Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netpfil/pf/if_pfsync.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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_kstate *, 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_kstate *, 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_kstate *, void *); 157 static void pfsync_out_iack(struct pf_kstate *, void *); 158 static void pfsync_out_upd_c(struct pf_kstate *, void *); 159 static void pfsync_out_del(struct pf_kstate *, 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_kstate *, int, bool); 170 static void pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *); 171 172 static void pfsync_update_state(struct pf_kstate *); 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_kstate *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_kstate) 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 uint8_t sc_maxupdates; 219 struct ip sc_template; 220 struct mtx sc_mtx; 221 222 /* Queued data */ 223 struct pfsync_bucket *sc_buckets; 224 225 /* Bulk update info */ 226 struct mtx sc_bulk_mtx; 227 uint32_t sc_ureq_sent; 228 int sc_bulk_tries; 229 uint32_t sc_ureq_received; 230 int sc_bulk_hashid; 231 uint64_t sc_bulk_stateid; 232 uint32_t sc_bulk_creatorid; 233 struct callout sc_bulk_tmo; 234 struct callout sc_bulkfail_tmo; 235 }; 236 237 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 238 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 239 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 240 241 #define PFSYNC_BUCKET_LOCK(b) mtx_lock(&(b)->b_mtx) 242 #define PFSYNC_BUCKET_UNLOCK(b) mtx_unlock(&(b)->b_mtx) 243 #define PFSYNC_BUCKET_LOCK_ASSERT(b) mtx_assert(&(b)->b_mtx, MA_OWNED) 244 245 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 246 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 247 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 248 249 static const char pfsyncname[] = "pfsync"; 250 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); 251 VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL; 252 #define V_pfsyncif VNET(pfsyncif) 253 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL; 254 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 255 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie); 256 #define V_pfsync_swi_ie VNET(pfsync_swi_ie) 257 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats); 258 #define V_pfsyncstats VNET(pfsyncstats) 259 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW; 260 #define V_pfsync_carp_adj VNET(pfsync_carp_adj) 261 262 static void pfsync_timeout(void *); 263 static void pfsync_push(struct pfsync_bucket *); 264 static void pfsync_push_all(struct pfsync_softc *); 265 static void pfsyncintr(void *); 266 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 267 struct in_mfilter *imf); 268 static void pfsync_multicast_cleanup(struct pfsync_softc *); 269 static void pfsync_pointers_init(void); 270 static void pfsync_pointers_uninit(void); 271 static int pfsync_init(void); 272 static void pfsync_uninit(void); 273 274 static unsigned long pfsync_buckets; 275 276 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 277 "PFSYNC"); 278 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW, 279 &VNET_NAME(pfsyncstats), pfsyncstats, 280 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 281 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW, 282 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 283 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN, 284 &pfsync_buckets, 0, "Number of pfsync hash buckets"); 285 286 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 287 static void pfsync_clone_destroy(struct ifnet *); 288 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 289 struct pf_state_peer *); 290 static int pfsyncoutput(struct ifnet *, struct mbuf *, 291 const struct sockaddr *, struct route *); 292 static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 293 294 static int pfsync_defer(struct pf_kstate *, struct mbuf *); 295 static void pfsync_undefer(struct pfsync_deferral *, int); 296 static void pfsync_undefer_state(struct pf_kstate *, int); 297 static void pfsync_defer_tmo(void *); 298 299 static void pfsync_request_update(u_int32_t, u_int64_t); 300 static bool pfsync_update_state_req(struct pf_kstate *); 301 302 static void pfsync_drop(struct pfsync_softc *); 303 static void pfsync_sendout(int, int); 304 static void pfsync_send_plus(void *, size_t); 305 306 static void pfsync_bulk_start(void); 307 static void pfsync_bulk_status(u_int8_t); 308 static void pfsync_bulk_update(void *); 309 static void pfsync_bulk_fail(void *); 310 311 static void pfsync_detach_ifnet(struct ifnet *); 312 #ifdef IPSEC 313 static void pfsync_update_net_tdb(struct pfsync_tdb *); 314 #endif 315 static struct pfsync_bucket *pfsync_get_bucket(struct pfsync_softc *, 316 struct pf_kstate *); 317 318 #define PFSYNC_MAX_BULKTRIES 12 319 #define PFSYNC_DEFER_TIMEOUT ((20 * hz) / 1000) 320 321 VNET_DEFINE(struct if_clone *, pfsync_cloner); 322 #define V_pfsync_cloner VNET(pfsync_cloner) 323 324 static int 325 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 326 { 327 struct pfsync_softc *sc; 328 struct ifnet *ifp; 329 struct pfsync_bucket *b; 330 int c, q; 331 332 if (unit != 0) 333 return (EINVAL); 334 335 if (! pfsync_buckets) 336 pfsync_buckets = mp_ncpus * 2; 337 338 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 339 sc->sc_flags |= PFSYNCF_OK; 340 sc->sc_maxupdates = 128; 341 342 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 343 if (ifp == NULL) { 344 free(sc, M_PFSYNC); 345 return (ENOSPC); 346 } 347 if_initname(ifp, pfsyncname, unit); 348 ifp->if_softc = sc; 349 ifp->if_ioctl = pfsyncioctl; 350 ifp->if_output = pfsyncoutput; 351 ifp->if_type = IFT_PFSYNC; 352 ifp->if_hdrlen = sizeof(struct pfsync_header); 353 ifp->if_mtu = ETHERMTU; 354 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); 355 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 356 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 357 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 358 359 if_attach(ifp); 360 361 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 362 363 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets), 364 M_PFSYNC, M_ZERO | M_WAITOK); 365 for (c = 0; c < pfsync_buckets; c++) { 366 b = &sc->sc_buckets[c]; 367 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF); 368 369 b->b_id = c; 370 b->b_sc = sc; 371 b->b_len = PFSYNC_MINPKT; 372 373 for (q = 0; q < PFSYNC_S_COUNT; q++) 374 TAILQ_INIT(&b->b_qs[q]); 375 376 TAILQ_INIT(&b->b_upd_req_list); 377 TAILQ_INIT(&b->b_deferrals); 378 379 callout_init(&b->b_tmo, 1); 380 381 b->b_snd.ifq_maxlen = ifqmaxlen; 382 } 383 384 V_pfsyncif = sc; 385 386 return (0); 387 } 388 389 static void 390 pfsync_clone_destroy(struct ifnet *ifp) 391 { 392 struct pfsync_softc *sc = ifp->if_softc; 393 struct pfsync_bucket *b; 394 int c; 395 396 for (c = 0; c < pfsync_buckets; c++) { 397 b = &sc->sc_buckets[c]; 398 /* 399 * At this stage, everything should have already been 400 * cleared by pfsync_uninit(), and we have only to 401 * drain callouts. 402 */ 403 while (b->b_deferred > 0) { 404 struct pfsync_deferral *pd = 405 TAILQ_FIRST(&b->b_deferrals); 406 407 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 408 b->b_deferred--; 409 if (callout_stop(&pd->pd_tmo) > 0) { 410 pf_release_state(pd->pd_st); 411 m_freem(pd->pd_m); 412 free(pd, M_PFSYNC); 413 } else { 414 pd->pd_refs++; 415 callout_drain(&pd->pd_tmo); 416 free(pd, M_PFSYNC); 417 } 418 } 419 420 callout_drain(&b->b_tmo); 421 } 422 423 callout_drain(&sc->sc_bulkfail_tmo); 424 callout_drain(&sc->sc_bulk_tmo); 425 426 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 427 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 428 bpfdetach(ifp); 429 if_detach(ifp); 430 431 pfsync_drop(sc); 432 433 if_free(ifp); 434 pfsync_multicast_cleanup(sc); 435 mtx_destroy(&sc->sc_mtx); 436 mtx_destroy(&sc->sc_bulk_mtx); 437 438 free(sc->sc_buckets, M_PFSYNC); 439 free(sc, M_PFSYNC); 440 441 V_pfsyncif = NULL; 442 } 443 444 static int 445 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 446 struct pf_state_peer *d) 447 { 448 if (s->scrub.scrub_flag && d->scrub == NULL) { 449 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 450 if (d->scrub == NULL) 451 return (ENOMEM); 452 } 453 454 return (0); 455 } 456 457 static int 458 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 459 { 460 struct pfsync_softc *sc = V_pfsyncif; 461 #ifndef __NO_STRICT_ALIGNMENT 462 struct pfsync_state_key key[2]; 463 #endif 464 struct pfsync_state_key *kw, *ks; 465 struct pf_kstate *st = NULL; 466 struct pf_state_key *skw = NULL, *sks = NULL; 467 struct pf_krule *r = NULL; 468 struct pfi_kkif *kif; 469 int error; 470 471 PF_RULES_RASSERT(); 472 473 if (sp->creatorid == 0) { 474 if (V_pf_status.debug >= PF_DEBUG_MISC) 475 printf("%s: invalid creator id: %08x\n", __func__, 476 ntohl(sp->creatorid)); 477 return (EINVAL); 478 } 479 480 if ((kif = pfi_kkif_find(sp->ifname)) == NULL) { 481 if (V_pf_status.debug >= PF_DEBUG_MISC) 482 printf("%s: unknown interface: %s\n", __func__, 483 sp->ifname); 484 if (flags & PFSYNC_SI_IOCTL) 485 return (EINVAL); 486 return (0); /* skip this state */ 487 } 488 489 /* 490 * If the ruleset checksums match or the state is coming from the ioctl, 491 * it's safe to associate the state with the rule of that number. 492 */ 493 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 494 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 495 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 496 r = pf_main_ruleset.rules[ 497 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 498 else 499 r = &V_pf_default_rule; 500 501 if ((r->max_states && 502 counter_u64_fetch(r->states_cur) >= r->max_states)) 503 goto cleanup; 504 505 /* 506 * XXXGL: consider M_WAITOK in ioctl path after. 507 */ 508 st = pf_alloc_state(M_NOWAIT); 509 if (__predict_false(st == NULL)) 510 goto cleanup; 511 512 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 513 goto cleanup; 514 515 #ifndef __NO_STRICT_ALIGNMENT 516 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2); 517 kw = &key[PF_SK_WIRE]; 518 ks = &key[PF_SK_STACK]; 519 #else 520 kw = &sp->key[PF_SK_WIRE]; 521 ks = &sp->key[PF_SK_STACK]; 522 #endif 523 524 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) || 525 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) || 526 kw->port[0] != ks->port[0] || 527 kw->port[1] != ks->port[1]) { 528 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 529 if (sks == NULL) 530 goto cleanup; 531 } else 532 sks = skw; 533 534 /* allocate memory for scrub info */ 535 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 536 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 537 goto cleanup; 538 539 /* Copy to state key(s). */ 540 skw->addr[0] = kw->addr[0]; 541 skw->addr[1] = kw->addr[1]; 542 skw->port[0] = kw->port[0]; 543 skw->port[1] = kw->port[1]; 544 skw->proto = sp->proto; 545 skw->af = sp->af; 546 if (sks != skw) { 547 sks->addr[0] = ks->addr[0]; 548 sks->addr[1] = ks->addr[1]; 549 sks->port[0] = ks->port[0]; 550 sks->port[1] = ks->port[1]; 551 sks->proto = sp->proto; 552 sks->af = sp->af; 553 } 554 555 /* copy to state */ 556 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 557 st->creation = time_uptime - ntohl(sp->creation); 558 st->expire = time_uptime; 559 if (sp->expire) { 560 uint32_t timeout; 561 562 timeout = r->timeout[sp->timeout]; 563 if (!timeout) 564 timeout = V_pf_default_rule.timeout[sp->timeout]; 565 566 /* sp->expire may have been adaptively scaled by export. */ 567 st->expire -= timeout - ntohl(sp->expire); 568 } 569 570 st->direction = sp->direction; 571 st->log = sp->log; 572 st->timeout = sp->timeout; 573 st->state_flags = sp->state_flags; 574 575 st->id = sp->id; 576 st->creatorid = sp->creatorid; 577 pf_state_peer_ntoh(&sp->src, &st->src); 578 pf_state_peer_ntoh(&sp->dst, &st->dst); 579 580 st->rule.ptr = r; 581 st->nat_rule.ptr = NULL; 582 st->anchor.ptr = NULL; 583 st->rt_kif = NULL; 584 585 st->pfsync_time = time_uptime; 586 st->sync_state = PFSYNC_S_NONE; 587 588 if (!(flags & PFSYNC_SI_IOCTL)) 589 st->state_flags |= PFSTATE_NOSYNC; 590 591 if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0) 592 goto cleanup_state; 593 594 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 595 counter_u64_add(r->states_cur, 1); 596 counter_u64_add(r->states_tot, 1); 597 598 if (!(flags & PFSYNC_SI_IOCTL)) { 599 st->state_flags &= ~PFSTATE_NOSYNC; 600 if (st->state_flags & PFSTATE_ACK) { 601 pfsync_q_ins(st, PFSYNC_S_IACK, true); 602 pfsync_push_all(sc); 603 } 604 } 605 st->state_flags &= ~PFSTATE_ACK; 606 PF_STATE_UNLOCK(st); 607 608 return (0); 609 610 cleanup: 611 error = ENOMEM; 612 if (skw == sks) 613 sks = NULL; 614 if (skw != NULL) 615 uma_zfree(V_pf_state_key_z, skw); 616 if (sks != NULL) 617 uma_zfree(V_pf_state_key_z, sks); 618 619 cleanup_state: /* pf_state_insert() frees the state keys. */ 620 if (st) { 621 st->timeout = PFTM_UNLINKED; /* appease an assert */ 622 pf_free_state(st); 623 } 624 return (error); 625 } 626 627 static int 628 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused) 629 { 630 struct pfsync_softc *sc = V_pfsyncif; 631 struct pfsync_pkt pkt; 632 struct mbuf *m = *mp; 633 struct ip *ip = mtod(m, struct ip *); 634 struct pfsync_header *ph; 635 struct pfsync_subheader subh; 636 637 int offset, len; 638 int rv; 639 uint16_t count; 640 641 PF_RULES_RLOCK_TRACKER; 642 643 *mp = NULL; 644 V_pfsyncstats.pfsyncs_ipackets++; 645 646 /* Verify that we have a sync interface configured. */ 647 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 648 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 649 goto done; 650 651 /* verify that the packet came in on the right interface */ 652 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 653 V_pfsyncstats.pfsyncs_badif++; 654 goto done; 655 } 656 657 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1); 658 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); 659 /* verify that the IP TTL is 255. */ 660 if (ip->ip_ttl != PFSYNC_DFLTTL) { 661 V_pfsyncstats.pfsyncs_badttl++; 662 goto done; 663 } 664 665 offset = ip->ip_hl << 2; 666 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 667 V_pfsyncstats.pfsyncs_hdrops++; 668 goto done; 669 } 670 671 if (offset + sizeof(*ph) > m->m_len) { 672 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 673 V_pfsyncstats.pfsyncs_hdrops++; 674 return (IPPROTO_DONE); 675 } 676 ip = mtod(m, struct ip *); 677 } 678 ph = (struct pfsync_header *)((char *)ip + offset); 679 680 /* verify the version */ 681 if (ph->version != PFSYNC_VERSION) { 682 V_pfsyncstats.pfsyncs_badver++; 683 goto done; 684 } 685 686 len = ntohs(ph->len) + offset; 687 if (m->m_pkthdr.len < len) { 688 V_pfsyncstats.pfsyncs_badlen++; 689 goto done; 690 } 691 692 /* Cheaper to grab this now than having to mess with mbufs later */ 693 pkt.ip = ip; 694 pkt.src = ip->ip_src; 695 pkt.flags = 0; 696 697 /* 698 * Trusting pf_chksum during packet processing, as well as seeking 699 * in interface name tree, require holding PF_RULES_RLOCK(). 700 */ 701 PF_RULES_RLOCK(); 702 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 703 pkt.flags |= PFSYNC_SI_CKSUM; 704 705 offset += sizeof(*ph); 706 while (offset <= len - sizeof(subh)) { 707 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 708 offset += sizeof(subh); 709 710 if (subh.action >= PFSYNC_ACT_MAX) { 711 V_pfsyncstats.pfsyncs_badact++; 712 PF_RULES_RUNLOCK(); 713 goto done; 714 } 715 716 count = ntohs(subh.count); 717 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 718 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); 719 if (rv == -1) { 720 PF_RULES_RUNLOCK(); 721 return (IPPROTO_DONE); 722 } 723 724 offset += rv; 725 } 726 PF_RULES_RUNLOCK(); 727 728 done: 729 m_freem(m); 730 return (IPPROTO_DONE); 731 } 732 733 static int 734 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 735 { 736 struct pfsync_clr *clr; 737 struct mbuf *mp; 738 int len = sizeof(*clr) * count; 739 int i, offp; 740 u_int32_t creatorid; 741 742 mp = m_pulldown(m, offset, len, &offp); 743 if (mp == NULL) { 744 V_pfsyncstats.pfsyncs_badlen++; 745 return (-1); 746 } 747 clr = (struct pfsync_clr *)(mp->m_data + offp); 748 749 for (i = 0; i < count; i++) { 750 creatorid = clr[i].creatorid; 751 752 if (clr[i].ifname[0] != '\0' && 753 pfi_kkif_find(clr[i].ifname) == NULL) 754 continue; 755 756 for (int i = 0; i <= pf_hashmask; i++) { 757 struct pf_idhash *ih = &V_pf_idhash[i]; 758 struct pf_kstate *s; 759 relock: 760 PF_HASHROW_LOCK(ih); 761 LIST_FOREACH(s, &ih->states, entry) { 762 if (s->creatorid == creatorid) { 763 s->state_flags |= PFSTATE_NOSYNC; 764 pf_unlink_state(s, PF_ENTER_LOCKED); 765 goto relock; 766 } 767 } 768 PF_HASHROW_UNLOCK(ih); 769 } 770 } 771 772 return (len); 773 } 774 775 static int 776 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 777 { 778 struct mbuf *mp; 779 struct pfsync_state *sa, *sp; 780 int len = sizeof(*sp) * count; 781 int i, offp; 782 783 mp = m_pulldown(m, offset, len, &offp); 784 if (mp == NULL) { 785 V_pfsyncstats.pfsyncs_badlen++; 786 return (-1); 787 } 788 sa = (struct pfsync_state *)(mp->m_data + offp); 789 790 for (i = 0; i < count; i++) { 791 sp = &sa[i]; 792 793 /* Check for invalid values. */ 794 if (sp->timeout >= PFTM_MAX || 795 sp->src.state > PF_TCPS_PROXY_DST || 796 sp->dst.state > PF_TCPS_PROXY_DST || 797 sp->direction > PF_OUT || 798 (sp->af != AF_INET && sp->af != AF_INET6)) { 799 if (V_pf_status.debug >= PF_DEBUG_MISC) 800 printf("%s: invalid value\n", __func__); 801 V_pfsyncstats.pfsyncs_badval++; 802 continue; 803 } 804 805 if (pfsync_state_import(sp, pkt->flags) == ENOMEM) 806 /* Drop out, but process the rest of the actions. */ 807 break; 808 } 809 810 return (len); 811 } 812 813 static int 814 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 815 { 816 struct pfsync_ins_ack *ia, *iaa; 817 struct pf_kstate *st; 818 819 struct mbuf *mp; 820 int len = count * sizeof(*ia); 821 int offp, i; 822 823 mp = m_pulldown(m, offset, len, &offp); 824 if (mp == NULL) { 825 V_pfsyncstats.pfsyncs_badlen++; 826 return (-1); 827 } 828 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 829 830 for (i = 0; i < count; i++) { 831 ia = &iaa[i]; 832 833 st = pf_find_state_byid(ia->id, ia->creatorid); 834 if (st == NULL) 835 continue; 836 837 if (st->state_flags & PFSTATE_ACK) { 838 pfsync_undefer_state(st, 0); 839 } 840 PF_STATE_UNLOCK(st); 841 } 842 /* 843 * XXX this is not yet implemented, but we know the size of the 844 * message so we can skip it. 845 */ 846 847 return (count * sizeof(struct pfsync_ins_ack)); 848 } 849 850 static int 851 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src, 852 struct pfsync_state_peer *dst) 853 { 854 int sync = 0; 855 856 PF_STATE_LOCK_ASSERT(st); 857 858 /* 859 * The state should never go backwards except 860 * for syn-proxy states. Neither should the 861 * sequence window slide backwards. 862 */ 863 if ((st->src.state > src->state && 864 (st->src.state < PF_TCPS_PROXY_SRC || 865 src->state >= PF_TCPS_PROXY_SRC)) || 866 867 (st->src.state == src->state && 868 SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))) 869 sync++; 870 else 871 pf_state_peer_ntoh(src, &st->src); 872 873 if ((st->dst.state > dst->state) || 874 875 (st->dst.state >= TCPS_SYN_SENT && 876 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) 877 sync++; 878 else 879 pf_state_peer_ntoh(dst, &st->dst); 880 881 return (sync); 882 } 883 884 static int 885 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 886 { 887 struct pfsync_softc *sc = V_pfsyncif; 888 struct pfsync_state *sa, *sp; 889 struct pf_kstate *st; 890 int sync; 891 892 struct mbuf *mp; 893 int len = count * sizeof(*sp); 894 int offp, i; 895 896 mp = m_pulldown(m, offset, len, &offp); 897 if (mp == NULL) { 898 V_pfsyncstats.pfsyncs_badlen++; 899 return (-1); 900 } 901 sa = (struct pfsync_state *)(mp->m_data + offp); 902 903 for (i = 0; i < count; i++) { 904 sp = &sa[i]; 905 906 /* check for invalid values */ 907 if (sp->timeout >= PFTM_MAX || 908 sp->src.state > PF_TCPS_PROXY_DST || 909 sp->dst.state > PF_TCPS_PROXY_DST) { 910 if (V_pf_status.debug >= PF_DEBUG_MISC) { 911 printf("pfsync_input: PFSYNC_ACT_UPD: " 912 "invalid value\n"); 913 } 914 V_pfsyncstats.pfsyncs_badval++; 915 continue; 916 } 917 918 st = pf_find_state_byid(sp->id, sp->creatorid); 919 if (st == NULL) { 920 /* insert the update */ 921 if (pfsync_state_import(sp, pkt->flags)) 922 V_pfsyncstats.pfsyncs_badstate++; 923 continue; 924 } 925 926 if (st->state_flags & PFSTATE_ACK) { 927 pfsync_undefer_state(st, 1); 928 } 929 930 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 931 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst); 932 else { 933 sync = 0; 934 935 /* 936 * Non-TCP protocol state machine always go 937 * forwards 938 */ 939 if (st->src.state > sp->src.state) 940 sync++; 941 else 942 pf_state_peer_ntoh(&sp->src, &st->src); 943 if (st->dst.state > sp->dst.state) 944 sync++; 945 else 946 pf_state_peer_ntoh(&sp->dst, &st->dst); 947 } 948 if (sync < 2) { 949 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 950 pf_state_peer_ntoh(&sp->dst, &st->dst); 951 st->expire = time_uptime; 952 st->timeout = sp->timeout; 953 } 954 st->pfsync_time = time_uptime; 955 956 if (sync) { 957 V_pfsyncstats.pfsyncs_stale++; 958 959 pfsync_update_state(st); 960 PF_STATE_UNLOCK(st); 961 pfsync_push_all(sc); 962 continue; 963 } 964 PF_STATE_UNLOCK(st); 965 } 966 967 return (len); 968 } 969 970 static int 971 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 972 { 973 struct pfsync_softc *sc = V_pfsyncif; 974 struct pfsync_upd_c *ua, *up; 975 struct pf_kstate *st; 976 int len = count * sizeof(*up); 977 int sync; 978 struct mbuf *mp; 979 int offp, i; 980 981 mp = m_pulldown(m, offset, len, &offp); 982 if (mp == NULL) { 983 V_pfsyncstats.pfsyncs_badlen++; 984 return (-1); 985 } 986 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 987 988 for (i = 0; i < count; i++) { 989 up = &ua[i]; 990 991 /* check for invalid values */ 992 if (up->timeout >= PFTM_MAX || 993 up->src.state > PF_TCPS_PROXY_DST || 994 up->dst.state > PF_TCPS_PROXY_DST) { 995 if (V_pf_status.debug >= PF_DEBUG_MISC) { 996 printf("pfsync_input: " 997 "PFSYNC_ACT_UPD_C: " 998 "invalid value\n"); 999 } 1000 V_pfsyncstats.pfsyncs_badval++; 1001 continue; 1002 } 1003 1004 st = pf_find_state_byid(up->id, up->creatorid); 1005 if (st == NULL) { 1006 /* We don't have this state. Ask for it. */ 1007 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 1008 pfsync_request_update(up->creatorid, up->id); 1009 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 1010 continue; 1011 } 1012 1013 if (st->state_flags & PFSTATE_ACK) { 1014 pfsync_undefer_state(st, 1); 1015 } 1016 1017 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 1018 sync = pfsync_upd_tcp(st, &up->src, &up->dst); 1019 else { 1020 sync = 0; 1021 1022 /* 1023 * Non-TCP protocol state machine always go 1024 * forwards 1025 */ 1026 if (st->src.state > up->src.state) 1027 sync++; 1028 else 1029 pf_state_peer_ntoh(&up->src, &st->src); 1030 if (st->dst.state > up->dst.state) 1031 sync++; 1032 else 1033 pf_state_peer_ntoh(&up->dst, &st->dst); 1034 } 1035 if (sync < 2) { 1036 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 1037 pf_state_peer_ntoh(&up->dst, &st->dst); 1038 st->expire = time_uptime; 1039 st->timeout = up->timeout; 1040 } 1041 st->pfsync_time = time_uptime; 1042 1043 if (sync) { 1044 V_pfsyncstats.pfsyncs_stale++; 1045 1046 pfsync_update_state(st); 1047 PF_STATE_UNLOCK(st); 1048 pfsync_push_all(sc); 1049 continue; 1050 } 1051 PF_STATE_UNLOCK(st); 1052 } 1053 1054 return (len); 1055 } 1056 1057 static int 1058 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1059 { 1060 struct pfsync_upd_req *ur, *ura; 1061 struct mbuf *mp; 1062 int len = count * sizeof(*ur); 1063 int i, offp; 1064 1065 struct pf_kstate *st; 1066 1067 mp = m_pulldown(m, offset, len, &offp); 1068 if (mp == NULL) { 1069 V_pfsyncstats.pfsyncs_badlen++; 1070 return (-1); 1071 } 1072 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1073 1074 for (i = 0; i < count; i++) { 1075 ur = &ura[i]; 1076 1077 if (ur->id == 0 && ur->creatorid == 0) 1078 pfsync_bulk_start(); 1079 else { 1080 st = pf_find_state_byid(ur->id, ur->creatorid); 1081 if (st == NULL) { 1082 V_pfsyncstats.pfsyncs_badstate++; 1083 continue; 1084 } 1085 if (st->state_flags & PFSTATE_NOSYNC) { 1086 PF_STATE_UNLOCK(st); 1087 continue; 1088 } 1089 1090 pfsync_update_state_req(st); 1091 PF_STATE_UNLOCK(st); 1092 } 1093 } 1094 1095 return (len); 1096 } 1097 1098 static int 1099 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1100 { 1101 struct mbuf *mp; 1102 struct pfsync_state *sa, *sp; 1103 struct pf_kstate *st; 1104 int len = count * sizeof(*sp); 1105 int offp, i; 1106 1107 mp = m_pulldown(m, offset, len, &offp); 1108 if (mp == NULL) { 1109 V_pfsyncstats.pfsyncs_badlen++; 1110 return (-1); 1111 } 1112 sa = (struct pfsync_state *)(mp->m_data + offp); 1113 1114 for (i = 0; i < count; i++) { 1115 sp = &sa[i]; 1116 1117 st = pf_find_state_byid(sp->id, sp->creatorid); 1118 if (st == NULL) { 1119 V_pfsyncstats.pfsyncs_badstate++; 1120 continue; 1121 } 1122 st->state_flags |= PFSTATE_NOSYNC; 1123 pf_unlink_state(st, PF_ENTER_LOCKED); 1124 } 1125 1126 return (len); 1127 } 1128 1129 static int 1130 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1131 { 1132 struct mbuf *mp; 1133 struct pfsync_del_c *sa, *sp; 1134 struct pf_kstate *st; 1135 int len = count * sizeof(*sp); 1136 int offp, i; 1137 1138 mp = m_pulldown(m, offset, len, &offp); 1139 if (mp == NULL) { 1140 V_pfsyncstats.pfsyncs_badlen++; 1141 return (-1); 1142 } 1143 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1144 1145 for (i = 0; i < count; i++) { 1146 sp = &sa[i]; 1147 1148 st = pf_find_state_byid(sp->id, sp->creatorid); 1149 if (st == NULL) { 1150 V_pfsyncstats.pfsyncs_badstate++; 1151 continue; 1152 } 1153 1154 st->state_flags |= PFSTATE_NOSYNC; 1155 pf_unlink_state(st, PF_ENTER_LOCKED); 1156 } 1157 1158 return (len); 1159 } 1160 1161 static int 1162 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1163 { 1164 struct pfsync_softc *sc = V_pfsyncif; 1165 struct pfsync_bus *bus; 1166 struct mbuf *mp; 1167 int len = count * sizeof(*bus); 1168 int offp; 1169 1170 PFSYNC_BLOCK(sc); 1171 1172 /* If we're not waiting for a bulk update, who cares. */ 1173 if (sc->sc_ureq_sent == 0) { 1174 PFSYNC_BUNLOCK(sc); 1175 return (len); 1176 } 1177 1178 mp = m_pulldown(m, offset, len, &offp); 1179 if (mp == NULL) { 1180 PFSYNC_BUNLOCK(sc); 1181 V_pfsyncstats.pfsyncs_badlen++; 1182 return (-1); 1183 } 1184 bus = (struct pfsync_bus *)(mp->m_data + offp); 1185 1186 switch (bus->status) { 1187 case PFSYNC_BUS_START: 1188 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1189 V_pf_limits[PF_LIMIT_STATES].limit / 1190 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1191 sizeof(struct pfsync_state)), 1192 pfsync_bulk_fail, sc); 1193 if (V_pf_status.debug >= PF_DEBUG_MISC) 1194 printf("pfsync: received bulk update start\n"); 1195 break; 1196 1197 case PFSYNC_BUS_END: 1198 if (time_uptime - ntohl(bus->endtime) >= 1199 sc->sc_ureq_sent) { 1200 /* that's it, we're happy */ 1201 sc->sc_ureq_sent = 0; 1202 sc->sc_bulk_tries = 0; 1203 callout_stop(&sc->sc_bulkfail_tmo); 1204 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1205 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1206 "pfsync bulk done"); 1207 sc->sc_flags |= PFSYNCF_OK; 1208 if (V_pf_status.debug >= PF_DEBUG_MISC) 1209 printf("pfsync: received valid " 1210 "bulk update end\n"); 1211 } else { 1212 if (V_pf_status.debug >= PF_DEBUG_MISC) 1213 printf("pfsync: received invalid " 1214 "bulk update end: bad timestamp\n"); 1215 } 1216 break; 1217 } 1218 PFSYNC_BUNLOCK(sc); 1219 1220 return (len); 1221 } 1222 1223 static int 1224 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1225 { 1226 int len = count * sizeof(struct pfsync_tdb); 1227 1228 #if defined(IPSEC) 1229 struct pfsync_tdb *tp; 1230 struct mbuf *mp; 1231 int offp; 1232 int i; 1233 int s; 1234 1235 mp = m_pulldown(m, offset, len, &offp); 1236 if (mp == NULL) { 1237 V_pfsyncstats.pfsyncs_badlen++; 1238 return (-1); 1239 } 1240 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1241 1242 for (i = 0; i < count; i++) 1243 pfsync_update_net_tdb(&tp[i]); 1244 #endif 1245 1246 return (len); 1247 } 1248 1249 #if defined(IPSEC) 1250 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1251 static void 1252 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1253 { 1254 struct tdb *tdb; 1255 int s; 1256 1257 /* check for invalid values */ 1258 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1259 (pt->dst.sa.sa_family != AF_INET && 1260 pt->dst.sa.sa_family != AF_INET6)) 1261 goto bad; 1262 1263 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1264 if (tdb) { 1265 pt->rpl = ntohl(pt->rpl); 1266 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1267 1268 /* Neither replay nor byte counter should ever decrease. */ 1269 if (pt->rpl < tdb->tdb_rpl || 1270 pt->cur_bytes < tdb->tdb_cur_bytes) { 1271 goto bad; 1272 } 1273 1274 tdb->tdb_rpl = pt->rpl; 1275 tdb->tdb_cur_bytes = pt->cur_bytes; 1276 } 1277 return; 1278 1279 bad: 1280 if (V_pf_status.debug >= PF_DEBUG_MISC) 1281 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1282 "invalid value\n"); 1283 V_pfsyncstats.pfsyncs_badstate++; 1284 return; 1285 } 1286 #endif 1287 1288 static int 1289 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1290 { 1291 /* check if we are at the right place in the packet */ 1292 if (offset != m->m_pkthdr.len) 1293 V_pfsyncstats.pfsyncs_badlen++; 1294 1295 /* we're done. free and let the caller return */ 1296 m_freem(m); 1297 return (-1); 1298 } 1299 1300 static int 1301 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1302 { 1303 V_pfsyncstats.pfsyncs_badact++; 1304 1305 m_freem(m); 1306 return (-1); 1307 } 1308 1309 static int 1310 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 1311 struct route *rt) 1312 { 1313 m_freem(m); 1314 return (0); 1315 } 1316 1317 /* ARGSUSED */ 1318 static int 1319 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1320 { 1321 struct pfsync_softc *sc = ifp->if_softc; 1322 struct ifreq *ifr = (struct ifreq *)data; 1323 struct pfsyncreq pfsyncr; 1324 int error; 1325 int c; 1326 1327 switch (cmd) { 1328 case SIOCSIFFLAGS: 1329 PFSYNC_LOCK(sc); 1330 if (ifp->if_flags & IFF_UP) { 1331 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1332 PFSYNC_UNLOCK(sc); 1333 pfsync_pointers_init(); 1334 } else { 1335 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1336 PFSYNC_UNLOCK(sc); 1337 pfsync_pointers_uninit(); 1338 } 1339 break; 1340 case SIOCSIFMTU: 1341 if (!sc->sc_sync_if || 1342 ifr->ifr_mtu <= PFSYNC_MINPKT || 1343 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1344 return (EINVAL); 1345 if (ifr->ifr_mtu < ifp->if_mtu) { 1346 for (c = 0; c < pfsync_buckets; c++) { 1347 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 1348 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT) 1349 pfsync_sendout(1, c); 1350 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 1351 } 1352 } 1353 ifp->if_mtu = ifr->ifr_mtu; 1354 break; 1355 case SIOCGETPFSYNC: 1356 bzero(&pfsyncr, sizeof(pfsyncr)); 1357 PFSYNC_LOCK(sc); 1358 if (sc->sc_sync_if) { 1359 strlcpy(pfsyncr.pfsyncr_syncdev, 1360 sc->sc_sync_if->if_xname, IFNAMSIZ); 1361 } 1362 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1363 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1364 pfsyncr.pfsyncr_defer = sc->sc_flags; 1365 PFSYNC_UNLOCK(sc); 1366 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr), 1367 sizeof(pfsyncr))); 1368 1369 case SIOCSETPFSYNC: 1370 { 1371 struct in_mfilter *imf = NULL; 1372 struct ifnet *sifp; 1373 struct ip *ip; 1374 1375 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1376 return (error); 1377 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr, 1378 sizeof(pfsyncr)))) 1379 return (error); 1380 1381 if (pfsyncr.pfsyncr_maxupdates > 255) 1382 return (EINVAL); 1383 1384 if (pfsyncr.pfsyncr_syncdev[0] == 0) 1385 sifp = NULL; 1386 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) 1387 return (EINVAL); 1388 1389 if (sifp != NULL && ( 1390 pfsyncr.pfsyncr_syncpeer.s_addr == 0 || 1391 pfsyncr.pfsyncr_syncpeer.s_addr == 1392 htonl(INADDR_PFSYNC_GROUP))) 1393 imf = ip_mfilter_alloc(M_WAITOK, 0, 0); 1394 1395 PFSYNC_LOCK(sc); 1396 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1397 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); 1398 else 1399 sc->sc_sync_peer.s_addr = 1400 pfsyncr.pfsyncr_syncpeer.s_addr; 1401 1402 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1403 if (pfsyncr.pfsyncr_defer & PFSYNCF_DEFER) { 1404 sc->sc_flags |= PFSYNCF_DEFER; 1405 V_pfsync_defer_ptr = pfsync_defer; 1406 } else { 1407 sc->sc_flags &= ~PFSYNCF_DEFER; 1408 V_pfsync_defer_ptr = NULL; 1409 } 1410 1411 if (sifp == NULL) { 1412 if (sc->sc_sync_if) 1413 if_rele(sc->sc_sync_if); 1414 sc->sc_sync_if = NULL; 1415 pfsync_multicast_cleanup(sc); 1416 PFSYNC_UNLOCK(sc); 1417 break; 1418 } 1419 1420 for (c = 0; c < pfsync_buckets; c++) { 1421 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 1422 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT && 1423 (sifp->if_mtu < sc->sc_ifp->if_mtu || 1424 (sc->sc_sync_if != NULL && 1425 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 1426 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 1427 pfsync_sendout(1, c); 1428 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 1429 } 1430 1431 pfsync_multicast_cleanup(sc); 1432 1433 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 1434 error = pfsync_multicast_setup(sc, sifp, imf); 1435 if (error) { 1436 if_rele(sifp); 1437 ip_mfilter_free(imf); 1438 PFSYNC_UNLOCK(sc); 1439 return (error); 1440 } 1441 } 1442 if (sc->sc_sync_if) 1443 if_rele(sc->sc_sync_if); 1444 sc->sc_sync_if = sifp; 1445 1446 ip = &sc->sc_template; 1447 bzero(ip, sizeof(*ip)); 1448 ip->ip_v = IPVERSION; 1449 ip->ip_hl = sizeof(sc->sc_template) >> 2; 1450 ip->ip_tos = IPTOS_LOWDELAY; 1451 /* len and id are set later. */ 1452 ip->ip_off = htons(IP_DF); 1453 ip->ip_ttl = PFSYNC_DFLTTL; 1454 ip->ip_p = IPPROTO_PFSYNC; 1455 ip->ip_src.s_addr = INADDR_ANY; 1456 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; 1457 1458 /* Request a full state table update. */ 1459 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1460 (*carp_demote_adj_p)(V_pfsync_carp_adj, 1461 "pfsync bulk start"); 1462 sc->sc_flags &= ~PFSYNCF_OK; 1463 if (V_pf_status.debug >= PF_DEBUG_MISC) 1464 printf("pfsync: requesting bulk update\n"); 1465 PFSYNC_UNLOCK(sc); 1466 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 1467 pfsync_request_update(0, 0); 1468 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 1469 PFSYNC_BLOCK(sc); 1470 sc->sc_ureq_sent = time_uptime; 1471 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, 1472 sc); 1473 PFSYNC_BUNLOCK(sc); 1474 1475 break; 1476 } 1477 default: 1478 return (ENOTTY); 1479 } 1480 1481 return (0); 1482 } 1483 1484 static void 1485 pfsync_out_state(struct pf_kstate *st, void *buf) 1486 { 1487 struct pfsync_state *sp = buf; 1488 1489 pfsync_state_export(sp, st); 1490 } 1491 1492 static void 1493 pfsync_out_iack(struct pf_kstate *st, void *buf) 1494 { 1495 struct pfsync_ins_ack *iack = buf; 1496 1497 iack->id = st->id; 1498 iack->creatorid = st->creatorid; 1499 } 1500 1501 static void 1502 pfsync_out_upd_c(struct pf_kstate *st, void *buf) 1503 { 1504 struct pfsync_upd_c *up = buf; 1505 1506 bzero(up, sizeof(*up)); 1507 up->id = st->id; 1508 pf_state_peer_hton(&st->src, &up->src); 1509 pf_state_peer_hton(&st->dst, &up->dst); 1510 up->creatorid = st->creatorid; 1511 up->timeout = st->timeout; 1512 } 1513 1514 static void 1515 pfsync_out_del(struct pf_kstate *st, void *buf) 1516 { 1517 struct pfsync_del_c *dp = buf; 1518 1519 dp->id = st->id; 1520 dp->creatorid = st->creatorid; 1521 st->state_flags |= PFSTATE_NOSYNC; 1522 } 1523 1524 static void 1525 pfsync_drop(struct pfsync_softc *sc) 1526 { 1527 struct pf_kstate *st, *next; 1528 struct pfsync_upd_req_item *ur; 1529 struct pfsync_bucket *b; 1530 int c, q; 1531 1532 for (c = 0; c < pfsync_buckets; c++) { 1533 b = &sc->sc_buckets[c]; 1534 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1535 if (TAILQ_EMPTY(&b->b_qs[q])) 1536 continue; 1537 1538 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) { 1539 KASSERT(st->sync_state == q, 1540 ("%s: st->sync_state == q", 1541 __func__)); 1542 st->sync_state = PFSYNC_S_NONE; 1543 pf_release_state(st); 1544 } 1545 TAILQ_INIT(&b->b_qs[q]); 1546 } 1547 1548 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1549 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1550 free(ur, M_PFSYNC); 1551 } 1552 1553 b->b_len = PFSYNC_MINPKT; 1554 b->b_plus = NULL; 1555 } 1556 } 1557 1558 static void 1559 pfsync_sendout(int schedswi, int c) 1560 { 1561 struct pfsync_softc *sc = V_pfsyncif; 1562 struct ifnet *ifp = sc->sc_ifp; 1563 struct mbuf *m; 1564 struct ip *ip; 1565 struct pfsync_header *ph; 1566 struct pfsync_subheader *subh; 1567 struct pf_kstate *st, *st_next; 1568 struct pfsync_upd_req_item *ur; 1569 struct pfsync_bucket *b = &sc->sc_buckets[c]; 1570 int offset; 1571 int q, count = 0; 1572 1573 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1574 KASSERT(b->b_len > PFSYNC_MINPKT, 1575 ("%s: sc_len %zu", __func__, b->b_len)); 1576 PFSYNC_BUCKET_LOCK_ASSERT(b); 1577 1578 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1579 pfsync_drop(sc); 1580 return; 1581 } 1582 1583 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR); 1584 if (m == NULL) { 1585 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); 1586 V_pfsyncstats.pfsyncs_onomem++; 1587 return; 1588 } 1589 m->m_data += max_linkhdr; 1590 m->m_len = m->m_pkthdr.len = b->b_len; 1591 1592 /* build the ip header */ 1593 ip = (struct ip *)m->m_data; 1594 bcopy(&sc->sc_template, ip, sizeof(*ip)); 1595 offset = sizeof(*ip); 1596 1597 ip->ip_len = htons(m->m_pkthdr.len); 1598 ip_fillid(ip); 1599 1600 /* build the pfsync header */ 1601 ph = (struct pfsync_header *)(m->m_data + offset); 1602 bzero(ph, sizeof(*ph)); 1603 offset += sizeof(*ph); 1604 1605 ph->version = PFSYNC_VERSION; 1606 ph->len = htons(b->b_len - sizeof(*ip)); 1607 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1608 1609 /* walk the queues */ 1610 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1611 if (TAILQ_EMPTY(&b->b_qs[q])) 1612 continue; 1613 1614 subh = (struct pfsync_subheader *)(m->m_data + offset); 1615 offset += sizeof(*subh); 1616 1617 count = 0; 1618 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) { 1619 KASSERT(st->sync_state == q, 1620 ("%s: st->sync_state == q", 1621 __func__)); 1622 /* 1623 * XXXGL: some of write methods do unlocked reads 1624 * of state data :( 1625 */ 1626 pfsync_qs[q].write(st, m->m_data + offset); 1627 offset += pfsync_qs[q].len; 1628 st->sync_state = PFSYNC_S_NONE; 1629 pf_release_state(st); 1630 count++; 1631 } 1632 TAILQ_INIT(&b->b_qs[q]); 1633 1634 bzero(subh, sizeof(*subh)); 1635 subh->action = pfsync_qs[q].action; 1636 subh->count = htons(count); 1637 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1638 } 1639 1640 if (!TAILQ_EMPTY(&b->b_upd_req_list)) { 1641 subh = (struct pfsync_subheader *)(m->m_data + offset); 1642 offset += sizeof(*subh); 1643 1644 count = 0; 1645 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1646 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1647 1648 bcopy(&ur->ur_msg, m->m_data + offset, 1649 sizeof(ur->ur_msg)); 1650 offset += sizeof(ur->ur_msg); 1651 free(ur, M_PFSYNC); 1652 count++; 1653 } 1654 1655 bzero(subh, sizeof(*subh)); 1656 subh->action = PFSYNC_ACT_UPD_REQ; 1657 subh->count = htons(count); 1658 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1659 } 1660 1661 /* has someone built a custom region for us to add? */ 1662 if (b->b_plus != NULL) { 1663 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen); 1664 offset += b->b_pluslen; 1665 1666 b->b_plus = NULL; 1667 } 1668 1669 subh = (struct pfsync_subheader *)(m->m_data + offset); 1670 offset += sizeof(*subh); 1671 1672 bzero(subh, sizeof(*subh)); 1673 subh->action = PFSYNC_ACT_EOF; 1674 subh->count = htons(1); 1675 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1676 1677 /* we're done, let's put it on the wire */ 1678 if (ifp->if_bpf) { 1679 m->m_data += sizeof(*ip); 1680 m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip); 1681 BPF_MTAP(ifp, m); 1682 m->m_data -= sizeof(*ip); 1683 m->m_len = m->m_pkthdr.len = b->b_len; 1684 } 1685 1686 if (sc->sc_sync_if == NULL) { 1687 b->b_len = PFSYNC_MINPKT; 1688 m_freem(m); 1689 return; 1690 } 1691 1692 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1); 1693 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); 1694 b->b_len = PFSYNC_MINPKT; 1695 1696 if (!_IF_QFULL(&b->b_snd)) 1697 _IF_ENQUEUE(&b->b_snd, m); 1698 else { 1699 m_freem(m); 1700 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1); 1701 } 1702 if (schedswi) 1703 swi_sched(V_pfsync_swi_cookie, 0); 1704 } 1705 1706 static void 1707 pfsync_insert_state(struct pf_kstate *st) 1708 { 1709 struct pfsync_softc *sc = V_pfsyncif; 1710 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1711 1712 if (st->state_flags & PFSTATE_NOSYNC) 1713 return; 1714 1715 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1716 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1717 st->state_flags |= PFSTATE_NOSYNC; 1718 return; 1719 } 1720 1721 KASSERT(st->sync_state == PFSYNC_S_NONE, 1722 ("%s: st->sync_state %u", __func__, st->sync_state)); 1723 1724 PFSYNC_BUCKET_LOCK(b); 1725 if (b->b_len == PFSYNC_MINPKT) 1726 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 1727 1728 pfsync_q_ins(st, PFSYNC_S_INS, true); 1729 PFSYNC_BUCKET_UNLOCK(b); 1730 1731 st->sync_updates = 0; 1732 } 1733 1734 static int 1735 pfsync_defer(struct pf_kstate *st, struct mbuf *m) 1736 { 1737 struct pfsync_softc *sc = V_pfsyncif; 1738 struct pfsync_deferral *pd; 1739 struct pfsync_bucket *b; 1740 1741 if (m->m_flags & (M_BCAST|M_MCAST)) 1742 return (0); 1743 1744 if (sc == NULL) 1745 return (0); 1746 1747 b = pfsync_get_bucket(sc, st); 1748 1749 PFSYNC_LOCK(sc); 1750 1751 if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) || 1752 !(sc->sc_flags & PFSYNCF_DEFER)) { 1753 PFSYNC_UNLOCK(sc); 1754 return (0); 1755 } 1756 1757 PFSYNC_BUCKET_LOCK(b); 1758 PFSYNC_UNLOCK(sc); 1759 1760 if (b->b_deferred >= 128) 1761 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0); 1762 1763 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1764 if (pd == NULL) { 1765 PFSYNC_BUCKET_UNLOCK(b); 1766 return (0); 1767 } 1768 b->b_deferred++; 1769 1770 m->m_flags |= M_SKIP_FIREWALL; 1771 st->state_flags |= PFSTATE_ACK; 1772 1773 pd->pd_sc = sc; 1774 pd->pd_refs = 0; 1775 pd->pd_st = st; 1776 pf_ref_state(st); 1777 pd->pd_m = m; 1778 1779 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry); 1780 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED); 1781 callout_reset(&pd->pd_tmo, PFSYNC_DEFER_TIMEOUT, pfsync_defer_tmo, pd); 1782 1783 pfsync_push(b); 1784 PFSYNC_BUCKET_UNLOCK(b); 1785 1786 return (1); 1787 } 1788 1789 static void 1790 pfsync_undefer(struct pfsync_deferral *pd, int drop) 1791 { 1792 struct pfsync_softc *sc = pd->pd_sc; 1793 struct mbuf *m = pd->pd_m; 1794 struct pf_kstate *st = pd->pd_st; 1795 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1796 1797 PFSYNC_BUCKET_LOCK_ASSERT(b); 1798 1799 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1800 b->b_deferred--; 1801 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1802 free(pd, M_PFSYNC); 1803 pf_release_state(st); 1804 1805 if (drop) 1806 m_freem(m); 1807 else { 1808 _IF_ENQUEUE(&b->b_snd, m); 1809 pfsync_push(b); 1810 } 1811 } 1812 1813 static void 1814 pfsync_defer_tmo(void *arg) 1815 { 1816 struct epoch_tracker et; 1817 struct pfsync_deferral *pd = arg; 1818 struct pfsync_softc *sc = pd->pd_sc; 1819 struct mbuf *m = pd->pd_m; 1820 struct pf_kstate *st = pd->pd_st; 1821 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1822 1823 PFSYNC_BUCKET_LOCK_ASSERT(b); 1824 1825 if (sc->sc_sync_if == NULL) 1826 return; 1827 1828 NET_EPOCH_ENTER(et); 1829 CURVNET_SET(sc->sc_sync_if->if_vnet); 1830 1831 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1832 b->b_deferred--; 1833 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1834 if (pd->pd_refs == 0) 1835 free(pd, M_PFSYNC); 1836 PFSYNC_BUCKET_UNLOCK(b); 1837 1838 ip_output(m, NULL, NULL, 0, NULL, NULL); 1839 1840 pf_release_state(st); 1841 1842 CURVNET_RESTORE(); 1843 NET_EPOCH_EXIT(et); 1844 } 1845 1846 static void 1847 pfsync_undefer_state(struct pf_kstate *st, int drop) 1848 { 1849 struct pfsync_softc *sc = V_pfsyncif; 1850 struct pfsync_deferral *pd; 1851 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1852 1853 PFSYNC_BUCKET_LOCK(b); 1854 1855 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) { 1856 if (pd->pd_st == st) { 1857 if (callout_stop(&pd->pd_tmo) > 0) 1858 pfsync_undefer(pd, drop); 1859 1860 PFSYNC_BUCKET_UNLOCK(b); 1861 return; 1862 } 1863 } 1864 PFSYNC_BUCKET_UNLOCK(b); 1865 1866 panic("%s: unable to find deferred state", __func__); 1867 } 1868 1869 static struct pfsync_bucket* 1870 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st) 1871 { 1872 int c = PF_IDHASH(st) % pfsync_buckets; 1873 return &sc->sc_buckets[c]; 1874 } 1875 1876 static void 1877 pfsync_update_state(struct pf_kstate *st) 1878 { 1879 struct pfsync_softc *sc = V_pfsyncif; 1880 bool sync = false, ref = true; 1881 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1882 1883 PF_STATE_LOCK_ASSERT(st); 1884 PFSYNC_BUCKET_LOCK(b); 1885 1886 if (st->state_flags & PFSTATE_ACK) 1887 pfsync_undefer_state(st, 0); 1888 if (st->state_flags & PFSTATE_NOSYNC) { 1889 if (st->sync_state != PFSYNC_S_NONE) 1890 pfsync_q_del(st, true, b); 1891 PFSYNC_BUCKET_UNLOCK(b); 1892 return; 1893 } 1894 1895 if (b->b_len == PFSYNC_MINPKT) 1896 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 1897 1898 switch (st->sync_state) { 1899 case PFSYNC_S_UPD_C: 1900 case PFSYNC_S_UPD: 1901 case PFSYNC_S_INS: 1902 /* we're already handling it */ 1903 1904 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 1905 st->sync_updates++; 1906 if (st->sync_updates >= sc->sc_maxupdates) 1907 sync = true; 1908 } 1909 break; 1910 1911 case PFSYNC_S_IACK: 1912 pfsync_q_del(st, false, b); 1913 ref = false; 1914 /* FALLTHROUGH */ 1915 1916 case PFSYNC_S_NONE: 1917 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref); 1918 st->sync_updates = 0; 1919 break; 1920 1921 default: 1922 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1923 } 1924 1925 if (sync || (time_uptime - st->pfsync_time) < 2) 1926 pfsync_push(b); 1927 1928 PFSYNC_BUCKET_UNLOCK(b); 1929 } 1930 1931 static void 1932 pfsync_request_update(u_int32_t creatorid, u_int64_t id) 1933 { 1934 struct pfsync_softc *sc = V_pfsyncif; 1935 struct pfsync_bucket *b = &sc->sc_buckets[0]; 1936 struct pfsync_upd_req_item *item; 1937 size_t nlen = sizeof(struct pfsync_upd_req); 1938 1939 PFSYNC_BUCKET_LOCK_ASSERT(b); 1940 1941 /* 1942 * This code does a bit to prevent multiple update requests for the 1943 * same state being generated. It searches current subheader queue, 1944 * but it doesn't lookup into queue of already packed datagrams. 1945 */ 1946 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry) 1947 if (item->ur_msg.id == id && 1948 item->ur_msg.creatorid == creatorid) 1949 return; 1950 1951 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 1952 if (item == NULL) 1953 return; /* XXX stats */ 1954 1955 item->ur_msg.id = id; 1956 item->ur_msg.creatorid = creatorid; 1957 1958 if (TAILQ_EMPTY(&b->b_upd_req_list)) 1959 nlen += sizeof(struct pfsync_subheader); 1960 1961 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 1962 pfsync_sendout(0, 0); 1963 1964 nlen = sizeof(struct pfsync_subheader) + 1965 sizeof(struct pfsync_upd_req); 1966 } 1967 1968 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry); 1969 b->b_len += nlen; 1970 1971 pfsync_push(b); 1972 } 1973 1974 static bool 1975 pfsync_update_state_req(struct pf_kstate *st) 1976 { 1977 struct pfsync_softc *sc = V_pfsyncif; 1978 bool ref = true, full = false; 1979 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1980 1981 PF_STATE_LOCK_ASSERT(st); 1982 PFSYNC_BUCKET_LOCK(b); 1983 1984 if (st->state_flags & PFSTATE_NOSYNC) { 1985 if (st->sync_state != PFSYNC_S_NONE) 1986 pfsync_q_del(st, true, b); 1987 PFSYNC_BUCKET_UNLOCK(b); 1988 return (full); 1989 } 1990 1991 switch (st->sync_state) { 1992 case PFSYNC_S_UPD_C: 1993 case PFSYNC_S_IACK: 1994 pfsync_q_del(st, false, b); 1995 ref = false; 1996 /* FALLTHROUGH */ 1997 1998 case PFSYNC_S_NONE: 1999 pfsync_q_ins(st, PFSYNC_S_UPD, ref); 2000 pfsync_push(b); 2001 break; 2002 2003 case PFSYNC_S_INS: 2004 case PFSYNC_S_UPD: 2005 case PFSYNC_S_DEL: 2006 /* we're already handling it */ 2007 break; 2008 2009 default: 2010 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2011 } 2012 2013 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state)) 2014 full = true; 2015 2016 PFSYNC_BUCKET_UNLOCK(b); 2017 2018 return (full); 2019 } 2020 2021 static void 2022 pfsync_delete_state(struct pf_kstate *st) 2023 { 2024 struct pfsync_softc *sc = V_pfsyncif; 2025 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2026 bool ref = true; 2027 2028 PFSYNC_BUCKET_LOCK(b); 2029 if (st->state_flags & PFSTATE_ACK) 2030 pfsync_undefer_state(st, 1); 2031 if (st->state_flags & PFSTATE_NOSYNC) { 2032 if (st->sync_state != PFSYNC_S_NONE) 2033 pfsync_q_del(st, true, b); 2034 PFSYNC_BUCKET_UNLOCK(b); 2035 return; 2036 } 2037 2038 if (b->b_len == PFSYNC_MINPKT) 2039 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 2040 2041 switch (st->sync_state) { 2042 case PFSYNC_S_INS: 2043 /* We never got to tell the world so just forget about it. */ 2044 pfsync_q_del(st, true, b); 2045 break; 2046 2047 case PFSYNC_S_UPD_C: 2048 case PFSYNC_S_UPD: 2049 case PFSYNC_S_IACK: 2050 pfsync_q_del(st, false, b); 2051 ref = false; 2052 /* FALLTHROUGH */ 2053 2054 case PFSYNC_S_NONE: 2055 pfsync_q_ins(st, PFSYNC_S_DEL, ref); 2056 break; 2057 2058 default: 2059 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2060 } 2061 2062 PFSYNC_BUCKET_UNLOCK(b); 2063 } 2064 2065 static void 2066 pfsync_clear_states(u_int32_t creatorid, const char *ifname) 2067 { 2068 struct { 2069 struct pfsync_subheader subh; 2070 struct pfsync_clr clr; 2071 } __packed r; 2072 2073 bzero(&r, sizeof(r)); 2074 2075 r.subh.action = PFSYNC_ACT_CLR; 2076 r.subh.count = htons(1); 2077 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 2078 2079 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 2080 r.clr.creatorid = creatorid; 2081 2082 pfsync_send_plus(&r, sizeof(r)); 2083 } 2084 2085 static void 2086 pfsync_q_ins(struct pf_kstate *st, int q, bool ref) 2087 { 2088 struct pfsync_softc *sc = V_pfsyncif; 2089 size_t nlen = pfsync_qs[q].len; 2090 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2091 2092 PFSYNC_BUCKET_LOCK_ASSERT(b); 2093 2094 KASSERT(st->sync_state == PFSYNC_S_NONE, 2095 ("%s: st->sync_state %u", __func__, st->sync_state)); 2096 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 2097 b->b_len)); 2098 2099 if (TAILQ_EMPTY(&b->b_qs[q])) 2100 nlen += sizeof(struct pfsync_subheader); 2101 2102 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 2103 pfsync_sendout(1, b->b_id); 2104 2105 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 2106 } 2107 2108 b->b_len += nlen; 2109 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list); 2110 st->sync_state = q; 2111 if (ref) 2112 pf_ref_state(st); 2113 } 2114 2115 static void 2116 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b) 2117 { 2118 int q = st->sync_state; 2119 2120 PFSYNC_BUCKET_LOCK_ASSERT(b); 2121 KASSERT(st->sync_state != PFSYNC_S_NONE, 2122 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 2123 2124 b->b_len -= pfsync_qs[q].len; 2125 TAILQ_REMOVE(&b->b_qs[q], st, sync_list); 2126 st->sync_state = PFSYNC_S_NONE; 2127 if (unref) 2128 pf_release_state(st); 2129 2130 if (TAILQ_EMPTY(&b->b_qs[q])) 2131 b->b_len -= sizeof(struct pfsync_subheader); 2132 } 2133 2134 static void 2135 pfsync_bulk_start(void) 2136 { 2137 struct pfsync_softc *sc = V_pfsyncif; 2138 2139 if (V_pf_status.debug >= PF_DEBUG_MISC) 2140 printf("pfsync: received bulk update request\n"); 2141 2142 PFSYNC_BLOCK(sc); 2143 2144 sc->sc_ureq_received = time_uptime; 2145 sc->sc_bulk_hashid = 0; 2146 sc->sc_bulk_stateid = 0; 2147 pfsync_bulk_status(PFSYNC_BUS_START); 2148 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2149 PFSYNC_BUNLOCK(sc); 2150 } 2151 2152 static void 2153 pfsync_bulk_update(void *arg) 2154 { 2155 struct pfsync_softc *sc = arg; 2156 struct pf_kstate *s; 2157 int i; 2158 2159 PFSYNC_BLOCK_ASSERT(sc); 2160 CURVNET_SET(sc->sc_ifp->if_vnet); 2161 2162 /* 2163 * Start with last state from previous invocation. 2164 * It may had gone, in this case start from the 2165 * hash slot. 2166 */ 2167 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2168 2169 if (s != NULL) 2170 i = PF_IDHASH(s); 2171 else 2172 i = sc->sc_bulk_hashid; 2173 2174 for (; i <= pf_hashmask; i++) { 2175 struct pf_idhash *ih = &V_pf_idhash[i]; 2176 2177 if (s != NULL) 2178 PF_HASHROW_ASSERT(ih); 2179 else { 2180 PF_HASHROW_LOCK(ih); 2181 s = LIST_FIRST(&ih->states); 2182 } 2183 2184 for (; s; s = LIST_NEXT(s, entry)) { 2185 if (s->sync_state == PFSYNC_S_NONE && 2186 s->timeout < PFTM_MAX && 2187 s->pfsync_time <= sc->sc_ureq_received) { 2188 if (pfsync_update_state_req(s)) { 2189 /* We've filled a packet. */ 2190 sc->sc_bulk_hashid = i; 2191 sc->sc_bulk_stateid = s->id; 2192 sc->sc_bulk_creatorid = s->creatorid; 2193 PF_HASHROW_UNLOCK(ih); 2194 callout_reset(&sc->sc_bulk_tmo, 1, 2195 pfsync_bulk_update, sc); 2196 goto full; 2197 } 2198 } 2199 } 2200 PF_HASHROW_UNLOCK(ih); 2201 } 2202 2203 /* We're done. */ 2204 pfsync_bulk_status(PFSYNC_BUS_END); 2205 full: 2206 CURVNET_RESTORE(); 2207 } 2208 2209 static void 2210 pfsync_bulk_status(u_int8_t status) 2211 { 2212 struct { 2213 struct pfsync_subheader subh; 2214 struct pfsync_bus bus; 2215 } __packed r; 2216 2217 struct pfsync_softc *sc = V_pfsyncif; 2218 2219 bzero(&r, sizeof(r)); 2220 2221 r.subh.action = PFSYNC_ACT_BUS; 2222 r.subh.count = htons(1); 2223 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2224 2225 r.bus.creatorid = V_pf_status.hostid; 2226 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2227 r.bus.status = status; 2228 2229 pfsync_send_plus(&r, sizeof(r)); 2230 } 2231 2232 static void 2233 pfsync_bulk_fail(void *arg) 2234 { 2235 struct pfsync_softc *sc = arg; 2236 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2237 2238 CURVNET_SET(sc->sc_ifp->if_vnet); 2239 2240 PFSYNC_BLOCK_ASSERT(sc); 2241 2242 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2243 /* Try again */ 2244 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2245 pfsync_bulk_fail, V_pfsyncif); 2246 PFSYNC_BUCKET_LOCK(b); 2247 pfsync_request_update(0, 0); 2248 PFSYNC_BUCKET_UNLOCK(b); 2249 } else { 2250 /* Pretend like the transfer was ok. */ 2251 sc->sc_ureq_sent = 0; 2252 sc->sc_bulk_tries = 0; 2253 PFSYNC_LOCK(sc); 2254 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2255 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2256 "pfsync bulk fail"); 2257 sc->sc_flags |= PFSYNCF_OK; 2258 PFSYNC_UNLOCK(sc); 2259 if (V_pf_status.debug >= PF_DEBUG_MISC) 2260 printf("pfsync: failed to receive bulk update\n"); 2261 } 2262 2263 CURVNET_RESTORE(); 2264 } 2265 2266 static void 2267 pfsync_send_plus(void *plus, size_t pluslen) 2268 { 2269 struct pfsync_softc *sc = V_pfsyncif; 2270 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2271 2272 PFSYNC_BUCKET_LOCK(b); 2273 2274 if (b->b_len + pluslen > sc->sc_ifp->if_mtu) 2275 pfsync_sendout(1, b->b_id); 2276 2277 b->b_plus = plus; 2278 b->b_len += (b->b_pluslen = pluslen); 2279 2280 pfsync_sendout(1, b->b_id); 2281 PFSYNC_BUCKET_UNLOCK(b); 2282 } 2283 2284 static void 2285 pfsync_timeout(void *arg) 2286 { 2287 struct pfsync_bucket *b = arg; 2288 2289 CURVNET_SET(b->b_sc->sc_ifp->if_vnet); 2290 PFSYNC_BUCKET_LOCK(b); 2291 pfsync_push(b); 2292 PFSYNC_BUCKET_UNLOCK(b); 2293 CURVNET_RESTORE(); 2294 } 2295 2296 static void 2297 pfsync_push(struct pfsync_bucket *b) 2298 { 2299 2300 PFSYNC_BUCKET_LOCK_ASSERT(b); 2301 2302 b->b_flags |= PFSYNCF_BUCKET_PUSH; 2303 swi_sched(V_pfsync_swi_cookie, 0); 2304 } 2305 2306 static void 2307 pfsync_push_all(struct pfsync_softc *sc) 2308 { 2309 int c; 2310 struct pfsync_bucket *b; 2311 2312 for (c = 0; c < pfsync_buckets; c++) { 2313 b = &sc->sc_buckets[c]; 2314 2315 PFSYNC_BUCKET_LOCK(b); 2316 pfsync_push(b); 2317 PFSYNC_BUCKET_UNLOCK(b); 2318 } 2319 } 2320 2321 static void 2322 pfsyncintr(void *arg) 2323 { 2324 struct epoch_tracker et; 2325 struct pfsync_softc *sc = arg; 2326 struct pfsync_bucket *b; 2327 struct mbuf *m, *n; 2328 int c; 2329 2330 NET_EPOCH_ENTER(et); 2331 CURVNET_SET(sc->sc_ifp->if_vnet); 2332 2333 for (c = 0; c < pfsync_buckets; c++) { 2334 b = &sc->sc_buckets[c]; 2335 2336 PFSYNC_BUCKET_LOCK(b); 2337 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) { 2338 pfsync_sendout(0, b->b_id); 2339 b->b_flags &= ~PFSYNCF_BUCKET_PUSH; 2340 } 2341 _IF_DEQUEUE_ALL(&b->b_snd, m); 2342 PFSYNC_BUCKET_UNLOCK(b); 2343 2344 for (; m != NULL; m = n) { 2345 n = m->m_nextpkt; 2346 m->m_nextpkt = NULL; 2347 2348 /* 2349 * We distinguish between a deferral packet and our 2350 * own pfsync packet based on M_SKIP_FIREWALL 2351 * flag. This is XXX. 2352 */ 2353 if (m->m_flags & M_SKIP_FIREWALL) 2354 ip_output(m, NULL, NULL, 0, NULL, NULL); 2355 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 2356 NULL) == 0) 2357 V_pfsyncstats.pfsyncs_opackets++; 2358 else 2359 V_pfsyncstats.pfsyncs_oerrors++; 2360 } 2361 } 2362 CURVNET_RESTORE(); 2363 NET_EPOCH_EXIT(et); 2364 } 2365 2366 static int 2367 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, 2368 struct in_mfilter *imf) 2369 { 2370 struct ip_moptions *imo = &sc->sc_imo; 2371 int error; 2372 2373 if (!(ifp->if_flags & IFF_MULTICAST)) 2374 return (EADDRNOTAVAIL); 2375 2376 imo->imo_multicast_vif = -1; 2377 2378 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, 2379 &imf->imf_inm)) != 0) 2380 return (error); 2381 2382 ip_mfilter_init(&imo->imo_head); 2383 ip_mfilter_insert(&imo->imo_head, imf); 2384 imo->imo_multicast_ifp = ifp; 2385 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2386 imo->imo_multicast_loop = 0; 2387 2388 return (0); 2389 } 2390 2391 static void 2392 pfsync_multicast_cleanup(struct pfsync_softc *sc) 2393 { 2394 struct ip_moptions *imo = &sc->sc_imo; 2395 struct in_mfilter *imf; 2396 2397 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 2398 ip_mfilter_remove(&imo->imo_head, imf); 2399 in_leavegroup(imf->imf_inm, NULL); 2400 ip_mfilter_free(imf); 2401 } 2402 imo->imo_multicast_ifp = NULL; 2403 } 2404 2405 void 2406 pfsync_detach_ifnet(struct ifnet *ifp) 2407 { 2408 struct pfsync_softc *sc = V_pfsyncif; 2409 2410 if (sc == NULL) 2411 return; 2412 2413 PFSYNC_LOCK(sc); 2414 2415 if (sc->sc_sync_if == ifp) { 2416 /* We don't need mutlicast cleanup here, because the interface 2417 * is going away. We do need to ensure we don't try to do 2418 * cleanup later. 2419 */ 2420 ip_mfilter_init(&sc->sc_imo.imo_head); 2421 sc->sc_imo.imo_multicast_ifp = NULL; 2422 sc->sc_sync_if = NULL; 2423 } 2424 2425 PFSYNC_UNLOCK(sc); 2426 } 2427 2428 #ifdef INET 2429 extern struct domain inetdomain; 2430 static struct protosw in_pfsync_protosw = { 2431 .pr_type = SOCK_RAW, 2432 .pr_domain = &inetdomain, 2433 .pr_protocol = IPPROTO_PFSYNC, 2434 .pr_flags = PR_ATOMIC|PR_ADDR, 2435 .pr_input = pfsync_input, 2436 .pr_output = rip_output, 2437 .pr_ctloutput = rip_ctloutput, 2438 .pr_usrreqs = &rip_usrreqs 2439 }; 2440 #endif 2441 2442 static void 2443 pfsync_pointers_init(void) 2444 { 2445 2446 PF_RULES_WLOCK(); 2447 V_pfsync_state_import_ptr = pfsync_state_import; 2448 V_pfsync_insert_state_ptr = pfsync_insert_state; 2449 V_pfsync_update_state_ptr = pfsync_update_state; 2450 V_pfsync_delete_state_ptr = pfsync_delete_state; 2451 V_pfsync_clear_states_ptr = pfsync_clear_states; 2452 V_pfsync_defer_ptr = pfsync_defer; 2453 PF_RULES_WUNLOCK(); 2454 } 2455 2456 static void 2457 pfsync_pointers_uninit(void) 2458 { 2459 2460 PF_RULES_WLOCK(); 2461 V_pfsync_state_import_ptr = NULL; 2462 V_pfsync_insert_state_ptr = NULL; 2463 V_pfsync_update_state_ptr = NULL; 2464 V_pfsync_delete_state_ptr = NULL; 2465 V_pfsync_clear_states_ptr = NULL; 2466 V_pfsync_defer_ptr = NULL; 2467 PF_RULES_WUNLOCK(); 2468 } 2469 2470 static void 2471 vnet_pfsync_init(const void *unused __unused) 2472 { 2473 int error; 2474 2475 V_pfsync_cloner = if_clone_simple(pfsyncname, 2476 pfsync_clone_create, pfsync_clone_destroy, 1); 2477 error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif, 2478 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2479 if (error) { 2480 if_clone_detach(V_pfsync_cloner); 2481 log(LOG_INFO, "swi_add() failed in %s\n", __func__); 2482 } 2483 2484 pfsync_pointers_init(); 2485 } 2486 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY, 2487 vnet_pfsync_init, NULL); 2488 2489 static void 2490 vnet_pfsync_uninit(const void *unused __unused) 2491 { 2492 int ret; 2493 2494 pfsync_pointers_uninit(); 2495 2496 if_clone_detach(V_pfsync_cloner); 2497 ret = swi_remove(V_pfsync_swi_cookie); 2498 MPASS(ret == 0); 2499 ret = intr_event_destroy(V_pfsync_swi_ie); 2500 MPASS(ret == 0); 2501 } 2502 2503 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH, 2504 vnet_pfsync_uninit, NULL); 2505 2506 static int 2507 pfsync_init(void) 2508 { 2509 #ifdef INET 2510 int error; 2511 2512 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet; 2513 2514 error = pf_proto_register(PF_INET, &in_pfsync_protosw); 2515 if (error) 2516 return (error); 2517 error = ipproto_register(IPPROTO_PFSYNC); 2518 if (error) { 2519 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2520 return (error); 2521 } 2522 #endif 2523 2524 return (0); 2525 } 2526 2527 static void 2528 pfsync_uninit(void) 2529 { 2530 pfsync_detach_ifnet_ptr = NULL; 2531 2532 #ifdef INET 2533 ipproto_unregister(IPPROTO_PFSYNC); 2534 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2535 #endif 2536 } 2537 2538 static int 2539 pfsync_modevent(module_t mod, int type, void *data) 2540 { 2541 int error = 0; 2542 2543 switch (type) { 2544 case MOD_LOAD: 2545 error = pfsync_init(); 2546 break; 2547 case MOD_UNLOAD: 2548 pfsync_uninit(); 2549 break; 2550 default: 2551 error = EINVAL; 2552 break; 2553 } 2554 2555 return (error); 2556 } 2557 2558 static moduledata_t pfsync_mod = { 2559 pfsyncname, 2560 pfsync_modevent, 2561 0 2562 }; 2563 2564 #define PFSYNC_MODVER 1 2565 2566 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */ 2567 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 2568 MODULE_VERSION(pfsync, PFSYNC_MODVER); 2569 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
Cache object: ef01d1022bac4f44b9c7f02e93835bad
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]