if_pfsync.c revision 241056
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 * 1.128 - cleanups 51 * 1.146 - bzero() mbuf before sparsely filling it with data 52 * 1.170 - SIOCSIFMTU checks 53 * 1.126, 1.142 - deferred packets processing 54 * 1.173 - correct expire time processing 55 */ 56 57#include <sys/cdefs.h> 58__FBSDID("$FreeBSD: head/sys/netpfil/pf/if_pfsync.c 241056 2012-09-29 20:02:26Z glebius $"); 59 60#include "opt_inet.h" 61#include "opt_inet6.h" 62#include "opt_pf.h" 63 64#include <sys/param.h> 65#include <sys/bus.h> 66#include <sys/endian.h> 67#include <sys/interrupt.h> 68#include <sys/kernel.h> 69#include <sys/lock.h> 70#include <sys/mbuf.h> 71#include <sys/module.h> 72#include <sys/mutex.h> 73#include <sys/priv.h> 74#include <sys/protosw.h> 75#include <sys/socket.h> 76#include <sys/sockio.h> 77#include <sys/sysctl.h> 78 79#include <net/bpf.h> 80#include <net/if.h> 81#include <net/if_clone.h> 82#include <net/if_types.h> 83#include <net/pfvar.h> 84#include <net/if_pfsync.h> 85 86#include <netinet/if_ether.h> 87#include <netinet/in.h> 88#include <netinet/in_var.h> 89#include <netinet/ip.h> 90#include <netinet/ip_carp.h> 91#include <netinet/ip_var.h> 92#include <netinet/tcp.h> 93#include <netinet/tcp_fsm.h> 94#include <netinet/tcp_seq.h> 95 96#define PFSYNC_MINPKT ( \ 97 sizeof(struct ip) + \ 98 sizeof(struct pfsync_header) + \ 99 sizeof(struct pfsync_subheader) + \ 100 sizeof(struct pfsync_eof)) 101 102struct pfsync_pkt { 103 struct ip *ip; 104 struct in_addr src; 105 u_int8_t flags; 106}; 107 108static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *, 109 struct pfsync_state_peer *); 110static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int); 111static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int); 112static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int); 113static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int); 114static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int); 115static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int); 116static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int); 117static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int); 118static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int); 119static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int); 120static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int); 121static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int); 122 123static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = { 124 pfsync_in_clr, /* PFSYNC_ACT_CLR */ 125 pfsync_in_ins, /* PFSYNC_ACT_INS */ 126 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ 127 pfsync_in_upd, /* PFSYNC_ACT_UPD */ 128 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ 129 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ 130 pfsync_in_del, /* PFSYNC_ACT_DEL */ 131 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ 132 pfsync_in_error, /* PFSYNC_ACT_INS_F */ 133 pfsync_in_error, /* PFSYNC_ACT_DEL_F */ 134 pfsync_in_bus, /* PFSYNC_ACT_BUS */ 135 pfsync_in_tdb, /* PFSYNC_ACT_TDB */ 136 pfsync_in_eof /* PFSYNC_ACT_EOF */ 137}; 138 139struct pfsync_q { 140 void (*write)(struct pf_state *, void *); 141 size_t len; 142 u_int8_t action; 143}; 144 145/* we have one of these for every PFSYNC_S_ */ 146static void pfsync_out_state(struct pf_state *, void *); 147static void pfsync_out_iack(struct pf_state *, void *); 148static void pfsync_out_upd_c(struct pf_state *, void *); 149static void pfsync_out_del(struct pf_state *, void *); 150 151static struct pfsync_q pfsync_qs[] = { 152 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS }, 153 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, 154 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD }, 155 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, 156 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } 157}; 158 159static void pfsync_q_ins(struct pf_state *, int); 160static void pfsync_q_del(struct pf_state *); 161 162static void pfsync_update_state(struct pf_state *); 163 164struct pfsync_upd_req_item { 165 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; 166 struct pfsync_upd_req ur_msg; 167}; 168 169struct pfsync_deferral { 170 struct pfsync_softc *pd_sc; 171 TAILQ_ENTRY(pfsync_deferral) pd_entry; 172 u_int pd_refs; 173 struct callout pd_tmo; 174 175 struct pf_state *pd_st; 176 struct mbuf *pd_m; 177}; 178 179struct pfsync_softc { 180 /* Configuration */ 181 struct ifnet *sc_ifp; 182 struct ifnet *sc_sync_if; 183 struct ip_moptions sc_imo; 184 struct in_addr sc_sync_peer; 185 uint32_t sc_flags; 186#define PFSYNCF_OK 0x00000001 187#define PFSYNCF_DEFER 0x00000002 188#define PFSYNCF_PUSH 0x00000004 189 uint8_t sc_maxupdates; 190 struct ip sc_template; 191 struct callout sc_tmo; 192 struct mtx sc_mtx; 193 194 /* Queued data */ 195 size_t sc_len; 196 TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT]; 197 TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list; 198 TAILQ_HEAD(, pfsync_deferral) sc_deferrals; 199 u_int sc_deferred; 200 void *sc_plus; 201 size_t sc_pluslen; 202 203 /* Bulk update info */ 204 struct mtx sc_bulk_mtx; 205 uint32_t sc_ureq_sent; 206 int sc_bulk_tries; 207 uint32_t sc_ureq_received; 208 int sc_bulk_hashid; 209 uint64_t sc_bulk_stateid; 210 uint32_t sc_bulk_creatorid; 211 struct callout sc_bulk_tmo; 212 struct callout sc_bulkfail_tmo; 213}; 214 215#define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 216#define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 217#define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 218 219#define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 220#define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 221#define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 222 223static MALLOC_DEFINE(M_PFSYNC, "pfsync", "pfsync(4) data"); 224static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL; 225#define V_pfsyncif VNET(pfsyncif) 226static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL; 227#define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 228static VNET_DEFINE(struct pfsyncstats, pfsyncstats); 229#define V_pfsyncstats VNET(pfsyncstats) 230static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW; 231#define V_pfsync_carp_adj VNET(pfsync_carp_adj) 232 233static void pfsync_timeout(void *); 234static void pfsync_push(struct pfsync_softc *); 235static void pfsyncintr(void *); 236static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 237 void *); 238static void pfsync_multicast_cleanup(struct pfsync_softc *); 239static int pfsync_init(void); 240static void pfsync_uninit(void); 241 242SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC"); 243SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW, 244 &VNET_NAME(pfsyncstats), pfsyncstats, 245 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 246SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW, 247 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 248 249static int pfsync_clone_create(struct if_clone *, int, caddr_t); 250static void pfsync_clone_destroy(struct ifnet *); 251static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 252 struct pf_state_peer *); 253static int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 254 struct route *); 255static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 256 257static int pfsync_defer(struct pf_state *, struct mbuf *); 258static void pfsync_undefer(struct pfsync_deferral *, int); 259static void pfsync_undefer_state(struct pf_state *, int); 260static void pfsync_defer_tmo(void *); 261 262static void pfsync_request_update(u_int32_t, u_int64_t); 263static void pfsync_update_state_req(struct pf_state *); 264 265static void pfsync_drop(struct pfsync_softc *); 266static void pfsync_sendout(int); 267static void pfsync_send_plus(void *, size_t); 268 269static void pfsync_bulk_start(void); 270static void pfsync_bulk_status(u_int8_t); 271static void pfsync_bulk_update(void *); 272static void pfsync_bulk_fail(void *); 273 274#ifdef IPSEC 275static void pfsync_update_net_tdb(struct pfsync_tdb *); 276#endif 277 278#define PFSYNC_MAX_BULKTRIES 12 279 280VNET_DEFINE(struct ifc_simple_data, pfsync_cloner_data); 281VNET_DEFINE(struct if_clone, pfsync_cloner); 282#define V_pfsync_cloner_data VNET(pfsync_cloner_data) 283#define V_pfsync_cloner VNET(pfsync_cloner) 284IFC_SIMPLE_DECLARE(pfsync, 1); 285 286static int 287pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 288{ 289 struct pfsync_softc *sc; 290 struct ifnet *ifp; 291 int q; 292 293 if (unit != 0) 294 return (EINVAL); 295 296 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 297 sc->sc_flags |= PFSYNCF_OK; 298 299 for (q = 0; q < PFSYNC_S_COUNT; q++) 300 TAILQ_INIT(&sc->sc_qs[q]); 301 302 TAILQ_INIT(&sc->sc_upd_req_list); 303 TAILQ_INIT(&sc->sc_deferrals); 304 305 sc->sc_len = PFSYNC_MINPKT; 306 sc->sc_maxupdates = 128; 307 308 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 309 if (ifp == NULL) { 310 free(sc, M_PFSYNC); 311 return (ENOSPC); 312 } 313 if_initname(ifp, ifc->ifc_name, unit); 314 ifp->if_softc = sc; 315 ifp->if_ioctl = pfsyncioctl; 316 ifp->if_output = pfsyncoutput; 317 ifp->if_type = IFT_PFSYNC; 318 ifp->if_snd.ifq_maxlen = ifqmaxlen; 319 ifp->if_hdrlen = sizeof(struct pfsync_header); 320 ifp->if_mtu = ETHERMTU; 321 mtx_init(&sc->sc_mtx, "pfsync", NULL, MTX_DEF); 322 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 323 callout_init(&sc->sc_tmo, CALLOUT_MPSAFE); 324 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 325 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 326 327 if_attach(ifp); 328 329 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 330 331 V_pfsyncif = sc; 332 333 return (0); 334} 335 336static void 337pfsync_clone_destroy(struct ifnet *ifp) 338{ 339 struct pfsync_softc *sc = ifp->if_softc; 340 341 /* 342 * At this stage, everything should have already been 343 * cleared by pfsync_uninit(), and we have only to 344 * drain callouts. 345 */ 346 while (sc->sc_deferred > 0) { 347 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals); 348 349 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 350 sc->sc_deferred--; 351 if (callout_stop(&pd->pd_tmo)) { 352 pf_release_state(pd->pd_st); 353 m_freem(pd->pd_m); 354 free(pd, M_PFSYNC); 355 } else { 356 pd->pd_refs++; 357 callout_drain(&pd->pd_tmo); 358 free(pd, M_PFSYNC); 359 } 360 } 361 362 callout_drain(&sc->sc_tmo); 363 callout_drain(&sc->sc_bulkfail_tmo); 364 callout_drain(&sc->sc_bulk_tmo); 365 366 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 367 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 368 bpfdetach(ifp); 369 if_detach(ifp); 370 371 pfsync_drop(sc); 372 373 if_free(ifp); 374 if (sc->sc_imo.imo_membership) 375 pfsync_multicast_cleanup(sc); 376 mtx_destroy(&sc->sc_mtx); 377 mtx_destroy(&sc->sc_bulk_mtx); 378 free(sc, M_PFSYNC); 379 380 V_pfsyncif = NULL; 381} 382 383static int 384pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 385 struct pf_state_peer *d) 386{ 387 if (s->scrub.scrub_flag && d->scrub == NULL) { 388 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 389 if (d->scrub == NULL) 390 return (ENOMEM); 391 } 392 393 return (0); 394} 395 396 397static int 398pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 399{ 400 struct pfsync_softc *sc = V_pfsyncif; 401 struct pf_state *st = NULL; 402 struct pf_state_key *skw = NULL, *sks = NULL; 403 struct pf_rule *r = NULL; 404 struct pfi_kif *kif; 405 int error; 406 407 PF_RULES_RASSERT(); 408 409 if (sp->creatorid == 0 && V_pf_status.debug >= PF_DEBUG_MISC) { 410 printf("%s: invalid creator id: %08x\n", __func__, 411 ntohl(sp->creatorid)); 412 return (EINVAL); 413 } 414 415 if ((kif = pfi_kif_find(sp->ifname)) == NULL) { 416 if (V_pf_status.debug >= PF_DEBUG_MISC) 417 printf("%s: unknown interface: %s\n", __func__, 418 sp->ifname); 419 if (flags & PFSYNC_SI_IOCTL) 420 return (EINVAL); 421 return (0); /* skip this state */ 422 } 423 424 /* 425 * If the ruleset checksums match or the state is coming from the ioctl, 426 * it's safe to associate the state with the rule of that number. 427 */ 428 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 429 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 430 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 431 r = pf_main_ruleset.rules[ 432 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 433 else 434 r = &V_pf_default_rule; 435 436 if ((r->max_states && r->states_cur >= r->max_states)) 437 goto cleanup; 438 439 /* 440 * XXXGL: consider M_WAITOK in ioctl path after. 441 */ 442 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL) 443 goto cleanup; 444 445 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 446 goto cleanup; 447 448 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 449 &sp->key[PF_SK_STACK].addr[0], sp->af) || 450 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 451 &sp->key[PF_SK_STACK].addr[1], sp->af) || 452 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 453 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) { 454 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 455 if (sks == NULL) 456 goto cleanup; 457 } else 458 sks = skw; 459 460 /* allocate memory for scrub info */ 461 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 462 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 463 goto cleanup; 464 465 /* copy to state key(s) */ 466 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 467 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 468 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 469 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 470 skw->proto = sp->proto; 471 skw->af = sp->af; 472 if (sks != skw) { 473 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 474 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 475 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 476 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 477 sks->proto = sp->proto; 478 sks->af = sp->af; 479 } 480 481 /* copy to state */ 482 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 483 st->creation = time_uptime - ntohl(sp->creation); 484 st->expire = time_uptime; 485 if (sp->expire) { 486 uint32_t timeout; 487 488 timeout = r->timeout[sp->timeout]; 489 if (!timeout) 490 timeout = V_pf_default_rule.timeout[sp->timeout]; 491 492 /* sp->expire may have been adaptively scaled by export. */ 493 st->expire -= timeout - ntohl(sp->expire); 494 } 495 496 st->direction = sp->direction; 497 st->log = sp->log; 498 st->timeout = sp->timeout; 499 st->state_flags = sp->state_flags; 500 501 st->id = sp->id; 502 st->creatorid = sp->creatorid; 503 pf_state_peer_ntoh(&sp->src, &st->src); 504 pf_state_peer_ntoh(&sp->dst, &st->dst); 505 506 st->rule.ptr = r; 507 st->nat_rule.ptr = NULL; 508 st->anchor.ptr = NULL; 509 st->rt_kif = NULL; 510 511 st->pfsync_time = time_uptime; 512 st->sync_state = PFSYNC_S_NONE; 513 514 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 515 r->states_cur++; 516 r->states_tot++; 517 518 if (!(flags & PFSYNC_SI_IOCTL)) 519 st->state_flags |= PFSTATE_NOSYNC; 520 521 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) { 522 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 523 r->states_cur--; 524 goto cleanup_state; 525 } 526 527 if (!(flags & PFSYNC_SI_IOCTL)) { 528 st->state_flags &= ~PFSTATE_NOSYNC; 529 if (st->state_flags & PFSTATE_ACK) { 530 pfsync_q_ins(st, PFSYNC_S_IACK); 531 pfsync_push(sc); 532 } 533 } 534 st->state_flags &= ~PFSTATE_ACK; 535 PF_STATE_UNLOCK(st); 536 537 return (0); 538 539cleanup: 540 error = ENOMEM; 541 if (skw == sks) 542 sks = NULL; 543 if (skw != NULL) 544 uma_zfree(V_pf_state_key_z, skw); 545 if (sks != NULL) 546 uma_zfree(V_pf_state_key_z, sks); 547 548cleanup_state: /* pf_state_insert() frees the state keys. */ 549 if (st) { 550 if (st->dst.scrub) 551 uma_zfree(V_pf_state_scrub_z, st->dst.scrub); 552 if (st->src.scrub) 553 uma_zfree(V_pf_state_scrub_z, st->src.scrub); 554 uma_zfree(V_pf_state_z, st); 555 } 556 return (error); 557} 558 559static void 560pfsync_input(struct mbuf *m, __unused int off) 561{ 562 struct pfsync_softc *sc = V_pfsyncif; 563 struct pfsync_pkt pkt; 564 struct ip *ip = mtod(m, struct ip *); 565 struct pfsync_header *ph; 566 struct pfsync_subheader subh; 567 568 int offset; 569 int rv; 570 uint16_t count; 571 572 V_pfsyncstats.pfsyncs_ipackets++; 573 574 /* Verify that we have a sync interface configured. */ 575 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 576 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 577 goto done; 578 579 /* verify that the packet came in on the right interface */ 580 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 581 V_pfsyncstats.pfsyncs_badif++; 582 goto done; 583 } 584 585 sc->sc_ifp->if_ipackets++; 586 sc->sc_ifp->if_ibytes += m->m_pkthdr.len; 587 /* verify that the IP TTL is 255. */ 588 if (ip->ip_ttl != PFSYNC_DFLTTL) { 589 V_pfsyncstats.pfsyncs_badttl++; 590 goto done; 591 } 592 593 offset = ip->ip_hl << 2; 594 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 595 V_pfsyncstats.pfsyncs_hdrops++; 596 goto done; 597 } 598 599 if (offset + sizeof(*ph) > m->m_len) { 600 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 601 V_pfsyncstats.pfsyncs_hdrops++; 602 return; 603 } 604 ip = mtod(m, struct ip *); 605 } 606 ph = (struct pfsync_header *)((char *)ip + offset); 607 608 /* verify the version */ 609 if (ph->version != PFSYNC_VERSION) { 610 V_pfsyncstats.pfsyncs_badver++; 611 goto done; 612 } 613 614 /* Cheaper to grab this now than having to mess with mbufs later */ 615 pkt.ip = ip; 616 pkt.src = ip->ip_src; 617 pkt.flags = 0; 618 619 /* 620 * Trusting pf_chksum during packet processing, as well as seeking 621 * in interface name tree, require holding PF_RULES_RLOCK(). 622 */ 623 PF_RULES_RLOCK(); 624 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 625 pkt.flags |= PFSYNC_SI_CKSUM; 626 627 offset += sizeof(*ph); 628 for (;;) { 629 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 630 offset += sizeof(subh); 631 632 if (subh.action >= PFSYNC_ACT_MAX) { 633 V_pfsyncstats.pfsyncs_badact++; 634 PF_RULES_RUNLOCK(); 635 goto done; 636 } 637 638 count = ntohs(subh.count); 639 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 640 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); 641 if (rv == -1) { 642 PF_RULES_RUNLOCK(); 643 return; 644 } 645 646 offset += rv; 647 } 648 PF_RULES_RUNLOCK(); 649 650done: 651 m_freem(m); 652} 653 654static int 655pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 656{ 657 struct pfsync_clr *clr; 658 struct mbuf *mp; 659 int len = sizeof(*clr) * count; 660 int i, offp; 661 u_int32_t creatorid; 662 663 mp = m_pulldown(m, offset, len, &offp); 664 if (mp == NULL) { 665 V_pfsyncstats.pfsyncs_badlen++; 666 return (-1); 667 } 668 clr = (struct pfsync_clr *)(mp->m_data + offp); 669 670 for (i = 0; i < count; i++) { 671 creatorid = clr[i].creatorid; 672 673 if (clr[i].ifname[0] != '\0' && 674 pfi_kif_find(clr[i].ifname) == NULL) 675 continue; 676 677 for (int i = 0; i <= V_pf_hashmask; i++) { 678 struct pf_idhash *ih = &V_pf_idhash[i]; 679 struct pf_state *s; 680relock: 681 PF_HASHROW_LOCK(ih); 682 LIST_FOREACH(s, &ih->states, entry) { 683 if (s->creatorid == creatorid) { 684 s->state_flags |= PFSTATE_NOSYNC; 685 pf_unlink_state(s, PF_ENTER_LOCKED); 686 goto relock; 687 } 688 } 689 PF_HASHROW_UNLOCK(ih); 690 } 691 } 692 693 return (len); 694} 695 696static int 697pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 698{ 699 struct mbuf *mp; 700 struct pfsync_state *sa, *sp; 701 int len = sizeof(*sp) * count; 702 int i, offp; 703 704 mp = m_pulldown(m, offset, len, &offp); 705 if (mp == NULL) { 706 V_pfsyncstats.pfsyncs_badlen++; 707 return (-1); 708 } 709 sa = (struct pfsync_state *)(mp->m_data + offp); 710 711 for (i = 0; i < count; i++) { 712 sp = &sa[i]; 713 714 /* Check for invalid values. */ 715 if (sp->timeout >= PFTM_MAX || 716 sp->src.state > PF_TCPS_PROXY_DST || 717 sp->dst.state > PF_TCPS_PROXY_DST || 718 sp->direction > PF_OUT || 719 (sp->af != AF_INET && sp->af != AF_INET6)) { 720 if (V_pf_status.debug >= PF_DEBUG_MISC) 721 printf("%s: invalid value\n", __func__); 722 V_pfsyncstats.pfsyncs_badval++; 723 continue; 724 } 725 726 if (pfsync_state_import(sp, pkt->flags) == ENOMEM) 727 /* Drop out, but process the rest of the actions. */ 728 break; 729 } 730 731 return (len); 732} 733 734static int 735pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 736{ 737 struct pfsync_ins_ack *ia, *iaa; 738 struct pf_state *st; 739 740 struct mbuf *mp; 741 int len = count * sizeof(*ia); 742 int offp, i; 743 744 mp = m_pulldown(m, offset, len, &offp); 745 if (mp == NULL) { 746 V_pfsyncstats.pfsyncs_badlen++; 747 return (-1); 748 } 749 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 750 751 for (i = 0; i < count; i++) { 752 ia = &iaa[i]; 753 754 st = pf_find_state_byid(ia->id, ia->creatorid); 755 if (st == NULL) 756 continue; 757 758 if (st->state_flags & PFSTATE_ACK) { 759 PFSYNC_LOCK(V_pfsyncif); 760 pfsync_undefer_state(st, 0); 761 PFSYNC_UNLOCK(V_pfsyncif); 762 } 763 PF_STATE_UNLOCK(st); 764 } 765 /* 766 * XXX this is not yet implemented, but we know the size of the 767 * message so we can skip it. 768 */ 769 770 return (count * sizeof(struct pfsync_ins_ack)); 771} 772 773static int 774pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src, 775 struct pfsync_state_peer *dst) 776{ 777 int sfail = 0; 778 779 PF_STATE_LOCK_ASSERT(st); 780 781 /* 782 * The state should never go backwards except 783 * for syn-proxy states. Neither should the 784 * sequence window slide backwards. 785 */ 786 if (st->src.state > src->state && 787 (st->src.state < PF_TCPS_PROXY_SRC || 788 src->state >= PF_TCPS_PROXY_SRC)) 789 sfail = 1; 790 else if (SEQ_GT(st->src.seqlo, ntohl(src->seqlo))) 791 sfail = 3; 792 else if (st->dst.state > dst->state) { 793 /* There might still be useful 794 * information about the src state here, 795 * so import that part of the update, 796 * then "fail" so we send the updated 797 * state back to the peer who is missing 798 * our what we know. */ 799 pf_state_peer_ntoh(src, &st->src); 800 /* XXX do anything with timeouts? */ 801 sfail = 7; 802 } else if (st->dst.state >= TCPS_SYN_SENT && 803 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))) 804 sfail = 4; 805 806 return (sfail); 807} 808 809static int 810pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 811{ 812 struct pfsync_softc *sc = V_pfsyncif; 813 struct pfsync_state *sa, *sp; 814 struct pf_state_key *sk; 815 struct pf_state *st; 816 int sfail; 817 818 struct mbuf *mp; 819 int len = count * sizeof(*sp); 820 int offp, i; 821 822 mp = m_pulldown(m, offset, len, &offp); 823 if (mp == NULL) { 824 V_pfsyncstats.pfsyncs_badlen++; 825 return (-1); 826 } 827 sa = (struct pfsync_state *)(mp->m_data + offp); 828 829 for (i = 0; i < count; i++) { 830 sp = &sa[i]; 831 832 /* check for invalid values */ 833 if (sp->timeout >= PFTM_MAX || 834 sp->src.state > PF_TCPS_PROXY_DST || 835 sp->dst.state > PF_TCPS_PROXY_DST) { 836 if (V_pf_status.debug >= PF_DEBUG_MISC) { 837 printf("pfsync_input: PFSYNC_ACT_UPD: " 838 "invalid value\n"); 839 } 840 V_pfsyncstats.pfsyncs_badval++; 841 continue; 842 } 843 844 st = pf_find_state_byid(sp->id, sp->creatorid); 845 if (st == NULL) { 846 /* insert the update */ 847 if (pfsync_state_import(sp, 0)) 848 V_pfsyncstats.pfsyncs_badstate++; 849 continue; 850 } 851 852 if (st->state_flags & PFSTATE_ACK) { 853 PFSYNC_LOCK(sc); 854 pfsync_undefer_state(st, 1); 855 PFSYNC_UNLOCK(sc); 856 } 857 858 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 859 sfail = 0; 860 if (sk->proto == IPPROTO_TCP) 861 sfail = pfsync_upd_tcp(st, &sp->src, &sp->dst); 862 else { 863 /* 864 * Non-TCP protocol state machine always go 865 * forwards 866 */ 867 if (st->src.state > sp->src.state) 868 sfail = 5; 869 else if (st->dst.state > sp->dst.state) 870 sfail = 6; 871 } 872 873 if (sfail) { 874 if (V_pf_status.debug >= PF_DEBUG_MISC) { 875 printf("pfsync: %s stale update (%d)" 876 " id: %016llx creatorid: %08x\n", 877 (sfail < 7 ? "ignoring" : "partial"), 878 sfail, (unsigned long long)be64toh(st->id), 879 ntohl(st->creatorid)); 880 } 881 V_pfsyncstats.pfsyncs_stale++; 882 883 pfsync_update_state(st); 884 PF_STATE_UNLOCK(st); 885 PFSYNC_LOCK(sc); 886 pfsync_push(sc); 887 PFSYNC_UNLOCK(sc); 888 continue; 889 } 890 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 891 pf_state_peer_ntoh(&sp->src, &st->src); 892 pf_state_peer_ntoh(&sp->dst, &st->dst); 893 st->expire = time_uptime; 894 st->timeout = sp->timeout; 895 st->pfsync_time = time_uptime; 896 PF_STATE_UNLOCK(st); 897 } 898 899 return (len); 900} 901 902static int 903pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 904{ 905 struct pfsync_softc *sc = V_pfsyncif; 906 struct pfsync_upd_c *ua, *up; 907 struct pf_state_key *sk; 908 struct pf_state *st; 909 910 int len = count * sizeof(*up); 911 int sfail; 912 913 struct mbuf *mp; 914 int offp, i; 915 916 mp = m_pulldown(m, offset, len, &offp); 917 if (mp == NULL) { 918 V_pfsyncstats.pfsyncs_badlen++; 919 return (-1); 920 } 921 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 922 923 for (i = 0; i < count; i++) { 924 up = &ua[i]; 925 926 /* check for invalid values */ 927 if (up->timeout >= PFTM_MAX || 928 up->src.state > PF_TCPS_PROXY_DST || 929 up->dst.state > PF_TCPS_PROXY_DST) { 930 if (V_pf_status.debug >= PF_DEBUG_MISC) { 931 printf("pfsync_input: " 932 "PFSYNC_ACT_UPD_C: " 933 "invalid value\n"); 934 } 935 V_pfsyncstats.pfsyncs_badval++; 936 continue; 937 } 938 939 st = pf_find_state_byid(up->id, up->creatorid); 940 if (st == NULL) { 941 /* We don't have this state. Ask for it. */ 942 PFSYNC_LOCK(sc); 943 pfsync_request_update(up->creatorid, up->id); 944 PFSYNC_UNLOCK(sc); 945 continue; 946 } 947 948 if (st->state_flags & PFSTATE_ACK) { 949 PFSYNC_LOCK(sc); 950 pfsync_undefer_state(st, 1); 951 PFSYNC_UNLOCK(sc); 952 } 953 954 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 955 sfail = 0; 956 if (sk->proto == IPPROTO_TCP) 957 sfail = pfsync_upd_tcp(st, &up->src, &up->dst); 958 else { 959 /* 960 * Non-TCP protocol state machine always go forwards 961 */ 962 if (st->src.state > up->src.state) 963 sfail = 5; 964 else if (st->dst.state > up->dst.state) 965 sfail = 6; 966 } 967 968 if (sfail) { 969 if (V_pf_status.debug >= PF_DEBUG_MISC) { 970 printf("pfsync: ignoring stale update " 971 "(%d) id: %016llx " 972 "creatorid: %08x\n", sfail, 973 (unsigned long long)be64toh(st->id), 974 ntohl(st->creatorid)); 975 } 976 V_pfsyncstats.pfsyncs_stale++; 977 978 pfsync_update_state(st); 979 PF_STATE_UNLOCK(st); 980 PFSYNC_LOCK(sc); 981 pfsync_push(sc); 982 PFSYNC_UNLOCK(sc); 983 continue; 984 } 985 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 986 pf_state_peer_ntoh(&up->src, &st->src); 987 pf_state_peer_ntoh(&up->dst, &st->dst); 988 st->expire = time_uptime; 989 st->timeout = up->timeout; 990 st->pfsync_time = time_uptime; 991 PF_STATE_UNLOCK(st); 992 } 993 994 return (len); 995} 996 997static int 998pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 999{ 1000 struct pfsync_upd_req *ur, *ura; 1001 struct mbuf *mp; 1002 int len = count * sizeof(*ur); 1003 int i, offp; 1004 1005 struct pf_state *st; 1006 1007 mp = m_pulldown(m, offset, len, &offp); 1008 if (mp == NULL) { 1009 V_pfsyncstats.pfsyncs_badlen++; 1010 return (-1); 1011 } 1012 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1013 1014 for (i = 0; i < count; i++) { 1015 ur = &ura[i]; 1016 1017 if (ur->id == 0 && ur->creatorid == 0) 1018 pfsync_bulk_start(); 1019 else { 1020 st = pf_find_state_byid(ur->id, ur->creatorid); 1021 if (st == NULL) { 1022 V_pfsyncstats.pfsyncs_badstate++; 1023 continue; 1024 } 1025 if (st->state_flags & PFSTATE_NOSYNC) { 1026 PF_STATE_UNLOCK(st); 1027 continue; 1028 } 1029 1030 pfsync_update_state_req(st); 1031 PF_STATE_UNLOCK(st); 1032 } 1033 } 1034 1035 return (len); 1036} 1037 1038static int 1039pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1040{ 1041 struct mbuf *mp; 1042 struct pfsync_state *sa, *sp; 1043 struct pf_state *st; 1044 int len = count * sizeof(*sp); 1045 int offp, i; 1046 1047 mp = m_pulldown(m, offset, len, &offp); 1048 if (mp == NULL) { 1049 V_pfsyncstats.pfsyncs_badlen++; 1050 return (-1); 1051 } 1052 sa = (struct pfsync_state *)(mp->m_data + offp); 1053 1054 for (i = 0; i < count; i++) { 1055 sp = &sa[i]; 1056 1057 st = pf_find_state_byid(sp->id, sp->creatorid); 1058 if (st == NULL) { 1059 V_pfsyncstats.pfsyncs_badstate++; 1060 continue; 1061 } 1062 st->state_flags |= PFSTATE_NOSYNC; 1063 pf_unlink_state(st, PF_ENTER_LOCKED); 1064 } 1065 1066 return (len); 1067} 1068 1069static int 1070pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1071{ 1072 struct mbuf *mp; 1073 struct pfsync_del_c *sa, *sp; 1074 struct pf_state *st; 1075 int len = count * sizeof(*sp); 1076 int offp, i; 1077 1078 mp = m_pulldown(m, offset, len, &offp); 1079 if (mp == NULL) { 1080 V_pfsyncstats.pfsyncs_badlen++; 1081 return (-1); 1082 } 1083 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1084 1085 for (i = 0; i < count; i++) { 1086 sp = &sa[i]; 1087 1088 st = pf_find_state_byid(sp->id, sp->creatorid); 1089 if (st == NULL) { 1090 V_pfsyncstats.pfsyncs_badstate++; 1091 continue; 1092 } 1093 1094 st->state_flags |= PFSTATE_NOSYNC; 1095 pf_unlink_state(st, PF_ENTER_LOCKED); 1096 } 1097 1098 return (len); 1099} 1100 1101static int 1102pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1103{ 1104 struct pfsync_softc *sc = V_pfsyncif; 1105 struct pfsync_bus *bus; 1106 struct mbuf *mp; 1107 int len = count * sizeof(*bus); 1108 int offp; 1109 1110 PFSYNC_BLOCK(sc); 1111 1112 /* If we're not waiting for a bulk update, who cares. */ 1113 if (sc->sc_ureq_sent == 0) { 1114 PFSYNC_BUNLOCK(sc); 1115 return (len); 1116 } 1117 1118 mp = m_pulldown(m, offset, len, &offp); 1119 if (mp == NULL) { 1120 PFSYNC_BUNLOCK(sc); 1121 V_pfsyncstats.pfsyncs_badlen++; 1122 return (-1); 1123 } 1124 bus = (struct pfsync_bus *)(mp->m_data + offp); 1125 1126 switch (bus->status) { 1127 case PFSYNC_BUS_START: 1128 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1129 V_pf_limits[PF_LIMIT_STATES].limit / 1130 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1131 sizeof(struct pfsync_state)), 1132 pfsync_bulk_fail, sc); 1133 if (V_pf_status.debug >= PF_DEBUG_MISC) 1134 printf("pfsync: received bulk update start\n"); 1135 break; 1136 1137 case PFSYNC_BUS_END: 1138 if (time_uptime - ntohl(bus->endtime) >= 1139 sc->sc_ureq_sent) { 1140 /* that's it, we're happy */ 1141 sc->sc_ureq_sent = 0; 1142 sc->sc_bulk_tries = 0; 1143 callout_stop(&sc->sc_bulkfail_tmo); 1144 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1145 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1146 "pfsync bulk done"); 1147 sc->sc_flags |= PFSYNCF_OK; 1148 if (V_pf_status.debug >= PF_DEBUG_MISC) 1149 printf("pfsync: received valid " 1150 "bulk update end\n"); 1151 } else { 1152 if (V_pf_status.debug >= PF_DEBUG_MISC) 1153 printf("pfsync: received invalid " 1154 "bulk update end: bad timestamp\n"); 1155 } 1156 break; 1157 } 1158 PFSYNC_BUNLOCK(sc); 1159 1160 return (len); 1161} 1162 1163static int 1164pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1165{ 1166 int len = count * sizeof(struct pfsync_tdb); 1167 1168#if defined(IPSEC) 1169 struct pfsync_tdb *tp; 1170 struct mbuf *mp; 1171 int offp; 1172 int i; 1173 int s; 1174 1175 mp = m_pulldown(m, offset, len, &offp); 1176 if (mp == NULL) { 1177 V_pfsyncstats.pfsyncs_badlen++; 1178 return (-1); 1179 } 1180 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1181 1182 for (i = 0; i < count; i++) 1183 pfsync_update_net_tdb(&tp[i]); 1184#endif 1185 1186 return (len); 1187} 1188 1189#if defined(IPSEC) 1190/* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1191static void 1192pfsync_update_net_tdb(struct pfsync_tdb *pt) 1193{ 1194 struct tdb *tdb; 1195 int s; 1196 1197 /* check for invalid values */ 1198 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1199 (pt->dst.sa.sa_family != AF_INET && 1200 pt->dst.sa.sa_family != AF_INET6)) 1201 goto bad; 1202 1203 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1204 if (tdb) { 1205 pt->rpl = ntohl(pt->rpl); 1206 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1207 1208 /* Neither replay nor byte counter should ever decrease. */ 1209 if (pt->rpl < tdb->tdb_rpl || 1210 pt->cur_bytes < tdb->tdb_cur_bytes) { 1211 goto bad; 1212 } 1213 1214 tdb->tdb_rpl = pt->rpl; 1215 tdb->tdb_cur_bytes = pt->cur_bytes; 1216 } 1217 return; 1218 1219bad: 1220 if (V_pf_status.debug >= PF_DEBUG_MISC) 1221 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1222 "invalid value\n"); 1223 V_pfsyncstats.pfsyncs_badstate++; 1224 return; 1225} 1226#endif 1227 1228 1229static int 1230pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1231{ 1232 /* check if we are at the right place in the packet */ 1233 if (offset != m->m_pkthdr.len - sizeof(struct pfsync_eof)) 1234 V_pfsyncstats.pfsyncs_badact++; 1235 1236 /* we're done. free and let the caller return */ 1237 m_freem(m); 1238 return (-1); 1239} 1240 1241static int 1242pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1243{ 1244 V_pfsyncstats.pfsyncs_badact++; 1245 1246 m_freem(m); 1247 return (-1); 1248} 1249 1250static int 1251pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 1252 struct route *rt) 1253{ 1254 m_freem(m); 1255 return (0); 1256} 1257 1258/* ARGSUSED */ 1259static int 1260pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1261{ 1262 struct pfsync_softc *sc = ifp->if_softc; 1263 struct ifreq *ifr = (struct ifreq *)data; 1264 struct pfsyncreq pfsyncr; 1265 int error; 1266 1267 switch (cmd) { 1268 case SIOCSIFFLAGS: 1269 PFSYNC_LOCK(sc); 1270 if (ifp->if_flags & IFF_UP) 1271 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1272 else 1273 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1274 PFSYNC_UNLOCK(sc); 1275 break; 1276 case SIOCSIFMTU: 1277 if (!sc->sc_sync_if || 1278 ifr->ifr_mtu <= PFSYNC_MINPKT || 1279 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1280 return (EINVAL); 1281 if (ifr->ifr_mtu < ifp->if_mtu) { 1282 PFSYNC_LOCK(sc); 1283 if (sc->sc_len > PFSYNC_MINPKT) 1284 pfsync_sendout(1); 1285 PFSYNC_UNLOCK(sc); 1286 } 1287 ifp->if_mtu = ifr->ifr_mtu; 1288 break; 1289 case SIOCGETPFSYNC: 1290 bzero(&pfsyncr, sizeof(pfsyncr)); 1291 PFSYNC_LOCK(sc); 1292 if (sc->sc_sync_if) { 1293 strlcpy(pfsyncr.pfsyncr_syncdev, 1294 sc->sc_sync_if->if_xname, IFNAMSIZ); 1295 } 1296 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1297 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1298 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER == 1299 (sc->sc_flags & PFSYNCF_DEFER)); 1300 PFSYNC_UNLOCK(sc); 1301 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))); 1302 1303 case SIOCSETPFSYNC: 1304 { 1305 struct ip_moptions *imo = &sc->sc_imo; 1306 struct ifnet *sifp; 1307 struct ip *ip; 1308 void *mship = NULL; 1309 1310 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1311 return (error); 1312 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 1313 return (error); 1314 1315 if (pfsyncr.pfsyncr_maxupdates > 255) 1316 return (EINVAL); 1317 1318 if (pfsyncr.pfsyncr_syncdev[0] == 0) 1319 sifp = NULL; 1320 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) 1321 return (EINVAL); 1322 1323 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0 && sifp != NULL) 1324 mship = malloc((sizeof(struct in_multi *) * 1325 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO); 1326 1327 PFSYNC_LOCK(sc); 1328 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1329 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); 1330 else 1331 sc->sc_sync_peer.s_addr = 1332 pfsyncr.pfsyncr_syncpeer.s_addr; 1333 1334 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1335 if (pfsyncr.pfsyncr_defer) { 1336 sc->sc_flags |= PFSYNCF_DEFER; 1337 pfsync_defer_ptr = pfsync_defer; 1338 } else { 1339 sc->sc_flags &= ~PFSYNCF_DEFER; 1340 pfsync_defer_ptr = NULL; 1341 } 1342 1343 if (sifp == NULL) { 1344 if (sc->sc_sync_if) 1345 if_rele(sc->sc_sync_if); 1346 sc->sc_sync_if = NULL; 1347 if (imo->imo_membership) 1348 pfsync_multicast_cleanup(sc); 1349 PFSYNC_UNLOCK(sc); 1350 break; 1351 } 1352 1353 if (sc->sc_len > PFSYNC_MINPKT && 1354 (sifp->if_mtu < sc->sc_ifp->if_mtu || 1355 (sc->sc_sync_if != NULL && 1356 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 1357 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 1358 pfsync_sendout(1); 1359 1360 if (imo->imo_membership) 1361 pfsync_multicast_cleanup(sc); 1362 1363 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 1364 error = pfsync_multicast_setup(sc, sifp, mship); 1365 if (error) { 1366 if_rele(sifp); 1367 free(mship, M_PFSYNC); 1368 return (error); 1369 } 1370 } 1371 if (sc->sc_sync_if) 1372 if_rele(sc->sc_sync_if); 1373 sc->sc_sync_if = sifp; 1374 1375 ip = &sc->sc_template; 1376 bzero(ip, sizeof(*ip)); 1377 ip->ip_v = IPVERSION; 1378 ip->ip_hl = sizeof(sc->sc_template) >> 2; 1379 ip->ip_tos = IPTOS_LOWDELAY; 1380 /* len and id are set later. */ 1381 ip->ip_off = IP_DF; 1382 ip->ip_ttl = PFSYNC_DFLTTL; 1383 ip->ip_p = IPPROTO_PFSYNC; 1384 ip->ip_src.s_addr = INADDR_ANY; 1385 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; 1386 1387 /* Request a full state table update. */ 1388 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1389 (*carp_demote_adj_p)(V_pfsync_carp_adj, 1390 "pfsync bulk start"); 1391 sc->sc_flags &= ~PFSYNCF_OK; 1392 if (V_pf_status.debug >= PF_DEBUG_MISC) 1393 printf("pfsync: requesting bulk update\n"); 1394 pfsync_request_update(0, 0); 1395 PFSYNC_UNLOCK(sc); 1396 PFSYNC_BLOCK(sc); 1397 sc->sc_ureq_sent = time_uptime; 1398 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, 1399 sc); 1400 PFSYNC_BUNLOCK(sc); 1401 1402 break; 1403 } 1404 default: 1405 return (ENOTTY); 1406 } 1407 1408 return (0); 1409} 1410 1411static void 1412pfsync_out_state(struct pf_state *st, void *buf) 1413{ 1414 struct pfsync_state *sp = buf; 1415 1416 pfsync_state_export(sp, st); 1417} 1418 1419static void 1420pfsync_out_iack(struct pf_state *st, void *buf) 1421{ 1422 struct pfsync_ins_ack *iack = buf; 1423 1424 iack->id = st->id; 1425 iack->creatorid = st->creatorid; 1426} 1427 1428static void 1429pfsync_out_upd_c(struct pf_state *st, void *buf) 1430{ 1431 struct pfsync_upd_c *up = buf; 1432 1433 bzero(up, sizeof(*up)); 1434 up->id = st->id; 1435 pf_state_peer_hton(&st->src, &up->src); 1436 pf_state_peer_hton(&st->dst, &up->dst); 1437 up->creatorid = st->creatorid; 1438 up->timeout = st->timeout; 1439} 1440 1441static void 1442pfsync_out_del(struct pf_state *st, void *buf) 1443{ 1444 struct pfsync_del_c *dp = buf; 1445 1446 dp->id = st->id; 1447 dp->creatorid = st->creatorid; 1448 st->state_flags |= PFSTATE_NOSYNC; 1449} 1450 1451static void 1452pfsync_drop(struct pfsync_softc *sc) 1453{ 1454 struct pf_state *st, *next; 1455 struct pfsync_upd_req_item *ur; 1456 int q; 1457 1458 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1459 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1460 continue; 1461 1462 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) { 1463 KASSERT(st->sync_state == q, 1464 ("%s: st->sync_state == q", 1465 __func__)); 1466 st->sync_state = PFSYNC_S_NONE; 1467 pf_release_state(st); 1468 } 1469 TAILQ_INIT(&sc->sc_qs[q]); 1470 } 1471 1472 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1473 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1474 free(ur, M_PFSYNC); 1475 } 1476 1477 sc->sc_plus = NULL; 1478 sc->sc_len = PFSYNC_MINPKT; 1479} 1480 1481static void 1482pfsync_sendout(int schedswi) 1483{ 1484 struct pfsync_softc *sc = V_pfsyncif; 1485 struct ifnet *ifp = sc->sc_ifp; 1486 struct mbuf *m; 1487 struct ip *ip; 1488 struct pfsync_header *ph; 1489 struct pfsync_subheader *subh; 1490 struct pf_state *st; 1491 struct pfsync_upd_req_item *ur; 1492 int offset; 1493 int q, count = 0; 1494 1495 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1496 KASSERT(sc->sc_len > PFSYNC_MINPKT, 1497 ("%s: sc_len %zu", __func__, sc->sc_len)); 1498 PFSYNC_LOCK_ASSERT(sc); 1499 1500 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1501 pfsync_drop(sc); 1502 return; 1503 } 1504 1505 m = m_get2(M_NOWAIT, MT_DATA, M_PKTHDR, max_linkhdr + sc->sc_len); 1506 if (m == NULL) { 1507 sc->sc_ifp->if_oerrors++; 1508 V_pfsyncstats.pfsyncs_onomem++; 1509 return; 1510 } 1511 m->m_data += max_linkhdr; 1512 m->m_len = m->m_pkthdr.len = sc->sc_len; 1513 1514 /* build the ip header */ 1515 ip = (struct ip *)m->m_data; 1516 bcopy(&sc->sc_template, ip, sizeof(*ip)); 1517 offset = sizeof(*ip); 1518 1519 ip->ip_len = m->m_pkthdr.len; 1520 ip->ip_id = htons(ip_randomid()); 1521 1522 /* build the pfsync header */ 1523 ph = (struct pfsync_header *)(m->m_data + offset); 1524 bzero(ph, sizeof(*ph)); 1525 offset += sizeof(*ph); 1526 1527 ph->version = PFSYNC_VERSION; 1528 ph->len = htons(sc->sc_len - sizeof(*ip)); 1529 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1530 1531 /* walk the queues */ 1532 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1533 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1534 continue; 1535 1536 subh = (struct pfsync_subheader *)(m->m_data + offset); 1537 offset += sizeof(*subh); 1538 1539 count = 0; 1540 TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) { 1541 KASSERT(st->sync_state == q, 1542 ("%s: st->sync_state == q", 1543 __func__)); 1544 /* 1545 * XXXGL: some of write methods do unlocked reads 1546 * of state data :( 1547 */ 1548 pfsync_qs[q].write(st, m->m_data + offset); 1549 offset += pfsync_qs[q].len; 1550 st->sync_state = PFSYNC_S_NONE; 1551 pf_release_state(st); 1552 count++; 1553 } 1554 TAILQ_INIT(&sc->sc_qs[q]); 1555 1556 bzero(subh, sizeof(*subh)); 1557 subh->action = pfsync_qs[q].action; 1558 subh->count = htons(count); 1559 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1560 } 1561 1562 if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) { 1563 subh = (struct pfsync_subheader *)(m->m_data + offset); 1564 offset += sizeof(*subh); 1565 1566 count = 0; 1567 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { 1568 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); 1569 1570 bcopy(&ur->ur_msg, m->m_data + offset, 1571 sizeof(ur->ur_msg)); 1572 offset += sizeof(ur->ur_msg); 1573 free(ur, M_PFSYNC); 1574 count++; 1575 } 1576 1577 bzero(subh, sizeof(*subh)); 1578 subh->action = PFSYNC_ACT_UPD_REQ; 1579 subh->count = htons(count); 1580 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1581 } 1582 1583 /* has someone built a custom region for us to add? */ 1584 if (sc->sc_plus != NULL) { 1585 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen); 1586 offset += sc->sc_pluslen; 1587 1588 sc->sc_plus = NULL; 1589 } 1590 1591 subh = (struct pfsync_subheader *)(m->m_data + offset); 1592 offset += sizeof(*subh); 1593 1594 bzero(subh, sizeof(*subh)); 1595 subh->action = PFSYNC_ACT_EOF; 1596 subh->count = htons(1); 1597 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1598 1599 /* XXX write checksum in EOF here */ 1600 1601 /* we're done, let's put it on the wire */ 1602 if (ifp->if_bpf) { 1603 m->m_data += sizeof(*ip); 1604 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip); 1605 BPF_MTAP(ifp, m); 1606 m->m_data -= sizeof(*ip); 1607 m->m_len = m->m_pkthdr.len = sc->sc_len; 1608 } 1609 1610 if (sc->sc_sync_if == NULL) { 1611 sc->sc_len = PFSYNC_MINPKT; 1612 m_freem(m); 1613 return; 1614 } 1615 1616 sc->sc_ifp->if_opackets++; 1617 sc->sc_ifp->if_obytes += m->m_pkthdr.len; 1618 sc->sc_len = PFSYNC_MINPKT; 1619 1620 if (!_IF_QFULL(&sc->sc_ifp->if_snd)) 1621 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1622 else { 1623 m_freem(m); 1624 sc->sc_ifp->if_snd.ifq_drops++; 1625 } 1626 if (schedswi) 1627 swi_sched(V_pfsync_swi_cookie, 0); 1628} 1629 1630static void 1631pfsync_insert_state(struct pf_state *st) 1632{ 1633 struct pfsync_softc *sc = V_pfsyncif; 1634 1635 if (st->state_flags & PFSTATE_NOSYNC) 1636 return; 1637 1638 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1639 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1640 st->state_flags |= PFSTATE_NOSYNC; 1641 return; 1642 } 1643 1644 KASSERT(st->sync_state == PFSYNC_S_NONE, 1645 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1646 1647 PFSYNC_LOCK(sc); 1648 if (sc->sc_len == PFSYNC_MINPKT) 1649 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1650 1651 pfsync_q_ins(st, PFSYNC_S_INS); 1652 PFSYNC_UNLOCK(sc); 1653 1654 st->sync_updates = 0; 1655} 1656 1657static int 1658pfsync_defer(struct pf_state *st, struct mbuf *m) 1659{ 1660 struct pfsync_softc *sc = V_pfsyncif; 1661 struct pfsync_deferral *pd; 1662 1663 if (m->m_flags & (M_BCAST|M_MCAST)) 1664 return (0); 1665 1666 PFSYNC_LOCK(sc); 1667 1668 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) || 1669 !(sc->sc_flags & PFSYNCF_DEFER)) { 1670 PFSYNC_UNLOCK(sc); 1671 return (0); 1672 } 1673 1674 if (sc->sc_deferred >= 128) 1675 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0); 1676 1677 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1678 if (pd == NULL) 1679 return (0); 1680 sc->sc_deferred++; 1681 1682 m->m_flags |= M_SKIP_FIREWALL; 1683 st->state_flags |= PFSTATE_ACK; 1684 1685 pd->pd_sc = sc; 1686 pd->pd_refs = 0; 1687 pd->pd_st = st; 1688 pf_ref_state(st); 1689 pd->pd_m = m; 1690 1691 TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry); 1692 callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1693 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd); 1694 1695 pfsync_push(sc); 1696 1697 return (1); 1698} 1699 1700static void 1701pfsync_undefer(struct pfsync_deferral *pd, int drop) 1702{ 1703 struct pfsync_softc *sc = pd->pd_sc; 1704 struct mbuf *m = pd->pd_m; 1705 struct pf_state *st = pd->pd_st; 1706 1707 PFSYNC_LOCK_ASSERT(sc); 1708 1709 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1710 sc->sc_deferred--; 1711 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1712 free(pd, M_PFSYNC); 1713 pf_release_state(st); 1714 1715 if (drop) 1716 m_freem(m); 1717 else { 1718 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); 1719 pfsync_push(sc); 1720 } 1721} 1722 1723static void 1724pfsync_defer_tmo(void *arg) 1725{ 1726 struct pfsync_deferral *pd = arg; 1727 struct pfsync_softc *sc = pd->pd_sc; 1728 struct mbuf *m = pd->pd_m; 1729 struct pf_state *st = pd->pd_st; 1730 1731 PFSYNC_LOCK_ASSERT(sc); 1732 1733 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); 1734 1735 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); 1736 sc->sc_deferred--; 1737 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1738 if (pd->pd_refs == 0) 1739 free(pd, M_PFSYNC); 1740 PFSYNC_UNLOCK(sc); 1741 1742 ip_output(m, NULL, NULL, 0, NULL, NULL); 1743 1744 pf_release_state(st); 1745 1746 CURVNET_RESTORE(); 1747} 1748 1749static void 1750pfsync_undefer_state(struct pf_state *st, int drop) 1751{ 1752 struct pfsync_softc *sc = V_pfsyncif; 1753 struct pfsync_deferral *pd; 1754 1755 PFSYNC_LOCK_ASSERT(sc); 1756 1757 TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) { 1758 if (pd->pd_st == st) { 1759 if (callout_stop(&pd->pd_tmo)) 1760 pfsync_undefer(pd, drop); 1761 return; 1762 } 1763 } 1764 1765 panic("%s: unable to find deferred state", __func__); 1766} 1767 1768static void 1769pfsync_update_state(struct pf_state *st) 1770{ 1771 struct pfsync_softc *sc = V_pfsyncif; 1772 int sync = 0; 1773 1774 PF_STATE_LOCK_ASSERT(st); 1775 PFSYNC_LOCK(sc); 1776 1777 if (st->state_flags & PFSTATE_ACK) 1778 pfsync_undefer_state(st, 0); 1779 if (st->state_flags & PFSTATE_NOSYNC) { 1780 if (st->sync_state != PFSYNC_S_NONE) 1781 pfsync_q_del(st); 1782 PFSYNC_UNLOCK(sc); 1783 return; 1784 } 1785 1786 if (sc->sc_len == PFSYNC_MINPKT) 1787 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1788 1789 switch (st->sync_state) { 1790 case PFSYNC_S_UPD_C: 1791 case PFSYNC_S_UPD: 1792 case PFSYNC_S_INS: 1793 /* we're already handling it */ 1794 1795 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 1796 st->sync_updates++; 1797 if (st->sync_updates >= sc->sc_maxupdates) 1798 sync = 1; 1799 } 1800 break; 1801 1802 case PFSYNC_S_IACK: 1803 pfsync_q_del(st); 1804 case PFSYNC_S_NONE: 1805 pfsync_q_ins(st, PFSYNC_S_UPD_C); 1806 st->sync_updates = 0; 1807 break; 1808 1809 default: 1810 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1811 } 1812 1813 if (sync || (time_uptime - st->pfsync_time) < 2) 1814 pfsync_push(sc); 1815 1816 PFSYNC_UNLOCK(sc); 1817} 1818 1819static void 1820pfsync_request_update(u_int32_t creatorid, u_int64_t id) 1821{ 1822 struct pfsync_softc *sc = V_pfsyncif; 1823 struct pfsync_upd_req_item *item; 1824 size_t nlen = sizeof(struct pfsync_upd_req); 1825 1826 PFSYNC_LOCK_ASSERT(sc); 1827 1828 /* 1829 * This code does nothing to prevent multiple update requests for the 1830 * same state being generated. 1831 */ 1832 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 1833 if (item == NULL) 1834 return; /* XXX stats */ 1835 1836 item->ur_msg.id = id; 1837 item->ur_msg.creatorid = creatorid; 1838 1839 if (TAILQ_EMPTY(&sc->sc_upd_req_list)) 1840 nlen += sizeof(struct pfsync_subheader); 1841 1842 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1843 pfsync_sendout(1); 1844 1845 nlen = sizeof(struct pfsync_subheader) + 1846 sizeof(struct pfsync_upd_req); 1847 } 1848 1849 TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry); 1850 sc->sc_len += nlen; 1851 1852 pfsync_push(sc); 1853} 1854 1855static void 1856pfsync_update_state_req(struct pf_state *st) 1857{ 1858 struct pfsync_softc *sc = V_pfsyncif; 1859 1860 PF_STATE_LOCK_ASSERT(st); 1861 PFSYNC_LOCK(sc); 1862 1863 if (st->state_flags & PFSTATE_NOSYNC) { 1864 if (st->sync_state != PFSYNC_S_NONE) 1865 pfsync_q_del(st); 1866 PFSYNC_UNLOCK(sc); 1867 return; 1868 } 1869 1870 switch (st->sync_state) { 1871 case PFSYNC_S_UPD_C: 1872 case PFSYNC_S_IACK: 1873 pfsync_q_del(st); 1874 case PFSYNC_S_NONE: 1875 pfsync_q_ins(st, PFSYNC_S_UPD); 1876 pfsync_push(sc); 1877 break; 1878 1879 case PFSYNC_S_INS: 1880 case PFSYNC_S_UPD: 1881 case PFSYNC_S_DEL: 1882 /* we're already handling it */ 1883 break; 1884 1885 default: 1886 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1887 } 1888 1889 PFSYNC_UNLOCK(sc); 1890} 1891 1892static void 1893pfsync_delete_state(struct pf_state *st) 1894{ 1895 struct pfsync_softc *sc = V_pfsyncif; 1896 1897 PFSYNC_LOCK(sc); 1898 if (st->state_flags & PFSTATE_ACK) 1899 pfsync_undefer_state(st, 1); 1900 if (st->state_flags & PFSTATE_NOSYNC) { 1901 if (st->sync_state != PFSYNC_S_NONE) 1902 pfsync_q_del(st); 1903 PFSYNC_UNLOCK(sc); 1904 return; 1905 } 1906 1907 if (sc->sc_len == PFSYNC_MINPKT) 1908 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); 1909 1910 switch (st->sync_state) { 1911 case PFSYNC_S_INS: 1912 /* We never got to tell the world so just forget about it. */ 1913 pfsync_q_del(st); 1914 break; 1915 1916 case PFSYNC_S_UPD_C: 1917 case PFSYNC_S_UPD: 1918 case PFSYNC_S_IACK: 1919 pfsync_q_del(st); 1920 /* FALLTHROUGH to putting it on the del list */ 1921 1922 case PFSYNC_S_NONE: 1923 pfsync_q_ins(st, PFSYNC_S_DEL); 1924 break; 1925 1926 default: 1927 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1928 } 1929 PFSYNC_UNLOCK(sc); 1930} 1931 1932static void 1933pfsync_clear_states(u_int32_t creatorid, const char *ifname) 1934{ 1935 struct pfsync_softc *sc = V_pfsyncif; 1936 struct { 1937 struct pfsync_subheader subh; 1938 struct pfsync_clr clr; 1939 } __packed r; 1940 1941 bzero(&r, sizeof(r)); 1942 1943 r.subh.action = PFSYNC_ACT_CLR; 1944 r.subh.count = htons(1); 1945 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 1946 1947 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 1948 r.clr.creatorid = creatorid; 1949 1950 PFSYNC_LOCK(sc); 1951 pfsync_send_plus(&r, sizeof(r)); 1952 PFSYNC_UNLOCK(sc); 1953} 1954 1955static void 1956pfsync_q_ins(struct pf_state *st, int q) 1957{ 1958 struct pfsync_softc *sc = V_pfsyncif; 1959 size_t nlen = pfsync_qs[q].len; 1960 1961 PFSYNC_LOCK_ASSERT(sc); 1962 1963 KASSERT(st->sync_state == PFSYNC_S_NONE, 1964 ("%s: st->sync_state == PFSYNC_S_NONE", __func__)); 1965 KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 1966 sc->sc_len)); 1967 1968 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1969 nlen += sizeof(struct pfsync_subheader); 1970 1971 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { 1972 pfsync_sendout(1); 1973 1974 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 1975 } 1976 1977 sc->sc_len += nlen; 1978 TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list); 1979 st->sync_state = q; 1980 pf_ref_state(st); 1981} 1982 1983static void 1984pfsync_q_del(struct pf_state *st) 1985{ 1986 struct pfsync_softc *sc = V_pfsyncif; 1987 int q = st->sync_state; 1988 1989 PFSYNC_LOCK_ASSERT(sc); 1990 KASSERT(st->sync_state != PFSYNC_S_NONE, 1991 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 1992 1993 sc->sc_len -= pfsync_qs[q].len; 1994 TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list); 1995 st->sync_state = PFSYNC_S_NONE; 1996 pf_release_state(st); 1997 1998 if (TAILQ_EMPTY(&sc->sc_qs[q])) 1999 sc->sc_len -= sizeof(struct pfsync_subheader); 2000} 2001 2002static void 2003pfsync_bulk_start(void) 2004{ 2005 struct pfsync_softc *sc = V_pfsyncif; 2006 2007 if (V_pf_status.debug >= PF_DEBUG_MISC) 2008 printf("pfsync: received bulk update request\n"); 2009 2010 PFSYNC_BLOCK(sc); 2011 2012 sc->sc_ureq_received = time_uptime; 2013 sc->sc_bulk_hashid = 0; 2014 sc->sc_bulk_stateid = 0; 2015 pfsync_bulk_status(PFSYNC_BUS_START); 2016 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2017 PFSYNC_BUNLOCK(sc); 2018} 2019 2020static void 2021pfsync_bulk_update(void *arg) 2022{ 2023 struct pfsync_softc *sc = arg; 2024 struct pf_state *s; 2025 int i, sent = 0; 2026 2027 PFSYNC_BLOCK_ASSERT(sc); 2028 CURVNET_SET(sc->sc_ifp->if_vnet); 2029 2030 /* 2031 * Start with last state from previous invocation. 2032 * It may had gone, in this case start from the 2033 * hash slot. 2034 */ 2035 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2036 2037 if (s != NULL) 2038 i = PF_IDHASH(s); 2039 else 2040 i = sc->sc_bulk_hashid; 2041 2042 for (; i <= V_pf_hashmask; i++) { 2043 struct pf_idhash *ih = &V_pf_idhash[i]; 2044 2045 if (s != NULL) 2046 PF_HASHROW_ASSERT(ih); 2047 else { 2048 PF_HASHROW_LOCK(ih); 2049 s = LIST_FIRST(&ih->states); 2050 } 2051 2052 for (; s; s = LIST_NEXT(s, entry)) { 2053 2054 if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) < 2055 sizeof(struct pfsync_state)) { 2056 /* We've filled a packet. */ 2057 sc->sc_bulk_hashid = i; 2058 sc->sc_bulk_stateid = s->id; 2059 sc->sc_bulk_creatorid = s->creatorid; 2060 PF_HASHROW_UNLOCK(ih); 2061 callout_reset(&sc->sc_bulk_tmo, 1, 2062 pfsync_bulk_update, sc); 2063 goto full; 2064 } 2065 2066 if (s->sync_state == PFSYNC_S_NONE && 2067 s->timeout < PFTM_MAX && 2068 s->pfsync_time <= sc->sc_ureq_received) { 2069 PFSYNC_LOCK(sc); 2070 pfsync_update_state_req(s); 2071 PFSYNC_UNLOCK(sc); 2072 sent++; 2073 } 2074 } 2075 PF_HASHROW_UNLOCK(ih); 2076 } 2077 2078 /* We're done. */ 2079 pfsync_bulk_status(PFSYNC_BUS_END); 2080 2081full: 2082 CURVNET_RESTORE(); 2083} 2084 2085static void 2086pfsync_bulk_status(u_int8_t status) 2087{ 2088 struct { 2089 struct pfsync_subheader subh; 2090 struct pfsync_bus bus; 2091 } __packed r; 2092 2093 struct pfsync_softc *sc = V_pfsyncif; 2094 2095 bzero(&r, sizeof(r)); 2096 2097 r.subh.action = PFSYNC_ACT_BUS; 2098 r.subh.count = htons(1); 2099 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2100 2101 r.bus.creatorid = V_pf_status.hostid; 2102 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2103 r.bus.status = status; 2104 2105 PFSYNC_LOCK(sc); 2106 pfsync_send_plus(&r, sizeof(r)); 2107 PFSYNC_UNLOCK(sc); 2108} 2109 2110static void 2111pfsync_bulk_fail(void *arg) 2112{ 2113 struct pfsync_softc *sc = arg; 2114 2115 CURVNET_SET(sc->sc_ifp->if_vnet); 2116 2117 PFSYNC_BLOCK_ASSERT(sc); 2118 2119 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2120 /* Try again */ 2121 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2122 pfsync_bulk_fail, V_pfsyncif); 2123 PFSYNC_LOCK(sc); 2124 pfsync_request_update(0, 0); 2125 PFSYNC_UNLOCK(sc); 2126 } else { 2127 /* Pretend like the transfer was ok. */ 2128 sc->sc_ureq_sent = 0; 2129 sc->sc_bulk_tries = 0; 2130 PFSYNC_LOCK(sc); 2131 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2132 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2133 "pfsync bulk fail"); 2134 sc->sc_flags |= PFSYNCF_OK; 2135 PFSYNC_UNLOCK(sc); 2136 if (V_pf_status.debug >= PF_DEBUG_MISC) 2137 printf("pfsync: failed to receive bulk update\n"); 2138 } 2139 2140 CURVNET_RESTORE(); 2141} 2142 2143static void 2144pfsync_send_plus(void *plus, size_t pluslen) 2145{ 2146 struct pfsync_softc *sc = V_pfsyncif; 2147 2148 PFSYNC_LOCK_ASSERT(sc); 2149 2150 if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu) 2151 pfsync_sendout(1); 2152 2153 sc->sc_plus = plus; 2154 sc->sc_len += (sc->sc_pluslen = pluslen); 2155 2156 pfsync_sendout(1); 2157} 2158 2159static void 2160pfsync_timeout(void *arg) 2161{ 2162 struct pfsync_softc *sc = arg; 2163 2164 CURVNET_SET(sc->sc_ifp->if_vnet); 2165 PFSYNC_LOCK(sc); 2166 pfsync_push(sc); 2167 PFSYNC_UNLOCK(sc); 2168 CURVNET_RESTORE(); 2169} 2170 2171static void 2172pfsync_push(struct pfsync_softc *sc) 2173{ 2174 2175 PFSYNC_LOCK_ASSERT(sc); 2176 2177 sc->sc_flags |= PFSYNCF_PUSH; 2178 swi_sched(V_pfsync_swi_cookie, 0); 2179} 2180 2181static void 2182pfsyncintr(void *arg) 2183{ 2184 struct pfsync_softc *sc = arg; 2185 struct mbuf *m, *n; 2186 2187 CURVNET_SET(sc->sc_ifp->if_vnet); 2188 2189 PFSYNC_LOCK(sc); 2190 if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) { 2191 pfsync_sendout(0); 2192 sc->sc_flags &= ~PFSYNCF_PUSH; 2193 } 2194 _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m); 2195 PFSYNC_UNLOCK(sc); 2196 2197 for (; m != NULL; m = n) { 2198 2199 n = m->m_nextpkt; 2200 m->m_nextpkt = NULL; 2201 2202 /* 2203 * We distinguish between a deferral packet and our 2204 * own pfsync packet based on M_SKIP_FIREWALL 2205 * flag. This is XXX. 2206 */ 2207 if (m->m_flags & M_SKIP_FIREWALL) 2208 ip_output(m, NULL, NULL, 0, NULL, NULL); 2209 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 2210 NULL) == 0) 2211 V_pfsyncstats.pfsyncs_opackets++; 2212 else 2213 V_pfsyncstats.pfsyncs_oerrors++; 2214 } 2215 CURVNET_RESTORE(); 2216} 2217 2218static int 2219pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship) 2220{ 2221 struct ip_moptions *imo = &sc->sc_imo; 2222 int error; 2223 2224 if (!(ifp->if_flags & IFF_MULTICAST)) 2225 return (EADDRNOTAVAIL); 2226 2227 imo->imo_membership = (struct in_multi **)mship; 2228 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 2229 imo->imo_multicast_vif = -1; 2230 2231 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, 2232 &imo->imo_membership[0])) != 0) { 2233 imo->imo_membership = NULL; 2234 return (error); 2235 } 2236 imo->imo_num_memberships++; 2237 imo->imo_multicast_ifp = ifp; 2238 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2239 imo->imo_multicast_loop = 0; 2240 2241 return (0); 2242} 2243 2244static void 2245pfsync_multicast_cleanup(struct pfsync_softc *sc) 2246{ 2247 struct ip_moptions *imo = &sc->sc_imo; 2248 2249 in_leavegroup(imo->imo_membership[0], NULL); 2250 free(imo->imo_membership, M_PFSYNC); 2251 imo->imo_membership = NULL; 2252 imo->imo_multicast_ifp = NULL; 2253} 2254 2255#ifdef INET 2256extern struct domain inetdomain; 2257static struct protosw in_pfsync_protosw = { 2258 .pr_type = SOCK_RAW, 2259 .pr_domain = &inetdomain, 2260 .pr_protocol = IPPROTO_PFSYNC, 2261 .pr_flags = PR_ATOMIC|PR_ADDR, 2262 .pr_input = pfsync_input, 2263 .pr_output = (pr_output_t *)rip_output, 2264 .pr_ctloutput = rip_ctloutput, 2265 .pr_usrreqs = &rip_usrreqs 2266}; 2267#endif 2268 2269static int 2270pfsync_init() 2271{ 2272 VNET_ITERATOR_DECL(vnet_iter); 2273 int error = 0; 2274 2275 VNET_LIST_RLOCK(); 2276 VNET_FOREACH(vnet_iter) { 2277 CURVNET_SET(vnet_iter); 2278 V_pfsync_cloner = pfsync_cloner; 2279 V_pfsync_cloner_data = pfsync_cloner_data; 2280 V_pfsync_cloner.ifc_data = &V_pfsync_cloner_data; 2281 if_clone_attach(&V_pfsync_cloner); 2282 error = swi_add(NULL, "pfsync", pfsyncintr, V_pfsyncif, 2283 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2284 CURVNET_RESTORE(); 2285 if (error) 2286 goto fail_locked; 2287 } 2288 VNET_LIST_RUNLOCK(); 2289#ifdef INET 2290 error = pf_proto_register(PF_INET, &in_pfsync_protosw); 2291 if (error) 2292 goto fail; 2293 error = ipproto_register(IPPROTO_PFSYNC); 2294 if (error) { 2295 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2296 goto fail; 2297 } 2298#endif 2299 PF_RULES_WLOCK(); 2300 pfsync_state_import_ptr = pfsync_state_import; 2301 pfsync_insert_state_ptr = pfsync_insert_state; 2302 pfsync_update_state_ptr = pfsync_update_state; 2303 pfsync_delete_state_ptr = pfsync_delete_state; 2304 pfsync_clear_states_ptr = pfsync_clear_states; 2305 pfsync_defer_ptr = pfsync_defer; 2306 PF_RULES_WUNLOCK(); 2307 2308 return (0); 2309 2310fail: 2311 VNET_LIST_RLOCK(); 2312fail_locked: 2313 VNET_FOREACH(vnet_iter) { 2314 CURVNET_SET(vnet_iter); 2315 if (V_pfsync_swi_cookie) { 2316 swi_remove(V_pfsync_swi_cookie); 2317 if_clone_detach(&V_pfsync_cloner); 2318 } 2319 CURVNET_RESTORE(); 2320 } 2321 VNET_LIST_RUNLOCK(); 2322 2323 return (error); 2324} 2325 2326static void 2327pfsync_uninit() 2328{ 2329 VNET_ITERATOR_DECL(vnet_iter); 2330 2331 PF_RULES_WLOCK(); 2332 pfsync_state_import_ptr = NULL; 2333 pfsync_insert_state_ptr = NULL; 2334 pfsync_update_state_ptr = NULL; 2335 pfsync_delete_state_ptr = NULL; 2336 pfsync_clear_states_ptr = NULL; 2337 pfsync_defer_ptr = NULL; 2338 PF_RULES_WUNLOCK(); 2339 2340 ipproto_unregister(IPPROTO_PFSYNC); 2341 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2342 VNET_LIST_RLOCK(); 2343 VNET_FOREACH(vnet_iter) { 2344 CURVNET_SET(vnet_iter); 2345 if_clone_detach(&V_pfsync_cloner); 2346 swi_remove(V_pfsync_swi_cookie); 2347 CURVNET_RESTORE(); 2348 } 2349 VNET_LIST_RUNLOCK(); 2350} 2351 2352static int 2353pfsync_modevent(module_t mod, int type, void *data) 2354{ 2355 int error = 0; 2356 2357 switch (type) { 2358 case MOD_LOAD: 2359 error = pfsync_init(); 2360 break; 2361 case MOD_QUIESCE: 2362 /* 2363 * Module should not be unloaded due to race conditions. 2364 */ 2365 error = EBUSY; 2366 break; 2367 case MOD_UNLOAD: 2368 pfsync_uninit(); 2369 break; 2370 default: 2371 error = EINVAL; 2372 break; 2373 } 2374 2375 return (error); 2376} 2377 2378static moduledata_t pfsync_mod = { 2379 "pfsync", 2380 pfsync_modevent, 2381 0 2382}; 2383 2384#define PFSYNC_MODVER 1 2385 2386DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); 2387MODULE_VERSION(pfsync, PFSYNC_MODVER); 2388MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); 2389