1 /* $OpenBSD: session.c,v 1.420 2021/05/27 09:15:51 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 *, u_int8_t, u_int8_t); 70 int session_capa_add_mp(struct ibuf *, u_int8_t); 71 int session_capa_add_afi(struct peer *, struct ibuf *, u_int8_t, u_int8_t); 72 struct bgp_msg *session_newmsg(enum msg_type, u_int16_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(u_int32_t, void *, size_t); 77 void session_notification(struct peer *, u_int8_t, u_int8_t, void *, 78 ssize_t); 79 void session_rrefresh(struct peer *, u_int8_t, u_int8_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 *, u_int16_t *, u_int8_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 *, u_int16_t, u_int32_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, u_int32_t, void *, u_int16_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 u_int32_t, u_int32_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 fatal("poll error"); 458 459 /* 460 * If we previously saw fd exhaustion, we stop accept() 461 * for 1 second to throttle the accept() loop. 462 */ 463 if (pauseaccept && getmonotime() > pauseaccept + 1) 464 pauseaccept = 0; 465 466 if (handle_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main) == -1) { 467 log_warnx("SE: Lost connection to parent"); 468 session_quit = 1; 469 continue; 470 } else 471 session_dispatch_imsg(ibuf_main, PFD_PIPE_MAIN, 472 &listener_cnt); 473 474 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde) == -1) { 475 log_warnx("SE: Lost connection to RDE"); 476 msgbuf_clear(&ibuf_rde->w); 477 free(ibuf_rde); 478 ibuf_rde = NULL; 479 } else 480 session_dispatch_imsg(ibuf_rde, PFD_PIPE_ROUTE, 481 &listener_cnt); 482 483 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl) == 484 -1) { 485 log_warnx("SE: Lost connection to RDE control"); 486 msgbuf_clear(&ibuf_rde_ctl->w); 487 free(ibuf_rde_ctl); 488 ibuf_rde_ctl = NULL; 489 } else 490 session_dispatch_imsg(ibuf_rde_ctl, PFD_PIPE_ROUTE_CTL, 491 &listener_cnt); 492 493 if (pfd[PFD_SOCK_CTL].revents & POLLIN) 494 ctl_cnt += control_accept(csock, 0); 495 496 if (pfd[PFD_SOCK_RCTL].revents & POLLIN) 497 ctl_cnt += control_accept(rcsock, 1); 498 499 for (j = PFD_LISTENERS_START; j < idx_listeners; j++) 500 if (pfd[j].revents & POLLIN) 501 session_accept(pfd[j].fd); 502 503 for (; j < idx_peers; j++) 504 session_dispatch_msg(&pfd[j], 505 peer_l[j - idx_listeners]); 506 507 RB_FOREACH(p, peer_head, &conf->peers) 508 if (p->rbuf && p->rbuf->wpos) 509 session_process_msg(p); 510 511 for (; j < idx_mrts; j++) 512 if (pfd[j].revents & POLLOUT) 513 mrt_write(mrt_l[j - idx_peers]); 514 515 for (; j < i; j++) 516 ctl_cnt -= control_dispatch_msg(&pfd[j], &conf->peers); 517 } 518 519 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 520 RB_REMOVE(peer_head, &conf->peers, p); 521 strlcpy(p->conf.reason, 522 "bgpd shutting down", 523 sizeof(p->conf.reason)); 524 session_stop(p, ERR_CEASE_ADMIN_DOWN); 525 timer_remove_all(&p->timers); 526 free(p); 527 } 528 529 while ((m = LIST_FIRST(&mrthead)) != NULL) { 530 mrt_clean(m); 531 LIST_REMOVE(m, entry); 532 free(m); 533 } 534 535 free_config(conf); 536 free(peer_l); 537 free(mrt_l); 538 free(pfd); 539 540 /* close pipes */ 541 if (ibuf_rde) { 542 msgbuf_write(&ibuf_rde->w); 543 msgbuf_clear(&ibuf_rde->w); 544 close(ibuf_rde->fd); 545 free(ibuf_rde); 546 } 547 if (ibuf_rde_ctl) { 548 msgbuf_clear(&ibuf_rde_ctl->w); 549 close(ibuf_rde_ctl->fd); 550 free(ibuf_rde_ctl); 551 } 552 msgbuf_write(&ibuf_main->w); 553 msgbuf_clear(&ibuf_main->w); 554 close(ibuf_main->fd); 555 free(ibuf_main); 556 557 control_shutdown(csock); 558 control_shutdown(rcsock); 559 log_info("session engine exiting"); 560 exit(0); 561 } 562 563 void 564 init_peer(struct peer *p) 565 { 566 TAILQ_INIT(&p->timers); 567 p->fd = p->wbuf.fd = -1; 568 569 if (p->conf.if_depend[0]) 570 imsg_compose(ibuf_main, IMSG_IFINFO, 0, 0, -1, 571 p->conf.if_depend, sizeof(p->conf.if_depend)); 572 else 573 p->depend_ok = 1; 574 575 peer_cnt++; 576 577 change_state(p, STATE_IDLE, EVNT_NONE); 578 if (p->conf.down) 579 timer_stop(&p->timers, Timer_IdleHold); /* no autostart */ 580 else 581 timer_set(&p->timers, Timer_IdleHold, 0); /* start ASAP */ 582 583 /* 584 * on startup, demote if requested. 585 * do not handle new peers. they must reach ESTABLISHED beforehands. 586 * peers added at runtime have reconf_action set to RECONF_REINIT. 587 */ 588 if (p->reconf_action != RECONF_REINIT && p->conf.demote_group[0]) 589 session_demote(p, +1); 590 } 591 592 void 593 bgp_fsm(struct peer *peer, enum session_events event) 594 { 595 switch (peer->state) { 596 case STATE_NONE: 597 /* nothing */ 598 break; 599 case STATE_IDLE: 600 switch (event) { 601 case EVNT_START: 602 timer_stop(&peer->timers, Timer_Hold); 603 timer_stop(&peer->timers, Timer_SendHold); 604 timer_stop(&peer->timers, Timer_Keepalive); 605 timer_stop(&peer->timers, Timer_IdleHold); 606 607 /* allocate read buffer */ 608 peer->rbuf = calloc(1, sizeof(struct ibuf_read)); 609 if (peer->rbuf == NULL) 610 fatal(NULL); 611 612 /* init write buffer */ 613 msgbuf_init(&peer->wbuf); 614 615 peer->stats.last_sent_errcode = 0; 616 peer->stats.last_sent_suberr = 0; 617 peer->stats.last_rcvd_errcode = 0; 618 peer->stats.last_rcvd_suberr = 0; 619 620 if (!peer->depend_ok) 621 timer_stop(&peer->timers, Timer_ConnectRetry); 622 else if (peer->passive || peer->conf.passive || 623 peer->conf.template) { 624 change_state(peer, STATE_ACTIVE, event); 625 timer_stop(&peer->timers, Timer_ConnectRetry); 626 } else { 627 change_state(peer, STATE_CONNECT, event); 628 timer_set(&peer->timers, Timer_ConnectRetry, 629 conf->connectretry); 630 session_connect(peer); 631 } 632 peer->passive = 0; 633 break; 634 default: 635 /* ignore */ 636 break; 637 } 638 break; 639 case STATE_CONNECT: 640 switch (event) { 641 case EVNT_START: 642 /* ignore */ 643 break; 644 case EVNT_CON_OPEN: 645 session_tcp_established(peer); 646 session_open(peer); 647 timer_stop(&peer->timers, Timer_ConnectRetry); 648 peer->holdtime = INTERVAL_HOLD_INITIAL; 649 start_timer_holdtime(peer); 650 change_state(peer, STATE_OPENSENT, event); 651 break; 652 case EVNT_CON_OPENFAIL: 653 timer_set(&peer->timers, Timer_ConnectRetry, 654 conf->connectretry); 655 session_close_connection(peer); 656 change_state(peer, STATE_ACTIVE, event); 657 break; 658 case EVNT_TIMER_CONNRETRY: 659 timer_set(&peer->timers, Timer_ConnectRetry, 660 conf->connectretry); 661 session_connect(peer); 662 break; 663 default: 664 change_state(peer, STATE_IDLE, event); 665 break; 666 } 667 break; 668 case STATE_ACTIVE: 669 switch (event) { 670 case EVNT_START: 671 /* ignore */ 672 break; 673 case EVNT_CON_OPEN: 674 session_tcp_established(peer); 675 session_open(peer); 676 timer_stop(&peer->timers, Timer_ConnectRetry); 677 peer->holdtime = INTERVAL_HOLD_INITIAL; 678 start_timer_holdtime(peer); 679 change_state(peer, STATE_OPENSENT, event); 680 break; 681 case EVNT_CON_OPENFAIL: 682 timer_set(&peer->timers, Timer_ConnectRetry, 683 conf->connectretry); 684 session_close_connection(peer); 685 change_state(peer, STATE_ACTIVE, event); 686 break; 687 case EVNT_TIMER_CONNRETRY: 688 timer_set(&peer->timers, Timer_ConnectRetry, 689 peer->holdtime); 690 change_state(peer, STATE_CONNECT, event); 691 session_connect(peer); 692 break; 693 default: 694 change_state(peer, STATE_IDLE, event); 695 break; 696 } 697 break; 698 case STATE_OPENSENT: 699 switch (event) { 700 case EVNT_START: 701 /* ignore */ 702 break; 703 case EVNT_STOP: 704 change_state(peer, STATE_IDLE, event); 705 break; 706 case EVNT_CON_CLOSED: 707 session_close_connection(peer); 708 timer_set(&peer->timers, Timer_ConnectRetry, 709 conf->connectretry); 710 change_state(peer, STATE_ACTIVE, event); 711 break; 712 case EVNT_CON_FATAL: 713 change_state(peer, STATE_IDLE, event); 714 break; 715 case EVNT_TIMER_HOLDTIME: 716 case EVNT_TIMER_SENDHOLD: 717 session_notification(peer, ERR_HOLDTIMEREXPIRED, 718 0, NULL, 0); 719 change_state(peer, STATE_IDLE, event); 720 break; 721 case EVNT_RCVD_OPEN: 722 /* parse_open calls change_state itself on failure */ 723 if (parse_open(peer)) 724 break; 725 session_keepalive(peer); 726 change_state(peer, STATE_OPENCONFIRM, event); 727 break; 728 case EVNT_RCVD_NOTIFICATION: 729 if (parse_notification(peer)) { 730 change_state(peer, STATE_IDLE, event); 731 /* don't punish, capa negotiation */ 732 timer_set(&peer->timers, Timer_IdleHold, 0); 733 peer->IdleHoldTime /= 2; 734 } else 735 change_state(peer, STATE_IDLE, event); 736 break; 737 default: 738 session_notification(peer, 739 ERR_FSM, ERR_FSM_UNEX_OPENSENT, NULL, 0); 740 change_state(peer, STATE_IDLE, event); 741 break; 742 } 743 break; 744 case STATE_OPENCONFIRM: 745 switch (event) { 746 case EVNT_START: 747 /* ignore */ 748 break; 749 case EVNT_STOP: 750 change_state(peer, STATE_IDLE, event); 751 break; 752 case EVNT_CON_CLOSED: 753 case EVNT_CON_FATAL: 754 change_state(peer, STATE_IDLE, event); 755 break; 756 case EVNT_TIMER_HOLDTIME: 757 case EVNT_TIMER_SENDHOLD: 758 session_notification(peer, ERR_HOLDTIMEREXPIRED, 759 0, NULL, 0); 760 change_state(peer, STATE_IDLE, event); 761 break; 762 case EVNT_TIMER_KEEPALIVE: 763 session_keepalive(peer); 764 break; 765 case EVNT_RCVD_KEEPALIVE: 766 start_timer_holdtime(peer); 767 change_state(peer, STATE_ESTABLISHED, event); 768 break; 769 case EVNT_RCVD_NOTIFICATION: 770 parse_notification(peer); 771 change_state(peer, STATE_IDLE, event); 772 break; 773 default: 774 session_notification(peer, 775 ERR_FSM, ERR_FSM_UNEX_OPENCONFIRM, NULL, 0); 776 change_state(peer, STATE_IDLE, event); 777 break; 778 } 779 break; 780 case STATE_ESTABLISHED: 781 switch (event) { 782 case EVNT_START: 783 /* ignore */ 784 break; 785 case EVNT_STOP: 786 change_state(peer, STATE_IDLE, event); 787 break; 788 case EVNT_CON_CLOSED: 789 case EVNT_CON_FATAL: 790 change_state(peer, STATE_IDLE, event); 791 break; 792 case EVNT_TIMER_HOLDTIME: 793 case EVNT_TIMER_SENDHOLD: 794 session_notification(peer, ERR_HOLDTIMEREXPIRED, 795 0, NULL, 0); 796 change_state(peer, STATE_IDLE, event); 797 break; 798 case EVNT_TIMER_KEEPALIVE: 799 session_keepalive(peer); 800 break; 801 case EVNT_RCVD_KEEPALIVE: 802 start_timer_holdtime(peer); 803 break; 804 case EVNT_RCVD_UPDATE: 805 start_timer_holdtime(peer); 806 if (parse_update(peer)) 807 change_state(peer, STATE_IDLE, event); 808 else 809 start_timer_holdtime(peer); 810 break; 811 case EVNT_RCVD_NOTIFICATION: 812 parse_notification(peer); 813 change_state(peer, STATE_IDLE, event); 814 break; 815 default: 816 session_notification(peer, 817 ERR_FSM, ERR_FSM_UNEX_ESTABLISHED, NULL, 0); 818 change_state(peer, STATE_IDLE, event); 819 break; 820 } 821 break; 822 } 823 } 824 825 void 826 start_timer_holdtime(struct peer *peer) 827 { 828 if (peer->holdtime > 0) 829 timer_set(&peer->timers, Timer_Hold, peer->holdtime); 830 else 831 timer_stop(&peer->timers, Timer_Hold); 832 } 833 834 void 835 start_timer_keepalive(struct peer *peer) 836 { 837 if (peer->holdtime > 0) 838 timer_set(&peer->timers, Timer_Keepalive, peer->holdtime / 3); 839 else 840 timer_stop(&peer->timers, Timer_Keepalive); 841 } 842 843 void 844 session_close_connection(struct peer *peer) 845 { 846 if (peer->fd != -1) { 847 close(peer->fd); 848 pauseaccept = 0; 849 } 850 peer->fd = peer->wbuf.fd = -1; 851 } 852 853 void 854 change_state(struct peer *peer, enum session_state state, 855 enum session_events event) 856 { 857 struct mrt *mrt; 858 859 switch (state) { 860 case STATE_IDLE: 861 /* carp demotion first. new peers handled in init_peer */ 862 if (peer->state == STATE_ESTABLISHED && 863 peer->conf.demote_group[0] && !peer->demoted) 864 session_demote(peer, +1); 865 866 /* 867 * try to write out what's buffered (maybe a notification), 868 * don't bother if it fails 869 */ 870 if (peer->state >= STATE_OPENSENT && peer->wbuf.queued) 871 msgbuf_write(&peer->wbuf); 872 873 /* 874 * we must start the timer for the next EVNT_START 875 * if we are coming here due to an error and the 876 * session was not established successfully before, the 877 * starttimerinterval needs to be exponentially increased 878 */ 879 if (peer->IdleHoldTime == 0) 880 peer->IdleHoldTime = INTERVAL_IDLE_HOLD_INITIAL; 881 peer->holdtime = INTERVAL_HOLD_INITIAL; 882 timer_stop(&peer->timers, Timer_ConnectRetry); 883 timer_stop(&peer->timers, Timer_Keepalive); 884 timer_stop(&peer->timers, Timer_Hold); 885 timer_stop(&peer->timers, Timer_SendHold); 886 timer_stop(&peer->timers, Timer_IdleHold); 887 timer_stop(&peer->timers, Timer_IdleHoldReset); 888 session_close_connection(peer); 889 msgbuf_clear(&peer->wbuf); 890 free(peer->rbuf); 891 peer->rbuf = NULL; 892 peer->rpending = 0; 893 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 894 if (!peer->template) 895 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 896 peer->conf.id, 0, -1, NULL, 0); 897 898 if (event != EVNT_STOP) { 899 timer_set(&peer->timers, Timer_IdleHold, 900 peer->IdleHoldTime); 901 if (event != EVNT_NONE && 902 peer->IdleHoldTime < MAX_IDLE_HOLD/2) 903 peer->IdleHoldTime *= 2; 904 } 905 if (peer->state == STATE_ESTABLISHED) { 906 if (peer->capa.neg.grestart.restart == 2 && 907 (event == EVNT_CON_CLOSED || 908 event == EVNT_CON_FATAL)) { 909 /* don't punish graceful restart */ 910 timer_set(&peer->timers, Timer_IdleHold, 0); 911 peer->IdleHoldTime /= 2; 912 session_graceful_restart(peer); 913 } else 914 session_down(peer); 915 } 916 if (peer->state == STATE_NONE || 917 peer->state == STATE_ESTABLISHED) { 918 /* initialize capability negotiation structures */ 919 memcpy(&peer->capa.ann, &peer->conf.capabilities, 920 sizeof(peer->capa.ann)); 921 if (!peer->conf.announce_capa) 922 session_capa_ann_none(peer); 923 } 924 break; 925 case STATE_CONNECT: 926 if (peer->state == STATE_ESTABLISHED && 927 peer->capa.neg.grestart.restart == 2) { 928 /* do the graceful restart dance */ 929 session_graceful_restart(peer); 930 peer->holdtime = INTERVAL_HOLD_INITIAL; 931 timer_stop(&peer->timers, Timer_ConnectRetry); 932 timer_stop(&peer->timers, Timer_Keepalive); 933 timer_stop(&peer->timers, Timer_Hold); 934 timer_stop(&peer->timers, Timer_SendHold); 935 timer_stop(&peer->timers, Timer_IdleHold); 936 timer_stop(&peer->timers, Timer_IdleHoldReset); 937 session_close_connection(peer); 938 msgbuf_clear(&peer->wbuf); 939 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 940 } 941 break; 942 case STATE_ACTIVE: 943 if (!peer->template) 944 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 945 peer->conf.id, 0, -1, NULL, 0); 946 break; 947 case STATE_OPENSENT: 948 break; 949 case STATE_OPENCONFIRM: 950 break; 951 case STATE_ESTABLISHED: 952 timer_set(&peer->timers, Timer_IdleHoldReset, 953 peer->IdleHoldTime); 954 if (peer->demoted) 955 timer_set(&peer->timers, Timer_CarpUndemote, 956 INTERVAL_HOLD_DEMOTED); 957 session_up(peer); 958 break; 959 default: /* something seriously fucked */ 960 break; 961 } 962 963 log_statechange(peer, state, event); 964 LIST_FOREACH(mrt, &mrthead, entry) { 965 if (!(mrt->type == MRT_ALL_IN || mrt->type == MRT_ALL_OUT)) 966 continue; 967 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 968 mrt->peer_id == peer->conf.id || (mrt->group_id != 0 && 969 mrt->group_id == peer->conf.groupid)) 970 mrt_dump_state(mrt, peer->state, state, peer); 971 } 972 peer->prev_state = peer->state; 973 peer->state = state; 974 } 975 976 void 977 session_accept(int listenfd) 978 { 979 int connfd; 980 socklen_t len; 981 struct sockaddr_storage cliaddr; 982 struct peer *p = NULL; 983 984 len = sizeof(cliaddr); 985 if ((connfd = accept4(listenfd, 986 (struct sockaddr *)&cliaddr, &len, 987 SOCK_CLOEXEC | SOCK_NONBLOCK)) == -1) { 988 if (errno == ENFILE || errno == EMFILE) 989 pauseaccept = getmonotime(); 990 else if (errno != EWOULDBLOCK && errno != EINTR && 991 errno != ECONNABORTED) 992 log_warn("accept"); 993 return; 994 } 995 996 p = getpeerbyip(conf, (struct sockaddr *)&cliaddr); 997 998 if (p != NULL && p->state == STATE_IDLE && p->errcnt < 2) { 999 if (timer_running(&p->timers, Timer_IdleHold, NULL)) { 1000 /* fast reconnect after clear */ 1001 p->passive = 1; 1002 bgp_fsm(p, EVNT_START); 1003 } 1004 } 1005 1006 if (p != NULL && 1007 (p->state == STATE_CONNECT || p->state == STATE_ACTIVE)) { 1008 if (p->fd != -1) { 1009 if (p->state == STATE_CONNECT) 1010 session_close_connection(p); 1011 else { 1012 close(connfd); 1013 return; 1014 } 1015 } 1016 1017 open: 1018 if (p->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1019 log_peer_warnx(&p->conf, 1020 "ipsec or md5sig configured but not available"); 1021 close(connfd); 1022 return; 1023 } 1024 1025 if (tcp_md5_check(connfd, p) == -1) { 1026 close(connfd); 1027 return; 1028 } 1029 p->fd = p->wbuf.fd = connfd; 1030 if (session_setup_socket(p)) { 1031 close(connfd); 1032 return; 1033 } 1034 bgp_fsm(p, EVNT_CON_OPEN); 1035 return; 1036 } else if (p != NULL && p->state == STATE_ESTABLISHED && 1037 p->capa.neg.grestart.restart == 2) { 1038 /* first do the graceful restart dance */ 1039 change_state(p, STATE_CONNECT, EVNT_CON_CLOSED); 1040 /* then do part of the open dance */ 1041 goto open; 1042 } else { 1043 log_conn_attempt(p, (struct sockaddr *)&cliaddr, len); 1044 close(connfd); 1045 } 1046 } 1047 1048 int 1049 session_connect(struct peer *peer) 1050 { 1051 struct sockaddr *sa; 1052 struct bgpd_addr *bind_addr = NULL; 1053 socklen_t sa_len; 1054 1055 /* 1056 * we do not need the overcomplicated collision detection RFC 1771 1057 * describes; we simply make sure there is only ever one concurrent 1058 * tcp connection per peer. 1059 */ 1060 if (peer->fd != -1) 1061 return (-1); 1062 1063 if ((peer->fd = socket(aid2af(peer->conf.remote_addr.aid), 1064 SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, IPPROTO_TCP)) == -1) { 1065 log_peer_warn(&peer->conf, "session_connect socket"); 1066 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1067 return (-1); 1068 } 1069 1070 if (peer->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1071 log_peer_warnx(&peer->conf, 1072 "ipsec or md5sig configured but not available"); 1073 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1074 return (-1); 1075 } 1076 1077 tcp_md5_set(peer->fd, peer); 1078 peer->wbuf.fd = peer->fd; 1079 1080 /* if local-address is set we need to bind() */ 1081 switch (peer->conf.remote_addr.aid) { 1082 case AID_INET: 1083 bind_addr = &peer->conf.local_addr_v4; 1084 break; 1085 case AID_INET6: 1086 bind_addr = &peer->conf.local_addr_v6; 1087 break; 1088 } 1089 if ((sa = addr2sa(bind_addr, 0, &sa_len)) != NULL) { 1090 if (bind(peer->fd, sa, sa_len) == -1) { 1091 log_peer_warn(&peer->conf, "session_connect bind"); 1092 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1093 return (-1); 1094 } 1095 } 1096 1097 if (session_setup_socket(peer)) { 1098 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1099 return (-1); 1100 } 1101 1102 sa = addr2sa(&peer->conf.remote_addr, BGP_PORT, &sa_len); 1103 if (connect(peer->fd, sa, sa_len) == -1) { 1104 if (errno != EINPROGRESS) { 1105 if (errno != peer->lasterr) 1106 log_peer_warn(&peer->conf, "connect"); 1107 peer->lasterr = errno; 1108 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1109 return (-1); 1110 } 1111 } else 1112 bgp_fsm(peer, EVNT_CON_OPEN); 1113 1114 return (0); 1115 } 1116 1117 int 1118 session_setup_socket(struct peer *p) 1119 { 1120 int ttl = p->conf.distance; 1121 int pre = IPTOS_PREC_INTERNETCONTROL; 1122 int nodelay = 1; 1123 int bsize; 1124 1125 switch (p->conf.remote_addr.aid) { 1126 case AID_INET: 1127 /* set precedence, see RFC 1771 appendix 5 */ 1128 if (setsockopt(p->fd, IPPROTO_IP, IP_TOS, &pre, sizeof(pre)) == 1129 -1) { 1130 log_peer_warn(&p->conf, 1131 "session_setup_socket setsockopt TOS"); 1132 return (-1); 1133 } 1134 1135 if (p->conf.ebgp) { 1136 /* 1137 * set TTL to foreign router's distance 1138 * 1=direct n=multihop with ttlsec, we always use 255 1139 */ 1140 if (p->conf.ttlsec) { 1141 ttl = 256 - p->conf.distance; 1142 if (setsockopt(p->fd, IPPROTO_IP, IP_MINTTL, 1143 &ttl, sizeof(ttl)) == -1) { 1144 log_peer_warn(&p->conf, 1145 "session_setup_socket: " 1146 "setsockopt MINTTL"); 1147 return (-1); 1148 } 1149 ttl = 255; 1150 } 1151 1152 if (setsockopt(p->fd, IPPROTO_IP, IP_TTL, &ttl, 1153 sizeof(ttl)) == -1) { 1154 log_peer_warn(&p->conf, 1155 "session_setup_socket setsockopt TTL"); 1156 return (-1); 1157 } 1158 } 1159 break; 1160 case AID_INET6: 1161 if (p->conf.ebgp) { 1162 /* 1163 * set hoplimit to foreign router's distance 1164 * 1=direct n=multihop with ttlsec, we always use 255 1165 */ 1166 if (p->conf.ttlsec) { 1167 ttl = 256 - p->conf.distance; 1168 if (setsockopt(p->fd, IPPROTO_IPV6, 1169 IPV6_MINHOPCOUNT, &ttl, sizeof(ttl)) 1170 == -1) { 1171 log_peer_warn(&p->conf, 1172 "session_setup_socket: " 1173 "setsockopt MINHOPCOUNT"); 1174 return (-1); 1175 } 1176 ttl = 255; 1177 } 1178 if (setsockopt(p->fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, 1179 &ttl, sizeof(ttl)) == -1) { 1180 log_peer_warn(&p->conf, 1181 "session_setup_socket setsockopt hoplimit"); 1182 return (-1); 1183 } 1184 } 1185 break; 1186 } 1187 1188 /* set TCP_NODELAY */ 1189 if (setsockopt(p->fd, IPPROTO_TCP, TCP_NODELAY, &nodelay, 1190 sizeof(nodelay)) == -1) { 1191 log_peer_warn(&p->conf, 1192 "session_setup_socket setsockopt TCP_NODELAY"); 1193 return (-1); 1194 } 1195 1196 /* only increase bufsize (and thus window) if md5 or ipsec is in use */ 1197 if (p->conf.auth.method != AUTH_NONE) { 1198 /* try to increase bufsize. no biggie if it fails */ 1199 bsize = 65535; 1200 while (bsize > 8192 && 1201 setsockopt(p->fd, SOL_SOCKET, SO_RCVBUF, &bsize, 1202 sizeof(bsize)) == -1 && errno != EINVAL) 1203 bsize /= 2; 1204 bsize = 65535; 1205 while (bsize > 8192 && 1206 setsockopt(p->fd, SOL_SOCKET, SO_SNDBUF, &bsize, 1207 sizeof(bsize)) == -1 && errno != EINVAL) 1208 bsize /= 2; 1209 } 1210 1211 return (0); 1212 } 1213 1214 /* compare two sockaddrs by converting them into bgpd_addr */ 1215 static int 1216 sa_cmp(struct sockaddr *a, struct sockaddr *b) 1217 { 1218 struct bgpd_addr ba, bb; 1219 1220 sa2addr(a, &ba, NULL); 1221 sa2addr(b, &bb, NULL); 1222 1223 return (memcmp(&ba, &bb, sizeof(ba)) == 0); 1224 } 1225 1226 static void 1227 get_alternate_addr(struct sockaddr *sa, struct bgpd_addr *alt) 1228 { 1229 struct ifaddrs *ifap, *ifa, *match; 1230 1231 if (getifaddrs(&ifap) == -1) 1232 fatal("getifaddrs"); 1233 1234 for (match = ifap; match != NULL; match = match->ifa_next) 1235 if (match->ifa_addr != NULL && 1236 sa_cmp(sa, match->ifa_addr) == 0) 1237 break; 1238 1239 if (match == NULL) { 1240 log_warnx("%s: local address not found", __func__); 1241 return; 1242 } 1243 1244 switch (sa->sa_family) { 1245 case AF_INET6: 1246 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1247 if (ifa->ifa_addr != NULL && 1248 ifa->ifa_addr->sa_family == AF_INET && 1249 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1250 sa2addr(ifa->ifa_addr, alt, NULL); 1251 break; 1252 } 1253 } 1254 break; 1255 case AF_INET: 1256 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1257 if (ifa->ifa_addr != NULL && 1258 ifa->ifa_addr->sa_family == AF_INET6 && 1259 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1260 struct sockaddr_in6 *s = 1261 (struct sockaddr_in6 *)ifa->ifa_addr; 1262 1263 /* only accept global scope addresses */ 1264 if (IN6_IS_ADDR_LINKLOCAL(&s->sin6_addr) || 1265 IN6_IS_ADDR_SITELOCAL(&s->sin6_addr)) 1266 continue; 1267 sa2addr(ifa->ifa_addr, alt, NULL); 1268 break; 1269 } 1270 } 1271 break; 1272 default: 1273 log_warnx("%s: unsupported address family %d", __func__, 1274 sa->sa_family); 1275 break; 1276 } 1277 1278 freeifaddrs(ifap); 1279 } 1280 1281 void 1282 session_tcp_established(struct peer *peer) 1283 { 1284 struct sockaddr_storage ss; 1285 socklen_t len; 1286 1287 len = sizeof(ss); 1288 if (getsockname(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1289 log_warn("getsockname"); 1290 sa2addr((struct sockaddr *)&ss, &peer->local, &peer->local_port); 1291 get_alternate_addr((struct sockaddr *)&ss, &peer->local_alt); 1292 len = sizeof(ss); 1293 if (getpeername(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1294 log_warn("getpeername"); 1295 sa2addr((struct sockaddr *)&ss, &peer->remote, &peer->remote_port); 1296 } 1297 1298 void 1299 session_capa_ann_none(struct peer *peer) 1300 { 1301 bzero(&peer->capa.ann, sizeof(peer->capa.ann)); 1302 } 1303 1304 int 1305 session_capa_add(struct ibuf *opb, u_int8_t capa_code, u_int8_t capa_len) 1306 { 1307 int errs = 0; 1308 1309 errs += ibuf_add(opb, &capa_code, sizeof(capa_code)); 1310 errs += ibuf_add(opb, &capa_len, sizeof(capa_len)); 1311 return (errs); 1312 } 1313 1314 int 1315 session_capa_add_mp(struct ibuf *buf, u_int8_t aid) 1316 { 1317 u_int8_t safi, pad = 0; 1318 u_int16_t afi; 1319 int errs = 0; 1320 1321 if (aid2afi(aid, &afi, &safi) == -1) 1322 fatalx("session_capa_add_mp: bad afi/safi pair"); 1323 afi = htons(afi); 1324 errs += ibuf_add(buf, &afi, sizeof(afi)); 1325 errs += ibuf_add(buf, &pad, sizeof(pad)); 1326 errs += ibuf_add(buf, &safi, sizeof(safi)); 1327 1328 return (errs); 1329 } 1330 1331 int 1332 session_capa_add_afi(struct peer *p, struct ibuf *b, u_int8_t aid, 1333 u_int8_t flags) 1334 { 1335 u_int errs = 0; 1336 u_int16_t afi; 1337 u_int8_t safi; 1338 1339 if (aid2afi(aid, &afi, &safi)) { 1340 log_warn("session_capa_add_afi: bad AID"); 1341 return (1); 1342 } 1343 1344 afi = htons(afi); 1345 errs += ibuf_add(b, &afi, sizeof(afi)); 1346 errs += ibuf_add(b, &safi, sizeof(safi)); 1347 errs += ibuf_add(b, &flags, sizeof(flags)); 1348 1349 return (errs); 1350 } 1351 1352 struct bgp_msg * 1353 session_newmsg(enum msg_type msgtype, u_int16_t len) 1354 { 1355 struct bgp_msg *msg; 1356 struct msg_header hdr; 1357 struct ibuf *buf; 1358 int errs = 0; 1359 1360 memset(&hdr.marker, 0xff, sizeof(hdr.marker)); 1361 hdr.len = htons(len); 1362 hdr.type = msgtype; 1363 1364 if ((buf = ibuf_open(len)) == NULL) 1365 return (NULL); 1366 1367 errs += ibuf_add(buf, &hdr.marker, sizeof(hdr.marker)); 1368 errs += ibuf_add(buf, &hdr.len, sizeof(hdr.len)); 1369 errs += ibuf_add(buf, &hdr.type, sizeof(hdr.type)); 1370 1371 if (errs || (msg = calloc(1, sizeof(*msg))) == NULL) { 1372 ibuf_free(buf); 1373 return (NULL); 1374 } 1375 1376 msg->buf = buf; 1377 msg->type = msgtype; 1378 msg->len = len; 1379 1380 return (msg); 1381 } 1382 1383 int 1384 session_sendmsg(struct bgp_msg *msg, struct peer *p) 1385 { 1386 struct mrt *mrt; 1387 1388 LIST_FOREACH(mrt, &mrthead, entry) { 1389 if (!(mrt->type == MRT_ALL_OUT || (msg->type == UPDATE && 1390 mrt->type == MRT_UPDATE_OUT))) 1391 continue; 1392 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1393 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1394 mrt->group_id == p->conf.groupid)) 1395 mrt_dump_bgp_msg(mrt, msg->buf->buf, msg->len, p); 1396 } 1397 1398 ibuf_close(&p->wbuf, msg->buf); 1399 if (!p->throttled && p->wbuf.queued > SESS_MSG_HIGH_MARK) { 1400 if (imsg_rde(IMSG_XOFF, p->conf.id, NULL, 0) == -1) 1401 log_peer_warn(&p->conf, "imsg_compose XOFF"); 1402 else 1403 p->throttled = 1; 1404 } 1405 1406 free(msg); 1407 return (0); 1408 } 1409 1410 void 1411 session_open(struct peer *p) 1412 { 1413 struct bgp_msg *buf; 1414 struct ibuf *opb; 1415 struct msg_open msg; 1416 u_int16_t len; 1417 u_int8_t i, op_type, optparamlen = 0; 1418 int errs = 0; 1419 int mpcapa = 0; 1420 1421 1422 if ((opb = ibuf_dynamic(0, UCHAR_MAX - sizeof(op_type) - 1423 sizeof(optparamlen))) == NULL) { 1424 bgp_fsm(p, EVNT_CON_FATAL); 1425 return; 1426 } 1427 1428 /* multiprotocol extensions, RFC 4760 */ 1429 for (i = 0; i < AID_MAX; i++) 1430 if (p->capa.ann.mp[i]) { /* 4 bytes data */ 1431 errs += session_capa_add(opb, CAPA_MP, 4); 1432 errs += session_capa_add_mp(opb, i); 1433 mpcapa++; 1434 } 1435 1436 /* route refresh, RFC 2918 */ 1437 if (p->capa.ann.refresh) /* no data */ 1438 errs += session_capa_add(opb, CAPA_REFRESH, 0); 1439 1440 /* graceful restart and End-of-RIB marker, RFC 4724 */ 1441 if (p->capa.ann.grestart.restart) { 1442 int rst = 0; 1443 u_int16_t hdr = 0; 1444 1445 for (i = 0; i < AID_MAX; i++) { 1446 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) 1447 rst++; 1448 } 1449 1450 /* Only set the R-flag if no graceful restart is ongoing */ 1451 if (!rst) 1452 hdr |= CAPA_GR_R_FLAG; 1453 hdr = htons(hdr); 1454 1455 errs += session_capa_add(opb, CAPA_RESTART, sizeof(hdr)); 1456 errs += ibuf_add(opb, &hdr, sizeof(hdr)); 1457 } 1458 1459 /* 4-bytes AS numbers, RFC6793 */ 1460 if (p->capa.ann.as4byte) { /* 4 bytes data */ 1461 u_int32_t nas; 1462 1463 nas = htonl(p->conf.local_as); 1464 errs += session_capa_add(opb, CAPA_AS4BYTE, sizeof(nas)); 1465 errs += ibuf_add(opb, &nas, sizeof(nas)); 1466 } 1467 1468 /* advertisement of multiple paths, RFC7911 */ 1469 if (p->capa.ann.add_path[0]) { /* variable */ 1470 u_int8_t aplen; 1471 1472 if (mpcapa) 1473 aplen = 2 + 4 * mpcapa; 1474 else /* AID_INET */ 1475 aplen = 2 + 4; 1476 errs += session_capa_add(opb, CAPA_ADD_PATH, aplen); 1477 if (mpcapa) { 1478 for (i = AID_MIN; i < AID_MAX; i++) { 1479 if (p->capa.ann.mp[i]) { 1480 errs += session_capa_add_afi(p, opb, 1481 i, p->capa.ann.add_path[i]); 1482 } 1483 } 1484 } else { /* AID_INET */ 1485 errs += session_capa_add_afi(p, opb, AID_INET, 1486 p->capa.ann.add_path[AID_INET]); 1487 } 1488 } 1489 1490 /* enhanced route-refresh, RFC7313 */ 1491 if (p->capa.ann.enhanced_rr) /* no data */ 1492 errs += session_capa_add(opb, CAPA_ENHANCED_RR, 0); 1493 1494 if (ibuf_size(opb)) 1495 optparamlen = ibuf_size(opb) + sizeof(op_type) + 1496 sizeof(optparamlen); 1497 1498 len = MSGSIZE_OPEN_MIN + optparamlen; 1499 if (errs || (buf = session_newmsg(OPEN, len)) == NULL) { 1500 ibuf_free(opb); 1501 bgp_fsm(p, EVNT_CON_FATAL); 1502 return; 1503 } 1504 1505 msg.version = 4; 1506 msg.myas = htons(p->conf.local_short_as); 1507 if (p->conf.holdtime) 1508 msg.holdtime = htons(p->conf.holdtime); 1509 else 1510 msg.holdtime = htons(conf->holdtime); 1511 msg.bgpid = conf->bgpid; /* is already in network byte order */ 1512 msg.optparamlen = optparamlen; 1513 1514 errs += ibuf_add(buf->buf, &msg.version, sizeof(msg.version)); 1515 errs += ibuf_add(buf->buf, &msg.myas, sizeof(msg.myas)); 1516 errs += ibuf_add(buf->buf, &msg.holdtime, sizeof(msg.holdtime)); 1517 errs += ibuf_add(buf->buf, &msg.bgpid, sizeof(msg.bgpid)); 1518 errs += ibuf_add(buf->buf, &msg.optparamlen, sizeof(msg.optparamlen)); 1519 1520 if (optparamlen) { 1521 op_type = OPT_PARAM_CAPABILITIES; 1522 optparamlen = ibuf_size(opb); 1523 errs += ibuf_add(buf->buf, &op_type, sizeof(op_type)); 1524 errs += ibuf_add(buf->buf, &optparamlen, sizeof(optparamlen)); 1525 errs += ibuf_add(buf->buf, opb->buf, ibuf_size(opb)); 1526 } 1527 1528 ibuf_free(opb); 1529 1530 if (errs) { 1531 ibuf_free(buf->buf); 1532 free(buf); 1533 bgp_fsm(p, EVNT_CON_FATAL); 1534 return; 1535 } 1536 1537 if (session_sendmsg(buf, p) == -1) { 1538 bgp_fsm(p, EVNT_CON_FATAL); 1539 return; 1540 } 1541 1542 p->stats.msg_sent_open++; 1543 } 1544 1545 void 1546 session_keepalive(struct peer *p) 1547 { 1548 struct bgp_msg *buf; 1549 1550 if ((buf = session_newmsg(KEEPALIVE, MSGSIZE_KEEPALIVE)) == NULL || 1551 session_sendmsg(buf, p) == -1) { 1552 bgp_fsm(p, EVNT_CON_FATAL); 1553 return; 1554 } 1555 1556 start_timer_keepalive(p); 1557 p->stats.msg_sent_keepalive++; 1558 } 1559 1560 void 1561 session_update(u_int32_t peerid, void *data, size_t datalen) 1562 { 1563 struct peer *p; 1564 struct bgp_msg *buf; 1565 1566 if ((p = getpeerbyid(conf, peerid)) == NULL) { 1567 log_warnx("no such peer: id=%u", peerid); 1568 return; 1569 } 1570 1571 if (p->state != STATE_ESTABLISHED) 1572 return; 1573 1574 if ((buf = session_newmsg(UPDATE, MSGSIZE_HEADER + datalen)) == NULL) { 1575 bgp_fsm(p, EVNT_CON_FATAL); 1576 return; 1577 } 1578 1579 if (ibuf_add(buf->buf, data, datalen)) { 1580 ibuf_free(buf->buf); 1581 free(buf); 1582 bgp_fsm(p, EVNT_CON_FATAL); 1583 return; 1584 } 1585 1586 if (session_sendmsg(buf, p) == -1) { 1587 bgp_fsm(p, EVNT_CON_FATAL); 1588 return; 1589 } 1590 1591 start_timer_keepalive(p); 1592 p->stats.msg_sent_update++; 1593 } 1594 1595 void 1596 session_notification(struct peer *p, u_int8_t errcode, u_int8_t subcode, 1597 void *data, ssize_t datalen) 1598 { 1599 struct bgp_msg *buf; 1600 int errs = 0; 1601 1602 if (p->stats.last_sent_errcode) /* some notification already sent */ 1603 return; 1604 1605 log_notification(p, errcode, subcode, data, datalen, "sending"); 1606 1607 /* cap to maximum size */ 1608 if (datalen > MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN) { 1609 log_peer_warnx(&p->conf, 1610 "oversized notification, data trunkated"); 1611 datalen = MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN; 1612 } 1613 1614 if ((buf = session_newmsg(NOTIFICATION, 1615 MSGSIZE_NOTIFICATION_MIN + datalen)) == NULL) { 1616 bgp_fsm(p, EVNT_CON_FATAL); 1617 return; 1618 } 1619 1620 errs += ibuf_add(buf->buf, &errcode, sizeof(errcode)); 1621 errs += ibuf_add(buf->buf, &subcode, sizeof(subcode)); 1622 1623 if (datalen > 0) 1624 errs += ibuf_add(buf->buf, data, datalen); 1625 1626 if (errs) { 1627 ibuf_free(buf->buf); 1628 free(buf); 1629 bgp_fsm(p, EVNT_CON_FATAL); 1630 return; 1631 } 1632 1633 if (session_sendmsg(buf, p) == -1) { 1634 bgp_fsm(p, EVNT_CON_FATAL); 1635 return; 1636 } 1637 1638 p->stats.msg_sent_notification++; 1639 p->stats.last_sent_errcode = errcode; 1640 p->stats.last_sent_suberr = subcode; 1641 } 1642 1643 int 1644 session_neighbor_rrefresh(struct peer *p) 1645 { 1646 u_int8_t i; 1647 1648 if (!(p->capa.neg.refresh || p->capa.neg.enhanced_rr)) 1649 return (-1); 1650 1651 for (i = 0; i < AID_MAX; i++) { 1652 if (p->capa.neg.mp[i] != 0) 1653 session_rrefresh(p, i, ROUTE_REFRESH_REQUEST); 1654 } 1655 1656 return (0); 1657 } 1658 1659 void 1660 session_rrefresh(struct peer *p, u_int8_t aid, u_int8_t subtype) 1661 { 1662 struct bgp_msg *buf; 1663 int errs = 0; 1664 u_int16_t afi; 1665 u_int8_t safi; 1666 1667 switch (subtype) { 1668 case ROUTE_REFRESH_REQUEST: 1669 p->stats.refresh_sent_req++; 1670 break; 1671 case ROUTE_REFRESH_BEGIN_RR: 1672 case ROUTE_REFRESH_END_RR: 1673 /* requires enhanced route refresh */ 1674 if (!p->capa.neg.enhanced_rr) 1675 return; 1676 if (subtype == ROUTE_REFRESH_BEGIN_RR) 1677 p->stats.refresh_sent_borr++; 1678 else 1679 p->stats.refresh_sent_eorr++; 1680 break; 1681 default: 1682 fatalx("session_rrefresh: bad subtype %d", subtype); 1683 } 1684 1685 if (aid2afi(aid, &afi, &safi) == -1) 1686 fatalx("session_rrefresh: bad afi/safi pair"); 1687 1688 if ((buf = session_newmsg(RREFRESH, MSGSIZE_RREFRESH)) == NULL) { 1689 bgp_fsm(p, EVNT_CON_FATAL); 1690 return; 1691 } 1692 1693 afi = htons(afi); 1694 errs += ibuf_add(buf->buf, &afi, sizeof(afi)); 1695 errs += ibuf_add(buf->buf, &subtype, sizeof(subtype)); 1696 errs += ibuf_add(buf->buf, &safi, sizeof(safi)); 1697 1698 if (errs) { 1699 ibuf_free(buf->buf); 1700 free(buf); 1701 bgp_fsm(p, EVNT_CON_FATAL); 1702 return; 1703 } 1704 1705 if (session_sendmsg(buf, p) == -1) { 1706 bgp_fsm(p, EVNT_CON_FATAL); 1707 return; 1708 } 1709 1710 p->stats.msg_sent_rrefresh++; 1711 } 1712 1713 int 1714 session_graceful_restart(struct peer *p) 1715 { 1716 u_int8_t i; 1717 1718 timer_set(&p->timers, Timer_RestartTimeout, 1719 p->capa.neg.grestart.timeout); 1720 1721 for (i = 0; i < AID_MAX; i++) { 1722 if (p->capa.neg.grestart.flags[i] & CAPA_GR_PRESENT) { 1723 if (imsg_rde(IMSG_SESSION_STALE, p->conf.id, 1724 &i, sizeof(i)) == -1) 1725 return (-1); 1726 log_peer_warnx(&p->conf, 1727 "graceful restart of %s, keeping routes", 1728 aid2str(i)); 1729 p->capa.neg.grestart.flags[i] |= CAPA_GR_RESTARTING; 1730 } else if (p->capa.neg.mp[i]) { 1731 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1732 &i, sizeof(i)) == -1) 1733 return (-1); 1734 log_peer_warnx(&p->conf, 1735 "graceful restart of %s, flushing routes", 1736 aid2str(i)); 1737 } 1738 } 1739 return (0); 1740 } 1741 1742 int 1743 session_graceful_stop(struct peer *p) 1744 { 1745 u_int8_t i; 1746 1747 for (i = 0; i < AID_MAX; i++) { 1748 /* 1749 * Only flush if the peer is restarting and the timeout fired. 1750 * In all other cases the session was already flushed when the 1751 * session went down or when the new open message was parsed. 1752 */ 1753 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) { 1754 log_peer_warnx(&p->conf, "graceful restart of %s, " 1755 "time-out, flushing", aid2str(i)); 1756 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1757 &i, sizeof(i)) == -1) 1758 return (-1); 1759 } 1760 p->capa.neg.grestart.flags[i] &= ~CAPA_GR_RESTARTING; 1761 } 1762 return (0); 1763 } 1764 1765 int 1766 session_dispatch_msg(struct pollfd *pfd, struct peer *p) 1767 { 1768 ssize_t n; 1769 socklen_t len; 1770 int error; 1771 1772 if (p->state == STATE_CONNECT) { 1773 if (pfd->revents & POLLOUT) { 1774 if (pfd->revents & POLLIN) { 1775 /* error occurred */ 1776 len = sizeof(error); 1777 if (getsockopt(pfd->fd, SOL_SOCKET, SO_ERROR, 1778 &error, &len) == -1 || error) { 1779 if (error) 1780 errno = error; 1781 if (errno != p->lasterr) { 1782 log_peer_warn(&p->conf, 1783 "socket error"); 1784 p->lasterr = errno; 1785 } 1786 bgp_fsm(p, EVNT_CON_OPENFAIL); 1787 return (1); 1788 } 1789 } 1790 bgp_fsm(p, EVNT_CON_OPEN); 1791 return (1); 1792 } 1793 if (pfd->revents & POLLHUP) { 1794 bgp_fsm(p, EVNT_CON_OPENFAIL); 1795 return (1); 1796 } 1797 if (pfd->revents & (POLLERR|POLLNVAL)) { 1798 bgp_fsm(p, EVNT_CON_FATAL); 1799 return (1); 1800 } 1801 return (0); 1802 } 1803 1804 if (pfd->revents & POLLHUP) { 1805 bgp_fsm(p, EVNT_CON_CLOSED); 1806 return (1); 1807 } 1808 if (pfd->revents & (POLLERR|POLLNVAL)) { 1809 bgp_fsm(p, EVNT_CON_FATAL); 1810 return (1); 1811 } 1812 1813 if (pfd->revents & POLLOUT && p->wbuf.queued) { 1814 if ((error = msgbuf_write(&p->wbuf)) <= 0 && errno != EAGAIN) { 1815 if (error == 0) 1816 log_peer_warnx(&p->conf, "Connection closed"); 1817 else if (error == -1) 1818 log_peer_warn(&p->conf, "write error"); 1819 bgp_fsm(p, EVNT_CON_FATAL); 1820 return (1); 1821 } 1822 p->stats.last_write = getmonotime(); 1823 if (p->holdtime > 0) 1824 timer_set(&p->timers, Timer_SendHold, 1825 p->holdtime < INTERVAL_HOLD ? INTERVAL_HOLD : 1826 p->holdtime); 1827 if (p->throttled && p->wbuf.queued < SESS_MSG_LOW_MARK) { 1828 if (imsg_rde(IMSG_XON, p->conf.id, NULL, 0) == -1) 1829 log_peer_warn(&p->conf, "imsg_compose XON"); 1830 else 1831 p->throttled = 0; 1832 } 1833 if (!(pfd->revents & POLLIN)) 1834 return (1); 1835 } 1836 1837 if (p->rbuf && pfd->revents & POLLIN) { 1838 if ((n = read(p->fd, p->rbuf->buf + p->rbuf->wpos, 1839 sizeof(p->rbuf->buf) - p->rbuf->wpos)) == -1) { 1840 if (errno != EINTR && errno != EAGAIN) { 1841 log_peer_warn(&p->conf, "read error"); 1842 bgp_fsm(p, EVNT_CON_FATAL); 1843 } 1844 return (1); 1845 } 1846 if (n == 0) { /* connection closed */ 1847 bgp_fsm(p, EVNT_CON_CLOSED); 1848 return (1); 1849 } 1850 1851 p->rbuf->wpos += n; 1852 p->stats.last_read = getmonotime(); 1853 return (1); 1854 } 1855 return (0); 1856 } 1857 1858 void 1859 session_process_msg(struct peer *p) 1860 { 1861 struct mrt *mrt; 1862 ssize_t rpos, av, left; 1863 int processed = 0; 1864 u_int16_t msglen; 1865 u_int8_t msgtype; 1866 1867 rpos = 0; 1868 av = p->rbuf->wpos; 1869 p->rpending = 0; 1870 1871 /* 1872 * session might drop to IDLE -> buffers deallocated 1873 * we MUST check rbuf != NULL before use 1874 */ 1875 for (;;) { 1876 if (p->rbuf == NULL) 1877 return; 1878 if (rpos + MSGSIZE_HEADER > av) 1879 break; 1880 if (parse_header(p, p->rbuf->buf + rpos, &msglen, 1881 &msgtype) == -1) 1882 return; 1883 if (rpos + msglen > av) 1884 break; 1885 p->rbuf->rptr = p->rbuf->buf + rpos; 1886 1887 /* dump to MRT as soon as we have a full packet */ 1888 LIST_FOREACH(mrt, &mrthead, entry) { 1889 if (!(mrt->type == MRT_ALL_IN || (msgtype == UPDATE && 1890 mrt->type == MRT_UPDATE_IN))) 1891 continue; 1892 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1893 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1894 mrt->group_id == p->conf.groupid)) 1895 mrt_dump_bgp_msg(mrt, p->rbuf->rptr, msglen, p); 1896 } 1897 1898 switch (msgtype) { 1899 case OPEN: 1900 bgp_fsm(p, EVNT_RCVD_OPEN); 1901 p->stats.msg_rcvd_open++; 1902 break; 1903 case UPDATE: 1904 bgp_fsm(p, EVNT_RCVD_UPDATE); 1905 p->stats.msg_rcvd_update++; 1906 break; 1907 case NOTIFICATION: 1908 bgp_fsm(p, EVNT_RCVD_NOTIFICATION); 1909 p->stats.msg_rcvd_notification++; 1910 break; 1911 case KEEPALIVE: 1912 bgp_fsm(p, EVNT_RCVD_KEEPALIVE); 1913 p->stats.msg_rcvd_keepalive++; 1914 break; 1915 case RREFRESH: 1916 parse_rrefresh(p); 1917 p->stats.msg_rcvd_rrefresh++; 1918 break; 1919 default: /* cannot happen */ 1920 session_notification(p, ERR_HEADER, ERR_HDR_TYPE, 1921 &msgtype, 1); 1922 log_warnx("received message with unknown type %u", 1923 msgtype); 1924 bgp_fsm(p, EVNT_CON_FATAL); 1925 } 1926 rpos += msglen; 1927 if (++processed > MSG_PROCESS_LIMIT) { 1928 p->rpending = 1; 1929 break; 1930 } 1931 } 1932 1933 if (rpos < av) { 1934 left = av - rpos; 1935 memmove(&p->rbuf->buf, p->rbuf->buf + rpos, left); 1936 p->rbuf->wpos = left; 1937 } else 1938 p->rbuf->wpos = 0; 1939 } 1940 1941 int 1942 parse_header(struct peer *peer, u_char *data, u_int16_t *len, u_int8_t *type) 1943 { 1944 u_char *p; 1945 u_int16_t olen; 1946 static const u_int8_t marker[MSGSIZE_HEADER_MARKER] = { 0xff, 0xff, 1947 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 1948 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; 1949 1950 /* caller MUST make sure we are getting 19 bytes! */ 1951 p = data; 1952 if (memcmp(p, marker, sizeof(marker))) { 1953 log_peer_warnx(&peer->conf, "sync error"); 1954 session_notification(peer, ERR_HEADER, ERR_HDR_SYNC, NULL, 0); 1955 bgp_fsm(peer, EVNT_CON_FATAL); 1956 return (-1); 1957 } 1958 p += MSGSIZE_HEADER_MARKER; 1959 1960 memcpy(&olen, p, 2); 1961 *len = ntohs(olen); 1962 p += 2; 1963 memcpy(type, p, 1); 1964 1965 if (*len < MSGSIZE_HEADER || *len > MAX_PKTSIZE) { 1966 log_peer_warnx(&peer->conf, 1967 "received message: illegal length: %u byte", *len); 1968 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1969 &olen, sizeof(olen)); 1970 bgp_fsm(peer, EVNT_CON_FATAL); 1971 return (-1); 1972 } 1973 1974 switch (*type) { 1975 case OPEN: 1976 if (*len < MSGSIZE_OPEN_MIN) { 1977 log_peer_warnx(&peer->conf, 1978 "received OPEN: illegal len: %u byte", *len); 1979 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1980 &olen, sizeof(olen)); 1981 bgp_fsm(peer, EVNT_CON_FATAL); 1982 return (-1); 1983 } 1984 break; 1985 case NOTIFICATION: 1986 if (*len < MSGSIZE_NOTIFICATION_MIN) { 1987 log_peer_warnx(&peer->conf, 1988 "received NOTIFICATION: illegal len: %u byte", 1989 *len); 1990 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1991 &olen, sizeof(olen)); 1992 bgp_fsm(peer, EVNT_CON_FATAL); 1993 return (-1); 1994 } 1995 break; 1996 case UPDATE: 1997 if (*len < MSGSIZE_UPDATE_MIN) { 1998 log_peer_warnx(&peer->conf, 1999 "received UPDATE: illegal len: %u byte", *len); 2000 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2001 &olen, sizeof(olen)); 2002 bgp_fsm(peer, EVNT_CON_FATAL); 2003 return (-1); 2004 } 2005 break; 2006 case KEEPALIVE: 2007 if (*len != MSGSIZE_KEEPALIVE) { 2008 log_peer_warnx(&peer->conf, 2009 "received KEEPALIVE: illegal len: %u byte", *len); 2010 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2011 &olen, sizeof(olen)); 2012 bgp_fsm(peer, EVNT_CON_FATAL); 2013 return (-1); 2014 } 2015 break; 2016 case RREFRESH: 2017 if (*len < MSGSIZE_RREFRESH_MIN) { 2018 log_peer_warnx(&peer->conf, 2019 "received RREFRESH: illegal len: %u byte", *len); 2020 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2021 &olen, sizeof(olen)); 2022 bgp_fsm(peer, EVNT_CON_FATAL); 2023 return (-1); 2024 } 2025 break; 2026 default: 2027 log_peer_warnx(&peer->conf, 2028 "received msg with unknown type %u", *type); 2029 session_notification(peer, ERR_HEADER, ERR_HDR_TYPE, 2030 type, 1); 2031 bgp_fsm(peer, EVNT_CON_FATAL); 2032 return (-1); 2033 } 2034 return (0); 2035 } 2036 2037 int 2038 parse_open(struct peer *peer) 2039 { 2040 u_char *p, *op_val; 2041 u_int8_t version, rversion; 2042 u_int16_t short_as, msglen; 2043 u_int16_t holdtime, oholdtime, myholdtime; 2044 u_int32_t as, bgpid; 2045 u_int8_t optparamlen, plen; 2046 u_int8_t op_type, op_len; 2047 2048 p = peer->rbuf->rptr; 2049 p += MSGSIZE_HEADER_MARKER; 2050 memcpy(&msglen, p, sizeof(msglen)); 2051 msglen = ntohs(msglen); 2052 2053 p = peer->rbuf->rptr; 2054 p += MSGSIZE_HEADER; /* header is already checked */ 2055 2056 memcpy(&version, p, sizeof(version)); 2057 p += sizeof(version); 2058 2059 if (version != BGP_VERSION) { 2060 log_peer_warnx(&peer->conf, 2061 "peer wants unrecognized version %u", version); 2062 if (version > BGP_VERSION) 2063 rversion = version - BGP_VERSION; 2064 else 2065 rversion = BGP_VERSION; 2066 session_notification(peer, ERR_OPEN, ERR_OPEN_VERSION, 2067 &rversion, sizeof(rversion)); 2068 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2069 return (-1); 2070 } 2071 2072 memcpy(&short_as, p, sizeof(short_as)); 2073 p += sizeof(short_as); 2074 as = peer->short_as = ntohs(short_as); 2075 if (as == 0) { 2076 log_peer_warnx(&peer->conf, 2077 "peer requests unacceptable AS %u", as); 2078 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, 2079 NULL, 0); 2080 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2081 return (-1); 2082 } 2083 2084 memcpy(&oholdtime, p, sizeof(oholdtime)); 2085 p += sizeof(oholdtime); 2086 2087 holdtime = ntohs(oholdtime); 2088 if (holdtime && holdtime < peer->conf.min_holdtime) { 2089 log_peer_warnx(&peer->conf, 2090 "peer requests unacceptable holdtime %u", holdtime); 2091 session_notification(peer, ERR_OPEN, ERR_OPEN_HOLDTIME, 2092 NULL, 0); 2093 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2094 return (-1); 2095 } 2096 2097 myholdtime = peer->conf.holdtime; 2098 if (!myholdtime) 2099 myholdtime = conf->holdtime; 2100 if (holdtime < myholdtime) 2101 peer->holdtime = holdtime; 2102 else 2103 peer->holdtime = myholdtime; 2104 2105 memcpy(&bgpid, p, sizeof(bgpid)); 2106 p += sizeof(bgpid); 2107 2108 /* check bgpid for validity - just disallow 0 */ 2109 if (ntohl(bgpid) == 0) { 2110 log_peer_warnx(&peer->conf, "peer BGPID %u unacceptable", 2111 ntohl(bgpid)); 2112 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2113 NULL, 0); 2114 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2115 return (-1); 2116 } 2117 peer->remote_bgpid = bgpid; 2118 2119 memcpy(&optparamlen, p, sizeof(optparamlen)); 2120 p += sizeof(optparamlen); 2121 2122 if (optparamlen != msglen - MSGSIZE_OPEN_MIN) { 2123 log_peer_warnx(&peer->conf, 2124 "corrupt OPEN message received: length mismatch"); 2125 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2126 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2127 return (-1); 2128 } 2129 2130 plen = optparamlen; 2131 while (plen > 0) { 2132 if (plen < 2) { 2133 log_peer_warnx(&peer->conf, 2134 "corrupt OPEN message received, len wrong"); 2135 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2136 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2137 return (-1); 2138 } 2139 memcpy(&op_type, p, sizeof(op_type)); 2140 p += sizeof(op_type); 2141 plen -= sizeof(op_type); 2142 memcpy(&op_len, p, sizeof(op_len)); 2143 p += sizeof(op_len); 2144 plen -= sizeof(op_len); 2145 if (op_len > 0) { 2146 if (plen < op_len) { 2147 log_peer_warnx(&peer->conf, 2148 "corrupt OPEN message received, len wrong"); 2149 session_notification(peer, ERR_OPEN, 0, 2150 NULL, 0); 2151 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2152 return (-1); 2153 } 2154 op_val = p; 2155 p += op_len; 2156 plen -= op_len; 2157 } else 2158 op_val = NULL; 2159 2160 switch (op_type) { 2161 case OPT_PARAM_CAPABILITIES: /* RFC 3392 */ 2162 if (parse_capabilities(peer, op_val, op_len, 2163 &as) == -1) { 2164 session_notification(peer, ERR_OPEN, 0, 2165 NULL, 0); 2166 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2167 return (-1); 2168 } 2169 break; 2170 case OPT_PARAM_AUTH: /* deprecated */ 2171 default: 2172 /* 2173 * unsupported type 2174 * the RFCs tell us to leave the data section empty 2175 * and notify the peer with ERR_OPEN, ERR_OPEN_OPT. 2176 * How the peer should know _which_ optional parameter 2177 * we don't support is beyond me. 2178 */ 2179 log_peer_warnx(&peer->conf, 2180 "received OPEN message with unsupported optional " 2181 "parameter: type %u", op_type); 2182 session_notification(peer, ERR_OPEN, ERR_OPEN_OPT, 2183 NULL, 0); 2184 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2185 /* no punish */ 2186 timer_set(&peer->timers, Timer_IdleHold, 0); 2187 peer->IdleHoldTime /= 2; 2188 return (-1); 2189 } 2190 } 2191 2192 /* if remote-as is zero and it's a cloned neighbor, accept any */ 2193 if (peer->template && !peer->conf.remote_as && as != AS_TRANS) { 2194 peer->conf.remote_as = as; 2195 peer->conf.ebgp = (peer->conf.remote_as != peer->conf.local_as); 2196 if (!peer->conf.ebgp) 2197 /* force enforce_as off for iBGP sessions */ 2198 peer->conf.enforce_as = ENFORCE_AS_OFF; 2199 } 2200 2201 if (peer->conf.remote_as != as) { 2202 log_peer_warnx(&peer->conf, "peer sent wrong AS %s", 2203 log_as(as)); 2204 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, NULL, 0); 2205 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2206 return (-1); 2207 } 2208 2209 /* on iBGP sessions check for bgpid collision */ 2210 if (!peer->conf.ebgp && peer->remote_bgpid == conf->bgpid) { 2211 log_peer_warnx(&peer->conf, "peer BGPID %u conflicts with ours", 2212 ntohl(bgpid)); 2213 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2214 NULL, 0); 2215 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2216 return (-1); 2217 } 2218 2219 if (capa_neg_calc(peer) == -1) { 2220 log_peer_warnx(&peer->conf, 2221 "capability negotiation calculation failed"); 2222 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2223 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2224 return (-1); 2225 } 2226 2227 return (0); 2228 } 2229 2230 int 2231 parse_update(struct peer *peer) 2232 { 2233 u_char *p; 2234 u_int16_t datalen; 2235 2236 /* 2237 * we pass the message verbatim to the rde. 2238 * in case of errors the whole session is reset with a 2239 * notification anyway, we only need to know the peer 2240 */ 2241 p = peer->rbuf->rptr; 2242 p += MSGSIZE_HEADER_MARKER; 2243 memcpy(&datalen, p, sizeof(datalen)); 2244 datalen = ntohs(datalen); 2245 2246 p = peer->rbuf->rptr; 2247 p += MSGSIZE_HEADER; /* header is already checked */ 2248 datalen -= MSGSIZE_HEADER; 2249 2250 if (imsg_rde(IMSG_UPDATE, peer->conf.id, p, datalen) == -1) 2251 return (-1); 2252 2253 return (0); 2254 } 2255 2256 int 2257 parse_rrefresh(struct peer *peer) 2258 { 2259 u_int16_t afi, datalen; 2260 u_int8_t aid, safi, subtype; 2261 u_char *p; 2262 2263 p = peer->rbuf->rptr; 2264 p += MSGSIZE_HEADER_MARKER; 2265 memcpy(&datalen, p, sizeof(datalen)); 2266 datalen = ntohs(datalen); 2267 2268 p = peer->rbuf->rptr; 2269 p += MSGSIZE_HEADER; /* header is already checked */ 2270 2271 /* 2272 * We could check if we actually announced the capability but 2273 * as long as the message is correctly encoded we don't care. 2274 */ 2275 2276 /* afi, 2 byte */ 2277 memcpy(&afi, p, sizeof(afi)); 2278 afi = ntohs(afi); 2279 p += 2; 2280 /* subtype, 1 byte */ 2281 subtype = *p; 2282 p += 1; 2283 /* safi, 1 byte */ 2284 safi = *p; 2285 2286 /* check subtype if peer announced enhanced route refresh */ 2287 if (peer->capa.neg.enhanced_rr) { 2288 switch (subtype) { 2289 case ROUTE_REFRESH_REQUEST: 2290 /* no ORF support, so no oversized RREFRESH msgs */ 2291 if (datalen != MSGSIZE_RREFRESH) { 2292 log_peer_warnx(&peer->conf, 2293 "received RREFRESH: illegal len: %u byte", 2294 datalen); 2295 datalen = htons(datalen); 2296 session_notification(peer, ERR_HEADER, 2297 ERR_HDR_LEN, &datalen, sizeof(datalen)); 2298 bgp_fsm(peer, EVNT_CON_FATAL); 2299 return (-1); 2300 } 2301 peer->stats.refresh_rcvd_req++; 2302 break; 2303 case ROUTE_REFRESH_BEGIN_RR: 2304 case ROUTE_REFRESH_END_RR: 2305 /* special handling for RFC7313 */ 2306 if (datalen != MSGSIZE_RREFRESH) { 2307 log_peer_warnx(&peer->conf, 2308 "received RREFRESH: illegal len: %u byte", 2309 datalen); 2310 p = peer->rbuf->rptr; 2311 p += MSGSIZE_HEADER; 2312 datalen -= MSGSIZE_HEADER; 2313 session_notification(peer, ERR_RREFRESH, 2314 ERR_RR_INV_LEN, p, datalen); 2315 bgp_fsm(peer, EVNT_CON_FATAL); 2316 return (-1); 2317 } 2318 if (subtype == ROUTE_REFRESH_BEGIN_RR) 2319 peer->stats.refresh_rcvd_borr++; 2320 else 2321 peer->stats.refresh_rcvd_eorr++; 2322 break; 2323 default: 2324 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2325 "bad subtype %d", subtype); 2326 return (0); 2327 } 2328 } else { 2329 /* force subtype to default */ 2330 subtype = ROUTE_REFRESH_REQUEST; 2331 peer->stats.refresh_rcvd_req++; 2332 } 2333 2334 /* afi/safi unchecked - unrecognized values will be ignored anyway */ 2335 if (afi2aid(afi, safi, &aid) == -1) { 2336 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2337 "invalid afi/safi pair"); 2338 return (0); 2339 } 2340 2341 if (!peer->capa.neg.refresh && !peer->capa.neg.enhanced_rr) { 2342 log_peer_warnx(&peer->conf, "peer sent unexpected refresh"); 2343 return (0); 2344 } 2345 2346 if (imsg_rde(IMSG_REFRESH, peer->conf.id, &aid, sizeof(aid)) == -1) 2347 return (-1); 2348 2349 return (0); 2350 } 2351 2352 int 2353 parse_notification(struct peer *peer) 2354 { 2355 u_char *p; 2356 u_int16_t datalen; 2357 u_int8_t errcode; 2358 u_int8_t subcode; 2359 u_int8_t capa_code; 2360 u_int8_t capa_len; 2361 size_t reason_len; 2362 u_int8_t i; 2363 2364 /* just log */ 2365 p = peer->rbuf->rptr; 2366 p += MSGSIZE_HEADER_MARKER; 2367 memcpy(&datalen, p, sizeof(datalen)); 2368 datalen = ntohs(datalen); 2369 2370 p = peer->rbuf->rptr; 2371 p += MSGSIZE_HEADER; /* header is already checked */ 2372 datalen -= MSGSIZE_HEADER; 2373 2374 memcpy(&errcode, p, sizeof(errcode)); 2375 p += sizeof(errcode); 2376 datalen -= sizeof(errcode); 2377 2378 memcpy(&subcode, p, sizeof(subcode)); 2379 p += sizeof(subcode); 2380 datalen -= sizeof(subcode); 2381 2382 log_notification(peer, errcode, subcode, p, datalen, "received"); 2383 peer->errcnt++; 2384 peer->stats.last_rcvd_errcode = errcode; 2385 peer->stats.last_rcvd_suberr = subcode; 2386 2387 if (errcode == ERR_OPEN && subcode == ERR_OPEN_CAPA) { 2388 if (datalen == 0) { /* zebra likes to send those.. humbug */ 2389 log_peer_warnx(&peer->conf, "received \"unsupported " 2390 "capability\" notification without data part, " 2391 "disabling capability announcements altogether"); 2392 session_capa_ann_none(peer); 2393 } 2394 2395 while (datalen > 0) { 2396 if (datalen < 2) { 2397 log_peer_warnx(&peer->conf, 2398 "parse_notification: " 2399 "expect len >= 2, len is %u", datalen); 2400 return (-1); 2401 } 2402 memcpy(&capa_code, p, sizeof(capa_code)); 2403 p += sizeof(capa_code); 2404 datalen -= sizeof(capa_code); 2405 memcpy(&capa_len, p, sizeof(capa_len)); 2406 p += sizeof(capa_len); 2407 datalen -= sizeof(capa_len); 2408 if (datalen < capa_len) { 2409 log_peer_warnx(&peer->conf, 2410 "parse_notification: capa_len %u exceeds " 2411 "remaining msg length %u", capa_len, 2412 datalen); 2413 return (-1); 2414 } 2415 p += capa_len; 2416 datalen -= capa_len; 2417 switch (capa_code) { 2418 case CAPA_MP: 2419 for (i = 0; i < AID_MAX; i++) 2420 peer->capa.ann.mp[i] = 0; 2421 log_peer_warnx(&peer->conf, 2422 "disabling multiprotocol capability"); 2423 break; 2424 case CAPA_REFRESH: 2425 peer->capa.ann.refresh = 0; 2426 log_peer_warnx(&peer->conf, 2427 "disabling route refresh capability"); 2428 break; 2429 case CAPA_RESTART: 2430 peer->capa.ann.grestart.restart = 0; 2431 log_peer_warnx(&peer->conf, 2432 "disabling restart capability"); 2433 break; 2434 case CAPA_AS4BYTE: 2435 peer->capa.ann.as4byte = 0; 2436 log_peer_warnx(&peer->conf, 2437 "disabling 4-byte AS num capability"); 2438 break; 2439 case CAPA_ADD_PATH: 2440 memset(peer->capa.ann.add_path, 0, 2441 sizeof(peer->capa.ann.add_path)); 2442 log_peer_warnx(&peer->conf, 2443 "disabling ADD-PATH capability"); 2444 break; 2445 case CAPA_ENHANCED_RR: 2446 peer->capa.ann.enhanced_rr = 0; 2447 log_peer_warnx(&peer->conf, 2448 "disabling enhanced route refresh " 2449 "capability"); 2450 break; 2451 default: /* should not happen... */ 2452 log_peer_warnx(&peer->conf, "received " 2453 "\"unsupported capability\" notification " 2454 "for unknown capability %u, disabling " 2455 "capability announcements altogether", 2456 capa_code); 2457 session_capa_ann_none(peer); 2458 break; 2459 } 2460 } 2461 2462 return (1); 2463 } 2464 2465 if (errcode == ERR_OPEN && subcode == ERR_OPEN_OPT) { 2466 session_capa_ann_none(peer); 2467 return (1); 2468 } 2469 2470 if (errcode == ERR_CEASE && 2471 (subcode == ERR_CEASE_ADMIN_DOWN || 2472 subcode == ERR_CEASE_ADMIN_RESET)) { 2473 if (datalen > 1) { 2474 reason_len = *p++; 2475 datalen--; 2476 if (datalen < reason_len) { 2477 log_peer_warnx(&peer->conf, 2478 "received truncated shutdown reason"); 2479 return (0); 2480 } 2481 if (reason_len > REASON_LEN - 1) { 2482 log_peer_warnx(&peer->conf, 2483 "received overly long shutdown reason"); 2484 return (0); 2485 } 2486 memcpy(peer->stats.last_reason, p, reason_len); 2487 peer->stats.last_reason[reason_len] = '\0'; 2488 log_peer_warnx(&peer->conf, 2489 "received shutdown reason: \"%s\"", 2490 log_reason(peer->stats.last_reason)); 2491 p += reason_len; 2492 datalen -= reason_len; 2493 } 2494 } 2495 2496 return (0); 2497 } 2498 2499 int 2500 parse_capabilities(struct peer *peer, u_char *d, u_int16_t dlen, u_int32_t *as) 2501 { 2502 u_char *capa_val; 2503 u_int32_t remote_as; 2504 u_int16_t len; 2505 u_int16_t afi; 2506 u_int16_t gr_header; 2507 u_int8_t safi; 2508 u_int8_t aid; 2509 u_int8_t flags; 2510 u_int8_t capa_code; 2511 u_int8_t capa_len; 2512 u_int8_t i; 2513 2514 len = dlen; 2515 while (len > 0) { 2516 if (len < 2) { 2517 log_peer_warnx(&peer->conf, "Bad capabilities attr " 2518 "length: %u, too short", len); 2519 return (-1); 2520 } 2521 memcpy(&capa_code, d, sizeof(capa_code)); 2522 d += sizeof(capa_code); 2523 len -= sizeof(capa_code); 2524 memcpy(&capa_len, d, sizeof(capa_len)); 2525 d += sizeof(capa_len); 2526 len -= sizeof(capa_len); 2527 if (capa_len > 0) { 2528 if (len < capa_len) { 2529 log_peer_warnx(&peer->conf, 2530 "Bad capabilities attr length: " 2531 "len %u smaller than capa_len %u", 2532 len, capa_len); 2533 return (-1); 2534 } 2535 capa_val = d; 2536 d += capa_len; 2537 len -= capa_len; 2538 } else 2539 capa_val = NULL; 2540 2541 switch (capa_code) { 2542 case CAPA_MP: /* RFC 4760 */ 2543 if (capa_len != 4) { 2544 log_peer_warnx(&peer->conf, 2545 "Bad multi protocol capability length: " 2546 "%u", capa_len); 2547 break; 2548 } 2549 memcpy(&afi, capa_val, sizeof(afi)); 2550 afi = ntohs(afi); 2551 memcpy(&safi, capa_val + 3, sizeof(safi)); 2552 if (afi2aid(afi, safi, &aid) == -1) { 2553 log_peer_warnx(&peer->conf, 2554 "Received multi protocol capability: " 2555 " unknown AFI %u, safi %u pair", 2556 afi, safi); 2557 break; 2558 } 2559 peer->capa.peer.mp[aid] = 1; 2560 break; 2561 case CAPA_REFRESH: 2562 peer->capa.peer.refresh = 1; 2563 break; 2564 case CAPA_RESTART: 2565 if (capa_len == 2) { 2566 /* peer only supports EoR marker */ 2567 peer->capa.peer.grestart.restart = 1; 2568 peer->capa.peer.grestart.timeout = 0; 2569 break; 2570 } else if (capa_len % 4 != 2) { 2571 log_peer_warnx(&peer->conf, 2572 "Bad graceful restart capability length: " 2573 "%u", capa_len); 2574 peer->capa.peer.grestart.restart = 0; 2575 peer->capa.peer.grestart.timeout = 0; 2576 break; 2577 } 2578 2579 memcpy(&gr_header, capa_val, sizeof(gr_header)); 2580 gr_header = ntohs(gr_header); 2581 peer->capa.peer.grestart.timeout = 2582 gr_header & CAPA_GR_TIMEMASK; 2583 if (peer->capa.peer.grestart.timeout == 0) { 2584 log_peer_warnx(&peer->conf, "Received " 2585 "graceful restart timeout is zero"); 2586 peer->capa.peer.grestart.restart = 0; 2587 break; 2588 } 2589 2590 for (i = 2; i <= capa_len - 4; i += 4) { 2591 memcpy(&afi, capa_val + i, sizeof(afi)); 2592 afi = ntohs(afi); 2593 safi = capa_val[i + 2]; 2594 flags = capa_val[i + 3]; 2595 if (afi2aid(afi, safi, &aid) == -1) { 2596 log_peer_warnx(&peer->conf, 2597 "Received graceful restart capa: " 2598 " unknown AFI %u, safi %u pair", 2599 afi, safi); 2600 continue; 2601 } 2602 peer->capa.peer.grestart.flags[aid] |= 2603 CAPA_GR_PRESENT; 2604 if (flags & CAPA_GR_F_FLAG) 2605 peer->capa.peer.grestart.flags[aid] |= 2606 CAPA_GR_FORWARD; 2607 if (gr_header & CAPA_GR_R_FLAG) 2608 peer->capa.peer.grestart.flags[aid] |= 2609 CAPA_GR_RESTART; 2610 peer->capa.peer.grestart.restart = 2; 2611 } 2612 break; 2613 case CAPA_AS4BYTE: 2614 if (capa_len != 4) { 2615 log_peer_warnx(&peer->conf, 2616 "Bad AS4BYTE capability length: " 2617 "%u", capa_len); 2618 peer->capa.peer.as4byte = 0; 2619 break; 2620 } 2621 memcpy(&remote_as, capa_val, sizeof(remote_as)); 2622 *as = ntohl(remote_as); 2623 if (*as == 0) { 2624 log_peer_warnx(&peer->conf, 2625 "peer requests unacceptable AS %u", *as); 2626 session_notification(peer, ERR_OPEN, 2627 ERR_OPEN_AS, NULL, 0); 2628 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2629 return (-1); 2630 } 2631 peer->capa.peer.as4byte = 1; 2632 break; 2633 case CAPA_ADD_PATH: 2634 if (capa_len % 4 != 0) { 2635 log_peer_warnx(&peer->conf, 2636 "Bad ADD-PATH capability length: " 2637 "%u", capa_len); 2638 memset(peer->capa.peer.add_path, 0, 2639 sizeof(peer->capa.peer.add_path)); 2640 break; 2641 } 2642 for (i = 0; i <= capa_len - 4; i += 4) { 2643 memcpy(&afi, capa_val + i, sizeof(afi)); 2644 afi = ntohs(afi); 2645 safi = capa_val[i + 2]; 2646 flags = capa_val[i + 3]; 2647 if (afi2aid(afi, safi, &aid) == -1) { 2648 log_peer_warnx(&peer->conf, 2649 "Received ADD-PATH capa: " 2650 " unknown AFI %u, safi %u pair", 2651 afi, safi); 2652 memset(peer->capa.peer.add_path, 0, 2653 sizeof(peer->capa.peer.add_path)); 2654 break; 2655 } 2656 if (flags & ~CAPA_AP_BIDIR) { 2657 log_peer_warnx(&peer->conf, 2658 "Received ADD-PATH capa: " 2659 " bad flags %x", flags); 2660 memset(peer->capa.peer.add_path, 0, 2661 sizeof(peer->capa.peer.add_path)); 2662 break; 2663 } 2664 peer->capa.peer.add_path[aid] = flags; 2665 } 2666 break; 2667 case CAPA_ENHANCED_RR: 2668 peer->capa.peer.enhanced_rr = 1; 2669 break; 2670 default: 2671 break; 2672 } 2673 } 2674 2675 return (0); 2676 } 2677 2678 int 2679 capa_neg_calc(struct peer *p) 2680 { 2681 u_int8_t i, hasmp = 0; 2682 2683 /* a capability is accepted only if both sides announced it */ 2684 2685 p->capa.neg.refresh = 2686 (p->capa.ann.refresh && p->capa.peer.refresh) != 0; 2687 p->capa.neg.enhanced_rr = 2688 (p->capa.ann.enhanced_rr && p->capa.peer.enhanced_rr) != 0; 2689 2690 p->capa.neg.as4byte = 2691 (p->capa.ann.as4byte && p->capa.peer.as4byte) != 0; 2692 2693 /* MP: both side must agree on the AFI,SAFI pair */ 2694 for (i = 0; i < AID_MAX; i++) { 2695 if (p->capa.ann.mp[i] && p->capa.peer.mp[i]) 2696 p->capa.neg.mp[i] = 1; 2697 else 2698 p->capa.neg.mp[i] = 0; 2699 if (p->capa.ann.mp[i]) 2700 hasmp = 1; 2701 } 2702 /* if no MP capability present default to IPv4 unicast mode */ 2703 if (!hasmp) 2704 p->capa.neg.mp[AID_INET] = 1; 2705 2706 /* 2707 * graceful restart: the peer capabilities are of interest here. 2708 * It is necessary to compare the new values with the previous ones 2709 * and act acordingly. AFI/SAFI that are not part in the MP capability 2710 * are treated as not being present. 2711 * Also make sure that a flush happens if the session stopped 2712 * supporting graceful restart. 2713 */ 2714 2715 for (i = 0; i < AID_MAX; i++) { 2716 int8_t negflags; 2717 2718 /* disable GR if the AFI/SAFI is not present */ 2719 if ((p->capa.peer.grestart.flags[i] & CAPA_GR_PRESENT && 2720 p->capa.neg.mp[i] == 0)) 2721 p->capa.peer.grestart.flags[i] = 0; /* disable */ 2722 /* look at current GR state and decide what to do */ 2723 negflags = p->capa.neg.grestart.flags[i]; 2724 p->capa.neg.grestart.flags[i] = p->capa.peer.grestart.flags[i]; 2725 if (negflags & CAPA_GR_RESTARTING) { 2726 if (p->capa.ann.grestart.restart != 0 && 2727 p->capa.peer.grestart.flags[i] & CAPA_GR_FORWARD) { 2728 p->capa.neg.grestart.flags[i] |= 2729 CAPA_GR_RESTARTING; 2730 } else { 2731 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 2732 &i, sizeof(i)) == -1) 2733 return (-1); 2734 log_peer_warnx(&p->conf, "graceful restart of " 2735 "%s, not restarted, flushing", aid2str(i)); 2736 } 2737 } 2738 } 2739 p->capa.neg.grestart.timeout = p->capa.peer.grestart.timeout; 2740 p->capa.neg.grestart.restart = p->capa.peer.grestart.restart; 2741 if (p->capa.ann.grestart.restart == 0) 2742 p->capa.neg.grestart.restart = 0; 2743 2744 2745 /* 2746 * ADD-PATH: set only those bits where both sides agree. 2747 * For this compare our send bit with the recv bit from the peer 2748 * and vice versa. 2749 * The flags are stored from this systems view point. 2750 */ 2751 memset(p->capa.neg.add_path, 0, sizeof(p->capa.neg.add_path)); 2752 if (p->capa.ann.add_path[0]) { 2753 for (i = AID_MIN; i < AID_MAX; i++) { 2754 if ((p->capa.ann.add_path[i] & CAPA_AP_RECV) && 2755 (p->capa.peer.add_path[i] & CAPA_AP_SEND)) 2756 p->capa.neg.add_path[i] |= CAPA_AP_RECV; 2757 if ((p->capa.ann.add_path[i] & CAPA_AP_SEND) && 2758 (p->capa.peer.add_path[i] & CAPA_AP_RECV)) 2759 p->capa.neg.add_path[i] |= CAPA_AP_SEND; 2760 } 2761 } 2762 2763 return (0); 2764 } 2765 2766 void 2767 session_dispatch_imsg(struct imsgbuf *ibuf, int idx, u_int *listener_cnt) 2768 { 2769 struct imsg imsg; 2770 struct mrt xmrt; 2771 struct mrt *mrt; 2772 struct imsgbuf *i; 2773 struct peer *p; 2774 struct listen_addr *la, *nla; 2775 struct kif *kif; 2776 u_char *data; 2777 int n, fd, depend_ok, restricted; 2778 u_int16_t t; 2779 u_int8_t aid, errcode, subcode; 2780 2781 while (ibuf) { 2782 if ((n = imsg_get(ibuf, &imsg)) == -1) 2783 fatal("session_dispatch_imsg: imsg_get error"); 2784 2785 if (n == 0) 2786 break; 2787 2788 switch (imsg.hdr.type) { 2789 case IMSG_SOCKET_CONN: 2790 case IMSG_SOCKET_CONN_CTL: 2791 if (idx != PFD_PIPE_MAIN) 2792 fatalx("reconf request not from parent"); 2793 if ((fd = imsg.fd) == -1) { 2794 log_warnx("expected to receive imsg fd to " 2795 "RDE but didn't receive any"); 2796 break; 2797 } 2798 if ((i = malloc(sizeof(struct imsgbuf))) == NULL) 2799 fatal(NULL); 2800 imsg_init(i, fd); 2801 if (imsg.hdr.type == IMSG_SOCKET_CONN) { 2802 if (ibuf_rde) { 2803 log_warnx("Unexpected imsg connection " 2804 "to RDE received"); 2805 msgbuf_clear(&ibuf_rde->w); 2806 free(ibuf_rde); 2807 } 2808 ibuf_rde = i; 2809 } else { 2810 if (ibuf_rde_ctl) { 2811 log_warnx("Unexpected imsg ctl " 2812 "connection to RDE received"); 2813 msgbuf_clear(&ibuf_rde_ctl->w); 2814 free(ibuf_rde_ctl); 2815 } 2816 ibuf_rde_ctl = i; 2817 } 2818 break; 2819 case IMSG_RECONF_CONF: 2820 if (idx != PFD_PIPE_MAIN) 2821 fatalx("reconf request not from parent"); 2822 nconf = new_config(); 2823 2824 copy_config(nconf, imsg.data); 2825 pending_reconf = 1; 2826 break; 2827 case IMSG_RECONF_PEER: 2828 if (idx != PFD_PIPE_MAIN) 2829 fatalx("reconf request not from parent"); 2830 if ((p = calloc(1, sizeof(struct peer))) == NULL) 2831 fatal("new_peer"); 2832 memcpy(&p->conf, imsg.data, sizeof(struct peer_config)); 2833 p->state = p->prev_state = STATE_NONE; 2834 p->reconf_action = RECONF_REINIT; 2835 if (RB_INSERT(peer_head, &nconf->peers, p) != NULL) 2836 fatalx("%s: peer tree is corrupt", __func__); 2837 break; 2838 case IMSG_RECONF_LISTENER: 2839 if (idx != PFD_PIPE_MAIN) 2840 fatalx("reconf request not from parent"); 2841 if (nconf == NULL) 2842 fatalx("IMSG_RECONF_LISTENER but no config"); 2843 nla = imsg.data; 2844 TAILQ_FOREACH(la, conf->listen_addrs, entry) 2845 if (!la_cmp(la, nla)) 2846 break; 2847 2848 if (la == NULL) { 2849 if (nla->reconf != RECONF_REINIT) 2850 fatalx("king bula sez: " 2851 "expected REINIT"); 2852 2853 if ((nla->fd = imsg.fd) == -1) 2854 log_warnx("expected to receive fd for " 2855 "%s but didn't receive any", 2856 log_sockaddr((struct sockaddr *) 2857 &nla->sa, nla->sa_len)); 2858 2859 la = calloc(1, sizeof(struct listen_addr)); 2860 if (la == NULL) 2861 fatal(NULL); 2862 memcpy(&la->sa, &nla->sa, sizeof(la->sa)); 2863 la->flags = nla->flags; 2864 la->fd = nla->fd; 2865 la->reconf = RECONF_REINIT; 2866 TAILQ_INSERT_TAIL(nconf->listen_addrs, la, 2867 entry); 2868 } else { 2869 if (nla->reconf != RECONF_KEEP) 2870 fatalx("king bula sez: expected KEEP"); 2871 la->reconf = RECONF_KEEP; 2872 } 2873 2874 break; 2875 case IMSG_RECONF_CTRL: 2876 if (idx != PFD_PIPE_MAIN) 2877 fatalx("reconf request not from parent"); 2878 if (imsg.hdr.len != IMSG_HEADER_SIZE + 2879 sizeof(restricted)) 2880 fatalx("IFINFO imsg with wrong len"); 2881 memcpy(&restricted, imsg.data, sizeof(restricted)); 2882 if (imsg.fd == -1) { 2883 log_warnx("expected to receive fd for control " 2884 "socket but didn't receive any"); 2885 break; 2886 } 2887 if (restricted) { 2888 control_shutdown(rcsock); 2889 rcsock = imsg.fd; 2890 } else { 2891 control_shutdown(csock); 2892 csock = imsg.fd; 2893 } 2894 break; 2895 case IMSG_RECONF_DRAIN: 2896 switch (idx) { 2897 case PFD_PIPE_ROUTE: 2898 if (nconf != NULL) 2899 fatalx("got unexpected %s from RDE", 2900 "IMSG_RECONF_DONE"); 2901 imsg_compose(ibuf_main, IMSG_RECONF_DONE, 0, 0, 2902 -1, NULL, 0); 2903 break; 2904 case PFD_PIPE_MAIN: 2905 if (nconf == NULL) 2906 fatalx("got unexpected %s from parent", 2907 "IMSG_RECONF_DONE"); 2908 imsg_compose(ibuf_main, IMSG_RECONF_DRAIN, 0, 0, 2909 -1, NULL, 0); 2910 break; 2911 default: 2912 fatalx("reconf request not from parent or RDE"); 2913 } 2914 break; 2915 case IMSG_RECONF_DONE: 2916 if (idx != PFD_PIPE_MAIN) 2917 fatalx("reconf request not from parent"); 2918 if (nconf == NULL) 2919 fatalx("got IMSG_RECONF_DONE but no config"); 2920 copy_config(conf, nconf); 2921 merge_peers(conf, nconf); 2922 2923 /* delete old listeners */ 2924 for (la = TAILQ_FIRST(conf->listen_addrs); la != NULL; 2925 la = nla) { 2926 nla = TAILQ_NEXT(la, entry); 2927 if (la->reconf == RECONF_NONE) { 2928 log_info("not listening on %s any more", 2929 log_sockaddr((struct sockaddr *) 2930 &la->sa, la->sa_len)); 2931 TAILQ_REMOVE(conf->listen_addrs, la, 2932 entry); 2933 close(la->fd); 2934 free(la); 2935 } 2936 } 2937 2938 /* add new listeners */ 2939 TAILQ_CONCAT(conf->listen_addrs, nconf->listen_addrs, 2940 entry); 2941 2942 setup_listeners(listener_cnt); 2943 free_config(nconf); 2944 nconf = NULL; 2945 pending_reconf = 0; 2946 log_info("SE reconfigured"); 2947 /* 2948 * IMSG_RECONF_DONE is sent when the RDE drained 2949 * the peer config sent in merge_peers(). 2950 */ 2951 break; 2952 case IMSG_IFINFO: 2953 if (idx != PFD_PIPE_MAIN) 2954 fatalx("IFINFO message not from parent"); 2955 if (imsg.hdr.len != IMSG_HEADER_SIZE + 2956 sizeof(struct kif)) 2957 fatalx("IFINFO imsg with wrong len"); 2958 kif = imsg.data; 2959 depend_ok = kif->depend_state; 2960 2961 RB_FOREACH(p, peer_head, &conf->peers) 2962 if (!strcmp(p->conf.if_depend, kif->ifname)) { 2963 if (depend_ok && !p->depend_ok) { 2964 p->depend_ok = depend_ok; 2965 bgp_fsm(p, EVNT_START); 2966 } else if (!depend_ok && p->depend_ok) { 2967 p->depend_ok = depend_ok; 2968 session_stop(p, 2969 ERR_CEASE_OTHER_CHANGE); 2970 } 2971 } 2972 break; 2973 case IMSG_MRT_OPEN: 2974 case IMSG_MRT_REOPEN: 2975 if (imsg.hdr.len > IMSG_HEADER_SIZE + 2976 sizeof(struct mrt)) { 2977 log_warnx("wrong imsg len"); 2978 break; 2979 } 2980 2981 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 2982 if ((xmrt.wbuf.fd = imsg.fd) == -1) 2983 log_warnx("expected to receive fd for mrt dump " 2984 "but didn't receive any"); 2985 2986 mrt = mrt_get(&mrthead, &xmrt); 2987 if (mrt == NULL) { 2988 /* new dump */ 2989 mrt = calloc(1, sizeof(struct mrt)); 2990 if (mrt == NULL) 2991 fatal("session_dispatch_imsg"); 2992 memcpy(mrt, &xmrt, sizeof(struct mrt)); 2993 TAILQ_INIT(&mrt->wbuf.bufs); 2994 LIST_INSERT_HEAD(&mrthead, mrt, entry); 2995 } else { 2996 /* old dump reopened */ 2997 close(mrt->wbuf.fd); 2998 mrt->wbuf.fd = xmrt.wbuf.fd; 2999 } 3000 break; 3001 case IMSG_MRT_CLOSE: 3002 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3003 sizeof(struct mrt)) { 3004 log_warnx("wrong imsg len"); 3005 break; 3006 } 3007 3008 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 3009 mrt = mrt_get(&mrthead, &xmrt); 3010 if (mrt != NULL) 3011 mrt_done(mrt); 3012 break; 3013 case IMSG_CTL_KROUTE: 3014 case IMSG_CTL_KROUTE_ADDR: 3015 case IMSG_CTL_SHOW_NEXTHOP: 3016 case IMSG_CTL_SHOW_INTERFACE: 3017 case IMSG_CTL_SHOW_FIB_TABLES: 3018 case IMSG_CTL_SHOW_RTR: 3019 case IMSG_CTL_SHOW_TIMER: 3020 if (idx != PFD_PIPE_MAIN) 3021 fatalx("ctl kroute request not from parent"); 3022 control_imsg_relay(&imsg); 3023 break; 3024 case IMSG_CTL_SHOW_RIB: 3025 case IMSG_CTL_SHOW_RIB_PREFIX: 3026 case IMSG_CTL_SHOW_RIB_COMMUNITIES: 3027 case IMSG_CTL_SHOW_RIB_ATTR: 3028 case IMSG_CTL_SHOW_RIB_MEM: 3029 case IMSG_CTL_SHOW_RIB_HASH: 3030 case IMSG_CTL_SHOW_NETWORK: 3031 case IMSG_CTL_SHOW_NEIGHBOR: 3032 case IMSG_CTL_SHOW_SET: 3033 if (idx != PFD_PIPE_ROUTE_CTL) 3034 fatalx("ctl rib request not from RDE"); 3035 control_imsg_relay(&imsg); 3036 break; 3037 case IMSG_CTL_END: 3038 case IMSG_CTL_RESULT: 3039 control_imsg_relay(&imsg); 3040 break; 3041 case IMSG_UPDATE: 3042 if (idx != PFD_PIPE_ROUTE) 3043 fatalx("update request not from RDE"); 3044 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3045 MAX_PKTSIZE - MSGSIZE_HEADER || 3046 imsg.hdr.len < IMSG_HEADER_SIZE + 3047 MSGSIZE_UPDATE_MIN - MSGSIZE_HEADER) 3048 log_warnx("RDE sent invalid update"); 3049 else 3050 session_update(imsg.hdr.peerid, imsg.data, 3051 imsg.hdr.len - IMSG_HEADER_SIZE); 3052 break; 3053 case IMSG_UPDATE_ERR: 3054 if (idx != PFD_PIPE_ROUTE) 3055 fatalx("update request not from RDE"); 3056 if (imsg.hdr.len < IMSG_HEADER_SIZE + 2) { 3057 log_warnx("RDE sent invalid notification"); 3058 break; 3059 } 3060 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3061 log_warnx("no such peer: id=%u", 3062 imsg.hdr.peerid); 3063 break; 3064 } 3065 data = imsg.data; 3066 errcode = *data++; 3067 subcode = *data++; 3068 3069 if (imsg.hdr.len == IMSG_HEADER_SIZE + 2) 3070 data = NULL; 3071 3072 session_notification(p, errcode, subcode, 3073 data, imsg.hdr.len - IMSG_HEADER_SIZE - 2); 3074 switch (errcode) { 3075 case ERR_CEASE: 3076 switch (subcode) { 3077 case ERR_CEASE_MAX_PREFIX: 3078 case ERR_CEASE_MAX_SENT_PREFIX: 3079 t = p->conf.max_out_prefix_restart; 3080 if (subcode == ERR_CEASE_MAX_PREFIX) 3081 t = p->conf.max_prefix_restart; 3082 3083 bgp_fsm(p, EVNT_STOP); 3084 if (t) 3085 timer_set(&p->timers, 3086 Timer_IdleHold, 60 * t); 3087 break; 3088 default: 3089 bgp_fsm(p, EVNT_CON_FATAL); 3090 break; 3091 } 3092 break; 3093 default: 3094 bgp_fsm(p, EVNT_CON_FATAL); 3095 break; 3096 } 3097 break; 3098 case IMSG_SESSION_RESTARTED: 3099 if (idx != PFD_PIPE_ROUTE) 3100 fatalx("update request not from RDE"); 3101 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(aid)) { 3102 log_warnx("RDE sent invalid restart msg"); 3103 break; 3104 } 3105 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3106 log_warnx("no such peer: id=%u", 3107 imsg.hdr.peerid); 3108 break; 3109 } 3110 memcpy(&aid, imsg.data, sizeof(aid)); 3111 if (aid >= AID_MAX) 3112 fatalx("IMSG_SESSION_RESTARTED: bad AID"); 3113 if (p->capa.neg.grestart.flags[aid] & 3114 CAPA_GR_RESTARTING) { 3115 log_peer_warnx(&p->conf, 3116 "graceful restart of %s finished", 3117 aid2str(aid)); 3118 p->capa.neg.grestart.flags[aid] &= 3119 ~CAPA_GR_RESTARTING; 3120 timer_stop(&p->timers, Timer_RestartTimeout); 3121 3122 /* signal back to RDE to cleanup stale routes */ 3123 if (imsg_rde(IMSG_SESSION_RESTARTED, 3124 imsg.hdr.peerid, &aid, sizeof(aid)) == -1) 3125 fatal("imsg_compose: " 3126 "IMSG_SESSION_RESTARTED"); 3127 } 3128 break; 3129 case IMSG_SESSION_DOWN: 3130 if (idx != PFD_PIPE_ROUTE) 3131 fatalx("update request not from RDE"); 3132 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3133 log_warnx("no such peer: id=%u", 3134 imsg.hdr.peerid); 3135 break; 3136 } 3137 session_stop(p, ERR_CEASE_ADMIN_DOWN); 3138 break; 3139 default: 3140 break; 3141 } 3142 imsg_free(&imsg); 3143 } 3144 } 3145 3146 int 3147 la_cmp(struct listen_addr *a, struct listen_addr *b) 3148 { 3149 struct sockaddr_in *in_a, *in_b; 3150 struct sockaddr_in6 *in6_a, *in6_b; 3151 3152 if (a->sa.ss_family != b->sa.ss_family) 3153 return (1); 3154 3155 switch (a->sa.ss_family) { 3156 case AF_INET: 3157 in_a = (struct sockaddr_in *)&a->sa; 3158 in_b = (struct sockaddr_in *)&b->sa; 3159 if (in_a->sin_addr.s_addr != in_b->sin_addr.s_addr) 3160 return (1); 3161 if (in_a->sin_port != in_b->sin_port) 3162 return (1); 3163 break; 3164 case AF_INET6: 3165 in6_a = (struct sockaddr_in6 *)&a->sa; 3166 in6_b = (struct sockaddr_in6 *)&b->sa; 3167 if (bcmp(&in6_a->sin6_addr, &in6_b->sin6_addr, 3168 sizeof(struct in6_addr))) 3169 return (1); 3170 if (in6_a->sin6_port != in6_b->sin6_port) 3171 return (1); 3172 break; 3173 default: 3174 fatal("king bula sez: unknown address family"); 3175 /* NOTREACHED */ 3176 } 3177 3178 return (0); 3179 } 3180 3181 struct peer * 3182 getpeerbydesc(struct bgpd_config *c, const char *descr) 3183 { 3184 struct peer *p, *res = NULL; 3185 int match = 0; 3186 3187 RB_FOREACH(p, peer_head, &c->peers) 3188 if (!strcmp(p->conf.descr, descr)) { 3189 res = p; 3190 match++; 3191 } 3192 3193 if (match > 1) 3194 log_info("neighbor description \"%s\" not unique, request " 3195 "aborted", descr); 3196 3197 if (match == 1) 3198 return (res); 3199 else 3200 return (NULL); 3201 } 3202 3203 struct peer * 3204 getpeerbyip(struct bgpd_config *c, struct sockaddr *ip) 3205 { 3206 struct bgpd_addr addr; 3207 struct peer *p, *newpeer, *loose = NULL; 3208 u_int32_t id; 3209 3210 sa2addr(ip, &addr, NULL); 3211 3212 /* we might want a more effective way to find peers by IP */ 3213 RB_FOREACH(p, peer_head, &c->peers) 3214 if (!p->conf.template && 3215 !memcmp(&addr, &p->conf.remote_addr, sizeof(addr))) 3216 return (p); 3217 3218 /* try template matching */ 3219 RB_FOREACH(p, peer_head, &c->peers) 3220 if (p->conf.template && 3221 p->conf.remote_addr.aid == addr.aid && 3222 session_match_mask(p, &addr)) 3223 if (loose == NULL || loose->conf.remote_masklen < 3224 p->conf.remote_masklen) 3225 loose = p; 3226 3227 if (loose != NULL) { 3228 /* clone */ 3229 if ((newpeer = malloc(sizeof(struct peer))) == NULL) 3230 fatal(NULL); 3231 memcpy(newpeer, loose, sizeof(struct peer)); 3232 for (id = PEER_ID_DYN_MAX; id > PEER_ID_STATIC_MAX; id--) { 3233 if (getpeerbyid(c, id) == NULL) /* we found a free id */ 3234 break; 3235 } 3236 newpeer->template = loose; 3237 session_template_clone(newpeer, ip, id, 0); 3238 newpeer->state = newpeer->prev_state = STATE_NONE; 3239 newpeer->reconf_action = RECONF_KEEP; 3240 newpeer->rbuf = NULL; 3241 newpeer->rpending = 0; 3242 init_peer(newpeer); 3243 bgp_fsm(newpeer, EVNT_START); 3244 if (RB_INSERT(peer_head, &c->peers, newpeer) != NULL) 3245 fatalx("%s: peer tree is corrupt", __func__); 3246 return (newpeer); 3247 } 3248 3249 return (NULL); 3250 } 3251 3252 struct peer * 3253 getpeerbyid(struct bgpd_config *c, u_int32_t peerid) 3254 { 3255 static struct peer lookup; 3256 3257 lookup.conf.id = peerid; 3258 3259 return RB_FIND(peer_head, &c->peers, &lookup); 3260 } 3261 3262 int 3263 peer_matched(struct peer *p, struct ctl_neighbor *n) 3264 { 3265 char *s; 3266 3267 if (n && n->addr.aid) { 3268 if (memcmp(&p->conf.remote_addr, &n->addr, 3269 sizeof(p->conf.remote_addr))) 3270 return 0; 3271 } else if (n && n->descr[0]) { 3272 s = n->is_group ? p->conf.group : p->conf.descr; 3273 if (strcmp(s, n->descr)) 3274 return 0; 3275 } 3276 return 1; 3277 } 3278 3279 void 3280 session_template_clone(struct peer *p, struct sockaddr *ip, u_int32_t id, 3281 u_int32_t as) 3282 { 3283 struct bgpd_addr remote_addr; 3284 3285 if (ip) 3286 sa2addr(ip, &remote_addr, NULL); 3287 else 3288 memcpy(&remote_addr, &p->conf.remote_addr, sizeof(remote_addr)); 3289 3290 memcpy(&p->conf, &p->template->conf, sizeof(struct peer_config)); 3291 3292 p->conf.id = id; 3293 3294 if (as) { 3295 p->conf.remote_as = as; 3296 p->conf.ebgp = (p->conf.remote_as != p->conf.local_as); 3297 if (!p->conf.ebgp) 3298 /* force enforce_as off for iBGP sessions */ 3299 p->conf.enforce_as = ENFORCE_AS_OFF; 3300 } 3301 3302 memcpy(&p->conf.remote_addr, &remote_addr, sizeof(remote_addr)); 3303 switch (p->conf.remote_addr.aid) { 3304 case AID_INET: 3305 p->conf.remote_masklen = 32; 3306 break; 3307 case AID_INET6: 3308 p->conf.remote_masklen = 128; 3309 break; 3310 } 3311 p->conf.template = 0; 3312 } 3313 3314 int 3315 session_match_mask(struct peer *p, struct bgpd_addr *a) 3316 { 3317 struct in_addr v4masked; 3318 struct in6_addr v6masked; 3319 3320 switch (p->conf.remote_addr.aid) { 3321 case AID_INET: 3322 inet4applymask(&v4masked, &a->v4, p->conf.remote_masklen); 3323 if (p->conf.remote_addr.v4.s_addr == v4masked.s_addr) 3324 return (1); 3325 return (0); 3326 case AID_INET6: 3327 inet6applymask(&v6masked, &a->v6, p->conf.remote_masklen); 3328 3329 if (memcmp(&v6masked, &p->conf.remote_addr.v6, 3330 sizeof(v6masked)) == 0) 3331 return (1); 3332 return (0); 3333 } 3334 return (0); 3335 } 3336 3337 void 3338 session_down(struct peer *peer) 3339 { 3340 bzero(&peer->capa.neg, sizeof(peer->capa.neg)); 3341 peer->stats.last_updown = getmonotime(); 3342 /* 3343 * session_down is called in the exit code path so check 3344 * if the RDE is still around, if not there is no need to 3345 * send the message. 3346 */ 3347 if (ibuf_rde == NULL) 3348 return; 3349 if (imsg_rde(IMSG_SESSION_DOWN, peer->conf.id, NULL, 0) == -1) 3350 fatalx("imsg_compose error"); 3351 } 3352 3353 void 3354 session_up(struct peer *p) 3355 { 3356 struct session_up sup; 3357 3358 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3359 &p->conf, sizeof(p->conf)) == -1) 3360 fatalx("imsg_compose error"); 3361 3362 if (p->local.aid == AID_INET) { 3363 sup.local_v4_addr = p->local; 3364 sup.local_v6_addr = p->local_alt; 3365 } else { 3366 sup.local_v6_addr = p->local; 3367 sup.local_v4_addr = p->local_alt; 3368 } 3369 sup.remote_addr = p->remote; 3370 3371 sup.remote_bgpid = p->remote_bgpid; 3372 sup.short_as = p->short_as; 3373 memcpy(&sup.capa, &p->capa.neg, sizeof(sup.capa)); 3374 p->stats.last_updown = getmonotime(); 3375 if (imsg_rde(IMSG_SESSION_UP, p->conf.id, &sup, sizeof(sup)) == -1) 3376 fatalx("imsg_compose error"); 3377 } 3378 3379 int 3380 imsg_ctl_parent(int type, u_int32_t peerid, pid_t pid, void *data, 3381 u_int16_t datalen) 3382 { 3383 return (imsg_compose(ibuf_main, type, peerid, pid, -1, data, datalen)); 3384 } 3385 3386 int 3387 imsg_ctl_rde(int type, pid_t pid, void *data, u_int16_t datalen) 3388 { 3389 if (ibuf_rde_ctl == NULL) { 3390 log_warnx("Can't send message %u to RDE, ctl pipe closed", 3391 type); 3392 return (0); 3393 } 3394 /* 3395 * Use control socket to talk to RDE to bypass the queue of the 3396 * regular imsg socket. 3397 */ 3398 return (imsg_compose(ibuf_rde_ctl, type, 0, pid, -1, data, datalen)); 3399 } 3400 3401 int 3402 imsg_rde(int type, uint32_t peerid, void *data, u_int16_t datalen) 3403 { 3404 if (ibuf_rde == NULL) { 3405 log_warnx("Can't send message %u to RDE, pipe closed", type); 3406 return (0); 3407 } 3408 3409 return (imsg_compose(ibuf_rde, type, peerid, 0, -1, data, datalen)); 3410 } 3411 3412 void 3413 session_demote(struct peer *p, int level) 3414 { 3415 struct demote_msg msg; 3416 3417 strlcpy(msg.demote_group, p->conf.demote_group, 3418 sizeof(msg.demote_group)); 3419 msg.level = level; 3420 if (imsg_compose(ibuf_main, IMSG_DEMOTE, p->conf.id, 0, -1, 3421 &msg, sizeof(msg)) == -1) 3422 fatalx("imsg_compose error"); 3423 3424 p->demoted += level; 3425 } 3426 3427 void 3428 session_stop(struct peer *peer, u_int8_t subcode) 3429 { 3430 char data[REASON_LEN]; 3431 size_t datalen; 3432 size_t reason_len; 3433 char *communication; 3434 3435 datalen = 0; 3436 communication = peer->conf.reason; 3437 3438 if ((subcode == ERR_CEASE_ADMIN_DOWN || 3439 subcode == ERR_CEASE_ADMIN_RESET) 3440 && communication && *communication) { 3441 reason_len = strlen(communication); 3442 if (reason_len > REASON_LEN - 1) { 3443 log_peer_warnx(&peer->conf, 3444 "trying to send overly long shutdown reason"); 3445 } else { 3446 data[0] = reason_len; 3447 datalen = reason_len + sizeof(data[0]); 3448 memcpy(data + 1, communication, reason_len); 3449 } 3450 } 3451 switch (peer->state) { 3452 case STATE_OPENSENT: 3453 case STATE_OPENCONFIRM: 3454 case STATE_ESTABLISHED: 3455 session_notification(peer, ERR_CEASE, subcode, data, datalen); 3456 break; 3457 default: 3458 /* session not open, no need to send notification */ 3459 break; 3460 } 3461 bgp_fsm(peer, EVNT_STOP); 3462 } 3463 3464 void 3465 merge_peers(struct bgpd_config *c, struct bgpd_config *nc) 3466 { 3467 struct peer *p, *np, *next; 3468 3469 RB_FOREACH(p, peer_head, &c->peers) { 3470 /* templates are handled specially */ 3471 if (p->template != NULL) 3472 continue; 3473 np = getpeerbyid(nc, p->conf.id); 3474 if (np == NULL) { 3475 p->reconf_action = RECONF_DELETE; 3476 continue; 3477 } 3478 3479 /* peer no longer uses TCP MD5SIG so deconfigure */ 3480 if (p->conf.auth.method == AUTH_MD5SIG && 3481 np->conf.auth.method != AUTH_MD5SIG) 3482 tcp_md5_del_listener(c, p); 3483 else if (np->conf.auth.method == AUTH_MD5SIG) 3484 tcp_md5_add_listener(c, np); 3485 3486 memcpy(&p->conf, &np->conf, sizeof(p->conf)); 3487 RB_REMOVE(peer_head, &nc->peers, np); 3488 free(np); 3489 3490 p->reconf_action = RECONF_KEEP; 3491 3492 /* had demotion, is demoted, demote removed? */ 3493 if (p->demoted && !p->conf.demote_group[0]) 3494 session_demote(p, -1); 3495 3496 /* if session is not open then refresh pfkey data */ 3497 if (p->state < STATE_OPENSENT && !p->template) 3498 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 3499 p->conf.id, 0, -1, NULL, 0); 3500 3501 /* sync the RDE in case we keep the peer */ 3502 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3503 &p->conf, sizeof(struct peer_config)) == -1) 3504 fatalx("imsg_compose error"); 3505 3506 /* apply the config to all clones of a template */ 3507 if (p->conf.template) { 3508 struct peer *xp; 3509 RB_FOREACH(xp, peer_head, &c->peers) { 3510 if (xp->template != p) 3511 continue; 3512 session_template_clone(xp, NULL, xp->conf.id, 3513 xp->conf.remote_as); 3514 if (imsg_rde(IMSG_SESSION_ADD, xp->conf.id, 3515 &xp->conf, sizeof(xp->conf)) == -1) 3516 fatalx("imsg_compose error"); 3517 } 3518 } 3519 } 3520 3521 if (imsg_rde(IMSG_RECONF_DRAIN, 0, NULL, 0) == -1) 3522 fatalx("imsg_compose error"); 3523 3524 /* pfkeys of new peers already loaded by the parent process */ 3525 RB_FOREACH_SAFE(np, peer_head, &nc->peers, next) { 3526 RB_REMOVE(peer_head, &nc->peers, np); 3527 if (RB_INSERT(peer_head, &c->peers, np) != NULL) 3528 fatalx("%s: peer tree is corrupt", __func__); 3529 if (np->conf.auth.method == AUTH_MD5SIG) 3530 tcp_md5_add_listener(c, np); 3531 } 3532 } 3533