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
  30 #include <sys/param.h>
  31 #include <sys/proc.h>
  32 #include <sys/systm.h>
  33 #include <sys/time.h>
  34 #include <sys/mbuf.h>
  35 #include <sys/socket.h>
  36 #include <sys/ioctl.h>
  37 #include <sys/timeout.h>
  38 #include <sys/kernel.h>
  39
  40 #include <net/if.h>
  41 #include <net/if_types.h>
  42 #include <net/route.h>
  43 #include <net/bpf.h>
  44 #include <netinet/in.h>
  45 #include <netinet/if_ether.h>
  46 #include <netinet/tcp.h>
  47 #include <netinet/tcp_seq.h>
  48
  49 #ifdef  INET
  50 #include <netinet/in_systm.h>
  51 #include <netinet/in_var.h>
  52 #include <netinet/ip.h>
  53 #include <netinet/ip_var.h>
  54 #endif
  55
  56 #ifdef INET6
  57 #include <netinet6/nd6.h>
  58 #endif /* INET6 */
  59
  60 #include "carp.h"
  61 #if NCARP > 0
  62 #include <netinet/ip_carp.h>
  63 #endif
  64
  65 #include <net/pfvar.h>
  66 #include <net/if_pfsync.h>
  67
  68 #include "bpfilter.h"
  69 #include "pfsync.h"
  70
  71 #define PFSYNC_MINMTU   \
  72     (sizeof(struct pfsync_header) + sizeof(struct pf_state))
  73
  74 #ifdef PFSYNCDEBUG
  75 #define DPRINTF(x)    do { if (pfsyncdebug) printf x ; } while (0)
  76 int pfsyncdebug;
  77 #else
  78 #define DPRINTF(x)
  79 #endif
  80
  81 struct pfsync_softc     *pfsyncif = NULL;
  82 struct pfsyncstats       pfsyncstats;
  83
  84 void    pfsyncattach(int);
  85 int     pfsync_clone_create(struct if_clone *, int);
  86 int     pfsync_clone_destroy(struct ifnet *);
  87 void    pfsync_setmtu(struct pfsync_softc *, int);
  88 int     pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
  89             struct pf_state_peer *);
  90 int     pfsync_insert_net_state(struct pfsync_state *, u_int8_t);
  91 void    pfsync_update_net_tdb(struct pfsync_tdb *);
  92 int     pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
  93             struct rtentry *);
  94 int     pfsyncioctl(struct ifnet *, u_long, caddr_t);
  95 void    pfsyncstart(struct ifnet *);
  96
  97 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **);
  98 int     pfsync_request_update(struct pfsync_state_upd *, struct in_addr *);
  99 int     pfsync_sendout(struct pfsync_softc *);
 100 int     pfsync_tdb_sendout(struct pfsync_softc *);
 101 int     pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *);
 102 void    pfsync_timeout(void *);
 103 void    pfsync_tdb_timeout(void *);
 104 void    pfsync_send_bus(struct pfsync_softc *, u_int8_t);
 105 void    pfsync_bulk_update(void *);
 106 void    pfsync_bulkfail(void *);
 107
 108 int     pfsync_sync_ok;
 109 extern int ifqmaxlen;
 110
 111 struct if_clone pfsync_cloner =
 112     IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy);
 113
 114 void
 115 pfsyncattach(int npfsync)
 116 {
 117         if_clone_attach(&pfsync_cloner);
 118 }
 119 int
 120 pfsync_clone_create(struct if_clone *ifc, int unit)
 121 {
 122         struct ifnet *ifp;
 123
 124         if (unit != 0)
 125                 return (EINVAL);
 126
 127         pfsync_sync_ok = 1;
 128         if ((pfsyncif = malloc(sizeof(*pfsyncif), M_DEVBUF, M_WAITOK)) == NULL)
 129                 return (ENOMEM);
 130         bzero(pfsyncif, sizeof(*pfsyncif));
 131         pfsyncif->sc_mbuf = NULL;
 132         pfsyncif->sc_mbuf_net = NULL;
 133         pfsyncif->sc_mbuf_tdb = NULL;
 134         pfsyncif->sc_statep.s = NULL;
 135         pfsyncif->sc_statep_net.s = NULL;
 136         pfsyncif->sc_statep_tdb.t = NULL;
 137         pfsyncif->sc_maxupdates = 128;
 138         pfsyncif->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP;
 139         pfsyncif->sc_sendaddr.s_addr = INADDR_PFSYNC_GROUP;
 140         pfsyncif->sc_ureq_received = 0;
 141         pfsyncif->sc_ureq_sent = 0;
 142         pfsyncif->sc_bulk_send_next = NULL;
 143         pfsyncif->sc_bulk_terminator = NULL;
 144         ifp = &pfsyncif->sc_if;
 145         snprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit);
 146         ifp->if_softc = pfsyncif;
 147         ifp->if_ioctl = pfsyncioctl;
 148         ifp->if_output = pfsyncoutput;
 149         ifp->if_start = pfsyncstart;
 150         ifp->if_type = IFT_PFSYNC;
 151         ifp->if_snd.ifq_maxlen = ifqmaxlen;
 152         ifp->if_hdrlen = PFSYNC_HDRLEN;
 153         pfsync_setmtu(pfsyncif, ETHERMTU);
 154         timeout_set(&pfsyncif->sc_tmo, pfsync_timeout, pfsyncif);
 155         timeout_set(&pfsyncif->sc_tdb_tmo, pfsync_tdb_timeout, pfsyncif);
 156         timeout_set(&pfsyncif->sc_bulk_tmo, pfsync_bulk_update, pfsyncif);
 157         timeout_set(&pfsyncif->sc_bulkfail_tmo, pfsync_bulkfail, pfsyncif);
 158         if_attach(ifp);
 159         if_alloc_sadl(ifp);
 160
 161 #if NCARP > 0
 162         if_addgroup(ifp, "carp");
 163 #endif
 164
 165 #if NBPFILTER > 0
 166         bpfattach(&pfsyncif->sc_if.if_bpf, ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
 167 #endif
 168
 169         return (0);
 170 }
 171
 172 int
 173 pfsync_clone_destroy(struct ifnet *ifp)
 174 {
 175 #if NBPFILTER > 0
 176         bpfdetach(ifp);
 177 #endif
 178         if_detach(ifp);
 179         free(pfsyncif, M_DEVBUF);
 180         pfsyncif = NULL;
 181         return (0);
 182 }
 183
 184 /*
 185  * Start output on the pfsync interface.
 186  */
 187 void
 188 pfsyncstart(struct ifnet *ifp)
 189 {
 190         struct mbuf *m;
 191         int s;
 192
 193         for (;;) {
 194                 s = splnet();
 195                 IF_DROP(&ifp->if_snd);
 196                 IF_DEQUEUE(&ifp->if_snd, m);
 197                 splx(s);
 198
 199                 if (m == NULL)
 200                         return;
 201                 else
 202                         m_freem(m);
 203         }
 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_rule *r = NULL;
 225         struct pfi_kif  *kif;
 226
 227         if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) {
 228                 printf("pfsync_insert_net_state: invalid creator id:"
 229                     " %08x\n", ntohl(sp->creatorid));
 230                 return (EINVAL);
 231         }
 232
 233         kif = pfi_kif_get(sp->ifname);
 234         if (kif == NULL) {
 235                 if (pf_status.debug >= PF_DEBUG_MISC)
 236                         printf("pfsync_insert_net_state: "
 237                             "unknown interface: %s\n", sp->ifname);
 238                 /* skip this state */
 239                 return (0);
 240         }
 241
 242         /*
 243          * If the ruleset checksums match, it's safe to associate the state
 244          * with the rule of that number.
 245          */
 246         if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag)
 247                 r = pf_main_ruleset.rules[
 248                     PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
 249         else
 250                 r = &pf_default_rule;
 251
 252         if (!r->max_states || r->states < r->max_states)
 253                 st = pool_get(&pf_state_pl, PR_NOWAIT);
 254         if (st == NULL) {
 255                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 256                 return (ENOMEM);
 257         }
 258         bzero(st, sizeof(*st));
 259
 260         /* allocate memory for scrub info */
 261         if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
 262             pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) {
 263                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 264                 if (st->src.scrub)
 265                         pool_put(&pf_state_scrub_pl, st->src.scrub);
 266                 pool_put(&pf_state_pl, st);
 267                 return (ENOMEM);
 268         }
 269
 270         st->rule.ptr = r;
 271         /* XXX get pointers to nat_rule and anchor */
 272
 273         /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
 274         r->states++;
 275
 276         /* fill in the rest of the state entry */
 277         pf_state_host_ntoh(&sp->lan, &st->lan);
 278         pf_state_host_ntoh(&sp->gwy, &st->gwy);
 279         pf_state_host_ntoh(&sp->ext, &st->ext);
 280
 281         pf_state_peer_ntoh(&sp->src, &st->src);
 282         pf_state_peer_ntoh(&sp->dst, &st->dst);
 283
 284         bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
 285         st->creation = time_second - ntohl(sp->creation);
 286         st->expire = ntohl(sp->expire) + time_second;
 287
 288         st->af = sp->af;
 289         st->proto = sp->proto;
 290         st->direction = sp->direction;
 291         st->log = sp->log;
 292         st->timeout = sp->timeout;
 293         st->allow_opts = sp->allow_opts;
 294
 295         bcopy(sp->id, &st->id, sizeof(st->id));
 296         st->creatorid = sp->creatorid;
 297         st->sync_flags = PFSTATE_FROMSYNC;
 298
 299         if (pf_insert_state(kif, st)) {
 300                 pfi_kif_unref(kif, PFI_KIF_REF_NONE);
 301                 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */
 302                 r->states--;
 303                 if (st->dst.scrub)
 304                         pool_put(&pf_state_scrub_pl, st->dst.scrub);
 305                 if (st->src.scrub)
 306                         pool_put(&pf_state_scrub_pl, st->src.scrub);
 307                 pool_put(&pf_state_pl, st);
 308                 return (EINVAL);
 309         }
 310
 311         return (0);
 312 }
 313
 314 void
 315 pfsync_input(struct mbuf *m, ...)
 316 {
 317         struct ip *ip = mtod(m, struct ip *);
 318         struct pfsync_header *ph;
 319         struct pfsync_softc *sc = pfsyncif;
 320         struct pf_state *st;
 321         struct pf_state_cmp key;
 322         struct pfsync_state *sp;
 323         struct pfsync_state_upd *up;
 324         struct pfsync_state_del *dp;
 325         struct pfsync_state_clr *cp;
 326         struct pfsync_state_upd_req *rup;
 327         struct pfsync_state_bus *bus;
 328         struct pfsync_tdb *pt;
 329         struct in_addr src;
 330         struct mbuf *mp;
 331         int iplen, action, error, i, s, count, offp, sfail, stale = 0;
 332         u_int8_t chksum_flag = 0;
 333
 334         pfsyncstats.pfsyncs_ipackets++;
 335
 336         /* verify that we have a sync interface configured */
 337         if (!sc || !sc->sc_sync_ifp || !pf_status.running)
 338                 goto done;
 339
 340         /* verify that the packet came in on the right interface */
 341         if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) {
 342                 pfsyncstats.pfsyncs_badif++;
 343                 goto done;
 344         }
 345
 346         /* verify that the IP TTL is 255.  */
 347         if (ip->ip_ttl != PFSYNC_DFLTTL) {
 348                 pfsyncstats.pfsyncs_badttl++;
 349                 goto done;
 350         }
 351
 352         iplen = ip->ip_hl << 2;
 353
 354         if (m->m_pkthdr.len < iplen + sizeof(*ph)) {
 355                 pfsyncstats.pfsyncs_hdrops++;
 356                 goto done;
 357         }
 358
 359         if (iplen + sizeof(*ph) > m->m_len) {
 360                 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) {
 361                         pfsyncstats.pfsyncs_hdrops++;
 362                         goto done;
 363                 }
 364                 ip = mtod(m, struct ip *);
 365         }
 366         ph = (struct pfsync_header *)((char *)ip + iplen);
 367
 368         /* verify the version */
 369         if (ph->version != PFSYNC_VERSION) {
 370                 pfsyncstats.pfsyncs_badver++;
 371                 goto done;
 372         }
 373
 374         action = ph->action;
 375         count = ph->count;
 376
 377         /* make sure it's a valid action code */
 378         if (action >= PFSYNC_ACT_MAX) {
 379                 pfsyncstats.pfsyncs_badact++;
 380                 goto done;
 381         }
 382
 383         /* Cheaper to grab this now than having to mess with mbufs later */
 384         src = ip->ip_src;
 385
 386         if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
 387                 chksum_flag++;
 388
 389         switch (action) {
 390         case PFSYNC_ACT_CLR: {
 391                 struct pf_state *nexts;
 392                 struct pfi_kif  *kif;
 393                 u_int32_t creatorid;
 394                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 395                     sizeof(*cp), &offp)) == NULL) {
 396                         pfsyncstats.pfsyncs_badlen++;
 397                         return;
 398                 }
 399                 cp = (struct pfsync_state_clr *)(mp->m_data + offp);
 400                 creatorid = cp->creatorid;
 401
 402                 s = splsoftnet();
 403                 if (cp->ifname[0] == '\0') {
 404                         for (st = RB_MIN(pf_state_tree_id, &tree_id);
 405                             st; st = nexts) {
 406                                 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st);
 407                                 if (st->creatorid == creatorid) {
 408                                         st->sync_flags |= PFSTATE_FROMSYNC;
 409                                         pf_unlink_state(st);
 410                                 }
 411                         }
 412                 } else {
 413                         if ((kif = pfi_kif_get(cp->ifname)) == NULL) {
 414                                 splx(s);
 415                                 return;
 416                         }
 417                         for (st = RB_MIN(pf_state_tree_lan_ext,
 418                             &kif->pfik_lan_ext); st; st = nexts) {
 419                                 nexts = RB_NEXT(pf_state_tree_lan_ext,
 420                                     &kif->pfik_lan_ext, st);
 421                                 if (st->creatorid == creatorid) {
 422                                         st->sync_flags |= PFSTATE_FROMSYNC;
 423                                         pf_unlink_state(st);
 424                                 }
 425                         }
 426                 }
 427                 splx(s);
 428
 429                 break;
 430         }
 431         case PFSYNC_ACT_INS:
 432                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 433                     count * sizeof(*sp), &offp)) == NULL) {
 434                         pfsyncstats.pfsyncs_badlen++;
 435                         return;
 436                 }
 437
 438                 s = splsoftnet();
 439                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 440                     i < count; i++, sp++) {
 441                         /* check for invalid values */
 442                         if (sp->timeout >= PFTM_MAX ||
 443                             sp->src.state > PF_TCPS_PROXY_DST ||
 444                             sp->dst.state > PF_TCPS_PROXY_DST ||
 445                             sp->direction > PF_OUT ||
 446                             (sp->af != AF_INET && sp->af != AF_INET6)) {
 447                                 if (pf_status.debug >= PF_DEBUG_MISC)
 448                                         printf("pfsync_insert: PFSYNC_ACT_INS: "
 449                                             "invalid value\n");
 450                                 pfsyncstats.pfsyncs_badstate++;
 451                                 continue;
 452                         }
 453
 454                         if ((error = pfsync_insert_net_state(sp,
 455                             chksum_flag))) {
 456                                 if (error == ENOMEM) {
 457                                         splx(s);
 458                                         goto done;
 459                                 }
 460                                 continue;
 461                         }
 462                 }
 463                 splx(s);
 464                 break;
 465         case PFSYNC_ACT_UPD:
 466                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 467                     count * sizeof(*sp), &offp)) == NULL) {
 468                         pfsyncstats.pfsyncs_badlen++;
 469                         return;
 470                 }
 471
 472                 s = splsoftnet();
 473                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 474                     i < count; i++, sp++) {
 475                         int flags = PFSYNC_FLAG_STALE;
 476
 477                         /* check for invalid values */
 478                         if (sp->timeout >= PFTM_MAX ||
 479                             sp->src.state > PF_TCPS_PROXY_DST ||
 480                             sp->dst.state > PF_TCPS_PROXY_DST) {
 481                                 if (pf_status.debug >= PF_DEBUG_MISC)
 482                                         printf("pfsync_insert: PFSYNC_ACT_UPD: "
 483                                             "invalid value\n");
 484                                 pfsyncstats.pfsyncs_badstate++;
 485                                 continue;
 486                         }
 487
 488                         bcopy(sp->id, &key.id, sizeof(key.id));
 489                         key.creatorid = sp->creatorid;
 490
 491                         st = pf_find_state_byid(&key);
 492                         if (st == NULL) {
 493                                 /* insert the update */
 494                                 if (pfsync_insert_net_state(sp, chksum_flag))
 495                                         pfsyncstats.pfsyncs_badstate++;
 496                                 continue;
 497                         }
 498                         sfail = 0;
 499                         if (st->proto == IPPROTO_TCP) {
 500                                 /*
 501                                  * The state should never go backwards except
 502                                  * for syn-proxy states.  Neither should the
 503                                  * sequence window slide backwards.
 504                                  */
 505                                 if (st->src.state > sp->src.state &&
 506                                     (st->src.state < PF_TCPS_PROXY_SRC ||
 507                                     sp->src.state >= PF_TCPS_PROXY_SRC))
 508                                         sfail = 1;
 509                                 else if (SEQ_GT(st->src.seqlo,
 510                                     ntohl(sp->src.seqlo)))
 511                                         sfail = 3;
 512                                 else if (st->dst.state > sp->dst.state) {
 513                                         /* There might still be useful
 514                                          * information about the src state here,
 515                                          * so import that part of the update,
 516                                          * then "fail" so we send the updated
 517                                          * state back to the peer who is missing
 518                                          * our what we know. */
 519                                         pf_state_peer_ntoh(&sp->src, &st->src);
 520                                         /* XXX do anything with timeouts? */
 521                                         sfail = 7;
 522                                         flags = 0;
 523                                 } else if (st->dst.state >= TCPS_SYN_SENT &&
 524                                     SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo)))
 525                                         sfail = 4;
 526                         } else {
 527                                 /*
 528                                  * Non-TCP protocol state machine always go
 529                                  * forwards
 530                                  */
 531                                 if (st->src.state > sp->src.state)
 532                                         sfail = 5;
 533                                 else if (st->dst.state > sp->dst.state)
 534                                         sfail = 6;
 535                         }
 536                         if (sfail) {
 537                                 if (pf_status.debug >= PF_DEBUG_MISC)
 538                                         printf("pfsync: %s stale update "
 539                                             "(%d) id: %016llx "
 540                                             "creatorid: %08x\n",
 541                                             (sfail < 7 ?  "ignoring"
 542                                              : "partial"), sfail,
 543                                             betoh64(st->id),
 544                                             ntohl(st->creatorid));
 545                                 pfsyncstats.pfsyncs_badstate++;
 546
 547                                 if (!(sp->sync_flags & PFSTATE_STALE)) {
 548                                         /* we have a better state, send it */
 549                                         if (sc->sc_mbuf != NULL && !stale)
 550                                                 pfsync_sendout(sc);
 551                                         stale++;
 552                                         if (!st->sync_flags)
 553                                                 pfsync_pack_state(
 554                                                     PFSYNC_ACT_UPD, st, flags);
 555                                 }
 556                                 continue;
 557                         }
 558                         pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
 559                         pf_state_peer_ntoh(&sp->src, &st->src);
 560                         pf_state_peer_ntoh(&sp->dst, &st->dst);
 561                         st->expire = ntohl(sp->expire) + time_second;
 562                         st->timeout = sp->timeout;
 563                 }
 564                 if (stale && sc->sc_mbuf != NULL)
 565                         pfsync_sendout(sc);
 566                 splx(s);
 567                 break;
 568         /*
 569          * It's not strictly necessary for us to support the "uncompressed"
 570          * delete action, but it's relatively simple and maintains consistency.
 571          */
 572         case PFSYNC_ACT_DEL:
 573                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 574                     count * sizeof(*sp), &offp)) == NULL) {
 575                         pfsyncstats.pfsyncs_badlen++;
 576                         return;
 577                 }
 578
 579                 s = splsoftnet();
 580                 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
 581                     i < count; i++, sp++) {
 582                         bcopy(sp->id, &key.id, sizeof(key.id));
 583                         key.creatorid = sp->creatorid;
 584
 585                         st = pf_find_state_byid(&key);
 586                         if (st == NULL) {
 587                                 pfsyncstats.pfsyncs_badstate++;
 588                                 continue;
 589                         }
 590                         st->sync_flags |= PFSTATE_FROMSYNC;
 591                         pf_unlink_state(st);
 592                 }
 593                 splx(s);
 594                 break;
 595         case PFSYNC_ACT_UPD_C: {
 596                 int update_requested = 0;
 597
 598                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 599                     count * sizeof(*up), &offp)) == NULL) {
 600                         pfsyncstats.pfsyncs_badlen++;
 601                         return;
 602                 }
 603
 604                 s = splsoftnet();
 605                 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp);
 606                     i < count; i++, up++) {
 607                         /* check for invalid values */
 608                         if (up->timeout >= PFTM_MAX ||
 609                             up->src.state > PF_TCPS_PROXY_DST ||
 610                             up->dst.state > PF_TCPS_PROXY_DST) {
 611                                 if (pf_status.debug >= PF_DEBUG_MISC)
 612                                         printf("pfsync_insert: "
 613                                             "PFSYNC_ACT_UPD_C: "
 614                                             "invalid value\n");
 615                                 pfsyncstats.pfsyncs_badstate++;
 616                                 continue;
 617                         }
 618
 619                         bcopy(up->id, &key.id, sizeof(key.id));
 620                         key.creatorid = up->creatorid;
 621
 622                         st = pf_find_state_byid(&key);
 623                         if (st == NULL) {
 624                                 /* We don't have this state. Ask for it. */
 625                                 error = pfsync_request_update(up, &src);
 626                                 if (error == ENOMEM) {
 627                                         splx(s);
 628                                         goto done;
 629                                 }
 630                                 update_requested = 1;
 631                                 pfsyncstats.pfsyncs_badstate++;
 632                                 continue;
 633                         }
 634                         sfail = 0;
 635                         if (st->proto == IPPROTO_TCP) {
 636                                 /*
 637                                  * The state should never go backwards except
 638                                  * for syn-proxy states.  Neither should the
 639                                  * sequence window slide backwards.
 640                                  */
 641                                 if (st->src.state > up->src.state &&
 642                                     (st->src.state < PF_TCPS_PROXY_SRC ||
 643                                     up->src.state >= PF_TCPS_PROXY_SRC))
 644                                         sfail = 1;
 645                                 else if (st->dst.state > up->dst.state)
 646                                         sfail = 2;
 647                                 else if (SEQ_GT(st->src.seqlo,
 648                                     ntohl(up->src.seqlo)))
 649                                         sfail = 3;
 650                                 else if (st->dst.state >= TCPS_SYN_SENT &&
 651                                     SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo)))
 652                                         sfail = 4;
 653                         } else {
 654                                 /*
 655                                  * Non-TCP protocol state machine always go
 656                                  * forwards
 657                                  */
 658                                 if (st->src.state > up->src.state)
 659                                         sfail = 5;
 660                                 else if (st->dst.state > up->dst.state)
 661                                         sfail = 6;
 662                         }
 663                         if (sfail) {
 664                                 if (pf_status.debug >= PF_DEBUG_MISC)
 665                                         printf("pfsync: ignoring stale update "
 666                                             "(%d) id: %016llx "
 667                                             "creatorid: %08x\n", sfail,
 668                                             betoh64(st->id),
 669                                             ntohl(st->creatorid));
 670                                 pfsyncstats.pfsyncs_badstate++;
 671
 672                                 /* we have a better state, send it out */
 673                                 if ((!stale || update_requested) &&
 674                                     sc->sc_mbuf != NULL) {
 675                                         pfsync_sendout(sc);
 676                                         update_requested = 0;
 677                                 }
 678                                 stale++;
 679                                 if (!st->sync_flags)
 680                                         pfsync_pack_state(PFSYNC_ACT_UPD, st,
 681                                             PFSYNC_FLAG_STALE);
 682                                 continue;
 683                         }
 684                         pfsync_alloc_scrub_memory(&up->dst, &st->dst);
 685                         pf_state_peer_ntoh(&up->src, &st->src);
 686                         pf_state_peer_ntoh(&up->dst, &st->dst);
 687                         st->expire = ntohl(up->expire) + time_second;
 688                         st->timeout = up->timeout;
 689                 }
 690                 if ((update_requested || stale) && sc->sc_mbuf)
 691                         pfsync_sendout(sc);
 692                 splx(s);
 693                 break;
 694         }
 695         case PFSYNC_ACT_DEL_C:
 696                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 697                     count * sizeof(*dp), &offp)) == NULL) {
 698                         pfsyncstats.pfsyncs_badlen++;
 699                         return;
 700                 }
 701
 702                 s = splsoftnet();
 703                 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp);
 704                     i < count; i++, dp++) {
 705                         bcopy(dp->id, &key.id, sizeof(key.id));
 706                         key.creatorid = dp->creatorid;
 707
 708                         st = pf_find_state_byid(&key);
 709                         if (st == NULL) {
 710                                 pfsyncstats.pfsyncs_badstate++;
 711                                 continue;
 712                         }
 713                         st->sync_flags |= PFSTATE_FROMSYNC;
 714                         pf_unlink_state(st);
 715                 }
 716                 splx(s);
 717                 break;
 718         case PFSYNC_ACT_INS_F:
 719         case PFSYNC_ACT_DEL_F:
 720                 /* not implemented */
 721                 break;
 722         case PFSYNC_ACT_UREQ:
 723                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 724                     count * sizeof(*rup), &offp)) == NULL) {
 725                         pfsyncstats.pfsyncs_badlen++;
 726                         return;
 727                 }
 728
 729                 s = splsoftnet();
 730                 if (sc->sc_mbuf != NULL)
 731                         pfsync_sendout(sc);
 732                 for (i = 0,
 733                     rup = (struct pfsync_state_upd_req *)(mp->m_data + offp);
 734                     i < count; i++, rup++) {
 735                         bcopy(rup->id, &key.id, sizeof(key.id));
 736                         key.creatorid = rup->creatorid;
 737
 738                         if (key.id == 0 && key.creatorid == 0) {
 739                                 sc->sc_ureq_received = time_uptime;
 740                                 if (sc->sc_bulk_send_next == NULL)
 741                                         sc->sc_bulk_send_next =
 742                                             TAILQ_FIRST(&state_list);
 743                                 sc->sc_bulk_terminator = sc->sc_bulk_send_next;
 744                                 if (pf_status.debug >= PF_DEBUG_MISC)
 745                                         printf("pfsync: received "
 746                                             "bulk update request\n");
 747                                 pfsync_send_bus(sc, PFSYNC_BUS_START);
 748                                 timeout_add(&sc->sc_bulk_tmo, 1 * hz);
 749                         } else {
 750                                 st = pf_find_state_byid(&key);
 751                                 if (st == NULL) {
 752                                         pfsyncstats.pfsyncs_badstate++;
 753                                         continue;
 754                                 }
 755                                 if (!st->sync_flags)
 756                                         pfsync_pack_state(PFSYNC_ACT_UPD,
 757                                             st, 0);
 758                         }
 759                 }
 760                 if (sc->sc_mbuf != NULL)
 761                         pfsync_sendout(sc);
 762                 splx(s);
 763                 break;
 764         case PFSYNC_ACT_BUS:
 765                 /* If we're not waiting for a bulk update, who cares. */
 766                 if (sc->sc_ureq_sent == 0)
 767                         break;
 768
 769                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 770                     sizeof(*bus), &offp)) == NULL) {
 771                         pfsyncstats.pfsyncs_badlen++;
 772                         return;
 773                 }
 774                 bus = (struct pfsync_state_bus *)(mp->m_data + offp);
 775                 switch (bus->status) {
 776                 case PFSYNC_BUS_START:
 777                         timeout_add(&sc->sc_bulkfail_tmo,
 778                             pf_pool_limits[PF_LIMIT_STATES].limit /
 779                             (PFSYNC_BULKPACKETS * sc->sc_maxcount));
 780                         if (pf_status.debug >= PF_DEBUG_MISC)
 781                                 printf("pfsync: received bulk "
 782                                     "update start\n");
 783                         break;
 784                 case PFSYNC_BUS_END:
 785                         if (time_uptime - ntohl(bus->endtime) >=
 786                             sc->sc_ureq_sent) {
 787                                 /* that's it, we're happy */
 788                                 sc->sc_ureq_sent = 0;
 789                                 sc->sc_bulk_tries = 0;
 790                                 timeout_del(&sc->sc_bulkfail_tmo);
 791 #if NCARP > 0
 792                                 if (!pfsync_sync_ok)
 793                                         carp_group_demote_adj(&sc->sc_if, -1);
 794 #endif
 795                                 pfsync_sync_ok = 1;
 796                                 if (pf_status.debug >= PF_DEBUG_MISC)
 797                                         printf("pfsync: received valid "
 798                                             "bulk update end\n");
 799                         } else {
 800                                 if (pf_status.debug >= PF_DEBUG_MISC)
 801                                         printf("pfsync: received invalid "
 802                                             "bulk update end: bad timestamp\n");
 803                         }
 804                         break;
 805                 }
 806                 break;
 807         case PFSYNC_ACT_TDB_UPD:
 808                 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
 809                     count * sizeof(*pt), &offp)) == NULL) {
 810                         pfsyncstats.pfsyncs_badlen++;
 811                         return;
 812                 }
 813                 s = splsoftnet();
 814                 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp);
 815                     i < count; i++, pt++)
 816                         pfsync_update_net_tdb(pt);
 817                 splx(s);
 818                 break;
 819         }
 820
 821 done:
 822         if (m)
 823                 m_freem(m);
 824 }
 825
 826 int
 827 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
 828         struct rtentry *rt)
 829 {
 830         m_freem(m);
 831         return (0);
 832 }
 833
 834 /* ARGSUSED */
 835 int
 836 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 837 {
 838         struct proc *p = curproc;
 839         struct pfsync_softc *sc = ifp->if_softc;
 840         struct ifreq *ifr = (struct ifreq *)data;
 841         struct ip_moptions *imo = &sc->sc_imo;
 842         struct pfsyncreq pfsyncr;
 843         struct ifnet    *sifp;
 844         int s, error;
 845
 846         switch (cmd) {
 847         case SIOCSIFADDR:
 848         case SIOCAIFADDR:
 849         case SIOCSIFDSTADDR:
 850         case SIOCSIFFLAGS:
 851                 if (ifp->if_flags & IFF_UP)
 852                         ifp->if_flags |= IFF_RUNNING;
 853                 else
 854                         ifp->if_flags &= ~IFF_RUNNING;
 855                 break;
 856         case SIOCSIFMTU:
 857                 if (ifr->ifr_mtu < PFSYNC_MINMTU)
 858                         return (EINVAL);
 859                 if (ifr->ifr_mtu > MCLBYTES)
 860                         ifr->ifr_mtu = MCLBYTES;
 861                 s = splnet();
 862                 if (ifr->ifr_mtu < ifp->if_mtu)
 863                         pfsync_sendout(sc);
 864                 pfsync_setmtu(sc, ifr->ifr_mtu);
 865                 splx(s);
 866                 break;
 867         case SIOCGETPFSYNC:
 868                 bzero(&pfsyncr, sizeof(pfsyncr));
 869                 if (sc->sc_sync_ifp)
 870                         strlcpy(pfsyncr.pfsyncr_syncdev,
 871                             sc->sc_sync_ifp->if_xname, IFNAMSIZ);
 872                 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
 873                 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
 874                 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))))
 875                         return (error);
 876                 break;
 877         case SIOCSETPFSYNC:
 878                 if ((error = suser(p, p->p_acflag)) != 0)
 879                         return (error);
 880                 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
 881                         return (error);
 882
 883                 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
 884                         sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP;
 885                 else
 886                         sc->sc_sync_peer.s_addr =
 887                             pfsyncr.pfsyncr_syncpeer.s_addr;
 888
 889                 if (pfsyncr.pfsyncr_maxupdates > 255)
 890                         return (EINVAL);
 891                 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
 892
 893                 if (pfsyncr.pfsyncr_syncdev[0] == 0) {
 894                         sc->sc_sync_ifp = NULL;
 895                         if (sc->sc_mbuf_net != NULL) {
 896                                 /* Don't keep stale pfsync packets around. */
 897                                 s = splnet();
 898                                 m_freem(sc->sc_mbuf_net);
 899                                 sc->sc_mbuf_net = NULL;
 900                                 sc->sc_statep_net.s = NULL;
 901                                 splx(s);
 902                         }
 903                         if (imo->imo_num_memberships > 0) {
 904                                 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
 905                                 imo->imo_multicast_ifp = NULL;
 906                         }
 907                         break;
 908                 }
 909
 910                 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL)
 911                         return (EINVAL);
 912
 913                 s = splnet();
 914                 if (sifp->if_mtu < sc->sc_if.if_mtu ||
 915                     (sc->sc_sync_ifp != NULL &&
 916                     sifp->if_mtu < sc->sc_sync_ifp->if_mtu) ||
 917                     sifp->if_mtu < MCLBYTES - sizeof(struct ip))
 918                         pfsync_sendout(sc);
 919                 sc->sc_sync_ifp = sifp;
 920
 921                 pfsync_setmtu(sc, sc->sc_if.if_mtu);
 922
 923                 if (imo->imo_num_memberships > 0) {
 924                         in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
 925                         imo->imo_multicast_ifp = NULL;
 926                 }
 927
 928                 if (sc->sc_sync_ifp &&
 929                     sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
 930                         struct in_addr addr;
 931
 932                         if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) {
 933                                 sc->sc_sync_ifp = NULL;
 934                                 splx(s);
 935                                 return (EADDRNOTAVAIL);
 936                         }
 937
 938                         addr.s_addr = INADDR_PFSYNC_GROUP;
 939
 940                         if ((imo->imo_membership[0] =
 941                             in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) {
 942                                 sc->sc_sync_ifp = NULL;
 943                                 splx(s);
 944                                 return (ENOBUFS);
 945                         }
 946                         imo->imo_num_memberships++;
 947                         imo->imo_multicast_ifp = sc->sc_sync_ifp;
 948                         imo->imo_multicast_ttl = PFSYNC_DFLTTL;
 949                         imo->imo_multicast_loop = 0;
 950                 }
 951
 952                 if (sc->sc_sync_ifp ||
 953                     sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) {
 954                         /* Request a full state table update. */
 955                         sc->sc_ureq_sent = time_uptime;
 956 #if NCARP > 0
 957                         if (pfsync_sync_ok)
 958                                 carp_group_demote_adj(&sc->sc_if, 1);
 959 #endif
 960                         pfsync_sync_ok = 0;
 961                         if (pf_status.debug >= PF_DEBUG_MISC)
 962                                 printf("pfsync: requesting bulk update\n");
 963                         timeout_add(&sc->sc_bulkfail_tmo, 5 * hz);
 964                         error = pfsync_request_update(NULL, NULL);
 965                         if (error == ENOMEM) {
 966                                 splx(s);
 967                                 return (ENOMEM);
 968                         }
 969                         pfsync_sendout(sc);
 970                 }
 971                 splx(s);
 972
 973                 break;
 974
 975         default:
 976                 return (ENOTTY);
 977         }
 978
 979         return (0);
 980 }
 981
 982 void
 983 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req)
 984 {
 985         int mtu;
 986
 987         if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req)
 988                 mtu = sc->sc_sync_ifp->if_mtu;
 989         else
 990                 mtu = mtu_req;
 991
 992         sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) /
 993             sizeof(struct pfsync_state);
 994         if (sc->sc_maxcount > 254)
 995             sc->sc_maxcount = 254;
 996         sc->sc_if.if_mtu = sizeof(struct pfsync_header) +
 997             sc->sc_maxcount * sizeof(struct pfsync_state);
 998 }
 999
1000 struct mbuf *
1001 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp)
1002 {
1003         struct pfsync_header *h;
1004         struct mbuf *m;
1005         int len;
1006
1007         MGETHDR(m, M_DONTWAIT, MT_DATA);
1008         if (m == NULL) {
1009                 sc->sc_if.if_oerrors++;
1010                 return (NULL);
1011         }
1012
1013         switch (action) {
1014         case PFSYNC_ACT_CLR:
1015                 len = sizeof(struct pfsync_header) +
1016                     sizeof(struct pfsync_state_clr);
1017                 break;
1018         case PFSYNC_ACT_UPD_C:
1019                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) +
1020                     sizeof(struct pfsync_header);
1021                 break;
1022         case PFSYNC_ACT_DEL_C:
1023                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) +
1024                     sizeof(struct pfsync_header);
1025                 break;
1026         case PFSYNC_ACT_UREQ:
1027                 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) +
1028                     sizeof(struct pfsync_header);
1029                 break;
1030         case PFSYNC_ACT_BUS:
1031                 len = sizeof(struct pfsync_header) +
1032                     sizeof(struct pfsync_state_bus);
1033                 break;
1034         case PFSYNC_ACT_TDB_UPD:
1035                 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) +
1036                     sizeof(struct pfsync_header);
1037                 break;
1038         default:
1039                 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) +
1040                     sizeof(struct pfsync_header);
1041                 break;
1042         }
1043
1044         if (len > MHLEN) {
1045                 MCLGET(m, M_DONTWAIT);
1046                 if ((m->m_flags & M_EXT) == 0) {
1047                         m_free(m);
1048                         sc->sc_if.if_oerrors++;
1049                         return (NULL);
1050                 }
1051                 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1);
1052         } else
1053                 MH_ALIGN(m, len);
1054
1055         m->m_pkthdr.rcvif = NULL;
1056         m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header);
1057         h = mtod(m, struct pfsync_header *);
1058         h->version = PFSYNC_VERSION;
1059         h->af = 0;
1060         h->count = 0;
1061         h->action = action;
1062         if (action != PFSYNC_ACT_TDB_UPD)
1063                 bcopy(&pf_status.pf_chksum, &h->pf_chksum,
1064                     PF_MD5_DIGEST_LENGTH);
1065
1066         *sp = (void *)((char *)h + PFSYNC_HDRLEN);
1067         if (action == PFSYNC_ACT_TDB_UPD)
1068                 timeout_add(&sc->sc_tdb_tmo, hz);
1069         else
1070                 timeout_add(&sc->sc_tmo, hz);
1071         return (m);
1072 }
1073
1074 int
1075 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags)
1076 {
1077         struct ifnet *ifp = NULL;
1078         struct pfsync_softc *sc = pfsyncif;
1079         struct pfsync_header *h, *h_net;
1080         struct pfsync_state *sp = NULL;
1081         struct pfsync_state_upd *up = NULL;
1082         struct pfsync_state_del *dp = NULL;
1083         struct pf_rule *r;
1084         u_long secs;
1085         int s, ret = 0;
1086         u_int8_t i = 255, newaction = 0;
1087
1088         if (sc == NULL)
1089                 return (0);
1090         ifp = &sc->sc_if;
1091
1092         /*
1093          * If a packet falls in the forest and there's nobody around to
1094          * hear, does it make a sound?
1095          */
1096         if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL &&
1097             sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
1098                 /* Don't leave any stale pfsync packets hanging around. */
1099                 if (sc->sc_mbuf != NULL) {
1100                         m_freem(sc->sc_mbuf);
1101                         sc->sc_mbuf = NULL;
1102                         sc->sc_statep.s = NULL;
1103                 }
1104                 return (0);
1105         }
1106
1107         if (action >= PFSYNC_ACT_MAX)
1108                 return (EINVAL);
1109
1110         s = splnet();
1111         if (sc->sc_mbuf == NULL) {
1112                 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1113                     (void *)&sc->sc_statep.s)) == NULL) {
1114                         splx(s);
1115                         return (ENOMEM);
1116                 }
1117                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1118         } else {
1119                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1120                 if (h->action != action) {
1121                         pfsync_sendout(sc);
1122                         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1123                             (void *)&sc->sc_statep.s)) == NULL) {
1124                                 splx(s);
1125                                 return (ENOMEM);
1126                         }
1127                         h = mtod(sc->sc_mbuf, struct pfsync_header *);
1128                 } else {
1129                         /*
1130                          * If it's an update, look in the packet to see if
1131                          * we already have an update for the state.
1132                          */
1133                         if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) {
1134                                 struct pfsync_state *usp =
1135                                     (void *)((char *)h + PFSYNC_HDRLEN);
1136
1137                                 for (i = 0; i < h->count; i++) {
1138                                         if (!memcmp(usp->id, &st->id,
1139                                             PFSYNC_ID_LEN) &&
1140                                             usp->creatorid == st->creatorid) {
1141                                                 sp = usp;
1142                                                 sp->updates++;
1143                                                 break;
1144                                         }
1145                                         usp++;
1146                                 }
1147                         }
1148                 }
1149         }
1150
1151         secs = time_second;
1152
1153         st->pfsync_time = time_uptime;
1154
1155         if (sp == NULL) {
1156                 /* not a "duplicate" update */
1157                 i = 255;
1158                 sp = sc->sc_statep.s++;
1159                 sc->sc_mbuf->m_pkthdr.len =
1160                     sc->sc_mbuf->m_len += sizeof(struct pfsync_state);
1161                 h->count++;
1162                 bzero(sp, sizeof(*sp));
1163
1164                 bcopy(&st->id, sp->id, sizeof(sp->id));
1165                 sp->creatorid = st->creatorid;
1166
1167                 strlcpy(sp->ifname, st->u.s.kif->pfik_name, sizeof(sp->ifname));
1168                 pf_state_host_hton(&st->lan, &sp->lan);
1169                 pf_state_host_hton(&st->gwy, &sp->gwy);
1170                 pf_state_host_hton(&st->ext, &sp->ext);
1171
1172                 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr));
1173
1174                 sp->creation = htonl(secs - st->creation);
1175                 pf_state_counter_hton(st->packets[0], sp->packets[0]);
1176                 pf_state_counter_hton(st->packets[1], sp->packets[1]);
1177                 pf_state_counter_hton(st->bytes[0], sp->bytes[0]);
1178                 pf_state_counter_hton(st->bytes[1], sp->bytes[1]);
1179                 if ((r = st->rule.ptr) == NULL)
1180                         sp->rule = htonl(-1);
1181                 else
1182                         sp->rule = htonl(r->nr);
1183                 if ((r = st->anchor.ptr) == NULL)
1184                         sp->anchor = htonl(-1);
1185                 else
1186                         sp->anchor = htonl(r->nr);
1187                 sp->af = st->af;
1188                 sp->proto = st->proto;
1189                 sp->direction = st->direction;
1190                 sp->log = st->log;
1191                 sp->allow_opts = st->allow_opts;
1192                 sp->timeout = st->timeout;
1193
1194                 if (flags & PFSYNC_FLAG_STALE)
1195                         sp->sync_flags |= PFSTATE_STALE;
1196         }
1197
1198         pf_state_peer_hton(&st->src, &sp->src);
1199         pf_state_peer_hton(&st->dst, &sp->dst);
1200
1201         if (st->expire <= secs)
1202                 sp->expire = htonl(0);
1203         else
1204                 sp->expire = htonl(st->expire - secs);
1205
1206         /* do we need to build "compressed" actions for network transfer? */
1207         if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) {
1208                 switch (action) {
1209                 case PFSYNC_ACT_UPD:
1210                         newaction = PFSYNC_ACT_UPD_C;
1211                         break;
1212                 case PFSYNC_ACT_DEL:
1213                         newaction = PFSYNC_ACT_DEL_C;
1214                         break;
1215                 default:
1216                         /* by default we just send the uncompressed states */
1217                         break;
1218                 }
1219         }
1220
1221         if (newaction) {
1222                 if (sc->sc_mbuf_net == NULL) {
1223                         if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction,
1224                             (void *)&sc->sc_statep_net.s)) == NULL) {
1225                                 splx(s);
1226                                 return (ENOMEM);
1227                         }
1228                 }
1229                 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *);
1230
1231                 switch (newaction) {
1232                 case PFSYNC_ACT_UPD_C:
1233                         if (i != 255) {
1234                                 up = (void *)((char *)h_net +
1235                                     PFSYNC_HDRLEN + (i * sizeof(*up)));
1236                                 up->updates++;
1237                         } else {
1238                                 h_net->count++;
1239                                 sc->sc_mbuf_net->m_pkthdr.len =
1240                                     sc->sc_mbuf_net->m_len += sizeof(*up);
1241                                 up = sc->sc_statep_net.u++;
1242
1243                                 bzero(up, sizeof(*up));
1244                                 bcopy(&st->id, up->id, sizeof(up->id));
1245                                 up->creatorid = st->creatorid;
1246                         }
1247                         up->timeout = st->timeout;
1248                         up->expire = sp->expire;
1249                         up->src = sp->src;
1250                         up->dst = sp->dst;
1251                         break;
1252                 case PFSYNC_ACT_DEL_C:
1253                         sc->sc_mbuf_net->m_pkthdr.len =
1254                             sc->sc_mbuf_net->m_len += sizeof(*dp);
1255                         dp = sc->sc_statep_net.d++;
1256                         h_net->count++;
1257
1258                         bzero(dp, sizeof(*dp));
1259                         bcopy(&st->id, dp->id, sizeof(dp->id));
1260                         dp->creatorid = st->creatorid;
1261                         break;
1262                 }
1263         }
1264
1265         if (h->count == sc->sc_maxcount ||
1266             (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates)))
1267                 ret = pfsync_sendout(sc);
1268
1269         splx(s);
1270         return (ret);
1271 }
1272
1273 /* This must be called in splnet() */
1274 int
1275 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src)
1276 {
1277         struct ifnet *ifp = NULL;
1278         struct pfsync_header *h;
1279         struct pfsync_softc *sc = pfsyncif;
1280         struct pfsync_state_upd_req *rup;
1281         int ret = 0;
1282
1283         if (sc == NULL)
1284                 return (0);
1285
1286         ifp = &sc->sc_if;
1287         if (sc->sc_mbuf == NULL) {
1288                 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1289                     (void *)&sc->sc_statep.s)) == NULL)
1290                         return (ENOMEM);
1291                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1292         } else {
1293                 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1294                 if (h->action != PFSYNC_ACT_UREQ) {
1295                         pfsync_sendout(sc);
1296                         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1297                             (void *)&sc->sc_statep.s)) == NULL)
1298                                 return (ENOMEM);
1299                         h = mtod(sc->sc_mbuf, struct pfsync_header *);
1300                 }
1301         }
1302
1303         if (src != NULL)
1304                 sc->sc_sendaddr = *src;
1305         sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup);
1306         h->count++;
1307         rup = sc->sc_statep.r++;
1308         bzero(rup, sizeof(*rup));
1309         if (up != NULL) {
1310                 bcopy(up->id, rup->id, sizeof(rup->id));
1311                 rup->creatorid = up->creatorid;
1312         }
1313
1314         if (h->count == sc->sc_maxcount)
1315                 ret = pfsync_sendout(sc);
1316
1317         return (ret);
1318 }
1319
1320 int
1321 pfsync_clear_states(u_int32_t creatorid, char *ifname)
1322 {
1323         struct ifnet *ifp = NULL;
1324         struct pfsync_softc *sc = pfsyncif;
1325         struct pfsync_state_clr *cp;
1326         int s, ret;
1327
1328         if (sc == NULL)
1329                 return (0);
1330
1331         ifp = &sc->sc_if;
1332         s = splnet();
1333         if (sc->sc_mbuf != NULL)
1334                 pfsync_sendout(sc);
1335         if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR,
1336             (void *)&sc->sc_statep.c)) == NULL) {
1337                 splx(s);
1338                 return (ENOMEM);
1339         }
1340         sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp);
1341         cp = sc->sc_statep.c;
1342         cp->creatorid = creatorid;
1343         if (ifname != NULL)
1344                 strlcpy(cp->ifname, ifname, IFNAMSIZ);
1345
1346         ret = (pfsync_sendout(sc));
1347         splx(s);
1348         return (ret);
1349 }
1350
1351 void
1352 pfsync_timeout(void *v)
1353 {
1354         struct pfsync_softc *sc = v;
1355         int s;
1356
1357         s = splnet();
1358         pfsync_sendout(sc);
1359         splx(s);
1360 }
1361
1362 void
1363 pfsync_tdb_timeout(void *v)
1364 {
1365         struct pfsync_softc *sc = v;
1366         int s;
1367
1368         s = splnet();
1369         pfsync_tdb_sendout(sc);
1370         splx(s);
1371 }
1372
1373 /* This must be called in splnet() */
1374 void
1375 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status)
1376 {
1377         struct pfsync_state_bus *bus;
1378
1379         if (sc->sc_mbuf != NULL)
1380                 pfsync_sendout(sc);
1381
1382         if (pfsync_sync_ok &&
1383             (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS,
1384             (void *)&sc->sc_statep.b)) != NULL) {
1385                 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus);
1386                 bus = sc->sc_statep.b;
1387                 bus->creatorid = pf_status.hostid;
1388                 bus->status = status;
1389                 bus->endtime = htonl(time_uptime - sc->sc_ureq_received);
1390                 pfsync_sendout(sc);
1391         }
1392 }
1393
1394 void
1395 pfsync_bulk_update(void *v)
1396 {
1397         struct pfsync_softc *sc = v;
1398         int s, i = 0;
1399         struct pf_state *state;
1400
1401         s = splnet();
1402         if (sc->sc_mbuf != NULL)
1403                 pfsync_sendout(sc);
1404
1405         /*
1406          * Grab at most PFSYNC_BULKPACKETS worth of states which have not
1407          * been sent since the latest request was made.
1408          */
1409         state = sc->sc_bulk_send_next;
1410         if (state)
1411                 do {
1412                         /* send state update if syncable and not already sent */
1413                         if (!state->sync_flags
1414                             && state->timeout < PFTM_MAX
1415                             && state->pfsync_time <= sc->sc_ureq_received) {
1416                                 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0);
1417                                 i++;
1418                         }
1419
1420                         /* figure next state to send */
1421                         state = TAILQ_NEXT(state, u.s.entry_list);
1422
1423                         /* wrap to start of list if we hit the end */
1424                         if (!state)
1425                                 state = TAILQ_FIRST(&state_list);
1426                 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS &&
1427                     state != sc->sc_bulk_terminator);
1428
1429         if (!state || state == sc->sc_bulk_terminator) {
1430                 /* we're done */
1431                 pfsync_send_bus(sc, PFSYNC_BUS_END);
1432                 sc->sc_ureq_received = 0;
1433                 sc->sc_bulk_send_next = NULL;
1434                 sc->sc_bulk_terminator = NULL;
1435                 timeout_del(&sc->sc_bulk_tmo);
1436                 if (pf_status.debug >= PF_DEBUG_MISC)
1437                         printf("pfsync: bulk update complete\n");
1438         } else {
1439                 /* look again for more in a bit */
1440                 timeout_add(&sc->sc_bulk_tmo, 1);
1441                 sc->sc_bulk_send_next = state;
1442         }
1443         if (sc->sc_mbuf != NULL)
1444                 pfsync_sendout(sc);
1445         splx(s);
1446 }
1447
1448 void
1449 pfsync_bulkfail(void *v)
1450 {
1451         struct pfsync_softc *sc = v;
1452         int s, error;
1453
1454         if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
1455                 /* Try again in a bit */
1456                 timeout_add(&sc->sc_bulkfail_tmo, 5 * hz);
1457                 s = splnet();
1458                 error = pfsync_request_update(NULL, NULL);
1459                 if (error == ENOMEM) {
1460                         if (pf_status.debug >= PF_DEBUG_MISC)
1461                                 printf("pfsync: cannot allocate mbufs for "
1462                                     "bulk update\n");
1463                 } else
1464                         pfsync_sendout(sc);
1465                 splx(s);
1466         } else {
1467                 /* Pretend like the transfer was ok */
1468                 sc->sc_ureq_sent = 0;
1469                 sc->sc_bulk_tries = 0;
1470 #if NCARP > 0
1471                 if (!pfsync_sync_ok)
1472                         carp_group_demote_adj(&sc->sc_if, -1);
1473 #endif
1474                 pfsync_sync_ok = 1;
1475                 if (pf_status.debug >= PF_DEBUG_MISC)
1476                         printf("pfsync: failed to receive "
1477                             "bulk update status\n");
1478                 timeout_del(&sc->sc_bulkfail_tmo);
1479         }
1480 }
1481
1482 /* This must be called in splnet() */
1483 int
1484 pfsync_sendout(struct pfsync_softc *sc)
1485 {
1486 #if NBPFILTER > 0
1487         struct ifnet *ifp = &sc->sc_if;
1488 #endif
1489         struct mbuf *m;
1490
1491         timeout_del(&sc->sc_tmo);
1492
1493         if (sc->sc_mbuf == NULL)
1494                 return (0);
1495         m = sc->sc_mbuf;
1496         sc->sc_mbuf = NULL;
1497         sc->sc_statep.s = NULL;
1498
1499 #if NBPFILTER > 0
1500         if (ifp->if_bpf)
1501                 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1502 #endif
1503
1504         if (sc->sc_mbuf_net) {
1505                 m_freem(m);
1506                 m = sc->sc_mbuf_net;
1507                 sc->sc_mbuf_net = NULL;
1508                 sc->sc_statep_net.s = NULL;
1509         }
1510
1511         return pfsync_sendout_mbuf(sc, m);
1512 }
1513
1514 int
1515 pfsync_tdb_sendout(struct pfsync_softc *sc)
1516 {
1517 #if NBPFILTER > 0
1518         struct ifnet *ifp = &sc->sc_if;
1519 #endif
1520         struct mbuf *m;
1521
1522         timeout_del(&sc->sc_tdb_tmo);
1523
1524         if (sc->sc_mbuf_tdb == NULL)
1525                 return (0);
1526         m = sc->sc_mbuf_tdb;
1527         sc->sc_mbuf_tdb = NULL;
1528         sc->sc_statep_tdb.t = NULL;
1529
1530 #if NBPFILTER > 0
1531         if (ifp->if_bpf)
1532                 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1533 #endif
1534
1535         return pfsync_sendout_mbuf(sc, m);
1536 }
1537
1538 int
1539 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m)
1540 {
1541         struct sockaddr sa;
1542         struct ip *ip;
1543
1544         if (sc->sc_sync_ifp ||
1545             sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) {
1546                 M_PREPEND(m, sizeof(struct ip), M_DONTWAIT);
1547                 if (m == NULL) {
1548                         pfsyncstats.pfsyncs_onomem++;
1549                         return (0);
1550                 }
1551                 ip = mtod(m, struct ip *);
1552                 ip->ip_v = IPVERSION;
1553                 ip->ip_hl = sizeof(*ip) >> 2;
1554                 ip->ip_tos = IPTOS_LOWDELAY;
1555                 ip->ip_len = htons(m->m_pkthdr.len);
1556                 ip->ip_id = htons(ip_randomid());
1557                 ip->ip_off = htons(IP_DF);
1558                 ip->ip_ttl = PFSYNC_DFLTTL;
1559                 ip->ip_p = IPPROTO_PFSYNC;
1560                 ip->ip_sum = 0;
1561
1562                 bzero(&sa, sizeof(sa));
1563                 ip->ip_src.s_addr = INADDR_ANY;
1564
1565                 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP)
1566                         m->m_flags |= M_MCAST;
1567                 ip->ip_dst = sc->sc_sendaddr;
1568                 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr;
1569
1570                 pfsyncstats.pfsyncs_opackets++;
1571
1572                 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL))
1573                         pfsyncstats.pfsyncs_oerrors++;
1574         } else
1575                 m_freem(m);
1576
1577         return (0);
1578 }
1579
1580 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1581 void
1582 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1583 {
1584         struct tdb              *tdb;
1585         int                      s;
1586
1587         /* check for invalid values */
1588         if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1589             (pt->dst.sa.sa_family != AF_INET &&
1590              pt->dst.sa.sa_family != AF_INET6))
1591                 goto bad;
1592
1593         s = spltdb();
1594         tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1595         if (tdb) {
1596                 pt->rpl = ntohl(pt->rpl);
1597                 pt->cur_bytes = betoh64(pt->cur_bytes);
1598
1599                 /* Neither replay nor byte counter should ever decrease. */
1600                 if (pt->rpl < tdb->tdb_rpl ||
1601                     pt->cur_bytes < tdb->tdb_cur_bytes) {
1602                         splx(s);
1603                         goto bad;
1604                 }
1605
1606                 tdb->tdb_rpl = pt->rpl;
1607                 tdb->tdb_cur_bytes = pt->cur_bytes;
1608         }
1609         splx(s);
1610         return;
1611
1612  bad:
1613         if (pf_status.debug >= PF_DEBUG_MISC)
1614                 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1615                     "invalid value\n");
1616         pfsyncstats.pfsyncs_badstate++;
1617         return;
1618 }
1619
1620 /* One of our local tdbs have been updated, need to sync rpl with others */
1621 int
1622 pfsync_update_tdb(struct tdb *tdb, int output)
1623 {
1624         struct ifnet *ifp = NULL;
1625         struct pfsync_softc *sc = pfsyncif;
1626         struct pfsync_header *h;
1627         struct pfsync_tdb *pt = NULL;
1628         int s, i, ret;
1629
1630         if (sc == NULL)
1631                 return (0);
1632
1633         ifp = &sc->sc_if;
1634         if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL &&
1635             sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
1636                 /* Don't leave any stale pfsync packets hanging around. */
1637                 if (sc->sc_mbuf_tdb != NULL) {
1638                         m_freem(sc->sc_mbuf_tdb);
1639                         sc->sc_mbuf_tdb = NULL;
1640                         sc->sc_statep_tdb.t = NULL;
1641                 }
1642                 return (0);
1643         }
1644
1645         s = splnet();
1646         if (sc->sc_mbuf_tdb == NULL) {
1647                 if ((sc->sc_mbuf_tdb = pfsync_get_mbuf(sc, PFSYNC_ACT_TDB_UPD,
1648                     (void *)&sc->sc_statep_tdb.t)) == NULL) {
1649                         splx(s);
1650                         return (ENOMEM);
1651                 }
1652                 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *);
1653         } else {
1654                 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *);
1655                 if (h->action != PFSYNC_ACT_TDB_UPD) {
1656                         /*
1657                          * XXX will never happen as long as there's
1658                          * only one "TDB action".
1659                          */
1660                         pfsync_tdb_sendout(sc);
1661                         sc->sc_mbuf_tdb = pfsync_get_mbuf(sc,
1662                             PFSYNC_ACT_TDB_UPD, (void *)&sc->sc_statep_tdb.t);
1663                         if (sc->sc_mbuf_tdb == NULL) {
1664                                 splx(s);
1665                                 return (ENOMEM);
1666                         }
1667                         h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *);
1668                 } else if (sc->sc_maxupdates) {
1669                         /*
1670                          * If it's an update, look in the packet to see if
1671                          * we already have an update for the state.
1672                          */
1673                         struct pfsync_tdb *u =
1674                             (void *)((char *)h + PFSYNC_HDRLEN);
1675
1676                         for (i = 0; !pt && i < h->count; i++) {
1677                                 if (tdb->tdb_spi == u->spi &&
1678                                     tdb->tdb_sproto == u->sproto &&
1679                                     !bcmp(&tdb->tdb_dst, &u->dst,
1680                                     SA_LEN(&u->dst.sa))) {
1681                                         pt = u;
1682                                         pt->updates++;
1683                                 }
1684                                 u++;
1685                         }
1686                 }
1687         }
1688
1689         if (pt == NULL) {
1690                 /* not a "duplicate" update */
1691                 pt = sc->sc_statep_tdb.t++;
1692                 sc->sc_mbuf_tdb->m_pkthdr.len =
1693                     sc->sc_mbuf_tdb->m_len += sizeof(struct pfsync_tdb);
1694                 h->count++;
1695                 bzero(pt, sizeof(*pt));
1696
1697                 pt->spi = tdb->tdb_spi;
1698                 memcpy(&pt->dst, &tdb->tdb_dst, sizeof pt->dst);
1699                 pt->sproto = tdb->tdb_sproto;
1700         }
1701
1702         /*
1703          * When a failover happens, the master's rpl is probably above
1704          * what we see here (we may be up to a second late), so
1705          * increase it a bit for outbound tdbs to manage most such
1706          * situations.
1707          *
1708          * For now, just add an offset that is likely to be larger
1709          * than the number of packets we can see in one second. The RFC
1710          * just says the next packet must have a higher seq value.
1711          *
1712          * XXX What is a good algorithm for this? We could use
1713          * a rate-determined increase, but to know it, we would have
1714          * to extend struct tdb.
1715          * XXX pt->rpl can wrap over MAXINT, but if so the real tdb
1716          * will soon be replaced anyway. For now, just don't handle
1717          * this edge case.
1718          */
1719 #define RPL_INCR 16384
1720         pt->rpl = htonl(tdb->tdb_rpl + (output ? RPL_INCR : 0));
1721         pt->cur_bytes = htobe64(tdb->tdb_cur_bytes);
1722
1723         if (h->count == sc->sc_maxcount ||
1724             (sc->sc_maxupdates && (pt->updates >= sc->sc_maxupdates)))
1725                 ret = pfsync_tdb_sendout(sc);
1726
1727         splx(s);
1728         return (ret);
1729 }