if_pfsync.c revision 243940
1/* $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $ */ 2 3/* 4 * Copyright (c) 2002 Michael Shalayeff 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 20 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 24 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 25 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 26 * THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29/* 30 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org> 31 * 32 * Permission to use, copy, modify, and distribute this software for any 33 * purpose with or without fee is hereby granted, provided that the above 34 * copyright notice and this permission notice appear in all copies. 35 * 36 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 37 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 38 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 39 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 40 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 41 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 42 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 43 */ 44 45/* 46 * Revisions picked from OpenBSD after revision 1.110 import: 47 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates 48 * 1.120, 1.175 - use monotonic time_uptime 49 * 1.122 - reduce number of updates for non-TCP sessions 50 * 1.125 - rewrite merge or stale processing 51 * 1.128 - cleanups 52 * 1.146 - bzero() mbuf before sparsely filling it with data 53 * 1.170 - SIOCSIFMTU checks 54 * 1.126, 1.142 - deferred packets processing 55 * 1.173 - correct expire time processing 56 */ 57 58#include <sys/cdefs.h> 59__FBSDID("$FreeBSD: head/sys/netpfil/pf/if_pfsync.c 243940 2012-12-06 08:22:08Z glebius $"); 60 61#include "opt_inet.h" 62#include "opt_inet6.h" 63#include "opt_pf.h" 64 65#include <sys/param.h> 66#include <sys/bus.h> 67#include <sys/endian.h> 68#include <sys/interrupt.h> 69#include <sys/kernel.h> 70#include <sys/lock.h> 71#include <sys/mbuf.h> 72#include <sys/module.h> 73#include <sys/mutex.h> 74#include <sys/priv.h> 75#include <sys/protosw.h> 76#include <sys/socket.h> 77#include <sys/sockio.h> 78#include <sys/sysctl.h> 79 80#include <net/bpf.h> 81#include <net/if.h> 82#include <net/if_clone.h> 83#include <net/if_types.h> 84#include <net/pfvar.h> 85#include <net/if_pfsync.h> 86 87#include <netinet/if_ether.h> 88#include <netinet/in.h> 89#include <netinet/in_var.h> 90#include <netinet/ip.h> 91#include <netinet/ip_carp.h> 92#include <netinet/ip_var.h> 93#include <netinet/tcp.h> 94#include <netinet/tcp_fsm.h> 95#include <netinet/tcp_seq.h> 96 97#define PFSYNC_MINPKT ( \ 98 sizeof(struct ip) + \ 99 sizeof(struct pfsync_header) + \ 100 sizeof(struct pfsync_subheader) + \ 101 sizeof(struct pfsync_eof)) 102 103struct pfsync_pkt { 104 struct ip *ip; 105 struct in_addr src; 106 u_int8_t flags; 107}; 108 109static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *, 110 struct pfsync_state_peer *); 111static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int); 112static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int); 113static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int); 114static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int); 115static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int); 116static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int); 117static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int); 118static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int); 119static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int); 120static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int); 121static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int); 122static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int); 123 124static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = { 125 pfsync_in_clr, /* PFSYNC_ACT_CLR */ 126 pfsync_in_ins, /* PFSYNC_ACT_INS */ 127 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ 128 pfsync_in_upd, /* PFSYNC_ACT_UPD */ 129 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ 130 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ 131 pfsync_in_del, /* PFSYNC_ACT_DEL */ 132 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ 133 pfsync_in_error, /* PFSYNC_ACT_INS_F */ 134 pfsync_in_error, /* PFSYNC_ACT_DEL_F */ 135 pfsync_in_bus, /* PFSYNC_ACT_BUS */ 136 pfsync_in_tdb, /* PFSYNC_ACT_TDB */ 137 pfsync_in_eof /* PFSYNC_ACT_EOF */ 138}; 139 140struct pfsync_q { 141 void (*write)(struct pf_state *, void *); 142 size_t len; 143 u_int8_t action; 144}; 145 146/* we have one of these for every PFSYNC_S_ */ 147static void pfsync_out_state(struct pf_state *, void *); 148static void pfsync_out_iack(struct pf_state *, void *); 149static void pfsync_out_upd_c(struct pf_state *, void *); 150static void pfsync_out_del(struct pf_state *, void *); 151 152static struct pfsync_q pfsync_qs[] = { 153 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS }, 154 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, 155 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD }, 156 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, 157 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } 158}; 159 160static void pfsync_q_ins(struct pf_state *, int); 161static void pfsync_q_del(struct pf_state *); 162 163static void pfsync_update_state(struct pf_state *); 164 165struct pfsync_upd_req_item { 166 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; 167 struct pfsync_upd_req ur_msg; 168}; 169 170struct pfsync_deferral { 171 struct pfsync_softc *pd_sc; 172 TAILQ_ENTRY(pfsync_deferral) pd_entry; 173 u_int pd_refs; 174 struct callout pd_tmo; 175 176 struct pf_state *pd_st; 177 struct mbuf *pd_m; 178}; 179 180struct pfsync_softc { 181 /* Configuration */ 182 struct ifnet *sc_ifp; 183 struct ifnet *sc_sync_if; 184 struct ip_moptions sc_imo; 185 struct in_addr sc_sync_peer; 186 uint32_t sc_flags; 187#define PFSYNCF_OK 0x00000001 188#define PFSYNCF_DEFER 0x00000002 189#define PFSYNCF_PUSH 0x00000004 190 uint8_t sc_maxupdates; 191 struct ip sc_template; 192 struct callout sc_tmo; 193 struct mtx sc_mtx; 194 195 /* Queued data */ 196 size_t sc_len; 197 TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT]; 198 TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list; 199 TAILQ_HEAD(, pfsync_deferral) sc_deferrals; 200 u_int sc_deferred; 201 void *sc_plus; 202 size_t sc_pluslen; 203 204 /* Bulk update info */ 205 struct mtx sc_bulk_mtx; 206 uint32_t sc_ureq_sent; 207 int sc_bulk_tries; 208 uint32_t sc_ureq_received; 209 int sc_bulk_hashid; 210 uint64_t sc_bulk_stateid; 211 uint32_t sc_bulk_creatorid; 212 struct callout sc_bulk_tmo; 213 struct callout sc_bulkfail_tmo; 214}; 215 216#define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 217#define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 218#define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 219 220#define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 221#define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 222#define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 223 224static const char pfsyncname[] = "pfsync"; 225static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); 226static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL; 227#define V_pfsyncif VNET(pfsyncif) 228static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL; 229#define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 230static VNET_DEFINE(struct pfsyncstats, pfsyncstats); 231#define V_pfsyncstats VNET(pfsyncstats) 232static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW; 233#define V_pfsync_carp_adj VNET(pfsync_carp_adj) 234 235static void pfsync_timeout(void *); 236static void pfsync_push(struct pfsync_softc *); 237static void pfsyncintr(void *); 238static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 239 void *); 240static void pfsync_multicast_cleanup(struct pfsync_softc *); 241static void pfsync_pointers_init(void); 242static void pfsync_pointers_uninit(void); 243static int pfsync_init(void); 244static void pfsync_uninit(void); 245 246SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC"); 247SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW, 248 &VNET_NAME(pfsyncstats), pfsyncstats, 249 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 250SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW, 251 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 252 253static int pfsync_clone_create(struct if_clone *, int, caddr_t); 254static void pfsync_clone_destroy(struct ifnet *); 255static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 256 struct pf_state_peer *); 257static int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 258 struct route *); 259static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 260 261static int pfsync_defer(struct pf_state *, struct mbuf *); 262static void pfsync_undefer(struct pfsync_deferral *, int); 263static void pfsync_undefer_state(struct pf_state *, int); 264static void pfsync_defer_tmo(void *); 265 266static void pfsync_request_update(u_int32_t, u_int64_t); 267static void pfsync_update_state_req(struct pf_state *); 268 269static void pfsync_drop(struct pfsync_softc *); 270static void pfsync_sendout(int); 271static void pfsync_send_plus(void *, size_t); 272 273static void pfsync_bulk_start(void); 274static void pfsync_bulk_status(u_int8_t); 275static void pfsync_bulk_update(void *); 276static void pfsync_bulk_fail(void *); 277 278#ifdef IPSEC 279static void pfsync_update_net_tdb(struct pfsync_tdb *); 280#endif 281 282#define PFSYNC_MAX_BULKTRIES 12 283 284VNET_DEFINE(struct if_clone *, pfsync_cloner); 285#define V_pfsync_cloner VNET(pfsync_cloner) 286 287static int 288pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 289{ 290 struct pfsync_softc *sc; 291 struct ifnet *ifp; 292 int q; 293 294 if (unit != 0) 295 return (EINVAL); 296 297 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 298 sc->sc_flags |= PFSYNCF_OK; 299 300 for (q = 0; q < PFSYNC_S_COUNT; q++) 301 TAILQ_INIT(&sc->sc_qs[q]); 302 303 TAILQ_INIT(&sc->sc_upd_req_list); 304 TAILQ_INIT(&sc->sc_deferrals); 305 306 sc->sc_len = PFSYNC_MINPKT; 307 sc->sc_maxupdates = 128; 308 309 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 310 if (ifp == NULL) { 311 free(sc, M_PFSYNC); 312 return (ENOSPC); 313 } 314 if_initname(ifp, pfsyncname, unit); 315 ifp->if_softc = sc; 316 ifp->if_ioctl = pfsyncioctl; 317 ifp->if_output = pfsyncoutput; 318 ifp->if_type = IFT_PFSYNC; 319 ifp->if_snd.ifq_maxlen = ifqmaxlen; 320 ifp->if_hdrlen = sizeof(struct pfsync_header); 321 ifp->if_mtu = ETHERMTU; 322 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); 323 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 324 callout_init(&sc->sc_tmo, CALLOUT_MPSAFE); 325 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 326 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 327 328 if_attach(ifp); 329 330 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 331 332 V_pfsyncif = sc; 333 334 return (0); 335} 336 337static void 338pfsync_clone_destroy(struct ifnet *ifp) 339{ 340 struct pfsync_softc *sc = ifp->if_softc; 341 342 /* 343 * At this stage, everything should have already been 344 * cleared by pfsync_uninit(), and we have only to 345 * drain callouts. 346 */ 347 while (sc->sc_deferred > 0) { 348 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals); 349 350 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 351 sc->sc_deferred--; 352 if (callout_stop(&pd->pd_tmo)) { 353 pf_release_state(pd->pd_st); 354 m_freem(pd->pd_m); 355 free(pd, M_PFSYNC); 356 } else { 357 pd->pd_refs++; 358 callout_drain(&pd->pd_tmo); 359 free(pd, M_PFSYNC); 360 } 361 } 362 363 callout_drain(&sc->sc_tmo); 364 callout_drain(&sc->sc_bulkfail_tmo); 365 callout_drain(&sc->sc_bulk_tmo); 366 367 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 368 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 369 bpfdetach(ifp); 370 if_detach(ifp); 371 372 pfsync_drop(sc); 373 374 if_free(ifp); 375 if (sc->sc_imo.imo_membership) 376 pfsync_multicast_cleanup(sc); 377 mtx_destroy(&sc->sc_mtx); 378 mtx_destroy(&sc->sc_bulk_mtx); 379 free(sc, M_PFSYNC); 380 381 V_pfsyncif = NULL; 382} 383 384static int 385pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 386 struct pf_state_peer *d) 387{ 388 if (s->scrub.scrub_flag && d->scrub == NULL) { 389 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 390 if (d->scrub == NULL) 391 return (ENOMEM); 392 } 393 394 return (0); 395} 396 397 398static int 399pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 400{ 401 struct pfsync_softc *sc = V_pfsyncif; 402 struct pf_state *st = NULL; 403 struct pf_state_key *skw = NULL, *sks = NULL; 404 struct pf_rule *r = NULL; 405 struct pfi_kif *kif; 406 int error; 407 408 PF_RULES_RASSERT(); 409 410 if (sp->creatorid == 0 && V_pf_status.debug >= PF_DEBUG_MISC) { 411 printf("%s: invalid creator id: %08x\n", __func__, 412 ntohl(sp->creatorid)); 413 return (EINVAL); 414 } 415 416 if ((kif = pfi_kif_find(sp->ifname)) == NULL) { 417 if (V_pf_status.debug >= PF_DEBUG_MISC) 418 printf("%s: unknown interface: %s\n", __func__, 419 sp->ifname); 420 if (flags & PFSYNC_SI_IOCTL) 421 return (EINVAL); 422 return (0); /* skip this state */ 423 } 424 425 /* 426 * If the ruleset checksums match or the state is coming from the ioctl, 427 * it's safe to associate the state with the rule of that number. 428 */ 429 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 430 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 431 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 432 r = pf_main_ruleset.rules[ 433 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 434 else 435 r = &V_pf_default_rule; 436 437 if ((r->max_states && r->states_cur >= r->max_states)) 438 goto cleanup; 439 440 /* 441 * XXXGL: consider M_WAITOK in ioctl path after. 442 */ 443 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL) 444 goto cleanup; 445 446 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 447 goto cleanup; 448 449 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 450 &sp->key[PF_SK_STACK].addr[0], sp->af) || 451 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 452 &sp->key[PF_SK_STACK].addr[1], sp->af) || 453 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 454 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) { 455 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 456 if (sks == NULL) 457 goto cleanup; 458 } else 459 sks = skw; 460 461 /* allocate memory for scrub info */ 462 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 463 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 464 goto cleanup; 465 466 /* copy to state key(s) */ 467 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 468 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 469 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 470 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 471 skw->proto = sp->proto; 472 skw->af = sp->af; 473 if (sks != skw) { 474 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 475 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 476 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 477 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 478 sks->proto = sp->proto; 479 sks->af = sp->af; 480 } 481 482 /* copy to state */ 483 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 484 st->creation = time_uptime - ntohl(sp->creation); 485 st->expire = time_uptime; 486 if (sp->expire) { 487 uint32_t timeout; 488 489 timeout = r->timeout[sp->timeout]; 490 if (!timeout) 491 timeout = V_pf_default_rule.timeout[sp->timeout]; 492 493 /* sp->expire may have been adaptively scaled by export. */ 494 st->expire -= timeout - ntohl(sp->expire); 495 } 496 497 st->direction = sp->direction; 498 st->log = sp->log; 499 st->timeout = sp->timeout; 500 st->state_flags = sp->state_flags; 501 502 st->id = sp->id; 503 st->creatorid = sp->creatorid; 504 pf_state_peer_ntoh(&sp->src, &st->src); 505 pf_state_peer_ntoh(&sp->dst, &st->dst); 506 507 st->rule.ptr = r; 508 st->nat_rule.ptr = NULL; 509 st->anchor.ptr = NULL; 510 st->rt_kif = NULL; 511 512 st->pfsync_time = time_uptime; 513 st->sync_state = PFSYNC_S_NONE; 514 515 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 516 r->states_cur++; 517 r->states_tot++; 518 519 if (!(flags & PFSYNC_SI_IOCTL)) 520 st->state_flags |= PFSTATE_NOSYNC; 521 522 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) { 523 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 524 r->states_cur--; 525 goto cleanup_state; 526 } 527 528 if (!(flags & PFSYNC_SI_IOCTL)) { 529 st->state_flags &= ~PFSTATE_NOSYNC; 530 if (st->state_flags & PFSTATE_ACK) { 531 pfsync_q_ins(st, PFSYNC_S_IACK); 532 pfsync_push(sc); 533 } 534 } 535 st->state_flags &= ~PFSTATE_ACK; 536 PF_STATE_UNLOCK(st); 537 538 return (0); 539 540cleanup: 541 error = ENOMEM; 542 if (skw == sks) 543 sks = NULL; 544 if (skw != NULL) 545 uma_zfree(V_pf_state_key_z, skw); 546 if (sks != NULL) 547 uma_zfree(V_pf_state_key_z, sks); 548 549cleanup_state: /* pf_state_insert() frees the state keys. */ 550 if (st) { 551 if (st->dst.scrub) 552 uma_zfree(V_pf_state_scrub_z, st->dst.scrub); 553 if (st->src.scrub) 554 uma_zfree(V_pf_state_scrub_z, st->src.scrub); 555 uma_zfree(V_pf_state_z, st); 556 } 557 return (error); 558} 559 560static void 561pfsync_input(struct mbuf *m, __unused int off) 562{ 563 struct pfsync_softc *sc = V_pfsyncif; 564 struct pfsync_pkt pkt; 565 struct ip *ip = mtod(m, struct ip *); 566 struct pfsync_header *ph; 567 struct pfsync_subheader subh; 568 569 int offset; 570 int rv; 571 uint16_t count; 572 573 V_pfsyncstats.pfsyncs_ipackets++; 574 575 /* Verify that we have a sync interface configured. */ 576 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 577 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 578 goto done; 579 580 /* verify that the packet came in on the right interface */ 581 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 582 V_pfsyncstats.pfsyncs_badif++; 583 goto done; 584 } 585 586 sc->sc_ifp->if_ipackets++; 587 sc->sc_ifp->if_ibytes += m->m_pkthdr.len; 588 /* verify that the IP TTL is 255. */ 589 if (ip->ip_ttl != PFSYNC_DFLTTL) { 590 V_pfsyncstats.pfsyncs_badttl++; 591 goto done; 592 } 593 594 offset = ip->ip_hl << 2; 595 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 596 V_pfsyncstats.pfsyncs_hdrops++; 597 goto done; 598 } 599 600 if (offset + sizeof(*ph) > m->m_len) { 601 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 602 V_pfsyncstats.pfsyncs_hdrops++; 603 return; 604 } 605 ip = mtod(m, struct ip *); 606 } 607 ph = (struct pfsync_header *)((char *)ip + offset); 608 609 /* verify the version */ 610 if (ph->version != PFSYNC_VERSION) { 611 V_pfsyncstats.pfsyncs_badver++; 612 goto done; 613 } 614 615 /* Cheaper to grab this now than having to mess with mbufs later */ 616 pkt.ip = ip; 617 pkt.src = ip->ip_src; 618 pkt.flags = 0; 619 620 /* 621 * Trusting pf_chksum during packet processing, as well as seeking 622 * in interface name tree, require holding PF_RULES_RLOCK(). 623 */ 624 PF_RULES_RLOCK(); 625 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 626 pkt.flags |= PFSYNC_SI_CKSUM; 627 628 offset += sizeof(*ph); 629 for (;;) { 630 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 631 offset += sizeof(subh); 632 633 if (subh.action >= PFSYNC_ACT_MAX) { 634 V_pfsyncstats.pfsyncs_badact++; 635 PF_RULES_RUNLOCK(); 636 goto done; 637 } 638 639 count = ntohs(subh.count); 640 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 641 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); 642 if (rv == -1) { 643 PF_RULES_RUNLOCK(); 644 return; 645 } 646 647 offset += rv; 648 } 649 PF_RULES_RUNLOCK(); 650 651done: 652 m_freem(m); 653} 654 655static int 656pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 657{ 658 struct pfsync_clr *clr; 659 struct mbuf *mp; 660 int len = sizeof(*clr) * count; 661 int i, offp; 662 u_int32_t creatorid; 663 664 mp = m_pulldown(m, offset, len, &offp); 665 if (mp == NULL) { 666 V_pfsyncstats.pfsyncs_badlen++; 667 return (-1); 668 } 669 clr = (struct pfsync_clr *)(mp->m_data + offp); 670 671 for (i = 0; i < count; i++) { 672 creatorid = clr[i].creatorid; 673 674 if (clr[i].ifname[0] != '\0' && 675 pfi_kif_find(clr[i].ifname) == NULL) 676 continue; 677 678 for (int i = 0; i <= V_pf_hashmask; i++) { 679 struct pf_idhash *ih = &V_pf_idhash[i]; 680 struct pf_state *s; 681relock: 682 PF_HASHROW_LOCK(ih); 683 LIST_FOREACH(s, &ih->states, entry) { 684 if (s->creatorid == creatorid) { 685 s->state_flags |= PFSTATE_NOSYNC; 686 pf_unlink_state(s, PF_ENTER_LOCKED); 687 goto relock; 688 } 689 } 690 PF_HASHROW_UNLOCK(ih); 691 } 692 } 693 694 return (len); 695} 696 697static int 698pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 699{ 700 struct mbuf *mp; 701 struct pfsync_state *sa, *sp; 702 int len = sizeof(*sp) * count; 703 int i, offp; 704 705 mp = m_pulldown(m, offset, len, &offp); 706 if (mp == NULL) { 707 V_pfsyncstats.pfsyncs_badlen++; 708 return (-1); 709 } 710 sa = (struct pfsync_state *)(mp->m_data + offp); 711 712 for (i = 0; i < count; i++) { 713 sp = &sa[i]; 714 715 /* Check for invalid values. */ 716 if (sp->timeout >= PFTM_MAX || 717 sp->src.state > PF_TCPS_PROXY_DST || 718 sp->dst.state > PF_TCPS_PROXY_DST || 719 sp->direction > PF_OUT || 720 (sp->af != AF_INET && sp->af != AF_INET6)) { 721 if (V_pf_status.debug >= PF_DEBUG_MISC) 722 printf("%s: invalid value\n", __func__); 723 V_pfsyncstats.pfsyncs_badval++; 724 continue; 725 } 726 727 if (pfsync_state_import(sp, pkt->flags) == ENOMEM) 728 /* Drop out, but process the rest of the actions. */ 729 break; 730 } 731 732 return (len); 733} 734 735static int 736pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 737{ 738 struct pfsync_ins_ack *ia, *iaa; 739 struct pf_state *st; 740 741 struct mbuf *mp; 742 int len = count * sizeof(*ia); 743 int offp, i; 744 745 mp = m_pulldown(m, offset, len, &offp); 746 if (mp == NULL) { 747 V_pfsyncstats.pfsyncs_badlen++; 748 return (-1); 749 } 750 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 751 752 for (i = 0; i < count; i++) { 753 ia = &iaa[i]; 754 755 st = pf_find_state_byid(ia->id, ia->creatorid); 756 if (st == NULL) 757 continue; 758 759 if (st->state_flags & PFSTATE_ACK) { 760 PFSYNC_LOCK(V_pfsyncif); 761 pfsync_undefer_state(st, 0); 762 PFSYNC_UNLOCK(V_pfsyncif); 763 } 764 PF_STATE_UNLOCK(st); 765 } 766 /* 767 * XXX this is not yet implemented, but we know the size of the 768 * message so we can skip it. 769 */ 770 771 return (count * sizeof(struct pfsync_ins_ack)); 772} 773 774static int 775pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src, 776 struct pfsync_state_peer *dst) 777{ 778 int sync = 0; 779 780 PF_STATE_LOCK_ASSERT(st); 781 782 /* 783 * The state should never go backwards except 784 * for syn-proxy states. Neither should the 785 * sequence window slide backwards. 786 */ 787 if ((st->src.state > src->state && 788 (st->src.state < PF_TCPS_PROXY_SRC || 789 src->state >= PF_TCPS_PROXY_SRC)) || 790 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))) 791 sync++; 792 else 793 pf_state_peer_ntoh(src, &st->src); 794 795 if (st->dst.state > dst->state || 796 (st->dst.state >= TCPS_SYN_SENT && 797 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) 798 sync++; 799 else 800 pf_state_peer_ntoh(dst, &st->dst); 801 802 return (sync); 803} 804 805static int 806pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 807{ 808 struct pfsync_softc *sc = V_pfsyncif; 809 struct pfsync_state *sa, *sp; 810 struct pf_state *st; 811 int sync; 812 813 struct mbuf *mp; 814 int len = count * sizeof(*sp); 815 int offp, i; 816 817 mp = m_pulldown(m, offset, len, &offp); 818 if (mp == NULL) { 819 V_pfsyncstats.pfsyncs_badlen++; 820 return (-1); 821 } 822 sa = (struct pfsync_state *)(mp->m_data + offp); 823 824 for (i = 0; i < count; i++) { 825 sp = &sa[i]; 826 827 /* check for invalid values */ 828 if (sp->timeout >= PFTM_MAX || 829 sp->src.state > PF_TCPS_PROXY_DST || 830 sp->dst.state > PF_TCPS_PROXY_DST) { 831 if (V_pf_status.debug >= PF_DEBUG_MISC) { 832 printf("pfsync_input: PFSYNC_ACT_UPD: " 833 "invalid value\n"); 834 } 835 V_pfsyncstats.pfsyncs_badval++; 836 continue; 837 } 838 839 st = pf_find_state_byid(sp->id, sp->creatorid); 840 if (st == NULL) { 841 /* insert the update */ 842 if (pfsync_state_import(sp, 0)) 843 V_pfsyncstats.pfsyncs_badstate++; 844 continue; 845 } 846 847 if (st->state_flags & PFSTATE_ACK) { 848 PFSYNC_LOCK(sc); 849 pfsync_undefer_state(st, 1); 850 PFSYNC_UNLOCK(sc); 851 } 852 853 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 854 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst); 855 else { 856 sync = 0; 857 858 /* 859 * Non-TCP protocol state machine always go 860 * forwards 861 */ 862 if (st->src.state > sp->src.state) 863 sync++; 864 else 865 pf_state_peer_ntoh(&sp->src, &st->src); 866 if (st->dst.state > sp->dst.state) 867 sync++; 868 else 869 pf_state_peer_ntoh(&sp->dst, &st->dst); 870 } 871 if (sync < 2) { 872 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 873 pf_state_peer_ntoh(&sp->dst, &st->dst); 874 st->expire = time_uptime; 875 st->timeout = sp->timeout; 876 } 877 st->pfsync_time = time_uptime; 878 879 if (sync) { 880 V_pfsyncstats.pfsyncs_stale++; 881 882 pfsync_update_state(st); 883 PF_STATE_UNLOCK(st); 884 PFSYNC_LOCK(sc); 885 pfsync_push(sc); 886 PFSYNC_UNLOCK(sc); 887 continue; 888 } 889 PF_STATE_UNLOCK(st); 890 } 891 892 return (len); 893} 894 895static int 896pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 897{ 898 struct pfsync_softc *sc = V_pfsyncif; 899 struct pfsync_upd_c *ua, *up; 900 struct pf_state *st; 901 int len = count * sizeof(*up); 902 int sync; 903 struct mbuf *mp; 904 int offp, i; 905 906 mp = m_pulldown(m, offset, len, &offp); 907 if (mp == NULL) { 908 V_pfsyncstats.pfsyncs_badlen++; 909 return (-1); 910 } 911 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 912 913 for (i = 0; i < count; i++) { 914 up = &ua[i]; 915 916 /* check for invalid values */ 917 if (up->timeout >= PFTM_MAX || 918 up->src.state > PF_TCPS_PROXY_DST || 919 up->dst.state > PF_TCPS_PROXY_DST) { 920 if (V_pf_status.debug >= PF_DEBUG_MISC) { 921 printf("pfsync_input: " 922 "PFSYNC_ACT_UPD_C: " 923 "invalid value\n"); 924 } 925 V_pfsyncstats.pfsyncs_badval++; 926 continue; 927 } 928 929 st = pf_find_state_byid(up->id, up->creatorid); 930 if (st == NULL) { 931 /* We don't have this state. Ask for it. */ 932 PFSYNC_LOCK(sc); 933 pfsync_request_update(up->creatorid, up->id); 934 PFSYNC_UNLOCK(sc); 935 continue; 936 } 937 938 if (st->state_flags & PFSTATE_ACK) { 939 PFSYNC_LOCK(sc); 940 pfsync_undefer_state(st, 1); 941 PFSYNC_UNLOCK(sc); 942 } 943 944 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 945 sync = pfsync_upd_tcp(st, &up->src, &up->dst); 946 else { 947 sync = 0; 948 949 /* 950 * Non-TCP protocol state machine always go 951 * forwards 952 */ 953 if (st->src.state > up->src.state) 954 sync++; 955 else 956 pf_state_peer_ntoh(&up->src, &st->src); 957 if (st->dst.state > up->dst.state) 958 sync++; 959 else 960 pf_state_peer_ntoh(&up->dst, &st->dst); 961 } 962 if (sync < 2) { 963 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 964 pf_state_peer_ntoh(&up->dst, &st->dst); 965 st->expire = time_uptime; 966 st->timeout = up->timeout; 967 } 968 st->pfsync_time = time_uptime; 969 970 if (sync) { 971 V_pfsyncstats.pfsyncs_stale++; 972 973 pfsync_update_state(st); 974 PF_STATE_UNLOCK(st); 975 PFSYNC_LOCK(sc); 976 pfsync_push(sc); 977 PFSYNC_UNLOCK(sc); 978 continue; 979 } 980 PF_STATE_UNLOCK(st); 981 } 982 983 return (len); 984} 985 986static int 987pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 988{ 989 struct pfsync_upd_req *ur, *ura; 990 struct mbuf *mp; 991 int len = count * sizeof(*ur); 992 int i, offp; 993 994 struct pf_state *st; 995 996 mp = m_pulldown(m, offset, len, &offp); 997 if (mp == NULL) { 998 V_pfsyncstats.pfsyncs_badlen++; 999 return (-1); 1000 } 1001 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1002 1003 for (i = 0; i < count; i++) { 1004 ur = &ura[i]; 1005 1006 if (ur->id == 0 && ur->creatorid == 0) 1007 pfsync_bulk_start(); 1008 else { 1009 st = pf_find_state_byid(ur->id, ur->creatorid); 1010 if (st == NULL) { 1011 V_pfsyncstats.pfsyncs_badstate++; 1012 continue; 1013 } 1014 if (st->state_flags & PFSTATE_NOSYNC) { 1015 PF_STATE_UNLOCK(st); 1016 continue; 1017 } 1018 1019 pfsync_update_state_req(st); 1020 PF_STATE_UNLOCK(st); 1021 } 1022 } 1023 1024 return (len); 1025} 1026 1027static int 1028pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1029{ 1030 struct mbuf *mp; 1031 struct pfsync_state *sa, *sp; 1032 struct pf_state *st; 1033 int len = count * sizeof(*sp); 1034 int offp, i; 1035 1036 mp = m_pulldown(m, offset, len, &offp); 1037 if (mp == NULL) { 1038 V_pfsyncstats.pfsyncs_badlen++; 1039 return (-1); 1040 } 1041 sa = (struct pfsync_state *)(mp->m_data + offp); 1042 1043 for (i = 0; i < count; i++) { 1044 sp = &sa[i]; 1045 1046 st = pf_find_state_byid(sp->id, sp->creatorid); 1047 if (st == NULL) { 1048 V_pfsyncstats.pfsyncs_badstate++; 1049 continue; 1050 } 1051 st->state_flags |= PFSTATE_NOSYNC; 1052 pf_unlink_state(st, PF_ENTER_LOCKED); 1053 } 1054 1055 return (len); 1056} 1057 1058static int 1059pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1060{ 1061 struct mbuf *mp; 1062 struct pfsync_del_c *sa, *sp; 1063 struct pf_state *st; 1064 int len = count * sizeof(*sp); 1065 int offp, i; 1066 1067 mp = m_pulldown(m, offset, len, &offp); 1068 if (mp == NULL) { 1069 V_pfsyncstats.pfsyncs_badlen++; 1070 return (-1); 1071 } 1072 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1073 1074 for (i = 0; i < count; i++) { 1075 sp = &sa[i]; 1076 1077 st = pf_find_state_byid(sp->id, sp->creatorid); 1078 if (st == NULL) { 1079 V_pfsyncstats.pfsyncs_badstate++; 1080 continue; 1081 } 1082 1083 st->state_flags |= PFSTATE_NOSYNC; 1084 pf_unlink_state(st, PF_ENTER_LOCKED); 1085 } 1086 1087 return (len); 1088} 1089 1090static int 1091pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1092{ 1093 struct pfsync_softc *sc = V_pfsyncif; 1094 struct pfsync_bus *bus; 1095 struct mbuf *mp; 1096 int len = count * sizeof(*bus); 1097 int offp; 1098 1099 PFSYNC_BLOCK(sc); 1100 1101 /* If we're not waiting for a bulk update, who cares. */ 1102 if (sc->sc_ureq_sent == 0) { 1103 PFSYNC_BUNLOCK(sc); 1104 return (len); 1105 } 1106 1107 mp = m_pulldown(m, offset, len, &offp); 1108 if (mp == NULL) { 1109 PFSYNC_BUNLOCK(sc); 1110 V_pfsyncstats.pfsyncs_badlen++; 1111 return (-1); 1112 } 1113 bus = (struct pfsync_bus *)(mp->m_data + offp); 1114 1115 switch (bus->status) { 1116 case PFSYNC_BUS_START: 1117 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1118 V_pf_limits[PF_LIMIT_STATES].limit / 1119 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1120 sizeof(struct pfsync_state)), 1121 pfsync_bulk_fail, sc); 1122 if (V_pf_status.debug >= PF_DEBUG_MISC) 1123 printf("pfsync: received bulk update start\n"); 1124 break; 1125 1126 case PFSYNC_BUS_END: 1127 if (time_uptime - ntohl(bus->endtime) >= 1128 sc->sc_ureq_sent) { 1129 /* that's it, we're happy */ 1130 sc->sc_ureq_sent = 0; 1131 sc->sc_bulk_tries = 0; 1132 callout_stop(&sc->sc_bulkfail_tmo); 1133 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1134 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1135 "pfsync bulk done"); 1136 sc->sc_flags |= PFSYNCF_OK; 1137 if (V_pf_status.debug >= PF_DEBUG_MISC) 1138 printf("pfsync: received valid " 1139 "bulk update end\n"); 1140 } else { 1141 if (V_pf_status.debug >= PF_DEBUG_MISC) 1142 printf("pfsync: received invalid " 1143 "bulk update end: bad timestamp\n"); 1144 } 1145 break; 1146 } 1147 PFSYNC_BUNLOCK(sc); 1148 1149 return (len); 1150} 1151 1152static int 1153pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1154{ 1155 int len = count * sizeof(struct pfsync_tdb); 1156 1157#if defined(IPSEC) 1158 struct pfsync_tdb *tp; 1159 struct mbuf *mp; 1160 int offp; 1161 int i; 1162 int s; 1163 1164 mp = m_pulldown(m, offset, len, &offp); 1165 if (mp == NULL) { 1166 V_pfsyncstats.pfsyncs_badlen++; 1167 return (-1); 1168 } 1169 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1170 1171 for (i = 0; i < count; i++) 1172 pfsync_update_net_tdb(&tp[i]); 1173#endif 1174 1175 return (len); 1176} 1177 1178#if defined(IPSEC) 1179/* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1180static void 1181pfsync_update_net_tdb(struct pfsync_tdb *pt) 1182{ 1183 struct tdb *tdb; 1184 int s; 1185 1186 /* check for invalid values */ 1187 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1188 (pt->dst.sa.sa_family != AF_INET && 1189 pt->dst.sa.sa_family != AF_INET6)) 1190 goto bad; 1191 1192 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1193 if (tdb) { 1194 pt->rpl = ntohl(pt->rpl); 1195 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1196 1197 /* Neither replay nor byte counter should ever decrease. */ 1198 if (pt->rpl < tdb->tdb_rpl || 1199 pt->cur_bytes < tdb->tdb_cur_bytes) { 1200 goto bad; 1201 } 1202 1203 tdb->tdb_rpl = pt->rpl; 1204 tdb->tdb_cur_bytes = pt->cur_bytes; 1205 } 1206 return; 1207 1208bad: 1209 if (V_pf_status.debug >= PF_DEBUG_MISC) 1210 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1211 "invalid value\n"); 1212 V_pfsyncstats.pfsyncs_badstate++; 1213 return; 1214} 1215#endif 1216 1217 1218static int 1219pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1220{ 1221 /* check if we are at the right place in the packet */ 1222 if (offset != m->m_pkthdr.len - sizeof(struct pfsync_eof)) 1223 V_pfsyncstats.pfsyncs_badact++; 1224 1225 /* we're done. free and let the caller return */ 1226 m_freem(m); 1227 return (-1); 1228} 1229 1230static int 1231pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1232{ 1233 V_pfsyncstats.pfsyncs_badact++; 1234 1235 m_freem(m); 1236 return (-1); 1237} 1238 1239static int 1240pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 1241 struct route *rt) 1242{ 1243 m_freem(m); 1244 return (0); 1245} 1246 1247/* ARGSUSED */ 1248static int 1249pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1250{ 1251 struct pfsync_softc *sc = ifp->if_softc; 1252 struct ifreq *ifr = (struct ifreq *)data; 1253 struct pfsyncreq pfsyncr; 1254 int error; 1255 1256 switch (cmd) { 1257 case SIOCSIFFLAGS: 1258 PFSYNC_LOCK(sc); 1259 if (ifp->if_flags & IFF_UP) { 1260 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1261 PFSYNC_UNLOCK(sc); 1262 pfsync_pointers_init(); 1263 } else { 1264 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1265 PFSYNC_UNLOCK(sc); 1266 pfsync_pointers_uninit(); 1267 } 1268 break; 1269 case SIOCSIFMTU: 1270 if (!sc->sc_sync_if || 1271 ifr->ifr_mtu <= PFSYNC_MINPKT || 1272 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1273 return (EINVAL); 1274 if (ifr->ifr_mtu < ifp->if_mtu) { 1275 PFSYNC_LOCK(sc); 1276 if (sc->sc_len > PFSYNC_MINPKT) 1277 pfsync_sendout(1); 1278 PFSYNC_UNLOCK(sc); 1279 } 1280 ifp->if_mtu = ifr->ifr_mtu; 1281 break; 1282 case SIOCGETPFSYNC: 1283 bzero(&pfsyncr, sizeof(pfsyncr)); 1284 PFSYNC_LOCK(sc); 1285 if (sc->sc_sync_if) { 1286 strlcpy(pfsyncr.pfsyncr_syncdev, 1287 sc->sc_sync_if->if_xname, IFNAMSIZ); 1288 } 1289 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1290 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1291 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER == 1292 (sc->sc_flags & PFSYNCF_DEFER)); 1293 PFSYNC_UNLOCK(sc); 1294 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))); 1295 1296 case SIOCSETPFSYNC: 1297 { 1298 struct ip_moptions *imo = &sc->sc_imo; 1299 struct ifnet *sifp; 1300 struct ip *ip; 1301 void *mship = NULL; 1302 1303 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1304 return (error); 1305 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 1306 return (error); 1307 1308 if (pfsyncr.pfsyncr_maxupdates > 255) 1309 return (EINVAL); 1310 1311 if (pfsyncr.pfsyncr_syncdev[0] == 0) 1312 sifp = NULL; 1313 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) 1314 return (EINVAL); 1315 1316 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0 && sifp != NULL) 1317 mship = malloc((sizeof(struct in_multi *) * 1318 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO); 1319 1320 PFSYNC_LOCK(sc); 1321 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1322 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); 1323 else 1324 sc->sc_sync_peer.s_addr = 1325 pfsyncr.pfsyncr_syncpeer.s_addr; 1326 1327 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1328 if (pfsyncr.pfsyncr_defer) { 1329 sc->sc_flags |= PFSYNCF_DEFER; 1330 pfsync_defer_ptr = pfsync_defer; 1331 } else { 1332 sc->sc_flags &= ~PFSYNCF_DEFER; 1333 pfsync_defer_ptr = NULL; 1334 } 1335 1336 if (sifp == NULL) { 1337 if (sc->sc_sync_if) 1338 if_rele(sc->sc_sync_if); 1339 sc->sc_sync_if = NULL; 1340 if (imo->imo_membership) 1341 pfsync_multicast_cleanup(sc); 1342 PFSYNC_UNLOCK(sc); 1343 break; 1344 } 1345 1346 if (sc->sc_len > PFSYNC_MINPKT && 1347 (sifp->if_mtu < sc->sc_ifp->if_mtu || 1348 (sc->sc_sync_if != NULL && 1349 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 1350 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 1351 pfsync_sendout(1); 1352 1353 if (imo->imo_membership) 1354 pfsync_multicast_cleanup(sc); 1355 1356 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 1357 error = pfsync_multicast_setup(sc, sifp, mship); 1358 if (error) { 1359 if_rele(sifp); 1360 free(mship, M_PFSYNC); 1361 return (error); 1362 } 1363 } 1364 if (sc->sc_sync_if) 1365 if_rele(sc->sc_sync_if); 1366 sc->sc_sync_if = sifp; 1367 1368 ip = &sc->sc_template; 1369 bzero(ip, sizeof(*ip)); 1370 ip->ip_v = IPVERSION; 1371 ip->ip_hl = sizeof(sc->sc_template) >> 2; 1372 ip->ip_tos = IPTOS_LOWDELAY; 1373 /* len and id are set later. */ 1374 ip->ip_off = htons(IP_DF); 1375 ip->ip_ttl = PFSYNC_DFLTTL; 1376 ip->ip_p = IPPROTO_PFSYNC; 1377 ip->ip_src.s_addr = INADDR_ANY; 1378 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; 1379 1380 /* Request a full state table update. */ 1381 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1382 (*carp_demote_adj_p)(V_pfsync_carp_adj, 1383 "pfsync bulk start"); 1384 sc->sc_flags &= ~PFSYNCF_OK; 1385 if (V_pf_status.debug >= PF_DEBUG_MISC) 1386 printf("pfsync: requesting bulk update\n"); 1387 pfsync_request_update(0, 0); 1388 PFSYNC_UNLOCK(sc); 1389 PFSYNC_BLOCK(sc); 1390 sc->sc_ureq_sent = time_uptime; 1391 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, 1392 sc); 1393 PFSYNC_BUNLOCK(sc); 1394 1395 break; 1396 } 1397 default: 1398 return (ENOTTY); 1399 } 1400 1401 return (0); 1402} 1403 1404static void 1405pfsync_out_state(struct pf_state *st, void *buf) 1406{ 1407 struct pfsync_state *sp = buf; 1408 1409 pfsync_state_export(sp, st); 1410} 1411 1412static void 1413pfsync_out_iack(struct pf_state *st, void *buf) 1414{ 1415 struct pfsync_ins_ack *iack = buf; 1416 1417 iack->id = st->id; 1418 iack->creatorid = st->creatorid; 1419} 1420 1421static void 1422pfsync_out_upd_c(struct pf_state *st, void *buf) 1423{ 1424 struct pfsync_upd_c *up = buf; 1425 1426 bzero(up, sizeof(*up)); 1427 up->id = st->id; 1428 pf_state_peer_hton(&st->src, &up->src); 1429 pf_state_peer_hton(&st->dst, &up->dst); 1430 up->creatorid = st->creatorid; 1431 up->timeout = st->timeout; 1432} 1433 1434static void 1435pfsync_out_del(struct pf_state *st, void *buf) 1436{ 1437 struct pfsync_del_c *dp = buf; 1438 1439 dp->id = st->id; 1440 dp->creatorid = st->creatorid; 1441 st->state_flags |= PFSTATE_NOSYNC; 1442} 1443 1444static void 1445pfsync_drop(struct pfsync_softc *sc) 1446{ 1447 struct pf_state *st, *next; 1448 struct pfsync_upd_req_item *ur; 1449 int q; 1450 1451 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1452 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1453 continue; 1454 1455 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) { 1456 KASSERT(st->sync_state == q, 1457 ("%s: st->sync_state == q", 1458 __func__)); 1459 st->sync_state = PFSYNC_S_NONE; 1460 pf_release_state(st); 1461 } 1462 TAILQ_INIT(&sc->sc_qs[q]); 1463 } 1464 1465 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1466 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1467 free(ur, M_PFSYNC); 1468 } 1469 1470 sc->sc_plus = NULL; 1471 sc->sc_len = PFSYNC_MINPKT; 1472} 1473 1474static void 1475pfsync_sendout(int schedswi) 1476{ 1477 struct pfsync_softc *sc = V_pfsyncif; 1478 struct ifnet *ifp = sc->sc_ifp; 1479 struct mbuf *m; 1480 struct ip *ip; 1481 struct pfsync_header *ph; 1482 struct pfsync_subheader *subh; 1483 struct pf_state *st; 1484 struct pfsync_upd_req_item *ur; 1485 int offset; 1486 int q, count = 0; 1487 1488 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1489 KASSERT(sc->sc_len > PFSYNC_MINPKT, 1490 ("%s: sc_len %zu", __func__, sc->sc_len)); 1491 PFSYNC_LOCK_ASSERT(sc); 1492 1493 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1494 pfsync_drop(sc); 1495 return; 1496 } 1497 1498 m = m_get2(M_NOWAIT, MT_DATA, M_PKTHDR, max_linkhdr + sc->sc_len); 1499 if (m == NULL) { 1500 sc->sc_ifp->if_oerrors++; 1501 V_pfsyncstats.pfsyncs_onomem++; 1502 return; 1503 } 1504 m->m_data += max_linkhdr; 1505 m->m_len = m->m_pkthdr.len = sc->sc_len; 1506 1507 /* build the ip header */ 1508 ip = (struct ip *)m->m_data; 1509 bcopy(&sc->sc_template, ip, sizeof(*ip)); 1510 offset = sizeof(*ip); 1511 1512 ip->ip_len = htons(m->m_pkthdr.len); 1513 ip->ip_id = htons(ip_randomid()); 1514 1515 /* build the pfsync header */ 1516 ph = (struct pfsync_header *)(m->m_data + offset); 1517 bzero(ph, sizeof(*ph)); 1518 offset += sizeof(*ph); 1519 1520 ph->version = PFSYNC_VERSION; 1521 ph->len = htons(sc->sc_len - sizeof(*ip)); 1522 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1523 1524 /* walk the queues */ 1525 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1526 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1527 continue; 1528 1529 subh = (struct pfsync_subheader *)(m->m_data + offset); 1530 offset += sizeof(*subh); 1531 1532 count = 0; 1533 TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) { 1534 KASSERT(st->sync_state == q, 1535 ("%s: st->sync_state == q", 1536 __func__)); 1537 /* 1538 * XXXGL: some of write methods do unlocked reads 1539 * of state data :( 1540 */ 1541 pfsync_qs[q].write(st, m->m_data + offset); 1542 offset += pfsync_qs[q].len; 1543 st->sync_state = PFSYNC_S_NONE; 1544 pf_release_state(st); 1545 count++; 1546 } 1547 TAILQ_INIT(&sc->sc_qs[q]); 1548 1549 bzero(subh, sizeof(*subh)); 1550 subh->action = pfsync_qs[q].action; 1551 subh->count = htons(count); 1552 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1553 } 1554 1555 if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) { 1556 subh = (struct pfsync_subheader *)(m->m_data + offset); 1557 offset += sizeof(*subh); 1558 1559 count = 0; 1560 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1561 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1562 1563 bcopy(&ur->ur_msg, m->m_data + offset, 1564 sizeof(ur->ur_msg)); 1565 offset += sizeof(ur->ur_msg); 1566 free(ur, M_PFSYNC); 1567 count++; 1568 } 1569 1570 bzero(subh, sizeof(*subh)); 1571 subh->action = PFSYNC_ACT_UPD_REQ; 1572 subh->count = htons(count); 1573 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1574 } 1575 1576 /* has someone built a custom region for us to add? */ 1577 if (sc->sc_plus != NULL) { 1578 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen); 1579 offset += sc->sc_pluslen; 1580 1581 sc->sc_plus = NULL; 1582 } 1583 1584 subh = (struct pfsync_subheader *)(m->m_data + offset); 1585 offset += sizeof(*subh); 1586 1587 bzero(subh, sizeof(*subh)); 1588 subh->action = PFSYNC_ACT_EOF; 1589 subh->count = htons(1); 1590 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1591 1592 /* XXX write checksum in EOF here */ 1593 1594 /* we're done, let's put it on the wire */ 1595 if (ifp->if_bpf) { 1596 m->m_data += sizeof(*ip); 1597 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip); 1598 BPF_MTAP(ifp, m); 1599 m->m_data -= sizeof(*ip); 1600 m->m_len = m->m_pkthdr.len = sc->sc_len; 1601 } 1602 1603 if (sc->sc_sync_if == NULL) { 1604 sc->sc_len = PFSYNC_MINPKT; 1605 m_freem(m); 1606 return; 1607 } 1608 1609 sc->sc_ifp->if_opackets++; 1610 sc->sc_ifp->if_obytes += m->m_pkthdr.len; 1611 sc->sc_len = PFSYNC_MINPKT; 1612 1613 if (!_IF_QFULL(&sc->sc_ifp->if_snd)) 1614 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1615 else { 1616 m_freem(m); 1617 sc->sc_ifp->if_snd.ifq_drops++; 1618 } 1619 if (schedswi) 1620 swi_sched(V_pfsync_swi_cookie, 0); 1621} 1622 1623static void 1624pfsync_insert_state(struct pf_state *st) 1625{ 1626 struct pfsync_softc *sc = V_pfsyncif; 1627 1628 if (st->state_flags & PFSTATE_NOSYNC) 1629 return; 1630 1631 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1632 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1633 st->state_flags |= PFSTATE_NOSYNC; 1634 return; 1635 } 1636 1637 KASSERT(st->sync_state == PFSYNC_S_NONE, 1638 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1639 1640 PFSYNC_LOCK(sc); 1641 if (sc->sc_len == PFSYNC_MINPKT) 1642 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1643 1644 pfsync_q_ins(st, PFSYNC_S_INS); 1645 PFSYNC_UNLOCK(sc); 1646 1647 st->sync_updates = 0; 1648} 1649 1650static int 1651pfsync_defer(struct pf_state *st, struct mbuf *m) 1652{ 1653 struct pfsync_softc *sc = V_pfsyncif; 1654 struct pfsync_deferral *pd; 1655 1656 if (m->m_flags & (M_BCAST|M_MCAST)) 1657 return (0); 1658 1659 PFSYNC_LOCK(sc); 1660 1661 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) || 1662 !(sc->sc_flags & PFSYNCF_DEFER)) { 1663 PFSYNC_UNLOCK(sc); 1664 return (0); 1665 } 1666 1667 if (sc->sc_deferred >= 128) 1668 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0); 1669 1670 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1671 if (pd == NULL) 1672 return (0); 1673 sc->sc_deferred++; 1674 1675 m->m_flags |= M_SKIP_FIREWALL; 1676 st->state_flags |= PFSTATE_ACK; 1677 1678 pd->pd_sc = sc; 1679 pd->pd_refs = 0; 1680 pd->pd_st = st; 1681 pf_ref_state(st); 1682 pd->pd_m = m; 1683 1684 TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry); 1685 callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1686 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd); 1687 1688 pfsync_push(sc); 1689 1690 return (1); 1691} 1692 1693static void 1694pfsync_undefer(struct pfsync_deferral *pd, int drop) 1695{ 1696 struct pfsync_softc *sc = pd->pd_sc; 1697 struct mbuf *m = pd->pd_m; 1698 struct pf_state *st = pd->pd_st; 1699 1700 PFSYNC_LOCK_ASSERT(sc); 1701 1702 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1703 sc->sc_deferred--; 1704 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1705 free(pd, M_PFSYNC); 1706 pf_release_state(st); 1707 1708 if (drop) 1709 m_freem(m); 1710 else { 1711 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1712 pfsync_push(sc); 1713 } 1714} 1715 1716static void 1717pfsync_defer_tmo(void *arg) 1718{ 1719 struct pfsync_deferral *pd = arg; 1720 struct pfsync_softc *sc = pd->pd_sc; 1721 struct mbuf *m = pd->pd_m; 1722 struct pf_state *st = pd->pd_st; 1723 1724 PFSYNC_LOCK_ASSERT(sc); 1725 1726 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); 1727 1728 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1729 sc->sc_deferred--; 1730 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1731 if (pd->pd_refs == 0) 1732 free(pd, M_PFSYNC); 1733 PFSYNC_UNLOCK(sc); 1734 1735 ip_output(m, NULL, NULL, 0, NULL, NULL); 1736 1737 pf_release_state(st); 1738 1739 CURVNET_RESTORE(); 1740} 1741 1742static void 1743pfsync_undefer_state(struct pf_state *st, int drop) 1744{ 1745 struct pfsync_softc *sc = V_pfsyncif; 1746 struct pfsync_deferral *pd; 1747 1748 PFSYNC_LOCK_ASSERT(sc); 1749 1750 TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) { 1751 if (pd->pd_st == st) { 1752 if (callout_stop(&pd->pd_tmo)) 1753 pfsync_undefer(pd, drop); 1754 return; 1755 } 1756 } 1757 1758 panic("%s: unable to find deferred state", __func__); 1759} 1760 1761static void 1762pfsync_update_state(struct pf_state *st) 1763{ 1764 struct pfsync_softc *sc = V_pfsyncif; 1765 int sync = 0; 1766 1767 PF_STATE_LOCK_ASSERT(st); 1768 PFSYNC_LOCK(sc); 1769 1770 if (st->state_flags & PFSTATE_ACK) 1771 pfsync_undefer_state(st, 0); 1772 if (st->state_flags & PFSTATE_NOSYNC) { 1773 if (st->sync_state != PFSYNC_S_NONE) 1774 pfsync_q_del(st); 1775 PFSYNC_UNLOCK(sc); 1776 return; 1777 } 1778 1779 if (sc->sc_len == PFSYNC_MINPKT) 1780 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1781 1782 switch (st->sync_state) { 1783 case PFSYNC_S_UPD_C: 1784 case PFSYNC_S_UPD: 1785 case PFSYNC_S_INS: 1786 /* we're already handling it */ 1787 1788 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 1789 st->sync_updates++; 1790 if (st->sync_updates >= sc->sc_maxupdates) 1791 sync = 1; 1792 } 1793 break; 1794 1795 case PFSYNC_S_IACK: 1796 pfsync_q_del(st); 1797 case PFSYNC_S_NONE: 1798 pfsync_q_ins(st, PFSYNC_S_UPD_C); 1799 st->sync_updates = 0; 1800 break; 1801 1802 default: 1803 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1804 } 1805 1806 if (sync || (time_uptime - st->pfsync_time) < 2) 1807 pfsync_push(sc); 1808 1809 PFSYNC_UNLOCK(sc); 1810} 1811 1812static void 1813pfsync_request_update(u_int32_t creatorid, u_int64_t id) 1814{ 1815 struct pfsync_softc *sc = V_pfsyncif; 1816 struct pfsync_upd_req_item *item; 1817 size_t nlen = sizeof(struct pfsync_upd_req); 1818 1819 PFSYNC_LOCK_ASSERT(sc); 1820 1821 /* 1822 * This code does a bit to prevent multiple update requests for the 1823 * same state being generated. It searches current subheader queue, 1824 * but it doesn't lookup into queue of already packed datagrams. 1825 */ 1826 TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry) 1827 if (item->ur_msg.id == id && 1828 item->ur_msg.creatorid == creatorid) 1829 return; 1830 1831 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 1832 if (item == NULL) 1833 return; /* XXX stats */ 1834 1835 item->ur_msg.id = id; 1836 item->ur_msg.creatorid = creatorid; 1837 1838 if (TAILQ_EMPTY(&sc->sc_upd_req_list)) 1839 nlen += sizeof(struct pfsync_subheader); 1840 1841 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1842 pfsync_sendout(1); 1843 1844 nlen = sizeof(struct pfsync_subheader) + 1845 sizeof(struct pfsync_upd_req); 1846 } 1847 1848 TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry); 1849 sc->sc_len += nlen; 1850} 1851 1852static void 1853pfsync_update_state_req(struct pf_state *st) 1854{ 1855 struct pfsync_softc *sc = V_pfsyncif; 1856 1857 PF_STATE_LOCK_ASSERT(st); 1858 PFSYNC_LOCK(sc); 1859 1860 if (st->state_flags & PFSTATE_NOSYNC) { 1861 if (st->sync_state != PFSYNC_S_NONE) 1862 pfsync_q_del(st); 1863 PFSYNC_UNLOCK(sc); 1864 return; 1865 } 1866 1867 switch (st->sync_state) { 1868 case PFSYNC_S_UPD_C: 1869 case PFSYNC_S_IACK: 1870 pfsync_q_del(st); 1871 case PFSYNC_S_NONE: 1872 pfsync_q_ins(st, PFSYNC_S_UPD); 1873 pfsync_push(sc); 1874 break; 1875 1876 case PFSYNC_S_INS: 1877 case PFSYNC_S_UPD: 1878 case PFSYNC_S_DEL: 1879 /* we're already handling it */ 1880 break; 1881 1882 default: 1883 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1884 } 1885 1886 PFSYNC_UNLOCK(sc); 1887} 1888 1889static void 1890pfsync_delete_state(struct pf_state *st) 1891{ 1892 struct pfsync_softc *sc = V_pfsyncif; 1893 1894 PFSYNC_LOCK(sc); 1895 if (st->state_flags & PFSTATE_ACK) 1896 pfsync_undefer_state(st, 1); 1897 if (st->state_flags & PFSTATE_NOSYNC) { 1898 if (st->sync_state != PFSYNC_S_NONE) 1899 pfsync_q_del(st); 1900 PFSYNC_UNLOCK(sc); 1901 return; 1902 } 1903 1904 if (sc->sc_len == PFSYNC_MINPKT) 1905 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1906 1907 switch (st->sync_state) { 1908 case PFSYNC_S_INS: 1909 /* We never got to tell the world so just forget about it. */ 1910 pfsync_q_del(st); 1911 break; 1912 1913 case PFSYNC_S_UPD_C: 1914 case PFSYNC_S_UPD: 1915 case PFSYNC_S_IACK: 1916 pfsync_q_del(st); 1917 /* FALLTHROUGH to putting it on the del list */ 1918 1919 case PFSYNC_S_NONE: 1920 pfsync_q_ins(st, PFSYNC_S_DEL); 1921 break; 1922 1923 default: 1924 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1925 } 1926 PFSYNC_UNLOCK(sc); 1927} 1928 1929static void 1930pfsync_clear_states(u_int32_t creatorid, const char *ifname) 1931{ 1932 struct pfsync_softc *sc = V_pfsyncif; 1933 struct { 1934 struct pfsync_subheader subh; 1935 struct pfsync_clr clr; 1936 } __packed r; 1937 1938 bzero(&r, sizeof(r)); 1939 1940 r.subh.action = PFSYNC_ACT_CLR; 1941 r.subh.count = htons(1); 1942 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 1943 1944 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 1945 r.clr.creatorid = creatorid; 1946 1947 PFSYNC_LOCK(sc); 1948 pfsync_send_plus(&r, sizeof(r)); 1949 PFSYNC_UNLOCK(sc); 1950} 1951 1952static void 1953pfsync_q_ins(struct pf_state *st, int q) 1954{ 1955 struct pfsync_softc *sc = V_pfsyncif; 1956 size_t nlen = pfsync_qs[q].len; 1957 1958 PFSYNC_LOCK_ASSERT(sc); 1959 1960 KASSERT(st->sync_state == PFSYNC_S_NONE, 1961 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1962 KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 1963 sc->sc_len)); 1964 1965 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1966 nlen += sizeof(struct pfsync_subheader); 1967 1968 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1969 pfsync_sendout(1); 1970 1971 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 1972 } 1973 1974 sc->sc_len += nlen; 1975 TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list); 1976 st->sync_state = q; 1977 pf_ref_state(st); 1978} 1979 1980static void 1981pfsync_q_del(struct pf_state *st) 1982{ 1983 struct pfsync_softc *sc = V_pfsyncif; 1984 int q = st->sync_state; 1985 1986 PFSYNC_LOCK_ASSERT(sc); 1987 KASSERT(st->sync_state != PFSYNC_S_NONE, 1988 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 1989 1990 sc->sc_len -= pfsync_qs[q].len; 1991 TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list); 1992 st->sync_state = PFSYNC_S_NONE; 1993 pf_release_state(st); 1994 1995 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1996 sc->sc_len -= sizeof(struct pfsync_subheader); 1997} 1998 1999static void 2000pfsync_bulk_start(void) 2001{ 2002 struct pfsync_softc *sc = V_pfsyncif; 2003 2004 if (V_pf_status.debug >= PF_DEBUG_MISC) 2005 printf("pfsync: received bulk update request\n"); 2006 2007 PFSYNC_BLOCK(sc); 2008 2009 sc->sc_ureq_received = time_uptime; 2010 sc->sc_bulk_hashid = 0; 2011 sc->sc_bulk_stateid = 0; 2012 pfsync_bulk_status(PFSYNC_BUS_START); 2013 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2014 PFSYNC_BUNLOCK(sc); 2015} 2016 2017static void 2018pfsync_bulk_update(void *arg) 2019{ 2020 struct pfsync_softc *sc = arg; 2021 struct pf_state *s; 2022 int i, sent = 0; 2023 2024 PFSYNC_BLOCK_ASSERT(sc); 2025 CURVNET_SET(sc->sc_ifp->if_vnet); 2026 2027 /* 2028 * Start with last state from previous invocation. 2029 * It may had gone, in this case start from the 2030 * hash slot. 2031 */ 2032 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2033 2034 if (s != NULL) 2035 i = PF_IDHASH(s); 2036 else 2037 i = sc->sc_bulk_hashid; 2038 2039 for (; i <= V_pf_hashmask; i++) { 2040 struct pf_idhash *ih = &V_pf_idhash[i]; 2041 2042 if (s != NULL) 2043 PF_HASHROW_ASSERT(ih); 2044 else { 2045 PF_HASHROW_LOCK(ih); 2046 s = LIST_FIRST(&ih->states); 2047 } 2048 2049 for (; s; s = LIST_NEXT(s, entry)) { 2050 2051 if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) < 2052 sizeof(struct pfsync_state)) { 2053 /* We've filled a packet. */ 2054 sc->sc_bulk_hashid = i; 2055 sc->sc_bulk_stateid = s->id; 2056 sc->sc_bulk_creatorid = s->creatorid; 2057 PF_HASHROW_UNLOCK(ih); 2058 callout_reset(&sc->sc_bulk_tmo, 1, 2059 pfsync_bulk_update, sc); 2060 goto full; 2061 } 2062 2063 if (s->sync_state == PFSYNC_S_NONE && 2064 s->timeout < PFTM_MAX && 2065 s->pfsync_time <= sc->sc_ureq_received) { 2066 pfsync_update_state_req(s); 2067 sent++; 2068 } 2069 } 2070 PF_HASHROW_UNLOCK(ih); 2071 } 2072 2073 /* We're done. */ 2074 pfsync_bulk_status(PFSYNC_BUS_END); 2075 2076full: 2077 CURVNET_RESTORE(); 2078} 2079 2080static void 2081pfsync_bulk_status(u_int8_t status) 2082{ 2083 struct { 2084 struct pfsync_subheader subh; 2085 struct pfsync_bus bus; 2086 } __packed r; 2087 2088 struct pfsync_softc *sc = V_pfsyncif; 2089 2090 bzero(&r, sizeof(r)); 2091 2092 r.subh.action = PFSYNC_ACT_BUS; 2093 r.subh.count = htons(1); 2094 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2095 2096 r.bus.creatorid = V_pf_status.hostid; 2097 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2098 r.bus.status = status; 2099 2100 PFSYNC_LOCK(sc); 2101 pfsync_send_plus(&r, sizeof(r)); 2102 PFSYNC_UNLOCK(sc); 2103} 2104 2105static void 2106pfsync_bulk_fail(void *arg) 2107{ 2108 struct pfsync_softc *sc = arg; 2109 2110 CURVNET_SET(sc->sc_ifp->if_vnet); 2111 2112 PFSYNC_BLOCK_ASSERT(sc); 2113 2114 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2115 /* Try again */ 2116 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2117 pfsync_bulk_fail, V_pfsyncif); 2118 PFSYNC_LOCK(sc); 2119 pfsync_request_update(0, 0); 2120 PFSYNC_UNLOCK(sc); 2121 } else { 2122 /* Pretend like the transfer was ok. */ 2123 sc->sc_ureq_sent = 0; 2124 sc->sc_bulk_tries = 0; 2125 PFSYNC_LOCK(sc); 2126 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2127 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2128 "pfsync bulk fail"); 2129 sc->sc_flags |= PFSYNCF_OK; 2130 PFSYNC_UNLOCK(sc); 2131 if (V_pf_status.debug >= PF_DEBUG_MISC) 2132 printf("pfsync: failed to receive bulk update\n"); 2133 } 2134 2135 CURVNET_RESTORE(); 2136} 2137 2138static void 2139pfsync_send_plus(void *plus, size_t pluslen) 2140{ 2141 struct pfsync_softc *sc = V_pfsyncif; 2142 2143 PFSYNC_LOCK_ASSERT(sc); 2144 2145 if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu) 2146 pfsync_sendout(1); 2147 2148 sc->sc_plus = plus; 2149 sc->sc_len += (sc->sc_pluslen = pluslen); 2150 2151 pfsync_sendout(1); 2152} 2153 2154static void 2155pfsync_timeout(void *arg) 2156{ 2157 struct pfsync_softc *sc = arg; 2158 2159 CURVNET_SET(sc->sc_ifp->if_vnet); 2160 PFSYNC_LOCK(sc); 2161 pfsync_push(sc); 2162 PFSYNC_UNLOCK(sc); 2163 CURVNET_RESTORE(); 2164} 2165 2166static void 2167pfsync_push(struct pfsync_softc *sc) 2168{ 2169 2170 PFSYNC_LOCK_ASSERT(sc); 2171 2172 sc->sc_flags |= PFSYNCF_PUSH; 2173 swi_sched(V_pfsync_swi_cookie, 0); 2174} 2175 2176static void 2177pfsyncintr(void *arg) 2178{ 2179 struct pfsync_softc *sc = arg; 2180 struct mbuf *m, *n; 2181 2182 CURVNET_SET(sc->sc_ifp->if_vnet); 2183 2184 PFSYNC_LOCK(sc); 2185 if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) { 2186 pfsync_sendout(0); 2187 sc->sc_flags &= ~PFSYNCF_PUSH; 2188 } 2189 _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m); 2190 PFSYNC_UNLOCK(sc); 2191 2192 for (; m != NULL; m = n) { 2193 2194 n = m->m_nextpkt; 2195 m->m_nextpkt = NULL; 2196 2197 /* 2198 * We distinguish between a deferral packet and our 2199 * own pfsync packet based on M_SKIP_FIREWALL 2200 * flag. This is XXX. 2201 */ 2202 if (m->m_flags & M_SKIP_FIREWALL) 2203 ip_output(m, NULL, NULL, 0, NULL, NULL); 2204 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 2205 NULL) == 0) 2206 V_pfsyncstats.pfsyncs_opackets++; 2207 else 2208 V_pfsyncstats.pfsyncs_oerrors++; 2209 } 2210 CURVNET_RESTORE(); 2211} 2212 2213static int 2214pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship) 2215{ 2216 struct ip_moptions *imo = &sc->sc_imo; 2217 int error; 2218 2219 if (!(ifp->if_flags & IFF_MULTICAST)) 2220 return (EADDRNOTAVAIL); 2221 2222 imo->imo_membership = (struct in_multi **)mship; 2223 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 2224 imo->imo_multicast_vif = -1; 2225 2226 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, 2227 &imo->imo_membership[0])) != 0) { 2228 imo->imo_membership = NULL; 2229 return (error); 2230 } 2231 imo->imo_num_memberships++; 2232 imo->imo_multicast_ifp = ifp; 2233 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2234 imo->imo_multicast_loop = 0; 2235 2236 return (0); 2237} 2238 2239static void 2240pfsync_multicast_cleanup(struct pfsync_softc *sc) 2241{ 2242 struct ip_moptions *imo = &sc->sc_imo; 2243 2244 in_leavegroup(imo->imo_membership[0], NULL); 2245 free(imo->imo_membership, M_PFSYNC); 2246 imo->imo_membership = NULL; 2247 imo->imo_multicast_ifp = NULL; 2248} 2249 2250#ifdef INET 2251extern struct domain inetdomain; 2252static struct protosw in_pfsync_protosw = { 2253 .pr_type = SOCK_RAW, 2254 .pr_domain = &inetdomain, 2255 .pr_protocol = IPPROTO_PFSYNC, 2256 .pr_flags = PR_ATOMIC|PR_ADDR, 2257 .pr_input = pfsync_input, 2258 .pr_output = (pr_output_t *)rip_output, 2259 .pr_ctloutput = rip_ctloutput, 2260 .pr_usrreqs = &rip_usrreqs 2261}; 2262#endif 2263 2264static void 2265pfsync_pointers_init() 2266{ 2267 2268 PF_RULES_WLOCK(); 2269 pfsync_state_import_ptr = pfsync_state_import; 2270 pfsync_insert_state_ptr = pfsync_insert_state; 2271 pfsync_update_state_ptr = pfsync_update_state; 2272 pfsync_delete_state_ptr = pfsync_delete_state; 2273 pfsync_clear_states_ptr = pfsync_clear_states; 2274 pfsync_defer_ptr = pfsync_defer; 2275 PF_RULES_WUNLOCK(); 2276} 2277 2278static void 2279pfsync_pointers_uninit() 2280{ 2281 2282 PF_RULES_WLOCK(); 2283 pfsync_state_import_ptr = NULL; 2284 pfsync_insert_state_ptr = NULL; 2285 pfsync_update_state_ptr = NULL; 2286 pfsync_delete_state_ptr = NULL; 2287 pfsync_clear_states_ptr = NULL; 2288 pfsync_defer_ptr = NULL; 2289 PF_RULES_WUNLOCK(); 2290} 2291 2292static int 2293pfsync_init() 2294{ 2295 VNET_ITERATOR_DECL(vnet_iter); 2296 int error = 0; 2297 2298 VNET_LIST_RLOCK(); 2299 VNET_FOREACH(vnet_iter) { 2300 CURVNET_SET(vnet_iter); 2301 V_pfsync_cloner = if_clone_simple(pfsyncname, 2302 pfsync_clone_create, pfsync_clone_destroy, 1); 2303 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif, 2304 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2305 CURVNET_RESTORE(); 2306 if (error) 2307 goto fail_locked; 2308 } 2309 VNET_LIST_RUNLOCK(); 2310#ifdef INET 2311 error = pf_proto_register(PF_INET, &in_pfsync_protosw); 2312 if (error) 2313 goto fail; 2314 error = ipproto_register(IPPROTO_PFSYNC); 2315 if (error) { 2316 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2317 goto fail; 2318 } 2319#endif 2320 pfsync_pointers_init(); 2321 2322 return (0); 2323 2324fail: 2325 VNET_LIST_RLOCK(); 2326fail_locked: 2327 VNET_FOREACH(vnet_iter) { 2328 CURVNET_SET(vnet_iter); 2329 if (V_pfsync_swi_cookie) { 2330 swi_remove(V_pfsync_swi_cookie); 2331 if_clone_detach(V_pfsync_cloner); 2332 } 2333 CURVNET_RESTORE(); 2334 } 2335 VNET_LIST_RUNLOCK(); 2336 2337 return (error); 2338} 2339 2340static void 2341pfsync_uninit() 2342{ 2343 VNET_ITERATOR_DECL(vnet_iter); 2344 2345 pfsync_pointers_uninit(); 2346 2347 ipproto_unregister(IPPROTO_PFSYNC); 2348 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2349 VNET_LIST_RLOCK(); 2350 VNET_FOREACH(vnet_iter) { 2351 CURVNET_SET(vnet_iter); 2352 if_clone_detach(V_pfsync_cloner); 2353 swi_remove(V_pfsync_swi_cookie); 2354 CURVNET_RESTORE(); 2355 } 2356 VNET_LIST_RUNLOCK(); 2357} 2358 2359static int 2360pfsync_modevent(module_t mod, int type, void *data) 2361{ 2362 int error = 0; 2363 2364 switch (type) { 2365 case MOD_LOAD: 2366 error = pfsync_init(); 2367 break; 2368 case MOD_QUIESCE: 2369 /* 2370 * Module should not be unloaded due to race conditions. 2371 */ 2372 error = EBUSY; 2373 break; 2374 case MOD_UNLOAD: 2375 pfsync_uninit(); 2376 break; 2377 default: 2378 error = EINVAL; 2379 break; 2380 } 2381 2382 return (error); 2383} 2384 2385static moduledata_t pfsync_mod = { 2386 pfsyncname, 2387 pfsync_modevent, 2388 0 2389}; 2390 2391#define PFSYNC_MODVER 1 2392 2393DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); 2394MODULE_VERSION(pfsync, PFSYNC_MODVER); 2395MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); 2396