sys/netpfil/pf/if_pfsync.c

   1 /*-
   2  * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC)
   3  *
   4  * Copyright (c) 2002 Michael Shalayeff
   5  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  20  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
  21  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  22  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  23  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  25  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  26  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
  27  * THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29
  30 /*-
  31  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
  32  *
  33  * Permission to use, copy, modify, and distribute this software for any
  34  * purpose with or without fee is hereby granted, provided that the above
  35  * copyright notice and this permission notice appear in all copies.
  36  *
  37  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  38  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  39  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  40  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  41  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  42  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  43  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  44  */
  45
  46 /*
  47  * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
  48  *
  49  * Revisions picked from OpenBSD after revision 1.110 import:
  50  * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
  51  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
  52  * 1.120, 1.175 - use monotonic time_uptime
  53  * 1.122 - reduce number of updates for non-TCP sessions
  54  * 1.125, 1.127 - rewrite merge or stale processing
  55  * 1.128 - cleanups
  56  * 1.146 - bzero() mbuf before sparsely filling it with data
  57  * 1.170 - SIOCSIFMTU checks
  58  * 1.126, 1.142 - deferred packets processing
  59  * 1.173 - correct expire time processing
  60  */
  61
  62 #include <sys/cdefs.h>
  63 __FBSDID("$FreeBSD$");
  64
  65 #include "opt_inet.h"
  66 #include "opt_inet6.h"
  67 #include "opt_pf.h"
  68
  69 #include <sys/param.h>
  70 #include <sys/bus.h>
  71 #include <sys/endian.h>
  72 #include <sys/interrupt.h>
  73 #include <sys/kernel.h>
  74 #include <sys/lock.h>
  75 #include <sys/mbuf.h>
  76 #include <sys/module.h>
  77 #include <sys/mutex.h>
  78 #include <sys/priv.h>
  79 #include <sys/protosw.h>
  80 #include <sys/smp.h>
  81 #include <sys/socket.h>
  82 #include <sys/sockio.h>
  83 #include <sys/sysctl.h>
  84 #include <sys/syslog.h>
  85
  86 #include <net/bpf.h>
  87 #include <net/if.h>
  88 #include <net/if_var.h>
  89 #include <net/if_clone.h>
  90 #include <net/if_types.h>
  91 #include <net/vnet.h>
  92 #include <net/pfvar.h>
  93 #include <net/if_pfsync.h>
  94
  95 #include <netinet/if_ether.h>
  96 #include <netinet/in.h>
  97 #include <netinet/in_var.h>
  98 #include <netinet/ip.h>
  99 #include <netinet/ip_carp.h>
 100 #include <netinet/ip_var.h>
 101 #include <netinet/tcp.h>
 102 #include <netinet/tcp_fsm.h>
 103 #include <netinet/tcp_seq.h>
 104
 105 #define PFSYNC_MINPKT ( \
 106         sizeof(struct ip) + \
 107         sizeof(struct pfsync_header) + \
 108         sizeof(struct pfsync_subheader) )
 109
 110 struct pfsync_bucket;
 111
 112 struct pfsync_pkt {
 113         struct ip *ip;
 114         struct in_addr src;
 115         u_int8_t flags;
 116 };
 117
 118 static int      pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
 119                     struct pfsync_state_peer *);
 120 static int      pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
 121 static int      pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
 122 static int      pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
 123 static int      pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
 124 static int      pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
 125 static int      pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
 126 static int      pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
 127 static int      pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
 128 static int      pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
 129 static int      pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
 130 static int      pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
 131 static int      pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
 132
 133 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
 134         pfsync_in_clr,                  /* PFSYNC_ACT_CLR */
 135         pfsync_in_ins,                  /* PFSYNC_ACT_INS */
 136         pfsync_in_iack,                 /* PFSYNC_ACT_INS_ACK */
 137         pfsync_in_upd,                  /* PFSYNC_ACT_UPD */
 138         pfsync_in_upd_c,                /* PFSYNC_ACT_UPD_C */
 139         pfsync_in_ureq,                 /* PFSYNC_ACT_UPD_REQ */
 140         pfsync_in_del,                  /* PFSYNC_ACT_DEL */
 141         pfsync_in_del_c,                /* PFSYNC_ACT_DEL_C */
 142         pfsync_in_error,                /* PFSYNC_ACT_INS_F */
 143         pfsync_in_error,                /* PFSYNC_ACT_DEL_F */
 144         pfsync_in_bus,                  /* PFSYNC_ACT_BUS */
 145         pfsync_in_tdb,                  /* PFSYNC_ACT_TDB */
 146         pfsync_in_eof                   /* PFSYNC_ACT_EOF */
 147 };
 148
 149 struct pfsync_q {
 150         void            (*write)(struct pf_state *, void *);
 151         size_t          len;
 152         u_int8_t        action;
 153 };
 154
 155 /* we have one of these for every PFSYNC_S_ */
 156 static void     pfsync_out_state(struct pf_state *, void *);
 157 static void     pfsync_out_iack(struct pf_state *, void *);
 158 static void     pfsync_out_upd_c(struct pf_state *, void *);
 159 static void     pfsync_out_del(struct pf_state *, void *);
 160
 161 static struct pfsync_q pfsync_qs[] = {
 162         { pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
 163         { pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
 164         { pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
 165         { pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
 166         { pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
 167 };
 168
 169 static void     pfsync_q_ins(struct pf_state *, int, bool);
 170 static void     pfsync_q_del(struct pf_state *, bool, struct pfsync_bucket *);
 171
 172 static void     pfsync_update_state(struct pf_state *);
 173
 174 struct pfsync_upd_req_item {
 175         TAILQ_ENTRY(pfsync_upd_req_item)        ur_entry;
 176         struct pfsync_upd_req                   ur_msg;
 177 };
 178
 179 struct pfsync_deferral {
 180         struct pfsync_softc             *pd_sc;
 181         TAILQ_ENTRY(pfsync_deferral)    pd_entry;
 182         u_int                           pd_refs;
 183         struct callout                  pd_tmo;
 184
 185         struct pf_state                 *pd_st;
 186         struct mbuf                     *pd_m;
 187 };
 188
 189 struct pfsync_sofct;
 190
 191 struct pfsync_bucket
 192 {
 193         int                     b_id;
 194         struct pfsync_softc     *b_sc;
 195         struct mtx              b_mtx;
 196         struct callout          b_tmo;
 197         int                     b_flags;
 198 #define PFSYNCF_BUCKET_PUSH     0x00000001
 199
 200         size_t                  b_len;
 201         TAILQ_HEAD(, pf_state)                  b_qs[PFSYNC_S_COUNT];
 202         TAILQ_HEAD(, pfsync_upd_req_item)       b_upd_req_list;
 203         TAILQ_HEAD(, pfsync_deferral)           b_deferrals;
 204         u_int                   b_deferred;
 205         void                    *b_plus;
 206         size_t                  b_pluslen;
 207
 208         struct  ifaltq b_snd;
 209 };
 210
 211 struct pfsync_softc {
 212         /* Configuration */
 213         struct ifnet            *sc_ifp;
 214         struct ifnet            *sc_sync_if;
 215         struct ip_moptions      sc_imo;
 216         struct in_addr          sc_sync_peer;
 217         uint32_t                sc_flags;
 218 #define PFSYNCF_OK              0x00000001
 219 #define PFSYNCF_DEFER           0x00000002
 220         uint8_t                 sc_maxupdates;
 221         struct ip               sc_template;
 222         struct mtx              sc_mtx;
 223
 224         /* Queued data */
 225         struct pfsync_bucket    *sc_buckets;
 226
 227         /* Bulk update info */
 228         struct mtx              sc_bulk_mtx;
 229         uint32_t                sc_ureq_sent;
 230         int                     sc_bulk_tries;
 231         uint32_t                sc_ureq_received;
 232         int                     sc_bulk_hashid;
 233         uint64_t                sc_bulk_stateid;
 234         uint32_t                sc_bulk_creatorid;
 235         struct callout          sc_bulk_tmo;
 236         struct callout          sc_bulkfail_tmo;
 237 };
 238
 239 #define PFSYNC_LOCK(sc)         mtx_lock(&(sc)->sc_mtx)
 240 #define PFSYNC_UNLOCK(sc)       mtx_unlock(&(sc)->sc_mtx)
 241 #define PFSYNC_LOCK_ASSERT(sc)  mtx_assert(&(sc)->sc_mtx, MA_OWNED)
 242
 243 #define PFSYNC_BUCKET_LOCK(b)           mtx_lock(&(b)->b_mtx)
 244 #define PFSYNC_BUCKET_UNLOCK(b)         mtx_unlock(&(b)->b_mtx)
 245 #define PFSYNC_BUCKET_LOCK_ASSERT(b)    mtx_assert(&(b)->b_mtx, MA_OWNED)
 246
 247 #define PFSYNC_BLOCK(sc)        mtx_lock(&(sc)->sc_bulk_mtx)
 248 #define PFSYNC_BUNLOCK(sc)      mtx_unlock(&(sc)->sc_bulk_mtx)
 249 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
 250
 251 static const char pfsyncname[] = "pfsync";
 252 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
 253 VNET_DEFINE_STATIC(struct pfsync_softc  *, pfsyncif) = NULL;
 254 #define V_pfsyncif              VNET(pfsyncif)
 255 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
 256 #define V_pfsync_swi_cookie     VNET(pfsync_swi_cookie)
 257 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie);
 258 #define V_pfsync_swi_ie         VNET(pfsync_swi_ie)
 259 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
 260 #define V_pfsyncstats           VNET(pfsyncstats)
 261 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
 262 #define V_pfsync_carp_adj       VNET(pfsync_carp_adj)
 263
 264 static void     pfsync_timeout(void *);
 265 static void     pfsync_push(struct pfsync_bucket *);
 266 static void     pfsync_push_all(struct pfsync_softc *);
 267 static void     pfsyncintr(void *);
 268 static int      pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
 269                     struct in_mfilter *imf);
 270 static void     pfsync_multicast_cleanup(struct pfsync_softc *);
 271 static void     pfsync_pointers_init(void);
 272 static void     pfsync_pointers_uninit(void);
 273 static int      pfsync_init(void);
 274 static void     pfsync_uninit(void);
 275
 276 static unsigned long pfsync_buckets;
 277
 278 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 279     "PFSYNC");
 280 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
 281     &VNET_NAME(pfsyncstats), pfsyncstats,
 282     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
 283 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
 284     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
 285 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
 286     &pfsync_buckets, 0, "Number of pfsync hash buckets");
 287
 288 static int      pfsync_clone_create(struct if_clone *, int, caddr_t);
 289 static void     pfsync_clone_destroy(struct ifnet *);
 290 static int      pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
 291                     struct pf_state_peer *);
 292 static int      pfsyncoutput(struct ifnet *, struct mbuf *,
 293                     const struct sockaddr *, struct route *);
 294 static int      pfsyncioctl(struct ifnet *, u_long, caddr_t);
 295
 296 static int      pfsync_defer(struct pf_state *, struct mbuf *);
 297 static void     pfsync_undefer(struct pfsync_deferral *, int);
 298 static void     pfsync_undefer_state(struct pf_state *, int);
 299 static void     pfsync_defer_tmo(void *);
 300
 301 static void     pfsync_request_update(u_int32_t, u_int64_t);
 302 static bool     pfsync_update_state_req(struct pf_state *);
 303
 304 static void     pfsync_drop(struct pfsync_softc *);
 305 static void     pfsync_sendout(int, int);
 306 static void     pfsync_send_plus(void *, size_t);
 307
 308 static void     pfsync_bulk_start(void);
 309 static void     pfsync_bulk_status(u_int8_t);
 310 static void     pfsync_bulk_update(void *);
 311 static void     pfsync_bulk_fail(void *);
 312
 313 static void     pfsync_detach_ifnet(struct ifnet *);
 314 #ifdef IPSEC
 315 static void     pfsync_update_net_tdb(struct pfsync_tdb *);
 316 #endif
 317 static struct pfsync_bucket     *pfsync_get_bucket(struct pfsync_softc *,
 318                     struct pf_state *);
 319
 320 #define PFSYNC_MAX_BULKTRIES    12
 321
 322 VNET_DEFINE(struct if_clone *, pfsync_cloner);
 323 #define V_pfsync_cloner VNET(pfsync_cloner)
 324
 325 static int
 326 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
 327 {
 328         struct pfsync_softc *sc;
 329         struct ifnet *ifp;
 330         struct pfsync_bucket *b;
 331         int c, q;
 332
 333         if (unit != 0)
 334                 return (EINVAL);
 335
 336         if (! pfsync_buckets)
 337                 pfsync_buckets = mp_ncpus * 2;
 338
 339         sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
 340         sc->sc_flags |= PFSYNCF_OK;
 341         sc->sc_maxupdates = 128;
 342
 343         ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
 344         if (ifp == NULL) {
 345                 free(sc, M_PFSYNC);
 346                 return (ENOSPC);
 347         }
 348         if_initname(ifp, pfsyncname, unit);
 349         ifp->if_softc = sc;
 350         ifp->if_ioctl = pfsyncioctl;
 351         ifp->if_output = pfsyncoutput;
 352         ifp->if_type = IFT_PFSYNC;
 353         ifp->if_hdrlen = sizeof(struct pfsync_header);
 354         ifp->if_mtu = ETHERMTU;
 355         mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
 356         mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
 357         callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
 358         callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
 359
 360         if_attach(ifp);
 361
 362         bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
 363
 364         sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
 365             M_PFSYNC, M_ZERO | M_WAITOK);
 366         for (c = 0; c < pfsync_buckets; c++) {
 367                 b = &sc->sc_buckets[c];
 368                 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
 369
 370                 b->b_id = c;
 371                 b->b_sc = sc;
 372                 b->b_len = PFSYNC_MINPKT;
 373
 374                 for (q = 0; q < PFSYNC_S_COUNT; q++)
 375                         TAILQ_INIT(&b->b_qs[q]);
 376
 377                 TAILQ_INIT(&b->b_upd_req_list);
 378                 TAILQ_INIT(&b->b_deferrals);
 379
 380                 callout_init(&b->b_tmo, 1);
 381
 382                 b->b_snd.ifq_maxlen = ifqmaxlen;
 383         }
 384
 385         V_pfsyncif = sc;
 386
 387         return (0);
 388 }
 389
 390 static void
 391 pfsync_clone_destroy(struct ifnet *ifp)
 392 {
 393         struct pfsync_softc *sc = ifp->if_softc;
 394         struct pfsync_bucket *b;
 395         int c;
 396
 397         for (c = 0; c < pfsync_buckets; c++) {
 398                 b = &sc->sc_buckets[c];
 399                 /*
 400                  * At this stage, everything should have already been
 401                  * cleared by pfsync_uninit(), and we have only to
 402                  * drain callouts.
 403                  */
 404                 while (b->b_deferred > 0) {
 405                         struct pfsync_deferral *pd =
 406                             TAILQ_FIRST(&b->b_deferrals);
 407
 408                         TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
 409                         b->b_deferred--;
 410                         if (callout_stop(&pd->pd_tmo) > 0) {
 411                                 pf_release_state(pd->pd_st);
 412                                 m_freem(pd->pd_m);
 413                                 free(pd, M_PFSYNC);
 414                         } else {
 415                                 pd->pd_refs++;
 416                                 callout_drain(&pd->pd_tmo);
 417                                 free(pd, M_PFSYNC);
 418                         }
 419                 }
 420
 421                 callout_drain(&b->b_tmo);
 422         }
 423
 424         callout_drain(&sc->sc_bulkfail_tmo);
 425         callout_drain(&sc->sc_bulk_tmo);
 426
 427         if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
 428                 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
 429         bpfdetach(ifp);
 430         if_detach(ifp);
 431
 432         pfsync_drop(sc);
 433
 434         if_free(ifp);
 435         pfsync_multicast_cleanup(sc);
 436         mtx_destroy(&sc->sc_mtx);
 437         mtx_destroy(&sc->sc_bulk_mtx);
 438
 439         free(sc->sc_buckets, M_PFSYNC);
 440         free(sc, M_PFSYNC);
 441
 442         V_pfsyncif = NULL;
 443 }
 444
 445 static int
 446 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
 447     struct pf_state_peer *d)
 448 {
 449         if (s->scrub.scrub_flag && d->scrub == NULL) {
 450                 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
 451                 if (d->scrub == NULL)
 452                         return (ENOMEM);
 453         }
 454
 455         return (0);
 456 }
 457
 458 static int
 459 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
 460 {
 461         struct pfsync_softc *sc = V_pfsyncif;
 462 #ifndef __NO_STRICT_ALIGNMENT
 463         struct pfsync_state_key key[2];
 464 #endif
 465         struct pfsync_state_key *kw, *ks;
 466         struct pf_state *st = NULL;
 467         struct pf_state_key *skw = NULL, *sks = NULL;
 468         struct pf_krule *r = NULL;
 469         struct pfi_kkif *kif;
 470         int error;
 471
 472         PF_RULES_RASSERT();
 473
 474         if (sp->creatorid == 0) {
 475                 if (V_pf_status.debug >= PF_DEBUG_MISC)
 476                         printf("%s: invalid creator id: %08x\n", __func__,
 477                             ntohl(sp->creatorid));
 478                 return (EINVAL);
 479         }
 480
 481         if ((kif = pfi_kkif_find(sp->ifname)) == NULL) {
 482                 if (V_pf_status.debug >= PF_DEBUG_MISC)
 483                         printf("%s: unknown interface: %s\n", __func__,
 484                             sp->ifname);
 485                 if (flags & PFSYNC_SI_IOCTL)
 486                         return (EINVAL);
 487                 return (0);     /* skip this state */
 488         }
 489
 490         /*
 491          * If the ruleset checksums match or the state is coming from the ioctl,
 492          * it's safe to associate the state with the rule of that number.
 493          */
 494         if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
 495             (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
 496             pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
 497                 r = pf_main_ruleset.rules[
 498                     PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
 499         else
 500                 r = &V_pf_default_rule;
 501
 502         if ((r->max_states &&
 503             counter_u64_fetch(r->states_cur) >= r->max_states))
 504                 goto cleanup;
 505
 506         /*
 507          * XXXGL: consider M_WAITOK in ioctl path after.
 508          */
 509         if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
 510                 goto cleanup;
 511
 512         for (int i = 0; i < 2; i++) {
 513                 st->packets[i] = counter_u64_alloc(M_NOWAIT);
 514                 st->bytes[i] = counter_u64_alloc(M_NOWAIT);
 515                 if (st->packets[i] == NULL || st->bytes[i] == NULL)
 516                         goto cleanup;
 517         }
 518
 519         if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
 520                 goto cleanup;
 521
 522 #ifndef __NO_STRICT_ALIGNMENT
 523         bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
 524         kw = &key[PF_SK_WIRE];
 525         ks = &key[PF_SK_STACK];
 526 #else
 527         kw = &sp->key[PF_SK_WIRE];
 528         ks = &sp->key[PF_SK_STACK];
 529 #endif
 530
 531         if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
 532             PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
 533             kw->port[0] != ks->port[0] ||
 534             kw->port[1] != ks->port[1]) {
 535                 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
 536                 if (sks == NULL)
 537                         goto cleanup;
 538         } else
 539                 sks = skw;
 540
 541         /* allocate memory for scrub info */
 542         if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
 543             pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
 544                 goto cleanup;
 545
 546         /* Copy to state key(s). */
 547         skw->addr[0] = kw->addr[0];
 548         skw->addr[1] = kw->addr[1];
 549         skw->port[0] = kw->port[0];
 550         skw->port[1] = kw->port[1];
 551         skw->proto = sp->proto;
 552         skw->af = sp->af;
 553         if (sks != skw) {
 554                 sks->addr[0] = ks->addr[0];
 555                 sks->addr[1] = ks->addr[1];
 556                 sks->port[0] = ks->port[0];
 557                 sks->port[1] = ks->port[1];
 558                 sks->proto = sp->proto;
 559                 sks->af = sp->af;
 560         }
 561
 562         /* copy to state */
 563         bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
 564         st->creation = time_uptime - ntohl(sp->creation);
 565         st->expire = time_uptime;
 566         if (sp->expire) {
 567                 uint32_t timeout;
 568
 569                 timeout = r->timeout[sp->timeout];
 570                 if (!timeout)
 571                         timeout = V_pf_default_rule.timeout[sp->timeout];
 572
 573                 /* sp->expire may have been adaptively scaled by export. */
 574                 st->expire -= timeout - ntohl(sp->expire);
 575         }
 576
 577         st->direction = sp->direction;
 578         st->log = sp->log;
 579         st->timeout = sp->timeout;
 580         st->state_flags = sp->state_flags;
 581
 582         st->id = sp->id;
 583         st->creatorid = sp->creatorid;
 584         pf_state_peer_ntoh(&sp->src, &st->src);
 585         pf_state_peer_ntoh(&sp->dst, &st->dst);
 586
 587         st->rule.ptr = r;
 588         st->nat_rule.ptr = NULL;
 589         st->anchor.ptr = NULL;
 590         st->rt_kif = NULL;
 591
 592         st->pfsync_time = time_uptime;
 593         st->sync_state = PFSYNC_S_NONE;
 594
 595         if (!(flags & PFSYNC_SI_IOCTL))
 596                 st->state_flags |= PFSTATE_NOSYNC;
 597
 598         if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
 599                 goto cleanup_state;
 600
 601         /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
 602         counter_u64_add(r->states_cur, 1);
 603         counter_u64_add(r->states_tot, 1);
 604
 605         if (!(flags & PFSYNC_SI_IOCTL)) {
 606                 st->state_flags &= ~PFSTATE_NOSYNC;
 607                 if (st->state_flags & PFSTATE_ACK) {
 608                         pfsync_q_ins(st, PFSYNC_S_IACK, true);
 609                         pfsync_push_all(sc);
 610                 }
 611         }
 612         st->state_flags &= ~PFSTATE_ACK;
 613         PF_STATE_UNLOCK(st);
 614
 615         return (0);
 616
 617 cleanup:
 618         error = ENOMEM;
 619         if (skw == sks)
 620                 sks = NULL;
 621         if (skw != NULL)
 622                 uma_zfree(V_pf_state_key_z, skw);
 623         if (sks != NULL)
 624                 uma_zfree(V_pf_state_key_z, sks);
 625
 626 cleanup_state:  /* pf_state_insert() frees the state keys. */
 627         if (st) {
 628                 for (int i = 0; i < 2; i++) {
 629                         counter_u64_free(st->packets[i]);
 630                         counter_u64_free(st->bytes[i]);
 631                 }
 632                 if (st->dst.scrub)
 633                         uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
 634                 if (st->src.scrub)
 635                         uma_zfree(V_pf_state_scrub_z, st->src.scrub);
 636                 uma_zfree(V_pf_state_z, st);
 637         }
 638         return (error);
 639 }
 640
 641 static int
 642 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
 643 {
 644         struct pfsync_softc *sc = V_pfsyncif;
 645         struct pfsync_pkt pkt;
 646         struct mbuf *m = *mp;
 647         struct ip *ip = mtod(m, struct ip *);
 648         struct pfsync_header *ph;
 649         struct pfsync_subheader subh;
 650
 651         int offset, len;
 652         int rv;
 653         uint16_t count;
 654
 655         PF_RULES_RLOCK_TRACKER;
 656
 657         *mp = NULL;
 658         V_pfsyncstats.pfsyncs_ipackets++;
 659
 660         /* Verify that we have a sync interface configured. */
 661         if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
 662             (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
 663                 goto done;
 664
 665         /* verify that the packet came in on the right interface */
 666         if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
 667                 V_pfsyncstats.pfsyncs_badif++;
 668                 goto done;
 669         }
 670
 671         if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
 672         if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
 673         /* verify that the IP TTL is 255. */
 674         if (ip->ip_ttl != PFSYNC_DFLTTL) {
 675                 V_pfsyncstats.pfsyncs_badttl++;
 676                 goto done;
 677         }
 678
 679         offset = ip->ip_hl << 2;
 680         if (m->m_pkthdr.len < offset + sizeof(*ph)) {
 681                 V_pfsyncstats.pfsyncs_hdrops++;
 682                 goto done;
 683         }
 684
 685         if (offset + sizeof(*ph) > m->m_len) {
 686                 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
 687                         V_pfsyncstats.pfsyncs_hdrops++;
 688                         return (IPPROTO_DONE);
 689                 }
 690                 ip = mtod(m, struct ip *);
 691         }
 692         ph = (struct pfsync_header *)((char *)ip + offset);
 693
 694         /* verify the version */
 695         if (ph->version != PFSYNC_VERSION) {
 696                 V_pfsyncstats.pfsyncs_badver++;
 697                 goto done;
 698         }
 699
 700         len = ntohs(ph->len) + offset;
 701         if (m->m_pkthdr.len < len) {
 702                 V_pfsyncstats.pfsyncs_badlen++;
 703                 goto done;
 704         }
 705
 706         /* Cheaper to grab this now than having to mess with mbufs later */
 707         pkt.ip = ip;
 708         pkt.src = ip->ip_src;
 709         pkt.flags = 0;
 710
 711         /*
 712          * Trusting pf_chksum during packet processing, as well as seeking
 713          * in interface name tree, require holding PF_RULES_RLOCK().
 714          */
 715         PF_RULES_RLOCK();
 716         if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
 717                 pkt.flags |= PFSYNC_SI_CKSUM;
 718
 719         offset += sizeof(*ph);
 720         while (offset <= len - sizeof(subh)) {
 721                 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
 722                 offset += sizeof(subh);
 723
 724                 if (subh.action >= PFSYNC_ACT_MAX) {
 725                         V_pfsyncstats.pfsyncs_badact++;
 726                         PF_RULES_RUNLOCK();
 727                         goto done;
 728                 }
 729
 730                 count = ntohs(subh.count);
 731                 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
 732                 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
 733                 if (rv == -1) {
 734                         PF_RULES_RUNLOCK();
 735                         return (IPPROTO_DONE);
 736                 }
 737
 738                 offset += rv;
 739         }
 740         PF_RULES_RUNLOCK();
 741
 742 done:
 743         m_freem(m);
 744         return (IPPROTO_DONE);
 745 }
 746
 747 static int
 748 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
 749 {
 750         struct pfsync_clr *clr;
 751         struct mbuf *mp;
 752         int len = sizeof(*clr) * count;
 753         int i, offp;
 754         u_int32_t creatorid;
 755
 756         mp = m_pulldown(m, offset, len, &offp);
 757         if (mp == NULL) {
 758                 V_pfsyncstats.pfsyncs_badlen++;
 759                 return (-1);
 760         }
 761         clr = (struct pfsync_clr *)(mp->m_data + offp);
 762
 763         for (i = 0; i < count; i++) {
 764                 creatorid = clr[i].creatorid;
 765
 766                 if (clr[i].ifname[0] != '\0' &&
 767                     pfi_kkif_find(clr[i].ifname) == NULL)
 768                         continue;
 769
 770                 for (int i = 0; i <= pf_hashmask; i++) {
 771                         struct pf_idhash *ih = &V_pf_idhash[i];
 772                         struct pf_state *s;
 773 relock:
 774                         PF_HASHROW_LOCK(ih);
 775                         LIST_FOREACH(s, &ih->states, entry) {
 776                                 if (s->creatorid == creatorid) {
 777                                         s->state_flags |= PFSTATE_NOSYNC;
 778                                         pf_unlink_state(s, PF_ENTER_LOCKED);
 779                                         goto relock;
 780                                 }
 781                         }
 782                         PF_HASHROW_UNLOCK(ih);
 783                 }
 784         }
 785
 786         return (len);
 787 }
 788
 789 static int
 790 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
 791 {
 792         struct mbuf *mp;
 793         struct pfsync_state *sa, *sp;
 794         int len = sizeof(*sp) * count;
 795         int i, offp;
 796
 797         mp = m_pulldown(m, offset, len, &offp);
 798         if (mp == NULL) {
 799                 V_pfsyncstats.pfsyncs_badlen++;
 800                 return (-1);
 801         }
 802         sa = (struct pfsync_state *)(mp->m_data + offp);
 803
 804         for (i = 0; i < count; i++) {
 805                 sp = &sa[i];
 806
 807                 /* Check for invalid values. */
 808                 if (sp->timeout >= PFTM_MAX ||
 809                     sp->src.state > PF_TCPS_PROXY_DST ||
 810                     sp->dst.state > PF_TCPS_PROXY_DST ||
 811                     sp->direction > PF_OUT ||
 812                     (sp->af != AF_INET && sp->af != AF_INET6)) {
 813                         if (V_pf_status.debug >= PF_DEBUG_MISC)
 814                                 printf("%s: invalid value\n", __func__);
 815                         V_pfsyncstats.pfsyncs_badval++;
 816                         continue;
 817                 }
 818
 819                 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
 820                         /* Drop out, but process the rest of the actions. */
 821                         break;
 822         }
 823
 824         return (len);
 825 }
 826
 827 static int
 828 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
 829 {
 830         struct pfsync_ins_ack *ia, *iaa;
 831         struct pf_state *st;
 832
 833         struct mbuf *mp;
 834         int len = count * sizeof(*ia);
 835         int offp, i;
 836
 837         mp = m_pulldown(m, offset, len, &offp);
 838         if (mp == NULL) {
 839                 V_pfsyncstats.pfsyncs_badlen++;
 840                 return (-1);
 841         }
 842         iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
 843
 844         for (i = 0; i < count; i++) {
 845                 ia = &iaa[i];
 846
 847                 st = pf_find_state_byid(ia->id, ia->creatorid);
 848                 if (st == NULL)
 849                         continue;
 850
 851                 if (st->state_flags & PFSTATE_ACK) {
 852                         pfsync_undefer_state(st, 0);
 853                 }
 854                 PF_STATE_UNLOCK(st);
 855         }
 856         /*
 857          * XXX this is not yet implemented, but we know the size of the
 858          * message so we can skip it.
 859          */
 860
 861         return (count * sizeof(struct pfsync_ins_ack));
 862 }
 863
 864 static int
 865 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
 866     struct pfsync_state_peer *dst)
 867 {
 868         int sync = 0;
 869
 870         PF_STATE_LOCK_ASSERT(st);
 871
 872         /*
 873          * The state should never go backwards except
 874          * for syn-proxy states.  Neither should the
 875          * sequence window slide backwards.
 876          */
 877         if ((st->src.state > src->state &&
 878             (st->src.state < PF_TCPS_PROXY_SRC ||
 879             src->state >= PF_TCPS_PROXY_SRC)) ||
 880
 881             (st->src.state == src->state &&
 882             SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
 883                 sync++;
 884         else
 885                 pf_state_peer_ntoh(src, &st->src);
 886
 887         if ((st->dst.state > dst->state) ||
 888
 889             (st->dst.state >= TCPS_SYN_SENT &&
 890             SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
 891                 sync++;
 892         else
 893                 pf_state_peer_ntoh(dst, &st->dst);
 894
 895         return (sync);
 896 }
 897
 898 static int
 899 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
 900 {
 901         struct pfsync_softc *sc = V_pfsyncif;
 902         struct pfsync_state *sa, *sp;
 903         struct pf_state *st;
 904         int sync;
 905
 906         struct mbuf *mp;
 907         int len = count * sizeof(*sp);
 908         int offp, i;
 909
 910         mp = m_pulldown(m, offset, len, &offp);
 911         if (mp == NULL) {
 912                 V_pfsyncstats.pfsyncs_badlen++;
 913                 return (-1);
 914         }
 915         sa = (struct pfsync_state *)(mp->m_data + offp);
 916
 917         for (i = 0; i < count; i++) {
 918                 sp = &sa[i];
 919
 920                 /* check for invalid values */
 921                 if (sp->timeout >= PFTM_MAX ||
 922                     sp->src.state > PF_TCPS_PROXY_DST ||
 923                     sp->dst.state > PF_TCPS_PROXY_DST) {
 924                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
 925                                 printf("pfsync_input: PFSYNC_ACT_UPD: "
 926                                     "invalid value\n");
 927                         }
 928                         V_pfsyncstats.pfsyncs_badval++;
 929                         continue;
 930                 }
 931
 932                 st = pf_find_state_byid(sp->id, sp->creatorid);
 933                 if (st == NULL) {
 934                         /* insert the update */
 935                         if (pfsync_state_import(sp, pkt->flags))
 936                                 V_pfsyncstats.pfsyncs_badstate++;
 937                         continue;
 938                 }
 939
 940                 if (st->state_flags & PFSTATE_ACK) {
 941                         pfsync_undefer_state(st, 1);
 942                 }
 943
 944                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
 945                         sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
 946                 else {
 947                         sync = 0;
 948
 949                         /*
 950                          * Non-TCP protocol state machine always go
 951                          * forwards
 952                          */
 953                         if (st->src.state > sp->src.state)
 954                                 sync++;
 955                         else
 956                                 pf_state_peer_ntoh(&sp->src, &st->src);
 957                         if (st->dst.state > sp->dst.state)
 958                                 sync++;
 959                         else
 960                                 pf_state_peer_ntoh(&sp->dst, &st->dst);
 961                 }
 962                 if (sync < 2) {
 963                         pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
 964                         pf_state_peer_ntoh(&sp->dst, &st->dst);
 965                         st->expire = time_uptime;
 966                         st->timeout = sp->timeout;
 967                 }
 968                 st->pfsync_time = time_uptime;
 969
 970                 if (sync) {
 971                         V_pfsyncstats.pfsyncs_stale++;
 972
 973                         pfsync_update_state(st);
 974                         PF_STATE_UNLOCK(st);
 975                         pfsync_push_all(sc);
 976                         continue;
 977                 }
 978                 PF_STATE_UNLOCK(st);
 979         }
 980
 981         return (len);
 982 }
 983
 984 static int
 985 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
 986 {
 987         struct pfsync_softc *sc = V_pfsyncif;
 988         struct pfsync_upd_c *ua, *up;
 989         struct pf_state *st;
 990         int len = count * sizeof(*up);
 991         int sync;
 992         struct mbuf *mp;
 993         int offp, i;
 994
 995         mp = m_pulldown(m, offset, len, &offp);
 996         if (mp == NULL) {
 997                 V_pfsyncstats.pfsyncs_badlen++;
 998                 return (-1);
 999         }
1000         ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1001
1002         for (i = 0; i < count; i++) {
1003                 up = &ua[i];
1004
1005                 /* check for invalid values */
1006                 if (up->timeout >= PFTM_MAX ||
1007                     up->src.state > PF_TCPS_PROXY_DST ||
1008                     up->dst.state > PF_TCPS_PROXY_DST) {
1009                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1010                                 printf("pfsync_input: "
1011                                     "PFSYNC_ACT_UPD_C: "
1012                                     "invalid value\n");
1013                         }
1014                         V_pfsyncstats.pfsyncs_badval++;
1015                         continue;
1016                 }
1017
1018                 st = pf_find_state_byid(up->id, up->creatorid);
1019                 if (st == NULL) {
1020                         /* We don't have this state. Ask for it. */
1021                         PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1022                         pfsync_request_update(up->creatorid, up->id);
1023                         PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1024                         continue;
1025                 }
1026
1027                 if (st->state_flags & PFSTATE_ACK) {
1028                         pfsync_undefer_state(st, 1);
1029                 }
1030
1031                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1032                         sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1033                 else {
1034                         sync = 0;
1035
1036                         /*
1037                          * Non-TCP protocol state machine always go
1038                          * forwards
1039                          */
1040                         if (st->src.state > up->src.state)
1041                                 sync++;
1042                         else
1043                                 pf_state_peer_ntoh(&up->src, &st->src);
1044                         if (st->dst.state > up->dst.state)
1045                                 sync++;
1046                         else
1047                                 pf_state_peer_ntoh(&up->dst, &st->dst);
1048                 }
1049                 if (sync < 2) {
1050                         pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1051                         pf_state_peer_ntoh(&up->dst, &st->dst);
1052                         st->expire = time_uptime;
1053                         st->timeout = up->timeout;
1054                 }
1055                 st->pfsync_time = time_uptime;
1056
1057                 if (sync) {
1058                         V_pfsyncstats.pfsyncs_stale++;
1059
1060                         pfsync_update_state(st);
1061                         PF_STATE_UNLOCK(st);
1062                         pfsync_push_all(sc);
1063                         continue;
1064                 }
1065                 PF_STATE_UNLOCK(st);
1066         }
1067
1068         return (len);
1069 }
1070
1071 static int
1072 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1073 {
1074         struct pfsync_upd_req *ur, *ura;
1075         struct mbuf *mp;
1076         int len = count * sizeof(*ur);
1077         int i, offp;
1078
1079         struct pf_state *st;
1080
1081         mp = m_pulldown(m, offset, len, &offp);
1082         if (mp == NULL) {
1083                 V_pfsyncstats.pfsyncs_badlen++;
1084                 return (-1);
1085         }
1086         ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1087
1088         for (i = 0; i < count; i++) {
1089                 ur = &ura[i];
1090
1091                 if (ur->id == 0 && ur->creatorid == 0)
1092                         pfsync_bulk_start();
1093                 else {
1094                         st = pf_find_state_byid(ur->id, ur->creatorid);
1095                         if (st == NULL) {
1096                                 V_pfsyncstats.pfsyncs_badstate++;
1097                                 continue;
1098                         }
1099                         if (st->state_flags & PFSTATE_NOSYNC) {
1100                                 PF_STATE_UNLOCK(st);
1101                                 continue;
1102                         }
1103
1104                         pfsync_update_state_req(st);
1105                         PF_STATE_UNLOCK(st);
1106                 }
1107         }
1108
1109         return (len);
1110 }
1111
1112 static int
1113 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1114 {
1115         struct mbuf *mp;
1116         struct pfsync_state *sa, *sp;
1117         struct pf_state *st;
1118         int len = count * sizeof(*sp);
1119         int offp, i;
1120
1121         mp = m_pulldown(m, offset, len, &offp);
1122         if (mp == NULL) {
1123                 V_pfsyncstats.pfsyncs_badlen++;
1124                 return (-1);
1125         }
1126         sa = (struct pfsync_state *)(mp->m_data + offp);
1127
1128         for (i = 0; i < count; i++) {
1129                 sp = &sa[i];
1130
1131                 st = pf_find_state_byid(sp->id, sp->creatorid);
1132                 if (st == NULL) {
1133                         V_pfsyncstats.pfsyncs_badstate++;
1134                         continue;
1135                 }
1136                 st->state_flags |= PFSTATE_NOSYNC;
1137                 pf_unlink_state(st, PF_ENTER_LOCKED);
1138         }
1139
1140         return (len);
1141 }
1142
1143 static int
1144 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1145 {
1146         struct mbuf *mp;
1147         struct pfsync_del_c *sa, *sp;
1148         struct pf_state *st;
1149         int len = count * sizeof(*sp);
1150         int offp, i;
1151
1152         mp = m_pulldown(m, offset, len, &offp);
1153         if (mp == NULL) {
1154                 V_pfsyncstats.pfsyncs_badlen++;
1155                 return (-1);
1156         }
1157         sa = (struct pfsync_del_c *)(mp->m_data + offp);
1158
1159         for (i = 0; i < count; i++) {
1160                 sp = &sa[i];
1161
1162                 st = pf_find_state_byid(sp->id, sp->creatorid);
1163                 if (st == NULL) {
1164                         V_pfsyncstats.pfsyncs_badstate++;
1165                         continue;
1166                 }
1167
1168                 st->state_flags |= PFSTATE_NOSYNC;
1169                 pf_unlink_state(st, PF_ENTER_LOCKED);
1170         }
1171
1172         return (len);
1173 }
1174
1175 static int
1176 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1177 {
1178         struct pfsync_softc *sc = V_pfsyncif;
1179         struct pfsync_bus *bus;
1180         struct mbuf *mp;
1181         int len = count * sizeof(*bus);
1182         int offp;
1183
1184         PFSYNC_BLOCK(sc);
1185
1186         /* If we're not waiting for a bulk update, who cares. */
1187         if (sc->sc_ureq_sent == 0) {
1188                 PFSYNC_BUNLOCK(sc);
1189                 return (len);
1190         }
1191
1192         mp = m_pulldown(m, offset, len, &offp);
1193         if (mp == NULL) {
1194                 PFSYNC_BUNLOCK(sc);
1195                 V_pfsyncstats.pfsyncs_badlen++;
1196                 return (-1);
1197         }
1198         bus = (struct pfsync_bus *)(mp->m_data + offp);
1199
1200         switch (bus->status) {
1201         case PFSYNC_BUS_START:
1202                 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1203                     V_pf_limits[PF_LIMIT_STATES].limit /
1204                     ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1205                     sizeof(struct pfsync_state)),
1206                     pfsync_bulk_fail, sc);
1207                 if (V_pf_status.debug >= PF_DEBUG_MISC)
1208                         printf("pfsync: received bulk update start\n");
1209                 break;
1210
1211         case PFSYNC_BUS_END:
1212                 if (time_uptime - ntohl(bus->endtime) >=
1213                     sc->sc_ureq_sent) {
1214                         /* that's it, we're happy */
1215                         sc->sc_ureq_sent = 0;
1216                         sc->sc_bulk_tries = 0;
1217                         callout_stop(&sc->sc_bulkfail_tmo);
1218                         if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1219                                 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1220                                     "pfsync bulk done");
1221                         sc->sc_flags |= PFSYNCF_OK;
1222                         if (V_pf_status.debug >= PF_DEBUG_MISC)
1223                                 printf("pfsync: received valid "
1224                                     "bulk update end\n");
1225                 } else {
1226                         if (V_pf_status.debug >= PF_DEBUG_MISC)
1227                                 printf("pfsync: received invalid "
1228                                     "bulk update end: bad timestamp\n");
1229                 }
1230                 break;
1231         }
1232         PFSYNC_BUNLOCK(sc);
1233
1234         return (len);
1235 }
1236
1237 static int
1238 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1239 {
1240         int len = count * sizeof(struct pfsync_tdb);
1241
1242 #if defined(IPSEC)
1243         struct pfsync_tdb *tp;
1244         struct mbuf *mp;
1245         int offp;
1246         int i;
1247         int s;
1248
1249         mp = m_pulldown(m, offset, len, &offp);
1250         if (mp == NULL) {
1251                 V_pfsyncstats.pfsyncs_badlen++;
1252                 return (-1);
1253         }
1254         tp = (struct pfsync_tdb *)(mp->m_data + offp);
1255
1256         for (i = 0; i < count; i++)
1257                 pfsync_update_net_tdb(&tp[i]);
1258 #endif
1259
1260         return (len);
1261 }
1262
1263 #if defined(IPSEC)
1264 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1265 static void
1266 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1267 {
1268         struct tdb              *tdb;
1269         int                      s;
1270
1271         /* check for invalid values */
1272         if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1273             (pt->dst.sa.sa_family != AF_INET &&
1274             pt->dst.sa.sa_family != AF_INET6))
1275                 goto bad;
1276
1277         tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1278         if (tdb) {
1279                 pt->rpl = ntohl(pt->rpl);
1280                 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1281
1282                 /* Neither replay nor byte counter should ever decrease. */
1283                 if (pt->rpl < tdb->tdb_rpl ||
1284                     pt->cur_bytes < tdb->tdb_cur_bytes) {
1285                         goto bad;
1286                 }
1287
1288                 tdb->tdb_rpl = pt->rpl;
1289                 tdb->tdb_cur_bytes = pt->cur_bytes;
1290         }
1291         return;
1292
1293 bad:
1294         if (V_pf_status.debug >= PF_DEBUG_MISC)
1295                 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1296                     "invalid value\n");
1297         V_pfsyncstats.pfsyncs_badstate++;
1298         return;
1299 }
1300 #endif
1301
1302 static int
1303 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1304 {
1305         /* check if we are at the right place in the packet */
1306         if (offset != m->m_pkthdr.len)
1307                 V_pfsyncstats.pfsyncs_badlen++;
1308
1309         /* we're done. free and let the caller return */
1310         m_freem(m);
1311         return (-1);
1312 }
1313
1314 static int
1315 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1316 {
1317         V_pfsyncstats.pfsyncs_badact++;
1318
1319         m_freem(m);
1320         return (-1);
1321 }
1322
1323 static int
1324 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1325         struct route *rt)
1326 {
1327         m_freem(m);
1328         return (0);
1329 }
1330
1331 /* ARGSUSED */
1332 static int
1333 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1334 {
1335         struct pfsync_softc *sc = ifp->if_softc;
1336         struct ifreq *ifr = (struct ifreq *)data;
1337         struct pfsyncreq pfsyncr;
1338         int error;
1339         int c;
1340
1341         switch (cmd) {
1342         case SIOCSIFFLAGS:
1343                 PFSYNC_LOCK(sc);
1344                 if (ifp->if_flags & IFF_UP) {
1345                         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1346                         PFSYNC_UNLOCK(sc);
1347                         pfsync_pointers_init();
1348                 } else {
1349                         ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1350                         PFSYNC_UNLOCK(sc);
1351                         pfsync_pointers_uninit();
1352                 }
1353                 break;
1354         case SIOCSIFMTU:
1355                 if (!sc->sc_sync_if ||
1356                     ifr->ifr_mtu <= PFSYNC_MINPKT ||
1357                     ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1358                         return (EINVAL);
1359                 if (ifr->ifr_mtu < ifp->if_mtu) {
1360                         for (c = 0; c < pfsync_buckets; c++) {
1361                                 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1362                                 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1363                                         pfsync_sendout(1, c);
1364                                 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1365                         }
1366                 }
1367                 ifp->if_mtu = ifr->ifr_mtu;
1368                 break;
1369         case SIOCGETPFSYNC:
1370                 bzero(&pfsyncr, sizeof(pfsyncr));
1371                 PFSYNC_LOCK(sc);
1372                 if (sc->sc_sync_if) {
1373                         strlcpy(pfsyncr.pfsyncr_syncdev,
1374                             sc->sc_sync_if->if_xname, IFNAMSIZ);
1375                 }
1376                 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1377                 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1378                 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1379                     (sc->sc_flags & PFSYNCF_DEFER));
1380                 PFSYNC_UNLOCK(sc);
1381                 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1382                     sizeof(pfsyncr)));
1383
1384         case SIOCSETPFSYNC:
1385             {
1386                 struct in_mfilter *imf = NULL;
1387                 struct ifnet *sifp;
1388                 struct ip *ip;
1389
1390                 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1391                         return (error);
1392                 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1393                     sizeof(pfsyncr))))
1394                         return (error);
1395
1396                 if (pfsyncr.pfsyncr_maxupdates > 255)
1397                         return (EINVAL);
1398
1399                 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1400                         sifp = NULL;
1401                 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1402                         return (EINVAL);
1403
1404                 if (sifp != NULL && (
1405                     pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1406                     pfsyncr.pfsyncr_syncpeer.s_addr ==
1407                     htonl(INADDR_PFSYNC_GROUP)))
1408                         imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1409
1410                 PFSYNC_LOCK(sc);
1411                 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1412                         sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1413                 else
1414                         sc->sc_sync_peer.s_addr =
1415                             pfsyncr.pfsyncr_syncpeer.s_addr;
1416
1417                 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1418                 if (pfsyncr.pfsyncr_defer) {
1419                         sc->sc_flags |= PFSYNCF_DEFER;
1420                         V_pfsync_defer_ptr = pfsync_defer;
1421                 } else {
1422                         sc->sc_flags &= ~PFSYNCF_DEFER;
1423                         V_pfsync_defer_ptr = NULL;
1424                 }
1425
1426                 if (sifp == NULL) {
1427                         if (sc->sc_sync_if)
1428                                 if_rele(sc->sc_sync_if);
1429                         sc->sc_sync_if = NULL;
1430                         pfsync_multicast_cleanup(sc);
1431                         PFSYNC_UNLOCK(sc);
1432                         break;
1433                 }
1434
1435                 for (c = 0; c < pfsync_buckets; c++) {
1436                         PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1437                         if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
1438                             (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1439                             (sc->sc_sync_if != NULL &&
1440                             sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1441                             sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1442                                 pfsync_sendout(1, c);
1443                         PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1444                 }
1445
1446                 pfsync_multicast_cleanup(sc);
1447
1448                 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1449                         error = pfsync_multicast_setup(sc, sifp, imf);
1450                         if (error) {
1451                                 if_rele(sifp);
1452                                 ip_mfilter_free(imf);
1453                                 PFSYNC_UNLOCK(sc);
1454                                 return (error);
1455                         }
1456                 }
1457                 if (sc->sc_sync_if)
1458                         if_rele(sc->sc_sync_if);
1459                 sc->sc_sync_if = sifp;
1460
1461                 ip = &sc->sc_template;
1462                 bzero(ip, sizeof(*ip));
1463                 ip->ip_v = IPVERSION;
1464                 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1465                 ip->ip_tos = IPTOS_LOWDELAY;
1466                 /* len and id are set later. */
1467                 ip->ip_off = htons(IP_DF);
1468                 ip->ip_ttl = PFSYNC_DFLTTL;
1469                 ip->ip_p = IPPROTO_PFSYNC;
1470                 ip->ip_src.s_addr = INADDR_ANY;
1471                 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1472
1473                 /* Request a full state table update. */
1474                 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1475                         (*carp_demote_adj_p)(V_pfsync_carp_adj,
1476                             "pfsync bulk start");
1477                 sc->sc_flags &= ~PFSYNCF_OK;
1478                 if (V_pf_status.debug >= PF_DEBUG_MISC)
1479                         printf("pfsync: requesting bulk update\n");
1480                 PFSYNC_UNLOCK(sc);
1481                 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1482                 pfsync_request_update(0, 0);
1483                 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1484                 PFSYNC_BLOCK(sc);
1485                 sc->sc_ureq_sent = time_uptime;
1486                 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1487                     sc);
1488                 PFSYNC_BUNLOCK(sc);
1489
1490                 break;
1491             }
1492         default:
1493                 return (ENOTTY);
1494         }
1495
1496         return (0);
1497 }
1498
1499 static void
1500 pfsync_out_state(struct pf_state *st, void *buf)
1501 {
1502         struct pfsync_state *sp = buf;
1503
1504         pfsync_state_export(sp, st);
1505 }
1506
1507 static void
1508 pfsync_out_iack(struct pf_state *st, void *buf)
1509 {
1510         struct pfsync_ins_ack *iack = buf;
1511
1512         iack->id = st->id;
1513         iack->creatorid = st->creatorid;
1514 }
1515
1516 static void
1517 pfsync_out_upd_c(struct pf_state *st, void *buf)
1518 {
1519         struct pfsync_upd_c *up = buf;
1520
1521         bzero(up, sizeof(*up));
1522         up->id = st->id;
1523         pf_state_peer_hton(&st->src, &up->src);
1524         pf_state_peer_hton(&st->dst, &up->dst);
1525         up->creatorid = st->creatorid;
1526         up->timeout = st->timeout;
1527 }
1528
1529 static void
1530 pfsync_out_del(struct pf_state *st, void *buf)
1531 {
1532         struct pfsync_del_c *dp = buf;
1533
1534         dp->id = st->id;
1535         dp->creatorid = st->creatorid;
1536         st->state_flags |= PFSTATE_NOSYNC;
1537 }
1538
1539 static void
1540 pfsync_drop(struct pfsync_softc *sc)
1541 {
1542         struct pf_state *st, *next;
1543         struct pfsync_upd_req_item *ur;
1544         struct pfsync_bucket *b;
1545         int c, q;
1546
1547         for (c = 0; c < pfsync_buckets; c++) {
1548                 b = &sc->sc_buckets[c];
1549                 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1550                         if (TAILQ_EMPTY(&b->b_qs[q]))
1551                                 continue;
1552
1553                         TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1554                                 KASSERT(st->sync_state == q,
1555                                         ("%s: st->sync_state == q",
1556                                                 __func__));
1557                                 st->sync_state = PFSYNC_S_NONE;
1558                                 pf_release_state(st);
1559                         }
1560                         TAILQ_INIT(&b->b_qs[q]);
1561                 }
1562
1563                 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1564                         TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1565                         free(ur, M_PFSYNC);
1566                 }
1567
1568                 b->b_len = PFSYNC_MINPKT;
1569                 b->b_plus = NULL;
1570         }
1571 }
1572
1573 static void
1574 pfsync_sendout(int schedswi, int c)
1575 {
1576         struct pfsync_softc *sc = V_pfsyncif;
1577         struct ifnet *ifp = sc->sc_ifp;
1578         struct mbuf *m;
1579         struct ip *ip;
1580         struct pfsync_header *ph;
1581         struct pfsync_subheader *subh;
1582         struct pf_state *st, *st_next;
1583         struct pfsync_upd_req_item *ur;
1584         struct pfsync_bucket *b = &sc->sc_buckets[c];
1585         int offset;
1586         int q, count = 0;
1587
1588         KASSERT(sc != NULL, ("%s: null sc", __func__));
1589         KASSERT(b->b_len > PFSYNC_MINPKT,
1590             ("%s: sc_len %zu", __func__, b->b_len));
1591         PFSYNC_BUCKET_LOCK_ASSERT(b);
1592
1593         if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1594                 pfsync_drop(sc);
1595                 return;
1596         }
1597
1598         m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1599         if (m == NULL) {
1600                 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1601                 V_pfsyncstats.pfsyncs_onomem++;
1602                 return;
1603         }
1604         m->m_data += max_linkhdr;
1605         m->m_len = m->m_pkthdr.len = b->b_len;
1606
1607         /* build the ip header */
1608         ip = (struct ip *)m->m_data;
1609         bcopy(&sc->sc_template, ip, sizeof(*ip));
1610         offset = sizeof(*ip);
1611
1612         ip->ip_len = htons(m->m_pkthdr.len);
1613         ip_fillid(ip);
1614
1615         /* build the pfsync header */
1616         ph = (struct pfsync_header *)(m->m_data + offset);
1617         bzero(ph, sizeof(*ph));
1618         offset += sizeof(*ph);
1619
1620         ph->version = PFSYNC_VERSION;
1621         ph->len = htons(b->b_len - sizeof(*ip));
1622         bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1623
1624         /* walk the queues */
1625         for (q = 0; q < PFSYNC_S_COUNT; q++) {
1626                 if (TAILQ_EMPTY(&b->b_qs[q]))
1627                         continue;
1628
1629                 subh = (struct pfsync_subheader *)(m->m_data + offset);
1630                 offset += sizeof(*subh);
1631
1632                 count = 0;
1633                 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1634                         KASSERT(st->sync_state == q,
1635                                 ("%s: st->sync_state == q",
1636                                         __func__));
1637                         /*
1638                          * XXXGL: some of write methods do unlocked reads
1639                          * of state data :(
1640                          */
1641                         pfsync_qs[q].write(st, m->m_data + offset);
1642                         offset += pfsync_qs[q].len;
1643                         st->sync_state = PFSYNC_S_NONE;
1644                         pf_release_state(st);
1645                         count++;
1646                 }
1647                 TAILQ_INIT(&b->b_qs[q]);
1648
1649                 bzero(subh, sizeof(*subh));
1650                 subh->action = pfsync_qs[q].action;
1651                 subh->count = htons(count);
1652                 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1653         }
1654
1655         if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1656                 subh = (struct pfsync_subheader *)(m->m_data + offset);
1657                 offset += sizeof(*subh);
1658
1659                 count = 0;
1660                 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1661                         TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1662
1663                         bcopy(&ur->ur_msg, m->m_data + offset,
1664                             sizeof(ur->ur_msg));
1665                         offset += sizeof(ur->ur_msg);
1666                         free(ur, M_PFSYNC);
1667                         count++;
1668                 }
1669
1670                 bzero(subh, sizeof(*subh));
1671                 subh->action = PFSYNC_ACT_UPD_REQ;
1672                 subh->count = htons(count);
1673                 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1674         }
1675
1676         /* has someone built a custom region for us to add? */
1677         if (b->b_plus != NULL) {
1678                 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1679                 offset += b->b_pluslen;
1680
1681                 b->b_plus = NULL;
1682         }
1683
1684         subh = (struct pfsync_subheader *)(m->m_data + offset);
1685         offset += sizeof(*subh);
1686
1687         bzero(subh, sizeof(*subh));
1688         subh->action = PFSYNC_ACT_EOF;
1689         subh->count = htons(1);
1690         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1691
1692         /* we're done, let's put it on the wire */
1693         if (ifp->if_bpf) {
1694                 m->m_data += sizeof(*ip);
1695                 m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip);
1696                 BPF_MTAP(ifp, m);
1697                 m->m_data -= sizeof(*ip);
1698                 m->m_len = m->m_pkthdr.len = b->b_len;
1699         }
1700
1701         if (sc->sc_sync_if == NULL) {
1702                 b->b_len = PFSYNC_MINPKT;
1703                 m_freem(m);
1704                 return;
1705         }
1706
1707         if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1708         if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1709         b->b_len = PFSYNC_MINPKT;
1710
1711         if (!_IF_QFULL(&b->b_snd))
1712                 _IF_ENQUEUE(&b->b_snd, m);
1713         else {
1714                 m_freem(m);
1715                 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1716         }
1717         if (schedswi)
1718                 swi_sched(V_pfsync_swi_cookie, 0);
1719 }
1720
1721 static void
1722 pfsync_insert_state(struct pf_state *st)
1723 {
1724         struct pfsync_softc *sc = V_pfsyncif;
1725         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1726
1727         if (st->state_flags & PFSTATE_NOSYNC)
1728                 return;
1729
1730         if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1731             st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1732                 st->state_flags |= PFSTATE_NOSYNC;
1733                 return;
1734         }
1735
1736         KASSERT(st->sync_state == PFSYNC_S_NONE,
1737                 ("%s: st->sync_state %u", __func__, st->sync_state));
1738
1739         PFSYNC_BUCKET_LOCK(b);
1740         if (b->b_len == PFSYNC_MINPKT)
1741                 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1742
1743         pfsync_q_ins(st, PFSYNC_S_INS, true);
1744         PFSYNC_BUCKET_UNLOCK(b);
1745
1746         st->sync_updates = 0;
1747 }
1748
1749 static int
1750 pfsync_defer(struct pf_state *st, struct mbuf *m)
1751 {
1752         struct pfsync_softc *sc = V_pfsyncif;
1753         struct pfsync_deferral *pd;
1754         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1755
1756         if (m->m_flags & (M_BCAST|M_MCAST))
1757                 return (0);
1758
1759         PFSYNC_LOCK(sc);
1760
1761         if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1762             !(sc->sc_flags & PFSYNCF_DEFER)) {
1763                 PFSYNC_UNLOCK(sc);
1764                 return (0);
1765         }
1766
1767         if (b->b_deferred >= 128)
1768                 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1769
1770         pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1771         if (pd == NULL)
1772                 return (0);
1773         b->b_deferred++;
1774
1775         m->m_flags |= M_SKIP_FIREWALL;
1776         st->state_flags |= PFSTATE_ACK;
1777
1778         pd->pd_sc = sc;
1779         pd->pd_refs = 0;
1780         pd->pd_st = st;
1781         pf_ref_state(st);
1782         pd->pd_m = m;
1783
1784         TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
1785         callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
1786         callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1787
1788         pfsync_push(b);
1789
1790         return (1);
1791 }
1792
1793 static void
1794 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1795 {
1796         struct pfsync_softc *sc = pd->pd_sc;
1797         struct mbuf *m = pd->pd_m;
1798         struct pf_state *st = pd->pd_st;
1799         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1800
1801         PFSYNC_BUCKET_LOCK_ASSERT(b);
1802
1803         TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1804         b->b_deferred--;
1805         pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1806         free(pd, M_PFSYNC);
1807         pf_release_state(st);
1808
1809         if (drop)
1810                 m_freem(m);
1811         else {
1812                 _IF_ENQUEUE(&b->b_snd, m);
1813                 pfsync_push(b);
1814         }
1815 }
1816
1817 static void
1818 pfsync_defer_tmo(void *arg)
1819 {
1820         struct epoch_tracker et;
1821         struct pfsync_deferral *pd = arg;
1822         struct pfsync_softc *sc = pd->pd_sc;
1823         struct mbuf *m = pd->pd_m;
1824         struct pf_state *st = pd->pd_st;
1825         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1826
1827         PFSYNC_BUCKET_LOCK_ASSERT(b);
1828
1829         NET_EPOCH_ENTER(et);
1830         CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1831
1832         TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1833         b->b_deferred--;
1834         pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1835         if (pd->pd_refs == 0)
1836                 free(pd, M_PFSYNC);
1837         PFSYNC_UNLOCK(sc);
1838
1839         ip_output(m, NULL, NULL, 0, NULL, NULL);
1840
1841         pf_release_state(st);
1842
1843         CURVNET_RESTORE();
1844         NET_EPOCH_EXIT(et);
1845 }
1846
1847 static void
1848 pfsync_undefer_state(struct pf_state *st, int drop)
1849 {
1850         struct pfsync_softc *sc = V_pfsyncif;
1851         struct pfsync_deferral *pd;
1852         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1853
1854         PFSYNC_BUCKET_LOCK(b);
1855
1856         TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
1857                  if (pd->pd_st == st) {
1858                         if (callout_stop(&pd->pd_tmo) > 0)
1859                                 pfsync_undefer(pd, drop);
1860
1861                         PFSYNC_BUCKET_UNLOCK(b);
1862                         return;
1863                 }
1864         }
1865         PFSYNC_BUCKET_UNLOCK(b);
1866
1867         panic("%s: unable to find deferred state", __func__);
1868 }
1869
1870 static struct pfsync_bucket*
1871 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_state *st)
1872 {
1873         int c = PF_IDHASH(st) % pfsync_buckets;
1874         return &sc->sc_buckets[c];
1875 }
1876
1877 static void
1878 pfsync_update_state(struct pf_state *st)
1879 {
1880         struct pfsync_softc *sc = V_pfsyncif;
1881         bool sync = false, ref = true;
1882         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1883
1884         PF_STATE_LOCK_ASSERT(st);
1885         PFSYNC_BUCKET_LOCK(b);
1886
1887         if (st->state_flags & PFSTATE_ACK)
1888                 pfsync_undefer_state(st, 0);
1889         if (st->state_flags & PFSTATE_NOSYNC) {
1890                 if (st->sync_state != PFSYNC_S_NONE)
1891                         pfsync_q_del(st, true, b);
1892                 PFSYNC_BUCKET_UNLOCK(b);
1893                 return;
1894         }
1895
1896         if (b->b_len == PFSYNC_MINPKT)
1897                 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1898
1899         switch (st->sync_state) {
1900         case PFSYNC_S_UPD_C:
1901         case PFSYNC_S_UPD:
1902         case PFSYNC_S_INS:
1903                 /* we're already handling it */
1904
1905                 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1906                         st->sync_updates++;
1907                         if (st->sync_updates >= sc->sc_maxupdates)
1908                                 sync = true;
1909                 }
1910                 break;
1911
1912         case PFSYNC_S_IACK:
1913                 pfsync_q_del(st, false, b);
1914                 ref = false;
1915                 /* FALLTHROUGH */
1916
1917         case PFSYNC_S_NONE:
1918                 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1919                 st->sync_updates = 0;
1920                 break;
1921
1922         default:
1923                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1924         }
1925
1926         if (sync || (time_uptime - st->pfsync_time) < 2)
1927                 pfsync_push(b);
1928
1929         PFSYNC_BUCKET_UNLOCK(b);
1930 }
1931
1932 static void
1933 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1934 {
1935         struct pfsync_softc *sc = V_pfsyncif;
1936         struct pfsync_bucket *b = &sc->sc_buckets[0];
1937         struct pfsync_upd_req_item *item;
1938         size_t nlen = sizeof(struct pfsync_upd_req);
1939
1940         PFSYNC_BUCKET_LOCK_ASSERT(b);
1941
1942         /*
1943          * This code does a bit to prevent multiple update requests for the
1944          * same state being generated. It searches current subheader queue,
1945          * but it doesn't lookup into queue of already packed datagrams.
1946          */
1947         TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
1948                 if (item->ur_msg.id == id &&
1949                     item->ur_msg.creatorid == creatorid)
1950                         return;
1951
1952         item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1953         if (item == NULL)
1954                 return; /* XXX stats */
1955
1956         item->ur_msg.id = id;
1957         item->ur_msg.creatorid = creatorid;
1958
1959         if (TAILQ_EMPTY(&b->b_upd_req_list))
1960                 nlen += sizeof(struct pfsync_subheader);
1961
1962         if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
1963                 pfsync_sendout(0, 0);
1964
1965                 nlen = sizeof(struct pfsync_subheader) +
1966                     sizeof(struct pfsync_upd_req);
1967         }
1968
1969         TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
1970         b->b_len += nlen;
1971
1972         pfsync_push(b);
1973 }
1974
1975 static bool
1976 pfsync_update_state_req(struct pf_state *st)
1977 {
1978         struct pfsync_softc *sc = V_pfsyncif;
1979         bool ref = true, full = false;
1980         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1981
1982         PF_STATE_LOCK_ASSERT(st);
1983         PFSYNC_BUCKET_LOCK(b);
1984
1985         if (st->state_flags & PFSTATE_NOSYNC) {
1986                 if (st->sync_state != PFSYNC_S_NONE)
1987                         pfsync_q_del(st, true, b);
1988                 PFSYNC_BUCKET_UNLOCK(b);
1989                 return (full);
1990         }
1991
1992         switch (st->sync_state) {
1993         case PFSYNC_S_UPD_C:
1994         case PFSYNC_S_IACK:
1995                 pfsync_q_del(st, false, b);
1996                 ref = false;
1997                 /* FALLTHROUGH */
1998
1999         case PFSYNC_S_NONE:
2000                 pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2001                 pfsync_push(b);
2002                 break;
2003
2004         case PFSYNC_S_INS:
2005         case PFSYNC_S_UPD:
2006         case PFSYNC_S_DEL:
2007                 /* we're already handling it */
2008                 break;
2009
2010         default:
2011                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2012         }
2013
2014         if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state))
2015                 full = true;
2016
2017         PFSYNC_BUCKET_UNLOCK(b);
2018
2019         return (full);
2020 }
2021
2022 static void
2023 pfsync_delete_state(struct pf_state *st)
2024 {
2025         struct pfsync_softc *sc = V_pfsyncif;
2026         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2027         bool ref = true;
2028
2029         PFSYNC_BUCKET_LOCK(b);
2030         if (st->state_flags & PFSTATE_ACK)
2031                 pfsync_undefer_state(st, 1);
2032         if (st->state_flags & PFSTATE_NOSYNC) {
2033                 if (st->sync_state != PFSYNC_S_NONE)
2034                         pfsync_q_del(st, true, b);
2035                 PFSYNC_BUCKET_UNLOCK(b);
2036                 return;
2037         }
2038
2039         if (b->b_len == PFSYNC_MINPKT)
2040                 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2041
2042         switch (st->sync_state) {
2043         case PFSYNC_S_INS:
2044                 /* We never got to tell the world so just forget about it. */
2045                 pfsync_q_del(st, true, b);
2046                 break;
2047
2048         case PFSYNC_S_UPD_C:
2049         case PFSYNC_S_UPD:
2050         case PFSYNC_S_IACK:
2051                 pfsync_q_del(st, false, b);
2052                 ref = false;
2053                 /* FALLTHROUGH */
2054
2055         case PFSYNC_S_NONE:
2056                 pfsync_q_ins(st, PFSYNC_S_DEL, ref);
2057                 break;
2058
2059         default:
2060                 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2061         }
2062
2063         PFSYNC_BUCKET_UNLOCK(b);
2064 }
2065
2066 static void
2067 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2068 {
2069         struct {
2070                 struct pfsync_subheader subh;
2071                 struct pfsync_clr clr;
2072         } __packed r;
2073
2074         bzero(&r, sizeof(r));
2075
2076         r.subh.action = PFSYNC_ACT_CLR;
2077         r.subh.count = htons(1);
2078         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2079
2080         strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2081         r.clr.creatorid = creatorid;
2082
2083         pfsync_send_plus(&r, sizeof(r));
2084 }
2085
2086 static void
2087 pfsync_q_ins(struct pf_state *st, int q, bool ref)
2088 {
2089         struct pfsync_softc *sc = V_pfsyncif;
2090         size_t nlen = pfsync_qs[q].len;
2091         struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2092
2093         PFSYNC_BUCKET_LOCK_ASSERT(b);
2094
2095         KASSERT(st->sync_state == PFSYNC_S_NONE,
2096                 ("%s: st->sync_state %u", __func__, st->sync_state));
2097         KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2098             b->b_len));
2099
2100         if (TAILQ_EMPTY(&b->b_qs[q]))
2101                 nlen += sizeof(struct pfsync_subheader);
2102
2103         if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2104                 pfsync_sendout(1, b->b_id);
2105
2106                 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2107         }
2108
2109         b->b_len += nlen;
2110         TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2111         st->sync_state = q;
2112         if (ref)
2113                 pf_ref_state(st);
2114 }
2115
2116 static void
2117 pfsync_q_del(struct pf_state *st, bool unref, struct pfsync_bucket *b)
2118 {
2119         int q = st->sync_state;
2120
2121         PFSYNC_BUCKET_LOCK_ASSERT(b);
2122         KASSERT(st->sync_state != PFSYNC_S_NONE,
2123                 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2124
2125         b->b_len -= pfsync_qs[q].len;
2126         TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2127         st->sync_state = PFSYNC_S_NONE;
2128         if (unref)
2129                 pf_release_state(st);
2130
2131         if (TAILQ_EMPTY(&b->b_qs[q]))
2132                 b->b_len -= sizeof(struct pfsync_subheader);
2133 }
2134
2135 static void
2136 pfsync_bulk_start(void)
2137 {
2138         struct pfsync_softc *sc = V_pfsyncif;
2139
2140         if (V_pf_status.debug >= PF_DEBUG_MISC)
2141                 printf("pfsync: received bulk update request\n");
2142
2143         PFSYNC_BLOCK(sc);
2144
2145         sc->sc_ureq_received = time_uptime;
2146         sc->sc_bulk_hashid = 0;
2147         sc->sc_bulk_stateid = 0;
2148         pfsync_bulk_status(PFSYNC_BUS_START);
2149         callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2150         PFSYNC_BUNLOCK(sc);
2151 }
2152
2153 static void
2154 pfsync_bulk_update(void *arg)
2155 {
2156         struct pfsync_softc *sc = arg;
2157         struct pf_state *s;
2158         int i, sent = 0;
2159
2160         PFSYNC_BLOCK_ASSERT(sc);
2161         CURVNET_SET(sc->sc_ifp->if_vnet);
2162
2163         /*
2164          * Start with last state from previous invocation.
2165          * It may had gone, in this case start from the
2166          * hash slot.
2167          */
2168         s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2169
2170         if (s != NULL)
2171                 i = PF_IDHASH(s);
2172         else
2173                 i = sc->sc_bulk_hashid;
2174
2175         for (; i <= pf_hashmask; i++) {
2176                 struct pf_idhash *ih = &V_pf_idhash[i];
2177
2178                 if (s != NULL)
2179                         PF_HASHROW_ASSERT(ih);
2180                 else {
2181                         PF_HASHROW_LOCK(ih);
2182                         s = LIST_FIRST(&ih->states);
2183                 }
2184
2185                 for (; s; s = LIST_NEXT(s, entry)) {
2186                         if (s->sync_state == PFSYNC_S_NONE &&
2187                             s->timeout < PFTM_MAX &&
2188                             s->pfsync_time <= sc->sc_ureq_received) {
2189                                 if (pfsync_update_state_req(s)) {
2190                                         /* We've filled a packet. */
2191                                         sc->sc_bulk_hashid = i;
2192                                         sc->sc_bulk_stateid = s->id;
2193                                         sc->sc_bulk_creatorid = s->creatorid;
2194                                         PF_HASHROW_UNLOCK(ih);
2195                                         callout_reset(&sc->sc_bulk_tmo, 1,
2196                                             pfsync_bulk_update, sc);
2197                                         goto full;
2198                                 }
2199                                 sent++;
2200                         }
2201                 }
2202                 PF_HASHROW_UNLOCK(ih);
2203         }
2204
2205         /* We're done. */
2206         pfsync_bulk_status(PFSYNC_BUS_END);
2207 full:
2208         CURVNET_RESTORE();
2209 }
2210
2211 static void
2212 pfsync_bulk_status(u_int8_t status)
2213 {
2214         struct {
2215                 struct pfsync_subheader subh;
2216                 struct pfsync_bus bus;
2217         } __packed r;
2218
2219         struct pfsync_softc *sc = V_pfsyncif;
2220
2221         bzero(&r, sizeof(r));
2222
2223         r.subh.action = PFSYNC_ACT_BUS;
2224         r.subh.count = htons(1);
2225         V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2226
2227         r.bus.creatorid = V_pf_status.hostid;
2228         r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2229         r.bus.status = status;
2230
2231         pfsync_send_plus(&r, sizeof(r));
2232 }
2233
2234 static void
2235 pfsync_bulk_fail(void *arg)
2236 {
2237         struct pfsync_softc *sc = arg;
2238         struct pfsync_bucket *b = &sc->sc_buckets[0];
2239
2240         CURVNET_SET(sc->sc_ifp->if_vnet);
2241
2242         PFSYNC_BLOCK_ASSERT(sc);
2243
2244         if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2245                 /* Try again */
2246                 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2247                     pfsync_bulk_fail, V_pfsyncif);
2248                 PFSYNC_BUCKET_LOCK(b);
2249                 pfsync_request_update(0, 0);
2250                 PFSYNC_BUCKET_UNLOCK(b);
2251         } else {
2252                 /* Pretend like the transfer was ok. */
2253                 sc->sc_ureq_sent = 0;
2254                 sc->sc_bulk_tries = 0;
2255                 PFSYNC_LOCK(sc);
2256                 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2257                         (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2258                             "pfsync bulk fail");
2259                 sc->sc_flags |= PFSYNCF_OK;
2260                 PFSYNC_UNLOCK(sc);
2261                 if (V_pf_status.debug >= PF_DEBUG_MISC)
2262                         printf("pfsync: failed to receive bulk update\n");
2263         }
2264
2265         CURVNET_RESTORE();
2266 }
2267
2268 static void
2269 pfsync_send_plus(void *plus, size_t pluslen)
2270 {
2271         struct pfsync_softc *sc = V_pfsyncif;
2272         struct pfsync_bucket *b = &sc->sc_buckets[0];
2273
2274         PFSYNC_BUCKET_LOCK(b);
2275
2276         if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2277                 pfsync_sendout(1, b->b_id);
2278
2279         b->b_plus = plus;
2280         b->b_len += (b->b_pluslen = pluslen);
2281
2282         pfsync_sendout(1, b->b_id);
2283         PFSYNC_BUCKET_UNLOCK(b);
2284 }
2285
2286 static void
2287 pfsync_timeout(void *arg)
2288 {
2289         struct pfsync_bucket *b = arg;
2290
2291         CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2292         PFSYNC_BUCKET_LOCK(b);
2293         pfsync_push(b);
2294         PFSYNC_BUCKET_UNLOCK(b);
2295         CURVNET_RESTORE();
2296 }
2297
2298 static void
2299 pfsync_push(struct pfsync_bucket *b)
2300 {
2301
2302         PFSYNC_BUCKET_LOCK_ASSERT(b);
2303
2304         b->b_flags |= PFSYNCF_BUCKET_PUSH;
2305         swi_sched(V_pfsync_swi_cookie, 0);
2306 }
2307
2308 static void
2309 pfsync_push_all(struct pfsync_softc *sc)
2310 {
2311         int c;
2312         struct pfsync_bucket *b;
2313
2314         for (c = 0; c < pfsync_buckets; c++) {
2315                 b = &sc->sc_buckets[c];
2316
2317                 PFSYNC_BUCKET_LOCK(b);
2318                 pfsync_push(b);
2319                 PFSYNC_BUCKET_UNLOCK(b);
2320         }
2321 }
2322
2323 static void
2324 pfsyncintr(void *arg)
2325 {
2326         struct epoch_tracker et;
2327         struct pfsync_softc *sc = arg;
2328         struct pfsync_bucket *b;
2329         struct mbuf *m, *n;
2330         int c;
2331
2332         NET_EPOCH_ENTER(et);
2333         CURVNET_SET(sc->sc_ifp->if_vnet);
2334
2335         for (c = 0; c < pfsync_buckets; c++) {
2336                 b = &sc->sc_buckets[c];
2337
2338                 PFSYNC_BUCKET_LOCK(b);
2339                 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2340                         pfsync_sendout(0, b->b_id);
2341                         b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2342                 }
2343                 _IF_DEQUEUE_ALL(&b->b_snd, m);
2344                 PFSYNC_BUCKET_UNLOCK(b);
2345
2346                 for (; m != NULL; m = n) {
2347                         n = m->m_nextpkt;
2348                         m->m_nextpkt = NULL;
2349
2350                         /*
2351                          * We distinguish between a deferral packet and our
2352                          * own pfsync packet based on M_SKIP_FIREWALL
2353                          * flag. This is XXX.
2354                          */
2355                         if (m->m_flags & M_SKIP_FIREWALL)
2356                                 ip_output(m, NULL, NULL, 0, NULL, NULL);
2357                         else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2358                             NULL) == 0)
2359                                 V_pfsyncstats.pfsyncs_opackets++;
2360                         else
2361                                 V_pfsyncstats.pfsyncs_oerrors++;
2362                 }
2363         }
2364         CURVNET_RESTORE();
2365         NET_EPOCH_EXIT(et);
2366 }
2367
2368 static int
2369 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2370     struct in_mfilter *imf)
2371 {
2372         struct ip_moptions *imo = &sc->sc_imo;
2373         int error;
2374
2375         if (!(ifp->if_flags & IFF_MULTICAST))
2376                 return (EADDRNOTAVAIL);
2377
2378         imo->imo_multicast_vif = -1;
2379
2380         if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2381             &imf->imf_inm)) != 0)
2382                 return (error);
2383
2384         ip_mfilter_init(&imo->imo_head);
2385         ip_mfilter_insert(&imo->imo_head, imf);
2386         imo->imo_multicast_ifp = ifp;
2387         imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2388         imo->imo_multicast_loop = 0;
2389
2390         return (0);
2391 }
2392
2393 static void
2394 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2395 {
2396         struct ip_moptions *imo = &sc->sc_imo;
2397         struct in_mfilter *imf;
2398
2399         while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2400                 ip_mfilter_remove(&imo->imo_head, imf);
2401                 in_leavegroup(imf->imf_inm, NULL);
2402                 ip_mfilter_free(imf);
2403         }
2404         imo->imo_multicast_ifp = NULL;
2405 }
2406
2407 void
2408 pfsync_detach_ifnet(struct ifnet *ifp)
2409 {
2410         struct pfsync_softc *sc = V_pfsyncif;
2411
2412         if (sc == NULL)
2413                 return;
2414
2415         PFSYNC_LOCK(sc);
2416
2417         if (sc->sc_sync_if == ifp) {
2418                 /* We don't need mutlicast cleanup here, because the interface
2419                  * is going away. We do need to ensure we don't try to do
2420                  * cleanup later.
2421                  */
2422                 ip_mfilter_init(&sc->sc_imo.imo_head);
2423                 sc->sc_imo.imo_multicast_ifp = NULL;
2424                 sc->sc_sync_if = NULL;
2425         }
2426
2427         PFSYNC_UNLOCK(sc);
2428 }
2429
2430 #ifdef INET
2431 extern  struct domain inetdomain;
2432 static struct protosw in_pfsync_protosw = {
2433         .pr_type =              SOCK_RAW,
2434         .pr_domain =            &inetdomain,
2435         .pr_protocol =          IPPROTO_PFSYNC,
2436         .pr_flags =             PR_ATOMIC|PR_ADDR,
2437         .pr_input =             pfsync_input,
2438         .pr_output =            rip_output,
2439         .pr_ctloutput =         rip_ctloutput,
2440         .pr_usrreqs =           &rip_usrreqs
2441 };
2442 #endif
2443
2444 static void
2445 pfsync_pointers_init()
2446 {
2447
2448         PF_RULES_WLOCK();
2449         V_pfsync_state_import_ptr = pfsync_state_import;
2450         V_pfsync_insert_state_ptr = pfsync_insert_state;
2451         V_pfsync_update_state_ptr = pfsync_update_state;
2452         V_pfsync_delete_state_ptr = pfsync_delete_state;
2453         V_pfsync_clear_states_ptr = pfsync_clear_states;
2454         V_pfsync_defer_ptr = pfsync_defer;
2455         PF_RULES_WUNLOCK();
2456 }
2457
2458 static void
2459 pfsync_pointers_uninit()
2460 {
2461
2462         PF_RULES_WLOCK();
2463         V_pfsync_state_import_ptr = NULL;
2464         V_pfsync_insert_state_ptr = NULL;
2465         V_pfsync_update_state_ptr = NULL;
2466         V_pfsync_delete_state_ptr = NULL;
2467         V_pfsync_clear_states_ptr = NULL;
2468         V_pfsync_defer_ptr = NULL;
2469         PF_RULES_WUNLOCK();
2470 }
2471
2472 static void
2473 vnet_pfsync_init(const void *unused __unused)
2474 {
2475         int error;
2476
2477         V_pfsync_cloner = if_clone_simple(pfsyncname,
2478             pfsync_clone_create, pfsync_clone_destroy, 1);
2479         error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
2480             SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2481         if (error) {
2482                 if_clone_detach(V_pfsync_cloner);
2483                 log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2484         }
2485
2486         pfsync_pointers_init();
2487 }
2488 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2489     vnet_pfsync_init, NULL);
2490
2491 static void
2492 vnet_pfsync_uninit(const void *unused __unused)
2493 {
2494         int ret;
2495
2496         pfsync_pointers_uninit();
2497
2498         if_clone_detach(V_pfsync_cloner);
2499         ret = swi_remove(V_pfsync_swi_cookie);
2500         MPASS(ret == 0);
2501         ret = intr_event_destroy(V_pfsync_swi_ie);
2502         MPASS(ret == 0);
2503 }
2504
2505 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
2506     vnet_pfsync_uninit, NULL);
2507
2508 static int
2509 pfsync_init()
2510 {
2511 #ifdef INET
2512         int error;
2513
2514         pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
2515
2516         error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2517         if (error)
2518                 return (error);
2519         error = ipproto_register(IPPROTO_PFSYNC);
2520         if (error) {
2521                 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2522                 return (error);
2523         }
2524 #endif
2525
2526         return (0);
2527 }
2528
2529 static void
2530 pfsync_uninit()
2531 {
2532         pfsync_detach_ifnet_ptr = NULL;
2533
2534 #ifdef INET
2535         ipproto_unregister(IPPROTO_PFSYNC);
2536         pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2537 #endif
2538 }
2539
2540 static int
2541 pfsync_modevent(module_t mod, int type, void *data)
2542 {
2543         int error = 0;
2544
2545         switch (type) {
2546         case MOD_LOAD:
2547                 error = pfsync_init();
2548                 break;
2549         case MOD_UNLOAD:
2550                 pfsync_uninit();
2551                 break;
2552         default:
2553                 error = EINVAL;
2554                 break;
2555         }
2556
2557         return (error);
2558 }
2559
2560 static moduledata_t pfsync_mod = {
2561         pfsyncname,
2562         pfsync_modevent,
2563         0
2564 };
2565
2566 #define PFSYNC_MODVER 1
2567
2568 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2569 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2570 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2571 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);