1 /*- 2 * SPDX-License-Identifier: (BSD-2-Clause AND ISC) 3 * 4 * Copyright (c) 2002 Michael Shalayeff 5 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 21 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 25 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 26 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 27 * THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 /*- 31 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org> 32 * 33 * Permission to use, copy, modify, and distribute this software for any 34 * purpose with or without fee is hereby granted, provided that the above 35 * copyright notice and this permission notice appear in all copies. 36 * 37 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 38 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 39 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 40 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 41 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 42 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 43 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 44 */ 45 46 /* 47 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $ 48 * 49 * Revisions picked from OpenBSD after revision 1.110 import: 50 * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input() 51 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates 52 * 1.120, 1.175 - use monotonic time_uptime 53 * 1.122 - reduce number of updates for non-TCP sessions 54 * 1.125, 1.127 - rewrite merge or stale processing 55 * 1.128 - cleanups 56 * 1.146 - bzero() mbuf before sparsely filling it with data 57 * 1.170 - SIOCSIFMTU checks 58 * 1.126, 1.142 - deferred packets processing 59 * 1.173 - correct expire time processing 60 */ 61 62 #include <sys/cdefs.h> 63 __FBSDID("$FreeBSD$"); 64 65 #include "opt_inet.h" 66 #include "opt_inet6.h" 67 #include "opt_pf.h" 68 69 #include <sys/param.h> 70 #include <sys/bus.h> 71 #include <sys/endian.h> 72 #include <sys/interrupt.h> 73 #include <sys/kernel.h> 74 #include <sys/lock.h> 75 #include <sys/mbuf.h> 76 #include <sys/module.h> 77 #include <sys/mutex.h> 78 #include <sys/nv.h> 79 #include <sys/priv.h> 80 #include <sys/smp.h> 81 #include <sys/socket.h> 82 #include <sys/sockio.h> 83 #include <sys/sysctl.h> 84 #include <sys/syslog.h> 85 86 #include <net/bpf.h> 87 #include <net/if.h> 88 #include <net/if_var.h> 89 #include <net/if_clone.h> 90 #include <net/if_private.h> 91 #include <net/if_types.h> 92 #include <net/vnet.h> 93 #include <net/pfvar.h> 94 #include <net/if_pfsync.h> 95 96 #include <netinet/if_ether.h> 97 #include <netinet/in.h> 98 #include <netinet/in_var.h> 99 #include <netinet/ip.h> 100 #include <netinet/ip_carp.h> 101 #include <netinet/ip_var.h> 102 #include <netinet/tcp.h> 103 #include <netinet/tcp_fsm.h> 104 #include <netinet/tcp_seq.h> 105 106 #include <netinet/ip6.h> 107 #include <netinet6/ip6_var.h> 108 109 #include <netpfil/pf/pfsync_nv.h> 110 111 struct pfsync_bucket; 112 struct pfsync_softc; 113 114 union inet_template { 115 struct ip ipv4; 116 }; 117 118 #define PFSYNC_MINPKT ( \ 119 sizeof(union inet_template) + \ 120 sizeof(struct pfsync_header) + \ 121 sizeof(struct pfsync_subheader) ) 122 123 static int pfsync_upd_tcp(struct pf_kstate *, struct pfsync_state_peer *, 124 struct pfsync_state_peer *); 125 static int pfsync_in_clr(struct mbuf *, int, int, int, int); 126 static int pfsync_in_ins(struct mbuf *, int, int, int, int); 127 static int pfsync_in_iack(struct mbuf *, int, int, int, int); 128 static int pfsync_in_upd(struct mbuf *, int, int, int, int); 129 static int pfsync_in_upd_c(struct mbuf *, int, int, int, int); 130 static int pfsync_in_ureq(struct mbuf *, int, int, int, int); 131 static int pfsync_in_del_c(struct mbuf *, int, int, int, int); 132 static int pfsync_in_bus(struct mbuf *, int, int, int, int); 133 static int pfsync_in_tdb(struct mbuf *, int, int, int, int); 134 static int pfsync_in_eof(struct mbuf *, int, int, int, int); 135 static int pfsync_in_error(struct mbuf *, int, int, int, int); 136 137 static int (*pfsync_acts[])(struct mbuf *, int, int, int, int) = { 138 pfsync_in_clr, /* PFSYNC_ACT_CLR */ 139 pfsync_in_ins, /* PFSYNC_ACT_INS_1301 */ 140 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ 141 pfsync_in_upd, /* PFSYNC_ACT_UPD_1301 */ 142 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ 143 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ 144 pfsync_in_error, /* PFSYNC_ACT_DEL */ 145 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ 146 pfsync_in_error, /* PFSYNC_ACT_INS_F */ 147 pfsync_in_error, /* PFSYNC_ACT_DEL_F */ 148 pfsync_in_bus, /* PFSYNC_ACT_BUS */ 149 pfsync_in_tdb, /* PFSYNC_ACT_TDB */ 150 pfsync_in_eof, /* PFSYNC_ACT_EOF */ 151 pfsync_in_ins, /* PFSYNC_ACT_INS_1400 */ 152 pfsync_in_upd, /* PFSYNC_ACT_UPD_1400 */ 153 }; 154 155 struct pfsync_q { 156 void (*write)(struct pf_kstate *, void *); 157 size_t len; 158 u_int8_t action; 159 }; 160 161 /* We have the following sync queues */ 162 enum pfsync_q_id { 163 PFSYNC_Q_INS_1301, 164 PFSYNC_Q_INS_1400, 165 PFSYNC_Q_IACK, 166 PFSYNC_Q_UPD_1301, 167 PFSYNC_Q_UPD_1400, 168 PFSYNC_Q_UPD_C, 169 PFSYNC_Q_DEL_C, 170 PFSYNC_Q_COUNT, 171 }; 172 173 /* Functions for building messages for given queue */ 174 static void pfsync_out_state_1301(struct pf_kstate *, void *); 175 static void pfsync_out_state_1400(struct pf_kstate *, void *); 176 static void pfsync_out_iack(struct pf_kstate *, void *); 177 static void pfsync_out_upd_c(struct pf_kstate *, void *); 178 static void pfsync_out_del_c(struct pf_kstate *, void *); 179 180 /* Attach those functions to queue */ 181 static struct pfsync_q pfsync_qs[] = { 182 { pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_INS_1301 }, 183 { pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_INS_1400 }, 184 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, 185 { pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_UPD_1301 }, 186 { pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_UPD_1400 }, 187 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, 188 { pfsync_out_del_c, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } 189 }; 190 191 /* Map queue to pf_kstate->sync_state */ 192 static u_int8_t pfsync_qid_sstate[] = { 193 PFSYNC_S_INS, /* PFSYNC_Q_INS_1301 */ 194 PFSYNC_S_INS, /* PFSYNC_Q_INS_1400 */ 195 PFSYNC_S_IACK, /* PFSYNC_Q_IACK */ 196 PFSYNC_S_UPD, /* PFSYNC_Q_UPD_1301 */ 197 PFSYNC_S_UPD, /* PFSYNC_Q_UPD_1400 */ 198 PFSYNC_S_UPD_C, /* PFSYNC_Q_UPD_C */ 199 PFSYNC_S_DEL_C, /* PFSYNC_Q_DEL_C */ 200 }; 201 202 /* Map pf_kstate->sync_state to queue */ 203 static enum pfsync_q_id pfsync_sstate_to_qid(u_int8_t); 204 205 static void pfsync_q_ins(struct pf_kstate *, int sync_state, bool); 206 static void pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *); 207 208 static void pfsync_update_state(struct pf_kstate *); 209 static void pfsync_tx(struct pfsync_softc *, struct mbuf *); 210 211 struct pfsync_upd_req_item { 212 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; 213 struct pfsync_upd_req ur_msg; 214 }; 215 216 struct pfsync_deferral { 217 struct pfsync_softc *pd_sc; 218 TAILQ_ENTRY(pfsync_deferral) pd_entry; 219 struct callout pd_tmo; 220 221 struct pf_kstate *pd_st; 222 struct mbuf *pd_m; 223 }; 224 225 struct pfsync_bucket 226 { 227 int b_id; 228 struct pfsync_softc *b_sc; 229 struct mtx b_mtx; 230 struct callout b_tmo; 231 int b_flags; 232 #define PFSYNCF_BUCKET_PUSH 0x00000001 233 234 size_t b_len; 235 TAILQ_HEAD(, pf_kstate) b_qs[PFSYNC_Q_COUNT]; 236 TAILQ_HEAD(, pfsync_upd_req_item) b_upd_req_list; 237 TAILQ_HEAD(, pfsync_deferral) b_deferrals; 238 u_int b_deferred; 239 void *b_plus; 240 size_t b_pluslen; 241 242 struct ifaltq b_snd; 243 }; 244 245 struct pfsync_softc { 246 /* Configuration */ 247 struct ifnet *sc_ifp; 248 struct ifnet *sc_sync_if; 249 struct ip_moptions sc_imo; 250 struct sockaddr_storage sc_sync_peer; 251 uint32_t sc_flags; 252 uint8_t sc_maxupdates; 253 union inet_template sc_template; 254 struct mtx sc_mtx; 255 uint32_t sc_version; 256 257 /* Queued data */ 258 struct pfsync_bucket *sc_buckets; 259 260 /* Bulk update info */ 261 struct mtx sc_bulk_mtx; 262 uint32_t sc_ureq_sent; 263 int sc_bulk_tries; 264 uint32_t sc_ureq_received; 265 int sc_bulk_hashid; 266 uint64_t sc_bulk_stateid; 267 uint32_t sc_bulk_creatorid; 268 struct callout sc_bulk_tmo; 269 struct callout sc_bulkfail_tmo; 270 }; 271 272 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 273 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 274 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 275 276 #define PFSYNC_BUCKET_LOCK(b) mtx_lock(&(b)->b_mtx) 277 #define PFSYNC_BUCKET_UNLOCK(b) mtx_unlock(&(b)->b_mtx) 278 #define PFSYNC_BUCKET_LOCK_ASSERT(b) mtx_assert(&(b)->b_mtx, MA_OWNED) 279 280 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 281 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 282 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 283 284 static const char pfsyncname[] = "pfsync"; 285 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); 286 VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL; 287 #define V_pfsyncif VNET(pfsyncif) 288 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL; 289 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 290 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie); 291 #define V_pfsync_swi_ie VNET(pfsync_swi_ie) 292 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats); 293 #define V_pfsyncstats VNET(pfsyncstats) 294 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW; 295 #define V_pfsync_carp_adj VNET(pfsync_carp_adj) 296 297 static void pfsync_timeout(void *); 298 static void pfsync_push(struct pfsync_bucket *); 299 static void pfsync_push_all(struct pfsync_softc *); 300 static void pfsyncintr(void *); 301 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 302 struct in_mfilter *imf); 303 static void pfsync_multicast_cleanup(struct pfsync_softc *); 304 static void pfsync_pointers_init(void); 305 static void pfsync_pointers_uninit(void); 306 static int pfsync_init(void); 307 static void pfsync_uninit(void); 308 309 static unsigned long pfsync_buckets; 310 311 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 312 "PFSYNC"); 313 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW, 314 &VNET_NAME(pfsyncstats), pfsyncstats, 315 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 316 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW, 317 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 318 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN, 319 &pfsync_buckets, 0, "Number of pfsync hash buckets"); 320 321 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 322 static void pfsync_clone_destroy(struct ifnet *); 323 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 324 struct pf_state_peer *); 325 static int pfsyncoutput(struct ifnet *, struct mbuf *, 326 const struct sockaddr *, struct route *); 327 static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 328 329 static int pfsync_defer(struct pf_kstate *, struct mbuf *); 330 static void pfsync_undefer(struct pfsync_deferral *, int); 331 static void pfsync_undefer_state_locked(struct pf_kstate *, int); 332 static void pfsync_undefer_state(struct pf_kstate *, int); 333 static void pfsync_defer_tmo(void *); 334 335 static void pfsync_request_update(u_int32_t, u_int64_t); 336 static bool pfsync_update_state_req(struct pf_kstate *); 337 338 static void pfsync_drop(struct pfsync_softc *); 339 static void pfsync_sendout(int, int); 340 static void pfsync_send_plus(void *, size_t); 341 342 static void pfsync_bulk_start(void); 343 static void pfsync_bulk_status(u_int8_t); 344 static void pfsync_bulk_update(void *); 345 static void pfsync_bulk_fail(void *); 346 347 static void pfsync_detach_ifnet(struct ifnet *); 348 349 static int pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *, 350 struct pfsync_kstatus *); 351 static int pfsync_kstatus_to_softc(struct pfsync_kstatus *, 352 struct pfsync_softc *); 353 354 #ifdef IPSEC 355 static void pfsync_update_net_tdb(struct pfsync_tdb *); 356 #endif 357 static struct pfsync_bucket *pfsync_get_bucket(struct pfsync_softc *, 358 struct pf_kstate *); 359 360 #define PFSYNC_MAX_BULKTRIES 12 361 #define PFSYNC_DEFER_TIMEOUT ((20 * hz) / 1000) 362 363 VNET_DEFINE(struct if_clone *, pfsync_cloner); 364 #define V_pfsync_cloner VNET(pfsync_cloner) 365 366 static int 367 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 368 { 369 struct pfsync_softc *sc; 370 struct ifnet *ifp; 371 struct pfsync_bucket *b; 372 int c; 373 enum pfsync_q_id q; 374 375 if (unit != 0) 376 return (EINVAL); 377 378 if (! pfsync_buckets) 379 pfsync_buckets = mp_ncpus * 2; 380 381 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 382 sc->sc_flags |= PFSYNCF_OK; 383 sc->sc_maxupdates = 128; 384 sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT; 385 386 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 387 if (ifp == NULL) { 388 free(sc, M_PFSYNC); 389 return (ENOSPC); 390 } 391 if_initname(ifp, pfsyncname, unit); 392 ifp->if_softc = sc; 393 ifp->if_ioctl = pfsyncioctl; 394 ifp->if_output = pfsyncoutput; 395 ifp->if_type = IFT_PFSYNC; 396 ifp->if_hdrlen = sizeof(struct pfsync_header); 397 ifp->if_mtu = ETHERMTU; 398 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); 399 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 400 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 401 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 402 403 if_attach(ifp); 404 405 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 406 407 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets), 408 M_PFSYNC, M_ZERO | M_WAITOK); 409 for (c = 0; c < pfsync_buckets; c++) { 410 b = &sc->sc_buckets[c]; 411 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF); 412 413 b->b_id = c; 414 b->b_sc = sc; 415 b->b_len = PFSYNC_MINPKT; 416 417 for (q = 0; q < PFSYNC_Q_COUNT; q++) 418 TAILQ_INIT(&b->b_qs[q]); 419 420 TAILQ_INIT(&b->b_upd_req_list); 421 TAILQ_INIT(&b->b_deferrals); 422 423 callout_init(&b->b_tmo, 1); 424 425 b->b_snd.ifq_maxlen = ifqmaxlen; 426 } 427 428 V_pfsyncif = sc; 429 430 return (0); 431 } 432 433 static void 434 pfsync_clone_destroy(struct ifnet *ifp) 435 { 436 struct pfsync_softc *sc = ifp->if_softc; 437 struct pfsync_bucket *b; 438 int c, ret; 439 440 for (c = 0; c < pfsync_buckets; c++) { 441 b = &sc->sc_buckets[c]; 442 /* 443 * At this stage, everything should have already been 444 * cleared by pfsync_uninit(), and we have only to 445 * drain callouts. 446 */ 447 PFSYNC_BUCKET_LOCK(b); 448 while (b->b_deferred > 0) { 449 struct pfsync_deferral *pd = 450 TAILQ_FIRST(&b->b_deferrals); 451 452 ret = callout_stop(&pd->pd_tmo); 453 PFSYNC_BUCKET_UNLOCK(b); 454 if (ret > 0) { 455 pfsync_undefer(pd, 1); 456 } else { 457 callout_drain(&pd->pd_tmo); 458 } 459 PFSYNC_BUCKET_LOCK(b); 460 } 461 MPASS(b->b_deferred == 0); 462 MPASS(TAILQ_EMPTY(&b->b_deferrals)); 463 PFSYNC_BUCKET_UNLOCK(b); 464 465 callout_drain(&b->b_tmo); 466 } 467 468 callout_drain(&sc->sc_bulkfail_tmo); 469 callout_drain(&sc->sc_bulk_tmo); 470 471 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 472 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 473 bpfdetach(ifp); 474 if_detach(ifp); 475 476 pfsync_drop(sc); 477 478 if_free(ifp); 479 pfsync_multicast_cleanup(sc); 480 mtx_destroy(&sc->sc_mtx); 481 mtx_destroy(&sc->sc_bulk_mtx); 482 483 free(sc->sc_buckets, M_PFSYNC); 484 free(sc, M_PFSYNC); 485 486 V_pfsyncif = NULL; 487 } 488 489 static int 490 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 491 struct pf_state_peer *d) 492 { 493 if (s->scrub.scrub_flag && d->scrub == NULL) { 494 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 495 if (d->scrub == NULL) 496 return (ENOMEM); 497 } 498 499 return (0); 500 } 501 502 static int 503 pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version) 504 { 505 struct pfsync_softc *sc = V_pfsyncif; 506 #ifndef __NO_STRICT_ALIGNMENT 507 struct pfsync_state_key key[2]; 508 #endif 509 struct pfsync_state_key *kw, *ks; 510 struct pf_kstate *st = NULL; 511 struct pf_state_key *skw = NULL, *sks = NULL; 512 struct pf_krule *r = NULL; 513 struct pfi_kkif *kif; 514 int error; 515 516 PF_RULES_RASSERT(); 517 518 if (sp->pfs_1301.creatorid == 0) { 519 if (V_pf_status.debug >= PF_DEBUG_MISC) 520 printf("%s: invalid creator id: %08x\n", __func__, 521 ntohl(sp->pfs_1301.creatorid)); 522 return (EINVAL); 523 } 524 525 if ((kif = pfi_kkif_find(sp->pfs_1301.ifname)) == NULL) { 526 if (V_pf_status.debug >= PF_DEBUG_MISC) 527 printf("%s: unknown interface: %s\n", __func__, 528 sp->pfs_1301.ifname); 529 if (flags & PFSYNC_SI_IOCTL) 530 return (EINVAL); 531 return (0); /* skip this state */ 532 } 533 534 /* 535 * If the ruleset checksums match or the state is coming from the ioctl, 536 * it's safe to associate the state with the rule of that number. 537 */ 538 if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) && 539 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) < 540 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 541 r = pf_main_ruleset.rules[ 542 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->pfs_1301.rule)]; 543 else 544 r = &V_pf_default_rule; 545 546 if ((r->max_states && 547 counter_u64_fetch(r->states_cur) >= r->max_states)) 548 goto cleanup; 549 550 /* 551 * XXXGL: consider M_WAITOK in ioctl path after. 552 */ 553 st = pf_alloc_state(M_NOWAIT); 554 if (__predict_false(st == NULL)) 555 goto cleanup; 556 557 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 558 goto cleanup; 559 560 #ifndef __NO_STRICT_ALIGNMENT 561 bcopy(&sp->pfs_1301.key, key, sizeof(struct pfsync_state_key) * 2); 562 kw = &key[PF_SK_WIRE]; 563 ks = &key[PF_SK_STACK]; 564 #else 565 kw = &sp->pfs_1301.key[PF_SK_WIRE]; 566 ks = &sp->pfs_1301.key[PF_SK_STACK]; 567 #endif 568 569 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->pfs_1301.af) || 570 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->pfs_1301.af) || 571 kw->port[0] != ks->port[0] || 572 kw->port[1] != ks->port[1]) { 573 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 574 if (sks == NULL) 575 goto cleanup; 576 } else 577 sks = skw; 578 579 /* allocate memory for scrub info */ 580 if (pfsync_alloc_scrub_memory(&sp->pfs_1301.src, &st->src) || 581 pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst)) 582 goto cleanup; 583 584 /* Copy to state key(s). */ 585 skw->addr[0] = kw->addr[0]; 586 skw->addr[1] = kw->addr[1]; 587 skw->port[0] = kw->port[0]; 588 skw->port[1] = kw->port[1]; 589 skw->proto = sp->pfs_1301.proto; 590 skw->af = sp->pfs_1301.af; 591 if (sks != skw) { 592 sks->addr[0] = ks->addr[0]; 593 sks->addr[1] = ks->addr[1]; 594 sks->port[0] = ks->port[0]; 595 sks->port[1] = ks->port[1]; 596 sks->proto = sp->pfs_1301.proto; 597 sks->af = sp->pfs_1301.af; 598 } 599 600 /* copy to state */ 601 bcopy(&sp->pfs_1301.rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 602 st->creation = time_uptime - ntohl(sp->pfs_1301.creation); 603 st->expire = time_uptime; 604 if (sp->pfs_1301.expire) { 605 uint32_t timeout; 606 607 timeout = r->timeout[sp->pfs_1301.timeout]; 608 if (!timeout) 609 timeout = V_pf_default_rule.timeout[sp->pfs_1301.timeout]; 610 611 /* sp->expire may have been adaptively scaled by export. */ 612 st->expire -= timeout - ntohl(sp->pfs_1301.expire); 613 } 614 615 st->direction = sp->pfs_1301.direction; 616 st->log = sp->pfs_1301.log; 617 st->timeout = sp->pfs_1301.timeout; 618 619 switch (msg_version) { 620 case PFSYNC_MSG_VERSION_1301: 621 st->state_flags = sp->pfs_1301.state_flags; 622 /* 623 * In FreeBSD 13 pfsync lacks many attributes. Copy them 624 * from the rule if possible. If rule can't be matched 625 * clear any set options as we can't recover their 626 * parameters. 627 */ 628 if (r == &V_pf_default_rule) { 629 st->state_flags &= ~PFSTATE_SETMASK; 630 } else { 631 /* 632 * Similar to pf_rule_to_actions(). This code 633 * won't set the actions properly if they come 634 * from multiple "match" rules as only rule 635 * creating the state is send over pfsync. 636 */ 637 st->qid = r->qid; 638 st->pqid = r->pqid; 639 st->rtableid = r->rtableid; 640 if (r->scrub_flags & PFSTATE_SETTOS) 641 st->set_tos = r->set_tos; 642 st->min_ttl = r->min_ttl; 643 st->max_mss = r->max_mss; 644 st->state_flags |= (r->scrub_flags & 645 (PFSTATE_NODF|PFSTATE_RANDOMID| 646 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP| 647 PFSTATE_SETPRIO)); 648 if (r->dnpipe || r->dnrpipe) { 649 if (r->free_flags & PFRULE_DN_IS_PIPE) 650 st->state_flags |= PFSTATE_DN_IS_PIPE; 651 else 652 st->state_flags &= ~PFSTATE_DN_IS_PIPE; 653 } 654 st->dnpipe = r->dnpipe; 655 st->dnrpipe = r->dnrpipe; 656 } 657 break; 658 case PFSYNC_MSG_VERSION_1400: 659 st->state_flags = ntohs(sp->pfs_1400.state_flags); 660 st->qid = ntohs(sp->pfs_1400.qid); 661 st->pqid = ntohs(sp->pfs_1400.pqid); 662 st->dnpipe = ntohs(sp->pfs_1400.dnpipe); 663 st->dnrpipe = ntohs(sp->pfs_1400.dnrpipe); 664 st->rtableid = ntohl(sp->pfs_1400.rtableid); 665 st->min_ttl = sp->pfs_1400.min_ttl; 666 st->set_tos = sp->pfs_1400.set_tos; 667 st->max_mss = ntohs(sp->pfs_1400.max_mss); 668 st->set_prio[0] = sp->pfs_1400.set_prio[0]; 669 st->set_prio[1] = sp->pfs_1400.set_prio[1]; 670 st->rt = sp->pfs_1400.rt; 671 if (st->rt && (st->rt_kif = pfi_kkif_find(sp->pfs_1400.rt_ifname)) == NULL) { 672 if (V_pf_status.debug >= PF_DEBUG_MISC) 673 printf("%s: unknown route interface: %s\n", 674 __func__, sp->pfs_1400.rt_ifname); 675 if (flags & PFSYNC_SI_IOCTL) 676 return (EINVAL); 677 return (0); /* skip this state */ 678 } 679 break; 680 default: 681 panic("%s: Unsupported pfsync_msg_version %d", 682 __func__, msg_version); 683 } 684 685 st->id = sp->pfs_1301.id; 686 st->creatorid = sp->pfs_1301.creatorid; 687 pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src); 688 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst); 689 690 st->rule.ptr = r; 691 st->nat_rule.ptr = NULL; 692 st->anchor.ptr = NULL; 693 694 st->pfsync_time = time_uptime; 695 st->sync_state = PFSYNC_S_NONE; 696 697 if (!(flags & PFSYNC_SI_IOCTL)) 698 st->state_flags |= PFSTATE_NOSYNC; 699 700 if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0) 701 goto cleanup_state; 702 703 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 704 counter_u64_add(r->states_cur, 1); 705 counter_u64_add(r->states_tot, 1); 706 707 if (!(flags & PFSYNC_SI_IOCTL)) { 708 st->state_flags &= ~PFSTATE_NOSYNC; 709 if (st->state_flags & PFSTATE_ACK) { 710 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 711 PFSYNC_BUCKET_LOCK(b); 712 pfsync_q_ins(st, PFSYNC_S_IACK, true); 713 PFSYNC_BUCKET_UNLOCK(b); 714 715 pfsync_push_all(sc); 716 } 717 } 718 st->state_flags &= ~PFSTATE_ACK; 719 PF_STATE_UNLOCK(st); 720 721 return (0); 722 723 cleanup: 724 error = ENOMEM; 725 if (skw == sks) 726 sks = NULL; 727 uma_zfree(V_pf_state_key_z, skw); 728 uma_zfree(V_pf_state_key_z, sks); 729 730 cleanup_state: /* pf_state_insert() frees the state keys. */ 731 if (st) { 732 st->timeout = PFTM_UNLINKED; /* appease an assert */ 733 pf_free_state(st); 734 } 735 return (error); 736 } 737 738 #ifdef INET 739 static int 740 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused) 741 { 742 struct pfsync_softc *sc = V_pfsyncif; 743 struct mbuf *m = *mp; 744 struct ip *ip = mtod(m, struct ip *); 745 struct pfsync_header *ph; 746 struct pfsync_subheader subh; 747 748 int offset, len, flags = 0; 749 int rv; 750 uint16_t count; 751 752 PF_RULES_RLOCK_TRACKER; 753 754 *mp = NULL; 755 V_pfsyncstats.pfsyncs_ipackets++; 756 757 /* Verify that we have a sync interface configured. */ 758 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 759 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 760 goto done; 761 762 /* verify that the packet came in on the right interface */ 763 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 764 V_pfsyncstats.pfsyncs_badif++; 765 goto done; 766 } 767 768 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1); 769 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); 770 /* verify that the IP TTL is 255. */ 771 if (ip->ip_ttl != PFSYNC_DFLTTL) { 772 V_pfsyncstats.pfsyncs_badttl++; 773 goto done; 774 } 775 776 offset = ip->ip_hl << 2; 777 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 778 V_pfsyncstats.pfsyncs_hdrops++; 779 goto done; 780 } 781 782 if (offset + sizeof(*ph) > m->m_len) { 783 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 784 V_pfsyncstats.pfsyncs_hdrops++; 785 return (IPPROTO_DONE); 786 } 787 ip = mtod(m, struct ip *); 788 } 789 ph = (struct pfsync_header *)((char *)ip + offset); 790 791 /* verify the version */ 792 if (ph->version != PFSYNC_VERSION) { 793 V_pfsyncstats.pfsyncs_badver++; 794 goto done; 795 } 796 797 len = ntohs(ph->len) + offset; 798 if (m->m_pkthdr.len < len) { 799 V_pfsyncstats.pfsyncs_badlen++; 800 goto done; 801 } 802 803 /* 804 * Trusting pf_chksum during packet processing, as well as seeking 805 * in interface name tree, require holding PF_RULES_RLOCK(). 806 */ 807 PF_RULES_RLOCK(); 808 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 809 flags = PFSYNC_SI_CKSUM; 810 811 offset += sizeof(*ph); 812 while (offset <= len - sizeof(subh)) { 813 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 814 offset += sizeof(subh); 815 816 if (subh.action >= PFSYNC_ACT_MAX) { 817 V_pfsyncstats.pfsyncs_badact++; 818 PF_RULES_RUNLOCK(); 819 goto done; 820 } 821 822 count = ntohs(subh.count); 823 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 824 rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action); 825 if (rv == -1) { 826 PF_RULES_RUNLOCK(); 827 return (IPPROTO_DONE); 828 } 829 830 offset += rv; 831 } 832 PF_RULES_RUNLOCK(); 833 834 done: 835 m_freem(m); 836 return (IPPROTO_DONE); 837 } 838 #endif 839 840 static int 841 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action) 842 { 843 struct pfsync_clr *clr; 844 struct mbuf *mp; 845 int len = sizeof(*clr) * count; 846 int i, offp; 847 u_int32_t creatorid; 848 849 mp = m_pulldown(m, offset, len, &offp); 850 if (mp == NULL) { 851 V_pfsyncstats.pfsyncs_badlen++; 852 return (-1); 853 } 854 clr = (struct pfsync_clr *)(mp->m_data + offp); 855 856 for (i = 0; i < count; i++) { 857 creatorid = clr[i].creatorid; 858 859 if (clr[i].ifname[0] != '\0' && 860 pfi_kkif_find(clr[i].ifname) == NULL) 861 continue; 862 863 for (int i = 0; i <= pf_hashmask; i++) { 864 struct pf_idhash *ih = &V_pf_idhash[i]; 865 struct pf_kstate *s; 866 relock: 867 PF_HASHROW_LOCK(ih); 868 LIST_FOREACH(s, &ih->states, entry) { 869 if (s->creatorid == creatorid) { 870 s->state_flags |= PFSTATE_NOSYNC; 871 pf_unlink_state(s); 872 goto relock; 873 } 874 } 875 PF_HASHROW_UNLOCK(ih); 876 } 877 } 878 879 return (len); 880 } 881 882 static int 883 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action) 884 { 885 struct mbuf *mp; 886 union pfsync_state_union *sa, *sp; 887 int i, offp, len, msg_version; 888 889 switch (action) { 890 case PFSYNC_ACT_INS_1301: 891 len = sizeof(struct pfsync_state_1301) * count; 892 msg_version = PFSYNC_MSG_VERSION_1301; 893 break; 894 case PFSYNC_ACT_INS_1400: 895 len = sizeof(struct pfsync_state_1400) * count; 896 msg_version = PFSYNC_MSG_VERSION_1400; 897 break; 898 default: 899 V_pfsyncstats.pfsyncs_badact++; 900 return (-1); 901 } 902 903 mp = m_pulldown(m, offset, len, &offp); 904 if (mp == NULL) { 905 V_pfsyncstats.pfsyncs_badlen++; 906 return (-1); 907 } 908 sa = (union pfsync_state_union *)(mp->m_data + offp); 909 910 for (i = 0; i < count; i++) { 911 sp = &sa[i]; 912 913 /* Check for invalid values. */ 914 if (sp->pfs_1301.timeout >= PFTM_MAX || 915 sp->pfs_1301.src.state > PF_TCPS_PROXY_DST || 916 sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST || 917 sp->pfs_1301.direction > PF_OUT || 918 (sp->pfs_1301.af != AF_INET && 919 sp->pfs_1301.af != AF_INET6)) { 920 if (V_pf_status.debug >= PF_DEBUG_MISC) 921 printf("%s: invalid value\n", __func__); 922 V_pfsyncstats.pfsyncs_badval++; 923 continue; 924 } 925 926 if (pfsync_state_import(sp, flags, msg_version) == ENOMEM) 927 /* Drop out, but process the rest of the actions. */ 928 break; 929 } 930 931 return (len); 932 } 933 934 static int 935 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action) 936 { 937 struct pfsync_ins_ack *ia, *iaa; 938 struct pf_kstate *st; 939 940 struct mbuf *mp; 941 int len = count * sizeof(*ia); 942 int offp, i; 943 944 mp = m_pulldown(m, offset, len, &offp); 945 if (mp == NULL) { 946 V_pfsyncstats.pfsyncs_badlen++; 947 return (-1); 948 } 949 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 950 951 for (i = 0; i < count; i++) { 952 ia = &iaa[i]; 953 954 st = pf_find_state_byid(ia->id, ia->creatorid); 955 if (st == NULL) 956 continue; 957 958 if (st->state_flags & PFSTATE_ACK) { 959 pfsync_undefer_state(st, 0); 960 } 961 PF_STATE_UNLOCK(st); 962 } 963 /* 964 * XXX this is not yet implemented, but we know the size of the 965 * message so we can skip it. 966 */ 967 968 return (count * sizeof(struct pfsync_ins_ack)); 969 } 970 971 static int 972 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src, 973 struct pfsync_state_peer *dst) 974 { 975 int sync = 0; 976 977 PF_STATE_LOCK_ASSERT(st); 978 979 /* 980 * The state should never go backwards except 981 * for syn-proxy states. Neither should the 982 * sequence window slide backwards. 983 */ 984 if ((st->src.state > src->state && 985 (st->src.state < PF_TCPS_PROXY_SRC || 986 src->state >= PF_TCPS_PROXY_SRC)) || 987 988 (st->src.state == src->state && 989 SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))) 990 sync++; 991 else 992 pf_state_peer_ntoh(src, &st->src); 993 994 if ((st->dst.state > dst->state) || 995 996 (st->dst.state >= TCPS_SYN_SENT && 997 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) 998 sync++; 999 else 1000 pf_state_peer_ntoh(dst, &st->dst); 1001 1002 return (sync); 1003 } 1004 1005 static int 1006 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action) 1007 { 1008 struct pfsync_softc *sc = V_pfsyncif; 1009 union pfsync_state_union *sa, *sp; 1010 struct pf_kstate *st; 1011 struct mbuf *mp; 1012 int sync, offp, i, len, msg_version; 1013 1014 switch (action) { 1015 case PFSYNC_ACT_UPD_1301: 1016 len = sizeof(struct pfsync_state_1301) * count; 1017 msg_version = PFSYNC_MSG_VERSION_1301; 1018 break; 1019 case PFSYNC_ACT_UPD_1400: 1020 len = sizeof(struct pfsync_state_1400) * count; 1021 msg_version = PFSYNC_MSG_VERSION_1400; 1022 break; 1023 default: 1024 V_pfsyncstats.pfsyncs_badact++; 1025 return (-1); 1026 } 1027 1028 mp = m_pulldown(m, offset, len, &offp); 1029 if (mp == NULL) { 1030 V_pfsyncstats.pfsyncs_badlen++; 1031 return (-1); 1032 } 1033 sa = (union pfsync_state_union *)(mp->m_data + offp); 1034 1035 for (i = 0; i < count; i++) { 1036 sp = &sa[i]; 1037 1038 /* check for invalid values */ 1039 if (sp->pfs_1301.timeout >= PFTM_MAX || 1040 sp->pfs_1301.src.state > PF_TCPS_PROXY_DST || 1041 sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) { 1042 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1043 printf("pfsync_input: PFSYNC_ACT_UPD: " 1044 "invalid value\n"); 1045 } 1046 V_pfsyncstats.pfsyncs_badval++; 1047 continue; 1048 } 1049 1050 st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid); 1051 if (st == NULL) { 1052 /* insert the update */ 1053 if (pfsync_state_import(sp, flags, msg_version)) 1054 V_pfsyncstats.pfsyncs_badstate++; 1055 continue; 1056 } 1057 1058 if (st->state_flags & PFSTATE_ACK) { 1059 pfsync_undefer_state(st, 1); 1060 } 1061 1062 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 1063 sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst); 1064 else { 1065 sync = 0; 1066 1067 /* 1068 * Non-TCP protocol state machine always go 1069 * forwards 1070 */ 1071 if (st->src.state > sp->pfs_1301.src.state) 1072 sync++; 1073 else 1074 pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src); 1075 if (st->dst.state > sp->pfs_1301.dst.state) 1076 sync++; 1077 else 1078 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst); 1079 } 1080 if (sync < 2) { 1081 pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst); 1082 pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst); 1083 st->expire = time_uptime; 1084 st->timeout = sp->pfs_1301.timeout; 1085 } 1086 st->pfsync_time = time_uptime; 1087 1088 if (sync) { 1089 V_pfsyncstats.pfsyncs_stale++; 1090 1091 pfsync_update_state(st); 1092 PF_STATE_UNLOCK(st); 1093 pfsync_push_all(sc); 1094 continue; 1095 } 1096 PF_STATE_UNLOCK(st); 1097 } 1098 1099 return (len); 1100 } 1101 1102 static int 1103 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action) 1104 { 1105 struct pfsync_softc *sc = V_pfsyncif; 1106 struct pfsync_upd_c *ua, *up; 1107 struct pf_kstate *st; 1108 int len = count * sizeof(*up); 1109 int sync; 1110 struct mbuf *mp; 1111 int offp, i; 1112 1113 mp = m_pulldown(m, offset, len, &offp); 1114 if (mp == NULL) { 1115 V_pfsyncstats.pfsyncs_badlen++; 1116 return (-1); 1117 } 1118 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 1119 1120 for (i = 0; i < count; i++) { 1121 up = &ua[i]; 1122 1123 /* check for invalid values */ 1124 if (up->timeout >= PFTM_MAX || 1125 up->src.state > PF_TCPS_PROXY_DST || 1126 up->dst.state > PF_TCPS_PROXY_DST) { 1127 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1128 printf("pfsync_input: " 1129 "PFSYNC_ACT_UPD_C: " 1130 "invalid value\n"); 1131 } 1132 V_pfsyncstats.pfsyncs_badval++; 1133 continue; 1134 } 1135 1136 st = pf_find_state_byid(up->id, up->creatorid); 1137 if (st == NULL) { 1138 /* We don't have this state. Ask for it. */ 1139 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 1140 pfsync_request_update(up->creatorid, up->id); 1141 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 1142 continue; 1143 } 1144 1145 if (st->state_flags & PFSTATE_ACK) { 1146 pfsync_undefer_state(st, 1); 1147 } 1148 1149 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 1150 sync = pfsync_upd_tcp(st, &up->src, &up->dst); 1151 else { 1152 sync = 0; 1153 1154 /* 1155 * Non-TCP protocol state machine always go 1156 * forwards 1157 */ 1158 if (st->src.state > up->src.state) 1159 sync++; 1160 else 1161 pf_state_peer_ntoh(&up->src, &st->src); 1162 if (st->dst.state > up->dst.state) 1163 sync++; 1164 else 1165 pf_state_peer_ntoh(&up->dst, &st->dst); 1166 } 1167 if (sync < 2) { 1168 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 1169 pf_state_peer_ntoh(&up->dst, &st->dst); 1170 st->expire = time_uptime; 1171 st->timeout = up->timeout; 1172 } 1173 st->pfsync_time = time_uptime; 1174 1175 if (sync) { 1176 V_pfsyncstats.pfsyncs_stale++; 1177 1178 pfsync_update_state(st); 1179 PF_STATE_UNLOCK(st); 1180 pfsync_push_all(sc); 1181 continue; 1182 } 1183 PF_STATE_UNLOCK(st); 1184 } 1185 1186 return (len); 1187 } 1188 1189 static int 1190 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action) 1191 { 1192 struct pfsync_upd_req *ur, *ura; 1193 struct mbuf *mp; 1194 int len = count * sizeof(*ur); 1195 int i, offp; 1196 1197 struct pf_kstate *st; 1198 1199 mp = m_pulldown(m, offset, len, &offp); 1200 if (mp == NULL) { 1201 V_pfsyncstats.pfsyncs_badlen++; 1202 return (-1); 1203 } 1204 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1205 1206 for (i = 0; i < count; i++) { 1207 ur = &ura[i]; 1208 1209 if (ur->id == 0 && ur->creatorid == 0) 1210 pfsync_bulk_start(); 1211 else { 1212 st = pf_find_state_byid(ur->id, ur->creatorid); 1213 if (st == NULL) { 1214 V_pfsyncstats.pfsyncs_badstate++; 1215 continue; 1216 } 1217 if (st->state_flags & PFSTATE_NOSYNC) { 1218 PF_STATE_UNLOCK(st); 1219 continue; 1220 } 1221 1222 pfsync_update_state_req(st); 1223 PF_STATE_UNLOCK(st); 1224 } 1225 } 1226 1227 return (len); 1228 } 1229 1230 static int 1231 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action) 1232 { 1233 struct mbuf *mp; 1234 struct pfsync_del_c *sa, *sp; 1235 struct pf_kstate *st; 1236 int len = count * sizeof(*sp); 1237 int offp, i; 1238 1239 mp = m_pulldown(m, offset, len, &offp); 1240 if (mp == NULL) { 1241 V_pfsyncstats.pfsyncs_badlen++; 1242 return (-1); 1243 } 1244 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1245 1246 for (i = 0; i < count; i++) { 1247 sp = &sa[i]; 1248 1249 st = pf_find_state_byid(sp->id, sp->creatorid); 1250 if (st == NULL) { 1251 V_pfsyncstats.pfsyncs_badstate++; 1252 continue; 1253 } 1254 1255 st->state_flags |= PFSTATE_NOSYNC; 1256 pf_unlink_state(st); 1257 } 1258 1259 return (len); 1260 } 1261 1262 static int 1263 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action) 1264 { 1265 struct pfsync_softc *sc = V_pfsyncif; 1266 struct pfsync_bus *bus; 1267 struct mbuf *mp; 1268 int len = count * sizeof(*bus); 1269 int offp; 1270 1271 PFSYNC_BLOCK(sc); 1272 1273 /* If we're not waiting for a bulk update, who cares. */ 1274 if (sc->sc_ureq_sent == 0) { 1275 PFSYNC_BUNLOCK(sc); 1276 return (len); 1277 } 1278 1279 mp = m_pulldown(m, offset, len, &offp); 1280 if (mp == NULL) { 1281 PFSYNC_BUNLOCK(sc); 1282 V_pfsyncstats.pfsyncs_badlen++; 1283 return (-1); 1284 } 1285 bus = (struct pfsync_bus *)(mp->m_data + offp); 1286 1287 switch (bus->status) { 1288 case PFSYNC_BUS_START: 1289 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1290 V_pf_limits[PF_LIMIT_STATES].limit / 1291 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1292 sizeof(union pfsync_state_union)), 1293 pfsync_bulk_fail, sc); 1294 if (V_pf_status.debug >= PF_DEBUG_MISC) 1295 printf("pfsync: received bulk update start\n"); 1296 break; 1297 1298 case PFSYNC_BUS_END: 1299 if (time_uptime - ntohl(bus->endtime) >= 1300 sc->sc_ureq_sent) { 1301 /* that's it, we're happy */ 1302 sc->sc_ureq_sent = 0; 1303 sc->sc_bulk_tries = 0; 1304 callout_stop(&sc->sc_bulkfail_tmo); 1305 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1306 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1307 "pfsync bulk done"); 1308 sc->sc_flags |= PFSYNCF_OK; 1309 if (V_pf_status.debug >= PF_DEBUG_MISC) 1310 printf("pfsync: received valid " 1311 "bulk update end\n"); 1312 } else { 1313 if (V_pf_status.debug >= PF_DEBUG_MISC) 1314 printf("pfsync: received invalid " 1315 "bulk update end: bad timestamp\n"); 1316 } 1317 break; 1318 } 1319 PFSYNC_BUNLOCK(sc); 1320 1321 return (len); 1322 } 1323 1324 static int 1325 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action) 1326 { 1327 int len = count * sizeof(struct pfsync_tdb); 1328 1329 #if defined(IPSEC) 1330 struct pfsync_tdb *tp; 1331 struct mbuf *mp; 1332 int offp; 1333 int i; 1334 int s; 1335 1336 mp = m_pulldown(m, offset, len, &offp); 1337 if (mp == NULL) { 1338 V_pfsyncstats.pfsyncs_badlen++; 1339 return (-1); 1340 } 1341 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1342 1343 for (i = 0; i < count; i++) 1344 pfsync_update_net_tdb(&tp[i]); 1345 #endif 1346 1347 return (len); 1348 } 1349 1350 #if defined(IPSEC) 1351 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1352 static void 1353 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1354 { 1355 struct tdb *tdb; 1356 int s; 1357 1358 /* check for invalid values */ 1359 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1360 (pt->dst.sa.sa_family != AF_INET && 1361 pt->dst.sa.sa_family != AF_INET6)) 1362 goto bad; 1363 1364 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1365 if (tdb) { 1366 pt->rpl = ntohl(pt->rpl); 1367 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1368 1369 /* Neither replay nor byte counter should ever decrease. */ 1370 if (pt->rpl < tdb->tdb_rpl || 1371 pt->cur_bytes < tdb->tdb_cur_bytes) { 1372 goto bad; 1373 } 1374 1375 tdb->tdb_rpl = pt->rpl; 1376 tdb->tdb_cur_bytes = pt->cur_bytes; 1377 } 1378 return; 1379 1380 bad: 1381 if (V_pf_status.debug >= PF_DEBUG_MISC) 1382 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1383 "invalid value\n"); 1384 V_pfsyncstats.pfsyncs_badstate++; 1385 return; 1386 } 1387 #endif 1388 1389 static int 1390 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action) 1391 { 1392 /* check if we are at the right place in the packet */ 1393 if (offset != m->m_pkthdr.len) 1394 V_pfsyncstats.pfsyncs_badlen++; 1395 1396 /* we're done. free and let the caller return */ 1397 m_freem(m); 1398 return (-1); 1399 } 1400 1401 static int 1402 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action) 1403 { 1404 V_pfsyncstats.pfsyncs_badact++; 1405 1406 m_freem(m); 1407 return (-1); 1408 } 1409 1410 static int 1411 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 1412 struct route *rt) 1413 { 1414 m_freem(m); 1415 return (0); 1416 } 1417 1418 /* ARGSUSED */ 1419 static int 1420 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1421 { 1422 struct pfsync_softc *sc = ifp->if_softc; 1423 struct ifreq *ifr = (struct ifreq *)data; 1424 struct pfsyncreq pfsyncr; 1425 size_t nvbuflen; 1426 int error; 1427 int c; 1428 1429 switch (cmd) { 1430 case SIOCSIFFLAGS: 1431 PFSYNC_LOCK(sc); 1432 if (ifp->if_flags & IFF_UP) { 1433 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1434 PFSYNC_UNLOCK(sc); 1435 pfsync_pointers_init(); 1436 } else { 1437 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1438 PFSYNC_UNLOCK(sc); 1439 pfsync_pointers_uninit(); 1440 } 1441 break; 1442 case SIOCSIFMTU: 1443 if (!sc->sc_sync_if || 1444 ifr->ifr_mtu <= PFSYNC_MINPKT || 1445 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1446 return (EINVAL); 1447 if (ifr->ifr_mtu < ifp->if_mtu) { 1448 for (c = 0; c < pfsync_buckets; c++) { 1449 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 1450 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT) 1451 pfsync_sendout(1, c); 1452 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 1453 } 1454 } 1455 ifp->if_mtu = ifr->ifr_mtu; 1456 break; 1457 case SIOCGETPFSYNC: 1458 bzero(&pfsyncr, sizeof(pfsyncr)); 1459 PFSYNC_LOCK(sc); 1460 if (sc->sc_sync_if) { 1461 strlcpy(pfsyncr.pfsyncr_syncdev, 1462 sc->sc_sync_if->if_xname, IFNAMSIZ); 1463 } 1464 pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr; 1465 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1466 pfsyncr.pfsyncr_defer = sc->sc_flags; 1467 PFSYNC_UNLOCK(sc); 1468 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr), 1469 sizeof(pfsyncr))); 1470 1471 case SIOCGETPFSYNCNV: 1472 { 1473 nvlist_t *nvl_syncpeer; 1474 nvlist_t *nvl = nvlist_create(0); 1475 1476 if (nvl == NULL) 1477 return (ENOMEM); 1478 1479 if (sc->sc_sync_if) 1480 nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname); 1481 nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates); 1482 nvlist_add_number(nvl, "flags", sc->sc_flags); 1483 nvlist_add_number(nvl, "version", sc->sc_version); 1484 if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL) 1485 nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer); 1486 1487 void *packed = NULL; 1488 packed = nvlist_pack(nvl, &nvbuflen); 1489 if (packed == NULL) { 1490 free(packed, M_NVLIST); 1491 nvlist_destroy(nvl); 1492 return (ENOMEM); 1493 } 1494 1495 if (nvbuflen > ifr->ifr_cap_nv.buf_length) { 1496 ifr->ifr_cap_nv.length = nvbuflen; 1497 ifr->ifr_cap_nv.buffer = NULL; 1498 free(packed, M_NVLIST); 1499 nvlist_destroy(nvl); 1500 return (EFBIG); 1501 } 1502 1503 ifr->ifr_cap_nv.length = nvbuflen; 1504 error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen); 1505 1506 nvlist_destroy(nvl); 1507 nvlist_destroy(nvl_syncpeer); 1508 free(packed, M_NVLIST); 1509 break; 1510 } 1511 1512 case SIOCSETPFSYNC: 1513 { 1514 struct pfsync_kstatus status; 1515 1516 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1517 return (error); 1518 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr, 1519 sizeof(pfsyncr)))) 1520 return (error); 1521 1522 memset((char *)&status, 0, sizeof(struct pfsync_kstatus)); 1523 pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status); 1524 1525 error = pfsync_kstatus_to_softc(&status, sc); 1526 return (error); 1527 } 1528 case SIOCSETPFSYNCNV: 1529 { 1530 struct pfsync_kstatus status; 1531 void *data; 1532 nvlist_t *nvl; 1533 1534 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1535 return (error); 1536 if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE) 1537 return (EINVAL); 1538 1539 data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK); 1540 1541 if ((error = copyin(ifr->ifr_cap_nv.buffer, data, 1542 ifr->ifr_cap_nv.length)) != 0) { 1543 free(data, M_TEMP); 1544 return (error); 1545 } 1546 1547 if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) { 1548 free(data, M_TEMP); 1549 return (EINVAL); 1550 } 1551 1552 memset((char *)&status, 0, sizeof(struct pfsync_kstatus)); 1553 pfsync_nvstatus_to_kstatus(nvl, &status); 1554 1555 nvlist_destroy(nvl); 1556 free(data, M_TEMP); 1557 1558 error = pfsync_kstatus_to_softc(&status, sc); 1559 return (error); 1560 } 1561 default: 1562 return (ENOTTY); 1563 } 1564 1565 return (0); 1566 } 1567 1568 static void 1569 pfsync_out_state_1301(struct pf_kstate *st, void *buf) 1570 { 1571 union pfsync_state_union *sp = buf; 1572 1573 pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301); 1574 } 1575 1576 static void 1577 pfsync_out_state_1400(struct pf_kstate *st, void *buf) 1578 { 1579 union pfsync_state_union *sp = buf; 1580 1581 pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400); 1582 } 1583 1584 static void 1585 pfsync_out_iack(struct pf_kstate *st, void *buf) 1586 { 1587 struct pfsync_ins_ack *iack = buf; 1588 1589 iack->id = st->id; 1590 iack->creatorid = st->creatorid; 1591 } 1592 1593 static void 1594 pfsync_out_upd_c(struct pf_kstate *st, void *buf) 1595 { 1596 struct pfsync_upd_c *up = buf; 1597 1598 bzero(up, sizeof(*up)); 1599 up->id = st->id; 1600 pf_state_peer_hton(&st->src, &up->src); 1601 pf_state_peer_hton(&st->dst, &up->dst); 1602 up->creatorid = st->creatorid; 1603 up->timeout = st->timeout; 1604 } 1605 1606 static void 1607 pfsync_out_del_c(struct pf_kstate *st, void *buf) 1608 { 1609 struct pfsync_del_c *dp = buf; 1610 1611 dp->id = st->id; 1612 dp->creatorid = st->creatorid; 1613 st->state_flags |= PFSTATE_NOSYNC; 1614 } 1615 1616 static void 1617 pfsync_drop(struct pfsync_softc *sc) 1618 { 1619 struct pf_kstate *st, *next; 1620 struct pfsync_upd_req_item *ur; 1621 struct pfsync_bucket *b; 1622 int c; 1623 enum pfsync_q_id q; 1624 1625 for (c = 0; c < pfsync_buckets; c++) { 1626 b = &sc->sc_buckets[c]; 1627 for (q = 0; q < PFSYNC_Q_COUNT; q++) { 1628 if (TAILQ_EMPTY(&b->b_qs[q])) 1629 continue; 1630 1631 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) { 1632 KASSERT(st->sync_state == pfsync_qid_sstate[q], 1633 ("%s: st->sync_state == q", 1634 __func__)); 1635 st->sync_state = PFSYNC_S_NONE; 1636 pf_release_state(st); 1637 } 1638 TAILQ_INIT(&b->b_qs[q]); 1639 } 1640 1641 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1642 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1643 free(ur, M_PFSYNC); 1644 } 1645 1646 b->b_len = PFSYNC_MINPKT; 1647 b->b_plus = NULL; 1648 } 1649 } 1650 1651 static void 1652 pfsync_sendout(int schedswi, int c) 1653 { 1654 struct pfsync_softc *sc = V_pfsyncif; 1655 struct ifnet *ifp = sc->sc_ifp; 1656 struct mbuf *m; 1657 struct pfsync_header *ph; 1658 struct pfsync_subheader *subh; 1659 struct pf_kstate *st, *st_next; 1660 struct pfsync_upd_req_item *ur; 1661 struct pfsync_bucket *b = &sc->sc_buckets[c]; 1662 int aflen, offset, count = 0; 1663 enum pfsync_q_id q; 1664 1665 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1666 KASSERT(b->b_len > PFSYNC_MINPKT, 1667 ("%s: sc_len %zu", __func__, b->b_len)); 1668 PFSYNC_BUCKET_LOCK_ASSERT(b); 1669 1670 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1671 pfsync_drop(sc); 1672 return; 1673 } 1674 1675 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR); 1676 if (m == NULL) { 1677 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); 1678 V_pfsyncstats.pfsyncs_onomem++; 1679 return; 1680 } 1681 m->m_data += max_linkhdr; 1682 m->m_len = m->m_pkthdr.len = b->b_len; 1683 1684 /* build the ip header */ 1685 switch (sc->sc_sync_peer.ss_family) { 1686 #ifdef INET 1687 case AF_INET: 1688 { 1689 struct ip *ip; 1690 1691 ip = mtod(m, struct ip *); 1692 bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip)); 1693 aflen = offset = sizeof(*ip); 1694 1695 ip->ip_len = htons(m->m_pkthdr.len); 1696 ip_fillid(ip); 1697 break; 1698 } 1699 #endif 1700 default: 1701 m_freem(m); 1702 return; 1703 } 1704 1705 /* build the pfsync header */ 1706 ph = (struct pfsync_header *)(m->m_data + offset); 1707 bzero(ph, sizeof(*ph)); 1708 offset += sizeof(*ph); 1709 1710 ph->version = PFSYNC_VERSION; 1711 ph->len = htons(b->b_len - aflen); 1712 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1713 1714 /* walk the queues */ 1715 for (q = 0; q < PFSYNC_Q_COUNT; q++) { 1716 if (TAILQ_EMPTY(&b->b_qs[q])) 1717 continue; 1718 1719 subh = (struct pfsync_subheader *)(m->m_data + offset); 1720 offset += sizeof(*subh); 1721 1722 count = 0; 1723 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) { 1724 KASSERT(st->sync_state == pfsync_qid_sstate[q], 1725 ("%s: st->sync_state == q", 1726 __func__)); 1727 /* 1728 * XXXGL: some of write methods do unlocked reads 1729 * of state data :( 1730 */ 1731 pfsync_qs[q].write(st, m->m_data + offset); 1732 offset += pfsync_qs[q].len; 1733 st->sync_state = PFSYNC_S_NONE; 1734 pf_release_state(st); 1735 count++; 1736 } 1737 TAILQ_INIT(&b->b_qs[q]); 1738 1739 bzero(subh, sizeof(*subh)); 1740 subh->action = pfsync_qs[q].action; 1741 subh->count = htons(count); 1742 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1743 } 1744 1745 if (!TAILQ_EMPTY(&b->b_upd_req_list)) { 1746 subh = (struct pfsync_subheader *)(m->m_data + offset); 1747 offset += sizeof(*subh); 1748 1749 count = 0; 1750 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1751 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1752 1753 bcopy(&ur->ur_msg, m->m_data + offset, 1754 sizeof(ur->ur_msg)); 1755 offset += sizeof(ur->ur_msg); 1756 free(ur, M_PFSYNC); 1757 count++; 1758 } 1759 1760 bzero(subh, sizeof(*subh)); 1761 subh->action = PFSYNC_ACT_UPD_REQ; 1762 subh->count = htons(count); 1763 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1764 } 1765 1766 /* has someone built a custom region for us to add? */ 1767 if (b->b_plus != NULL) { 1768 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen); 1769 offset += b->b_pluslen; 1770 1771 b->b_plus = NULL; 1772 } 1773 1774 subh = (struct pfsync_subheader *)(m->m_data + offset); 1775 offset += sizeof(*subh); 1776 1777 bzero(subh, sizeof(*subh)); 1778 subh->action = PFSYNC_ACT_EOF; 1779 subh->count = htons(1); 1780 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1781 1782 /* we're done, let's put it on the wire */ 1783 if (ifp->if_bpf) { 1784 m->m_data += aflen; 1785 m->m_len = m->m_pkthdr.len = b->b_len - aflen; 1786 BPF_MTAP(ifp, m); 1787 m->m_data -= aflen; 1788 m->m_len = m->m_pkthdr.len = b->b_len; 1789 } 1790 1791 if (sc->sc_sync_if == NULL) { 1792 b->b_len = PFSYNC_MINPKT; 1793 m_freem(m); 1794 return; 1795 } 1796 1797 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1); 1798 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); 1799 b->b_len = PFSYNC_MINPKT; 1800 1801 if (!_IF_QFULL(&b->b_snd)) 1802 _IF_ENQUEUE(&b->b_snd, m); 1803 else { 1804 m_freem(m); 1805 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1); 1806 } 1807 if (schedswi) 1808 swi_sched(V_pfsync_swi_cookie, 0); 1809 } 1810 1811 static void 1812 pfsync_insert_state(struct pf_kstate *st) 1813 { 1814 struct pfsync_softc *sc = V_pfsyncif; 1815 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1816 1817 if (st->state_flags & PFSTATE_NOSYNC) 1818 return; 1819 1820 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1821 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1822 st->state_flags |= PFSTATE_NOSYNC; 1823 return; 1824 } 1825 1826 KASSERT(st->sync_state == PFSYNC_S_NONE, 1827 ("%s: st->sync_state %u", __func__, st->sync_state)); 1828 1829 PFSYNC_BUCKET_LOCK(b); 1830 if (b->b_len == PFSYNC_MINPKT) 1831 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 1832 1833 pfsync_q_ins(st, PFSYNC_S_INS, true); 1834 PFSYNC_BUCKET_UNLOCK(b); 1835 1836 st->sync_updates = 0; 1837 } 1838 1839 static int 1840 pfsync_defer(struct pf_kstate *st, struct mbuf *m) 1841 { 1842 struct pfsync_softc *sc = V_pfsyncif; 1843 struct pfsync_deferral *pd; 1844 struct pfsync_bucket *b; 1845 1846 if (m->m_flags & (M_BCAST|M_MCAST)) 1847 return (0); 1848 1849 if (sc == NULL) 1850 return (0); 1851 1852 b = pfsync_get_bucket(sc, st); 1853 1854 PFSYNC_LOCK(sc); 1855 1856 if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) || 1857 !(sc->sc_flags & PFSYNCF_DEFER)) { 1858 PFSYNC_UNLOCK(sc); 1859 return (0); 1860 } 1861 1862 PFSYNC_BUCKET_LOCK(b); 1863 PFSYNC_UNLOCK(sc); 1864 1865 if (b->b_deferred >= 128) 1866 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0); 1867 1868 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1869 if (pd == NULL) { 1870 PFSYNC_BUCKET_UNLOCK(b); 1871 return (0); 1872 } 1873 b->b_deferred++; 1874 1875 m->m_flags |= M_SKIP_FIREWALL; 1876 st->state_flags |= PFSTATE_ACK; 1877 1878 pd->pd_sc = sc; 1879 pd->pd_st = st; 1880 pf_ref_state(st); 1881 pd->pd_m = m; 1882 1883 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry); 1884 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED); 1885 callout_reset(&pd->pd_tmo, PFSYNC_DEFER_TIMEOUT, pfsync_defer_tmo, pd); 1886 1887 pfsync_push(b); 1888 PFSYNC_BUCKET_UNLOCK(b); 1889 1890 return (1); 1891 } 1892 1893 static void 1894 pfsync_undefer(struct pfsync_deferral *pd, int drop) 1895 { 1896 struct pfsync_softc *sc = pd->pd_sc; 1897 struct mbuf *m = pd->pd_m; 1898 struct pf_kstate *st = pd->pd_st; 1899 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1900 1901 PFSYNC_BUCKET_LOCK_ASSERT(b); 1902 1903 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1904 b->b_deferred--; 1905 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1906 free(pd, M_PFSYNC); 1907 pf_release_state(st); 1908 1909 if (drop) 1910 m_freem(m); 1911 else { 1912 _IF_ENQUEUE(&b->b_snd, m); 1913 pfsync_push(b); 1914 } 1915 } 1916 1917 static void 1918 pfsync_defer_tmo(void *arg) 1919 { 1920 struct epoch_tracker et; 1921 struct pfsync_deferral *pd = arg; 1922 struct pfsync_softc *sc = pd->pd_sc; 1923 struct mbuf *m = pd->pd_m; 1924 struct pf_kstate *st = pd->pd_st; 1925 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1926 1927 PFSYNC_BUCKET_LOCK_ASSERT(b); 1928 1929 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1930 b->b_deferred--; 1931 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1932 PFSYNC_BUCKET_UNLOCK(b); 1933 free(pd, M_PFSYNC); 1934 1935 if (sc->sc_sync_if == NULL) { 1936 pf_release_state(st); 1937 m_freem(m); 1938 return; 1939 } 1940 1941 NET_EPOCH_ENTER(et); 1942 CURVNET_SET(sc->sc_sync_if->if_vnet); 1943 1944 pfsync_tx(sc, m); 1945 1946 pf_release_state(st); 1947 1948 CURVNET_RESTORE(); 1949 NET_EPOCH_EXIT(et); 1950 } 1951 1952 static void 1953 pfsync_undefer_state_locked(struct pf_kstate *st, int drop) 1954 { 1955 struct pfsync_softc *sc = V_pfsyncif; 1956 struct pfsync_deferral *pd; 1957 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1958 1959 PFSYNC_BUCKET_LOCK_ASSERT(b); 1960 1961 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) { 1962 if (pd->pd_st == st) { 1963 if (callout_stop(&pd->pd_tmo) > 0) 1964 pfsync_undefer(pd, drop); 1965 1966 return; 1967 } 1968 } 1969 1970 panic("%s: unable to find deferred state", __func__); 1971 } 1972 1973 static void 1974 pfsync_undefer_state(struct pf_kstate *st, int drop) 1975 { 1976 struct pfsync_softc *sc = V_pfsyncif; 1977 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1978 1979 PFSYNC_BUCKET_LOCK(b); 1980 pfsync_undefer_state_locked(st, drop); 1981 PFSYNC_BUCKET_UNLOCK(b); 1982 } 1983 1984 static struct pfsync_bucket* 1985 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st) 1986 { 1987 int c = PF_IDHASH(st) % pfsync_buckets; 1988 return &sc->sc_buckets[c]; 1989 } 1990 1991 static void 1992 pfsync_update_state(struct pf_kstate *st) 1993 { 1994 struct pfsync_softc *sc = V_pfsyncif; 1995 bool sync = false, ref = true; 1996 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1997 1998 PF_STATE_LOCK_ASSERT(st); 1999 PFSYNC_BUCKET_LOCK(b); 2000 2001 if (st->state_flags & PFSTATE_ACK) 2002 pfsync_undefer_state_locked(st, 0); 2003 if (st->state_flags & PFSTATE_NOSYNC) { 2004 if (st->sync_state != PFSYNC_S_NONE) 2005 pfsync_q_del(st, true, b); 2006 PFSYNC_BUCKET_UNLOCK(b); 2007 return; 2008 } 2009 2010 if (b->b_len == PFSYNC_MINPKT) 2011 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 2012 2013 switch (st->sync_state) { 2014 case PFSYNC_S_UPD_C: 2015 case PFSYNC_S_UPD: 2016 case PFSYNC_S_INS: 2017 /* we're already handling it */ 2018 2019 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 2020 st->sync_updates++; 2021 if (st->sync_updates >= sc->sc_maxupdates) 2022 sync = true; 2023 } 2024 break; 2025 2026 case PFSYNC_S_IACK: 2027 pfsync_q_del(st, false, b); 2028 ref = false; 2029 /* FALLTHROUGH */ 2030 2031 case PFSYNC_S_NONE: 2032 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref); 2033 st->sync_updates = 0; 2034 break; 2035 2036 default: 2037 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2038 } 2039 2040 if (sync || (time_uptime - st->pfsync_time) < 2) 2041 pfsync_push(b); 2042 2043 PFSYNC_BUCKET_UNLOCK(b); 2044 } 2045 2046 static void 2047 pfsync_request_update(u_int32_t creatorid, u_int64_t id) 2048 { 2049 struct pfsync_softc *sc = V_pfsyncif; 2050 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2051 struct pfsync_upd_req_item *item; 2052 size_t nlen = sizeof(struct pfsync_upd_req); 2053 2054 PFSYNC_BUCKET_LOCK_ASSERT(b); 2055 2056 /* 2057 * This code does a bit to prevent multiple update requests for the 2058 * same state being generated. It searches current subheader queue, 2059 * but it doesn't lookup into queue of already packed datagrams. 2060 */ 2061 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry) 2062 if (item->ur_msg.id == id && 2063 item->ur_msg.creatorid == creatorid) 2064 return; 2065 2066 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 2067 if (item == NULL) 2068 return; /* XXX stats */ 2069 2070 item->ur_msg.id = id; 2071 item->ur_msg.creatorid = creatorid; 2072 2073 if (TAILQ_EMPTY(&b->b_upd_req_list)) 2074 nlen += sizeof(struct pfsync_subheader); 2075 2076 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 2077 pfsync_sendout(0, 0); 2078 2079 nlen = sizeof(struct pfsync_subheader) + 2080 sizeof(struct pfsync_upd_req); 2081 } 2082 2083 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry); 2084 b->b_len += nlen; 2085 2086 pfsync_push(b); 2087 } 2088 2089 static bool 2090 pfsync_update_state_req(struct pf_kstate *st) 2091 { 2092 struct pfsync_softc *sc = V_pfsyncif; 2093 bool ref = true, full = false; 2094 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2095 2096 PF_STATE_LOCK_ASSERT(st); 2097 PFSYNC_BUCKET_LOCK(b); 2098 2099 if (st->state_flags & PFSTATE_NOSYNC) { 2100 if (st->sync_state != PFSYNC_S_NONE) 2101 pfsync_q_del(st, true, b); 2102 PFSYNC_BUCKET_UNLOCK(b); 2103 return (full); 2104 } 2105 2106 switch (st->sync_state) { 2107 case PFSYNC_S_UPD_C: 2108 case PFSYNC_S_IACK: 2109 pfsync_q_del(st, false, b); 2110 ref = false; 2111 /* FALLTHROUGH */ 2112 2113 case PFSYNC_S_NONE: 2114 pfsync_q_ins(st, PFSYNC_S_UPD, ref); 2115 pfsync_push(b); 2116 break; 2117 2118 case PFSYNC_S_INS: 2119 case PFSYNC_S_UPD: 2120 case PFSYNC_S_DEL_C: 2121 /* we're already handling it */ 2122 break; 2123 2124 default: 2125 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2126 } 2127 2128 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union)) 2129 full = true; 2130 2131 PFSYNC_BUCKET_UNLOCK(b); 2132 2133 return (full); 2134 } 2135 2136 static void 2137 pfsync_delete_state(struct pf_kstate *st) 2138 { 2139 struct pfsync_softc *sc = V_pfsyncif; 2140 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2141 bool ref = true; 2142 2143 PFSYNC_BUCKET_LOCK(b); 2144 if (st->state_flags & PFSTATE_ACK) 2145 pfsync_undefer_state_locked(st, 1); 2146 if (st->state_flags & PFSTATE_NOSYNC) { 2147 if (st->sync_state != PFSYNC_S_NONE) 2148 pfsync_q_del(st, true, b); 2149 PFSYNC_BUCKET_UNLOCK(b); 2150 return; 2151 } 2152 2153 if (b->b_len == PFSYNC_MINPKT) 2154 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 2155 2156 switch (st->sync_state) { 2157 case PFSYNC_S_INS: 2158 /* We never got to tell the world so just forget about it. */ 2159 pfsync_q_del(st, true, b); 2160 break; 2161 2162 case PFSYNC_S_UPD_C: 2163 case PFSYNC_S_UPD: 2164 case PFSYNC_S_IACK: 2165 pfsync_q_del(st, false, b); 2166 ref = false; 2167 /* FALLTHROUGH */ 2168 2169 case PFSYNC_S_NONE: 2170 pfsync_q_ins(st, PFSYNC_S_DEL_C, ref); 2171 break; 2172 2173 default: 2174 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2175 } 2176 2177 PFSYNC_BUCKET_UNLOCK(b); 2178 } 2179 2180 static void 2181 pfsync_clear_states(u_int32_t creatorid, const char *ifname) 2182 { 2183 struct { 2184 struct pfsync_subheader subh; 2185 struct pfsync_clr clr; 2186 } __packed r; 2187 2188 bzero(&r, sizeof(r)); 2189 2190 r.subh.action = PFSYNC_ACT_CLR; 2191 r.subh.count = htons(1); 2192 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 2193 2194 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 2195 r.clr.creatorid = creatorid; 2196 2197 pfsync_send_plus(&r, sizeof(r)); 2198 } 2199 2200 static enum pfsync_q_id 2201 pfsync_sstate_to_qid(u_int8_t sync_state) 2202 { 2203 struct pfsync_softc *sc = V_pfsyncif; 2204 2205 switch (sync_state) { 2206 case PFSYNC_S_INS: 2207 switch (sc->sc_version) { 2208 case PFSYNC_MSG_VERSION_1301: 2209 return PFSYNC_Q_INS_1301; 2210 case PFSYNC_MSG_VERSION_1400: 2211 return PFSYNC_Q_INS_1400; 2212 } 2213 break; 2214 case PFSYNC_S_IACK: 2215 return PFSYNC_Q_IACK; 2216 case PFSYNC_S_UPD: 2217 switch (sc->sc_version) { 2218 case PFSYNC_MSG_VERSION_1301: 2219 return PFSYNC_Q_UPD_1301; 2220 case PFSYNC_MSG_VERSION_1400: 2221 return PFSYNC_Q_UPD_1400; 2222 } 2223 break; 2224 case PFSYNC_S_UPD_C: 2225 return PFSYNC_Q_UPD_C; 2226 case PFSYNC_S_DEL_C: 2227 return PFSYNC_Q_DEL_C; 2228 default: 2229 panic("%s: Unsupported st->sync_state 0x%02x", 2230 __func__, sync_state); 2231 } 2232 2233 panic("%s: Unsupported pfsync_msg_version %d", 2234 __func__, sc->sc_version); 2235 } 2236 2237 static void 2238 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref) 2239 { 2240 enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state); 2241 struct pfsync_softc *sc = V_pfsyncif; 2242 size_t nlen = pfsync_qs[q].len; 2243 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2244 2245 PFSYNC_BUCKET_LOCK_ASSERT(b); 2246 2247 KASSERT(st->sync_state == PFSYNC_S_NONE, 2248 ("%s: st->sync_state %u", __func__, st->sync_state)); 2249 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 2250 b->b_len)); 2251 2252 if (TAILQ_EMPTY(&b->b_qs[q])) 2253 nlen += sizeof(struct pfsync_subheader); 2254 2255 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 2256 pfsync_sendout(1, b->b_id); 2257 2258 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 2259 } 2260 2261 b->b_len += nlen; 2262 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list); 2263 st->sync_state = pfsync_qid_sstate[q]; 2264 if (ref) 2265 pf_ref_state(st); 2266 } 2267 2268 static void 2269 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b) 2270 { 2271 enum pfsync_q_id q; 2272 2273 PFSYNC_BUCKET_LOCK_ASSERT(b); 2274 KASSERT(st->sync_state != PFSYNC_S_NONE, 2275 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 2276 2277 q = pfsync_sstate_to_qid(st->sync_state); 2278 b->b_len -= pfsync_qs[q].len; 2279 TAILQ_REMOVE(&b->b_qs[q], st, sync_list); 2280 st->sync_state = PFSYNC_S_NONE; 2281 if (unref) 2282 pf_release_state(st); 2283 2284 if (TAILQ_EMPTY(&b->b_qs[q])) 2285 b->b_len -= sizeof(struct pfsync_subheader); 2286 } 2287 2288 static void 2289 pfsync_bulk_start(void) 2290 { 2291 struct pfsync_softc *sc = V_pfsyncif; 2292 2293 if (V_pf_status.debug >= PF_DEBUG_MISC) 2294 printf("pfsync: received bulk update request\n"); 2295 2296 PFSYNC_BLOCK(sc); 2297 2298 sc->sc_ureq_received = time_uptime; 2299 sc->sc_bulk_hashid = 0; 2300 sc->sc_bulk_stateid = 0; 2301 pfsync_bulk_status(PFSYNC_BUS_START); 2302 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2303 PFSYNC_BUNLOCK(sc); 2304 } 2305 2306 static void 2307 pfsync_bulk_update(void *arg) 2308 { 2309 struct pfsync_softc *sc = arg; 2310 struct pf_kstate *s; 2311 int i; 2312 2313 PFSYNC_BLOCK_ASSERT(sc); 2314 CURVNET_SET(sc->sc_ifp->if_vnet); 2315 2316 /* 2317 * Start with last state from previous invocation. 2318 * It may had gone, in this case start from the 2319 * hash slot. 2320 */ 2321 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2322 2323 if (s != NULL) 2324 i = PF_IDHASH(s); 2325 else 2326 i = sc->sc_bulk_hashid; 2327 2328 for (; i <= pf_hashmask; i++) { 2329 struct pf_idhash *ih = &V_pf_idhash[i]; 2330 2331 if (s != NULL) 2332 PF_HASHROW_ASSERT(ih); 2333 else { 2334 PF_HASHROW_LOCK(ih); 2335 s = LIST_FIRST(&ih->states); 2336 } 2337 2338 for (; s; s = LIST_NEXT(s, entry)) { 2339 if (s->sync_state == PFSYNC_S_NONE && 2340 s->timeout < PFTM_MAX && 2341 s->pfsync_time <= sc->sc_ureq_received) { 2342 if (pfsync_update_state_req(s)) { 2343 /* We've filled a packet. */ 2344 sc->sc_bulk_hashid = i; 2345 sc->sc_bulk_stateid = s->id; 2346 sc->sc_bulk_creatorid = s->creatorid; 2347 PF_HASHROW_UNLOCK(ih); 2348 callout_reset(&sc->sc_bulk_tmo, 1, 2349 pfsync_bulk_update, sc); 2350 goto full; 2351 } 2352 } 2353 } 2354 PF_HASHROW_UNLOCK(ih); 2355 } 2356 2357 /* We're done. */ 2358 pfsync_bulk_status(PFSYNC_BUS_END); 2359 full: 2360 CURVNET_RESTORE(); 2361 } 2362 2363 static void 2364 pfsync_bulk_status(u_int8_t status) 2365 { 2366 struct { 2367 struct pfsync_subheader subh; 2368 struct pfsync_bus bus; 2369 } __packed r; 2370 2371 struct pfsync_softc *sc = V_pfsyncif; 2372 2373 bzero(&r, sizeof(r)); 2374 2375 r.subh.action = PFSYNC_ACT_BUS; 2376 r.subh.count = htons(1); 2377 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2378 2379 r.bus.creatorid = V_pf_status.hostid; 2380 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2381 r.bus.status = status; 2382 2383 pfsync_send_plus(&r, sizeof(r)); 2384 } 2385 2386 static void 2387 pfsync_bulk_fail(void *arg) 2388 { 2389 struct pfsync_softc *sc = arg; 2390 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2391 2392 CURVNET_SET(sc->sc_ifp->if_vnet); 2393 2394 PFSYNC_BLOCK_ASSERT(sc); 2395 2396 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2397 /* Try again */ 2398 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2399 pfsync_bulk_fail, V_pfsyncif); 2400 PFSYNC_BUCKET_LOCK(b); 2401 pfsync_request_update(0, 0); 2402 PFSYNC_BUCKET_UNLOCK(b); 2403 } else { 2404 /* Pretend like the transfer was ok. */ 2405 sc->sc_ureq_sent = 0; 2406 sc->sc_bulk_tries = 0; 2407 PFSYNC_LOCK(sc); 2408 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2409 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2410 "pfsync bulk fail"); 2411 sc->sc_flags |= PFSYNCF_OK; 2412 PFSYNC_UNLOCK(sc); 2413 if (V_pf_status.debug >= PF_DEBUG_MISC) 2414 printf("pfsync: failed to receive bulk update\n"); 2415 } 2416 2417 CURVNET_RESTORE(); 2418 } 2419 2420 static void 2421 pfsync_send_plus(void *plus, size_t pluslen) 2422 { 2423 struct pfsync_softc *sc = V_pfsyncif; 2424 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2425 2426 PFSYNC_BUCKET_LOCK(b); 2427 2428 if (b->b_len + pluslen > sc->sc_ifp->if_mtu) 2429 pfsync_sendout(1, b->b_id); 2430 2431 b->b_plus = plus; 2432 b->b_len += (b->b_pluslen = pluslen); 2433 2434 pfsync_sendout(1, b->b_id); 2435 PFSYNC_BUCKET_UNLOCK(b); 2436 } 2437 2438 static void 2439 pfsync_timeout(void *arg) 2440 { 2441 struct pfsync_bucket *b = arg; 2442 2443 CURVNET_SET(b->b_sc->sc_ifp->if_vnet); 2444 PFSYNC_BUCKET_LOCK(b); 2445 pfsync_push(b); 2446 PFSYNC_BUCKET_UNLOCK(b); 2447 CURVNET_RESTORE(); 2448 } 2449 2450 static void 2451 pfsync_push(struct pfsync_bucket *b) 2452 { 2453 2454 PFSYNC_BUCKET_LOCK_ASSERT(b); 2455 2456 b->b_flags |= PFSYNCF_BUCKET_PUSH; 2457 swi_sched(V_pfsync_swi_cookie, 0); 2458 } 2459 2460 static void 2461 pfsync_push_all(struct pfsync_softc *sc) 2462 { 2463 int c; 2464 struct pfsync_bucket *b; 2465 2466 for (c = 0; c < pfsync_buckets; c++) { 2467 b = &sc->sc_buckets[c]; 2468 2469 PFSYNC_BUCKET_LOCK(b); 2470 pfsync_push(b); 2471 PFSYNC_BUCKET_UNLOCK(b); 2472 } 2473 } 2474 2475 static void 2476 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m) 2477 { 2478 struct ip *ip; 2479 int af, error = 0; 2480 2481 ip = mtod(m, struct ip *); 2482 MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4)); 2483 2484 af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6; 2485 2486 /* 2487 * We distinguish between a deferral packet and our 2488 * own pfsync packet based on M_SKIP_FIREWALL 2489 * flag. This is XXX. 2490 */ 2491 switch (af) { 2492 #ifdef INET 2493 case AF_INET: 2494 if (m->m_flags & M_SKIP_FIREWALL) { 2495 error = ip_output(m, NULL, NULL, 0, 2496 NULL, NULL); 2497 } else { 2498 error = ip_output(m, NULL, NULL, 2499 IP_RAWOUTPUT, &sc->sc_imo, NULL); 2500 } 2501 break; 2502 #endif 2503 #ifdef INET6 2504 case AF_INET6: 2505 if (m->m_flags & M_SKIP_FIREWALL) { 2506 error = ip6_output(m, NULL, NULL, 0, 2507 NULL, NULL, NULL); 2508 } else { 2509 MPASS(false); 2510 /* We don't support pfsync over IPv6. */ 2511 /*error = ip6_output(m, NULL, NULL, 2512 IP_RAWOUTPUT, &sc->sc_imo6, NULL);*/ 2513 } 2514 break; 2515 #endif 2516 } 2517 2518 if (error == 0) 2519 V_pfsyncstats.pfsyncs_opackets++; 2520 else 2521 V_pfsyncstats.pfsyncs_oerrors++; 2522 2523 } 2524 2525 static void 2526 pfsyncintr(void *arg) 2527 { 2528 struct epoch_tracker et; 2529 struct pfsync_softc *sc = arg; 2530 struct pfsync_bucket *b; 2531 struct mbuf *m, *n; 2532 int c; 2533 2534 NET_EPOCH_ENTER(et); 2535 CURVNET_SET(sc->sc_ifp->if_vnet); 2536 2537 for (c = 0; c < pfsync_buckets; c++) { 2538 b = &sc->sc_buckets[c]; 2539 2540 PFSYNC_BUCKET_LOCK(b); 2541 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) { 2542 pfsync_sendout(0, b->b_id); 2543 b->b_flags &= ~PFSYNCF_BUCKET_PUSH; 2544 } 2545 _IF_DEQUEUE_ALL(&b->b_snd, m); 2546 PFSYNC_BUCKET_UNLOCK(b); 2547 2548 for (; m != NULL; m = n) { 2549 n = m->m_nextpkt; 2550 m->m_nextpkt = NULL; 2551 2552 pfsync_tx(sc, m); 2553 } 2554 } 2555 CURVNET_RESTORE(); 2556 NET_EPOCH_EXIT(et); 2557 } 2558 2559 static int 2560 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, 2561 struct in_mfilter *imf) 2562 { 2563 struct ip_moptions *imo = &sc->sc_imo; 2564 int error; 2565 2566 if (!(ifp->if_flags & IFF_MULTICAST)) 2567 return (EADDRNOTAVAIL); 2568 2569 switch (sc->sc_sync_peer.ss_family) { 2570 #ifdef INET 2571 case AF_INET: 2572 { 2573 ip_mfilter_init(&imo->imo_head); 2574 imo->imo_multicast_vif = -1; 2575 if ((error = in_joingroup(ifp, &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL, 2576 &imf->imf_inm)) != 0) 2577 return (error); 2578 2579 ip_mfilter_insert(&imo->imo_head, imf); 2580 imo->imo_multicast_ifp = ifp; 2581 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2582 imo->imo_multicast_loop = 0; 2583 break; 2584 } 2585 #endif 2586 } 2587 2588 return (0); 2589 } 2590 2591 static void 2592 pfsync_multicast_cleanup(struct pfsync_softc *sc) 2593 { 2594 struct ip_moptions *imo = &sc->sc_imo; 2595 struct in_mfilter *imf; 2596 2597 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 2598 ip_mfilter_remove(&imo->imo_head, imf); 2599 in_leavegroup(imf->imf_inm, NULL); 2600 ip_mfilter_free(imf); 2601 } 2602 imo->imo_multicast_ifp = NULL; 2603 } 2604 2605 void 2606 pfsync_detach_ifnet(struct ifnet *ifp) 2607 { 2608 struct pfsync_softc *sc = V_pfsyncif; 2609 2610 if (sc == NULL) 2611 return; 2612 2613 PFSYNC_LOCK(sc); 2614 2615 if (sc->sc_sync_if == ifp) { 2616 /* We don't need mutlicast cleanup here, because the interface 2617 * is going away. We do need to ensure we don't try to do 2618 * cleanup later. 2619 */ 2620 ip_mfilter_init(&sc->sc_imo.imo_head); 2621 sc->sc_imo.imo_multicast_ifp = NULL; 2622 sc->sc_sync_if = NULL; 2623 } 2624 2625 PFSYNC_UNLOCK(sc); 2626 } 2627 2628 static int 2629 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status) 2630 { 2631 struct sockaddr_storage sa; 2632 status->maxupdates = pfsyncr->pfsyncr_maxupdates; 2633 status->flags = pfsyncr->pfsyncr_defer; 2634 2635 strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ); 2636 2637 memset(&sa, 0, sizeof(sa)); 2638 if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) { 2639 struct sockaddr_in *in = (struct sockaddr_in *)&sa; 2640 in->sin_family = AF_INET; 2641 in->sin_len = sizeof(*in); 2642 in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr; 2643 } 2644 status->syncpeer = sa; 2645 2646 return 0; 2647 } 2648 2649 static int 2650 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc) 2651 { 2652 struct in_mfilter *imf = NULL; 2653 struct ifnet *sifp; 2654 struct ip *ip; 2655 int error; 2656 int c; 2657 2658 if ((status->maxupdates < 0) || (status->maxupdates > 255)) 2659 return (EINVAL); 2660 2661 if (status->syncdev[0] == '\0') 2662 sifp = NULL; 2663 else if ((sifp = ifunit_ref(status->syncdev)) == NULL) 2664 return (EINVAL); 2665 2666 struct sockaddr_in *status_sin = 2667 (struct sockaddr_in *)&(status->syncpeer); 2668 if (sifp != NULL && (status_sin->sin_addr.s_addr == 0 || 2669 status_sin->sin_addr.s_addr == 2670 htonl(INADDR_PFSYNC_GROUP))) 2671 imf = ip_mfilter_alloc(M_WAITOK, 0, 0); 2672 2673 PFSYNC_LOCK(sc); 2674 2675 switch (status->version) { 2676 case PFSYNC_MSG_VERSION_UNSPECIFIED: 2677 sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT; 2678 break; 2679 case PFSYNC_MSG_VERSION_1301: 2680 case PFSYNC_MSG_VERSION_1400: 2681 sc->sc_version = status->version; 2682 break; 2683 default: 2684 PFSYNC_UNLOCK(sc); 2685 return (EINVAL); 2686 } 2687 2688 struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer; 2689 sc_sin->sin_family = AF_INET; 2690 sc_sin->sin_len = sizeof(*sc_sin); 2691 if (status_sin->sin_addr.s_addr == 0) { 2692 sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP); 2693 } else { 2694 sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr; 2695 } 2696 2697 sc->sc_maxupdates = status->maxupdates; 2698 if (status->flags & PFSYNCF_DEFER) { 2699 sc->sc_flags |= PFSYNCF_DEFER; 2700 V_pfsync_defer_ptr = pfsync_defer; 2701 } else { 2702 sc->sc_flags &= ~PFSYNCF_DEFER; 2703 V_pfsync_defer_ptr = NULL; 2704 } 2705 2706 if (sifp == NULL) { 2707 if (sc->sc_sync_if) 2708 if_rele(sc->sc_sync_if); 2709 sc->sc_sync_if = NULL; 2710 pfsync_multicast_cleanup(sc); 2711 PFSYNC_UNLOCK(sc); 2712 return (0); 2713 } 2714 2715 for (c = 0; c < pfsync_buckets; c++) { 2716 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 2717 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT && 2718 (sifp->if_mtu < sc->sc_ifp->if_mtu || 2719 (sc->sc_sync_if != NULL && 2720 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 2721 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 2722 pfsync_sendout(1, c); 2723 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 2724 } 2725 2726 pfsync_multicast_cleanup(sc); 2727 2728 if (sc_sin->sin_addr.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 2729 error = pfsync_multicast_setup(sc, sifp, imf); 2730 if (error) { 2731 if_rele(sifp); 2732 ip_mfilter_free(imf); 2733 PFSYNC_UNLOCK(sc); 2734 return (error); 2735 } 2736 } 2737 if (sc->sc_sync_if) 2738 if_rele(sc->sc_sync_if); 2739 sc->sc_sync_if = sifp; 2740 2741 ip = &sc->sc_template.ipv4; 2742 bzero(ip, sizeof(*ip)); 2743 ip->ip_v = IPVERSION; 2744 ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2; 2745 ip->ip_tos = IPTOS_LOWDELAY; 2746 /* len and id are set later. */ 2747 ip->ip_off = htons(IP_DF); 2748 ip->ip_ttl = PFSYNC_DFLTTL; 2749 ip->ip_p = IPPROTO_PFSYNC; 2750 ip->ip_src.s_addr = INADDR_ANY; 2751 ip->ip_dst.s_addr = sc_sin->sin_addr.s_addr; 2752 2753 /* Request a full state table update. */ 2754 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2755 (*carp_demote_adj_p)(V_pfsync_carp_adj, 2756 "pfsync bulk start"); 2757 sc->sc_flags &= ~PFSYNCF_OK; 2758 if (V_pf_status.debug >= PF_DEBUG_MISC) 2759 printf("pfsync: requesting bulk update\n"); 2760 PFSYNC_UNLOCK(sc); 2761 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 2762 pfsync_request_update(0, 0); 2763 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 2764 PFSYNC_BLOCK(sc); 2765 sc->sc_ureq_sent = time_uptime; 2766 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc); 2767 PFSYNC_BUNLOCK(sc); 2768 return (0); 2769 } 2770 2771 static void 2772 pfsync_pointers_init(void) 2773 { 2774 2775 PF_RULES_WLOCK(); 2776 V_pfsync_state_import_ptr = pfsync_state_import; 2777 V_pfsync_insert_state_ptr = pfsync_insert_state; 2778 V_pfsync_update_state_ptr = pfsync_update_state; 2779 V_pfsync_delete_state_ptr = pfsync_delete_state; 2780 V_pfsync_clear_states_ptr = pfsync_clear_states; 2781 V_pfsync_defer_ptr = pfsync_defer; 2782 PF_RULES_WUNLOCK(); 2783 } 2784 2785 static void 2786 pfsync_pointers_uninit(void) 2787 { 2788 2789 PF_RULES_WLOCK(); 2790 V_pfsync_state_import_ptr = NULL; 2791 V_pfsync_insert_state_ptr = NULL; 2792 V_pfsync_update_state_ptr = NULL; 2793 V_pfsync_delete_state_ptr = NULL; 2794 V_pfsync_clear_states_ptr = NULL; 2795 V_pfsync_defer_ptr = NULL; 2796 PF_RULES_WUNLOCK(); 2797 } 2798 2799 static void 2800 vnet_pfsync_init(const void *unused __unused) 2801 { 2802 int error; 2803 2804 V_pfsync_cloner = if_clone_simple(pfsyncname, 2805 pfsync_clone_create, pfsync_clone_destroy, 1); 2806 error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif, 2807 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2808 if (error) { 2809 if_clone_detach(V_pfsync_cloner); 2810 log(LOG_INFO, "swi_add() failed in %s\n", __func__); 2811 } 2812 2813 pfsync_pointers_init(); 2814 } 2815 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY, 2816 vnet_pfsync_init, NULL); 2817 2818 static void 2819 vnet_pfsync_uninit(const void *unused __unused) 2820 { 2821 int ret __diagused; 2822 2823 pfsync_pointers_uninit(); 2824 2825 if_clone_detach(V_pfsync_cloner); 2826 ret = swi_remove(V_pfsync_swi_cookie); 2827 MPASS(ret == 0); 2828 ret = intr_event_destroy(V_pfsync_swi_ie); 2829 MPASS(ret == 0); 2830 } 2831 2832 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH, 2833 vnet_pfsync_uninit, NULL); 2834 2835 static int 2836 pfsync_init(void) 2837 { 2838 #ifdef INET 2839 int error; 2840 2841 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet; 2842 2843 error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL); 2844 if (error) 2845 return (error); 2846 #endif 2847 2848 return (0); 2849 } 2850 2851 static void 2852 pfsync_uninit(void) 2853 { 2854 pfsync_detach_ifnet_ptr = NULL; 2855 2856 #ifdef INET 2857 ipproto_unregister(IPPROTO_PFSYNC); 2858 #endif 2859 } 2860 2861 static int 2862 pfsync_modevent(module_t mod, int type, void *data) 2863 { 2864 int error = 0; 2865 2866 switch (type) { 2867 case MOD_LOAD: 2868 error = pfsync_init(); 2869 break; 2870 case MOD_UNLOAD: 2871 pfsync_uninit(); 2872 break; 2873 default: 2874 error = EINVAL; 2875 break; 2876 } 2877 2878 return (error); 2879 } 2880 2881 static moduledata_t pfsync_mod = { 2882 pfsyncname, 2883 pfsync_modevent, 2884 0 2885 }; 2886 2887 #define PFSYNC_MODVER 1 2888 2889 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */ 2890 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 2891 MODULE_VERSION(pfsync, PFSYNC_MODVER); 2892 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); 2893