1 /*- 2 * Copyright (c) 2002 Michael Shalayeff 3 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 19 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 20 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 21 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 23 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 24 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 25 * THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 /*- 29 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org> 30 * 31 * Permission to use, copy, modify, and distribute this software for any 32 * purpose with or without fee is hereby granted, provided that the above 33 * copyright notice and this permission notice appear in all copies. 34 * 35 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 36 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 37 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 38 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 39 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 40 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 41 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 42 */ 43 44 /* 45 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $ 46 * 47 * Revisions picked from OpenBSD after revision 1.110 import: 48 * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input() 49 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates 50 * 1.120, 1.175 - use monotonic time_uptime 51 * 1.122 - reduce number of updates for non-TCP sessions 52 * 1.125, 1.127 - rewrite merge or stale processing 53 * 1.128 - cleanups 54 * 1.146 - bzero() mbuf before sparsely filling it with data 55 * 1.170 - SIOCSIFMTU checks 56 * 1.126, 1.142 - deferred packets processing 57 * 1.173 - correct expire time processing 58 */ 59 60 #include <sys/cdefs.h> 61 __FBSDID("$FreeBSD$"); 62 63 #include "opt_inet.h" 64 #include "opt_inet6.h" 65 #include "opt_pf.h" 66 67 #include <sys/param.h> 68 #include <sys/bus.h> 69 #include <sys/endian.h> 70 #include <sys/interrupt.h> 71 #include <sys/kernel.h> 72 #include <sys/lock.h> 73 #include <sys/mbuf.h> 74 #include <sys/module.h> 75 #include <sys/mutex.h> 76 #include <sys/priv.h> 77 #include <sys/protosw.h> 78 #include <sys/socket.h> 79 #include <sys/sockio.h> 80 #include <sys/sysctl.h> 81 82 #include <net/bpf.h> 83 #include <net/if.h> 84 #include <net/if_clone.h> 85 #include <net/if_types.h> 86 #include <net/pfvar.h> 87 #include <net/if_pfsync.h> 88 89 #include <netinet/if_ether.h> 90 #include <netinet/in.h> 91 #include <netinet/in_var.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_carp.h> 94 #include <netinet/ip_var.h> 95 #include <netinet/tcp.h> 96 #include <netinet/tcp_fsm.h> 97 #include <netinet/tcp_seq.h> 98 99 #define PFSYNC_MINPKT ( \ 100 sizeof(struct ip) + \ 101 sizeof(struct pfsync_header) + \ 102 sizeof(struct pfsync_subheader) + \ 103 sizeof(struct pfsync_eof)) 104 105 struct pfsync_pkt { 106 struct ip *ip; 107 struct in_addr src; 108 u_int8_t flags; 109 }; 110 111 static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *, 112 struct pfsync_state_peer *); 113 static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int); 114 static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int); 115 static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int); 116 static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int); 117 static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int); 118 static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int); 119 static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int); 120 static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int); 121 static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int); 122 static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int); 123 static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int); 124 static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int); 125 126 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = { 127 pfsync_in_clr, /* PFSYNC_ACT_CLR */ 128 pfsync_in_ins, /* PFSYNC_ACT_INS */ 129 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ 130 pfsync_in_upd, /* PFSYNC_ACT_UPD */ 131 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ 132 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ 133 pfsync_in_del, /* PFSYNC_ACT_DEL */ 134 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ 135 pfsync_in_error, /* PFSYNC_ACT_INS_F */ 136 pfsync_in_error, /* PFSYNC_ACT_DEL_F */ 137 pfsync_in_bus, /* PFSYNC_ACT_BUS */ 138 pfsync_in_tdb, /* PFSYNC_ACT_TDB */ 139 pfsync_in_eof /* PFSYNC_ACT_EOF */ 140 }; 141 142 struct pfsync_q { 143 void (*write)(struct pf_state *, void *); 144 size_t len; 145 u_int8_t action; 146 }; 147 148 /* we have one of these for every PFSYNC_S_ */ 149 static void pfsync_out_state(struct pf_state *, void *); 150 static void pfsync_out_iack(struct pf_state *, void *); 151 static void pfsync_out_upd_c(struct pf_state *, void *); 152 static void pfsync_out_del(struct pf_state *, void *); 153 154 static struct pfsync_q pfsync_qs[] = { 155 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS }, 156 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, 157 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD }, 158 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, 159 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } 160 }; 161 162 static void pfsync_q_ins(struct pf_state *, int); 163 static void pfsync_q_del(struct pf_state *); 164 165 static void pfsync_update_state(struct pf_state *); 166 167 struct pfsync_upd_req_item { 168 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; 169 struct pfsync_upd_req ur_msg; 170 }; 171 172 struct pfsync_deferral { 173 struct pfsync_softc *pd_sc; 174 TAILQ_ENTRY(pfsync_deferral) pd_entry; 175 u_int pd_refs; 176 struct callout pd_tmo; 177 178 struct pf_state *pd_st; 179 struct mbuf *pd_m; 180 }; 181 182 struct pfsync_softc { 183 /* Configuration */ 184 struct ifnet *sc_ifp; 185 struct ifnet *sc_sync_if; 186 struct ip_moptions sc_imo; 187 struct in_addr sc_sync_peer; 188 uint32_t sc_flags; 189 #define PFSYNCF_OK 0x00000001 190 #define PFSYNCF_DEFER 0x00000002 191 #define PFSYNCF_PUSH 0x00000004 192 uint8_t sc_maxupdates; 193 struct ip sc_template; 194 struct callout sc_tmo; 195 struct mtx sc_mtx; 196 197 /* Queued data */ 198 size_t sc_len; 199 TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT]; 200 TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list; 201 TAILQ_HEAD(, pfsync_deferral) sc_deferrals; 202 u_int sc_deferred; 203 void *sc_plus; 204 size_t sc_pluslen; 205 206 /* Bulk update info */ 207 struct mtx sc_bulk_mtx; 208 uint32_t sc_ureq_sent; 209 int sc_bulk_tries; 210 uint32_t sc_ureq_received; 211 int sc_bulk_hashid; 212 uint64_t sc_bulk_stateid; 213 uint32_t sc_bulk_creatorid; 214 struct callout sc_bulk_tmo; 215 struct callout sc_bulkfail_tmo; 216 }; 217 218 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 219 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 220 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 221 222 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 223 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 224 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 225 226 static const char pfsyncname[] = "pfsync"; 227 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); 228 static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL; 229 #define V_pfsyncif VNET(pfsyncif) 230 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL; 231 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 232 static VNET_DEFINE(struct pfsyncstats, pfsyncstats); 233 #define V_pfsyncstats VNET(pfsyncstats) 234 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW; 235 #define V_pfsync_carp_adj VNET(pfsync_carp_adj) 236 237 static void pfsync_timeout(void *); 238 static void pfsync_push(struct pfsync_softc *); 239 static void pfsyncintr(void *); 240 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 241 void *); 242 static void pfsync_multicast_cleanup(struct pfsync_softc *); 243 static void pfsync_pointers_init(void); 244 static void pfsync_pointers_uninit(void); 245 static int pfsync_init(void); 246 static void pfsync_uninit(void); 247 248 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC"); 249 SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW, 250 &VNET_NAME(pfsyncstats), pfsyncstats, 251 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 252 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW, 253 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 254 255 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 256 static void pfsync_clone_destroy(struct ifnet *); 257 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 258 struct pf_state_peer *); 259 static int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 260 struct route *); 261 static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 262 263 static int pfsync_defer(struct pf_state *, struct mbuf *); 264 static void pfsync_undefer(struct pfsync_deferral *, int); 265 static void pfsync_undefer_state(struct pf_state *, int); 266 static void pfsync_defer_tmo(void *); 267 268 static void pfsync_request_update(u_int32_t, u_int64_t); 269 static void pfsync_update_state_req(struct pf_state *); 270 271 static void pfsync_drop(struct pfsync_softc *); 272 static void pfsync_sendout(int); 273 static void pfsync_send_plus(void *, size_t); 274 275 static void pfsync_bulk_start(void); 276 static void pfsync_bulk_status(u_int8_t); 277 static void pfsync_bulk_update(void *); 278 static void pfsync_bulk_fail(void *); 279 280 #ifdef IPSEC 281 static void pfsync_update_net_tdb(struct pfsync_tdb *); 282 #endif 283 284 #define PFSYNC_MAX_BULKTRIES 12 285 286 VNET_DEFINE(struct if_clone *, pfsync_cloner); 287 #define V_pfsync_cloner VNET(pfsync_cloner) 288 289 static int 290 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 291 { 292 struct pfsync_softc *sc; 293 struct ifnet *ifp; 294 int q; 295 296 if (unit != 0) 297 return (EINVAL); 298 299 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 300 sc->sc_flags |= PFSYNCF_OK; 301 302 for (q = 0; q < PFSYNC_S_COUNT; q++) 303 TAILQ_INIT(&sc->sc_qs[q]); 304 305 TAILQ_INIT(&sc->sc_upd_req_list); 306 TAILQ_INIT(&sc->sc_deferrals); 307 308 sc->sc_len = PFSYNC_MINPKT; 309 sc->sc_maxupdates = 128; 310 311 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 312 if (ifp == NULL) { 313 free(sc, M_PFSYNC); 314 return (ENOSPC); 315 } 316 if_initname(ifp, pfsyncname, unit); 317 ifp->if_softc = sc; 318 ifp->if_ioctl = pfsyncioctl; 319 ifp->if_output = pfsyncoutput; 320 ifp->if_type = IFT_PFSYNC; 321 ifp->if_snd.ifq_maxlen = ifqmaxlen; 322 ifp->if_hdrlen = sizeof(struct pfsync_header); 323 ifp->if_mtu = ETHERMTU; 324 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); 325 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 326 callout_init(&sc->sc_tmo, CALLOUT_MPSAFE); 327 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 328 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 329 330 if_attach(ifp); 331 332 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 333 334 V_pfsyncif = sc; 335 336 return (0); 337 } 338 339 static void 340 pfsync_clone_destroy(struct ifnet *ifp) 341 { 342 struct pfsync_softc *sc = ifp->if_softc; 343 344 /* 345 * At this stage, everything should have already been 346 * cleared by pfsync_uninit(), and we have only to 347 * drain callouts. 348 */ 349 while (sc->sc_deferred > 0) { 350 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals); 351 352 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 353 sc->sc_deferred--; 354 if (callout_stop(&pd->pd_tmo)) { 355 pf_release_state(pd->pd_st); 356 m_freem(pd->pd_m); 357 free(pd, M_PFSYNC); 358 } else { 359 pd->pd_refs++; 360 callout_drain(&pd->pd_tmo); 361 free(pd, M_PFSYNC); 362 } 363 } 364 365 callout_drain(&sc->sc_tmo); 366 callout_drain(&sc->sc_bulkfail_tmo); 367 callout_drain(&sc->sc_bulk_tmo); 368 369 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 370 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 371 bpfdetach(ifp); 372 if_detach(ifp); 373 374 pfsync_drop(sc); 375 376 if_free(ifp); 377 if (sc->sc_imo.imo_membership) 378 pfsync_multicast_cleanup(sc); 379 mtx_destroy(&sc->sc_mtx); 380 mtx_destroy(&sc->sc_bulk_mtx); 381 free(sc, M_PFSYNC); 382 383 V_pfsyncif = NULL; 384 } 385 386 static int 387 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 388 struct pf_state_peer *d) 389 { 390 if (s->scrub.scrub_flag && d->scrub == NULL) { 391 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 392 if (d->scrub == NULL) 393 return (ENOMEM); 394 } 395 396 return (0); 397 } 398 399 400 static int 401 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 402 { 403 struct pfsync_softc *sc = V_pfsyncif; 404 struct pf_state *st = NULL; 405 struct pf_state_key *skw = NULL, *sks = NULL; 406 struct pf_rule *r = NULL; 407 struct pfi_kif *kif; 408 int error; 409 410 PF_RULES_RASSERT(); 411 412 if (sp->creatorid == 0 && V_pf_status.debug >= PF_DEBUG_MISC) { 413 printf("%s: invalid creator id: %08x\n", __func__, 414 ntohl(sp->creatorid)); 415 return (EINVAL); 416 } 417 418 if ((kif = pfi_kif_find(sp->ifname)) == NULL) { 419 if (V_pf_status.debug >= PF_DEBUG_MISC) 420 printf("%s: unknown interface: %s\n", __func__, 421 sp->ifname); 422 if (flags & PFSYNC_SI_IOCTL) 423 return (EINVAL); 424 return (0); /* skip this state */ 425 } 426 427 /* 428 * If the ruleset checksums match or the state is coming from the ioctl, 429 * it's safe to associate the state with the rule of that number. 430 */ 431 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 432 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 433 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 434 r = pf_main_ruleset.rules[ 435 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 436 else 437 r = &V_pf_default_rule; 438 439 if ((r->max_states && r->states_cur >= r->max_states)) 440 goto cleanup; 441 442 /* 443 * XXXGL: consider M_WAITOK in ioctl path after. 444 */ 445 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL) 446 goto cleanup; 447 448 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 449 goto cleanup; 450 451 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 452 &sp->key[PF_SK_STACK].addr[0], sp->af) || 453 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 454 &sp->key[PF_SK_STACK].addr[1], sp->af) || 455 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 456 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) { 457 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 458 if (sks == NULL) 459 goto cleanup; 460 } else 461 sks = skw; 462 463 /* allocate memory for scrub info */ 464 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 465 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 466 goto cleanup; 467 468 /* copy to state key(s) */ 469 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 470 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 471 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 472 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 473 skw->proto = sp->proto; 474 skw->af = sp->af; 475 if (sks != skw) { 476 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 477 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 478 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 479 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 480 sks->proto = sp->proto; 481 sks->af = sp->af; 482 } 483 484 /* copy to state */ 485 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 486 st->creation = time_uptime - ntohl(sp->creation); 487 st->expire = time_uptime; 488 if (sp->expire) { 489 uint32_t timeout; 490 491 timeout = r->timeout[sp->timeout]; 492 if (!timeout) 493 timeout = V_pf_default_rule.timeout[sp->timeout]; 494 495 /* sp->expire may have been adaptively scaled by export. */ 496 st->expire -= timeout - ntohl(sp->expire); 497 } 498 499 st->direction = sp->direction; 500 st->log = sp->log; 501 st->timeout = sp->timeout; 502 st->state_flags = sp->state_flags; 503 504 st->id = sp->id; 505 st->creatorid = sp->creatorid; 506 pf_state_peer_ntoh(&sp->src, &st->src); 507 pf_state_peer_ntoh(&sp->dst, &st->dst); 508 509 st->rule.ptr = r; 510 st->nat_rule.ptr = NULL; 511 st->anchor.ptr = NULL; 512 st->rt_kif = NULL; 513 514 st->pfsync_time = time_uptime; 515 st->sync_state = PFSYNC_S_NONE; 516 517 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 518 r->states_cur++; 519 r->states_tot++; 520 521 if (!(flags & PFSYNC_SI_IOCTL)) 522 st->state_flags |= PFSTATE_NOSYNC; 523 524 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) { 525 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 526 r->states_cur--; 527 goto cleanup_state; 528 } 529 530 if (!(flags & PFSYNC_SI_IOCTL)) { 531 st->state_flags &= ~PFSTATE_NOSYNC; 532 if (st->state_flags & PFSTATE_ACK) { 533 pfsync_q_ins(st, PFSYNC_S_IACK); 534 pfsync_push(sc); 535 } 536 } 537 st->state_flags &= ~PFSTATE_ACK; 538 PF_STATE_UNLOCK(st); 539 540 return (0); 541 542 cleanup: 543 error = ENOMEM; 544 if (skw == sks) 545 sks = NULL; 546 if (skw != NULL) 547 uma_zfree(V_pf_state_key_z, skw); 548 if (sks != NULL) 549 uma_zfree(V_pf_state_key_z, sks); 550 551 cleanup_state: /* pf_state_insert() frees the state keys. */ 552 if (st) { 553 if (st->dst.scrub) 554 uma_zfree(V_pf_state_scrub_z, st->dst.scrub); 555 if (st->src.scrub) 556 uma_zfree(V_pf_state_scrub_z, st->src.scrub); 557 uma_zfree(V_pf_state_z, st); 558 } 559 return (error); 560 } 561 562 static void 563 pfsync_input(struct mbuf *m, __unused int off) 564 { 565 struct pfsync_softc *sc = V_pfsyncif; 566 struct pfsync_pkt pkt; 567 struct ip *ip = mtod(m, struct ip *); 568 struct pfsync_header *ph; 569 struct pfsync_subheader subh; 570 571 int offset, len; 572 int rv; 573 uint16_t count; 574 575 V_pfsyncstats.pfsyncs_ipackets++; 576 577 /* Verify that we have a sync interface configured. */ 578 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 579 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 580 goto done; 581 582 /* verify that the packet came in on the right interface */ 583 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 584 V_pfsyncstats.pfsyncs_badif++; 585 goto done; 586 } 587 588 sc->sc_ifp->if_ipackets++; 589 sc->sc_ifp->if_ibytes += m->m_pkthdr.len; 590 /* verify that the IP TTL is 255. */ 591 if (ip->ip_ttl != PFSYNC_DFLTTL) { 592 V_pfsyncstats.pfsyncs_badttl++; 593 goto done; 594 } 595 596 offset = ip->ip_hl << 2; 597 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 598 V_pfsyncstats.pfsyncs_hdrops++; 599 goto done; 600 } 601 602 if (offset + sizeof(*ph) > m->m_len) { 603 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 604 V_pfsyncstats.pfsyncs_hdrops++; 605 return; 606 } 607 ip = mtod(m, struct ip *); 608 } 609 ph = (struct pfsync_header *)((char *)ip + offset); 610 611 /* verify the version */ 612 if (ph->version != PFSYNC_VERSION) { 613 V_pfsyncstats.pfsyncs_badver++; 614 goto done; 615 } 616 617 len = ntohs(ph->len) + offset; 618 if (m->m_pkthdr.len < len) { 619 V_pfsyncstats.pfsyncs_badlen++; 620 goto done; 621 } 622 623 /* Cheaper to grab this now than having to mess with mbufs later */ 624 pkt.ip = ip; 625 pkt.src = ip->ip_src; 626 pkt.flags = 0; 627 628 /* 629 * Trusting pf_chksum during packet processing, as well as seeking 630 * in interface name tree, require holding PF_RULES_RLOCK(). 631 */ 632 PF_RULES_RLOCK(); 633 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 634 pkt.flags |= PFSYNC_SI_CKSUM; 635 636 offset += sizeof(*ph); 637 while (offset <= len - sizeof(subh)) { 638 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 639 offset += sizeof(subh); 640 641 if (subh.action >= PFSYNC_ACT_MAX) { 642 V_pfsyncstats.pfsyncs_badact++; 643 PF_RULES_RUNLOCK(); 644 goto done; 645 } 646 647 count = ntohs(subh.count); 648 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 649 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); 650 if (rv == -1) { 651 PF_RULES_RUNLOCK(); 652 return; 653 } 654 655 offset += rv; 656 } 657 PF_RULES_RUNLOCK(); 658 659 done: 660 m_freem(m); 661 } 662 663 static int 664 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 665 { 666 struct pfsync_clr *clr; 667 struct mbuf *mp; 668 int len = sizeof(*clr) * count; 669 int i, offp; 670 u_int32_t creatorid; 671 672 mp = m_pulldown(m, offset, len, &offp); 673 if (mp == NULL) { 674 V_pfsyncstats.pfsyncs_badlen++; 675 return (-1); 676 } 677 clr = (struct pfsync_clr *)(mp->m_data + offp); 678 679 for (i = 0; i < count; i++) { 680 creatorid = clr[i].creatorid; 681 682 if (clr[i].ifname[0] != '\0' && 683 pfi_kif_find(clr[i].ifname) == NULL) 684 continue; 685 686 for (int i = 0; i <= V_pf_hashmask; i++) { 687 struct pf_idhash *ih = &V_pf_idhash[i]; 688 struct pf_state *s; 689 relock: 690 PF_HASHROW_LOCK(ih); 691 LIST_FOREACH(s, &ih->states, entry) { 692 if (s->creatorid == creatorid) { 693 s->state_flags |= PFSTATE_NOSYNC; 694 pf_unlink_state(s, PF_ENTER_LOCKED); 695 goto relock; 696 } 697 } 698 PF_HASHROW_UNLOCK(ih); 699 } 700 } 701 702 return (len); 703 } 704 705 static int 706 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 707 { 708 struct mbuf *mp; 709 struct pfsync_state *sa, *sp; 710 int len = sizeof(*sp) * count; 711 int i, offp; 712 713 mp = m_pulldown(m, offset, len, &offp); 714 if (mp == NULL) { 715 V_pfsyncstats.pfsyncs_badlen++; 716 return (-1); 717 } 718 sa = (struct pfsync_state *)(mp->m_data + offp); 719 720 for (i = 0; i < count; i++) { 721 sp = &sa[i]; 722 723 /* Check for invalid values. */ 724 if (sp->timeout >= PFTM_MAX || 725 sp->src.state > PF_TCPS_PROXY_DST || 726 sp->dst.state > PF_TCPS_PROXY_DST || 727 sp->direction > PF_OUT || 728 (sp->af != AF_INET && sp->af != AF_INET6)) { 729 if (V_pf_status.debug >= PF_DEBUG_MISC) 730 printf("%s: invalid value\n", __func__); 731 V_pfsyncstats.pfsyncs_badval++; 732 continue; 733 } 734 735 if (pfsync_state_import(sp, pkt->flags) == ENOMEM) 736 /* Drop out, but process the rest of the actions. */ 737 break; 738 } 739 740 return (len); 741 } 742 743 static int 744 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 745 { 746 struct pfsync_ins_ack *ia, *iaa; 747 struct pf_state *st; 748 749 struct mbuf *mp; 750 int len = count * sizeof(*ia); 751 int offp, i; 752 753 mp = m_pulldown(m, offset, len, &offp); 754 if (mp == NULL) { 755 V_pfsyncstats.pfsyncs_badlen++; 756 return (-1); 757 } 758 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 759 760 for (i = 0; i < count; i++) { 761 ia = &iaa[i]; 762 763 st = pf_find_state_byid(ia->id, ia->creatorid); 764 if (st == NULL) 765 continue; 766 767 if (st->state_flags & PFSTATE_ACK) { 768 PFSYNC_LOCK(V_pfsyncif); 769 pfsync_undefer_state(st, 0); 770 PFSYNC_UNLOCK(V_pfsyncif); 771 } 772 PF_STATE_UNLOCK(st); 773 } 774 /* 775 * XXX this is not yet implemented, but we know the size of the 776 * message so we can skip it. 777 */ 778 779 return (count * sizeof(struct pfsync_ins_ack)); 780 } 781 782 static int 783 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src, 784 struct pfsync_state_peer *dst) 785 { 786 int sync = 0; 787 788 PF_STATE_LOCK_ASSERT(st); 789 790 /* 791 * The state should never go backwards except 792 * for syn-proxy states. Neither should the 793 * sequence window slide backwards. 794 */ 795 if ((st->src.state > src->state && 796 (st->src.state < PF_TCPS_PROXY_SRC || 797 src->state >= PF_TCPS_PROXY_SRC)) || 798 799 (st->src.state == src->state && 800 SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))) 801 sync++; 802 else 803 pf_state_peer_ntoh(src, &st->src); 804 805 if ((st->dst.state > dst->state) || 806 807 (st->dst.state >= TCPS_SYN_SENT && 808 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) 809 sync++; 810 else 811 pf_state_peer_ntoh(dst, &st->dst); 812 813 return (sync); 814 } 815 816 static int 817 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 818 { 819 struct pfsync_softc *sc = V_pfsyncif; 820 struct pfsync_state *sa, *sp; 821 struct pf_state *st; 822 int sync; 823 824 struct mbuf *mp; 825 int len = count * sizeof(*sp); 826 int offp, i; 827 828 mp = m_pulldown(m, offset, len, &offp); 829 if (mp == NULL) { 830 V_pfsyncstats.pfsyncs_badlen++; 831 return (-1); 832 } 833 sa = (struct pfsync_state *)(mp->m_data + offp); 834 835 for (i = 0; i < count; i++) { 836 sp = &sa[i]; 837 838 /* check for invalid values */ 839 if (sp->timeout >= PFTM_MAX || 840 sp->src.state > PF_TCPS_PROXY_DST || 841 sp->dst.state > PF_TCPS_PROXY_DST) { 842 if (V_pf_status.debug >= PF_DEBUG_MISC) { 843 printf("pfsync_input: PFSYNC_ACT_UPD: " 844 "invalid value\n"); 845 } 846 V_pfsyncstats.pfsyncs_badval++; 847 continue; 848 } 849 850 st = pf_find_state_byid(sp->id, sp->creatorid); 851 if (st == NULL) { 852 /* insert the update */ 853 if (pfsync_state_import(sp, 0)) 854 V_pfsyncstats.pfsyncs_badstate++; 855 continue; 856 } 857 858 if (st->state_flags & PFSTATE_ACK) { 859 PFSYNC_LOCK(sc); 860 pfsync_undefer_state(st, 1); 861 PFSYNC_UNLOCK(sc); 862 } 863 864 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 865 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst); 866 else { 867 sync = 0; 868 869 /* 870 * Non-TCP protocol state machine always go 871 * forwards 872 */ 873 if (st->src.state > sp->src.state) 874 sync++; 875 else 876 pf_state_peer_ntoh(&sp->src, &st->src); 877 if (st->dst.state > sp->dst.state) 878 sync++; 879 else 880 pf_state_peer_ntoh(&sp->dst, &st->dst); 881 } 882 if (sync < 2) { 883 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 884 pf_state_peer_ntoh(&sp->dst, &st->dst); 885 st->expire = time_uptime; 886 st->timeout = sp->timeout; 887 } 888 st->pfsync_time = time_uptime; 889 890 if (sync) { 891 V_pfsyncstats.pfsyncs_stale++; 892 893 pfsync_update_state(st); 894 PF_STATE_UNLOCK(st); 895 PFSYNC_LOCK(sc); 896 pfsync_push(sc); 897 PFSYNC_UNLOCK(sc); 898 continue; 899 } 900 PF_STATE_UNLOCK(st); 901 } 902 903 return (len); 904 } 905 906 static int 907 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 908 { 909 struct pfsync_softc *sc = V_pfsyncif; 910 struct pfsync_upd_c *ua, *up; 911 struct pf_state *st; 912 int len = count * sizeof(*up); 913 int sync; 914 struct mbuf *mp; 915 int offp, i; 916 917 mp = m_pulldown(m, offset, len, &offp); 918 if (mp == NULL) { 919 V_pfsyncstats.pfsyncs_badlen++; 920 return (-1); 921 } 922 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 923 924 for (i = 0; i < count; i++) { 925 up = &ua[i]; 926 927 /* check for invalid values */ 928 if (up->timeout >= PFTM_MAX || 929 up->src.state > PF_TCPS_PROXY_DST || 930 up->dst.state > PF_TCPS_PROXY_DST) { 931 if (V_pf_status.debug >= PF_DEBUG_MISC) { 932 printf("pfsync_input: " 933 "PFSYNC_ACT_UPD_C: " 934 "invalid value\n"); 935 } 936 V_pfsyncstats.pfsyncs_badval++; 937 continue; 938 } 939 940 st = pf_find_state_byid(up->id, up->creatorid); 941 if (st == NULL) { 942 /* We don't have this state. Ask for it. */ 943 PFSYNC_LOCK(sc); 944 pfsync_request_update(up->creatorid, up->id); 945 PFSYNC_UNLOCK(sc); 946 continue; 947 } 948 949 if (st->state_flags & PFSTATE_ACK) { 950 PFSYNC_LOCK(sc); 951 pfsync_undefer_state(st, 1); 952 PFSYNC_UNLOCK(sc); 953 } 954 955 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 956 sync = pfsync_upd_tcp(st, &up->src, &up->dst); 957 else { 958 sync = 0; 959 960 /* 961 * Non-TCP protocol state machine always go 962 * forwards 963 */ 964 if (st->src.state > up->src.state) 965 sync++; 966 else 967 pf_state_peer_ntoh(&up->src, &st->src); 968 if (st->dst.state > up->dst.state) 969 sync++; 970 else 971 pf_state_peer_ntoh(&up->dst, &st->dst); 972 } 973 if (sync < 2) { 974 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 975 pf_state_peer_ntoh(&up->dst, &st->dst); 976 st->expire = time_uptime; 977 st->timeout = up->timeout; 978 } 979 st->pfsync_time = time_uptime; 980 981 if (sync) { 982 V_pfsyncstats.pfsyncs_stale++; 983 984 pfsync_update_state(st); 985 PF_STATE_UNLOCK(st); 986 PFSYNC_LOCK(sc); 987 pfsync_push(sc); 988 PFSYNC_UNLOCK(sc); 989 continue; 990 } 991 PF_STATE_UNLOCK(st); 992 } 993 994 return (len); 995 } 996 997 static int 998 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 999 { 1000 struct pfsync_upd_req *ur, *ura; 1001 struct mbuf *mp; 1002 int len = count * sizeof(*ur); 1003 int i, offp; 1004 1005 struct pf_state *st; 1006 1007 mp = m_pulldown(m, offset, len, &offp); 1008 if (mp == NULL) { 1009 V_pfsyncstats.pfsyncs_badlen++; 1010 return (-1); 1011 } 1012 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1013 1014 for (i = 0; i < count; i++) { 1015 ur = &ura[i]; 1016 1017 if (ur->id == 0 && ur->creatorid == 0) 1018 pfsync_bulk_start(); 1019 else { 1020 st = pf_find_state_byid(ur->id, ur->creatorid); 1021 if (st == NULL) { 1022 V_pfsyncstats.pfsyncs_badstate++; 1023 continue; 1024 } 1025 if (st->state_flags & PFSTATE_NOSYNC) { 1026 PF_STATE_UNLOCK(st); 1027 continue; 1028 } 1029 1030 pfsync_update_state_req(st); 1031 PF_STATE_UNLOCK(st); 1032 } 1033 } 1034 1035 return (len); 1036 } 1037 1038 static int 1039 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1040 { 1041 struct mbuf *mp; 1042 struct pfsync_state *sa, *sp; 1043 struct pf_state *st; 1044 int len = count * sizeof(*sp); 1045 int offp, i; 1046 1047 mp = m_pulldown(m, offset, len, &offp); 1048 if (mp == NULL) { 1049 V_pfsyncstats.pfsyncs_badlen++; 1050 return (-1); 1051 } 1052 sa = (struct pfsync_state *)(mp->m_data + offp); 1053 1054 for (i = 0; i < count; i++) { 1055 sp = &sa[i]; 1056 1057 st = pf_find_state_byid(sp->id, sp->creatorid); 1058 if (st == NULL) { 1059 V_pfsyncstats.pfsyncs_badstate++; 1060 continue; 1061 } 1062 st->state_flags |= PFSTATE_NOSYNC; 1063 pf_unlink_state(st, PF_ENTER_LOCKED); 1064 } 1065 1066 return (len); 1067 } 1068 1069 static int 1070 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1071 { 1072 struct mbuf *mp; 1073 struct pfsync_del_c *sa, *sp; 1074 struct pf_state *st; 1075 int len = count * sizeof(*sp); 1076 int offp, i; 1077 1078 mp = m_pulldown(m, offset, len, &offp); 1079 if (mp == NULL) { 1080 V_pfsyncstats.pfsyncs_badlen++; 1081 return (-1); 1082 } 1083 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1084 1085 for (i = 0; i < count; i++) { 1086 sp = &sa[i]; 1087 1088 st = pf_find_state_byid(sp->id, sp->creatorid); 1089 if (st == NULL) { 1090 V_pfsyncstats.pfsyncs_badstate++; 1091 continue; 1092 } 1093 1094 st->state_flags |= PFSTATE_NOSYNC; 1095 pf_unlink_state(st, PF_ENTER_LOCKED); 1096 } 1097 1098 return (len); 1099 } 1100 1101 static int 1102 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1103 { 1104 struct pfsync_softc *sc = V_pfsyncif; 1105 struct pfsync_bus *bus; 1106 struct mbuf *mp; 1107 int len = count * sizeof(*bus); 1108 int offp; 1109 1110 PFSYNC_BLOCK(sc); 1111 1112 /* If we're not waiting for a bulk update, who cares. */ 1113 if (sc->sc_ureq_sent == 0) { 1114 PFSYNC_BUNLOCK(sc); 1115 return (len); 1116 } 1117 1118 mp = m_pulldown(m, offset, len, &offp); 1119 if (mp == NULL) { 1120 PFSYNC_BUNLOCK(sc); 1121 V_pfsyncstats.pfsyncs_badlen++; 1122 return (-1); 1123 } 1124 bus = (struct pfsync_bus *)(mp->m_data + offp); 1125 1126 switch (bus->status) { 1127 case PFSYNC_BUS_START: 1128 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1129 V_pf_limits[PF_LIMIT_STATES].limit / 1130 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1131 sizeof(struct pfsync_state)), 1132 pfsync_bulk_fail, sc); 1133 if (V_pf_status.debug >= PF_DEBUG_MISC) 1134 printf("pfsync: received bulk update start\n"); 1135 break; 1136 1137 case PFSYNC_BUS_END: 1138 if (time_uptime - ntohl(bus->endtime) >= 1139 sc->sc_ureq_sent) { 1140 /* that's it, we're happy */ 1141 sc->sc_ureq_sent = 0; 1142 sc->sc_bulk_tries = 0; 1143 callout_stop(&sc->sc_bulkfail_tmo); 1144 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1145 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1146 "pfsync bulk done"); 1147 sc->sc_flags |= PFSYNCF_OK; 1148 if (V_pf_status.debug >= PF_DEBUG_MISC) 1149 printf("pfsync: received valid " 1150 "bulk update end\n"); 1151 } else { 1152 if (V_pf_status.debug >= PF_DEBUG_MISC) 1153 printf("pfsync: received invalid " 1154 "bulk update end: bad timestamp\n"); 1155 } 1156 break; 1157 } 1158 PFSYNC_BUNLOCK(sc); 1159 1160 return (len); 1161 } 1162 1163 static int 1164 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1165 { 1166 int len = count * sizeof(struct pfsync_tdb); 1167 1168 #if defined(IPSEC) 1169 struct pfsync_tdb *tp; 1170 struct mbuf *mp; 1171 int offp; 1172 int i; 1173 int s; 1174 1175 mp = m_pulldown(m, offset, len, &offp); 1176 if (mp == NULL) { 1177 V_pfsyncstats.pfsyncs_badlen++; 1178 return (-1); 1179 } 1180 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1181 1182 for (i = 0; i < count; i++) 1183 pfsync_update_net_tdb(&tp[i]); 1184 #endif 1185 1186 return (len); 1187 } 1188 1189 #if defined(IPSEC) 1190 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1191 static void 1192 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1193 { 1194 struct tdb *tdb; 1195 int s; 1196 1197 /* check for invalid values */ 1198 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1199 (pt->dst.sa.sa_family != AF_INET && 1200 pt->dst.sa.sa_family != AF_INET6)) 1201 goto bad; 1202 1203 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1204 if (tdb) { 1205 pt->rpl = ntohl(pt->rpl); 1206 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1207 1208 /* Neither replay nor byte counter should ever decrease. */ 1209 if (pt->rpl < tdb->tdb_rpl || 1210 pt->cur_bytes < tdb->tdb_cur_bytes) { 1211 goto bad; 1212 } 1213 1214 tdb->tdb_rpl = pt->rpl; 1215 tdb->tdb_cur_bytes = pt->cur_bytes; 1216 } 1217 return; 1218 1219 bad: 1220 if (V_pf_status.debug >= PF_DEBUG_MISC) 1221 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1222 "invalid value\n"); 1223 V_pfsyncstats.pfsyncs_badstate++; 1224 return; 1225 } 1226 #endif 1227 1228 1229 static int 1230 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1231 { 1232 /* check if we are at the right place in the packet */ 1233 if (offset != m->m_pkthdr.len) 1234 V_pfsyncstats.pfsyncs_badlen++; 1235 1236 /* we're done. free and let the caller return */ 1237 m_freem(m); 1238 return (-1); 1239 } 1240 1241 static int 1242 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1243 { 1244 V_pfsyncstats.pfsyncs_badact++; 1245 1246 m_freem(m); 1247 return (-1); 1248 } 1249 1250 static int 1251 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 1252 struct route *rt) 1253 { 1254 m_freem(m); 1255 return (0); 1256 } 1257 1258 /* ARGSUSED */ 1259 static int 1260 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1261 { 1262 struct pfsync_softc *sc = ifp->if_softc; 1263 struct ifreq *ifr = (struct ifreq *)data; 1264 struct pfsyncreq pfsyncr; 1265 int error; 1266 1267 switch (cmd) { 1268 case SIOCSIFFLAGS: 1269 PFSYNC_LOCK(sc); 1270 if (ifp->if_flags & IFF_UP) { 1271 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1272 PFSYNC_UNLOCK(sc); 1273 pfsync_pointers_init(); 1274 } else { 1275 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1276 PFSYNC_UNLOCK(sc); 1277 pfsync_pointers_uninit(); 1278 } 1279 break; 1280 case SIOCSIFMTU: 1281 if (!sc->sc_sync_if || 1282 ifr->ifr_mtu <= PFSYNC_MINPKT || 1283 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1284 return (EINVAL); 1285 if (ifr->ifr_mtu < ifp->if_mtu) { 1286 PFSYNC_LOCK(sc); 1287 if (sc->sc_len > PFSYNC_MINPKT) 1288 pfsync_sendout(1); 1289 PFSYNC_UNLOCK(sc); 1290 } 1291 ifp->if_mtu = ifr->ifr_mtu; 1292 break; 1293 case SIOCGETPFSYNC: 1294 bzero(&pfsyncr, sizeof(pfsyncr)); 1295 PFSYNC_LOCK(sc); 1296 if (sc->sc_sync_if) { 1297 strlcpy(pfsyncr.pfsyncr_syncdev, 1298 sc->sc_sync_if->if_xname, IFNAMSIZ); 1299 } 1300 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1301 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1302 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER == 1303 (sc->sc_flags & PFSYNCF_DEFER)); 1304 PFSYNC_UNLOCK(sc); 1305 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))); 1306 1307 case SIOCSETPFSYNC: 1308 { 1309 struct ip_moptions *imo = &sc->sc_imo; 1310 struct ifnet *sifp; 1311 struct ip *ip; 1312 void *mship = NULL; 1313 1314 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1315 return (error); 1316 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 1317 return (error); 1318 1319 if (pfsyncr.pfsyncr_maxupdates > 255) 1320 return (EINVAL); 1321 1322 if (pfsyncr.pfsyncr_syncdev[0] == 0) 1323 sifp = NULL; 1324 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) 1325 return (EINVAL); 1326 1327 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0 && sifp != NULL) 1328 mship = malloc((sizeof(struct in_multi *) * 1329 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO); 1330 1331 PFSYNC_LOCK(sc); 1332 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1333 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); 1334 else 1335 sc->sc_sync_peer.s_addr = 1336 pfsyncr.pfsyncr_syncpeer.s_addr; 1337 1338 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1339 if (pfsyncr.pfsyncr_defer) { 1340 sc->sc_flags |= PFSYNCF_DEFER; 1341 pfsync_defer_ptr = pfsync_defer; 1342 } else { 1343 sc->sc_flags &= ~PFSYNCF_DEFER; 1344 pfsync_defer_ptr = NULL; 1345 } 1346 1347 if (sifp == NULL) { 1348 if (sc->sc_sync_if) 1349 if_rele(sc->sc_sync_if); 1350 sc->sc_sync_if = NULL; 1351 if (imo->imo_membership) 1352 pfsync_multicast_cleanup(sc); 1353 PFSYNC_UNLOCK(sc); 1354 break; 1355 } 1356 1357 if (sc->sc_len > PFSYNC_MINPKT && 1358 (sifp->if_mtu < sc->sc_ifp->if_mtu || 1359 (sc->sc_sync_if != NULL && 1360 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 1361 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 1362 pfsync_sendout(1); 1363 1364 if (imo->imo_membership) 1365 pfsync_multicast_cleanup(sc); 1366 1367 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 1368 error = pfsync_multicast_setup(sc, sifp, mship); 1369 if (error) { 1370 if_rele(sifp); 1371 free(mship, M_PFSYNC); 1372 return (error); 1373 } 1374 } 1375 if (sc->sc_sync_if) 1376 if_rele(sc->sc_sync_if); 1377 sc->sc_sync_if = sifp; 1378 1379 ip = &sc->sc_template; 1380 bzero(ip, sizeof(*ip)); 1381 ip->ip_v = IPVERSION; 1382 ip->ip_hl = sizeof(sc->sc_template) >> 2; 1383 ip->ip_tos = IPTOS_LOWDELAY; 1384 /* len and id are set later. */ 1385 ip->ip_off = htons(IP_DF); 1386 ip->ip_ttl = PFSYNC_DFLTTL; 1387 ip->ip_p = IPPROTO_PFSYNC; 1388 ip->ip_src.s_addr = INADDR_ANY; 1389 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; 1390 1391 /* Request a full state table update. */ 1392 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1393 (*carp_demote_adj_p)(V_pfsync_carp_adj, 1394 "pfsync bulk start"); 1395 sc->sc_flags &= ~PFSYNCF_OK; 1396 if (V_pf_status.debug >= PF_DEBUG_MISC) 1397 printf("pfsync: requesting bulk update\n"); 1398 pfsync_request_update(0, 0); 1399 PFSYNC_UNLOCK(sc); 1400 PFSYNC_BLOCK(sc); 1401 sc->sc_ureq_sent = time_uptime; 1402 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, 1403 sc); 1404 PFSYNC_BUNLOCK(sc); 1405 1406 break; 1407 } 1408 default: 1409 return (ENOTTY); 1410 } 1411 1412 return (0); 1413 } 1414 1415 static void 1416 pfsync_out_state(struct pf_state *st, void *buf) 1417 { 1418 struct pfsync_state *sp = buf; 1419 1420 pfsync_state_export(sp, st); 1421 } 1422 1423 static void 1424 pfsync_out_iack(struct pf_state *st, void *buf) 1425 { 1426 struct pfsync_ins_ack *iack = buf; 1427 1428 iack->id = st->id; 1429 iack->creatorid = st->creatorid; 1430 } 1431 1432 static void 1433 pfsync_out_upd_c(struct pf_state *st, void *buf) 1434 { 1435 struct pfsync_upd_c *up = buf; 1436 1437 bzero(up, sizeof(*up)); 1438 up->id = st->id; 1439 pf_state_peer_hton(&st->src, &up->src); 1440 pf_state_peer_hton(&st->dst, &up->dst); 1441 up->creatorid = st->creatorid; 1442 up->timeout = st->timeout; 1443 } 1444 1445 static void 1446 pfsync_out_del(struct pf_state *st, void *buf) 1447 { 1448 struct pfsync_del_c *dp = buf; 1449 1450 dp->id = st->id; 1451 dp->creatorid = st->creatorid; 1452 st->state_flags |= PFSTATE_NOSYNC; 1453 } 1454 1455 static void 1456 pfsync_drop(struct pfsync_softc *sc) 1457 { 1458 struct pf_state *st, *next; 1459 struct pfsync_upd_req_item *ur; 1460 int q; 1461 1462 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1463 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1464 continue; 1465 1466 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) { 1467 KASSERT(st->sync_state == q, 1468 ("%s: st->sync_state == q", 1469 __func__)); 1470 st->sync_state = PFSYNC_S_NONE; 1471 pf_release_state(st); 1472 } 1473 TAILQ_INIT(&sc->sc_qs[q]); 1474 } 1475 1476 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1477 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1478 free(ur, M_PFSYNC); 1479 } 1480 1481 sc->sc_plus = NULL; 1482 sc->sc_len = PFSYNC_MINPKT; 1483 } 1484 1485 static void 1486 pfsync_sendout(int schedswi) 1487 { 1488 struct pfsync_softc *sc = V_pfsyncif; 1489 struct ifnet *ifp = sc->sc_ifp; 1490 struct mbuf *m; 1491 struct ip *ip; 1492 struct pfsync_header *ph; 1493 struct pfsync_subheader *subh; 1494 struct pf_state *st; 1495 struct pfsync_upd_req_item *ur; 1496 int offset; 1497 int q, count = 0; 1498 1499 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1500 KASSERT(sc->sc_len > PFSYNC_MINPKT, 1501 ("%s: sc_len %zu", __func__, sc->sc_len)); 1502 PFSYNC_LOCK_ASSERT(sc); 1503 1504 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1505 pfsync_drop(sc); 1506 return; 1507 } 1508 1509 m = m_get2(M_NOWAIT, MT_DATA, M_PKTHDR, max_linkhdr + sc->sc_len); 1510 if (m == NULL) { 1511 sc->sc_ifp->if_oerrors++; 1512 V_pfsyncstats.pfsyncs_onomem++; 1513 return; 1514 } 1515 m->m_data += max_linkhdr; 1516 m->m_len = m->m_pkthdr.len = sc->sc_len; 1517 1518 /* build the ip header */ 1519 ip = (struct ip *)m->m_data; 1520 bcopy(&sc->sc_template, ip, sizeof(*ip)); 1521 offset = sizeof(*ip); 1522 1523 ip->ip_len = htons(m->m_pkthdr.len); 1524 ip->ip_id = htons(ip_randomid()); 1525 1526 /* build the pfsync header */ 1527 ph = (struct pfsync_header *)(m->m_data + offset); 1528 bzero(ph, sizeof(*ph)); 1529 offset += sizeof(*ph); 1530 1531 ph->version = PFSYNC_VERSION; 1532 ph->len = htons(sc->sc_len - sizeof(*ip)); 1533 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1534 1535 /* walk the queues */ 1536 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1537 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1538 continue; 1539 1540 subh = (struct pfsync_subheader *)(m->m_data + offset); 1541 offset += sizeof(*subh); 1542 1543 count = 0; 1544 TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) { 1545 KASSERT(st->sync_state == q, 1546 ("%s: st->sync_state == q", 1547 __func__)); 1548 /* 1549 * XXXGL: some of write methods do unlocked reads 1550 * of state data :( 1551 */ 1552 pfsync_qs[q].write(st, m->m_data + offset); 1553 offset += pfsync_qs[q].len; 1554 st->sync_state = PFSYNC_S_NONE; 1555 pf_release_state(st); 1556 count++; 1557 } 1558 TAILQ_INIT(&sc->sc_qs[q]); 1559 1560 bzero(subh, sizeof(*subh)); 1561 subh->action = pfsync_qs[q].action; 1562 subh->count = htons(count); 1563 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1564 } 1565 1566 if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) { 1567 subh = (struct pfsync_subheader *)(m->m_data + offset); 1568 offset += sizeof(*subh); 1569 1570 count = 0; 1571 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1572 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1573 1574 bcopy(&ur->ur_msg, m->m_data + offset, 1575 sizeof(ur->ur_msg)); 1576 offset += sizeof(ur->ur_msg); 1577 free(ur, M_PFSYNC); 1578 count++; 1579 } 1580 1581 bzero(subh, sizeof(*subh)); 1582 subh->action = PFSYNC_ACT_UPD_REQ; 1583 subh->count = htons(count); 1584 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1585 } 1586 1587 /* has someone built a custom region for us to add? */ 1588 if (sc->sc_plus != NULL) { 1589 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen); 1590 offset += sc->sc_pluslen; 1591 1592 sc->sc_plus = NULL; 1593 } 1594 1595 subh = (struct pfsync_subheader *)(m->m_data + offset); 1596 offset += sizeof(*subh); 1597 1598 bzero(subh, sizeof(*subh)); 1599 subh->action = PFSYNC_ACT_EOF; 1600 subh->count = htons(1); 1601 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1602 1603 /* we're done, let's put it on the wire */ 1604 if (ifp->if_bpf) { 1605 m->m_data += sizeof(*ip); 1606 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip); 1607 BPF_MTAP(ifp, m); 1608 m->m_data -= sizeof(*ip); 1609 m->m_len = m->m_pkthdr.len = sc->sc_len; 1610 } 1611 1612 if (sc->sc_sync_if == NULL) { 1613 sc->sc_len = PFSYNC_MINPKT; 1614 m_freem(m); 1615 return; 1616 } 1617 1618 sc->sc_ifp->if_opackets++; 1619 sc->sc_ifp->if_obytes += m->m_pkthdr.len; 1620 sc->sc_len = PFSYNC_MINPKT; 1621 1622 if (!_IF_QFULL(&sc->sc_ifp->if_snd)) 1623 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1624 else { 1625 m_freem(m); 1626 sc->sc_ifp->if_snd.ifq_drops++; 1627 } 1628 if (schedswi) 1629 swi_sched(V_pfsync_swi_cookie, 0); 1630 } 1631 1632 static void 1633 pfsync_insert_state(struct pf_state *st) 1634 { 1635 struct pfsync_softc *sc = V_pfsyncif; 1636 1637 if (st->state_flags & PFSTATE_NOSYNC) 1638 return; 1639 1640 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1641 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1642 st->state_flags |= PFSTATE_NOSYNC; 1643 return; 1644 } 1645 1646 KASSERT(st->sync_state == PFSYNC_S_NONE, 1647 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1648 1649 PFSYNC_LOCK(sc); 1650 if (sc->sc_len == PFSYNC_MINPKT) 1651 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1652 1653 pfsync_q_ins(st, PFSYNC_S_INS); 1654 PFSYNC_UNLOCK(sc); 1655 1656 st->sync_updates = 0; 1657 } 1658 1659 static int 1660 pfsync_defer(struct pf_state *st, struct mbuf *m) 1661 { 1662 struct pfsync_softc *sc = V_pfsyncif; 1663 struct pfsync_deferral *pd; 1664 1665 if (m->m_flags & (M_BCAST|M_MCAST)) 1666 return (0); 1667 1668 PFSYNC_LOCK(sc); 1669 1670 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) || 1671 !(sc->sc_flags & PFSYNCF_DEFER)) { 1672 PFSYNC_UNLOCK(sc); 1673 return (0); 1674 } 1675 1676 if (sc->sc_deferred >= 128) 1677 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0); 1678 1679 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1680 if (pd == NULL) 1681 return (0); 1682 sc->sc_deferred++; 1683 1684 m->m_flags |= M_SKIP_FIREWALL; 1685 st->state_flags |= PFSTATE_ACK; 1686 1687 pd->pd_sc = sc; 1688 pd->pd_refs = 0; 1689 pd->pd_st = st; 1690 pf_ref_state(st); 1691 pd->pd_m = m; 1692 1693 TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry); 1694 callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1695 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd); 1696 1697 pfsync_push(sc); 1698 1699 return (1); 1700 } 1701 1702 static void 1703 pfsync_undefer(struct pfsync_deferral *pd, int drop) 1704 { 1705 struct pfsync_softc *sc = pd->pd_sc; 1706 struct mbuf *m = pd->pd_m; 1707 struct pf_state *st = pd->pd_st; 1708 1709 PFSYNC_LOCK_ASSERT(sc); 1710 1711 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1712 sc->sc_deferred--; 1713 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1714 free(pd, M_PFSYNC); 1715 pf_release_state(st); 1716 1717 if (drop) 1718 m_freem(m); 1719 else { 1720 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1721 pfsync_push(sc); 1722 } 1723 } 1724 1725 static void 1726 pfsync_defer_tmo(void *arg) 1727 { 1728 struct pfsync_deferral *pd = arg; 1729 struct pfsync_softc *sc = pd->pd_sc; 1730 struct mbuf *m = pd->pd_m; 1731 struct pf_state *st = pd->pd_st; 1732 1733 PFSYNC_LOCK_ASSERT(sc); 1734 1735 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); 1736 1737 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1738 sc->sc_deferred--; 1739 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1740 if (pd->pd_refs == 0) 1741 free(pd, M_PFSYNC); 1742 PFSYNC_UNLOCK(sc); 1743 1744 ip_output(m, NULL, NULL, 0, NULL, NULL); 1745 1746 pf_release_state(st); 1747 1748 CURVNET_RESTORE(); 1749 } 1750 1751 static void 1752 pfsync_undefer_state(struct pf_state *st, int drop) 1753 { 1754 struct pfsync_softc *sc = V_pfsyncif; 1755 struct pfsync_deferral *pd; 1756 1757 PFSYNC_LOCK_ASSERT(sc); 1758 1759 TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) { 1760 if (pd->pd_st == st) { 1761 if (callout_stop(&pd->pd_tmo)) 1762 pfsync_undefer(pd, drop); 1763 return; 1764 } 1765 } 1766 1767 panic("%s: unable to find deferred state", __func__); 1768 } 1769 1770 static void 1771 pfsync_update_state(struct pf_state *st) 1772 { 1773 struct pfsync_softc *sc = V_pfsyncif; 1774 int sync = 0; 1775 1776 PF_STATE_LOCK_ASSERT(st); 1777 PFSYNC_LOCK(sc); 1778 1779 if (st->state_flags & PFSTATE_ACK) 1780 pfsync_undefer_state(st, 0); 1781 if (st->state_flags & PFSTATE_NOSYNC) { 1782 if (st->sync_state != PFSYNC_S_NONE) 1783 pfsync_q_del(st); 1784 PFSYNC_UNLOCK(sc); 1785 return; 1786 } 1787 1788 if (sc->sc_len == PFSYNC_MINPKT) 1789 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1790 1791 switch (st->sync_state) { 1792 case PFSYNC_S_UPD_C: 1793 case PFSYNC_S_UPD: 1794 case PFSYNC_S_INS: 1795 /* we're already handling it */ 1796 1797 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 1798 st->sync_updates++; 1799 if (st->sync_updates >= sc->sc_maxupdates) 1800 sync = 1; 1801 } 1802 break; 1803 1804 case PFSYNC_S_IACK: 1805 pfsync_q_del(st); 1806 case PFSYNC_S_NONE: 1807 pfsync_q_ins(st, PFSYNC_S_UPD_C); 1808 st->sync_updates = 0; 1809 break; 1810 1811 default: 1812 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1813 } 1814 1815 if (sync || (time_uptime - st->pfsync_time) < 2) 1816 pfsync_push(sc); 1817 1818 PFSYNC_UNLOCK(sc); 1819 } 1820 1821 static void 1822 pfsync_request_update(u_int32_t creatorid, u_int64_t id) 1823 { 1824 struct pfsync_softc *sc = V_pfsyncif; 1825 struct pfsync_upd_req_item *item; 1826 size_t nlen = sizeof(struct pfsync_upd_req); 1827 1828 PFSYNC_LOCK_ASSERT(sc); 1829 1830 /* 1831 * This code does a bit to prevent multiple update requests for the 1832 * same state being generated. It searches current subheader queue, 1833 * but it doesn't lookup into queue of already packed datagrams. 1834 */ 1835 TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry) 1836 if (item->ur_msg.id == id && 1837 item->ur_msg.creatorid == creatorid) 1838 return; 1839 1840 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 1841 if (item == NULL) 1842 return; /* XXX stats */ 1843 1844 item->ur_msg.id = id; 1845 item->ur_msg.creatorid = creatorid; 1846 1847 if (TAILQ_EMPTY(&sc->sc_upd_req_list)) 1848 nlen += sizeof(struct pfsync_subheader); 1849 1850 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1851 pfsync_sendout(1); 1852 1853 nlen = sizeof(struct pfsync_subheader) + 1854 sizeof(struct pfsync_upd_req); 1855 } 1856 1857 TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry); 1858 sc->sc_len += nlen; 1859 } 1860 1861 static void 1862 pfsync_update_state_req(struct pf_state *st) 1863 { 1864 struct pfsync_softc *sc = V_pfsyncif; 1865 1866 PF_STATE_LOCK_ASSERT(st); 1867 PFSYNC_LOCK(sc); 1868 1869 if (st->state_flags & PFSTATE_NOSYNC) { 1870 if (st->sync_state != PFSYNC_S_NONE) 1871 pfsync_q_del(st); 1872 PFSYNC_UNLOCK(sc); 1873 return; 1874 } 1875 1876 switch (st->sync_state) { 1877 case PFSYNC_S_UPD_C: 1878 case PFSYNC_S_IACK: 1879 pfsync_q_del(st); 1880 case PFSYNC_S_NONE: 1881 pfsync_q_ins(st, PFSYNC_S_UPD); 1882 pfsync_push(sc); 1883 break; 1884 1885 case PFSYNC_S_INS: 1886 case PFSYNC_S_UPD: 1887 case PFSYNC_S_DEL: 1888 /* we're already handling it */ 1889 break; 1890 1891 default: 1892 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1893 } 1894 1895 PFSYNC_UNLOCK(sc); 1896 } 1897 1898 static void 1899 pfsync_delete_state(struct pf_state *st) 1900 { 1901 struct pfsync_softc *sc = V_pfsyncif; 1902 1903 PFSYNC_LOCK(sc); 1904 if (st->state_flags & PFSTATE_ACK) 1905 pfsync_undefer_state(st, 1); 1906 if (st->state_flags & PFSTATE_NOSYNC) { 1907 if (st->sync_state != PFSYNC_S_NONE) 1908 pfsync_q_del(st); 1909 PFSYNC_UNLOCK(sc); 1910 return; 1911 } 1912 1913 if (sc->sc_len == PFSYNC_MINPKT) 1914 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1915 1916 switch (st->sync_state) { 1917 case PFSYNC_S_INS: 1918 /* We never got to tell the world so just forget about it. */ 1919 pfsync_q_del(st); 1920 break; 1921 1922 case PFSYNC_S_UPD_C: 1923 case PFSYNC_S_UPD: 1924 case PFSYNC_S_IACK: 1925 pfsync_q_del(st); 1926 /* FALLTHROUGH to putting it on the del list */ 1927 1928 case PFSYNC_S_NONE: 1929 pfsync_q_ins(st, PFSYNC_S_DEL); 1930 break; 1931 1932 default: 1933 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1934 } 1935 PFSYNC_UNLOCK(sc); 1936 } 1937 1938 static void 1939 pfsync_clear_states(u_int32_t creatorid, const char *ifname) 1940 { 1941 struct pfsync_softc *sc = V_pfsyncif; 1942 struct { 1943 struct pfsync_subheader subh; 1944 struct pfsync_clr clr; 1945 } __packed r; 1946 1947 bzero(&r, sizeof(r)); 1948 1949 r.subh.action = PFSYNC_ACT_CLR; 1950 r.subh.count = htons(1); 1951 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 1952 1953 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 1954 r.clr.creatorid = creatorid; 1955 1956 PFSYNC_LOCK(sc); 1957 pfsync_send_plus(&r, sizeof(r)); 1958 PFSYNC_UNLOCK(sc); 1959 } 1960 1961 static void 1962 pfsync_q_ins(struct pf_state *st, int q) 1963 { 1964 struct pfsync_softc *sc = V_pfsyncif; 1965 size_t nlen = pfsync_qs[q].len; 1966 1967 PFSYNC_LOCK_ASSERT(sc); 1968 1969 KASSERT(st->sync_state == PFSYNC_S_NONE, 1970 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1971 KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 1972 sc->sc_len)); 1973 1974 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1975 nlen += sizeof(struct pfsync_subheader); 1976 1977 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1978 pfsync_sendout(1); 1979 1980 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 1981 } 1982 1983 sc->sc_len += nlen; 1984 TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list); 1985 st->sync_state = q; 1986 pf_ref_state(st); 1987 } 1988 1989 static void 1990 pfsync_q_del(struct pf_state *st) 1991 { 1992 struct pfsync_softc *sc = V_pfsyncif; 1993 int q = st->sync_state; 1994 1995 PFSYNC_LOCK_ASSERT(sc); 1996 KASSERT(st->sync_state != PFSYNC_S_NONE, 1997 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 1998 1999 sc->sc_len -= pfsync_qs[q].len; 2000 TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list); 2001 st->sync_state = PFSYNC_S_NONE; 2002 pf_release_state(st); 2003 2004 if (TAILQ_EMPTY(&sc->sc_qs[q])) 2005 sc->sc_len -= sizeof(struct pfsync_subheader); 2006 } 2007 2008 static void 2009 pfsync_bulk_start(void) 2010 { 2011 struct pfsync_softc *sc = V_pfsyncif; 2012 2013 if (V_pf_status.debug >= PF_DEBUG_MISC) 2014 printf("pfsync: received bulk update request\n"); 2015 2016 PFSYNC_BLOCK(sc); 2017 2018 sc->sc_ureq_received = time_uptime; 2019 sc->sc_bulk_hashid = 0; 2020 sc->sc_bulk_stateid = 0; 2021 pfsync_bulk_status(PFSYNC_BUS_START); 2022 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2023 PFSYNC_BUNLOCK(sc); 2024 } 2025 2026 static void 2027 pfsync_bulk_update(void *arg) 2028 { 2029 struct pfsync_softc *sc = arg; 2030 struct pf_state *s; 2031 int i, sent = 0; 2032 2033 PFSYNC_BLOCK_ASSERT(sc); 2034 CURVNET_SET(sc->sc_ifp->if_vnet); 2035 2036 /* 2037 * Start with last state from previous invocation. 2038 * It may had gone, in this case start from the 2039 * hash slot. 2040 */ 2041 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2042 2043 if (s != NULL) 2044 i = PF_IDHASH(s); 2045 else 2046 i = sc->sc_bulk_hashid; 2047 2048 for (; i <= V_pf_hashmask; i++) { 2049 struct pf_idhash *ih = &V_pf_idhash[i]; 2050 2051 if (s != NULL) 2052 PF_HASHROW_ASSERT(ih); 2053 else { 2054 PF_HASHROW_LOCK(ih); 2055 s = LIST_FIRST(&ih->states); 2056 } 2057 2058 for (; s; s = LIST_NEXT(s, entry)) { 2059 2060 if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) < 2061 sizeof(struct pfsync_state)) { 2062 /* We've filled a packet. */ 2063 sc->sc_bulk_hashid = i; 2064 sc->sc_bulk_stateid = s->id; 2065 sc->sc_bulk_creatorid = s->creatorid; 2066 PF_HASHROW_UNLOCK(ih); 2067 callout_reset(&sc->sc_bulk_tmo, 1, 2068 pfsync_bulk_update, sc); 2069 goto full; 2070 } 2071 2072 if (s->sync_state == PFSYNC_S_NONE && 2073 s->timeout < PFTM_MAX && 2074 s->pfsync_time <= sc->sc_ureq_received) { 2075 pfsync_update_state_req(s); 2076 sent++; 2077 } 2078 } 2079 PF_HASHROW_UNLOCK(ih); 2080 } 2081 2082 /* We're done. */ 2083 pfsync_bulk_status(PFSYNC_BUS_END); 2084 2085 full: 2086 CURVNET_RESTORE(); 2087 } 2088 2089 static void 2090 pfsync_bulk_status(u_int8_t status) 2091 { 2092 struct { 2093 struct pfsync_subheader subh; 2094 struct pfsync_bus bus; 2095 } __packed r; 2096 2097 struct pfsync_softc *sc = V_pfsyncif; 2098 2099 bzero(&r, sizeof(r)); 2100 2101 r.subh.action = PFSYNC_ACT_BUS; 2102 r.subh.count = htons(1); 2103 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2104 2105 r.bus.creatorid = V_pf_status.hostid; 2106 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2107 r.bus.status = status; 2108 2109 PFSYNC_LOCK(sc); 2110 pfsync_send_plus(&r, sizeof(r)); 2111 PFSYNC_UNLOCK(sc); 2112 } 2113 2114 static void 2115 pfsync_bulk_fail(void *arg) 2116 { 2117 struct pfsync_softc *sc = arg; 2118 2119 CURVNET_SET(sc->sc_ifp->if_vnet); 2120 2121 PFSYNC_BLOCK_ASSERT(sc); 2122 2123 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2124 /* Try again */ 2125 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2126 pfsync_bulk_fail, V_pfsyncif); 2127 PFSYNC_LOCK(sc); 2128 pfsync_request_update(0, 0); 2129 PFSYNC_UNLOCK(sc); 2130 } else { 2131 /* Pretend like the transfer was ok. */ 2132 sc->sc_ureq_sent = 0; 2133 sc->sc_bulk_tries = 0; 2134 PFSYNC_LOCK(sc); 2135 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2136 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2137 "pfsync bulk fail"); 2138 sc->sc_flags |= PFSYNCF_OK; 2139 PFSYNC_UNLOCK(sc); 2140 if (V_pf_status.debug >= PF_DEBUG_MISC) 2141 printf("pfsync: failed to receive bulk update\n"); 2142 } 2143 2144 CURVNET_RESTORE(); 2145 } 2146 2147 static void 2148 pfsync_send_plus(void *plus, size_t pluslen) 2149 { 2150 struct pfsync_softc *sc = V_pfsyncif; 2151 2152 PFSYNC_LOCK_ASSERT(sc); 2153 2154 if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu) 2155 pfsync_sendout(1); 2156 2157 sc->sc_plus = plus; 2158 sc->sc_len += (sc->sc_pluslen = pluslen); 2159 2160 pfsync_sendout(1); 2161 } 2162 2163 static void 2164 pfsync_timeout(void *arg) 2165 { 2166 struct pfsync_softc *sc = arg; 2167 2168 CURVNET_SET(sc->sc_ifp->if_vnet); 2169 PFSYNC_LOCK(sc); 2170 pfsync_push(sc); 2171 PFSYNC_UNLOCK(sc); 2172 CURVNET_RESTORE(); 2173 } 2174 2175 static void 2176 pfsync_push(struct pfsync_softc *sc) 2177 { 2178 2179 PFSYNC_LOCK_ASSERT(sc); 2180 2181 sc->sc_flags |= PFSYNCF_PUSH; 2182 swi_sched(V_pfsync_swi_cookie, 0); 2183 } 2184 2185 static void 2186 pfsyncintr(void *arg) 2187 { 2188 struct pfsync_softc *sc = arg; 2189 struct mbuf *m, *n; 2190 2191 CURVNET_SET(sc->sc_ifp->if_vnet); 2192 2193 PFSYNC_LOCK(sc); 2194 if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) { 2195 pfsync_sendout(0); 2196 sc->sc_flags &= ~PFSYNCF_PUSH; 2197 } 2198 _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m); 2199 PFSYNC_UNLOCK(sc); 2200 2201 for (; m != NULL; m = n) { 2202 2203 n = m->m_nextpkt; 2204 m->m_nextpkt = NULL; 2205 2206 /* 2207 * We distinguish between a deferral packet and our 2208 * own pfsync packet based on M_SKIP_FIREWALL 2209 * flag. This is XXX. 2210 */ 2211 if (m->m_flags & M_SKIP_FIREWALL) 2212 ip_output(m, NULL, NULL, 0, NULL, NULL); 2213 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 2214 NULL) == 0) 2215 V_pfsyncstats.pfsyncs_opackets++; 2216 else 2217 V_pfsyncstats.pfsyncs_oerrors++; 2218 } 2219 CURVNET_RESTORE(); 2220 } 2221 2222 static int 2223 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship) 2224 { 2225 struct ip_moptions *imo = &sc->sc_imo; 2226 int error; 2227 2228 if (!(ifp->if_flags & IFF_MULTICAST)) 2229 return (EADDRNOTAVAIL); 2230 2231 imo->imo_membership = (struct in_multi **)mship; 2232 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 2233 imo->imo_multicast_vif = -1; 2234 2235 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, 2236 &imo->imo_membership[0])) != 0) { 2237 imo->imo_membership = NULL; 2238 return (error); 2239 } 2240 imo->imo_num_memberships++; 2241 imo->imo_multicast_ifp = ifp; 2242 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2243 imo->imo_multicast_loop = 0; 2244 2245 return (0); 2246 } 2247 2248 static void 2249 pfsync_multicast_cleanup(struct pfsync_softc *sc) 2250 { 2251 struct ip_moptions *imo = &sc->sc_imo; 2252 2253 in_leavegroup(imo->imo_membership[0], NULL); 2254 free(imo->imo_membership, M_PFSYNC); 2255 imo->imo_membership = NULL; 2256 imo->imo_multicast_ifp = NULL; 2257 } 2258 2259 #ifdef INET 2260 extern struct domain inetdomain; 2261 static struct protosw in_pfsync_protosw = { 2262 .pr_type = SOCK_RAW, 2263 .pr_domain = &inetdomain, 2264 .pr_protocol = IPPROTO_PFSYNC, 2265 .pr_flags = PR_ATOMIC|PR_ADDR, 2266 .pr_input = pfsync_input, 2267 .pr_output = (pr_output_t *)rip_output, 2268 .pr_ctloutput = rip_ctloutput, 2269 .pr_usrreqs = &rip_usrreqs 2270 }; 2271 #endif 2272 2273 static void 2274 pfsync_pointers_init() 2275 { 2276 2277 PF_RULES_WLOCK(); 2278 pfsync_state_import_ptr = pfsync_state_import; 2279 pfsync_insert_state_ptr = pfsync_insert_state; 2280 pfsync_update_state_ptr = pfsync_update_state; 2281 pfsync_delete_state_ptr = pfsync_delete_state; 2282 pfsync_clear_states_ptr = pfsync_clear_states; 2283 pfsync_defer_ptr = pfsync_defer; 2284 PF_RULES_WUNLOCK(); 2285 } 2286 2287 static void 2288 pfsync_pointers_uninit() 2289 { 2290 2291 PF_RULES_WLOCK(); 2292 pfsync_state_import_ptr = NULL; 2293 pfsync_insert_state_ptr = NULL; 2294 pfsync_update_state_ptr = NULL; 2295 pfsync_delete_state_ptr = NULL; 2296 pfsync_clear_states_ptr = NULL; 2297 pfsync_defer_ptr = NULL; 2298 PF_RULES_WUNLOCK(); 2299 } 2300 2301 static int 2302 pfsync_init() 2303 { 2304 VNET_ITERATOR_DECL(vnet_iter); 2305 int error = 0; 2306 2307 VNET_LIST_RLOCK(); 2308 VNET_FOREACH(vnet_iter) { 2309 CURVNET_SET(vnet_iter); 2310 V_pfsync_cloner = if_clone_simple(pfsyncname, 2311 pfsync_clone_create, pfsync_clone_destroy, 1); 2312 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif, 2313 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2314 CURVNET_RESTORE(); 2315 if (error) 2316 goto fail_locked; 2317 } 2318 VNET_LIST_RUNLOCK(); 2319 #ifdef INET 2320 error = pf_proto_register(PF_INET, &in_pfsync_protosw); 2321 if (error) 2322 goto fail; 2323 error = ipproto_register(IPPROTO_PFSYNC); 2324 if (error) { 2325 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2326 goto fail; 2327 } 2328 #endif 2329 pfsync_pointers_init(); 2330 2331 return (0); 2332 2333 fail: 2334 VNET_LIST_RLOCK(); 2335 fail_locked: 2336 VNET_FOREACH(vnet_iter) { 2337 CURVNET_SET(vnet_iter); 2338 if (V_pfsync_swi_cookie) { 2339 swi_remove(V_pfsync_swi_cookie); 2340 if_clone_detach(V_pfsync_cloner); 2341 } 2342 CURVNET_RESTORE(); 2343 } 2344 VNET_LIST_RUNLOCK(); 2345 2346 return (error); 2347 } 2348 2349 static void 2350 pfsync_uninit() 2351 { 2352 VNET_ITERATOR_DECL(vnet_iter); 2353 2354 pfsync_pointers_uninit(); 2355 2356 ipproto_unregister(IPPROTO_PFSYNC); 2357 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2358 VNET_LIST_RLOCK(); 2359 VNET_FOREACH(vnet_iter) { 2360 CURVNET_SET(vnet_iter); 2361 if_clone_detach(V_pfsync_cloner); 2362 swi_remove(V_pfsync_swi_cookie); 2363 CURVNET_RESTORE(); 2364 } 2365 VNET_LIST_RUNLOCK(); 2366 } 2367 2368 static int 2369 pfsync_modevent(module_t mod, int type, void *data) 2370 { 2371 int error = 0; 2372 2373 switch (type) { 2374 case MOD_LOAD: 2375 error = pfsync_init(); 2376 break; 2377 case MOD_QUIESCE: 2378 /* 2379 * Module should not be unloaded due to race conditions. 2380 */ 2381 error = EBUSY; 2382 break; 2383 case MOD_UNLOAD: 2384 pfsync_uninit(); 2385 break; 2386 default: 2387 error = EINVAL; 2388 break; 2389 } 2390 2391 return (error); 2392 } 2393 2394 static moduledata_t pfsync_mod = { 2395 pfsyncname, 2396 pfsync_modevent, 2397 0 2398 }; 2399 2400 #define PFSYNC_MODVER 1 2401 2402 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); 2403 MODULE_VERSION(pfsync, PFSYNC_MODVER); 2404 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); 2405