sys/net/pf/if_pfsync.c

   1 /*      $OpenBSD: if_pfsync.c,v 1.73 2006/11/16 13:13:38 henning Exp $  */
   2
   3 /*
   4  * Copyright (c) 2002 Michael Shalayeff
   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  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  19  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
  20  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  22  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  24  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  25  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
  26  * THE POSSIBILITY OF SUCH DAMAGE.
  27  */
  28
  29 #include "opt_inet.h"
  30 #include "opt_inet6.h"
  31 #include "use_carp.h"
  32
  33 #include <sys/param.h>
  34 #include <sys/endian.h>
  35 #include <sys/proc.h>
  36 #include <sys/priv.h>
  37 #include <sys/systm.h>
  38 #include <sys/time.h>
  39 #include <sys/mbuf.h>
  40 #include <sys/socket.h>
  41 #include <sys/kernel.h>
  42 #include <sys/malloc.h>
  43 #include <sys/module.h>
  44 #include <sys/sockio.h>
  45 #include <sys/thread2.h>
  46 #include <vm/vm_zone.h>
  47
  48 #include <machine/inttypes.h>
  49
  50 #include <net/if.h>
  51 #include <net/if_types.h>
  52 #include <net/route.h>
  53 #include <net/bpf.h>
  54 #include <netinet/in.h>
  55 #include <netinet/if_ether.h>
  56 #include <netinet/ip_carp.h>
  57 #include <netinet/tcp.h>
  58 #include <netinet/tcp_seq.h>
  59
  60 #ifdef  INET
  61 #include <netinet/in_systm.h>
  62 #include <netinet/in_var.h>
  63 #include <netinet/ip.h>
  64 #include <netinet/ip_var.h>
  65 #endif
  66
  67 #ifdef INET6
  68 #include <netinet6/nd6.h>
  69 #endif /* INET6 */
  70
  71 #include <net/pf/pfvar.h>
  72 #include <net/pf/if_pfsync.h>
  73
  74 #define PFSYNCNAME      "pfsync"
  75
  76 #define PFSYNC_MINMTU   \
  77     (sizeof(struct pfsync_header) + sizeof(struct pf_state))
  78
  79 #ifdef PFSYNCDEBUG
  80 #define DPRINTF(x)    do { if (pfsyncdebug) kprintf x ; } while (0)
  81 int pfsyncdebug;
  82 #else
  83 #define DPRINTF(x)
  84 #endif
  85
  86 struct pfsync_softc     *pfsyncif = NULL;
  87 struct pfsyncstats       pfsyncstats;
  88
  89 void    pfsyncattach(int);
  90 static void     pfsync_clone_destroy(struct ifnet *);
  91 static int      pfsync_clone_create(struct if_clone *, int, caddr_t);
  92 void    pfsync_setmtu(struct pfsync_softc *, int);
  93 int     pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
  94             struct pf_state_peer *);
  95 int     pfsync_insert_net_state(struct pfsync_state *, u_int8_t);
  96 int     pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
  97             struct rtentry *);
  98 int     pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
  99 void    pfsyncstart(struct ifnet *);
 100
 101 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **);
 102 int     pfsync_request_update(struct pfsync_state_upd *, struct in_addr *);
 103 int     pfsync_sendout(struct pfsync_softc *);
 104 int     pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *);
 105 void    pfsync_timeout(void *);
 106 void    pfsync_send_bus(struct pfsync_softc *, u_int8_t);
 107 void    pfsync_bulk_update(void *);
 108 void    pfsync_bulkfail(void *);
 109
 110 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface");
 111 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list;
 112
 113 int     pfsync_sync_ok;
 114
 115 struct if_clone pfsync_cloner =
 116     IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1);
 117
 118 void
 119 pfsyncattach(int npfsync)
 120 {
 121         if_clone_attach(&pfsync_cloner);
 122 }
 123 static int
 124 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused)
 125 {
 126         struct pfsync_softc *sc;
 127         struct ifnet *ifp;
 128
 129         lwkt_gettoken(&pf_token);
 130
 131         MALLOC(sc, struct pfsync_softc *, sizeof(*sc), M_PFSYNC,
 132             M_WAITOK|M_ZERO);
 133
 134         pfsync_sync_ok = 1;
 135         sc->sc_mbuf = NULL;
 136         sc->sc_mbuf_net = NULL;
 137         sc->sc_mbuf_tdb = NULL;
 138         sc->sc_statep.s = NULL;
 139         sc->sc_statep_net.s = NULL;
 140         sc->sc_statep_tdb.t = NULL;
 141         sc->sc_maxupdates = 128;
 142         sc->sc_sync_peer.s_addr =htonl(INADDR_PFSYNC_GROUP);
 143         sc->sc_sendaddr.s_addr = htonl(INADDR_PFSYNC_GROUP);
 144         sc->sc_ureq_received = 0;
 145         sc->sc_ureq_sent = 0;
 146         sc->sc_bulk_send_next = NULL;
 147         sc->sc_bulk_terminator = NULL;
 148
 149         lwkt_reltoken(&pf_token);
 150         ifp = &sc->sc_if;
 151         ksnprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit);
 152         if_initname(ifp, ifc->ifc_name, unit);
 153         ifp->if_ioctl = pfsyncioctl;
 154         ifp->if_output = pfsyncoutput;
 155         ifp->if_start = pfsyncstart;
 156         ifp->if_type = IFT_PFSYNC;
 157         ifp->if_snd.ifq_maxlen = ifqmaxlen;
 158         ifp->if_hdrlen = PFSYNC_HDRLEN;
 159         ifp->if_baudrate = IF_Mbps(100);
 160         ifp->if_softc = sc;
 161         pfsync_setmtu(sc, MCLBYTES);
 162         callout_init(&sc->sc_tmo);
 163         callout_init(&sc->sc_tdb_tmo);
 164         callout_init(&sc->sc_bulk_tmo);
 165         callout_init(&sc->sc_bulkfail_tmo);
 166         if_attach(&sc->sc_if, NULL);
 167
 168         LIST_INSERT_HEAD(&pfsync_list, sc, sc_next);
 169         bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN);
 170
 171 #if NCARP > 0
 172         if_addgroup(ifp, "carp");
 173 #endif
 174         lwkt_gettoken(&pf_token);
 175
 176         lwkt_reltoken(&pf_token);
 177         return (0);
 178 }
 179
 180 static void
 181 pfsync_clone_destroy(struct ifnet *ifp)
 182 {
 183         lwkt_gettoken(&pf_token);
 184         lwkt_reltoken(&pf_token);
 185 #if NBPFILTER > 0
 186         bpfdetach(ifp);
 187 #endif
 188         if_detach(ifp);
 189         lwkt_gettoken(&pf_token);
 190         kfree(pfsyncif, M_DEVBUF);
 191         pfsyncif = NULL;
 192 }
 193
 194 /*
 195  * Start output on the pfsync interface.
 196  */
 197 void
 198 pfsyncstart(struct ifnet *ifp)
 199 {
 200         crit_enter();
 201         IF_DROP(&ifp->if_snd);
 202         IF_DRAIN(&ifp->if_snd);
 203         crit_exit();
 204 }
 205
 206 int
 207 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
 208     struct pf_state_peer *d)
 209 {
 210         if (s->scrub.scrub_flag && d->scrub == NULL) {
 211                 d->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT);
 212                 if (d->scrub == NULL)
 213                         return (ENOMEM);
 214                 bzero(d->scrub, sizeof(*d->scrub));
 215         }
 216
 217         return (0);
 218 }
 219
 220 int
 221 pfsync_insert_net_state(struct pfsync_state *sp, u_int8_t chksum_flag)
 222 {
 223         struct pf_state *st = NULL;
 224         struct pf_state_key *sk = NULL;
 225         struct pf_rule *r = NULL;
 226         struct pfi_kif  *kif;
 227
 228         if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) {
 229                 kprintf("pfsync_insert_net_state: invalid creator id:"
 230                     " %08x\n", ntohl(sp->creatorid));
 231                 return (EINVAL);
 232         }
 233
 234         kif = pfi_kif_get(sp->ifname);
 235         if (kif == NULL) {
 236                 if (pf_status.debug >= PF_DEBUG_MISC)
 237                         kprintf("pfsync_insert_net_state: "
 238                             "unknown interface: %s\n", sp->ifname);
 239                 /* skip this state */
 240                 return (0);
 241         }
 242
 243         /*
 244          * If the ruleset checksums match, it's safe to associate the state
 245          * with the rule of that number.
 246          */
 247         if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag &&
 248             ntohl(sp->rule) <
 249             pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
 250                 r = pf_main_ruleset.rules[
 251                     PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
 252         else
 253                 r = &pf_default_rule;
 254
 255         if (!r->max_states || r->states < r->max_states)
 256                 st = pool_get(&pf_state_pl, PR_NOWAIT);
 257         if (st == NULL) {
 258                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 259                 return (ENOMEM);
 260         }
 261         bzero(st, sizeof(*st));
 262
 263         if ((sk = pf_alloc_state_key(st)) == NULL) {
 264                 pool_put(&pf_state_pl, st);
 265                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 266                 return (ENOMEM);
 267         }
 268
 269         /* allocate memory for scrub info */
 270         if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
 271             pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) {
 272                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 273                 if (st->src.scrub)
 274                         pool_put(&pf_state_scrub_pl, st->src.scrub);
 275                 pool_put(&pf_state_pl, st);
 276                 pool_put(&pf_state_key_pl, sk);
 277                 return (ENOMEM);
 278         }
 279
 280         st->rule.ptr = r;
 281         /* XXX get pointers to nat_rule and anchor */
 282
 283         /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
 284         r->states++;
 285
 286         /* fill in the rest of the state entry */
 287         pf_state_host_ntoh(&sp->lan, &sk->lan);
 288         pf_state_host_ntoh(&sp->gwy, &sk->gwy);
 289         pf_state_host_ntoh(&sp->ext, &sk->ext);
 290
 291         pf_state_peer_ntoh(&sp->src, &st->src);
 292         pf_state_peer_ntoh(&sp->dst, &st->dst);
 293
 294         bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
 295         st->creation = time_second - ntohl(sp->creation);
 296         st->expire = ntohl(sp->expire) + time_second;
 297
 298         sk->af = sp->af;
 299         sk->proto = sp->proto;
 300         sk->direction = sp->direction;
 301         st->log = sp->log;
 302         st->timeout = sp->timeout;
 303         st->allow_opts = sp->allow_opts;
 304
 305         bcopy(sp->id, &st->id, sizeof(st->id));
 306         st->creatorid = sp->creatorid;
 307         st->sync_flags = PFSTATE_FROMSYNC;
 308
 309         if (pf_insert_state(kif, st)) {
 310                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 311                 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */
 312                 r->states--;
 313                 if (st->dst.scrub)
 314                         pool_put(&pf_state_scrub_pl, st->dst.scrub);
 315                 if (st->src.scrub)
 316                         pool_put(&pf_state_scrub_pl, st->src.scrub);
 317                 pool_put(&pf_state_pl, st);
 318                 return (EINVAL);
 319         }
 320
 321         return (0);
 322 }
 323
 324 void
 325 pfsync_input(struct mbuf *m, ...)
 326 {
 327         struct ip *ip = mtod(m, struct ip *);
 328         struct pfsync_header *ph;
 329         struct pfsync_softc *sc = pfsyncif;
 330         struct pf_state *st;
 331         struct pf_state_key *sk;
 332         struct pf_state_cmp id_key;
 333         struct pfsync_state *sp;
 334         struct pfsync_state_upd *up;
 335         struct pfsync_state_del *dp;
 336         struct pfsync_state_clr *cp;
 337         struct pfsync_state_upd_req *rup;
 338         struct pfsync_state_bus *bus;
 339 #ifdef IPSEC
 340         struct pfsync_tdb *pt;
 341 #endif
 342         struct in_addr src;
 343         struct mbuf *mp;
 344         int iplen, action, error, i, count, offp, sfail, stale = 0;
 345         u_int8_t chksum_flag = 0;
 346
 347         /* This function is not yet called from anywhere */
 348         /* Still we assume for safety that pf_token must be held */
 349         ASSERT_LWKT_TOKEN_HELD(&pf_token);
 350
 351         pfsyncstats.pfsyncs_ipackets++;
 352
 353         /* verify that we have a sync interface configured */
 354         if (!sc || !sc->sc_sync_ifp || !pf_status.running)
 355                 goto done;
 356
 357         /* verify that the packet came in on the right interface */
 358         if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) {
 359                 pfsyncstats.pfsyncs_badif++;
 360                 goto done;
 361         }
 362
 363         /* verify that the IP TTL is 255.  */
 364         if (ip->ip_ttl != PFSYNC_DFLTTL) {
 365                 pfsyncstats.pfsyncs_badttl++;
 366                 goto done;
 367         }
 368
 369         iplen = ip->ip_hl << 2;
 370
 371         if (m->m_pkthdr.len < iplen + sizeof(*ph)) {
 372                 pfsyncstats.pfsyncs_hdrops++;
 373                 goto done;
 374         }
 375
 376         if (iplen + sizeof(*ph) > m->m_len) {
 377                 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) {
 378                         pfsyncstats.pfsyncs_hdrops++;
 379                         goto done;
 380                 }
 381                 ip = mtod(m, struct ip *);
 382         }
 383         ph = (struct pfsync_header *)((char *)ip + iplen);
 384
 385         /* verify the version */
 386         if (ph->version != PFSYNC_VERSION) {
 387                 pfsyncstats.pfsyncs_badver++;
 388                 goto done;
 389         }
 390
 391         action = ph->action;
 392         count = ph->count;
 393
 394         /* make sure it's a valid action code */
 395         if (action >= PFSYNC_ACT_MAX) {
 396                 pfsyncstats.pfsyncs_badact++;
 397                 goto done;
 398         }
 399
 400         /* Cheaper to grab this now than having to mess with mbufs later */
 401         src = ip->ip_src;
 402
 403         if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
 404                 chksum_flag++;
 405
 406         switch (action) {
 407         case PFSYNC_ACT_CLR: {
 408                 struct pf_state *nexts;
 409                 struct pf_state_key *nextsk;
 410                 struct pfi_kif *kif;
 411                 u_int32_t creatorid;
 412                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 413                     sizeof(*cp), &offp)) == NULL) {
 414                         pfsyncstats.pfsyncs_badlen++;
 415                         return;
 416                 }
 417                 cp = (struct pfsync_state_clr *)(mp->m_data + offp);
 418                 creatorid = cp->creatorid;
 419
 420                 crit_enter();
 421                 if (cp->ifname[0] == '\0') {
 422                         for (st = RB_MIN(pf_state_tree_id, &tree_id);
 423                             st; st = nexts) {
 424                                 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st);
 425                                 if (st->creatorid == creatorid) {
 426                                         st->sync_flags |= PFSTATE_FROMSYNC;
 427                                         pf_unlink_state(st);
 428                                 }
 429                         }
 430                 } else {
 431                         if ((kif = pfi_kif_get(cp->ifname)) == NULL) {
 432                                 crit_exit();
 433                                 return;
 434                         }
 435                         for (sk = RB_MIN(pf_state_tree_lan_ext,
 436                             &pf_statetbl_lan_ext); sk; sk = nextsk) {
 437                                 nextsk = RB_NEXT(pf_state_tree_lan_ext,
 438                                     &pf_statetbl_lan_ext, sk);
 439                                 TAILQ_FOREACH(st, &sk->states, next) {
 440                                         if (st->creatorid == creatorid) {
 441                                                 st->sync_flags |=
 442                                                     PFSTATE_FROMSYNC;
 443                                                 pf_unlink_state(st);
 444                                         }
 445                                 }
 446                         }
 447                 }
 448                 crit_exit();
 449
 450                 break;
 451         }
 452         case PFSYNC_ACT_INS:
 453                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 454                     count * sizeof(*sp), &offp)) == NULL) {
 455                         pfsyncstats.pfsyncs_badlen++;
 456                         return;
 457                 }
 458
 459                 crit_enter();
 460                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 461                     i < count; i++, sp++) {
 462                         /* check for invalid values */
 463                         if (sp->timeout >= PFTM_MAX ||
 464                             sp->src.state > PF_TCPS_PROXY_DST ||
 465                             sp->dst.state > PF_TCPS_PROXY_DST ||
 466                             sp->direction > PF_OUT ||
 467                             (sp->af != AF_INET && sp->af != AF_INET6)) {
 468                                 if (pf_status.debug >= PF_DEBUG_MISC)
 469                                         kprintf("pfsync_insert: PFSYNC_ACT_INS: "
 470                                             "invalid value\n");
 471                                 pfsyncstats.pfsyncs_badstate++;
 472                                 continue;
 473                         }
 474
 475                         if ((error = pfsync_insert_net_state(sp,
 476                             chksum_flag))) {
 477                                 if (error == ENOMEM) {
 478                                         crit_exit();
 479                                         goto done;
 480                                 }
 481                                 continue;
 482                         }
 483                 }
 484                 crit_exit();
 485                 break;
 486         case PFSYNC_ACT_UPD:
 487                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 488                     count * sizeof(*sp), &offp)) == NULL) {
 489                         pfsyncstats.pfsyncs_badlen++;
 490                         return;
 491                 }
 492
 493                 crit_enter();
 494                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 495                     i < count; i++, sp++) {
 496                         int flags = PFSYNC_FLAG_STALE;
 497
 498                         /* check for invalid values */
 499                         if (sp->timeout >= PFTM_MAX ||
 500                             sp->src.state > PF_TCPS_PROXY_DST ||
 501                             sp->dst.state > PF_TCPS_PROXY_DST) {
 502                                 if (pf_status.debug >= PF_DEBUG_MISC)
 503                                         kprintf("pfsync_insert: PFSYNC_ACT_UPD: "
 504                                             "invalid value\n");
 505                                 pfsyncstats.pfsyncs_badstate++;
 506                                 continue;
 507                         }
 508
 509                         bcopy(sp->id, &id_key.id, sizeof(id_key.id));
 510                         id_key.creatorid = sp->creatorid;
 511
 512                         st = pf_find_state_byid(&id_key);
 513                         if (st == NULL) {
 514                                 /* insert the update */
 515                                 if (pfsync_insert_net_state(sp, chksum_flag))
 516                                         pfsyncstats.pfsyncs_badstate++;
 517                                 continue;
 518                         }
 519                         sk = st->state_key;
 520                         sfail = 0;
 521                         if (sk->proto == IPPROTO_TCP) {
 522                                 /*
 523                                  * The state should never go backwards except
 524                                  * for syn-proxy states.  Neither should the
 525                                  * sequence window slide backwards.
 526                                  */
 527                                 if (st->src.state > sp->src.state &&
 528                                     (st->src.state < PF_TCPS_PROXY_SRC ||
 529                                     sp->src.state >= PF_TCPS_PROXY_SRC))
 530                                         sfail = 1;
 531                                 else if (SEQ_GT(st->src.seqlo,
 532                                     ntohl(sp->src.seqlo)))
 533                                         sfail = 3;
 534                                 else if (st->dst.state > sp->dst.state) {
 535                                         /* There might still be useful
 536                                          * information about the src state here,
 537                                          * so import that part of the update,
 538                                          * then "fail" so we send the updated
 539                                          * state back to the peer who is missing
 540                                          * our what we know. */
 541                                         pf_state_peer_ntoh(&sp->src, &st->src);
 542                                         /* XXX do anything with timeouts? */
 543                                         sfail = 7;
 544                                         flags = 0;
 545                                 } else if (st->dst.state >= TCPS_SYN_SENT &&
 546                                     SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo)))
 547                                         sfail = 4;
 548                         } else {
 549                                 /*
 550                                  * Non-TCP protocol state machine always go
 551                                  * forwards
 552                                  */
 553                                 if (st->src.state > sp->src.state)
 554                                         sfail = 5;
 555                                 else if (st->dst.state > sp->dst.state)
 556                                         sfail = 6;
 557                         }
 558                         if (sfail) {
 559                                 if (pf_status.debug >= PF_DEBUG_MISC)
 560                                         kprintf("pfsync: %s stale update "
 561                                             "(%d) id: %016llx "
 562                                             "creatorid: %08x\n",
 563                                             (sfail < 7 ?  "ignoring"
 564                                              : "partial"), sfail,
 565                                             be64toh(st->id),
 566                                             ntohl(st->creatorid));
 567                                 pfsyncstats.pfsyncs_badstate++;
 568
 569                                 if (!(sp->sync_flags & PFSTATE_STALE)) {
 570                                         /* we have a better state, send it */
 571                                         if (sc->sc_mbuf != NULL && !stale)
 572                                                 pfsync_sendout(sc);
 573                                         stale++;
 574                                         if (!st->sync_flags)
 575                                                 pfsync_pack_state(
 576                                                     PFSYNC_ACT_UPD, st, flags);
 577                                 }
 578                                 continue;
 579                         }
 580                         pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
 581                         pf_state_peer_ntoh(&sp->src, &st->src);
 582                         pf_state_peer_ntoh(&sp->dst, &st->dst);
 583                         st->expire = ntohl(sp->expire) + time_second;
 584                         st->timeout = sp->timeout;
 585                 }
 586                 if (stale && sc->sc_mbuf != NULL)
 587                         pfsync_sendout(sc);
 588                 crit_exit();
 589                 break;
 590         /*
 591          * It's not strictly necessary for us to support the "uncompressed"
 592          * delete action, but it's relatively simple and maintains consistency.
 593          */
 594         case PFSYNC_ACT_DEL:
 595                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 596                     count * sizeof(*sp), &offp)) == NULL) {
 597                         pfsyncstats.pfsyncs_badlen++;
 598                         return;
 599                 }
 600
 601                 crit_enter();
 602                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 603                     i < count; i++, sp++) {
 604                         bcopy(sp->id, &id_key.id, sizeof(id_key.id));
 605                         id_key.creatorid = sp->creatorid;
 606
 607                         st = pf_find_state_byid(&id_key);
 608                         if (st == NULL) {
 609                                 pfsyncstats.pfsyncs_badstate++;
 610                                 continue;
 611                         }
 612                         st->sync_flags |= PFSTATE_FROMSYNC;
 613                         pf_unlink_state(st);
 614                 }
 615                 crit_exit();
 616                 break;
 617         case PFSYNC_ACT_UPD_C: {
 618                 int update_requested = 0;
 619
 620                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 621                     count * sizeof(*up), &offp)) == NULL) {
 622                         pfsyncstats.pfsyncs_badlen++;
 623                         return;
 624                 }
 625
 626                 crit_enter();
 627                 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp);
 628                     i < count; i++, up++) {
 629                         /* check for invalid values */
 630                         if (up->timeout >= PFTM_MAX ||
 631                             up->src.state > PF_TCPS_PROXY_DST ||
 632                             up->dst.state > PF_TCPS_PROXY_DST) {
 633                                 if (pf_status.debug >= PF_DEBUG_MISC)
 634                                         kprintf("pfsync_insert: "
 635                                             "PFSYNC_ACT_UPD_C: "
 636                                             "invalid value\n");
 637                                 pfsyncstats.pfsyncs_badstate++;
 638                                 continue;
 639                         }
 640
 641                         bcopy(up->id, &id_key.id, sizeof(id_key.id));
 642                         id_key.creatorid = up->creatorid;
 643
 644                         st = pf_find_state_byid(&id_key);
 645                         if (st == NULL) {
 646                                 /* We don't have this state. Ask for it. */
 647                                 error = pfsync_request_update(up, &src);
 648                                 if (error == ENOMEM) {
 649                                         crit_exit();
 650                                         goto done;
 651                                 }
 652                                 update_requested = 1;
 653                                 pfsyncstats.pfsyncs_badstate++;
 654                                 continue;
 655                         }
 656                         sk = st->state_key;
 657                         sfail = 0;
 658                         if (sk->proto == IPPROTO_TCP) {
 659                                 /*
 660                                  * The state should never go backwards except
 661                                  * for syn-proxy states.  Neither should the
 662                                  * sequence window slide backwards.
 663                                  */
 664                                 if (st->src.state > up->src.state &&
 665                                     (st->src.state < PF_TCPS_PROXY_SRC ||
 666                                     up->src.state >= PF_TCPS_PROXY_SRC))
 667                                         sfail = 1;
 668                                 else if (st->dst.state > up->dst.state)
 669                                         sfail = 2;
 670                                 else if (SEQ_GT(st->src.seqlo,
 671                                     ntohl(up->src.seqlo)))
 672                                         sfail = 3;
 673                                 else if (st->dst.state >= TCPS_SYN_SENT &&
 674                                     SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo)))
 675                                         sfail = 4;
 676                         } else {
 677                                 /*
 678                                  * Non-TCP protocol state machine always go
 679                                  * forwards
 680                                  */
 681                                 if (st->src.state > up->src.state)
 682                                         sfail = 5;
 683                                 else if (st->dst.state > up->dst.state)
 684                                         sfail = 6;
 685                         }
 686                         if (sfail) {
 687                                 if (pf_status.debug >= PF_DEBUG_MISC)
 688                                         kprintf("pfsync: ignoring stale update "
 689                                             "(%d) id: %016llx "
 690                                             "creatorid: %08x\n", sfail,
 691                                             be64toh(st->id),
 692                                             ntohl(st->creatorid));
 693                                 pfsyncstats.pfsyncs_badstate++;
 694
 695                                 /* we have a better state, send it out */
 696                                 if ((!stale || update_requested) &&
 697                                     sc->sc_mbuf != NULL) {
 698                                         pfsync_sendout(sc);
 699                                         update_requested = 0;
 700                                 }
 701                                 stale++;
 702                                 if (!st->sync_flags)
 703                                         pfsync_pack_state(PFSYNC_ACT_UPD, st,
 704                                             PFSYNC_FLAG_STALE);
 705                                 continue;
 706                         }
 707                         pfsync_alloc_scrub_memory(&up->dst, &st->dst);
 708                         pf_state_peer_ntoh(&up->src, &st->src);
 709                         pf_state_peer_ntoh(&up->dst, &st->dst);
 710                         st->expire = ntohl(up->expire) + time_second;
 711                         st->timeout = up->timeout;
 712                 }
 713                 if ((update_requested || stale) && sc->sc_mbuf)
 714                         pfsync_sendout(sc);
 715                 crit_exit();
 716                 break;
 717         }
 718         case PFSYNC_ACT_DEL_C:
 719                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 720                     count * sizeof(*dp), &offp)) == NULL) {
 721                         pfsyncstats.pfsyncs_badlen++;
 722                         return;
 723                 }
 724
 725                 crit_enter();
 726                 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp);
 727                     i < count; i++, dp++) {
 728                         bcopy(dp->id, &id_key.id, sizeof(id_key.id));
 729                         id_key.creatorid = dp->creatorid;
 730
 731                         st = pf_find_state_byid(&id_key);
 732                         if (st == NULL) {
 733                                 pfsyncstats.pfsyncs_badstate++;
 734                                 continue;
 735                         }
 736                         st->sync_flags |= PFSTATE_FROMSYNC;
 737                         pf_unlink_state(st);
 738                 }
 739                 crit_exit();
 740                 break;
 741         case PFSYNC_ACT_INS_F:
 742         case PFSYNC_ACT_DEL_F:
 743                 /* not implemented */
 744                 break;
 745         case PFSYNC_ACT_UREQ:
 746                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 747                     count * sizeof(*rup), &offp)) == NULL) {
 748                         pfsyncstats.pfsyncs_badlen++;
 749                         return;
 750                 }
 751
 752                 crit_enter();
 753                 if (sc->sc_mbuf != NULL)
 754                         pfsync_sendout(sc);
 755                 for (i = 0,
 756                     rup = (struct pfsync_state_upd_req *)(mp->m_data + offp);
 757                     i < count; i++, rup++) {
 758                         bcopy(rup->id, &id_key.id, sizeof(id_key.id));
 759                         id_key.creatorid = rup->creatorid;
 760
 761                         if (id_key.id == 0 && id_key.creatorid == 0) {
 762                                 sc->sc_ureq_received = mycpu->gd_time_seconds;
 763                                 if (sc->sc_bulk_send_next == NULL)
 764                                         sc->sc_bulk_send_next =
 765                                             TAILQ_FIRST(&state_list);
 766                                 sc->sc_bulk_terminator = sc->sc_bulk_send_next;
 767                                 if (pf_status.debug >= PF_DEBUG_MISC)
 768                                         kprintf("pfsync: received "
 769                                             "bulk update request\n");
 770                                 pfsync_send_bus(sc, PFSYNC_BUS_START);
 771                                 lwkt_reltoken(&pf_token);
 772                                 callout_reset(&sc->sc_bulk_tmo, 1 * hz,
 773                                     pfsync_bulk_update,
 774                                     LIST_FIRST(&pfsync_list));
 775                                 lwkt_gettoken(&pf_token);
 776                         } else {
 777                                 st = pf_find_state_byid(&id_key);
 778                                 if (st == NULL) {
 779                                         pfsyncstats.pfsyncs_badstate++;
 780                                         continue;
 781                                 }
 782                                 if (!st->sync_flags)
 783                                         pfsync_pack_state(PFSYNC_ACT_UPD,
 784                                             st, 0);
 785                         }
 786                 }
 787                 if (sc->sc_mbuf != NULL)
 788                         pfsync_sendout(sc);
 789                 crit_exit();
 790                 break;
 791         case PFSYNC_ACT_BUS:
 792                 /* If we're not waiting for a bulk update, who cares. */
 793                 if (sc->sc_ureq_sent == 0)
 794                         break;
 795
 796                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 797                     sizeof(*bus), &offp)) == NULL) {
 798                         pfsyncstats.pfsyncs_badlen++;
 799                         return;
 800                 }
 801                 bus = (struct pfsync_state_bus *)(mp->m_data + offp);
 802                 switch (bus->status) {
 803                 case PFSYNC_BUS_START:
 804                         lwkt_reltoken(&pf_token);
 805                         callout_reset(&sc->sc_bulkfail_tmo,
 806                             pf_pool_limits[PF_LIMIT_STATES].limit /
 807                             (PFSYNC_BULKPACKETS * sc->sc_maxcount),
 808                             pfsync_bulkfail, LIST_FIRST(&pfsync_list));
 809                         lwkt_gettoken(&pf_token);
 810                         if (pf_status.debug >= PF_DEBUG_MISC)
 811                                 kprintf("pfsync: received bulk "
 812                                     "update start\n");
 813                         break;
 814                 case PFSYNC_BUS_END:
 815                         if (mycpu->gd_time_seconds - ntohl(bus->endtime) >=
 816                             sc->sc_ureq_sent) {
 817                                 /* that's it, we're happy */
 818                                 sc->sc_ureq_sent = 0;
 819                                 sc->sc_bulk_tries = 0;
 820                                 lwkt_reltoken(&pf_token);
 821                                 callout_stop(&sc->sc_bulkfail_tmo);
 822                                 lwkt_gettoken(&pf_token);
 823 #if NCARP > 0
 824                                 if (!pfsync_sync_ok) {
 825                                         lwkt_reltoken(&pf_token);
 826                                         carp_group_demote_adj(&sc->sc_if, -1);
 827                                         lwkt_gettoken(&pf_token);
 828                                 }
 829 #endif
 830                                 pfsync_sync_ok = 1;
 831                                 if (pf_status.debug >= PF_DEBUG_MISC)
 832                                         kprintf("pfsync: received valid "
 833                                             "bulk update end\n");
 834                         } else {
 835                                 if (pf_status.debug >= PF_DEBUG_MISC)
 836                                         kprintf("pfsync: received invalid "
 837                                             "bulk update end: bad timestamp\n");
 838                         }
 839                         break;
 840                 }
 841                 break;
 842 #ifdef IPSEC
 843         case PFSYNC_ACT_TDB_UPD:
 844                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 845                     count * sizeof(*pt), &offp)) == NULL) {
 846                         pfsyncstats.pfsyncs_badlen++;
 847                         return;
 848                 }
 849                 crit_enter();
 850                 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp);
 851                     i < count; i++, pt++)
 852                         pfsync_update_net_tdb(pt);
 853                 crit_exit();
 854                 break;
 855 #endif
 856         }
 857
 858 done:
 859         if (m)
 860                 m_freem(m);
 861 }
 862
 863 int
 864 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
 865         struct rtentry *rt)
 866 {
 867         m_freem(m);
 868         return (0);
 869 }
 870
 871 /* ARGSUSED */
 872 int
 873 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
 874 {
 875         struct pfsync_softc *sc = ifp->if_softc;
 876         struct ifreq *ifr = (struct ifreq *)data;
 877         struct ip_moptions *imo = &sc->sc_imo;
 878         struct pfsyncreq pfsyncr;
 879         struct ifnet    *sifp;
 880         int error;
 881
 882         lwkt_gettoken(&pf_token);
 883
 884         switch (cmd) {
 885         case SIOCSIFADDR:
 886         case SIOCAIFADDR:
 887         case SIOCSIFDSTADDR:
 888         case SIOCSIFFLAGS:
 889                 if (ifp->if_flags & IFF_UP)
 890                         ifp->if_flags |= IFF_RUNNING;
 891                 else
 892                         ifp->if_flags &= ~IFF_RUNNING;
 893                 break;
 894         case SIOCSIFMTU:
 895                 if (ifr->ifr_mtu < PFSYNC_MINMTU) {
 896                         lwkt_reltoken(&pf_token);
 897                         return (EINVAL);
 898                 }
 899                 if (ifr->ifr_mtu > MCLBYTES)
 900                         ifr->ifr_mtu = MCLBYTES;
 901                 crit_enter();
 902                 if (ifr->ifr_mtu < ifp->if_mtu)
 903                         pfsync_sendout(sc);
 904                 pfsync_setmtu(sc, ifr->ifr_mtu);
 905                 crit_exit();
 906                 break;
 907         case SIOCGETPFSYNC:
 908                 bzero(&pfsyncr, sizeof(pfsyncr));
 909                 if (sc->sc_sync_ifp)
 910                         strlcpy(pfsyncr.pfsyncr_syncdev,
 911                             sc->sc_sync_ifp->if_xname, IFNAMSIZ);
 912                 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
 913                 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
 914                 lwkt_reltoken(&pf_token);
 915                 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))))
 916                         return (error);
 917                 lwkt_gettoken(&pf_token);
 918                 break;
 919         case SIOCSETPFSYNC:
 920                 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) {
 921                         lwkt_reltoken(&pf_token);
 922                         return (error);
 923                 }
 924                 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) {
 925                         lwkt_reltoken(&pf_token);
 926                         return (error);
 927                 }
 928
 929                 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
 930                         sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP;
 931                 else
 932                         sc->sc_sync_peer.s_addr =
 933                             pfsyncr.pfsyncr_syncpeer.s_addr;
 934
 935                 if (pfsyncr.pfsyncr_maxupdates > 255) {
 936                         lwkt_reltoken(&pf_token);
 937                         return (EINVAL);
 938                 }
 939                 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
 940
 941                 if (pfsyncr.pfsyncr_syncdev[0] == 0) {
 942                         sc->sc_sync_ifp = NULL;
 943                         if (sc->sc_mbuf_net != NULL) {
 944                                 /* Don't keep stale pfsync packets around. */
 945                                 crit_enter();
 946                                 m_freem(sc->sc_mbuf_net);
 947                                 sc->sc_mbuf_net = NULL;
 948                                 sc->sc_statep_net.s = NULL;
 949                                 crit_exit();
 950                         }
 951                         if (imo->imo_num_memberships > 0) {
 952                                 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
 953                                 imo->imo_multicast_ifp = NULL;
 954                         }
 955                         break;
 956                 }
 957
 958                 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) {
 959                         lwkt_reltoken(&pf_token);
 960                         return (EINVAL);
 961                 }
 962
 963                 crit_enter();
 964                 if (sifp->if_mtu < sc->sc_if.if_mtu ||
 965                     (sc->sc_sync_ifp != NULL &&
 966                     sifp->if_mtu < sc->sc_sync_ifp->if_mtu) ||
 967                     sifp->if_mtu < MCLBYTES - sizeof(struct ip))
 968                         pfsync_sendout(sc);
 969                 sc->sc_sync_ifp = sifp;
 970
 971                 pfsync_setmtu(sc, sc->sc_if.if_mtu);
 972
 973                 if (imo->imo_num_memberships > 0) {
 974                         in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
 975                         imo->imo_multicast_ifp = NULL;
 976                 }
 977
 978                 if (sc->sc_sync_ifp &&
 979                     sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
 980                         struct in_addr addr;
 981
 982                         if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) {
 983                                 sc->sc_sync_ifp = NULL;
 984                                 lwkt_reltoken(&pf_token);
 985                                 crit_exit();
 986                                 return (EADDRNOTAVAIL);
 987                         }
 988
 989                         addr.s_addr = INADDR_PFSYNC_GROUP;
 990
 991                         if ((imo->imo_membership[0] =
 992                             in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) {
 993                                 sc->sc_sync_ifp = NULL;
 994                                 lwkt_reltoken(&pf_token);
 995                                 crit_exit();
 996                                 return (ENOBUFS);
 997                         }
 998                         imo->imo_num_memberships++;
 999                         imo->imo_multicast_ifp = sc->sc_sync_ifp;
1000                         imo->imo_multicast_ttl = PFSYNC_DFLTTL;
1001                         imo->imo_multicast_loop = 0;
1002                 }
1003
1004                 if (sc->sc_sync_ifp ||
1005                     sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) {
1006                         /* Request a full state table update. */
1007                         sc->sc_ureq_sent = mycpu->gd_time_seconds;
1008 #if NCARP > 0
1009                         if (pfsync_sync_ok)
1010                                 carp_group_demote_adj(&sc->sc_if, 1);
1011 #endif
1012                         pfsync_sync_ok = 0;
1013                         if (pf_status.debug >= PF_DEBUG_MISC)
1014                                 kprintf("pfsync: requesting bulk update\n");
1015                         lwkt_reltoken(&pf_token);
1016                         callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
1017                             pfsync_bulkfail, LIST_FIRST(&pfsync_list));
1018                         lwkt_gettoken(&pf_token);
1019                         error = pfsync_request_update(NULL, NULL);
1020                         if (error == ENOMEM) {
1021                                 lwkt_reltoken(&pf_token);
1022                                 crit_exit();
1023                                 return (ENOMEM);
1024                         }
1025                         pfsync_sendout(sc);
1026                 }
1027                 crit_exit();
1028
1029                 break;
1030
1031         default:
1032                 lwkt_reltoken(&pf_token);
1033                 return (ENOTTY);
1034         }
1035
1036         lwkt_reltoken(&pf_token);
1037         return (0);
1038 }
1039
1040 void
1041 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req)
1042 {
1043         int mtu;
1044
1045         if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req)
1046                 mtu = sc->sc_sync_ifp->if_mtu;
1047         else
1048                 mtu = mtu_req;
1049
1050         sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) /
1051             sizeof(struct pfsync_state);
1052         if (sc->sc_maxcount > 254)
1053             sc->sc_maxcount = 254;
1054         sc->sc_if.if_mtu = sizeof(struct pfsync_header) +
1055             sc->sc_maxcount * sizeof(struct pfsync_state);
1056 }
1057
1058 struct mbuf *
1059 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp)
1060 {
1061         struct pfsync_header *h;
1062         struct mbuf *m;
1063         int len;
1064
1065         ASSERT_LWKT_TOKEN_HELD(&pf_token);
1066
1067         MGETHDR(m, M_WAITOK, MT_DATA);
1068         if (m == NULL) {
1069                 sc->sc_if.if_oerrors++;
1070                 return (NULL);
1071         }
1072
1073         switch (action) {
1074         case PFSYNC_ACT_CLR:
1075                 len = sizeof(struct pfsync_header) +
1076                     sizeof(struct pfsync_state_clr);
1077                 break;
1078         case PFSYNC_ACT_UPD_C:
1079                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) +
1080                     sizeof(struct pfsync_header);
1081                 break;
1082         case PFSYNC_ACT_DEL_C:
1083                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) +
1084                     sizeof(struct pfsync_header);
1085                 break;
1086         case PFSYNC_ACT_UREQ:
1087                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) +
1088                     sizeof(struct pfsync_header);
1089                 break;
1090         case PFSYNC_ACT_BUS:
1091                 len = sizeof(struct pfsync_header) +
1092                     sizeof(struct pfsync_state_bus);
1093                 break;
1094         case PFSYNC_ACT_TDB_UPD:
1095                 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) +
1096                     sizeof(struct pfsync_header);
1097                 break;
1098         default:
1099                 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) +
1100                     sizeof(struct pfsync_header);
1101                 break;
1102         }
1103
1104         if (len > MHLEN) {
1105                 MCLGET(m, M_WAITOK);
1106                 if ((m->m_flags & M_EXT) == 0) {
1107                         m_free(m);
1108                         sc->sc_if.if_oerrors++;
1109                         return (NULL);
1110                 }
1111                 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1);
1112         } else
1113                 MH_ALIGN(m, len);
1114
1115         m->m_pkthdr.rcvif = NULL;
1116         m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header);
1117         h = mtod(m, struct pfsync_header *);
1118         h->version = PFSYNC_VERSION;
1119         h->af = 0;
1120         h->count = 0;
1121         h->action = action;
1122
1123         *sp = (void *)((char *)h + PFSYNC_HDRLEN);
1124         lwkt_reltoken(&pf_token);
1125         callout_reset(&sc->sc_tmo, hz, pfsync_timeout,
1126             LIST_FIRST(&pfsync_list));
1127         lwkt_gettoken(&pf_token);
1128         return (m);
1129 }
1130
1131 int
1132 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags)
1133 {
1134         struct ifnet *ifp = NULL;
1135         struct pfsync_softc *sc = pfsyncif;
1136         struct pfsync_header *h, *h_net;
1137         struct pfsync_state *sp = NULL;
1138         struct pfsync_state_upd *up = NULL;
1139         struct pfsync_state_del *dp = NULL;
1140         struct pf_state_key *sk = st->state_key;
1141         struct pf_rule *r;
1142         u_long secs;
1143         int ret = 0;
1144         u_int8_t i = 255, newaction = 0;
1145
1146         if (sc == NULL)
1147                 return (0);
1148         ifp = &sc->sc_if;
1149
1150         /*
1151          * If a packet falls in the forest and there's nobody around to
1152          * hear, does it make a sound?
1153          */
1154         if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL &&
1155             sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
1156                 /* Don't leave any stale pfsync packets hanging around. */
1157                 if (sc->sc_mbuf != NULL) {
1158                         m_freem(sc->sc_mbuf);
1159                         sc->sc_mbuf = NULL;
1160                         sc->sc_statep.s = NULL;
1161                 }
1162                 return (0);
1163         }
1164
1165         if (action >= PFSYNC_ACT_MAX)
1166                 return (EINVAL);
1167
1168         crit_enter();
1169         if (sc->sc_mbuf == NULL) {
1170                 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1171                     (void *)&sc->sc_statep.s)) == NULL) {
1172                         crit_exit();
1173                         return (ENOMEM);
1174                 }
1175                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1176         } else {
1177                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1178                 if (h->action != action) {
1179                         pfsync_sendout(sc);
1180                         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1181                             (void *)&sc->sc_statep.s)) == NULL) {
1182                                 crit_exit();
1183                                 return (ENOMEM);
1184                         }
1185                         h = mtod(sc->sc_mbuf, struct pfsync_header *);
1186                 } else {
1187                         /*
1188                          * If it's an update, look in the packet to see if
1189                          * we already have an update for the state.
1190                          */
1191                         if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) {
1192                                 struct pfsync_state *usp =
1193                                     (void *)((char *)h + PFSYNC_HDRLEN);
1194
1195                                 for (i = 0; i < h->count; i++) {
1196                                         if (!memcmp(usp->id, &st->id,
1197                                             PFSYNC_ID_LEN) &&
1198                                             usp->creatorid == st->creatorid) {
1199                                                 sp = usp;
1200                                                 sp->updates++;
1201                                                 break;
1202                                         }
1203                                         usp++;
1204                                 }
1205                         }
1206                 }
1207         }
1208
1209         secs = time_second;
1210
1211         st->pfsync_time = mycpu->gd_time_seconds;;
1212
1213         if (sp == NULL) {
1214                 /* not a "duplicate" update */
1215                 i = 255;
1216                 sp = sc->sc_statep.s++;
1217                 sc->sc_mbuf->m_pkthdr.len =
1218                     sc->sc_mbuf->m_len += sizeof(struct pfsync_state);
1219                 h->count++;
1220                 bzero(sp, sizeof(*sp));
1221
1222                 bcopy(&st->id, sp->id, sizeof(sp->id));
1223                 sp->creatorid = st->creatorid;
1224
1225                 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname));
1226                 pf_state_host_hton(&sk->lan, &sp->lan);
1227                 pf_state_host_hton(&sk->gwy, &sp->gwy);
1228                 pf_state_host_hton(&sk->ext, &sp->ext);
1229
1230                 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr));
1231
1232                 sp->creation = htonl(secs - st->creation);
1233                 pf_state_counter_hton(st->packets[0], sp->packets[0]);
1234                 pf_state_counter_hton(st->packets[1], sp->packets[1]);
1235                 pf_state_counter_hton(st->bytes[0], sp->bytes[0]);
1236                 pf_state_counter_hton(st->bytes[1], sp->bytes[1]);
1237                 if ((r = st->rule.ptr) == NULL)
1238                         sp->rule = htonl(-1);
1239                 else
1240                         sp->rule = htonl(r->nr);
1241                 if ((r = st->anchor.ptr) == NULL)
1242                         sp->anchor = htonl(-1);
1243                 else
1244                         sp->anchor = htonl(r->nr);
1245                 sp->af = sk->af;
1246                 sp->proto = sk->proto;
1247                 sp->direction = sk->direction;
1248                 sp->log = st->log;
1249                 sp->allow_opts = st->allow_opts;
1250                 sp->timeout = st->timeout;
1251
1252                 if (flags & PFSYNC_FLAG_STALE)
1253                         sp->sync_flags |= PFSTATE_STALE;
1254         }
1255
1256         pf_state_peer_hton(&st->src, &sp->src);
1257         pf_state_peer_hton(&st->dst, &sp->dst);
1258
1259         if (st->expire <= secs)
1260                 sp->expire = htonl(0);
1261         else
1262                 sp->expire = htonl(st->expire - secs);
1263
1264         /* do we need to build "compressed" actions for network transfer? */
1265         if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) {
1266                 switch (action) {
1267                 case PFSYNC_ACT_UPD:
1268                         newaction = PFSYNC_ACT_UPD_C;
1269                         break;
1270                 case PFSYNC_ACT_DEL:
1271                         newaction = PFSYNC_ACT_DEL_C;
1272                         break;
1273                 default:
1274                         /* by default we just send the uncompressed states */
1275                         break;
1276                 }
1277         }
1278
1279         if (newaction) {
1280                 if (sc->sc_mbuf_net == NULL) {
1281                         if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction,
1282                             (void *)&sc->sc_statep_net.s)) == NULL) {
1283                                 crit_exit();
1284                                 return (ENOMEM);
1285                         }
1286                 }
1287                 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *);
1288
1289                 switch (newaction) {
1290                 case PFSYNC_ACT_UPD_C:
1291                         if (i != 255) {
1292                                 up = (void *)((char *)h_net +
1293                                     PFSYNC_HDRLEN + (i * sizeof(*up)));
1294                                 up->updates++;
1295                         } else {
1296                                 h_net->count++;
1297                                 sc->sc_mbuf_net->m_pkthdr.len =
1298                                     sc->sc_mbuf_net->m_len += sizeof(*up);
1299                                 up = sc->sc_statep_net.u++;
1300
1301                                 bzero(up, sizeof(*up));
1302                                 bcopy(&st->id, up->id, sizeof(up->id));
1303                                 up->creatorid = st->creatorid;
1304                         }
1305                         up->timeout = st->timeout;
1306                         up->expire = sp->expire;
1307                         up->src = sp->src;
1308                         up->dst = sp->dst;
1309                         break;
1310                 case PFSYNC_ACT_DEL_C:
1311                         sc->sc_mbuf_net->m_pkthdr.len =
1312                             sc->sc_mbuf_net->m_len += sizeof(*dp);
1313                         dp = sc->sc_statep_net.d++;
1314                         h_net->count++;
1315
1316                         bzero(dp, sizeof(*dp));
1317                         bcopy(&st->id, dp->id, sizeof(dp->id));
1318                         dp->creatorid = st->creatorid;
1319                         break;
1320                 }
1321         }
1322
1323         if (h->count == sc->sc_maxcount ||
1324             (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates)))
1325                 ret = pfsync_sendout(sc);
1326
1327         crit_exit();
1328         return (ret);
1329 }
1330
1331 int
1332 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src)
1333 {
1334         struct ifnet *ifp = NULL;
1335         struct pfsync_header *h;
1336         struct pfsync_softc *sc = pfsyncif;
1337         struct pfsync_state_upd_req *rup;
1338         int ret = 0;
1339
1340         if (sc == NULL)
1341                 return (0);
1342
1343         ifp = &sc->sc_if;
1344         if (sc->sc_mbuf == NULL) {
1345                 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1346                     (void *)&sc->sc_statep.s)) == NULL)
1347                         return (ENOMEM);
1348                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1349         } else {
1350                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1351                 if (h->action != PFSYNC_ACT_UREQ) {
1352                         pfsync_sendout(sc);
1353                         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1354                             (void *)&sc->sc_statep.s)) == NULL)
1355                                 return (ENOMEM);
1356                         h = mtod(sc->sc_mbuf, struct pfsync_header *);
1357                 }
1358         }
1359
1360         if (src != NULL)
1361                 sc->sc_sendaddr = *src;
1362         sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup);
1363         h->count++;
1364         rup = sc->sc_statep.r++;
1365         bzero(rup, sizeof(*rup));
1366         if (up != NULL) {
1367                 bcopy(up->id, rup->id, sizeof(rup->id));
1368                 rup->creatorid = up->creatorid;
1369         }
1370
1371         if (h->count == sc->sc_maxcount)
1372                 ret = pfsync_sendout(sc);
1373
1374         return (ret);
1375 }
1376
1377 int
1378 pfsync_clear_states(u_int32_t creatorid, char *ifname)
1379 {
1380         struct ifnet *ifp = NULL;
1381         struct pfsync_softc *sc = pfsyncif;
1382         struct pfsync_state_clr *cp;
1383         int ret;
1384
1385         if (sc == NULL)
1386                 return (0);
1387
1388         ifp = &sc->sc_if;
1389         crit_enter();
1390         if (sc->sc_mbuf != NULL)
1391                 pfsync_sendout(sc);
1392         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR,
1393             (void *)&sc->sc_statep.c)) == NULL) {
1394                 crit_exit();
1395                 return (ENOMEM);
1396         }
1397         sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp);
1398         cp = sc->sc_statep.c;
1399         cp->creatorid = creatorid;
1400         if (ifname != NULL)
1401                 strlcpy(cp->ifname, ifname, IFNAMSIZ);
1402
1403         ret = (pfsync_sendout(sc));
1404         crit_exit();
1405         return (ret);
1406 }
1407
1408 void
1409 pfsync_timeout(void *v)
1410 {
1411         struct pfsync_softc *sc = v;
1412
1413         crit_enter();
1414         pfsync_sendout(sc);
1415         crit_exit();
1416 }
1417
1418 void
1419 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status)
1420 {
1421         struct pfsync_state_bus *bus;
1422
1423         if (sc->sc_mbuf != NULL)
1424                 pfsync_sendout(sc);
1425
1426         if (pfsync_sync_ok &&
1427             (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS,
1428             (void *)&sc->sc_statep.b)) != NULL) {
1429                 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus);
1430                 bus = sc->sc_statep.b;
1431                 bus->creatorid = pf_status.hostid;
1432                 bus->status = status;
1433                 bus->endtime = htonl(mycpu->gd_time_seconds - sc->sc_ureq_received);
1434                 pfsync_sendout(sc);
1435         }
1436 }
1437
1438 void
1439 pfsync_bulk_update(void *v)
1440 {
1441         struct pfsync_softc *sc = v;
1442         int i = 0;
1443         struct pf_state *state;
1444
1445         ASSERT_LWKT_TOKEN_HELD(&pf_token);
1446
1447         crit_enter();
1448         if (sc->sc_mbuf != NULL)
1449                 pfsync_sendout(sc);
1450
1451         /*
1452          * Grab at most PFSYNC_BULKPACKETS worth of states which have not
1453          * been sent since the latest request was made.
1454          */
1455         state = sc->sc_bulk_send_next;
1456         if (state)
1457                 do {
1458                         /* send state update if syncable and not already sent */
1459                         if (!state->sync_flags
1460                             && state->timeout < PFTM_MAX
1461                             && state->pfsync_time <= sc->sc_ureq_received) {
1462                                 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0);
1463                                 i++;
1464                         }
1465
1466                         /* figure next state to send */
1467                         state = TAILQ_NEXT(state, entry_list);
1468
1469                         /* wrap to start of list if we hit the end */
1470                         if (!state)
1471                                 state = TAILQ_FIRST(&state_list);
1472                 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS &&
1473                     state != sc->sc_bulk_terminator);
1474
1475         if (!state || state == sc->sc_bulk_terminator) {
1476                 /* we're done */
1477                 pfsync_send_bus(sc, PFSYNC_BUS_END);
1478                 sc->sc_ureq_received = 0;
1479                 sc->sc_bulk_send_next = NULL;
1480                 sc->sc_bulk_terminator = NULL;
1481                 lwkt_reltoken(&pf_token);
1482                 callout_stop(&sc->sc_bulk_tmo);
1483                 lwkt_gettoken(&pf_token);
1484                 if (pf_status.debug >= PF_DEBUG_MISC)
1485                         kprintf("pfsync: bulk update complete\n");
1486         } else {
1487                 /* look again for more in a bit */
1488                 lwkt_reltoken(&pf_token);
1489                 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_timeout,
1490                             LIST_FIRST(&pfsync_list));
1491                 lwkt_gettoken(&pf_token);
1492                 sc->sc_bulk_send_next = state;
1493         }
1494         if (sc->sc_mbuf != NULL)
1495                 pfsync_sendout(sc);
1496         crit_exit();
1497 }
1498
1499 void
1500 pfsync_bulkfail(void *v)
1501 {
1502         struct pfsync_softc *sc = v;
1503         int error;
1504
1505         ASSERT_LWKT_TOKEN_HELD(&pf_token);
1506
1507         if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
1508                 /* Try again in a bit */
1509                 lwkt_reltoken(&pf_token);
1510                 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulkfail,
1511                     LIST_FIRST(&pfsync_list));
1512                 lwkt_gettoken(&pf_token);
1513                 crit_enter();
1514                 error = pfsync_request_update(NULL, NULL);
1515                 if (error == ENOMEM) {
1516                         if (pf_status.debug >= PF_DEBUG_MISC)
1517                                 kprintf("pfsync: cannot allocate mbufs for "
1518                                     "bulk update\n");
1519                 } else
1520                         pfsync_sendout(sc);
1521                 crit_exit();
1522         } else {
1523                 /* Pretend like the transfer was ok */
1524                 sc->sc_ureq_sent = 0;
1525                 sc->sc_bulk_tries = 0;
1526 #if NCARP > 0
1527                 if (!pfsync_sync_ok)
1528                         carp_group_demote_adj(&sc->sc_if, -1);
1529 #endif
1530                 pfsync_sync_ok = 1;
1531                 if (pf_status.debug >= PF_DEBUG_MISC)
1532                         kprintf("pfsync: failed to receive "
1533                             "bulk update status\n");
1534                 lwkt_reltoken(&pf_token);
1535                 callout_stop(&sc->sc_bulkfail_tmo);
1536                 lwkt_gettoken(&pf_token);
1537         }
1538 }
1539
1540 /* This must be called in splnet() */
1541 int
1542 pfsync_sendout(struct pfsync_softc *sc)
1543 {
1544 #if NBPFILTER > 0
1545         struct ifnet *ifp = &sc->sc_if;
1546 #endif
1547         struct mbuf *m;
1548
1549         ASSERT_LWKT_TOKEN_HELD(&pf_token);
1550
1551         lwkt_reltoken(&pf_token);
1552         callout_stop(&sc->sc_tmo);
1553         lwkt_gettoken(&pf_token);
1554
1555         if (sc->sc_mbuf == NULL)
1556                 return (0);
1557         m = sc->sc_mbuf;
1558         sc->sc_mbuf = NULL;
1559         sc->sc_statep.s = NULL;
1560
1561 #if NBPFILTER > 0
1562         if (ifp->if_bpf) {
1563                 lwkt_reltoken(&pf_token);
1564                 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1565                 lwkt_gettoken(&pf_token);
1566         }
1567 #endif
1568
1569         if (sc->sc_mbuf_net) {
1570                 m_freem(m);
1571                 m = sc->sc_mbuf_net;
1572                 sc->sc_mbuf_net = NULL;
1573                 sc->sc_statep_net.s = NULL;
1574         }
1575
1576         return pfsync_sendout_mbuf(sc, m);
1577 }
1578
1579 int
1580 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m)
1581 {
1582         struct sockaddr sa;
1583         struct ip *ip;
1584
1585         if (sc->sc_sync_ifp ||
1586             sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) {
1587                 M_PREPEND(m, sizeof(struct ip), M_WAITOK);
1588                 if (m == NULL) {
1589                         pfsyncstats.pfsyncs_onomem++;
1590                         return (0);
1591                 }
1592                 ip = mtod(m, struct ip *);
1593                 ip->ip_v = IPVERSION;
1594                 ip->ip_hl = sizeof(*ip) >> 2;
1595                 ip->ip_tos = IPTOS_LOWDELAY;
1596                 ip->ip_len = htons(m->m_pkthdr.len);
1597                 ip->ip_id = htons(ip_randomid());
1598                 ip->ip_off = htons(IP_DF);
1599                 ip->ip_ttl = PFSYNC_DFLTTL;
1600                 ip->ip_p = IPPROTO_PFSYNC;
1601                 ip->ip_sum = 0;
1602
1603                 bzero(&sa, sizeof(sa));
1604                 ip->ip_src.s_addr = INADDR_ANY;
1605
1606                 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP)
1607                         m->m_flags |= M_MCAST;
1608                 ip->ip_dst = sc->sc_sendaddr;
1609                 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr;
1610
1611                 pfsyncstats.pfsyncs_opackets++;
1612
1613                 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL))
1614                         pfsyncstats.pfsyncs_oerrors++;
1615         } else
1616                 m_freem(m);
1617
1618         return (0);
1619 }
1620
1621 static int
1622 pfsync_modevent(module_t mod, int type, void *data)
1623 {
1624         int error = 0;
1625
1626         lwkt_gettoken(&pf_token);
1627
1628         switch (type) {
1629         case MOD_LOAD:
1630                 LIST_INIT(&pfsync_list);
1631                 lwkt_reltoken(&pf_token);
1632                 if_clone_attach(&pfsync_cloner);
1633                 lwkt_gettoken(&pf_token);
1634                 break;
1635
1636         case MOD_UNLOAD:
1637                 lwkt_reltoken(&pf_token);
1638                 if_clone_detach(&pfsync_cloner);
1639                 lwkt_gettoken(&pf_token);
1640                 while (!LIST_EMPTY(&pfsync_list))
1641                         pfsync_clone_destroy(
1642                                 &LIST_FIRST(&pfsync_list)->sc_if);
1643                 break;
1644
1645         default:
1646                 error = EINVAL;
1647                 break;
1648         }
1649
1650         lwkt_reltoken(&pf_token);
1651         return error;
1652 }
1653
1654 static moduledata_t pfsync_mod = {
1655         "pfsync",
1656         pfsync_modevent,
1657         0
1658 };
1659
1660 #define PFSYNC_MODVER 1
1661
1662 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
1663 MODULE_VERSION(pfsync, PFSYNC_MODVER);
1664
1665
1666