ip_sum, icmpsum,
4804                                             pd->ip_sum, 0, pd2.af);
4805
4806                                 if (PF_ANEQ(pd2.dst,
4807                                     &nk->addr[pd2.didx], pd2.af) ||
4808                                     nk->port[pd2.didx] != th.th_dport)
4809                                         pf_change_icmp(pd2.dst, &th.th_dport,
4810                                             saddr, &nk->addr[pd2.didx],
4811                                             nk->port[pd2.didx], NULL,
4812                                             pd2.ip_sum, icmpsum,
4813                                             pd->ip_sum, 0, pd2.af);
4814                                 copyback = 1;
4815                         }
4816
4817                         if (copyback) {
4818                                 switch (pd2.af) {
4819 #ifdef INET
4820                                 case AF_INET:
4821                                         m_copyback(m, off, ICMP_MINLEN,
4822                                             (caddr_t )pd->hdr.icmp);
4823                                         m_copyback(m, ipoff2, sizeof(h2),
4824                                             (caddr_t )&h2);
4825                                         break;
4826 #endif /* INET */
4827 #ifdef INET6
4828                                 case AF_INET6:
4829                                         m_copyback(m, off,
4830                                             sizeof(struct icmp6_hdr),
4831                                             (caddr_t )pd->hdr.icmp6);
4832                                         m_copyback(m, ipoff2, sizeof(h2_6),
4833                                             (caddr_t )&h2_6);
4834                                         break;
4835 #endif /* INET6 */
4836                                 }
4837                                 m_copyback(m, off2, 8, (caddr_t)&th);
4838                         }
4839
4840                         return (PF_PASS);
4841                         break;
4842                 }
4843                 case IPPROTO_UDP: {
4844                         struct udphdr           uh;
4845
4846                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4847                             NULL, reason, pd2.af)) {
4848                                 DPFPRINTF(PF_DEBUG_MISC,
4849                                     ("pf: ICMP error message too short "
4850                                     "(udp)\n"));
4851                                 return (PF_DROP);
4852                         }
4853
4854                         key.af = pd2.af;
4855                         key.proto = IPPROTO_UDP;
4856                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4857                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4858                         key.port[pd2.sidx] = uh.uh_sport;
4859                         key.port[pd2.didx] = uh.uh_dport;
4860
4861                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4862
4863                         /* translate source/destination address, if necessary */
4864                         if ((*state)->key[PF_SK_WIRE] !=
4865                             (*state)->key[PF_SK_STACK]) {
4866                                 struct pf_state_key *nk =
4867                                     (*state)->key[pd->didx];
4868
4869                                 if (PF_ANEQ(pd2.src,
4870                                     &nk->addr[pd2.sidx], pd2.af) ||
4871                                     nk->port[pd2.sidx] != uh.uh_sport)
4872                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4873                                             daddr, &nk->addr[pd2.sidx],
4874                                             nk->port[pd2.sidx], &uh.uh_sum,
4875                                             pd2.ip_sum, icmpsum,
4876                                             pd->ip_sum, 1, pd2.af);
4877
4878                                 if (PF_ANEQ(pd2.dst,
4879                                     &nk->addr[pd2.didx], pd2.af) ||
4880                                     nk->port[pd2.didx] != uh.uh_dport)
4881                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4882                                             saddr, &nk->addr[pd2.didx],
4883                                             nk->port[pd2.didx], &uh.uh_sum,
4884                                             pd2.ip_sum, icmpsum,
4885                                             pd->ip_sum, 1, pd2.af);
4886
4887                                 switch (pd2.af) {
4888 #ifdef INET
4889                                 case AF_INET:
4890                                         m_copyback(m, off, ICMP_MINLEN,
4891                                             (caddr_t )pd->hdr.icmp);
4892                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4893                                         break;
4894 #endif /* INET */
4895 #ifdef INET6
4896                                 case AF_INET6:
4897                                         m_copyback(m, off,
4898                                             sizeof(struct icmp6_hdr),
4899                                             (caddr_t )pd->hdr.icmp6);
4900                                         m_copyback(m, ipoff2, sizeof(h2_6),
4901                                             (caddr_t )&h2_6);
4902                                         break;
4903 #endif /* INET6 */
4904                                 }
4905                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4906                         }
4907                         return (PF_PASS);
4908                         break;
4909                 }
4910 #ifdef INET
4911                 case IPPROTO_ICMP: {
4912                         struct icmp             iih;
4913
4914                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4915                             NULL, reason, pd2.af)) {
4916                                 DPFPRINTF(PF_DEBUG_MISC,
4917                                     ("pf: ICMP error message too short i"
4918                                     "(icmp)\n"));
4919                                 return (PF_DROP);
4920                         }
4921
4922                         key.af = pd2.af;
4923                         key.proto = IPPROTO_ICMP;
4924                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4925                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4926                         key.port[0] = key.port[1] = iih.icmp_id;
4927
4928                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4929
4930                         /* translate source/destination address, if necessary */
4931                         if ((*state)->key[PF_SK_WIRE] !=
4932                             (*state)->key[PF_SK_STACK]) {
4933                                 struct pf_state_key *nk =
4934                                     (*state)->key[pd->didx];
4935
4936                                 if (PF_ANEQ(pd2.src,
4937                                     &nk->addr[pd2.sidx], pd2.af) ||
4938                                     nk->port[pd2.sidx] != iih.icmp_id)
4939                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4940                                             daddr, &nk->addr[pd2.sidx],
4941                                             nk->port[pd2.sidx], NULL,
4942                                             pd2.ip_sum, icmpsum,
4943                                             pd->ip_sum, 0, AF_INET);
4944
4945                                 if (PF_ANEQ(pd2.dst,
4946                                     &nk->addr[pd2.didx], pd2.af) ||
4947                                     nk->port[pd2.didx] != iih.icmp_id)
4948                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4949                                             saddr, &nk->addr[pd2.didx],
4950                                             nk->port[pd2.didx], NULL,
4951                                             pd2.ip_sum, icmpsum,
4952                                             pd->ip_sum, 0, AF_INET);
4953
4954                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4955                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4956                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4957                         }
4958                         return (PF_PASS);
4959                         break;
4960                 }
4961 #endif /* INET */
4962 #ifdef INET6
4963                 case IPPROTO_ICMPV6: {
4964                         struct icmp6_hdr        iih;
4965
4966                         if (!pf_pull_hdr(m, off2, &iih,
4967                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4968                                 DPFPRINTF(PF_DEBUG_MISC,
4969                                     ("pf: ICMP error message too short "
4970                                     "(icmp6)\n"));
4971                                 return (PF_DROP);
4972                         }
4973
4974                         key.af = pd2.af;
4975                         key.proto = IPPROTO_ICMPV6;
4976                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4977                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4978                         key.port[0] = key.port[1] = iih.icmp6_id;
4979
4980                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4981
4982                         /* translate source/destination address, if necessary */
4983                         if ((*state)->key[PF_SK_WIRE] !=
4984                             (*state)->key[PF_SK_STACK]) {
4985                                 struct pf_state_key *nk =
4986                                     (*state)->key[pd->didx];
4987
4988                                 if (PF_ANEQ(pd2.src,
4989                                     &nk->addr[pd2.sidx], pd2.af) ||
4990                                     nk->port[pd2.sidx] != iih.icmp6_id)
4991                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4992                                             daddr, &nk->addr[pd2.sidx],
4993                                             nk->port[pd2.sidx], NULL,
4994                                             pd2.ip_sum, icmpsum,
4995                                             pd->ip_sum, 0, AF_INET6);
4996
4997                                 if (PF_ANEQ(pd2.dst,
4998                                     &nk->addr[pd2.didx], pd2.af) ||
4999                                     nk->port[pd2.didx] != iih.icmp6_id)
5000                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5001                                             saddr, &nk->addr[pd2.didx],
5002                                             nk->port[pd2.didx], NULL,
5003                                             pd2.ip_sum, icmpsum,
5004                                             pd->ip_sum, 0, AF_INET6);
5005
5006                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5007                                     (caddr_t)pd->hdr.icmp6);
5008                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5009                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5010                                     (caddr_t)&iih);
5011                         }
5012                         return (PF_PASS);
5013                         break;
5014                 }
5015 #endif /* INET6 */
5016                 default: {
5017                         key.af = pd2.af;
5018                         key.proto = pd2.proto;
5019                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5020                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5021                         key.port[0] = key.port[1] = 0;
5022
5023                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5024
5025                         /* translate source/destination address, if necessary */
5026                         if ((*state)->key[PF_SK_WIRE] !=
5027                             (*state)->key[PF_SK_STACK]) {
5028                                 struct pf_state_key *nk =
5029                                     (*state)->key[pd->didx];
5030
5031                                 if (PF_ANEQ(pd2.src,
5032                                     &nk->addr[pd2.sidx], pd2.af))
5033                                         pf_change_icmp(pd2.src, NULL, daddr,
5034                                             &nk->addr[pd2.sidx], 0, NULL,
5035                                             pd2.ip_sum, icmpsum,
5036                                             pd->ip_sum, 0, pd2.af);
5037
5038                                 if (PF_ANEQ(pd2.dst,
5039                                     &nk->addr[pd2.didx], pd2.af))
5040                                         pf_change_icmp(pd2.dst, NULL, saddr,
5041                                             &nk->addr[pd2.didx], 0, NULL,
5042                                             pd2.ip_sum, icmpsum,
5043                                             pd->ip_sum, 0, pd2.af);
5044
5045                                 switch (pd2.af) {
5046 #ifdef INET
5047                                 case AF_INET:
5048                                         m_copyback(m, off, ICMP_MINLEN,
5049                                             (caddr_t)pd->hdr.icmp);
5050                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5051                                         break;
5052 #endif /* INET */
5053 #ifdef INET6
5054                                 case AF_INET6:
5055                                         m_copyback(m, off,
5056                                             sizeof(struct icmp6_hdr),
5057                                             (caddr_t )pd->hdr.icmp6);
5058                                         m_copyback(m, ipoff2, sizeof(h2_6),
5059                                             (caddr_t )&h2_6);
5060                                         break;
5061 #endif /* INET6 */
5062                                 }
5063                         }
5064                         return (PF_PASS);
5065                         break;
5066                 }
5067                 }
5068         }
5069 }
5070
5071 static int
5072 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5073     struct mbuf *m, struct pf_pdesc *pd)
5074 {
5075         struct pf_state_peer    *src, *dst;
5076         struct pf_state_key_cmp  key;
5077
5078         bzero(&key, sizeof(key));
5079         key.af = pd->af;
5080         key.proto = pd->proto;
5081         if (direction == PF_IN) {
5082                 PF_ACPY(&key.addr[0], pd->src, key.af);
5083                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5084                 key.port[0] = key.port[1] = 0;
5085         } else {
5086                 PF_ACPY(&key.addr[1], pd->src, key.af);
5087                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5088                 key.port[1] = key.port[0] = 0;
5089         }
5090
5091         STATE_LOOKUP(kif, &key, direction, *state, pd);
5092
5093         if (direction == (*state)->direction) {
5094                 src = &(*state)->src;
5095                 dst = &(*state)->dst;
5096         } else {
5097                 src = &(*state)->dst;
5098                 dst = &(*state)->src;
5099         }
5100
5101         /* update states */
5102         if (src->state < PFOTHERS_SINGLE)
5103                 src->state = PFOTHERS_SINGLE;
5104         if (dst->state == PFOTHERS_SINGLE)
5105                 dst->state = PFOTHERS_MULTIPLE;
5106
5107         /* update expire time */
5108         (*state)->expire = time_uptime;
5109         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5110                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5111         else
5112                 (*state)->timeout = PFTM_OTHER_SINGLE;
5113
5114         /* translate source/destination address, if necessary */
5115         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5116                 struct pf_state_key *nk = (*state)->key[pd->didx];
5117
5118                 KASSERT(nk, ("%s: nk is null", __func__));
5119                 KASSERT(pd, ("%s: pd is null", __func__));
5120                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5121                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5122                 switch (pd->af) {
5123 #ifdef INET
5124                 case AF_INET:
5125                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5126                                 pf_change_a(&pd->src->v4.s_addr,
5127                                     pd->ip_sum,
5128                                     nk->addr[pd->sidx].v4.s_addr,
5129                                     0);
5130
5131
5132                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5133                                 pf_change_a(&pd->dst->v4.s_addr,
5134                                     pd->ip_sum,
5135                                     nk->addr[pd->didx].v4.s_addr,
5136                                     0);
5137
5138                                 break;
5139 #endif /* INET */
5140 #ifdef INET6
5141                 case AF_INET6:
5142                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5143                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5144
5145                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5146                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5147 #endif /* INET6 */
5148                 }
5149         }
5150         return (PF_PASS);
5151 }
5152
5153 /*
5154  * ipoff and off are measured from the start of the mbuf chain.
5155  * h must be at "ipoff" on the mbuf chain.
5156  */
5157 void *
5158 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5159     u_short *actionp, u_short *reasonp, sa_family_t af)
5160 {
5161         switch (af) {
5162 #ifdef INET
5163         case AF_INET: {
5164                 struct ip       *h = mtod(m, struct ip *);
5165                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5166
5167                 if (fragoff) {
5168                         if (fragoff >= len)
5169                                 ACTION_SET(actionp, PF_PASS);
5170                         else {
5171                                 ACTION_SET(actionp, PF_DROP);
5172                                 REASON_SET(reasonp, PFRES_FRAG);
5173                         }
5174                         return (NULL);
5175                 }
5176                 if (m->m_pkthdr.len < off + len ||
5177                     ntohs(h->ip_len) < off + len) {
5178                         ACTION_SET(actionp, PF_DROP);
5179                         REASON_SET(reasonp, PFRES_SHORT);
5180                         return (NULL);
5181                 }
5182                 break;
5183         }
5184 #endif /* INET */
5185 #ifdef INET6
5186         case AF_INET6: {
5187                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5188
5189                 if (m->m_pkthdr.len < off + len ||
5190                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5191                     (unsigned)(off + len)) {
5192                         ACTION_SET(actionp, PF_DROP);
5193                         REASON_SET(reasonp, PFRES_SHORT);
5194                         return (NULL);
5195                 }
5196                 break;
5197         }
5198 #endif /* INET6 */
5199         }
5200         m_copydata(m, off, len, p);
5201         return (p);
5202 }
5203
5204 int
5205 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5206     int rtableid)
5207 {
5208 #ifdef RADIX_MPATH
5209         struct radix_node_head  *rnh;
5210 #endif
5211         struct sockaddr_in      *dst;
5212         int                      ret = 1;
5213         int                      check_mpath;
5214 #ifdef INET6
5215         struct sockaddr_in6     *dst6;
5216         struct route_in6         ro;
5217 #else
5218         struct route             ro;
5219 #endif
5220         struct radix_node       *rn;
5221         struct rtentry          *rt;
5222         struct ifnet            *ifp;
5223
5224         check_mpath = 0;
5225 #ifdef RADIX_MPATH
5226         /* XXX: stick to table 0 for now */
5227         rnh = rt_tables_get_rnh(0, af);
5228         if (rnh != NULL && rn_mpath_capable(rnh))
5229                 check_mpath = 1;
5230 #endif
5231         bzero(&ro, sizeof(ro));
5232         switch (af) {
5233         case AF_INET:
5234                 dst = satosin(&ro.ro_dst);
5235                 dst->sin_family = AF_INET;
5236                 dst->sin_len = sizeof(*dst);
5237                 dst->sin_addr = addr->v4;
5238                 break;
5239 #ifdef INET6
5240         case AF_INET6:
5241                 /*
5242                  * Skip check for addresses with embedded interface scope,
5243                  * as they would always match anyway.
5244                  */
5245                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5246                         goto out;
5247                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5248                 dst6->sin6_family = AF_INET6;
5249                 dst6->sin6_len = sizeof(*dst6);
5250                 dst6->sin6_addr = addr->v6;
5251                 break;
5252 #endif /* INET6 */
5253         default:
5254                 return (0);
5255         }
5256
5257         /* Skip checks for ipsec interfaces */
5258         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5259                 goto out;
5260
5261         switch (af) {
5262 #ifdef INET6
5263         case AF_INET6:
5264                 in6_rtalloc_ign(&ro, 0, rtableid);
5265                 break;
5266 #endif
5267 #ifdef INET
5268         case AF_INET:
5269                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5270                 break;
5271 #endif
5272         default:
5273                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5274                 break;
5275         }
5276
5277         if (ro.ro_rt != NULL) {
5278                 /* No interface given, this is a no-route check */
5279                 if (kif == NULL)
5280                         goto out;
5281
5282                 if (kif->pfik_ifp == NULL) {
5283                         ret = 0;
5284                         goto out;
5285                 }
5286
5287                 /* Perform uRPF check if passed input interface */
5288                 ret = 0;
5289                 rn = (struct radix_node *)ro.ro_rt;
5290                 do {
5291                         rt = (struct rtentry *)rn;
5292                         ifp = rt->rt_ifp;
5293
5294                         if (kif->pfik_ifp == ifp)
5295                                 ret = 1;
5296 #ifdef RADIX_MPATH
5297                         rn = rn_mpath_next(rn);
5298 #endif
5299                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5300         } else
5301                 ret = 0;
5302 out:
5303         if (ro.ro_rt != NULL)
5304                 RTFREE(ro.ro_rt);
5305         return (ret);
5306 }
5307
5308 #ifdef INET
5309 static void
5310 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5311     struct pf_state *s, struct pf_pdesc *pd)
5312 {
5313         struct mbuf             *m0, *m1;
5314         struct sockaddr_in      dst;
5315         struct ip               *ip;
5316         struct ifnet            *ifp = NULL;
5317         struct pf_addr           naddr;
5318         struct pf_src_node      *sn = NULL;
5319         int                      error = 0;
5320         uint16_t                 ip_len, ip_off;
5321
5322         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5323         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5324             __func__));
5325
5326         if ((pd->pf_mtag == NULL &&
5327             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5328             pd->pf_mtag->routed++ > 3) {
5329                 m0 = *m;
5330                 *m = NULL;
5331                 goto bad_locked;
5332         }
5333
5334         if (r->rt == PF_DUPTO) {
5335                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5336                         if (s)
5337                                 PF_STATE_UNLOCK(s);
5338                         return;
5339                 }
5340         } else {
5341                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5342                         if (s)
5343                                 PF_STATE_UNLOCK(s);
5344                         return;
5345                 }
5346                 m0 = *m;
5347         }
5348
5349         ip = mtod(m0, struct ip *);
5350
5351         bzero(&dst, sizeof(dst));
5352         dst.sin_family = AF_INET;
5353         dst.sin_len = sizeof(dst);
5354         dst.sin_addr = ip->ip_dst;
5355
5356         if (r->rt == PF_FASTROUTE) {
5357                 struct rtentry *rt;
5358
5359                 if (s)
5360                         PF_STATE_UNLOCK(s);
5361                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5362                 if (rt == NULL) {
5363                         KMOD_IPSTAT_INC(ips_noroute);
5364                         error = EHOSTUNREACH;
5365                         goto bad;
5366                 }
5367
5368                 ifp = rt->rt_ifp;
5369                 counter_u64_add(rt->rt_pksent, 1);
5370
5371                 if (rt->rt_flags & RTF_GATEWAY)
5372                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5373                 RTFREE_LOCKED(rt);
5374         } else {
5375                 if (TAILQ_EMPTY(&r->rpool.list)) {
5376                         DPFPRINTF(PF_DEBUG_URGENT,
5377                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5378                         goto bad_locked;
5379                 }
5380                 if (s == NULL) {
5381                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5382                             &naddr, NULL, &sn);
5383                         if (!PF_AZERO(&naddr, AF_INET))
5384                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5385                         ifp = r->rpool.cur->kif ?
5386                             r->rpool.cur->kif->pfik_ifp : NULL;
5387                 } else {
5388                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5389                                 dst.sin_addr.s_addr =
5390                                     s->rt_addr.v4.s_addr;
5391                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5392                         PF_STATE_UNLOCK(s);
5393                 }
5394         }
5395         if (ifp == NULL)
5396                 goto bad;
5397
5398         if (oifp != ifp) {
5399                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5400                         goto bad;
5401                 else if (m0 == NULL)
5402                         goto done;
5403                 if (m0->m_len < sizeof(struct ip)) {
5404                         DPFPRINTF(PF_DEBUG_URGENT,
5405                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5406                         goto bad;
5407                 }
5408                 ip = mtod(m0, struct ip *);
5409         }
5410
5411         if (ifp->if_flags & IFF_LOOPBACK)
5412                 m0->m_flags |= M_SKIP_FIREWALL;
5413
5414         ip_len = ntohs(ip->ip_len);
5415         ip_off = ntohs(ip->ip_off);
5416
5417         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5418         m0->m_pkthdr.csum_flags |= CSUM_IP;
5419         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5420                 in_delayed_cksum(m0);
5421                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5422         }
5423 #ifdef SCTP
5424         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5425                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5426                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5427         }
5428 #endif
5429
5430         /*
5431          * If small enough for interface, or the interface will take
5432          * care of the fragmentation for us, we can just send directly.
5433          */
5434         if (ip_len <= ifp->if_mtu ||
5435             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5436             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5437                 ip->ip_sum = 0;
5438                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5439                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5440                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5441                 }
5442                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5443                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5444                 goto done;
5445         }
5446
5447         /* Balk when DF bit is set or the interface didn't support TSO. */
5448         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5449                 error = EMSGSIZE;
5450                 KMOD_IPSTAT_INC(ips_cantfrag);
5451                 if (r->rt != PF_DUPTO) {
5452                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5453                             ifp->if_mtu);
5454                         goto done;
5455                 } else
5456                         goto bad;
5457         }
5458
5459         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5460         if (error)
5461                 goto bad;
5462
5463         for (; m0; m0 = m1) {
5464                 m1 = m0->m_nextpkt;
5465                 m0->m_nextpkt = NULL;
5466                 if (error == 0) {
5467                         m_clrprotoflags(m0);
5468                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5469                 } else
5470                         m_freem(m0);
5471         }
5472
5473         if (error == 0)
5474                 KMOD_IPSTAT_INC(ips_fragmented);
5475
5476 done:
5477         if (r->rt != PF_DUPTO)
5478                 *m = NULL;
5479         return;
5480
5481 bad_locked:
5482         if (s)
5483                 PF_STATE_UNLOCK(s);
5484 bad:
5485         m_freem(m0);
5486         goto done;
5487 }
5488 #endif /* INET */
5489
5490 #ifdef INET6
5491 static void
5492 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5493     struct pf_state *s, struct pf_pdesc *pd)
5494 {
5495         struct mbuf             *m0;
5496         struct sockaddr_in6     dst;
5497         struct ip6_hdr          *ip6;
5498         struct ifnet            *ifp = NULL;
5499         struct pf_addr           naddr;
5500         struct pf_src_node      *sn = NULL;
5501
5502         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5503         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5504             __func__));
5505
5506         if ((pd->pf_mtag == NULL &&
5507             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5508             pd->pf_mtag->routed++ > 3) {
5509                 m0 = *m;
5510                 *m = NULL;
5511                 goto bad_locked;
5512         }
5513
5514         if (r->rt == PF_DUPTO) {
5515                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5516                         if (s)
5517                                 PF_STATE_UNLOCK(s);
5518                         return;
5519                 }
5520         } else {
5521                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5522                         if (s)
5523                                 PF_STATE_UNLOCK(s);
5524                         return;
5525                 }
5526                 m0 = *m;
5527         }
5528
5529         ip6 = mtod(m0, struct ip6_hdr *);
5530
5531         bzero(&dst, sizeof(dst));
5532         dst.sin6_family = AF_INET6;
5533         dst.sin6_len = sizeof(dst);
5534         dst.sin6_addr = ip6->ip6_dst;
5535
5536         /* Cheat. XXX why only in the v6 case??? */
5537         if (r->rt == PF_FASTROUTE) {
5538                 if (s)
5539                         PF_STATE_UNLOCK(s);
5540                 m0->m_flags |= M_SKIP_FIREWALL;
5541                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5542                 *m = NULL;
5543                 return;
5544         }
5545
5546         if (TAILQ_EMPTY(&r->rpool.list)) {
5547                 DPFPRINTF(PF_DEBUG_URGENT,
5548                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5549                 goto bad_locked;
5550         }
5551         if (s == NULL) {
5552                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5553                     &naddr, NULL, &sn);
5554                 if (!PF_AZERO(&naddr, AF_INET6))
5555                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5556                             &naddr, AF_INET6);
5557                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5558         } else {
5559                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5560                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5561                             &s->rt_addr, AF_INET6);
5562                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5563         }
5564
5565         if (s)
5566                 PF_STATE_UNLOCK(s);
5567
5568         if (ifp == NULL)
5569                 goto bad;
5570
5571         if (oifp != ifp) {
5572                 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS)
5573                         goto bad;
5574                 else if (m0 == NULL)
5575                         goto done;
5576                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5577                         DPFPRINTF(PF_DEBUG_URGENT,
5578                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5579                             __func__));
5580                         goto bad;
5581                 }
5582                 ip6 = mtod(m0, struct ip6_hdr *);
5583         }
5584
5585         if (ifp->if_flags & IFF_LOOPBACK)
5586                 m0->m_flags |= M_SKIP_FIREWALL;
5587
5588         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5589             ~ifp->if_hwassist) {
5590                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5591                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5592                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5593         }
5594
5595         /*
5596          * If the packet is too large for the outgoing interface,
5597          * send back an icmp6 error.
5598          */
5599         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5600                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5601         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5602                 nd6_output(ifp, ifp, m0, &dst, NULL);
5603         else {
5604                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5605                 if (r->rt != PF_DUPTO)
5606                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5607                 else
5608                         goto bad;
5609         }
5610
5611 done:
5612         if (r->rt != PF_DUPTO)
5613                 *m = NULL;
5614         return;
5615
5616 bad_locked:
5617         if (s)
5618                 PF_STATE_UNLOCK(s);
5619 bad:
5620         m_freem(m0);
5621         goto done;
5622 }
5623 #endif /* INET6 */
5624
5625 /*
5626  * FreeBSD supports cksum offloads for the following drivers.
5627  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5628  *   ti(4), txp(4), xl(4)
5629  *
5630  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5631  *  network driver performed cksum including pseudo header, need to verify
5632  *   csum_data
5633  * CSUM_DATA_VALID :
5634  *  network driver performed cksum, needs to additional pseudo header
5635  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5636  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5637  *
5638  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5639  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5640  * TCP/UDP layer.
5641  * Also, set csum_data to 0xffff to force cksum validation.
5642  */
5643 static int
5644 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5645 {
5646         u_int16_t sum = 0;
5647         int hw_assist = 0;
5648         struct ip *ip;
5649
5650         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5651                 return (1);
5652         if (m->m_pkthdr.len < off + len)
5653                 return (1);
5654
5655         switch (p) {
5656         case IPPROTO_TCP:
5657                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5658                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5659                                 sum = m->m_pkthdr.csum_data;
5660                         } else {
5661                                 ip = mtod(m, struct ip *);
5662                                 sum = in_pseudo(ip->ip_src.s_addr,
5663                                 ip->ip_dst.s_addr, htonl((u_short)len +
5664                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5665                         }
5666                         sum ^= 0xffff;
5667                         ++hw_assist;
5668                 }
5669                 break;
5670         case IPPROTO_UDP:
5671                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5672                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5673                                 sum = m->m_pkthdr.csum_data;
5674                         } else {
5675                                 ip = mtod(m, struct ip *);
5676                                 sum = in_pseudo(ip->ip_src.s_addr,
5677                                 ip->ip_dst.s_addr, htonl((u_short)len +
5678                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5679                         }
5680                         sum ^= 0xffff;
5681                         ++hw_assist;
5682                 }
5683                 break;
5684         case IPPROTO_ICMP:
5685 #ifdef INET6
5686         case IPPROTO_ICMPV6:
5687 #endif /* INET6 */
5688                 break;
5689         default:
5690                 return (1);
5691         }
5692
5693         if (!hw_assist) {
5694                 switch (af) {
5695                 case AF_INET:
5696                         if (p == IPPROTO_ICMP) {
5697                                 if (m->m_len < off)
5698                                         return (1);
5699                                 m->m_data += off;
5700                                 m->m_len -= off;
5701                                 sum = in_cksum(m, len);
5702                                 m->m_data -= off;
5703                                 m->m_len += off;
5704                         } else {
5705                                 if (m->m_len < sizeof(struct ip))
5706                                         return (1);
5707                                 sum = in4_cksum(m, p, off, len);
5708                         }
5709                         break;
5710 #ifdef INET6
5711                 case AF_INET6:
5712                         if (m->m_len < sizeof(struct ip6_hdr))
5713                                 return (1);
5714                         sum = in6_cksum(m, p, off, len);
5715                         break;
5716 #endif /* INET6 */
5717                 default:
5718                         return (1);
5719                 }
5720         }
5721         if (sum) {
5722                 switch (p) {
5723                 case IPPROTO_TCP:
5724                     {
5725                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5726                         break;
5727                     }
5728                 case IPPROTO_UDP:
5729                     {
5730                         KMOD_UDPSTAT_INC(udps_badsum);
5731                         break;
5732                     }
5733 #ifdef INET
5734                 case IPPROTO_ICMP:
5735                     {
5736                         KMOD_ICMPSTAT_INC(icps_checksum);
5737                         break;
5738                     }
5739 #endif
5740 #ifdef INET6
5741                 case IPPROTO_ICMPV6:
5742                     {
5743                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5744                         break;
5745                     }
5746 #endif /* INET6 */
5747                 }
5748                 return (1);
5749         } else {
5750                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5751                         m->m_pkthdr.csum_flags |=
5752                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5753                         m->m_pkthdr.csum_data = 0xffff;
5754                 }
5755         }
5756         return (0);
5757 }
5758
5759
5760 #ifdef INET
5761 int
5762 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5763 {
5764         struct pfi_kif          *kif;
5765         u_short                  action, reason = 0, log = 0;
5766         struct mbuf             *m = *m0;
5767         struct ip               *h = NULL;
5768         struct m_tag            *ipfwtag;
5769         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5770         struct pf_state         *s = NULL;
5771         struct pf_ruleset       *ruleset = NULL;
5772         struct pf_pdesc          pd;
5773         int                      off, dirndx, pqid = 0;
5774
5775         M_ASSERTPKTHDR(m);
5776
5777         if (!V_pf_status.running)
5778                 return (PF_PASS);
5779
5780         memset(&pd, 0, sizeof(pd));
5781
5782         kif = (struct pfi_kif *)ifp->if_pf_kif;
5783
5784         if (kif == NULL) {
5785                 DPFPRINTF(PF_DEBUG_URGENT,
5786                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5787                 return (PF_DROP);
5788         }
5789         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5790                 return (PF_PASS);
5791
5792         if (m->m_flags & M_SKIP_FIREWALL)
5793                 return (PF_PASS);
5794
5795         pd.pf_mtag = pf_find_mtag(m);
5796
5797         PF_RULES_RLOCK();
5798
5799         if (ip_divert_ptr != NULL &&
5800             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5801                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5802                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5803                         if (pd.pf_mtag == NULL &&
5804                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5805                                 action = PF_DROP;
5806                                 goto done;
5807                         }
5808                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5809                         m_tag_delete(m, ipfwtag);
5810                 }
5811                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5812                         m->m_flags |= M_FASTFWD_OURS;
5813                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5814                 }
5815         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5816                 /* We do IP header normalization and packet reassembly here */
5817                 action = PF_DROP;
5818                 goto done;
5819         }
5820         m = *m0;        /* pf_normalize messes with m0 */
5821         h = mtod(m, struct ip *);
5822
5823         off = h->ip_hl << 2;
5824         if (off < (int)sizeof(struct ip)) {
5825                 action = PF_DROP;
5826                 REASON_SET(&reason, PFRES_SHORT);
5827                 log = 1;
5828                 goto done;
5829         }
5830
5831         pd.src = (struct pf_addr *)&h->ip_src;
5832         pd.dst = (struct pf_addr *)&h->ip_dst;
5833         pd.sport = pd.dport = NULL;
5834         pd.ip_sum = &h->ip_sum;
5835         pd.proto_sum = NULL;
5836         pd.proto = h->ip_p;
5837         pd.dir = dir;
5838         pd.sidx = (dir == PF_IN) ? 0 : 1;
5839         pd.didx = (dir == PF_IN) ? 1 : 0;
5840         pd.af = AF_INET;
5841         pd.tos = h->ip_tos;
5842         pd.tot_len = ntohs(h->ip_len);
5843
5844         /* handle fragments that didn't get reassembled by normalization */
5845         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5846                 action = pf_test_fragment(&r, dir, kif, m, h,
5847                     &pd, &a, &ruleset);
5848                 goto done;
5849         }
5850
5851         switch (h->ip_p) {
5852
5853         case IPPROTO_TCP: {
5854                 struct tcphdr   th;
5855
5856                 pd.hdr.tcp = &th;
5857                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5858                     &action, &reason, AF_INET)) {
5859                         log = action != PF_PASS;
5860                         goto done;
5861                 }
5862                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5863                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5864                         pqid = 1;
5865                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5866                 if (action == PF_DROP)
5867                         goto done;
5868                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5869                     &reason);
5870                 if (action == PF_PASS) {
5871                         if (pfsync_update_state_ptr != NULL)
5872                                 pfsync_update_state_ptr(s);
5873                         r = s->rule.ptr;
5874                         a = s->anchor.ptr;
5875                         log = s->log;
5876                 } else if (s == NULL)
5877                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5878                             &a, &ruleset, inp);
5879                 break;
5880         }
5881
5882         case IPPROTO_UDP: {
5883                 struct udphdr   uh;
5884
5885                 pd.hdr.udp = &uh;
5886                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5887                     &action, &reason, AF_INET)) {
5888                         log = action != PF_PASS;
5889                         goto done;
5890                 }
5891                 if (uh.uh_dport == 0 ||
5892                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5893                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5894                         action = PF_DROP;
5895                         REASON_SET(&reason, PFRES_SHORT);
5896                         goto done;
5897                 }
5898                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5899                 if (action == PF_PASS) {
5900                         if (pfsync_update_state_ptr != NULL)
5901                                 pfsync_update_state_ptr(s);
5902                         r = s->rule.ptr;
5903                         a = s->anchor.ptr;
5904                         log = s->log;
5905                 } else if (s == NULL)
5906                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5907                             &a, &ruleset, inp);
5908                 break;
5909         }
5910
5911         case IPPROTO_ICMP: {
5912                 struct icmp     ih;
5913
5914                 pd.hdr.icmp = &ih;
5915                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5916                     &action, &reason, AF_INET)) {
5917                         log = action != PF_PASS;
5918                         goto done;
5919                 }
5920                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5921                     &reason);
5922                 if (action == PF_PASS) {
5923                         if (pfsync_update_state_ptr != NULL)
5924                                 pfsync_update_state_ptr(s);
5925                         r = s->rule.ptr;
5926                         a = s->anchor.ptr;
5927                         log = s->log;
5928                 } else if (s == NULL)
5929                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5930                             &a, &ruleset, inp);
5931                 break;
5932         }
5933
5934 #ifdef INET6
5935         case IPPROTO_ICMPV6: {
5936                 action = PF_DROP;
5937                 DPFPRINTF(PF_DEBUG_MISC,
5938                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5939                 goto done;
5940         }
5941 #endif
5942
5943         default:
5944                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5945                 if (action == PF_PASS) {
5946                         if (pfsync_update_state_ptr != NULL)
5947                                 pfsync_update_state_ptr(s);
5948                         r = s->rule.ptr;
5949                         a = s->anchor.ptr;
5950                         log = s->log;
5951                 } else if (s == NULL)
5952                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5953                             &a, &ruleset, inp);
5954                 break;
5955         }
5956
5957 done:
5958         PF_RULES_RUNLOCK();
5959         if (action == PF_PASS && h->ip_hl > 5 &&
5960             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5961                 action = PF_DROP;
5962                 REASON_SET(&reason, PFRES_IPOPTIONS);
5963                 log = r->log;
5964                 DPFPRINTF(PF_DEBUG_MISC,
5965                     ("pf: dropping packet with ip options\n"));
5966         }
5967
5968         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5969                 action = PF_DROP;
5970                 REASON_SET(&reason, PFRES_MEMORY);
5971         }
5972         if (r->rtableid >= 0)
5973                 M_SETFIB(m, r->rtableid);
5974
5975 #ifdef ALTQ
5976         if (action == PF_PASS && r->qid) {
5977                 if (pd.pf_mtag == NULL &&
5978                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5979                         action = PF_DROP;
5980                         REASON_SET(&reason, PFRES_MEMORY);
5981                 } else {
5982                         if (s != NULL)
5983                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
5984                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
5985                                 pd.pf_mtag->qid = r->pqid;
5986                         else
5987                                 pd.pf_mtag->qid = r->qid;
5988                         /* Add hints for ecn. */
5989                         pd.pf_mtag->hdr = h;
5990                 }
5991
5992         }
5993 #endif /* ALTQ */
5994
5995         /*
5996          * connections redirected to loopback should not match sockets
5997          * bound specifically to loopback due to security implications,
5998          * see tcp_input() and in_pcblookup_listen().
5999          */
6000         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6001             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6002             (s->nat_rule.ptr->action == PF_RDR ||
6003             s->nat_rule.ptr->action == PF_BINAT) &&
6004             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6005                 m->m_flags |= M_SKIP_FIREWALL;
6006
6007         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6008             !PACKET_LOOPED(&pd)) {
6009
6010                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6011                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6012                 if (ipfwtag != NULL) {
6013                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6014                             ntohs(r->divert.port);
6015                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6016
6017                         if (s)
6018                                 PF_STATE_UNLOCK(s);
6019
6020                         m_tag_prepend(m, ipfwtag);
6021                         if (m->m_flags & M_FASTFWD_OURS) {
6022                                 if (pd.pf_mtag == NULL &&
6023                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6024                                         action = PF_DROP;
6025                                         REASON_SET(&reason, PFRES_MEMORY);
6026                                         log = 1;
6027                                         DPFPRINTF(PF_DEBUG_MISC,
6028                                             ("pf: failed to allocate tag\n"));
6029                                 } else {
6030                                         pd.pf_mtag->flags |=
6031                                             PF_FASTFWD_OURS_PRESENT;
6032                                         m->m_flags &= ~M_FASTFWD_OURS;
6033                                 }
6034                         }
6035                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6036                         *m0 = NULL;
6037
6038                         return (action);
6039                 } else {
6040                         /* XXX: ipfw has the same behaviour! */
6041                         action = PF_DROP;
6042                         REASON_SET(&reason, PFRES_MEMORY);
6043                         log = 1;
6044                         DPFPRINTF(PF_DEBUG_MISC,
6045                             ("pf: failed to allocate divert tag\n"));
6046                 }
6047         }
6048
6049         if (log) {
6050                 struct pf_rule *lr;
6051
6052                 if (s != NULL && s->nat_rule.ptr != NULL &&
6053                     s->nat_rule.ptr->log & PF_LOG_ALL)
6054                         lr = s->nat_rule.ptr;
6055                 else
6056                         lr = r;
6057                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6058                     (s == NULL));
6059         }
6060
6061         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6062         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6063
6064         if (action == PF_PASS || r->action == PF_DROP) {
6065                 dirndx = (dir == PF_OUT);
6066                 r->packets[dirndx]++;
6067                 r->bytes[dirndx] += pd.tot_len;
6068                 if (a != NULL) {
6069                         a->packets[dirndx]++;
6070                         a->bytes[dirndx] += pd.tot_len;
6071                 }
6072                 if (s != NULL) {
6073                         if (s->nat_rule.ptr != NULL) {
6074                                 s->nat_rule.ptr->packets[dirndx]++;
6075                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6076                         }
6077                         if (s->src_node != NULL) {
6078                                 s->src_node->packets[dirndx]++;
6079                                 s->src_node->bytes[dirndx] += pd.tot_len;
6080                         }
6081                         if (s->nat_src_node != NULL) {
6082                                 s->nat_src_node->packets[dirndx]++;
6083                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6084                         }
6085                         dirndx = (dir == s->direction) ? 0 : 1;
6086                         s->packets[dirndx]++;
6087                         s->bytes[dirndx] += pd.tot_len;
6088                 }
6089                 tr = r;
6090                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6091                 if (nr != NULL && r == &V_pf_default_rule)
6092                         tr = nr;
6093                 if (tr->src.addr.type == PF_ADDR_TABLE)
6094                         pfr_update_stats(tr->src.addr.p.tbl,
6095                             (s == NULL) ? pd.src :
6096                             &s->key[(s->direction == PF_IN)]->
6097                                 addr[(s->direction == PF_OUT)],
6098                             pd.af, pd.tot_len, dir == PF_OUT,
6099                             r->action == PF_PASS, tr->src.neg);
6100                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6101                         pfr_update_stats(tr->dst.addr.p.tbl,
6102                             (s == NULL) ? pd.dst :
6103                             &s->key[(s->direction == PF_IN)]->
6104                                 addr[(s->direction == PF_IN)],
6105                             pd.af, pd.tot_len, dir == PF_OUT,
6106                             r->action == PF_PASS, tr->dst.neg);
6107         }
6108
6109         switch (action) {
6110         case PF_SYNPROXY_DROP:
6111                 m_freem(*m0);
6112         case PF_DEFER:
6113                 *m0 = NULL;
6114                 action = PF_PASS;
6115                 break;
6116         case PF_DROP:
6117                 m_freem(*m0);
6118                 *m0 = NULL;
6119                 break;
6120         default:
6121                 /* pf_route() returns unlocked. */
6122                 if (r->rt) {
6123                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6124                         return (action);
6125                 }
6126                 break;
6127         }
6128         if (s)
6129                 PF_STATE_UNLOCK(s);
6130
6131         return (action);
6132 }
6133 #endif /* INET */
6134
6135 #ifdef INET6
6136 int
6137 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6138 {
6139         struct pfi_kif          *kif;
6140         u_short                  action, reason = 0, log = 0;
6141         struct mbuf             *m = *m0, *n = NULL;
6142         struct m_tag            *mtag;
6143         struct ip6_hdr          *h = NULL;
6144         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6145         struct pf_state         *s = NULL;
6146         struct pf_ruleset       *ruleset = NULL;
6147         struct pf_pdesc          pd;
6148         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6149         int                      fwdir = dir;
6150
6151         M_ASSERTPKTHDR(m);
6152
6153         /* Detect packet forwarding.
6154          * If the input interface is different from the output interface we're
6155          * forwarding.
6156          * We do need to be careful about bridges. If the
6157          * net.link.bridge.pfil_bridge sysctl is set we can be filtering on a
6158          * bridge, so if the input interface is a bridge member and the output
6159          * interface is its bridge or a member of the same bridge we're not
6160          * actually forwarding but bridging.
6161          */
6162         if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif &&
6163             (m->m_pkthdr.rcvif->if_bridge == NULL ||
6164             (m->m_pkthdr.rcvif->if_bridge != ifp->if_softc &&
6165             m->m_pkthdr.rcvif->if_bridge != ifp->if_bridge)))
6166                 fwdir = PF_FWD;
6167
6168         if (dir == PF_FWD)
6169                 dir = PF_OUT;
6170
6171         if (!V_pf_status.running)
6172                 return (PF_PASS);
6173
6174         memset(&pd, 0, sizeof(pd));
6175         pd.pf_mtag = pf_find_mtag(m);
6176
6177         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6178                 return (PF_PASS);
6179
6180         kif = (struct pfi_kif *)ifp->if_pf_kif;
6181         if (kif == NULL) {
6182                 DPFPRINTF(PF_DEBUG_URGENT,
6183                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6184                 return (PF_DROP);
6185         }
6186         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6187                 return (PF_PASS);
6188
6189         if (m->m_flags & M_SKIP_FIREWALL)
6190                 return (PF_PASS);
6191
6192         PF_RULES_RLOCK();
6193
6194         /* We do IP header normalization and packet reassembly here */
6195         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6196                 action = PF_DROP;
6197                 goto done;
6198         }
6199         m = *m0;        /* pf_normalize messes with m0 */
6200         h = mtod(m, struct ip6_hdr *);
6201
6202 #if 1
6203         /*
6204          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6205          * will do something bad, so drop the packet for now.
6206          */
6207         if (htons(h->ip6_plen) == 0) {
6208                 action = PF_DROP;
6209                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6210                 goto done;
6211         }
6212 #endif
6213
6214         pd.src = (struct pf_addr *)&h->ip6_src;
6215         pd.dst = (struct pf_addr *)&h->ip6_dst;
6216         pd.sport = pd.dport = NULL;
6217         pd.ip_sum = NULL;
6218         pd.proto_sum = NULL;
6219         pd.dir = dir;
6220         pd.sidx = (dir == PF_IN) ? 0 : 1;
6221         pd.didx = (dir == PF_IN) ? 1 : 0;
6222         pd.af = AF_INET6;
6223         pd.tos = 0;
6224         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6225
6226         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6227         pd.proto = h->ip6_nxt;
6228         do {
6229                 switch (pd.proto) {
6230                 case IPPROTO_FRAGMENT:
6231                         action = pf_test_fragment(&r, dir, kif, m, h,
6232                             &pd, &a, &ruleset);
6233                         if (action == PF_DROP)
6234                                 REASON_SET(&reason, PFRES_FRAG);
6235                         goto done;
6236                 case IPPROTO_ROUTING: {
6237                         struct ip6_rthdr rthdr;
6238
6239                         if (rh_cnt++) {
6240                                 DPFPRINTF(PF_DEBUG_MISC,
6241                                     ("pf: IPv6 more than one rthdr\n"));
6242                                 action = PF_DROP;
6243                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6244                                 log = 1;
6245                                 goto done;
6246                         }
6247                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6248                             &reason, pd.af)) {
6249                                 DPFPRINTF(PF_DEBUG_MISC,
6250                                     ("pf: IPv6 short rthdr\n"));
6251                                 action = PF_DROP;
6252                                 REASON_SET(&reason, PFRES_SHORT);
6253                                 log = 1;
6254                                 goto done;
6255                         }
6256                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6257                                 DPFPRINTF(PF_DEBUG_MISC,
6258                                     ("pf: IPv6 rthdr0\n"));
6259                                 action = PF_DROP;
6260                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6261                                 log = 1;
6262                                 goto done;
6263                         }
6264                         /* FALLTHROUGH */
6265                 }
6266                 case IPPROTO_AH:
6267                 case IPPROTO_HOPOPTS:
6268                 case IPPROTO_DSTOPTS: {
6269                         /* get next header and header length */
6270                         struct ip6_ext  opt6;
6271
6272                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6273                             NULL, &reason, pd.af)) {
6274                                 DPFPRINTF(PF_DEBUG_MISC,
6275                                     ("pf: IPv6 short opt\n"));
6276                                 action = PF_DROP;
6277                                 log = 1;
6278                                 goto done;
6279                         }
6280                         if (pd.proto == IPPROTO_AH)
6281                                 off += (opt6.ip6e_len + 2) * 4;
6282                         else
6283                                 off += (opt6.ip6e_len + 1) * 8;
6284                         pd.proto = opt6.ip6e_nxt;
6285                         /* goto the next header */
6286                         break;
6287                 }
6288                 default:
6289                         terminal++;
6290                         break;
6291                 }
6292         } while (!terminal);
6293
6294         /* if there's no routing header, use unmodified mbuf for checksumming */
6295         if (!n)
6296                 n = m;
6297
6298         switch (pd.proto) {
6299
6300         case IPPROTO_TCP: {
6301                 struct tcphdr   th;
6302
6303                 pd.hdr.tcp = &th;
6304                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6305                     &action, &reason, AF_INET6)) {
6306                         log = action != PF_PASS;
6307                         goto done;
6308                 }
6309                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6310                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6311                 if (action == PF_DROP)
6312                         goto done;
6313                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6314                     &reason);
6315                 if (action == PF_PASS) {
6316                         if (pfsync_update_state_ptr != NULL)
6317                                 pfsync_update_state_ptr(s);
6318                         r = s->rule.ptr;
6319                         a = s->anchor.ptr;
6320                         log = s->log;
6321                 } else if (s == NULL)
6322                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6323                             &a, &ruleset, inp);
6324                 break;
6325         }
6326
6327         case IPPROTO_UDP: {
6328                 struct udphdr   uh;
6329
6330                 pd.hdr.udp = &uh;
6331                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6332                     &action, &reason, AF_INET6)) {
6333                         log = action != PF_PASS;
6334                         goto done;
6335                 }
6336                 if (uh.uh_dport == 0 ||
6337                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6338                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6339                         action = PF_DROP;
6340                         REASON_SET(&reason, PFRES_SHORT);
6341                         goto done;
6342                 }
6343                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6344                 if (action == PF_PASS) {
6345                         if (pfsync_update_state_ptr != NULL)
6346                                 pfsync_update_state_ptr(s);
6347                         r = s->rule.ptr;
6348                         a = s->anchor.ptr;
6349                         log = s->log;
6350                 } else if (s == NULL)
6351                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6352                             &a, &ruleset, inp);
6353                 break;
6354         }
6355
6356         case IPPROTO_ICMP: {
6357                 action = PF_DROP;
6358                 DPFPRINTF(PF_DEBUG_MISC,
6359                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6360                 goto done;
6361         }
6362
6363         case IPPROTO_ICMPV6: {
6364                 struct icmp6_hdr        ih;
6365
6366                 pd.hdr.icmp6 = &ih;
6367                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6368                     &action, &reason, AF_INET6)) {
6369                         log = action != PF_PASS;
6370                         goto done;
6371                 }
6372                 action = pf_test_state_icmp(&s, dir, kif,
6373                     m, off, h, &pd, &reason);
6374                 if (action == PF_PASS) {
6375                         if (pfsync_update_state_ptr != NULL)
6376                                 pfsync_update_state_ptr(s);
6377                         r = s->rule.ptr;
6378                         a = s->anchor.ptr;
6379                         log = s->log;
6380                 } else if (s == NULL)
6381                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6382                             &a, &ruleset, inp);
6383                 break;
6384         }
6385
6386         default:
6387                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6388                 if (action == PF_PASS) {
6389                         if (pfsync_update_state_ptr != NULL)
6390                                 pfsync_update_state_ptr(s);
6391                         r = s->rule.ptr;
6392                         a = s->anchor.ptr;
6393                         log = s->log;
6394                 } else if (s == NULL)
6395                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6396                             &a, &ruleset, inp);
6397                 break;
6398         }
6399
6400 done:
6401         PF_RULES_RUNLOCK();
6402         if (n != m) {
6403                 m_freem(n);
6404                 n = NULL;
6405         }
6406
6407         /* handle dangerous IPv6 extension headers. */
6408         if (action == PF_PASS && rh_cnt &&
6409             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6410                 action = PF_DROP;
6411                 REASON_SET(&reason, PFRES_IPOPTIONS);
6412                 log = r->log;
6413                 DPFPRINTF(PF_DEBUG_MISC,
6414                     ("pf: dropping packet with dangerous v6 headers\n"));
6415         }
6416
6417         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6418                 action = PF_DROP;
6419                 REASON_SET(&reason, PFRES_MEMORY);
6420         }
6421         if (r->rtableid >= 0)
6422                 M_SETFIB(m, r->rtableid);
6423
6424 #ifdef ALTQ
6425         if (action == PF_PASS && r->qid) {
6426                 if (pd.pf_mtag == NULL &&
6427                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6428                         action = PF_DROP;
6429                         REASON_SET(&reason, PFRES_MEMORY);
6430                 } else {
6431                         if (s != NULL)
6432                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6433                         if (pd.tos & IPTOS_LOWDELAY)
6434                                 pd.pf_mtag->qid = r->pqid;
6435                         else
6436                                 pd.pf_mtag->qid = r->qid;
6437                         /* Add hints for ecn. */
6438                         pd.pf_mtag->hdr = h;
6439                 }
6440         }
6441 #endif /* ALTQ */
6442
6443         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6444             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6445             (s->nat_rule.ptr->action == PF_RDR ||
6446             s->nat_rule.ptr->action == PF_BINAT) &&
6447             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6448                 m->m_flags |= M_SKIP_FIREWALL;
6449
6450         /* XXX: Anybody working on it?! */
6451         if (r->divert.port)
6452                 printf("pf: divert(9) is not supported for IPv6\n");
6453
6454         if (log) {
6455                 struct pf_rule *lr;
6456
6457                 if (s != NULL && s->nat_rule.ptr != NULL &&
6458                     s->nat_rule.ptr->log & PF_LOG_ALL)
6459                         lr = s->nat_rule.ptr;
6460                 else
6461                         lr = r;
6462                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6463                     &pd, (s == NULL));
6464         }
6465
6466         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6467         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6468
6469         if (action == PF_PASS || r->action == PF_DROP) {
6470                 dirndx = (dir == PF_OUT);
6471                 r->packets[dirndx]++;
6472                 r->bytes[dirndx] += pd.tot_len;
6473                 if (a != NULL) {
6474                         a->packets[dirndx]++;
6475                         a->bytes[dirndx] += pd.tot_len;
6476                 }
6477                 if (s != NULL) {
6478                         if (s->nat_rule.ptr != NULL) {
6479                                 s->nat_rule.ptr->packets[dirndx]++;
6480                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6481                         }
6482                         if (s->src_node != NULL) {
6483                                 s->src_node->packets[dirndx]++;
6484                                 s->src_node->bytes[dirndx] += pd.tot_len;
6485                         }
6486                         if (s->nat_src_node != NULL) {
6487                                 s->nat_src_node->packets[dirndx]++;
6488                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6489                         }
6490                         dirndx = (dir == s->direction) ? 0 : 1;
6491                         s->packets[dirndx]++;
6492                         s->bytes[dirndx] += pd.tot_len;
6493                 }
6494                 tr = r;
6495                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6496                 if (nr != NULL && r == &V_pf_default_rule)
6497                         tr = nr;
6498                 if (tr->src.addr.type == PF_ADDR_TABLE)
6499                         pfr_update_stats(tr->src.addr.p.tbl,
6500                             (s == NULL) ? pd.src :
6501                             &s->key[(s->direction == PF_IN)]->addr[0],
6502                             pd.af, pd.tot_len, dir == PF_OUT,
6503                             r->action == PF_PASS, tr->src.neg);
6504                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6505                         pfr_update_stats(tr->dst.addr.p.tbl,
6506                             (s == NULL) ? pd.dst :
6507                             &s->key[(s->direction == PF_IN)]->addr[1],
6508                             pd.af, pd.tot_len, dir == PF_OUT,
6509                             r->action == PF_PASS, tr->dst.neg);
6510         }
6511
6512         switch (action) {
6513         case PF_SYNPROXY_DROP:
6514                 m_freem(*m0);
6515         case PF_DEFER:
6516                 *m0 = NULL;
6517                 action = PF_PASS;
6518                 break;
6519         case PF_DROP:
6520                 m_freem(*m0);
6521                 *m0 = NULL;
6522                 break;
6523         default:
6524                 /* pf_route6() returns unlocked. */
6525                 if (r->rt) {
6526                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6527                         return (action);
6528                 }
6529                 break;
6530         }
6531
6532         if (s)
6533                 PF_STATE_UNLOCK(s);
6534
6535         /* If reassembled packet passed, create new fragments. */
6536         if (action == PF_PASS && *m0 && fwdir == PF_FWD &&
6537             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6538                 action = pf_refragment6(ifp, m0, mtag);
6539
6540         return (action);
6541 }
6542 #endif /* INET6 */