1 /* $OpenBSD: session.c,v 1.427 2022/02/23 11:20:35 claudio Exp $ */ 2 3 /* 4 * Copyright (c) 2003, 2004, 2005 Henning Brauer <henning@openbsd.org> 5 * Copyright (c) 2017 Peter van Dijk <peter.van.dijk@powerdns.com> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include <sys/types.h> 21 22 #include <sys/mman.h> 23 #include <sys/socket.h> 24 #include <sys/time.h> 25 #include <sys/resource.h> 26 #include <sys/un.h> 27 #include <netinet/in.h> 28 #include <netinet/ip.h> 29 #include <netinet/tcp.h> 30 #include <arpa/inet.h> 31 #include <limits.h> 32 33 #include <err.h> 34 #include <errno.h> 35 #include <fcntl.h> 36 #include <ifaddrs.h> 37 #include <poll.h> 38 #include <pwd.h> 39 #include <signal.h> 40 #include <stdio.h> 41 #include <stdlib.h> 42 #include <string.h> 43 #include <syslog.h> 44 #include <unistd.h> 45 46 #include "bgpd.h" 47 #include "session.h" 48 #include "log.h" 49 50 #define PFD_PIPE_MAIN 0 51 #define PFD_PIPE_ROUTE 1 52 #define PFD_PIPE_ROUTE_CTL 2 53 #define PFD_SOCK_CTL 3 54 #define PFD_SOCK_RCTL 4 55 #define PFD_LISTENERS_START 5 56 57 void session_sighdlr(int); 58 int setup_listeners(u_int *); 59 void init_peer(struct peer *); 60 void start_timer_holdtime(struct peer *); 61 void start_timer_keepalive(struct peer *); 62 void session_close_connection(struct peer *); 63 void change_state(struct peer *, enum session_state, enum session_events); 64 int session_setup_socket(struct peer *); 65 void session_accept(int); 66 int session_connect(struct peer *); 67 void session_tcp_established(struct peer *); 68 void session_capa_ann_none(struct peer *); 69 int session_capa_add(struct ibuf *, uint8_t, uint8_t); 70 int session_capa_add_mp(struct ibuf *, uint8_t); 71 int session_capa_add_afi(struct peer *, struct ibuf *, uint8_t, uint8_t); 72 struct bgp_msg *session_newmsg(enum msg_type, uint16_t); 73 int session_sendmsg(struct bgp_msg *, struct peer *); 74 void session_open(struct peer *); 75 void session_keepalive(struct peer *); 76 void session_update(uint32_t, void *, size_t); 77 void session_notification(struct peer *, uint8_t, uint8_t, void *, 78 ssize_t); 79 void session_rrefresh(struct peer *, uint8_t, uint8_t); 80 int session_graceful_restart(struct peer *); 81 int session_graceful_stop(struct peer *); 82 int session_dispatch_msg(struct pollfd *, struct peer *); 83 void session_process_msg(struct peer *); 84 int parse_header(struct peer *, u_char *, uint16_t *, uint8_t *); 85 int parse_open(struct peer *); 86 int parse_update(struct peer *); 87 int parse_rrefresh(struct peer *); 88 int parse_notification(struct peer *); 89 int parse_capabilities(struct peer *, u_char *, uint16_t, uint32_t *); 90 int capa_neg_calc(struct peer *); 91 void session_dispatch_imsg(struct imsgbuf *, int, u_int *); 92 void session_up(struct peer *); 93 void session_down(struct peer *); 94 int imsg_rde(int, uint32_t, void *, uint16_t); 95 void session_demote(struct peer *, int); 96 void merge_peers(struct bgpd_config *, struct bgpd_config *); 97 98 int la_cmp(struct listen_addr *, struct listen_addr *); 99 void session_template_clone(struct peer *, struct sockaddr *, 100 uint32_t, uint32_t); 101 int session_match_mask(struct peer *, struct bgpd_addr *); 102 103 static struct bgpd_config *conf, *nconf; 104 static struct imsgbuf *ibuf_rde; 105 static struct imsgbuf *ibuf_rde_ctl; 106 static struct imsgbuf *ibuf_main; 107 108 struct bgpd_sysdep sysdep; 109 volatile sig_atomic_t session_quit; 110 int pending_reconf; 111 int csock = -1, rcsock = -1; 112 u_int peer_cnt; 113 114 struct mrt_head mrthead; 115 time_t pauseaccept; 116 117 static inline int 118 peer_compare(const struct peer *a, const struct peer *b) 119 { 120 return a->conf.id - b->conf.id; 121 } 122 123 RB_GENERATE(peer_head, peer, entry, peer_compare); 124 125 void 126 session_sighdlr(int sig) 127 { 128 switch (sig) { 129 case SIGINT: 130 case SIGTERM: 131 session_quit = 1; 132 break; 133 } 134 } 135 136 int 137 setup_listeners(u_int *la_cnt) 138 { 139 int ttl = 255; 140 struct listen_addr *la; 141 u_int cnt = 0; 142 143 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 144 la->reconf = RECONF_NONE; 145 cnt++; 146 147 if (la->flags & LISTENER_LISTENING) 148 continue; 149 150 if (la->fd == -1) { 151 log_warn("cannot establish listener on %s: invalid fd", 152 log_sockaddr((struct sockaddr *)&la->sa, 153 la->sa_len)); 154 continue; 155 } 156 157 if (tcp_md5_prep_listener(la, &conf->peers) == -1) 158 fatal("tcp_md5_prep_listener"); 159 160 /* set ttl to 255 so that ttl-security works */ 161 if (la->sa.ss_family == AF_INET && setsockopt(la->fd, 162 IPPROTO_IP, IP_TTL, &ttl, sizeof(ttl)) == -1) { 163 log_warn("setup_listeners setsockopt TTL"); 164 continue; 165 } 166 if (la->sa.ss_family == AF_INET6 && setsockopt(la->fd, 167 IPPROTO_IPV6, IPV6_UNICAST_HOPS, &ttl, sizeof(ttl)) == -1) { 168 log_warn("setup_listeners setsockopt hoplimit"); 169 continue; 170 } 171 172 if (listen(la->fd, MAX_BACKLOG)) { 173 close(la->fd); 174 fatal("listen"); 175 } 176 177 la->flags |= LISTENER_LISTENING; 178 179 log_info("listening on %s", 180 log_sockaddr((struct sockaddr *)&la->sa, la->sa_len)); 181 } 182 183 *la_cnt = cnt; 184 185 return (0); 186 } 187 188 void 189 session_main(int debug, int verbose) 190 { 191 int timeout; 192 unsigned int i, j, idx_peers, idx_listeners, idx_mrts; 193 u_int pfd_elms = 0, peer_l_elms = 0, mrt_l_elms = 0; 194 u_int listener_cnt, ctl_cnt, mrt_cnt; 195 u_int new_cnt; 196 struct passwd *pw; 197 struct peer *p, **peer_l = NULL, *next; 198 struct mrt *m, *xm, **mrt_l = NULL; 199 struct pollfd *pfd = NULL; 200 struct listen_addr *la; 201 void *newp; 202 time_t now; 203 short events; 204 205 log_init(debug, LOG_DAEMON); 206 log_setverbose(verbose); 207 208 log_procinit(log_procnames[PROC_SE]); 209 210 if ((pw = getpwnam(BGPD_USER)) == NULL) 211 fatal(NULL); 212 213 if (chroot(pw->pw_dir) == -1) 214 fatal("chroot"); 215 if (chdir("/") == -1) 216 fatal("chdir(\"/\")"); 217 218 setproctitle("session engine"); 219 220 if (setgroups(1, &pw->pw_gid) || 221 setresgid(pw->pw_gid, pw->pw_gid, pw->pw_gid) || 222 setresuid(pw->pw_uid, pw->pw_uid, pw->pw_uid)) 223 fatal("can't drop privileges"); 224 225 if (pledge("stdio inet recvfd", NULL) == -1) 226 fatal("pledge"); 227 228 signal(SIGTERM, session_sighdlr); 229 signal(SIGINT, session_sighdlr); 230 signal(SIGPIPE, SIG_IGN); 231 signal(SIGHUP, SIG_IGN); 232 signal(SIGALRM, SIG_IGN); 233 signal(SIGUSR1, SIG_IGN); 234 235 if ((ibuf_main = malloc(sizeof(struct imsgbuf))) == NULL) 236 fatal(NULL); 237 imsg_init(ibuf_main, 3); 238 239 LIST_INIT(&mrthead); 240 listener_cnt = 0; 241 peer_cnt = 0; 242 ctl_cnt = 0; 243 244 conf = new_config(); 245 log_info("session engine ready"); 246 247 while (session_quit == 0) { 248 /* check for peers to be initialized or deleted */ 249 if (!pending_reconf) { 250 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 251 /* cloned peer that idled out? */ 252 if (p->template && (p->state == STATE_IDLE || 253 p->state == STATE_ACTIVE) && 254 getmonotime() - p->stats.last_updown >= 255 INTERVAL_HOLD_CLONED) 256 p->reconf_action = RECONF_DELETE; 257 258 /* new peer that needs init? */ 259 if (p->state == STATE_NONE) 260 init_peer(p); 261 262 /* reinit due? */ 263 if (p->reconf_action == RECONF_REINIT) { 264 session_stop(p, ERR_CEASE_ADMIN_RESET); 265 if (!p->conf.down) 266 timer_set(&p->timers, 267 Timer_IdleHold, 0); 268 } 269 270 /* deletion due? */ 271 if (p->reconf_action == RECONF_DELETE) { 272 if (p->demoted) 273 session_demote(p, -1); 274 p->conf.demote_group[0] = 0; 275 session_stop(p, ERR_CEASE_PEER_UNCONF); 276 timer_remove_all(&p->timers); 277 tcp_md5_del_listener(conf, p); 278 log_peer_warnx(&p->conf, "removed"); 279 RB_REMOVE(peer_head, &conf->peers, p); 280 free(p); 281 peer_cnt--; 282 continue; 283 } 284 p->reconf_action = RECONF_NONE; 285 } 286 } 287 288 if (peer_cnt > peer_l_elms) { 289 if ((newp = reallocarray(peer_l, peer_cnt, 290 sizeof(struct peer *))) == NULL) { 291 /* panic for now */ 292 log_warn("could not resize peer_l from %u -> %u" 293 " entries", peer_l_elms, peer_cnt); 294 fatalx("exiting"); 295 } 296 peer_l = newp; 297 peer_l_elms = peer_cnt; 298 } 299 300 mrt_cnt = 0; 301 for (m = LIST_FIRST(&mrthead); m != NULL; m = xm) { 302 xm = LIST_NEXT(m, entry); 303 if (m->state == MRT_STATE_REMOVE) { 304 mrt_clean(m); 305 LIST_REMOVE(m, entry); 306 free(m); 307 continue; 308 } 309 if (m->wbuf.queued) 310 mrt_cnt++; 311 } 312 313 if (mrt_cnt > mrt_l_elms) { 314 if ((newp = reallocarray(mrt_l, mrt_cnt, 315 sizeof(struct mrt *))) == NULL) { 316 /* panic for now */ 317 log_warn("could not resize mrt_l from %u -> %u" 318 " entries", mrt_l_elms, mrt_cnt); 319 fatalx("exiting"); 320 } 321 mrt_l = newp; 322 mrt_l_elms = mrt_cnt; 323 } 324 325 new_cnt = PFD_LISTENERS_START + listener_cnt + peer_cnt + 326 ctl_cnt + mrt_cnt; 327 if (new_cnt > pfd_elms) { 328 if ((newp = reallocarray(pfd, new_cnt, 329 sizeof(struct pollfd))) == NULL) { 330 /* panic for now */ 331 log_warn("could not resize pfd from %u -> %u" 332 " entries", pfd_elms, new_cnt); 333 fatalx("exiting"); 334 } 335 pfd = newp; 336 pfd_elms = new_cnt; 337 } 338 339 bzero(pfd, sizeof(struct pollfd) * pfd_elms); 340 341 set_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main); 342 set_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde); 343 set_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl); 344 345 if (pauseaccept == 0) { 346 pfd[PFD_SOCK_CTL].fd = csock; 347 pfd[PFD_SOCK_CTL].events = POLLIN; 348 pfd[PFD_SOCK_RCTL].fd = rcsock; 349 pfd[PFD_SOCK_RCTL].events = POLLIN; 350 } else { 351 pfd[PFD_SOCK_CTL].fd = -1; 352 pfd[PFD_SOCK_RCTL].fd = -1; 353 } 354 355 i = PFD_LISTENERS_START; 356 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 357 if (pauseaccept == 0) { 358 pfd[i].fd = la->fd; 359 pfd[i].events = POLLIN; 360 } else 361 pfd[i].fd = -1; 362 i++; 363 } 364 idx_listeners = i; 365 timeout = 240; /* loop every 240s at least */ 366 367 now = getmonotime(); 368 RB_FOREACH(p, peer_head, &conf->peers) { 369 time_t nextaction; 370 struct timer *pt; 371 372 /* check timers */ 373 if ((pt = timer_nextisdue(&p->timers, now)) != NULL) { 374 switch (pt->type) { 375 case Timer_Hold: 376 bgp_fsm(p, EVNT_TIMER_HOLDTIME); 377 break; 378 case Timer_SendHold: 379 bgp_fsm(p, EVNT_TIMER_SENDHOLD); 380 break; 381 case Timer_ConnectRetry: 382 bgp_fsm(p, EVNT_TIMER_CONNRETRY); 383 break; 384 case Timer_Keepalive: 385 bgp_fsm(p, EVNT_TIMER_KEEPALIVE); 386 break; 387 case Timer_IdleHold: 388 bgp_fsm(p, EVNT_START); 389 break; 390 case Timer_IdleHoldReset: 391 p->IdleHoldTime = 392 INTERVAL_IDLE_HOLD_INITIAL; 393 p->errcnt = 0; 394 timer_stop(&p->timers, 395 Timer_IdleHoldReset); 396 break; 397 case Timer_CarpUndemote: 398 timer_stop(&p->timers, 399 Timer_CarpUndemote); 400 if (p->demoted && 401 p->state == STATE_ESTABLISHED) 402 session_demote(p, -1); 403 break; 404 case Timer_RestartTimeout: 405 timer_stop(&p->timers, 406 Timer_RestartTimeout); 407 session_graceful_stop(p); 408 break; 409 default: 410 fatalx("King Bula lost in time"); 411 } 412 } 413 if ((nextaction = timer_nextduein(&p->timers, 414 now)) != -1 && nextaction < timeout) 415 timeout = nextaction; 416 417 /* are we waiting for a write? */ 418 events = POLLIN; 419 if (p->wbuf.queued > 0 || p->state == STATE_CONNECT) 420 events |= POLLOUT; 421 /* is there still work to do? */ 422 if (p->rpending && p->rbuf && p->rbuf->wpos) 423 timeout = 0; 424 425 /* poll events */ 426 if (p->fd != -1 && events != 0) { 427 pfd[i].fd = p->fd; 428 pfd[i].events = events; 429 peer_l[i - idx_listeners] = p; 430 i++; 431 } 432 } 433 434 idx_peers = i; 435 436 LIST_FOREACH(m, &mrthead, entry) 437 if (m->wbuf.queued) { 438 pfd[i].fd = m->wbuf.fd; 439 pfd[i].events = POLLOUT; 440 mrt_l[i - idx_peers] = m; 441 i++; 442 } 443 444 idx_mrts = i; 445 446 i += control_fill_pfds(pfd + i, pfd_elms -i); 447 448 if (i > pfd_elms) 449 fatalx("poll pfd overflow"); 450 451 if (pauseaccept && timeout > 1) 452 timeout = 1; 453 if (timeout < 0) 454 timeout = 0; 455 if (poll(pfd, i, timeout * 1000) == -1) { 456 if (errno == EINTR) 457 continue; 458 fatal("poll error"); 459 } 460 461 /* 462 * If we previously saw fd exhaustion, we stop accept() 463 * for 1 second to throttle the accept() loop. 464 */ 465 if (pauseaccept && getmonotime() > pauseaccept + 1) 466 pauseaccept = 0; 467 468 if (handle_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main) == -1) { 469 log_warnx("SE: Lost connection to parent"); 470 session_quit = 1; 471 continue; 472 } else 473 session_dispatch_imsg(ibuf_main, PFD_PIPE_MAIN, 474 &listener_cnt); 475 476 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde) == -1) { 477 log_warnx("SE: Lost connection to RDE"); 478 msgbuf_clear(&ibuf_rde->w); 479 free(ibuf_rde); 480 ibuf_rde = NULL; 481 } else 482 session_dispatch_imsg(ibuf_rde, PFD_PIPE_ROUTE, 483 &listener_cnt); 484 485 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl) == 486 -1) { 487 log_warnx("SE: Lost connection to RDE control"); 488 msgbuf_clear(&ibuf_rde_ctl->w); 489 free(ibuf_rde_ctl); 490 ibuf_rde_ctl = NULL; 491 } else 492 session_dispatch_imsg(ibuf_rde_ctl, PFD_PIPE_ROUTE_CTL, 493 &listener_cnt); 494 495 if (pfd[PFD_SOCK_CTL].revents & POLLIN) 496 ctl_cnt += control_accept(csock, 0); 497 498 if (pfd[PFD_SOCK_RCTL].revents & POLLIN) 499 ctl_cnt += control_accept(rcsock, 1); 500 501 for (j = PFD_LISTENERS_START; j < idx_listeners; j++) 502 if (pfd[j].revents & POLLIN) 503 session_accept(pfd[j].fd); 504 505 for (; j < idx_peers; j++) 506 session_dispatch_msg(&pfd[j], 507 peer_l[j - idx_listeners]); 508 509 RB_FOREACH(p, peer_head, &conf->peers) 510 if (p->rbuf && p->rbuf->wpos) 511 session_process_msg(p); 512 513 for (; j < idx_mrts; j++) 514 if (pfd[j].revents & POLLOUT) 515 mrt_write(mrt_l[j - idx_peers]); 516 517 for (; j < i; j++) 518 ctl_cnt -= control_dispatch_msg(&pfd[j], &conf->peers); 519 } 520 521 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 522 RB_REMOVE(peer_head, &conf->peers, p); 523 strlcpy(p->conf.reason, 524 "bgpd shutting down", 525 sizeof(p->conf.reason)); 526 session_stop(p, ERR_CEASE_ADMIN_DOWN); 527 timer_remove_all(&p->timers); 528 free(p); 529 } 530 531 while ((m = LIST_FIRST(&mrthead)) != NULL) { 532 mrt_clean(m); 533 LIST_REMOVE(m, entry); 534 free(m); 535 } 536 537 free_config(conf); 538 free(peer_l); 539 free(mrt_l); 540 free(pfd); 541 542 /* close pipes */ 543 if (ibuf_rde) { 544 msgbuf_write(&ibuf_rde->w); 545 msgbuf_clear(&ibuf_rde->w); 546 close(ibuf_rde->fd); 547 free(ibuf_rde); 548 } 549 if (ibuf_rde_ctl) { 550 msgbuf_clear(&ibuf_rde_ctl->w); 551 close(ibuf_rde_ctl->fd); 552 free(ibuf_rde_ctl); 553 } 554 msgbuf_write(&ibuf_main->w); 555 msgbuf_clear(&ibuf_main->w); 556 close(ibuf_main->fd); 557 free(ibuf_main); 558 559 control_shutdown(csock); 560 control_shutdown(rcsock); 561 log_info("session engine exiting"); 562 exit(0); 563 } 564 565 void 566 init_peer(struct peer *p) 567 { 568 TAILQ_INIT(&p->timers); 569 p->fd = p->wbuf.fd = -1; 570 571 if (p->conf.if_depend[0]) 572 imsg_compose(ibuf_main, IMSG_IFINFO, 0, 0, -1, 573 p->conf.if_depend, sizeof(p->conf.if_depend)); 574 else 575 p->depend_ok = 1; 576 577 peer_cnt++; 578 579 change_state(p, STATE_IDLE, EVNT_NONE); 580 if (p->conf.down) 581 timer_stop(&p->timers, Timer_IdleHold); /* no autostart */ 582 else 583 timer_set(&p->timers, Timer_IdleHold, 0); /* start ASAP */ 584 585 /* 586 * on startup, demote if requested. 587 * do not handle new peers. they must reach ESTABLISHED beforehands. 588 * peers added at runtime have reconf_action set to RECONF_REINIT. 589 */ 590 if (p->reconf_action != RECONF_REINIT && p->conf.demote_group[0]) 591 session_demote(p, +1); 592 } 593 594 void 595 bgp_fsm(struct peer *peer, enum session_events event) 596 { 597 switch (peer->state) { 598 case STATE_NONE: 599 /* nothing */ 600 break; 601 case STATE_IDLE: 602 switch (event) { 603 case EVNT_START: 604 timer_stop(&peer->timers, Timer_Hold); 605 timer_stop(&peer->timers, Timer_SendHold); 606 timer_stop(&peer->timers, Timer_Keepalive); 607 timer_stop(&peer->timers, Timer_IdleHold); 608 609 /* allocate read buffer */ 610 peer->rbuf = calloc(1, sizeof(struct ibuf_read)); 611 if (peer->rbuf == NULL) 612 fatal(NULL); 613 614 /* init write buffer */ 615 msgbuf_init(&peer->wbuf); 616 617 peer->stats.last_sent_errcode = 0; 618 peer->stats.last_sent_suberr = 0; 619 peer->stats.last_rcvd_errcode = 0; 620 peer->stats.last_rcvd_suberr = 0; 621 622 if (!peer->depend_ok) 623 timer_stop(&peer->timers, Timer_ConnectRetry); 624 else if (peer->passive || peer->conf.passive || 625 peer->conf.template) { 626 change_state(peer, STATE_ACTIVE, event); 627 timer_stop(&peer->timers, Timer_ConnectRetry); 628 } else { 629 change_state(peer, STATE_CONNECT, event); 630 timer_set(&peer->timers, Timer_ConnectRetry, 631 conf->connectretry); 632 session_connect(peer); 633 } 634 peer->passive = 0; 635 break; 636 default: 637 /* ignore */ 638 break; 639 } 640 break; 641 case STATE_CONNECT: 642 switch (event) { 643 case EVNT_START: 644 /* ignore */ 645 break; 646 case EVNT_CON_OPEN: 647 session_tcp_established(peer); 648 session_open(peer); 649 timer_stop(&peer->timers, Timer_ConnectRetry); 650 peer->holdtime = INTERVAL_HOLD_INITIAL; 651 start_timer_holdtime(peer); 652 change_state(peer, STATE_OPENSENT, event); 653 break; 654 case EVNT_CON_OPENFAIL: 655 timer_set(&peer->timers, Timer_ConnectRetry, 656 conf->connectretry); 657 session_close_connection(peer); 658 change_state(peer, STATE_ACTIVE, event); 659 break; 660 case EVNT_TIMER_CONNRETRY: 661 timer_set(&peer->timers, Timer_ConnectRetry, 662 conf->connectretry); 663 session_connect(peer); 664 break; 665 default: 666 change_state(peer, STATE_IDLE, event); 667 break; 668 } 669 break; 670 case STATE_ACTIVE: 671 switch (event) { 672 case EVNT_START: 673 /* ignore */ 674 break; 675 case EVNT_CON_OPEN: 676 session_tcp_established(peer); 677 session_open(peer); 678 timer_stop(&peer->timers, Timer_ConnectRetry); 679 peer->holdtime = INTERVAL_HOLD_INITIAL; 680 start_timer_holdtime(peer); 681 change_state(peer, STATE_OPENSENT, event); 682 break; 683 case EVNT_CON_OPENFAIL: 684 timer_set(&peer->timers, Timer_ConnectRetry, 685 conf->connectretry); 686 session_close_connection(peer); 687 change_state(peer, STATE_ACTIVE, event); 688 break; 689 case EVNT_TIMER_CONNRETRY: 690 timer_set(&peer->timers, Timer_ConnectRetry, 691 peer->holdtime); 692 change_state(peer, STATE_CONNECT, event); 693 session_connect(peer); 694 break; 695 default: 696 change_state(peer, STATE_IDLE, event); 697 break; 698 } 699 break; 700 case STATE_OPENSENT: 701 switch (event) { 702 case EVNT_START: 703 /* ignore */ 704 break; 705 case EVNT_STOP: 706 change_state(peer, STATE_IDLE, event); 707 break; 708 case EVNT_CON_CLOSED: 709 session_close_connection(peer); 710 timer_set(&peer->timers, Timer_ConnectRetry, 711 conf->connectretry); 712 change_state(peer, STATE_ACTIVE, event); 713 break; 714 case EVNT_CON_FATAL: 715 change_state(peer, STATE_IDLE, event); 716 break; 717 case EVNT_TIMER_HOLDTIME: 718 case EVNT_TIMER_SENDHOLD: 719 session_notification(peer, ERR_HOLDTIMEREXPIRED, 720 0, NULL, 0); 721 change_state(peer, STATE_IDLE, event); 722 break; 723 case EVNT_RCVD_OPEN: 724 /* parse_open calls change_state itself on failure */ 725 if (parse_open(peer)) 726 break; 727 session_keepalive(peer); 728 change_state(peer, STATE_OPENCONFIRM, event); 729 break; 730 case EVNT_RCVD_NOTIFICATION: 731 if (parse_notification(peer)) { 732 change_state(peer, STATE_IDLE, event); 733 /* don't punish, capa negotiation */ 734 timer_set(&peer->timers, Timer_IdleHold, 0); 735 peer->IdleHoldTime /= 2; 736 } else 737 change_state(peer, STATE_IDLE, event); 738 break; 739 default: 740 session_notification(peer, 741 ERR_FSM, ERR_FSM_UNEX_OPENSENT, NULL, 0); 742 change_state(peer, STATE_IDLE, event); 743 break; 744 } 745 break; 746 case STATE_OPENCONFIRM: 747 switch (event) { 748 case EVNT_START: 749 /* ignore */ 750 break; 751 case EVNT_STOP: 752 change_state(peer, STATE_IDLE, event); 753 break; 754 case EVNT_CON_CLOSED: 755 case EVNT_CON_FATAL: 756 change_state(peer, STATE_IDLE, event); 757 break; 758 case EVNT_TIMER_HOLDTIME: 759 case EVNT_TIMER_SENDHOLD: 760 session_notification(peer, ERR_HOLDTIMEREXPIRED, 761 0, NULL, 0); 762 change_state(peer, STATE_IDLE, event); 763 break; 764 case EVNT_TIMER_KEEPALIVE: 765 session_keepalive(peer); 766 break; 767 case EVNT_RCVD_KEEPALIVE: 768 start_timer_holdtime(peer); 769 change_state(peer, STATE_ESTABLISHED, event); 770 break; 771 case EVNT_RCVD_NOTIFICATION: 772 parse_notification(peer); 773 change_state(peer, STATE_IDLE, event); 774 break; 775 default: 776 session_notification(peer, 777 ERR_FSM, ERR_FSM_UNEX_OPENCONFIRM, NULL, 0); 778 change_state(peer, STATE_IDLE, event); 779 break; 780 } 781 break; 782 case STATE_ESTABLISHED: 783 switch (event) { 784 case EVNT_START: 785 /* ignore */ 786 break; 787 case EVNT_STOP: 788 change_state(peer, STATE_IDLE, event); 789 break; 790 case EVNT_CON_CLOSED: 791 case EVNT_CON_FATAL: 792 change_state(peer, STATE_IDLE, event); 793 break; 794 case EVNT_TIMER_HOLDTIME: 795 case EVNT_TIMER_SENDHOLD: 796 session_notification(peer, ERR_HOLDTIMEREXPIRED, 797 0, NULL, 0); 798 change_state(peer, STATE_IDLE, event); 799 break; 800 case EVNT_TIMER_KEEPALIVE: 801 session_keepalive(peer); 802 break; 803 case EVNT_RCVD_KEEPALIVE: 804 start_timer_holdtime(peer); 805 break; 806 case EVNT_RCVD_UPDATE: 807 start_timer_holdtime(peer); 808 if (parse_update(peer)) 809 change_state(peer, STATE_IDLE, event); 810 else 811 start_timer_holdtime(peer); 812 break; 813 case EVNT_RCVD_NOTIFICATION: 814 parse_notification(peer); 815 change_state(peer, STATE_IDLE, event); 816 break; 817 default: 818 session_notification(peer, 819 ERR_FSM, ERR_FSM_UNEX_ESTABLISHED, NULL, 0); 820 change_state(peer, STATE_IDLE, event); 821 break; 822 } 823 break; 824 } 825 } 826 827 void 828 start_timer_holdtime(struct peer *peer) 829 { 830 if (peer->holdtime > 0) 831 timer_set(&peer->timers, Timer_Hold, peer->holdtime); 832 else 833 timer_stop(&peer->timers, Timer_Hold); 834 } 835 836 void 837 start_timer_keepalive(struct peer *peer) 838 { 839 if (peer->holdtime > 0) 840 timer_set(&peer->timers, Timer_Keepalive, peer->holdtime / 3); 841 else 842 timer_stop(&peer->timers, Timer_Keepalive); 843 } 844 845 void 846 session_close_connection(struct peer *peer) 847 { 848 if (peer->fd != -1) { 849 close(peer->fd); 850 pauseaccept = 0; 851 } 852 peer->fd = peer->wbuf.fd = -1; 853 } 854 855 void 856 change_state(struct peer *peer, enum session_state state, 857 enum session_events event) 858 { 859 struct mrt *mrt; 860 861 switch (state) { 862 case STATE_IDLE: 863 /* carp demotion first. new peers handled in init_peer */ 864 if (peer->state == STATE_ESTABLISHED && 865 peer->conf.demote_group[0] && !peer->demoted) 866 session_demote(peer, +1); 867 868 /* 869 * try to write out what's buffered (maybe a notification), 870 * don't bother if it fails 871 */ 872 if (peer->state >= STATE_OPENSENT && peer->wbuf.queued) 873 msgbuf_write(&peer->wbuf); 874 875 /* 876 * we must start the timer for the next EVNT_START 877 * if we are coming here due to an error and the 878 * session was not established successfully before, the 879 * starttimerinterval needs to be exponentially increased 880 */ 881 if (peer->IdleHoldTime == 0) 882 peer->IdleHoldTime = INTERVAL_IDLE_HOLD_INITIAL; 883 peer->holdtime = INTERVAL_HOLD_INITIAL; 884 timer_stop(&peer->timers, Timer_ConnectRetry); 885 timer_stop(&peer->timers, Timer_Keepalive); 886 timer_stop(&peer->timers, Timer_Hold); 887 timer_stop(&peer->timers, Timer_SendHold); 888 timer_stop(&peer->timers, Timer_IdleHold); 889 timer_stop(&peer->timers, Timer_IdleHoldReset); 890 session_close_connection(peer); 891 msgbuf_clear(&peer->wbuf); 892 free(peer->rbuf); 893 peer->rbuf = NULL; 894 peer->rpending = 0; 895 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 896 if (!peer->template) 897 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 898 peer->conf.id, 0, -1, NULL, 0); 899 900 if (event != EVNT_STOP) { 901 timer_set(&peer->timers, Timer_IdleHold, 902 peer->IdleHoldTime); 903 if (event != EVNT_NONE && 904 peer->IdleHoldTime < MAX_IDLE_HOLD/2) 905 peer->IdleHoldTime *= 2; 906 } 907 if (peer->state == STATE_ESTABLISHED) { 908 if (peer->capa.neg.grestart.restart == 2 && 909 (event == EVNT_CON_CLOSED || 910 event == EVNT_CON_FATAL)) { 911 /* don't punish graceful restart */ 912 timer_set(&peer->timers, Timer_IdleHold, 0); 913 peer->IdleHoldTime /= 2; 914 session_graceful_restart(peer); 915 } else 916 session_down(peer); 917 } 918 if (peer->state == STATE_NONE || 919 peer->state == STATE_ESTABLISHED) { 920 /* initialize capability negotiation structures */ 921 memcpy(&peer->capa.ann, &peer->conf.capabilities, 922 sizeof(peer->capa.ann)); 923 if (!peer->conf.announce_capa) 924 session_capa_ann_none(peer); 925 } 926 break; 927 case STATE_CONNECT: 928 if (peer->state == STATE_ESTABLISHED && 929 peer->capa.neg.grestart.restart == 2) { 930 /* do the graceful restart dance */ 931 session_graceful_restart(peer); 932 peer->holdtime = INTERVAL_HOLD_INITIAL; 933 timer_stop(&peer->timers, Timer_ConnectRetry); 934 timer_stop(&peer->timers, Timer_Keepalive); 935 timer_stop(&peer->timers, Timer_Hold); 936 timer_stop(&peer->timers, Timer_SendHold); 937 timer_stop(&peer->timers, Timer_IdleHold); 938 timer_stop(&peer->timers, Timer_IdleHoldReset); 939 session_close_connection(peer); 940 msgbuf_clear(&peer->wbuf); 941 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 942 } 943 break; 944 case STATE_ACTIVE: 945 if (!peer->template) 946 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 947 peer->conf.id, 0, -1, NULL, 0); 948 break; 949 case STATE_OPENSENT: 950 break; 951 case STATE_OPENCONFIRM: 952 break; 953 case STATE_ESTABLISHED: 954 timer_set(&peer->timers, Timer_IdleHoldReset, 955 peer->IdleHoldTime); 956 if (peer->demoted) 957 timer_set(&peer->timers, Timer_CarpUndemote, 958 INTERVAL_HOLD_DEMOTED); 959 session_up(peer); 960 break; 961 default: /* something seriously fucked */ 962 break; 963 } 964 965 log_statechange(peer, state, event); 966 LIST_FOREACH(mrt, &mrthead, entry) { 967 if (!(mrt->type == MRT_ALL_IN || mrt->type == MRT_ALL_OUT)) 968 continue; 969 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 970 mrt->peer_id == peer->conf.id || (mrt->group_id != 0 && 971 mrt->group_id == peer->conf.groupid)) 972 mrt_dump_state(mrt, peer->state, state, peer); 973 } 974 peer->prev_state = peer->state; 975 peer->state = state; 976 } 977 978 void 979 session_accept(int listenfd) 980 { 981 int connfd; 982 socklen_t len; 983 struct sockaddr_storage cliaddr; 984 struct peer *p = NULL; 985 986 len = sizeof(cliaddr); 987 if ((connfd = accept4(listenfd, 988 (struct sockaddr *)&cliaddr, &len, 989 SOCK_CLOEXEC | SOCK_NONBLOCK)) == -1) { 990 if (errno == ENFILE || errno == EMFILE) 991 pauseaccept = getmonotime(); 992 else if (errno != EWOULDBLOCK && errno != EINTR && 993 errno != ECONNABORTED) 994 log_warn("accept"); 995 return; 996 } 997 998 p = getpeerbyip(conf, (struct sockaddr *)&cliaddr); 999 1000 if (p != NULL && p->state == STATE_IDLE && p->errcnt < 2) { 1001 if (timer_running(&p->timers, Timer_IdleHold, NULL)) { 1002 /* fast reconnect after clear */ 1003 p->passive = 1; 1004 bgp_fsm(p, EVNT_START); 1005 } 1006 } 1007 1008 if (p != NULL && 1009 (p->state == STATE_CONNECT || p->state == STATE_ACTIVE)) { 1010 if (p->fd != -1) { 1011 if (p->state == STATE_CONNECT) 1012 session_close_connection(p); 1013 else { 1014 close(connfd); 1015 return; 1016 } 1017 } 1018 1019 open: 1020 if (p->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1021 log_peer_warnx(&p->conf, 1022 "ipsec or md5sig configured but not available"); 1023 close(connfd); 1024 return; 1025 } 1026 1027 if (tcp_md5_check(connfd, p) == -1) { 1028 close(connfd); 1029 return; 1030 } 1031 p->fd = p->wbuf.fd = connfd; 1032 if (session_setup_socket(p)) { 1033 close(connfd); 1034 return; 1035 } 1036 bgp_fsm(p, EVNT_CON_OPEN); 1037 return; 1038 } else if (p != NULL && p->state == STATE_ESTABLISHED && 1039 p->capa.neg.grestart.restart == 2) { 1040 /* first do the graceful restart dance */ 1041 change_state(p, STATE_CONNECT, EVNT_CON_CLOSED); 1042 /* then do part of the open dance */ 1043 goto open; 1044 } else { 1045 log_conn_attempt(p, (struct sockaddr *)&cliaddr, len); 1046 close(connfd); 1047 } 1048 } 1049 1050 int 1051 session_connect(struct peer *peer) 1052 { 1053 struct sockaddr *sa; 1054 struct bgpd_addr *bind_addr = NULL; 1055 socklen_t sa_len; 1056 1057 /* 1058 * we do not need the overcomplicated collision detection RFC 1771 1059 * describes; we simply make sure there is only ever one concurrent 1060 * tcp connection per peer. 1061 */ 1062 if (peer->fd != -1) 1063 return (-1); 1064 1065 if ((peer->fd = socket(aid2af(peer->conf.remote_addr.aid), 1066 SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, IPPROTO_TCP)) == -1) { 1067 log_peer_warn(&peer->conf, "session_connect socket"); 1068 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1069 return (-1); 1070 } 1071 1072 if (peer->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1073 log_peer_warnx(&peer->conf, 1074 "ipsec or md5sig configured but not available"); 1075 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1076 return (-1); 1077 } 1078 1079 tcp_md5_set(peer->fd, peer); 1080 peer->wbuf.fd = peer->fd; 1081 1082 /* if local-address is set we need to bind() */ 1083 switch (peer->conf.remote_addr.aid) { 1084 case AID_INET: 1085 bind_addr = &peer->conf.local_addr_v4; 1086 break; 1087 case AID_INET6: 1088 bind_addr = &peer->conf.local_addr_v6; 1089 break; 1090 } 1091 if ((sa = addr2sa(bind_addr, 0, &sa_len)) != NULL) { 1092 if (bind(peer->fd, sa, sa_len) == -1) { 1093 log_peer_warn(&peer->conf, "session_connect bind"); 1094 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1095 return (-1); 1096 } 1097 } 1098 1099 if (session_setup_socket(peer)) { 1100 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1101 return (-1); 1102 } 1103 1104 sa = addr2sa(&peer->conf.remote_addr, peer->conf.remote_port, &sa_len); 1105 if (connect(peer->fd, sa, sa_len) == -1) { 1106 if (errno != EINPROGRESS) { 1107 if (errno != peer->lasterr) 1108 log_peer_warn(&peer->conf, "connect"); 1109 peer->lasterr = errno; 1110 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1111 return (-1); 1112 } 1113 } else 1114 bgp_fsm(peer, EVNT_CON_OPEN); 1115 1116 return (0); 1117 } 1118 1119 int 1120 session_setup_socket(struct peer *p) 1121 { 1122 int ttl = p->conf.distance; 1123 int pre = IPTOS_PREC_INTERNETCONTROL; 1124 int nodelay = 1; 1125 int bsize; 1126 1127 switch (p->conf.remote_addr.aid) { 1128 case AID_INET: 1129 /* set precedence, see RFC 1771 appendix 5 */ 1130 if (setsockopt(p->fd, IPPROTO_IP, IP_TOS, &pre, sizeof(pre)) == 1131 -1) { 1132 log_peer_warn(&p->conf, 1133 "session_setup_socket setsockopt TOS"); 1134 return (-1); 1135 } 1136 1137 if (p->conf.ebgp) { 1138 /* 1139 * set TTL to foreign router's distance 1140 * 1=direct n=multihop with ttlsec, we always use 255 1141 */ 1142 if (p->conf.ttlsec) { 1143 ttl = 256 - p->conf.distance; 1144 if (setsockopt(p->fd, IPPROTO_IP, IP_MINTTL, 1145 &ttl, sizeof(ttl)) == -1) { 1146 log_peer_warn(&p->conf, 1147 "session_setup_socket: " 1148 "setsockopt MINTTL"); 1149 return (-1); 1150 } 1151 ttl = 255; 1152 } 1153 1154 if (setsockopt(p->fd, IPPROTO_IP, IP_TTL, &ttl, 1155 sizeof(ttl)) == -1) { 1156 log_peer_warn(&p->conf, 1157 "session_setup_socket setsockopt TTL"); 1158 return (-1); 1159 } 1160 } 1161 break; 1162 case AID_INET6: 1163 if (p->conf.ebgp) { 1164 /* 1165 * set hoplimit to foreign router's distance 1166 * 1=direct n=multihop with ttlsec, we always use 255 1167 */ 1168 if (p->conf.ttlsec) { 1169 ttl = 256 - p->conf.distance; 1170 if (setsockopt(p->fd, IPPROTO_IPV6, 1171 IPV6_MINHOPCOUNT, &ttl, sizeof(ttl)) 1172 == -1) { 1173 log_peer_warn(&p->conf, 1174 "session_setup_socket: " 1175 "setsockopt MINHOPCOUNT"); 1176 return (-1); 1177 } 1178 ttl = 255; 1179 } 1180 if (setsockopt(p->fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, 1181 &ttl, sizeof(ttl)) == -1) { 1182 log_peer_warn(&p->conf, 1183 "session_setup_socket setsockopt hoplimit"); 1184 return (-1); 1185 } 1186 } 1187 break; 1188 } 1189 1190 /* set TCP_NODELAY */ 1191 if (setsockopt(p->fd, IPPROTO_TCP, TCP_NODELAY, &nodelay, 1192 sizeof(nodelay)) == -1) { 1193 log_peer_warn(&p->conf, 1194 "session_setup_socket setsockopt TCP_NODELAY"); 1195 return (-1); 1196 } 1197 1198 /* only increase bufsize (and thus window) if md5 or ipsec is in use */ 1199 if (p->conf.auth.method != AUTH_NONE) { 1200 /* try to increase bufsize. no biggie if it fails */ 1201 bsize = 65535; 1202 while (bsize > 8192 && 1203 setsockopt(p->fd, SOL_SOCKET, SO_RCVBUF, &bsize, 1204 sizeof(bsize)) == -1 && errno != EINVAL) 1205 bsize /= 2; 1206 bsize = 65535; 1207 while (bsize > 8192 && 1208 setsockopt(p->fd, SOL_SOCKET, SO_SNDBUF, &bsize, 1209 sizeof(bsize)) == -1 && errno != EINVAL) 1210 bsize /= 2; 1211 } 1212 1213 return (0); 1214 } 1215 1216 /* compare two sockaddrs by converting them into bgpd_addr */ 1217 static int 1218 sa_cmp(struct sockaddr *a, struct sockaddr *b) 1219 { 1220 struct bgpd_addr ba, bb; 1221 1222 sa2addr(a, &ba, NULL); 1223 sa2addr(b, &bb, NULL); 1224 1225 return (memcmp(&ba, &bb, sizeof(ba)) == 0); 1226 } 1227 1228 static void 1229 get_alternate_addr(struct sockaddr *sa, struct bgpd_addr *alt) 1230 { 1231 struct ifaddrs *ifap, *ifa, *match; 1232 1233 if (getifaddrs(&ifap) == -1) 1234 fatal("getifaddrs"); 1235 1236 for (match = ifap; match != NULL; match = match->ifa_next) 1237 if (match->ifa_addr != NULL && 1238 sa_cmp(sa, match->ifa_addr) == 0) 1239 break; 1240 1241 if (match == NULL) { 1242 log_warnx("%s: local address not found", __func__); 1243 return; 1244 } 1245 1246 switch (sa->sa_family) { 1247 case AF_INET6: 1248 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1249 if (ifa->ifa_addr != NULL && 1250 ifa->ifa_addr->sa_family == AF_INET && 1251 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1252 sa2addr(ifa->ifa_addr, alt, NULL); 1253 break; 1254 } 1255 } 1256 break; 1257 case AF_INET: 1258 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1259 if (ifa->ifa_addr != NULL && 1260 ifa->ifa_addr->sa_family == AF_INET6 && 1261 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1262 struct sockaddr_in6 *s = 1263 (struct sockaddr_in6 *)ifa->ifa_addr; 1264 1265 /* only accept global scope addresses */ 1266 if (IN6_IS_ADDR_LINKLOCAL(&s->sin6_addr) || 1267 IN6_IS_ADDR_SITELOCAL(&s->sin6_addr)) 1268 continue; 1269 sa2addr(ifa->ifa_addr, alt, NULL); 1270 break; 1271 } 1272 } 1273 break; 1274 default: 1275 log_warnx("%s: unsupported address family %d", __func__, 1276 sa->sa_family); 1277 break; 1278 } 1279 1280 freeifaddrs(ifap); 1281 } 1282 1283 void 1284 session_tcp_established(struct peer *peer) 1285 { 1286 struct sockaddr_storage ss; 1287 socklen_t len; 1288 1289 len = sizeof(ss); 1290 if (getsockname(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1291 log_warn("getsockname"); 1292 sa2addr((struct sockaddr *)&ss, &peer->local, &peer->local_port); 1293 get_alternate_addr((struct sockaddr *)&ss, &peer->local_alt); 1294 len = sizeof(ss); 1295 if (getpeername(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1296 log_warn("getpeername"); 1297 sa2addr((struct sockaddr *)&ss, &peer->remote, &peer->remote_port); 1298 } 1299 1300 void 1301 session_capa_ann_none(struct peer *peer) 1302 { 1303 bzero(&peer->capa.ann, sizeof(peer->capa.ann)); 1304 } 1305 1306 int 1307 session_capa_add(struct ibuf *opb, uint8_t capa_code, uint8_t capa_len) 1308 { 1309 int errs = 0; 1310 1311 errs += ibuf_add(opb, &capa_code, sizeof(capa_code)); 1312 errs += ibuf_add(opb, &capa_len, sizeof(capa_len)); 1313 return (errs); 1314 } 1315 1316 int 1317 session_capa_add_mp(struct ibuf *buf, uint8_t aid) 1318 { 1319 uint8_t safi, pad = 0; 1320 uint16_t afi; 1321 int errs = 0; 1322 1323 if (aid2afi(aid, &afi, &safi) == -1) 1324 fatalx("session_capa_add_mp: bad afi/safi pair"); 1325 afi = htons(afi); 1326 errs += ibuf_add(buf, &afi, sizeof(afi)); 1327 errs += ibuf_add(buf, &pad, sizeof(pad)); 1328 errs += ibuf_add(buf, &safi, sizeof(safi)); 1329 1330 return (errs); 1331 } 1332 1333 int 1334 session_capa_add_afi(struct peer *p, struct ibuf *b, uint8_t aid, 1335 uint8_t flags) 1336 { 1337 u_int errs = 0; 1338 uint16_t afi; 1339 uint8_t safi; 1340 1341 if (aid2afi(aid, &afi, &safi)) { 1342 log_warn("session_capa_add_afi: bad AID"); 1343 return (1); 1344 } 1345 1346 afi = htons(afi); 1347 errs += ibuf_add(b, &afi, sizeof(afi)); 1348 errs += ibuf_add(b, &safi, sizeof(safi)); 1349 errs += ibuf_add(b, &flags, sizeof(flags)); 1350 1351 return (errs); 1352 } 1353 1354 struct bgp_msg * 1355 session_newmsg(enum msg_type msgtype, uint16_t len) 1356 { 1357 struct bgp_msg *msg; 1358 struct msg_header hdr; 1359 struct ibuf *buf; 1360 int errs = 0; 1361 1362 memset(&hdr.marker, 0xff, sizeof(hdr.marker)); 1363 hdr.len = htons(len); 1364 hdr.type = msgtype; 1365 1366 if ((buf = ibuf_open(len)) == NULL) 1367 return (NULL); 1368 1369 errs += ibuf_add(buf, &hdr.marker, sizeof(hdr.marker)); 1370 errs += ibuf_add(buf, &hdr.len, sizeof(hdr.len)); 1371 errs += ibuf_add(buf, &hdr.type, sizeof(hdr.type)); 1372 1373 if (errs || (msg = calloc(1, sizeof(*msg))) == NULL) { 1374 ibuf_free(buf); 1375 return (NULL); 1376 } 1377 1378 msg->buf = buf; 1379 msg->type = msgtype; 1380 msg->len = len; 1381 1382 return (msg); 1383 } 1384 1385 int 1386 session_sendmsg(struct bgp_msg *msg, struct peer *p) 1387 { 1388 struct mrt *mrt; 1389 1390 LIST_FOREACH(mrt, &mrthead, entry) { 1391 if (!(mrt->type == MRT_ALL_OUT || (msg->type == UPDATE && 1392 mrt->type == MRT_UPDATE_OUT))) 1393 continue; 1394 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1395 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1396 mrt->group_id == p->conf.groupid)) 1397 mrt_dump_bgp_msg(mrt, msg->buf->buf, msg->len, p, 1398 msg->type); 1399 } 1400 1401 ibuf_close(&p->wbuf, msg->buf); 1402 if (!p->throttled && p->wbuf.queued > SESS_MSG_HIGH_MARK) { 1403 if (imsg_rde(IMSG_XOFF, p->conf.id, NULL, 0) == -1) 1404 log_peer_warn(&p->conf, "imsg_compose XOFF"); 1405 else 1406 p->throttled = 1; 1407 } 1408 1409 free(msg); 1410 return (0); 1411 } 1412 1413 void 1414 session_open(struct peer *p) 1415 { 1416 struct bgp_msg *buf; 1417 struct ibuf *opb; 1418 struct msg_open msg; 1419 uint16_t len, optparamlen = 0; 1420 uint8_t i, op_type; 1421 int errs = 0, extlen = 0; 1422 int mpcapa = 0; 1423 1424 1425 if ((opb = ibuf_dynamic(0, UINT16_MAX - 3)) == NULL) { 1426 bgp_fsm(p, EVNT_CON_FATAL); 1427 return; 1428 } 1429 1430 /* multiprotocol extensions, RFC 4760 */ 1431 for (i = 0; i < AID_MAX; i++) 1432 if (p->capa.ann.mp[i]) { /* 4 bytes data */ 1433 errs += session_capa_add(opb, CAPA_MP, 4); 1434 errs += session_capa_add_mp(opb, i); 1435 mpcapa++; 1436 } 1437 1438 /* route refresh, RFC 2918 */ 1439 if (p->capa.ann.refresh) /* no data */ 1440 errs += session_capa_add(opb, CAPA_REFRESH, 0); 1441 1442 /* graceful restart and End-of-RIB marker, RFC 4724 */ 1443 if (p->capa.ann.grestart.restart) { 1444 int rst = 0; 1445 uint16_t hdr = 0; 1446 1447 for (i = 0; i < AID_MAX; i++) { 1448 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) 1449 rst++; 1450 } 1451 1452 /* Only set the R-flag if no graceful restart is ongoing */ 1453 if (!rst) 1454 hdr |= CAPA_GR_R_FLAG; 1455 hdr = htons(hdr); 1456 1457 errs += session_capa_add(opb, CAPA_RESTART, sizeof(hdr)); 1458 errs += ibuf_add(opb, &hdr, sizeof(hdr)); 1459 } 1460 1461 /* 4-bytes AS numbers, RFC6793 */ 1462 if (p->capa.ann.as4byte) { /* 4 bytes data */ 1463 uint32_t nas; 1464 1465 nas = htonl(p->conf.local_as); 1466 errs += session_capa_add(opb, CAPA_AS4BYTE, sizeof(nas)); 1467 errs += ibuf_add(opb, &nas, sizeof(nas)); 1468 } 1469 1470 /* advertisement of multiple paths, RFC7911 */ 1471 if (p->capa.ann.add_path[0]) { /* variable */ 1472 uint8_t aplen; 1473 1474 if (mpcapa) 1475 aplen = 4 * mpcapa; 1476 else /* AID_INET */ 1477 aplen = 4; 1478 errs += session_capa_add(opb, CAPA_ADD_PATH, aplen); 1479 if (mpcapa) { 1480 for (i = AID_MIN; i < AID_MAX; i++) { 1481 if (p->capa.ann.mp[i]) { 1482 errs += session_capa_add_afi(p, opb, 1483 i, p->capa.ann.add_path[i]); 1484 } 1485 } 1486 } else { /* AID_INET */ 1487 errs += session_capa_add_afi(p, opb, AID_INET, 1488 p->capa.ann.add_path[AID_INET]); 1489 } 1490 } 1491 1492 /* enhanced route-refresh, RFC7313 */ 1493 if (p->capa.ann.enhanced_rr) /* no data */ 1494 errs += session_capa_add(opb, CAPA_ENHANCED_RR, 0); 1495 1496 optparamlen = ibuf_size(opb); 1497 if (optparamlen == 0) { 1498 /* nothing */ 1499 } else if (optparamlen + 2 >= 255) { 1500 /* RFC9072: 2 byte lenght instead of 1 + 3 byte extra header */ 1501 optparamlen += sizeof(op_type) + 2 + 3; 1502 msg.optparamlen = 255; 1503 extlen = 1; 1504 } else { 1505 optparamlen += sizeof(op_type) + 1; 1506 msg.optparamlen = optparamlen; 1507 } 1508 1509 len = MSGSIZE_OPEN_MIN + optparamlen; 1510 if (errs || (buf = session_newmsg(OPEN, len)) == NULL) { 1511 ibuf_free(opb); 1512 bgp_fsm(p, EVNT_CON_FATAL); 1513 return; 1514 } 1515 1516 msg.version = 4; 1517 msg.myas = htons(p->conf.local_short_as); 1518 if (p->conf.holdtime) 1519 msg.holdtime = htons(p->conf.holdtime); 1520 else 1521 msg.holdtime = htons(conf->holdtime); 1522 msg.bgpid = conf->bgpid; /* is already in network byte order */ 1523 1524 errs += ibuf_add(buf->buf, &msg.version, sizeof(msg.version)); 1525 errs += ibuf_add(buf->buf, &msg.myas, sizeof(msg.myas)); 1526 errs += ibuf_add(buf->buf, &msg.holdtime, sizeof(msg.holdtime)); 1527 errs += ibuf_add(buf->buf, &msg.bgpid, sizeof(msg.bgpid)); 1528 errs += ibuf_add(buf->buf, &msg.optparamlen, 1); 1529 1530 if (extlen) { 1531 /* write RFC9072 extra header */ 1532 uint16_t op_extlen = htons(optparamlen - 3); 1533 op_type = OPT_PARAM_EXT_LEN; 1534 errs += ibuf_add(buf->buf, &op_type, 1); 1535 errs += ibuf_add(buf->buf, &op_extlen, 2); 1536 } 1537 1538 if (optparamlen) { 1539 op_type = OPT_PARAM_CAPABILITIES; 1540 errs += ibuf_add(buf->buf, &op_type, sizeof(op_type)); 1541 1542 optparamlen = ibuf_size(opb); 1543 if (extlen) { 1544 /* RFC9072: 2-byte extended length */ 1545 uint16_t op_extlen = htons(optparamlen); 1546 errs += ibuf_add(buf->buf, &op_extlen, 2); 1547 } else { 1548 uint8_t op_len = optparamlen; 1549 errs += ibuf_add(buf->buf, &op_len, 1); 1550 } 1551 errs += ibuf_add(buf->buf, opb->buf, ibuf_size(opb)); 1552 } 1553 1554 ibuf_free(opb); 1555 1556 if (errs) { 1557 ibuf_free(buf->buf); 1558 free(buf); 1559 bgp_fsm(p, EVNT_CON_FATAL); 1560 return; 1561 } 1562 1563 if (session_sendmsg(buf, p) == -1) { 1564 bgp_fsm(p, EVNT_CON_FATAL); 1565 return; 1566 } 1567 1568 p->stats.msg_sent_open++; 1569 } 1570 1571 void 1572 session_keepalive(struct peer *p) 1573 { 1574 struct bgp_msg *buf; 1575 1576 if ((buf = session_newmsg(KEEPALIVE, MSGSIZE_KEEPALIVE)) == NULL || 1577 session_sendmsg(buf, p) == -1) { 1578 bgp_fsm(p, EVNT_CON_FATAL); 1579 return; 1580 } 1581 1582 start_timer_keepalive(p); 1583 p->stats.msg_sent_keepalive++; 1584 } 1585 1586 void 1587 session_update(uint32_t peerid, void *data, size_t datalen) 1588 { 1589 struct peer *p; 1590 struct bgp_msg *buf; 1591 1592 if ((p = getpeerbyid(conf, peerid)) == NULL) { 1593 log_warnx("no such peer: id=%u", peerid); 1594 return; 1595 } 1596 1597 if (p->state != STATE_ESTABLISHED) 1598 return; 1599 1600 if ((buf = session_newmsg(UPDATE, MSGSIZE_HEADER + datalen)) == NULL) { 1601 bgp_fsm(p, EVNT_CON_FATAL); 1602 return; 1603 } 1604 1605 if (ibuf_add(buf->buf, data, datalen)) { 1606 ibuf_free(buf->buf); 1607 free(buf); 1608 bgp_fsm(p, EVNT_CON_FATAL); 1609 return; 1610 } 1611 1612 if (session_sendmsg(buf, p) == -1) { 1613 bgp_fsm(p, EVNT_CON_FATAL); 1614 return; 1615 } 1616 1617 start_timer_keepalive(p); 1618 p->stats.msg_sent_update++; 1619 } 1620 1621 void 1622 session_notification(struct peer *p, uint8_t errcode, uint8_t subcode, 1623 void *data, ssize_t datalen) 1624 { 1625 struct bgp_msg *buf; 1626 int errs = 0; 1627 1628 if (p->stats.last_sent_errcode) /* some notification already sent */ 1629 return; 1630 1631 log_notification(p, errcode, subcode, data, datalen, "sending"); 1632 1633 /* cap to maximum size */ 1634 if (datalen > MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN) { 1635 log_peer_warnx(&p->conf, 1636 "oversized notification, data trunkated"); 1637 datalen = MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN; 1638 } 1639 1640 if ((buf = session_newmsg(NOTIFICATION, 1641 MSGSIZE_NOTIFICATION_MIN + datalen)) == NULL) { 1642 bgp_fsm(p, EVNT_CON_FATAL); 1643 return; 1644 } 1645 1646 errs += ibuf_add(buf->buf, &errcode, sizeof(errcode)); 1647 errs += ibuf_add(buf->buf, &subcode, sizeof(subcode)); 1648 1649 if (datalen > 0) 1650 errs += ibuf_add(buf->buf, data, datalen); 1651 1652 if (errs) { 1653 ibuf_free(buf->buf); 1654 free(buf); 1655 bgp_fsm(p, EVNT_CON_FATAL); 1656 return; 1657 } 1658 1659 if (session_sendmsg(buf, p) == -1) { 1660 bgp_fsm(p, EVNT_CON_FATAL); 1661 return; 1662 } 1663 1664 p->stats.msg_sent_notification++; 1665 p->stats.last_sent_errcode = errcode; 1666 p->stats.last_sent_suberr = subcode; 1667 } 1668 1669 int 1670 session_neighbor_rrefresh(struct peer *p) 1671 { 1672 uint8_t i; 1673 1674 if (!(p->capa.neg.refresh || p->capa.neg.enhanced_rr)) 1675 return (-1); 1676 1677 for (i = 0; i < AID_MAX; i++) { 1678 if (p->capa.neg.mp[i] != 0) 1679 session_rrefresh(p, i, ROUTE_REFRESH_REQUEST); 1680 } 1681 1682 return (0); 1683 } 1684 1685 void 1686 session_rrefresh(struct peer *p, uint8_t aid, uint8_t subtype) 1687 { 1688 struct bgp_msg *buf; 1689 int errs = 0; 1690 uint16_t afi; 1691 uint8_t safi; 1692 1693 switch (subtype) { 1694 case ROUTE_REFRESH_REQUEST: 1695 p->stats.refresh_sent_req++; 1696 break; 1697 case ROUTE_REFRESH_BEGIN_RR: 1698 case ROUTE_REFRESH_END_RR: 1699 /* requires enhanced route refresh */ 1700 if (!p->capa.neg.enhanced_rr) 1701 return; 1702 if (subtype == ROUTE_REFRESH_BEGIN_RR) 1703 p->stats.refresh_sent_borr++; 1704 else 1705 p->stats.refresh_sent_eorr++; 1706 break; 1707 default: 1708 fatalx("session_rrefresh: bad subtype %d", subtype); 1709 } 1710 1711 if (aid2afi(aid, &afi, &safi) == -1) 1712 fatalx("session_rrefresh: bad afi/safi pair"); 1713 1714 if ((buf = session_newmsg(RREFRESH, MSGSIZE_RREFRESH)) == NULL) { 1715 bgp_fsm(p, EVNT_CON_FATAL); 1716 return; 1717 } 1718 1719 afi = htons(afi); 1720 errs += ibuf_add(buf->buf, &afi, sizeof(afi)); 1721 errs += ibuf_add(buf->buf, &subtype, sizeof(subtype)); 1722 errs += ibuf_add(buf->buf, &safi, sizeof(safi)); 1723 1724 if (errs) { 1725 ibuf_free(buf->buf); 1726 free(buf); 1727 bgp_fsm(p, EVNT_CON_FATAL); 1728 return; 1729 } 1730 1731 if (session_sendmsg(buf, p) == -1) { 1732 bgp_fsm(p, EVNT_CON_FATAL); 1733 return; 1734 } 1735 1736 p->stats.msg_sent_rrefresh++; 1737 } 1738 1739 int 1740 session_graceful_restart(struct peer *p) 1741 { 1742 uint8_t i; 1743 1744 timer_set(&p->timers, Timer_RestartTimeout, 1745 p->capa.neg.grestart.timeout); 1746 1747 for (i = 0; i < AID_MAX; i++) { 1748 if (p->capa.neg.grestart.flags[i] & CAPA_GR_PRESENT) { 1749 if (imsg_rde(IMSG_SESSION_STALE, p->conf.id, 1750 &i, sizeof(i)) == -1) 1751 return (-1); 1752 log_peer_warnx(&p->conf, 1753 "graceful restart of %s, keeping routes", 1754 aid2str(i)); 1755 p->capa.neg.grestart.flags[i] |= CAPA_GR_RESTARTING; 1756 } else if (p->capa.neg.mp[i]) { 1757 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1758 &i, sizeof(i)) == -1) 1759 return (-1); 1760 log_peer_warnx(&p->conf, 1761 "graceful restart of %s, flushing routes", 1762 aid2str(i)); 1763 } 1764 } 1765 return (0); 1766 } 1767 1768 int 1769 session_graceful_stop(struct peer *p) 1770 { 1771 uint8_t i; 1772 1773 for (i = 0; i < AID_MAX; i++) { 1774 /* 1775 * Only flush if the peer is restarting and the timeout fired. 1776 * In all other cases the session was already flushed when the 1777 * session went down or when the new open message was parsed. 1778 */ 1779 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) { 1780 log_peer_warnx(&p->conf, "graceful restart of %s, " 1781 "time-out, flushing", aid2str(i)); 1782 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1783 &i, sizeof(i)) == -1) 1784 return (-1); 1785 } 1786 p->capa.neg.grestart.flags[i] &= ~CAPA_GR_RESTARTING; 1787 } 1788 return (0); 1789 } 1790 1791 int 1792 session_dispatch_msg(struct pollfd *pfd, struct peer *p) 1793 { 1794 ssize_t n; 1795 socklen_t len; 1796 int error; 1797 1798 if (p->state == STATE_CONNECT) { 1799 if (pfd->revents & POLLOUT) { 1800 if (pfd->revents & POLLIN) { 1801 /* error occurred */ 1802 len = sizeof(error); 1803 if (getsockopt(pfd->fd, SOL_SOCKET, SO_ERROR, 1804 &error, &len) == -1 || error) { 1805 if (error) 1806 errno = error; 1807 if (errno != p->lasterr) { 1808 log_peer_warn(&p->conf, 1809 "socket error"); 1810 p->lasterr = errno; 1811 } 1812 bgp_fsm(p, EVNT_CON_OPENFAIL); 1813 return (1); 1814 } 1815 } 1816 bgp_fsm(p, EVNT_CON_OPEN); 1817 return (1); 1818 } 1819 if (pfd->revents & POLLHUP) { 1820 bgp_fsm(p, EVNT_CON_OPENFAIL); 1821 return (1); 1822 } 1823 if (pfd->revents & (POLLERR|POLLNVAL)) { 1824 bgp_fsm(p, EVNT_CON_FATAL); 1825 return (1); 1826 } 1827 return (0); 1828 } 1829 1830 if (pfd->revents & POLLHUP) { 1831 bgp_fsm(p, EVNT_CON_CLOSED); 1832 return (1); 1833 } 1834 if (pfd->revents & (POLLERR|POLLNVAL)) { 1835 bgp_fsm(p, EVNT_CON_FATAL); 1836 return (1); 1837 } 1838 1839 if (pfd->revents & POLLOUT && p->wbuf.queued) { 1840 if ((error = msgbuf_write(&p->wbuf)) <= 0 && errno != EAGAIN) { 1841 if (error == 0) 1842 log_peer_warnx(&p->conf, "Connection closed"); 1843 else if (error == -1) 1844 log_peer_warn(&p->conf, "write error"); 1845 bgp_fsm(p, EVNT_CON_FATAL); 1846 return (1); 1847 } 1848 p->stats.last_write = getmonotime(); 1849 if (p->holdtime > 0) 1850 timer_set(&p->timers, Timer_SendHold, 1851 p->holdtime < INTERVAL_HOLD ? INTERVAL_HOLD : 1852 p->holdtime); 1853 if (p->throttled && p->wbuf.queued < SESS_MSG_LOW_MARK) { 1854 if (imsg_rde(IMSG_XON, p->conf.id, NULL, 0) == -1) 1855 log_peer_warn(&p->conf, "imsg_compose XON"); 1856 else 1857 p->throttled = 0; 1858 } 1859 if (!(pfd->revents & POLLIN)) 1860 return (1); 1861 } 1862 1863 if (p->rbuf && pfd->revents & POLLIN) { 1864 if ((n = read(p->fd, p->rbuf->buf + p->rbuf->wpos, 1865 sizeof(p->rbuf->buf) - p->rbuf->wpos)) == -1) { 1866 if (errno != EINTR && errno != EAGAIN) { 1867 log_peer_warn(&p->conf, "read error"); 1868 bgp_fsm(p, EVNT_CON_FATAL); 1869 } 1870 return (1); 1871 } 1872 if (n == 0) { /* connection closed */ 1873 bgp_fsm(p, EVNT_CON_CLOSED); 1874 return (1); 1875 } 1876 1877 p->rbuf->wpos += n; 1878 p->stats.last_read = getmonotime(); 1879 return (1); 1880 } 1881 return (0); 1882 } 1883 1884 void 1885 session_process_msg(struct peer *p) 1886 { 1887 struct mrt *mrt; 1888 ssize_t rpos, av, left; 1889 int processed = 0; 1890 uint16_t msglen; 1891 uint8_t msgtype; 1892 1893 rpos = 0; 1894 av = p->rbuf->wpos; 1895 p->rpending = 0; 1896 1897 /* 1898 * session might drop to IDLE -> buffers deallocated 1899 * we MUST check rbuf != NULL before use 1900 */ 1901 for (;;) { 1902 if (p->rbuf == NULL) 1903 return; 1904 if (rpos + MSGSIZE_HEADER > av) 1905 break; 1906 if (parse_header(p, p->rbuf->buf + rpos, &msglen, 1907 &msgtype) == -1) 1908 return; 1909 if (rpos + msglen > av) 1910 break; 1911 p->rbuf->rptr = p->rbuf->buf + rpos; 1912 1913 /* dump to MRT as soon as we have a full packet */ 1914 LIST_FOREACH(mrt, &mrthead, entry) { 1915 if (!(mrt->type == MRT_ALL_IN || (msgtype == UPDATE && 1916 mrt->type == MRT_UPDATE_IN))) 1917 continue; 1918 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1919 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1920 mrt->group_id == p->conf.groupid)) 1921 mrt_dump_bgp_msg(mrt, p->rbuf->rptr, msglen, p, 1922 msgtype); 1923 } 1924 1925 switch (msgtype) { 1926 case OPEN: 1927 bgp_fsm(p, EVNT_RCVD_OPEN); 1928 p->stats.msg_rcvd_open++; 1929 break; 1930 case UPDATE: 1931 bgp_fsm(p, EVNT_RCVD_UPDATE); 1932 p->stats.msg_rcvd_update++; 1933 break; 1934 case NOTIFICATION: 1935 bgp_fsm(p, EVNT_RCVD_NOTIFICATION); 1936 p->stats.msg_rcvd_notification++; 1937 break; 1938 case KEEPALIVE: 1939 bgp_fsm(p, EVNT_RCVD_KEEPALIVE); 1940 p->stats.msg_rcvd_keepalive++; 1941 break; 1942 case RREFRESH: 1943 parse_rrefresh(p); 1944 p->stats.msg_rcvd_rrefresh++; 1945 break; 1946 default: /* cannot happen */ 1947 session_notification(p, ERR_HEADER, ERR_HDR_TYPE, 1948 &msgtype, 1); 1949 log_warnx("received message with unknown type %u", 1950 msgtype); 1951 bgp_fsm(p, EVNT_CON_FATAL); 1952 } 1953 rpos += msglen; 1954 if (++processed > MSG_PROCESS_LIMIT) { 1955 p->rpending = 1; 1956 break; 1957 } 1958 } 1959 1960 if (rpos < av) { 1961 left = av - rpos; 1962 memmove(&p->rbuf->buf, p->rbuf->buf + rpos, left); 1963 p->rbuf->wpos = left; 1964 } else 1965 p->rbuf->wpos = 0; 1966 } 1967 1968 int 1969 parse_header(struct peer *peer, u_char *data, uint16_t *len, uint8_t *type) 1970 { 1971 u_char *p; 1972 uint16_t olen; 1973 static const uint8_t marker[MSGSIZE_HEADER_MARKER] = { 0xff, 0xff, 1974 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 1975 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; 1976 1977 /* caller MUST make sure we are getting 19 bytes! */ 1978 p = data; 1979 if (memcmp(p, marker, sizeof(marker))) { 1980 log_peer_warnx(&peer->conf, "sync error"); 1981 session_notification(peer, ERR_HEADER, ERR_HDR_SYNC, NULL, 0); 1982 bgp_fsm(peer, EVNT_CON_FATAL); 1983 return (-1); 1984 } 1985 p += MSGSIZE_HEADER_MARKER; 1986 1987 memcpy(&olen, p, 2); 1988 *len = ntohs(olen); 1989 p += 2; 1990 memcpy(type, p, 1); 1991 1992 if (*len < MSGSIZE_HEADER || *len > MAX_PKTSIZE) { 1993 log_peer_warnx(&peer->conf, 1994 "received message: illegal length: %u byte", *len); 1995 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1996 &olen, sizeof(olen)); 1997 bgp_fsm(peer, EVNT_CON_FATAL); 1998 return (-1); 1999 } 2000 2001 switch (*type) { 2002 case OPEN: 2003 if (*len < MSGSIZE_OPEN_MIN) { 2004 log_peer_warnx(&peer->conf, 2005 "received OPEN: illegal len: %u byte", *len); 2006 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2007 &olen, sizeof(olen)); 2008 bgp_fsm(peer, EVNT_CON_FATAL); 2009 return (-1); 2010 } 2011 break; 2012 case NOTIFICATION: 2013 if (*len < MSGSIZE_NOTIFICATION_MIN) { 2014 log_peer_warnx(&peer->conf, 2015 "received NOTIFICATION: illegal len: %u byte", 2016 *len); 2017 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2018 &olen, sizeof(olen)); 2019 bgp_fsm(peer, EVNT_CON_FATAL); 2020 return (-1); 2021 } 2022 break; 2023 case UPDATE: 2024 if (*len < MSGSIZE_UPDATE_MIN) { 2025 log_peer_warnx(&peer->conf, 2026 "received UPDATE: illegal len: %u byte", *len); 2027 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2028 &olen, sizeof(olen)); 2029 bgp_fsm(peer, EVNT_CON_FATAL); 2030 return (-1); 2031 } 2032 break; 2033 case KEEPALIVE: 2034 if (*len != MSGSIZE_KEEPALIVE) { 2035 log_peer_warnx(&peer->conf, 2036 "received KEEPALIVE: illegal len: %u byte", *len); 2037 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2038 &olen, sizeof(olen)); 2039 bgp_fsm(peer, EVNT_CON_FATAL); 2040 return (-1); 2041 } 2042 break; 2043 case RREFRESH: 2044 if (*len < MSGSIZE_RREFRESH_MIN) { 2045 log_peer_warnx(&peer->conf, 2046 "received RREFRESH: illegal len: %u byte", *len); 2047 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2048 &olen, sizeof(olen)); 2049 bgp_fsm(peer, EVNT_CON_FATAL); 2050 return (-1); 2051 } 2052 break; 2053 default: 2054 log_peer_warnx(&peer->conf, 2055 "received msg with unknown type %u", *type); 2056 session_notification(peer, ERR_HEADER, ERR_HDR_TYPE, 2057 type, 1); 2058 bgp_fsm(peer, EVNT_CON_FATAL); 2059 return (-1); 2060 } 2061 return (0); 2062 } 2063 2064 int 2065 parse_open(struct peer *peer) 2066 { 2067 u_char *p, *op_val; 2068 uint8_t version, rversion; 2069 uint16_t short_as, msglen; 2070 uint16_t holdtime, oholdtime, myholdtime; 2071 uint32_t as, bgpid; 2072 uint16_t optparamlen, extlen, plen, op_len; 2073 uint8_t op_type; 2074 2075 p = peer->rbuf->rptr; 2076 p += MSGSIZE_HEADER_MARKER; 2077 memcpy(&msglen, p, sizeof(msglen)); 2078 msglen = ntohs(msglen); 2079 2080 p = peer->rbuf->rptr; 2081 p += MSGSIZE_HEADER; /* header is already checked */ 2082 2083 memcpy(&version, p, sizeof(version)); 2084 p += sizeof(version); 2085 2086 if (version != BGP_VERSION) { 2087 log_peer_warnx(&peer->conf, 2088 "peer wants unrecognized version %u", version); 2089 if (version > BGP_VERSION) 2090 rversion = version - BGP_VERSION; 2091 else 2092 rversion = BGP_VERSION; 2093 session_notification(peer, ERR_OPEN, ERR_OPEN_VERSION, 2094 &rversion, sizeof(rversion)); 2095 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2096 return (-1); 2097 } 2098 2099 memcpy(&short_as, p, sizeof(short_as)); 2100 p += sizeof(short_as); 2101 as = peer->short_as = ntohs(short_as); 2102 if (as == 0) { 2103 log_peer_warnx(&peer->conf, 2104 "peer requests unacceptable AS %u", as); 2105 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, 2106 NULL, 0); 2107 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2108 return (-1); 2109 } 2110 2111 memcpy(&oholdtime, p, sizeof(oholdtime)); 2112 p += sizeof(oholdtime); 2113 2114 holdtime = ntohs(oholdtime); 2115 if (holdtime && holdtime < peer->conf.min_holdtime) { 2116 log_peer_warnx(&peer->conf, 2117 "peer requests unacceptable holdtime %u", holdtime); 2118 session_notification(peer, ERR_OPEN, ERR_OPEN_HOLDTIME, 2119 NULL, 0); 2120 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2121 return (-1); 2122 } 2123 2124 myholdtime = peer->conf.holdtime; 2125 if (!myholdtime) 2126 myholdtime = conf->holdtime; 2127 if (holdtime < myholdtime) 2128 peer->holdtime = holdtime; 2129 else 2130 peer->holdtime = myholdtime; 2131 2132 memcpy(&bgpid, p, sizeof(bgpid)); 2133 p += sizeof(bgpid); 2134 2135 /* check bgpid for validity - just disallow 0 */ 2136 if (ntohl(bgpid) == 0) { 2137 log_peer_warnx(&peer->conf, "peer BGPID %u unacceptable", 2138 ntohl(bgpid)); 2139 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2140 NULL, 0); 2141 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2142 return (-1); 2143 } 2144 peer->remote_bgpid = bgpid; 2145 2146 extlen = 0; 2147 optparamlen = *p++; 2148 2149 if (optparamlen == 0) { 2150 if (msglen != MSGSIZE_OPEN_MIN) { 2151 bad_len: 2152 log_peer_warnx(&peer->conf, 2153 "corrupt OPEN message received: length mismatch"); 2154 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2155 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2156 return (-1); 2157 } 2158 } else { 2159 if (msglen < MSGSIZE_OPEN_MIN + 1) 2160 goto bad_len; 2161 2162 op_type = *p; 2163 if (op_type == OPT_PARAM_EXT_LEN) { 2164 p++; 2165 memcpy(&optparamlen, p, sizeof(optparamlen)); 2166 optparamlen = ntohs(optparamlen); 2167 p += sizeof(optparamlen); 2168 extlen = 1; 2169 } 2170 2171 /* RFC9020 encoding has 3 extra bytes */ 2172 if (optparamlen + 3 * extlen != msglen - MSGSIZE_OPEN_MIN) 2173 goto bad_len; 2174 } 2175 2176 plen = optparamlen; 2177 while (plen > 0) { 2178 if (plen < 2 + extlen) 2179 goto bad_len; 2180 2181 memcpy(&op_type, p, sizeof(op_type)); 2182 p += sizeof(op_type); 2183 plen -= sizeof(op_type); 2184 if (!extlen) { 2185 op_len = *p++; 2186 plen--; 2187 } else { 2188 memcpy(&op_len, p, sizeof(op_len)); 2189 op_len = ntohs(op_len); 2190 p += sizeof(op_len); 2191 plen -= sizeof(op_len); 2192 } 2193 if (op_len > 0) { 2194 if (plen < op_len) 2195 goto bad_len; 2196 op_val = p; 2197 p += op_len; 2198 plen -= op_len; 2199 } else 2200 op_val = NULL; 2201 2202 switch (op_type) { 2203 case OPT_PARAM_CAPABILITIES: /* RFC 3392 */ 2204 if (parse_capabilities(peer, op_val, op_len, 2205 &as) == -1) { 2206 session_notification(peer, ERR_OPEN, 0, 2207 NULL, 0); 2208 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2209 return (-1); 2210 } 2211 break; 2212 case OPT_PARAM_AUTH: /* deprecated */ 2213 default: 2214 /* 2215 * unsupported type 2216 * the RFCs tell us to leave the data section empty 2217 * and notify the peer with ERR_OPEN, ERR_OPEN_OPT. 2218 * How the peer should know _which_ optional parameter 2219 * we don't support is beyond me. 2220 */ 2221 log_peer_warnx(&peer->conf, 2222 "received OPEN message with unsupported optional " 2223 "parameter: type %u", op_type); 2224 session_notification(peer, ERR_OPEN, ERR_OPEN_OPT, 2225 NULL, 0); 2226 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2227 /* no punish */ 2228 timer_set(&peer->timers, Timer_IdleHold, 0); 2229 peer->IdleHoldTime /= 2; 2230 return (-1); 2231 } 2232 } 2233 2234 /* if remote-as is zero and it's a cloned neighbor, accept any */ 2235 if (peer->template && !peer->conf.remote_as && as != AS_TRANS) { 2236 peer->conf.remote_as = as; 2237 peer->conf.ebgp = (peer->conf.remote_as != peer->conf.local_as); 2238 if (!peer->conf.ebgp) 2239 /* force enforce_as off for iBGP sessions */ 2240 peer->conf.enforce_as = ENFORCE_AS_OFF; 2241 } 2242 2243 if (peer->conf.remote_as != as) { 2244 log_peer_warnx(&peer->conf, "peer sent wrong AS %s", 2245 log_as(as)); 2246 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, NULL, 0); 2247 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2248 return (-1); 2249 } 2250 2251 /* on iBGP sessions check for bgpid collision */ 2252 if (!peer->conf.ebgp && peer->remote_bgpid == conf->bgpid) { 2253 log_peer_warnx(&peer->conf, "peer BGPID %u conflicts with ours", 2254 ntohl(bgpid)); 2255 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2256 NULL, 0); 2257 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2258 return (-1); 2259 } 2260 2261 if (capa_neg_calc(peer) == -1) { 2262 log_peer_warnx(&peer->conf, 2263 "capability negotiation calculation failed"); 2264 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2265 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2266 return (-1); 2267 } 2268 2269 return (0); 2270 } 2271 2272 int 2273 parse_update(struct peer *peer) 2274 { 2275 u_char *p; 2276 uint16_t datalen; 2277 2278 /* 2279 * we pass the message verbatim to the rde. 2280 * in case of errors the whole session is reset with a 2281 * notification anyway, we only need to know the peer 2282 */ 2283 p = peer->rbuf->rptr; 2284 p += MSGSIZE_HEADER_MARKER; 2285 memcpy(&datalen, p, sizeof(datalen)); 2286 datalen = ntohs(datalen); 2287 2288 p = peer->rbuf->rptr; 2289 p += MSGSIZE_HEADER; /* header is already checked */ 2290 datalen -= MSGSIZE_HEADER; 2291 2292 if (imsg_rde(IMSG_UPDATE, peer->conf.id, p, datalen) == -1) 2293 return (-1); 2294 2295 return (0); 2296 } 2297 2298 int 2299 parse_rrefresh(struct peer *peer) 2300 { 2301 struct route_refresh rr; 2302 uint16_t afi, datalen; 2303 uint8_t aid, safi, subtype; 2304 u_char *p; 2305 2306 p = peer->rbuf->rptr; 2307 p += MSGSIZE_HEADER_MARKER; 2308 memcpy(&datalen, p, sizeof(datalen)); 2309 datalen = ntohs(datalen); 2310 2311 p = peer->rbuf->rptr; 2312 p += MSGSIZE_HEADER; /* header is already checked */ 2313 2314 /* 2315 * We could check if we actually announced the capability but 2316 * as long as the message is correctly encoded we don't care. 2317 */ 2318 2319 /* afi, 2 byte */ 2320 memcpy(&afi, p, sizeof(afi)); 2321 afi = ntohs(afi); 2322 p += 2; 2323 /* subtype, 1 byte */ 2324 subtype = *p; 2325 p += 1; 2326 /* safi, 1 byte */ 2327 safi = *p; 2328 2329 /* check subtype if peer announced enhanced route refresh */ 2330 if (peer->capa.neg.enhanced_rr) { 2331 switch (subtype) { 2332 case ROUTE_REFRESH_REQUEST: 2333 /* no ORF support, so no oversized RREFRESH msgs */ 2334 if (datalen != MSGSIZE_RREFRESH) { 2335 log_peer_warnx(&peer->conf, 2336 "received RREFRESH: illegal len: %u byte", 2337 datalen); 2338 datalen = htons(datalen); 2339 session_notification(peer, ERR_HEADER, 2340 ERR_HDR_LEN, &datalen, sizeof(datalen)); 2341 bgp_fsm(peer, EVNT_CON_FATAL); 2342 return (-1); 2343 } 2344 peer->stats.refresh_rcvd_req++; 2345 break; 2346 case ROUTE_REFRESH_BEGIN_RR: 2347 case ROUTE_REFRESH_END_RR: 2348 /* special handling for RFC7313 */ 2349 if (datalen != MSGSIZE_RREFRESH) { 2350 log_peer_warnx(&peer->conf, 2351 "received RREFRESH: illegal len: %u byte", 2352 datalen); 2353 p = peer->rbuf->rptr; 2354 p += MSGSIZE_HEADER; 2355 datalen -= MSGSIZE_HEADER; 2356 session_notification(peer, ERR_RREFRESH, 2357 ERR_RR_INV_LEN, p, datalen); 2358 bgp_fsm(peer, EVNT_CON_FATAL); 2359 return (-1); 2360 } 2361 if (subtype == ROUTE_REFRESH_BEGIN_RR) 2362 peer->stats.refresh_rcvd_borr++; 2363 else 2364 peer->stats.refresh_rcvd_eorr++; 2365 break; 2366 default: 2367 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2368 "bad subtype %d", subtype); 2369 return (0); 2370 } 2371 } else { 2372 /* force subtype to default */ 2373 subtype = ROUTE_REFRESH_REQUEST; 2374 peer->stats.refresh_rcvd_req++; 2375 } 2376 2377 /* afi/safi unchecked - unrecognized values will be ignored anyway */ 2378 if (afi2aid(afi, safi, &aid) == -1) { 2379 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2380 "invalid afi/safi pair"); 2381 return (0); 2382 } 2383 2384 if (!peer->capa.neg.refresh && !peer->capa.neg.enhanced_rr) { 2385 log_peer_warnx(&peer->conf, "peer sent unexpected refresh"); 2386 return (0); 2387 } 2388 2389 rr.aid = aid; 2390 rr.subtype = subtype; 2391 2392 if (imsg_rde(IMSG_REFRESH, peer->conf.id, &rr, sizeof(rr)) == -1) 2393 return (-1); 2394 2395 return (0); 2396 } 2397 2398 int 2399 parse_notification(struct peer *peer) 2400 { 2401 u_char *p; 2402 uint16_t datalen; 2403 uint8_t errcode; 2404 uint8_t subcode; 2405 uint8_t capa_code; 2406 uint8_t capa_len; 2407 size_t reason_len; 2408 uint8_t i; 2409 2410 /* just log */ 2411 p = peer->rbuf->rptr; 2412 p += MSGSIZE_HEADER_MARKER; 2413 memcpy(&datalen, p, sizeof(datalen)); 2414 datalen = ntohs(datalen); 2415 2416 p = peer->rbuf->rptr; 2417 p += MSGSIZE_HEADER; /* header is already checked */ 2418 datalen -= MSGSIZE_HEADER; 2419 2420 memcpy(&errcode, p, sizeof(errcode)); 2421 p += sizeof(errcode); 2422 datalen -= sizeof(errcode); 2423 2424 memcpy(&subcode, p, sizeof(subcode)); 2425 p += sizeof(subcode); 2426 datalen -= sizeof(subcode); 2427 2428 log_notification(peer, errcode, subcode, p, datalen, "received"); 2429 peer->errcnt++; 2430 peer->stats.last_rcvd_errcode = errcode; 2431 peer->stats.last_rcvd_suberr = subcode; 2432 2433 if (errcode == ERR_OPEN && subcode == ERR_OPEN_CAPA) { 2434 if (datalen == 0) { /* zebra likes to send those.. humbug */ 2435 log_peer_warnx(&peer->conf, "received \"unsupported " 2436 "capability\" notification without data part, " 2437 "disabling capability announcements altogether"); 2438 session_capa_ann_none(peer); 2439 } 2440 2441 while (datalen > 0) { 2442 if (datalen < 2) { 2443 log_peer_warnx(&peer->conf, 2444 "parse_notification: " 2445 "expect len >= 2, len is %u", datalen); 2446 return (-1); 2447 } 2448 memcpy(&capa_code, p, sizeof(capa_code)); 2449 p += sizeof(capa_code); 2450 datalen -= sizeof(capa_code); 2451 memcpy(&capa_len, p, sizeof(capa_len)); 2452 p += sizeof(capa_len); 2453 datalen -= sizeof(capa_len); 2454 if (datalen < capa_len) { 2455 log_peer_warnx(&peer->conf, 2456 "parse_notification: capa_len %u exceeds " 2457 "remaining msg length %u", capa_len, 2458 datalen); 2459 return (-1); 2460 } 2461 p += capa_len; 2462 datalen -= capa_len; 2463 switch (capa_code) { 2464 case CAPA_MP: 2465 for (i = 0; i < AID_MAX; i++) 2466 peer->capa.ann.mp[i] = 0; 2467 log_peer_warnx(&peer->conf, 2468 "disabling multiprotocol capability"); 2469 break; 2470 case CAPA_REFRESH: 2471 peer->capa.ann.refresh = 0; 2472 log_peer_warnx(&peer->conf, 2473 "disabling route refresh capability"); 2474 break; 2475 case CAPA_RESTART: 2476 peer->capa.ann.grestart.restart = 0; 2477 log_peer_warnx(&peer->conf, 2478 "disabling restart capability"); 2479 break; 2480 case CAPA_AS4BYTE: 2481 peer->capa.ann.as4byte = 0; 2482 log_peer_warnx(&peer->conf, 2483 "disabling 4-byte AS num capability"); 2484 break; 2485 case CAPA_ADD_PATH: 2486 memset(peer->capa.ann.add_path, 0, 2487 sizeof(peer->capa.ann.add_path)); 2488 log_peer_warnx(&peer->conf, 2489 "disabling ADD-PATH capability"); 2490 break; 2491 case CAPA_ENHANCED_RR: 2492 peer->capa.ann.enhanced_rr = 0; 2493 log_peer_warnx(&peer->conf, 2494 "disabling enhanced route refresh " 2495 "capability"); 2496 break; 2497 default: /* should not happen... */ 2498 log_peer_warnx(&peer->conf, "received " 2499 "\"unsupported capability\" notification " 2500 "for unknown capability %u, disabling " 2501 "capability announcements altogether", 2502 capa_code); 2503 session_capa_ann_none(peer); 2504 break; 2505 } 2506 } 2507 2508 return (1); 2509 } 2510 2511 if (errcode == ERR_OPEN && subcode == ERR_OPEN_OPT) { 2512 session_capa_ann_none(peer); 2513 return (1); 2514 } 2515 2516 if (errcode == ERR_CEASE && 2517 (subcode == ERR_CEASE_ADMIN_DOWN || 2518 subcode == ERR_CEASE_ADMIN_RESET)) { 2519 if (datalen > 1) { 2520 reason_len = *p++; 2521 datalen--; 2522 if (datalen < reason_len) { 2523 log_peer_warnx(&peer->conf, 2524 "received truncated shutdown reason"); 2525 return (0); 2526 } 2527 if (reason_len > REASON_LEN - 1) { 2528 log_peer_warnx(&peer->conf, 2529 "received overly long shutdown reason"); 2530 return (0); 2531 } 2532 memcpy(peer->stats.last_reason, p, reason_len); 2533 peer->stats.last_reason[reason_len] = '\0'; 2534 log_peer_warnx(&peer->conf, 2535 "received shutdown reason: \"%s\"", 2536 log_reason(peer->stats.last_reason)); 2537 p += reason_len; 2538 datalen -= reason_len; 2539 } 2540 } 2541 2542 return (0); 2543 } 2544 2545 int 2546 parse_capabilities(struct peer *peer, u_char *d, uint16_t dlen, uint32_t *as) 2547 { 2548 u_char *capa_val; 2549 uint32_t remote_as; 2550 uint16_t len; 2551 uint16_t afi; 2552 uint16_t gr_header; 2553 uint8_t safi; 2554 uint8_t aid; 2555 uint8_t flags; 2556 uint8_t capa_code; 2557 uint8_t capa_len; 2558 uint8_t i; 2559 2560 len = dlen; 2561 while (len > 0) { 2562 if (len < 2) { 2563 log_peer_warnx(&peer->conf, "Bad capabilities attr " 2564 "length: %u, too short", len); 2565 return (-1); 2566 } 2567 memcpy(&capa_code, d, sizeof(capa_code)); 2568 d += sizeof(capa_code); 2569 len -= sizeof(capa_code); 2570 memcpy(&capa_len, d, sizeof(capa_len)); 2571 d += sizeof(capa_len); 2572 len -= sizeof(capa_len); 2573 if (capa_len > 0) { 2574 if (len < capa_len) { 2575 log_peer_warnx(&peer->conf, 2576 "Bad capabilities attr length: " 2577 "len %u smaller than capa_len %u", 2578 len, capa_len); 2579 return (-1); 2580 } 2581 capa_val = d; 2582 d += capa_len; 2583 len -= capa_len; 2584 } else 2585 capa_val = NULL; 2586 2587 switch (capa_code) { 2588 case CAPA_MP: /* RFC 4760 */ 2589 if (capa_len != 4) { 2590 log_peer_warnx(&peer->conf, 2591 "Bad multi protocol capability length: " 2592 "%u", capa_len); 2593 break; 2594 } 2595 memcpy(&afi, capa_val, sizeof(afi)); 2596 afi = ntohs(afi); 2597 memcpy(&safi, capa_val + 3, sizeof(safi)); 2598 if (afi2aid(afi, safi, &aid) == -1) { 2599 log_peer_warnx(&peer->conf, 2600 "Received multi protocol capability: " 2601 " unknown AFI %u, safi %u pair", 2602 afi, safi); 2603 break; 2604 } 2605 peer->capa.peer.mp[aid] = 1; 2606 break; 2607 case CAPA_REFRESH: 2608 peer->capa.peer.refresh = 1; 2609 break; 2610 case CAPA_RESTART: 2611 if (capa_len == 2) { 2612 /* peer only supports EoR marker */ 2613 peer->capa.peer.grestart.restart = 1; 2614 peer->capa.peer.grestart.timeout = 0; 2615 break; 2616 } else if (capa_len % 4 != 2) { 2617 log_peer_warnx(&peer->conf, 2618 "Bad graceful restart capability length: " 2619 "%u", capa_len); 2620 peer->capa.peer.grestart.restart = 0; 2621 peer->capa.peer.grestart.timeout = 0; 2622 break; 2623 } 2624 2625 memcpy(&gr_header, capa_val, sizeof(gr_header)); 2626 gr_header = ntohs(gr_header); 2627 peer->capa.peer.grestart.timeout = 2628 gr_header & CAPA_GR_TIMEMASK; 2629 if (peer->capa.peer.grestart.timeout == 0) { 2630 log_peer_warnx(&peer->conf, "Received " 2631 "graceful restart timeout is zero"); 2632 peer->capa.peer.grestart.restart = 0; 2633 break; 2634 } 2635 2636 for (i = 2; i <= capa_len - 4; i += 4) { 2637 memcpy(&afi, capa_val + i, sizeof(afi)); 2638 afi = ntohs(afi); 2639 safi = capa_val[i + 2]; 2640 flags = capa_val[i + 3]; 2641 if (afi2aid(afi, safi, &aid) == -1) { 2642 log_peer_warnx(&peer->conf, 2643 "Received graceful restart capa: " 2644 " unknown AFI %u, safi %u pair", 2645 afi, safi); 2646 continue; 2647 } 2648 peer->capa.peer.grestart.flags[aid] |= 2649 CAPA_GR_PRESENT; 2650 if (flags & CAPA_GR_F_FLAG) 2651 peer->capa.peer.grestart.flags[aid] |= 2652 CAPA_GR_FORWARD; 2653 if (gr_header & CAPA_GR_R_FLAG) 2654 peer->capa.peer.grestart.flags[aid] |= 2655 CAPA_GR_RESTART; 2656 peer->capa.peer.grestart.restart = 2; 2657 } 2658 break; 2659 case CAPA_AS4BYTE: 2660 if (capa_len != 4) { 2661 log_peer_warnx(&peer->conf, 2662 "Bad AS4BYTE capability length: " 2663 "%u", capa_len); 2664 peer->capa.peer.as4byte = 0; 2665 break; 2666 } 2667 memcpy(&remote_as, capa_val, sizeof(remote_as)); 2668 *as = ntohl(remote_as); 2669 if (*as == 0) { 2670 log_peer_warnx(&peer->conf, 2671 "peer requests unacceptable AS %u", *as); 2672 session_notification(peer, ERR_OPEN, 2673 ERR_OPEN_AS, NULL, 0); 2674 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2675 return (-1); 2676 } 2677 peer->capa.peer.as4byte = 1; 2678 break; 2679 case CAPA_ADD_PATH: 2680 if (capa_len % 4 != 0) { 2681 log_peer_warnx(&peer->conf, 2682 "Bad ADD-PATH capability length: " 2683 "%u", capa_len); 2684 memset(peer->capa.peer.add_path, 0, 2685 sizeof(peer->capa.peer.add_path)); 2686 break; 2687 } 2688 for (i = 0; i <= capa_len - 4; i += 4) { 2689 memcpy(&afi, capa_val + i, sizeof(afi)); 2690 afi = ntohs(afi); 2691 safi = capa_val[i + 2]; 2692 flags = capa_val[i + 3]; 2693 if (afi2aid(afi, safi, &aid) == -1) { 2694 log_peer_warnx(&peer->conf, 2695 "Received ADD-PATH capa: " 2696 " unknown AFI %u, safi %u pair", 2697 afi, safi); 2698 memset(peer->capa.peer.add_path, 0, 2699 sizeof(peer->capa.peer.add_path)); 2700 break; 2701 } 2702 if (flags & ~CAPA_AP_BIDIR) { 2703 log_peer_warnx(&peer->conf, 2704 "Received ADD-PATH capa: " 2705 " bad flags %x", flags); 2706 memset(peer->capa.peer.add_path, 0, 2707 sizeof(peer->capa.peer.add_path)); 2708 break; 2709 } 2710 peer->capa.peer.add_path[aid] = flags; 2711 } 2712 break; 2713 case CAPA_ENHANCED_RR: 2714 peer->capa.peer.enhanced_rr = 1; 2715 break; 2716 default: 2717 break; 2718 } 2719 } 2720 2721 return (0); 2722 } 2723 2724 int 2725 capa_neg_calc(struct peer *p) 2726 { 2727 uint8_t i, hasmp = 0; 2728 2729 /* a capability is accepted only if both sides announced it */ 2730 2731 p->capa.neg.refresh = 2732 (p->capa.ann.refresh && p->capa.peer.refresh) != 0; 2733 p->capa.neg.enhanced_rr = 2734 (p->capa.ann.enhanced_rr && p->capa.peer.enhanced_rr) != 0; 2735 2736 p->capa.neg.as4byte = 2737 (p->capa.ann.as4byte && p->capa.peer.as4byte) != 0; 2738 2739 /* MP: both side must agree on the AFI,SAFI pair */ 2740 for (i = 0; i < AID_MAX; i++) { 2741 if (p->capa.ann.mp[i] && p->capa.peer.mp[i]) 2742 p->capa.neg.mp[i] = 1; 2743 else 2744 p->capa.neg.mp[i] = 0; 2745 if (p->capa.ann.mp[i]) 2746 hasmp = 1; 2747 } 2748 /* if no MP capability present default to IPv4 unicast mode */ 2749 if (!hasmp) 2750 p->capa.neg.mp[AID_INET] = 1; 2751 2752 /* 2753 * graceful restart: the peer capabilities are of interest here. 2754 * It is necessary to compare the new values with the previous ones 2755 * and act acordingly. AFI/SAFI that are not part in the MP capability 2756 * are treated as not being present. 2757 * Also make sure that a flush happens if the session stopped 2758 * supporting graceful restart. 2759 */ 2760 2761 for (i = 0; i < AID_MAX; i++) { 2762 int8_t negflags; 2763 2764 /* disable GR if the AFI/SAFI is not present */ 2765 if ((p->capa.peer.grestart.flags[i] & CAPA_GR_PRESENT && 2766 p->capa.neg.mp[i] == 0)) 2767 p->capa.peer.grestart.flags[i] = 0; /* disable */ 2768 /* look at current GR state and decide what to do */ 2769 negflags = p->capa.neg.grestart.flags[i]; 2770 p->capa.neg.grestart.flags[i] = p->capa.peer.grestart.flags[i]; 2771 if (negflags & CAPA_GR_RESTARTING) { 2772 if (p->capa.ann.grestart.restart != 0 && 2773 p->capa.peer.grestart.flags[i] & CAPA_GR_FORWARD) { 2774 p->capa.neg.grestart.flags[i] |= 2775 CAPA_GR_RESTARTING; 2776 } else { 2777 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 2778 &i, sizeof(i)) == -1) 2779 return (-1); 2780 log_peer_warnx(&p->conf, "graceful restart of " 2781 "%s, not restarted, flushing", aid2str(i)); 2782 } 2783 } 2784 } 2785 p->capa.neg.grestart.timeout = p->capa.peer.grestart.timeout; 2786 p->capa.neg.grestart.restart = p->capa.peer.grestart.restart; 2787 if (p->capa.ann.grestart.restart == 0) 2788 p->capa.neg.grestart.restart = 0; 2789 2790 2791 /* 2792 * ADD-PATH: set only those bits where both sides agree. 2793 * For this compare our send bit with the recv bit from the peer 2794 * and vice versa. 2795 * The flags are stored from this systems view point. 2796 */ 2797 memset(p->capa.neg.add_path, 0, sizeof(p->capa.neg.add_path)); 2798 if (p->capa.ann.add_path[0]) { 2799 for (i = AID_MIN; i < AID_MAX; i++) { 2800 if ((p->capa.ann.add_path[i] & CAPA_AP_RECV) && 2801 (p->capa.peer.add_path[i] & CAPA_AP_SEND)) { 2802 p->capa.neg.add_path[i] |= CAPA_AP_RECV; 2803 p->capa.neg.add_path[0] |= CAPA_AP_RECV; 2804 } 2805 if ((p->capa.ann.add_path[i] & CAPA_AP_SEND) && 2806 (p->capa.peer.add_path[i] & CAPA_AP_RECV)) { 2807 p->capa.neg.add_path[i] |= CAPA_AP_SEND; 2808 p->capa.neg.add_path[0] |= CAPA_AP_SEND; 2809 } 2810 } 2811 } 2812 2813 return (0); 2814 } 2815 2816 void 2817 session_dispatch_imsg(struct imsgbuf *ibuf, int idx, u_int *listener_cnt) 2818 { 2819 struct imsg imsg; 2820 struct mrt xmrt; 2821 struct route_refresh rr; 2822 struct mrt *mrt; 2823 struct imsgbuf *i; 2824 struct peer *p; 2825 struct listen_addr *la, *nla; 2826 struct kif *kif; 2827 u_char *data; 2828 int n, fd, depend_ok, restricted; 2829 uint16_t t; 2830 uint8_t aid, errcode, subcode; 2831 2832 while (ibuf) { 2833 if ((n = imsg_get(ibuf, &imsg)) == -1) 2834 fatal("session_dispatch_imsg: imsg_get error"); 2835 2836 if (n == 0) 2837 break; 2838 2839 switch (imsg.hdr.type) { 2840 case IMSG_SOCKET_CONN: 2841 case IMSG_SOCKET_CONN_CTL: 2842 if (idx != PFD_PIPE_MAIN) 2843 fatalx("reconf request not from parent"); 2844 if ((fd = imsg.fd) == -1) { 2845 log_warnx("expected to receive imsg fd to " 2846 "RDE but didn't receive any"); 2847 break; 2848 } 2849 if ((i = malloc(sizeof(struct imsgbuf))) == NULL) 2850 fatal(NULL); 2851 imsg_init(i, fd); 2852 if (imsg.hdr.type == IMSG_SOCKET_CONN) { 2853 if (ibuf_rde) { 2854 log_warnx("Unexpected imsg connection " 2855 "to RDE received"); 2856 msgbuf_clear(&ibuf_rde->w); 2857 free(ibuf_rde); 2858 } 2859 ibuf_rde = i; 2860 } else { 2861 if (ibuf_rde_ctl) { 2862 log_warnx("Unexpected imsg ctl " 2863 "connection to RDE received"); 2864 msgbuf_clear(&ibuf_rde_ctl->w); 2865 free(ibuf_rde_ctl); 2866 } 2867 ibuf_rde_ctl = i; 2868 } 2869 break; 2870 case IMSG_RECONF_CONF: 2871 if (idx != PFD_PIPE_MAIN) 2872 fatalx("reconf request not from parent"); 2873 nconf = new_config(); 2874 2875 copy_config(nconf, imsg.data); 2876 pending_reconf = 1; 2877 break; 2878 case IMSG_RECONF_PEER: 2879 if (idx != PFD_PIPE_MAIN) 2880 fatalx("reconf request not from parent"); 2881 if ((p = calloc(1, sizeof(struct peer))) == NULL) 2882 fatal("new_peer"); 2883 memcpy(&p->conf, imsg.data, sizeof(struct peer_config)); 2884 p->state = p->prev_state = STATE_NONE; 2885 p->reconf_action = RECONF_REINIT; 2886 if (RB_INSERT(peer_head, &nconf->peers, p) != NULL) 2887 fatalx("%s: peer tree is corrupt", __func__); 2888 break; 2889 case IMSG_RECONF_LISTENER: 2890 if (idx != PFD_PIPE_MAIN) 2891 fatalx("reconf request not from parent"); 2892 if (nconf == NULL) 2893 fatalx("IMSG_RECONF_LISTENER but no config"); 2894 nla = imsg.data; 2895 TAILQ_FOREACH(la, conf->listen_addrs, entry) 2896 if (!la_cmp(la, nla)) 2897 break; 2898 2899 if (la == NULL) { 2900 if (nla->reconf != RECONF_REINIT) 2901 fatalx("king bula sez: " 2902 "expected REINIT"); 2903 2904 if ((nla->fd = imsg.fd) == -1) 2905 log_warnx("expected to receive fd for " 2906 "%s but didn't receive any", 2907 log_sockaddr((struct sockaddr *) 2908 &nla->sa, nla->sa_len)); 2909 2910 la = calloc(1, sizeof(struct listen_addr)); 2911 if (la == NULL) 2912 fatal(NULL); 2913 memcpy(&la->sa, &nla->sa, sizeof(la->sa)); 2914 la->flags = nla->flags; 2915 la->fd = nla->fd; 2916 la->reconf = RECONF_REINIT; 2917 TAILQ_INSERT_TAIL(nconf->listen_addrs, la, 2918 entry); 2919 } else { 2920 if (nla->reconf != RECONF_KEEP) 2921 fatalx("king bula sez: expected KEEP"); 2922 la->reconf = RECONF_KEEP; 2923 } 2924 2925 break; 2926 case IMSG_RECONF_CTRL: 2927 if (idx != PFD_PIPE_MAIN) 2928 fatalx("reconf request not from parent"); 2929 if (imsg.hdr.len != IMSG_HEADER_SIZE + 2930 sizeof(restricted)) 2931 fatalx("IFINFO imsg with wrong len"); 2932 memcpy(&restricted, imsg.data, sizeof(restricted)); 2933 if (imsg.fd == -1) { 2934 log_warnx("expected to receive fd for control " 2935 "socket but didn't receive any"); 2936 break; 2937 } 2938 if (restricted) { 2939 control_shutdown(rcsock); 2940 rcsock = imsg.fd; 2941 } else { 2942 control_shutdown(csock); 2943 csock = imsg.fd; 2944 } 2945 break; 2946 case IMSG_RECONF_DRAIN: 2947 switch (idx) { 2948 case PFD_PIPE_ROUTE: 2949 if (nconf != NULL) 2950 fatalx("got unexpected %s from RDE", 2951 "IMSG_RECONF_DONE"); 2952 imsg_compose(ibuf_main, IMSG_RECONF_DONE, 0, 0, 2953 -1, NULL, 0); 2954 break; 2955 case PFD_PIPE_MAIN: 2956 if (nconf == NULL) 2957 fatalx("got unexpected %s from parent", 2958 "IMSG_RECONF_DONE"); 2959 imsg_compose(ibuf_main, IMSG_RECONF_DRAIN, 0, 0, 2960 -1, NULL, 0); 2961 break; 2962 default: 2963 fatalx("reconf request not from parent or RDE"); 2964 } 2965 break; 2966 case IMSG_RECONF_DONE: 2967 if (idx != PFD_PIPE_MAIN) 2968 fatalx("reconf request not from parent"); 2969 if (nconf == NULL) 2970 fatalx("got IMSG_RECONF_DONE but no config"); 2971 copy_config(conf, nconf); 2972 merge_peers(conf, nconf); 2973 2974 /* delete old listeners */ 2975 for (la = TAILQ_FIRST(conf->listen_addrs); la != NULL; 2976 la = nla) { 2977 nla = TAILQ_NEXT(la, entry); 2978 if (la->reconf == RECONF_NONE) { 2979 log_info("not listening on %s any more", 2980 log_sockaddr((struct sockaddr *) 2981 &la->sa, la->sa_len)); 2982 TAILQ_REMOVE(conf->listen_addrs, la, 2983 entry); 2984 close(la->fd); 2985 free(la); 2986 } 2987 } 2988 2989 /* add new listeners */ 2990 TAILQ_CONCAT(conf->listen_addrs, nconf->listen_addrs, 2991 entry); 2992 2993 setup_listeners(listener_cnt); 2994 free_config(nconf); 2995 nconf = NULL; 2996 pending_reconf = 0; 2997 log_info("SE reconfigured"); 2998 /* 2999 * IMSG_RECONF_DONE is sent when the RDE drained 3000 * the peer config sent in merge_peers(). 3001 */ 3002 break; 3003 case IMSG_IFINFO: 3004 if (idx != PFD_PIPE_MAIN) 3005 fatalx("IFINFO message not from parent"); 3006 if (imsg.hdr.len != IMSG_HEADER_SIZE + 3007 sizeof(struct kif)) 3008 fatalx("IFINFO imsg with wrong len"); 3009 kif = imsg.data; 3010 depend_ok = kif->depend_state; 3011 3012 RB_FOREACH(p, peer_head, &conf->peers) 3013 if (!strcmp(p->conf.if_depend, kif->ifname)) { 3014 if (depend_ok && !p->depend_ok) { 3015 p->depend_ok = depend_ok; 3016 bgp_fsm(p, EVNT_START); 3017 } else if (!depend_ok && p->depend_ok) { 3018 p->depend_ok = depend_ok; 3019 session_stop(p, 3020 ERR_CEASE_OTHER_CHANGE); 3021 } 3022 } 3023 break; 3024 case IMSG_MRT_OPEN: 3025 case IMSG_MRT_REOPEN: 3026 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3027 sizeof(struct mrt)) { 3028 log_warnx("wrong imsg len"); 3029 break; 3030 } 3031 3032 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 3033 if ((xmrt.wbuf.fd = imsg.fd) == -1) 3034 log_warnx("expected to receive fd for mrt dump " 3035 "but didn't receive any"); 3036 3037 mrt = mrt_get(&mrthead, &xmrt); 3038 if (mrt == NULL) { 3039 /* new dump */ 3040 mrt = calloc(1, sizeof(struct mrt)); 3041 if (mrt == NULL) 3042 fatal("session_dispatch_imsg"); 3043 memcpy(mrt, &xmrt, sizeof(struct mrt)); 3044 TAILQ_INIT(&mrt->wbuf.bufs); 3045 LIST_INSERT_HEAD(&mrthead, mrt, entry); 3046 } else { 3047 /* old dump reopened */ 3048 close(mrt->wbuf.fd); 3049 mrt->wbuf.fd = xmrt.wbuf.fd; 3050 } 3051 break; 3052 case IMSG_MRT_CLOSE: 3053 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3054 sizeof(struct mrt)) { 3055 log_warnx("wrong imsg len"); 3056 break; 3057 } 3058 3059 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 3060 mrt = mrt_get(&mrthead, &xmrt); 3061 if (mrt != NULL) 3062 mrt_done(mrt); 3063 break; 3064 case IMSG_CTL_KROUTE: 3065 case IMSG_CTL_KROUTE_ADDR: 3066 case IMSG_CTL_SHOW_NEXTHOP: 3067 case IMSG_CTL_SHOW_INTERFACE: 3068 case IMSG_CTL_SHOW_FIB_TABLES: 3069 case IMSG_CTL_SHOW_RTR: 3070 case IMSG_CTL_SHOW_TIMER: 3071 if (idx != PFD_PIPE_MAIN) 3072 fatalx("ctl kroute request not from parent"); 3073 control_imsg_relay(&imsg); 3074 break; 3075 case IMSG_CTL_SHOW_RIB: 3076 case IMSG_CTL_SHOW_RIB_PREFIX: 3077 case IMSG_CTL_SHOW_RIB_COMMUNITIES: 3078 case IMSG_CTL_SHOW_RIB_ATTR: 3079 case IMSG_CTL_SHOW_RIB_MEM: 3080 case IMSG_CTL_SHOW_RIB_HASH: 3081 case IMSG_CTL_SHOW_NETWORK: 3082 case IMSG_CTL_SHOW_NEIGHBOR: 3083 case IMSG_CTL_SHOW_SET: 3084 if (idx != PFD_PIPE_ROUTE_CTL) 3085 fatalx("ctl rib request not from RDE"); 3086 control_imsg_relay(&imsg); 3087 break; 3088 case IMSG_CTL_END: 3089 case IMSG_CTL_RESULT: 3090 control_imsg_relay(&imsg); 3091 break; 3092 case IMSG_UPDATE: 3093 if (idx != PFD_PIPE_ROUTE) 3094 fatalx("update request not from RDE"); 3095 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3096 MAX_PKTSIZE - MSGSIZE_HEADER || 3097 imsg.hdr.len < IMSG_HEADER_SIZE + 3098 MSGSIZE_UPDATE_MIN - MSGSIZE_HEADER) 3099 log_warnx("RDE sent invalid update"); 3100 else 3101 session_update(imsg.hdr.peerid, imsg.data, 3102 imsg.hdr.len - IMSG_HEADER_SIZE); 3103 break; 3104 case IMSG_UPDATE_ERR: 3105 if (idx != PFD_PIPE_ROUTE) 3106 fatalx("update request not from RDE"); 3107 if (imsg.hdr.len < IMSG_HEADER_SIZE + 2) { 3108 log_warnx("RDE sent invalid notification"); 3109 break; 3110 } 3111 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3112 log_warnx("no such peer: id=%u", 3113 imsg.hdr.peerid); 3114 break; 3115 } 3116 data = imsg.data; 3117 errcode = *data++; 3118 subcode = *data++; 3119 3120 if (imsg.hdr.len == IMSG_HEADER_SIZE + 2) 3121 data = NULL; 3122 3123 session_notification(p, errcode, subcode, 3124 data, imsg.hdr.len - IMSG_HEADER_SIZE - 2); 3125 switch (errcode) { 3126 case ERR_CEASE: 3127 switch (subcode) { 3128 case ERR_CEASE_MAX_PREFIX: 3129 case ERR_CEASE_MAX_SENT_PREFIX: 3130 t = p->conf.max_out_prefix_restart; 3131 if (subcode == ERR_CEASE_MAX_PREFIX) 3132 t = p->conf.max_prefix_restart; 3133 3134 bgp_fsm(p, EVNT_STOP); 3135 if (t) 3136 timer_set(&p->timers, 3137 Timer_IdleHold, 60 * t); 3138 break; 3139 default: 3140 bgp_fsm(p, EVNT_CON_FATAL); 3141 break; 3142 } 3143 break; 3144 default: 3145 bgp_fsm(p, EVNT_CON_FATAL); 3146 break; 3147 } 3148 break; 3149 case IMSG_REFRESH: 3150 if (idx != PFD_PIPE_ROUTE) 3151 fatalx("route refresh request not from RDE"); 3152 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(rr)) { 3153 log_warnx("RDE sent invalid refresh msg"); 3154 break; 3155 } 3156 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3157 log_warnx("no such peer: id=%u", 3158 imsg.hdr.peerid); 3159 break; 3160 } 3161 memcpy(&rr, imsg.data, sizeof(rr)); 3162 if (rr.aid >= AID_MAX) 3163 fatalx("IMSG_REFRESH: bad AID"); 3164 session_rrefresh(p, rr.aid, rr.subtype); 3165 break; 3166 case IMSG_SESSION_RESTARTED: 3167 if (idx != PFD_PIPE_ROUTE) 3168 fatalx("update request not from RDE"); 3169 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(aid)) { 3170 log_warnx("RDE sent invalid restart msg"); 3171 break; 3172 } 3173 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3174 log_warnx("no such peer: id=%u", 3175 imsg.hdr.peerid); 3176 break; 3177 } 3178 memcpy(&aid, imsg.data, sizeof(aid)); 3179 if (aid >= AID_MAX) 3180 fatalx("IMSG_SESSION_RESTARTED: bad AID"); 3181 if (p->capa.neg.grestart.flags[aid] & 3182 CAPA_GR_RESTARTING) { 3183 log_peer_warnx(&p->conf, 3184 "graceful restart of %s finished", 3185 aid2str(aid)); 3186 p->capa.neg.grestart.flags[aid] &= 3187 ~CAPA_GR_RESTARTING; 3188 timer_stop(&p->timers, Timer_RestartTimeout); 3189 3190 /* signal back to RDE to cleanup stale routes */ 3191 if (imsg_rde(IMSG_SESSION_RESTARTED, 3192 imsg.hdr.peerid, &aid, sizeof(aid)) == -1) 3193 fatal("imsg_compose: " 3194 "IMSG_SESSION_RESTARTED"); 3195 } 3196 break; 3197 case IMSG_SESSION_DOWN: 3198 if (idx != PFD_PIPE_ROUTE) 3199 fatalx("update request not from RDE"); 3200 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3201 log_warnx("no such peer: id=%u", 3202 imsg.hdr.peerid); 3203 break; 3204 } 3205 session_stop(p, ERR_CEASE_ADMIN_DOWN); 3206 break; 3207 default: 3208 break; 3209 } 3210 imsg_free(&imsg); 3211 } 3212 } 3213 3214 int 3215 la_cmp(struct listen_addr *a, struct listen_addr *b) 3216 { 3217 struct sockaddr_in *in_a, *in_b; 3218 struct sockaddr_in6 *in6_a, *in6_b; 3219 3220 if (a->sa.ss_family != b->sa.ss_family) 3221 return (1); 3222 3223 switch (a->sa.ss_family) { 3224 case AF_INET: 3225 in_a = (struct sockaddr_in *)&a->sa; 3226 in_b = (struct sockaddr_in *)&b->sa; 3227 if (in_a->sin_addr.s_addr != in_b->sin_addr.s_addr) 3228 return (1); 3229 if (in_a->sin_port != in_b->sin_port) 3230 return (1); 3231 break; 3232 case AF_INET6: 3233 in6_a = (struct sockaddr_in6 *)&a->sa; 3234 in6_b = (struct sockaddr_in6 *)&b->sa; 3235 if (bcmp(&in6_a->sin6_addr, &in6_b->sin6_addr, 3236 sizeof(struct in6_addr))) 3237 return (1); 3238 if (in6_a->sin6_port != in6_b->sin6_port) 3239 return (1); 3240 break; 3241 default: 3242 fatal("king bula sez: unknown address family"); 3243 /* NOTREACHED */ 3244 } 3245 3246 return (0); 3247 } 3248 3249 struct peer * 3250 getpeerbydesc(struct bgpd_config *c, const char *descr) 3251 { 3252 struct peer *p, *res = NULL; 3253 int match = 0; 3254 3255 RB_FOREACH(p, peer_head, &c->peers) 3256 if (!strcmp(p->conf.descr, descr)) { 3257 res = p; 3258 match++; 3259 } 3260 3261 if (match > 1) 3262 log_info("neighbor description \"%s\" not unique, request " 3263 "aborted", descr); 3264 3265 if (match == 1) 3266 return (res); 3267 else 3268 return (NULL); 3269 } 3270 3271 struct peer * 3272 getpeerbyip(struct bgpd_config *c, struct sockaddr *ip) 3273 { 3274 struct bgpd_addr addr; 3275 struct peer *p, *newpeer, *loose = NULL; 3276 uint32_t id; 3277 3278 sa2addr(ip, &addr, NULL); 3279 3280 /* we might want a more effective way to find peers by IP */ 3281 RB_FOREACH(p, peer_head, &c->peers) 3282 if (!p->conf.template && 3283 !memcmp(&addr, &p->conf.remote_addr, sizeof(addr))) 3284 return (p); 3285 3286 /* try template matching */ 3287 RB_FOREACH(p, peer_head, &c->peers) 3288 if (p->conf.template && 3289 p->conf.remote_addr.aid == addr.aid && 3290 session_match_mask(p, &addr)) 3291 if (loose == NULL || loose->conf.remote_masklen < 3292 p->conf.remote_masklen) 3293 loose = p; 3294 3295 if (loose != NULL) { 3296 /* clone */ 3297 if ((newpeer = malloc(sizeof(struct peer))) == NULL) 3298 fatal(NULL); 3299 memcpy(newpeer, loose, sizeof(struct peer)); 3300 for (id = PEER_ID_DYN_MAX; id > PEER_ID_STATIC_MAX; id--) { 3301 if (getpeerbyid(c, id) == NULL) /* we found a free id */ 3302 break; 3303 } 3304 newpeer->template = loose; 3305 session_template_clone(newpeer, ip, id, 0); 3306 newpeer->state = newpeer->prev_state = STATE_NONE; 3307 newpeer->reconf_action = RECONF_KEEP; 3308 newpeer->rbuf = NULL; 3309 newpeer->rpending = 0; 3310 init_peer(newpeer); 3311 bgp_fsm(newpeer, EVNT_START); 3312 if (RB_INSERT(peer_head, &c->peers, newpeer) != NULL) 3313 fatalx("%s: peer tree is corrupt", __func__); 3314 return (newpeer); 3315 } 3316 3317 return (NULL); 3318 } 3319 3320 struct peer * 3321 getpeerbyid(struct bgpd_config *c, uint32_t peerid) 3322 { 3323 static struct peer lookup; 3324 3325 lookup.conf.id = peerid; 3326 3327 return RB_FIND(peer_head, &c->peers, &lookup); 3328 } 3329 3330 int 3331 peer_matched(struct peer *p, struct ctl_neighbor *n) 3332 { 3333 char *s; 3334 3335 if (n && n->addr.aid) { 3336 if (memcmp(&p->conf.remote_addr, &n->addr, 3337 sizeof(p->conf.remote_addr))) 3338 return 0; 3339 } else if (n && n->descr[0]) { 3340 s = n->is_group ? p->conf.group : p->conf.descr; 3341 if (strcmp(s, n->descr)) 3342 return 0; 3343 } 3344 return 1; 3345 } 3346 3347 void 3348 session_template_clone(struct peer *p, struct sockaddr *ip, uint32_t id, 3349 uint32_t as) 3350 { 3351 struct bgpd_addr remote_addr; 3352 3353 if (ip) 3354 sa2addr(ip, &remote_addr, NULL); 3355 else 3356 memcpy(&remote_addr, &p->conf.remote_addr, sizeof(remote_addr)); 3357 3358 memcpy(&p->conf, &p->template->conf, sizeof(struct peer_config)); 3359 3360 p->conf.id = id; 3361 3362 if (as) { 3363 p->conf.remote_as = as; 3364 p->conf.ebgp = (p->conf.remote_as != p->conf.local_as); 3365 if (!p->conf.ebgp) 3366 /* force enforce_as off for iBGP sessions */ 3367 p->conf.enforce_as = ENFORCE_AS_OFF; 3368 } 3369 3370 memcpy(&p->conf.remote_addr, &remote_addr, sizeof(remote_addr)); 3371 switch (p->conf.remote_addr.aid) { 3372 case AID_INET: 3373 p->conf.remote_masklen = 32; 3374 break; 3375 case AID_INET6: 3376 p->conf.remote_masklen = 128; 3377 break; 3378 } 3379 p->conf.template = 0; 3380 } 3381 3382 int 3383 session_match_mask(struct peer *p, struct bgpd_addr *a) 3384 { 3385 struct in_addr v4masked; 3386 struct in6_addr v6masked; 3387 3388 switch (p->conf.remote_addr.aid) { 3389 case AID_INET: 3390 inet4applymask(&v4masked, &a->v4, p->conf.remote_masklen); 3391 if (p->conf.remote_addr.v4.s_addr == v4masked.s_addr) 3392 return (1); 3393 return (0); 3394 case AID_INET6: 3395 inet6applymask(&v6masked, &a->v6, p->conf.remote_masklen); 3396 3397 if (memcmp(&v6masked, &p->conf.remote_addr.v6, 3398 sizeof(v6masked)) == 0) 3399 return (1); 3400 return (0); 3401 } 3402 return (0); 3403 } 3404 3405 void 3406 session_down(struct peer *peer) 3407 { 3408 bzero(&peer->capa.neg, sizeof(peer->capa.neg)); 3409 peer->stats.last_updown = getmonotime(); 3410 /* 3411 * session_down is called in the exit code path so check 3412 * if the RDE is still around, if not there is no need to 3413 * send the message. 3414 */ 3415 if (ibuf_rde == NULL) 3416 return; 3417 if (imsg_rde(IMSG_SESSION_DOWN, peer->conf.id, NULL, 0) == -1) 3418 fatalx("imsg_compose error"); 3419 } 3420 3421 void 3422 session_up(struct peer *p) 3423 { 3424 struct session_up sup; 3425 3426 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3427 &p->conf, sizeof(p->conf)) == -1) 3428 fatalx("imsg_compose error"); 3429 3430 if (p->local.aid == AID_INET) { 3431 sup.local_v4_addr = p->local; 3432 sup.local_v6_addr = p->local_alt; 3433 } else { 3434 sup.local_v6_addr = p->local; 3435 sup.local_v4_addr = p->local_alt; 3436 } 3437 sup.remote_addr = p->remote; 3438 3439 sup.remote_bgpid = p->remote_bgpid; 3440 sup.short_as = p->short_as; 3441 memcpy(&sup.capa, &p->capa.neg, sizeof(sup.capa)); 3442 p->stats.last_updown = getmonotime(); 3443 if (imsg_rde(IMSG_SESSION_UP, p->conf.id, &sup, sizeof(sup)) == -1) 3444 fatalx("imsg_compose error"); 3445 } 3446 3447 int 3448 imsg_ctl_parent(int type, uint32_t peerid, pid_t pid, void *data, 3449 uint16_t datalen) 3450 { 3451 return (imsg_compose(ibuf_main, type, peerid, pid, -1, data, datalen)); 3452 } 3453 3454 int 3455 imsg_ctl_rde(int type, pid_t pid, void *data, uint16_t datalen) 3456 { 3457 if (ibuf_rde_ctl == NULL) { 3458 log_warnx("Can't send message %u to RDE, ctl pipe closed", 3459 type); 3460 return (0); 3461 } 3462 /* 3463 * Use control socket to talk to RDE to bypass the queue of the 3464 * regular imsg socket. 3465 */ 3466 return (imsg_compose(ibuf_rde_ctl, type, 0, pid, -1, data, datalen)); 3467 } 3468 3469 int 3470 imsg_rde(int type, uint32_t peerid, void *data, uint16_t datalen) 3471 { 3472 if (ibuf_rde == NULL) { 3473 log_warnx("Can't send message %u to RDE, pipe closed", type); 3474 return (0); 3475 } 3476 3477 return (imsg_compose(ibuf_rde, type, peerid, 0, -1, data, datalen)); 3478 } 3479 3480 void 3481 session_demote(struct peer *p, int level) 3482 { 3483 struct demote_msg msg; 3484 3485 strlcpy(msg.demote_group, p->conf.demote_group, 3486 sizeof(msg.demote_group)); 3487 msg.level = level; 3488 if (imsg_compose(ibuf_main, IMSG_DEMOTE, p->conf.id, 0, -1, 3489 &msg, sizeof(msg)) == -1) 3490 fatalx("imsg_compose error"); 3491 3492 p->demoted += level; 3493 } 3494 3495 void 3496 session_stop(struct peer *peer, uint8_t subcode) 3497 { 3498 char data[REASON_LEN]; 3499 size_t datalen; 3500 size_t reason_len; 3501 char *communication; 3502 3503 datalen = 0; 3504 communication = peer->conf.reason; 3505 3506 if ((subcode == ERR_CEASE_ADMIN_DOWN || 3507 subcode == ERR_CEASE_ADMIN_RESET) 3508 && communication && *communication) { 3509 reason_len = strlen(communication); 3510 if (reason_len > REASON_LEN - 1) { 3511 log_peer_warnx(&peer->conf, 3512 "trying to send overly long shutdown reason"); 3513 } else { 3514 data[0] = reason_len; 3515 datalen = reason_len + sizeof(data[0]); 3516 memcpy(data + 1, communication, reason_len); 3517 } 3518 } 3519 switch (peer->state) { 3520 case STATE_OPENSENT: 3521 case STATE_OPENCONFIRM: 3522 case STATE_ESTABLISHED: 3523 session_notification(peer, ERR_CEASE, subcode, data, datalen); 3524 break; 3525 default: 3526 /* session not open, no need to send notification */ 3527 break; 3528 } 3529 bgp_fsm(peer, EVNT_STOP); 3530 } 3531 3532 void 3533 merge_peers(struct bgpd_config *c, struct bgpd_config *nc) 3534 { 3535 struct peer *p, *np, *next; 3536 3537 RB_FOREACH(p, peer_head, &c->peers) { 3538 /* templates are handled specially */ 3539 if (p->template != NULL) 3540 continue; 3541 np = getpeerbyid(nc, p->conf.id); 3542 if (np == NULL) { 3543 p->reconf_action = RECONF_DELETE; 3544 continue; 3545 } 3546 3547 /* peer no longer uses TCP MD5SIG so deconfigure */ 3548 if (p->conf.auth.method == AUTH_MD5SIG && 3549 np->conf.auth.method != AUTH_MD5SIG) 3550 tcp_md5_del_listener(c, p); 3551 else if (np->conf.auth.method == AUTH_MD5SIG) 3552 tcp_md5_add_listener(c, np); 3553 3554 memcpy(&p->conf, &np->conf, sizeof(p->conf)); 3555 RB_REMOVE(peer_head, &nc->peers, np); 3556 free(np); 3557 3558 p->reconf_action = RECONF_KEEP; 3559 3560 /* had demotion, is demoted, demote removed? */ 3561 if (p->demoted && !p->conf.demote_group[0]) 3562 session_demote(p, -1); 3563 3564 /* if session is not open then refresh pfkey data */ 3565 if (p->state < STATE_OPENSENT && !p->template) 3566 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 3567 p->conf.id, 0, -1, NULL, 0); 3568 3569 /* sync the RDE in case we keep the peer */ 3570 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3571 &p->conf, sizeof(struct peer_config)) == -1) 3572 fatalx("imsg_compose error"); 3573 3574 /* apply the config to all clones of a template */ 3575 if (p->conf.template) { 3576 struct peer *xp; 3577 RB_FOREACH(xp, peer_head, &c->peers) { 3578 if (xp->template != p) 3579 continue; 3580 session_template_clone(xp, NULL, xp->conf.id, 3581 xp->conf.remote_as); 3582 if (imsg_rde(IMSG_SESSION_ADD, xp->conf.id, 3583 &xp->conf, sizeof(xp->conf)) == -1) 3584 fatalx("imsg_compose error"); 3585 } 3586 } 3587 } 3588 3589 if (imsg_rde(IMSG_RECONF_DRAIN, 0, NULL, 0) == -1) 3590 fatalx("imsg_compose error"); 3591 3592 /* pfkeys of new peers already loaded by the parent process */ 3593 RB_FOREACH_SAFE(np, peer_head, &nc->peers, next) { 3594 RB_REMOVE(peer_head, &nc->peers, np); 3595 if (RB_INSERT(peer_head, &c->peers, np) != NULL) 3596 fatalx("%s: peer tree is corrupt", __func__); 3597 if (np->conf.auth.method == AUTH_MD5SIG) 3598 tcp_md5_add_listener(c, np); 3599 } 3600 } 3601