1 /* $OpenBSD: session.c,v 1.445 2023/05/25 14:20:25 claudio Exp $ */ 2 3 /* 4 * Copyright (c) 2003, 2004, 2005 Henning Brauer <henning@openbsd.org> 5 * Copyright (c) 2017 Peter van Dijk <peter.van.dijk@powerdns.com> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include <sys/types.h> 21 22 #include <sys/mman.h> 23 #include <sys/socket.h> 24 #include <sys/time.h> 25 #include <sys/resource.h> 26 #include <sys/un.h> 27 #include <netinet/in.h> 28 #include <netinet/ip.h> 29 #include <netinet/tcp.h> 30 #include <arpa/inet.h> 31 #include <limits.h> 32 33 #include <err.h> 34 #include <errno.h> 35 #include <fcntl.h> 36 #include <ifaddrs.h> 37 #include <poll.h> 38 #include <pwd.h> 39 #include <signal.h> 40 #include <stdio.h> 41 #include <stdlib.h> 42 #include <string.h> 43 #include <syslog.h> 44 #include <unistd.h> 45 46 #include "bgpd.h" 47 #include "session.h" 48 #include "log.h" 49 50 #define PFD_PIPE_MAIN 0 51 #define PFD_PIPE_ROUTE 1 52 #define PFD_PIPE_ROUTE_CTL 2 53 #define PFD_SOCK_CTL 3 54 #define PFD_SOCK_RCTL 4 55 #define PFD_LISTENERS_START 5 56 57 void session_sighdlr(int); 58 int setup_listeners(u_int *); 59 void init_peer(struct peer *); 60 void start_timer_holdtime(struct peer *); 61 void start_timer_keepalive(struct peer *); 62 void session_close_connection(struct peer *); 63 void change_state(struct peer *, enum session_state, enum session_events); 64 int session_setup_socket(struct peer *); 65 void session_accept(int); 66 int session_connect(struct peer *); 67 void session_tcp_established(struct peer *); 68 void session_capa_ann_none(struct peer *); 69 int session_capa_add(struct ibuf *, uint8_t, uint8_t); 70 int session_capa_add_mp(struct ibuf *, uint8_t); 71 int session_capa_add_afi(struct peer *, struct ibuf *, uint8_t, uint8_t); 72 struct bgp_msg *session_newmsg(enum msg_type, uint16_t); 73 int session_sendmsg(struct bgp_msg *, struct peer *); 74 void session_open(struct peer *); 75 void session_keepalive(struct peer *); 76 void session_update(uint32_t, void *, size_t); 77 void session_notification(struct peer *, uint8_t, uint8_t, void *, 78 ssize_t); 79 void session_rrefresh(struct peer *, uint8_t, uint8_t); 80 int session_graceful_restart(struct peer *); 81 int session_graceful_stop(struct peer *); 82 int session_dispatch_msg(struct pollfd *, struct peer *); 83 void session_process_msg(struct peer *); 84 int parse_header(struct peer *, u_char *, uint16_t *, uint8_t *); 85 int parse_open(struct peer *); 86 int parse_update(struct peer *); 87 int parse_rrefresh(struct peer *); 88 int parse_notification(struct peer *); 89 int parse_capabilities(struct peer *, u_char *, uint16_t, uint32_t *); 90 int capa_neg_calc(struct peer *, uint8_t *); 91 void session_dispatch_imsg(struct imsgbuf *, int, u_int *); 92 void session_up(struct peer *); 93 void session_down(struct peer *); 94 int imsg_rde(int, uint32_t, void *, uint16_t); 95 void session_demote(struct peer *, int); 96 void merge_peers(struct bgpd_config *, struct bgpd_config *); 97 98 int la_cmp(struct listen_addr *, struct listen_addr *); 99 void session_template_clone(struct peer *, struct sockaddr *, 100 uint32_t, uint32_t); 101 int session_match_mask(struct peer *, struct bgpd_addr *); 102 103 static struct bgpd_config *conf, *nconf; 104 static struct imsgbuf *ibuf_rde; 105 static struct imsgbuf *ibuf_rde_ctl; 106 static struct imsgbuf *ibuf_main; 107 108 struct bgpd_sysdep sysdep; 109 volatile sig_atomic_t session_quit; 110 int pending_reconf; 111 int csock = -1, rcsock = -1; 112 u_int peer_cnt; 113 114 struct mrt_head mrthead; 115 time_t pauseaccept; 116 117 static inline int 118 peer_compare(const struct peer *a, const struct peer *b) 119 { 120 return a->conf.id - b->conf.id; 121 } 122 123 RB_GENERATE(peer_head, peer, entry, peer_compare); 124 125 void 126 session_sighdlr(int sig) 127 { 128 switch (sig) { 129 case SIGINT: 130 case SIGTERM: 131 session_quit = 1; 132 break; 133 } 134 } 135 136 int 137 setup_listeners(u_int *la_cnt) 138 { 139 int ttl = 255; 140 struct listen_addr *la; 141 u_int cnt = 0; 142 143 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 144 la->reconf = RECONF_NONE; 145 cnt++; 146 147 if (la->flags & LISTENER_LISTENING) 148 continue; 149 150 if (la->fd == -1) { 151 log_warn("cannot establish listener on %s: invalid fd", 152 log_sockaddr((struct sockaddr *)&la->sa, 153 la->sa_len)); 154 continue; 155 } 156 157 if (tcp_md5_prep_listener(la, &conf->peers) == -1) 158 fatal("tcp_md5_prep_listener"); 159 160 /* set ttl to 255 so that ttl-security works */ 161 if (la->sa.ss_family == AF_INET && setsockopt(la->fd, 162 IPPROTO_IP, IP_TTL, &ttl, sizeof(ttl)) == -1) { 163 log_warn("setup_listeners setsockopt TTL"); 164 continue; 165 } 166 if (la->sa.ss_family == AF_INET6 && setsockopt(la->fd, 167 IPPROTO_IPV6, IPV6_UNICAST_HOPS, &ttl, sizeof(ttl)) == -1) { 168 log_warn("setup_listeners setsockopt hoplimit"); 169 continue; 170 } 171 172 if (listen(la->fd, MAX_BACKLOG)) { 173 close(la->fd); 174 fatal("listen"); 175 } 176 177 la->flags |= LISTENER_LISTENING; 178 179 log_info("listening on %s", 180 log_sockaddr((struct sockaddr *)&la->sa, la->sa_len)); 181 } 182 183 *la_cnt = cnt; 184 185 return (0); 186 } 187 188 void 189 session_main(int debug, int verbose) 190 { 191 int timeout; 192 unsigned int i, j, idx_peers, idx_listeners, idx_mrts; 193 u_int pfd_elms = 0, peer_l_elms = 0, mrt_l_elms = 0; 194 u_int listener_cnt, ctl_cnt, mrt_cnt; 195 u_int new_cnt; 196 struct passwd *pw; 197 struct peer *p, **peer_l = NULL, *next; 198 struct mrt *m, *xm, **mrt_l = NULL; 199 struct pollfd *pfd = NULL; 200 struct listen_addr *la; 201 void *newp; 202 time_t now; 203 short events; 204 205 log_init(debug, LOG_DAEMON); 206 log_setverbose(verbose); 207 208 log_procinit(log_procnames[PROC_SE]); 209 210 if ((pw = getpwnam(BGPD_USER)) == NULL) 211 fatal(NULL); 212 213 if (chroot(pw->pw_dir) == -1) 214 fatal("chroot"); 215 if (chdir("/") == -1) 216 fatal("chdir(\"/\")"); 217 218 setproctitle("session engine"); 219 220 if (setgroups(1, &pw->pw_gid) || 221 setresgid(pw->pw_gid, pw->pw_gid, pw->pw_gid) || 222 setresuid(pw->pw_uid, pw->pw_uid, pw->pw_uid)) 223 fatal("can't drop privileges"); 224 225 if (pledge("stdio inet recvfd", NULL) == -1) 226 fatal("pledge"); 227 228 signal(SIGTERM, session_sighdlr); 229 signal(SIGINT, session_sighdlr); 230 signal(SIGPIPE, SIG_IGN); 231 signal(SIGHUP, SIG_IGN); 232 signal(SIGALRM, SIG_IGN); 233 signal(SIGUSR1, SIG_IGN); 234 235 if ((ibuf_main = malloc(sizeof(struct imsgbuf))) == NULL) 236 fatal(NULL); 237 imsg_init(ibuf_main, 3); 238 239 LIST_INIT(&mrthead); 240 listener_cnt = 0; 241 peer_cnt = 0; 242 ctl_cnt = 0; 243 244 conf = new_config(); 245 log_info("session engine ready"); 246 247 while (session_quit == 0) { 248 /* check for peers to be initialized or deleted */ 249 if (!pending_reconf) { 250 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 251 /* cloned peer that idled out? */ 252 if (p->template && (p->state == STATE_IDLE || 253 p->state == STATE_ACTIVE) && 254 getmonotime() - p->stats.last_updown >= 255 INTERVAL_HOLD_CLONED) 256 p->reconf_action = RECONF_DELETE; 257 258 /* new peer that needs init? */ 259 if (p->state == STATE_NONE) 260 init_peer(p); 261 262 /* deletion due? */ 263 if (p->reconf_action == RECONF_DELETE) { 264 if (p->demoted) 265 session_demote(p, -1); 266 p->conf.demote_group[0] = 0; 267 session_stop(p, ERR_CEASE_PEER_UNCONF); 268 timer_remove_all(&p->timers); 269 tcp_md5_del_listener(conf, p); 270 log_peer_warnx(&p->conf, "removed"); 271 RB_REMOVE(peer_head, &conf->peers, p); 272 free(p); 273 peer_cnt--; 274 continue; 275 } 276 p->reconf_action = RECONF_NONE; 277 } 278 } 279 280 if (peer_cnt > peer_l_elms) { 281 if ((newp = reallocarray(peer_l, peer_cnt, 282 sizeof(struct peer *))) == NULL) { 283 /* panic for now */ 284 log_warn("could not resize peer_l from %u -> %u" 285 " entries", peer_l_elms, peer_cnt); 286 fatalx("exiting"); 287 } 288 peer_l = newp; 289 peer_l_elms = peer_cnt; 290 } 291 292 mrt_cnt = 0; 293 for (m = LIST_FIRST(&mrthead); m != NULL; m = xm) { 294 xm = LIST_NEXT(m, entry); 295 if (m->state == MRT_STATE_REMOVE) { 296 mrt_clean(m); 297 LIST_REMOVE(m, entry); 298 free(m); 299 continue; 300 } 301 if (m->wbuf.queued) 302 mrt_cnt++; 303 } 304 305 if (mrt_cnt > mrt_l_elms) { 306 if ((newp = reallocarray(mrt_l, mrt_cnt, 307 sizeof(struct mrt *))) == NULL) { 308 /* panic for now */ 309 log_warn("could not resize mrt_l from %u -> %u" 310 " entries", mrt_l_elms, mrt_cnt); 311 fatalx("exiting"); 312 } 313 mrt_l = newp; 314 mrt_l_elms = mrt_cnt; 315 } 316 317 new_cnt = PFD_LISTENERS_START + listener_cnt + peer_cnt + 318 ctl_cnt + mrt_cnt; 319 if (new_cnt > pfd_elms) { 320 if ((newp = reallocarray(pfd, new_cnt, 321 sizeof(struct pollfd))) == NULL) { 322 /* panic for now */ 323 log_warn("could not resize pfd from %u -> %u" 324 " entries", pfd_elms, new_cnt); 325 fatalx("exiting"); 326 } 327 pfd = newp; 328 pfd_elms = new_cnt; 329 } 330 331 memset(pfd, 0, sizeof(struct pollfd) * pfd_elms); 332 333 set_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main); 334 set_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde); 335 set_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl); 336 337 if (pauseaccept == 0) { 338 pfd[PFD_SOCK_CTL].fd = csock; 339 pfd[PFD_SOCK_CTL].events = POLLIN; 340 pfd[PFD_SOCK_RCTL].fd = rcsock; 341 pfd[PFD_SOCK_RCTL].events = POLLIN; 342 } else { 343 pfd[PFD_SOCK_CTL].fd = -1; 344 pfd[PFD_SOCK_RCTL].fd = -1; 345 } 346 347 i = PFD_LISTENERS_START; 348 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 349 if (pauseaccept == 0) { 350 pfd[i].fd = la->fd; 351 pfd[i].events = POLLIN; 352 } else 353 pfd[i].fd = -1; 354 i++; 355 } 356 idx_listeners = i; 357 timeout = 240; /* loop every 240s at least */ 358 359 now = getmonotime(); 360 RB_FOREACH(p, peer_head, &conf->peers) { 361 time_t nextaction; 362 struct timer *pt; 363 364 /* check timers */ 365 if ((pt = timer_nextisdue(&p->timers, now)) != NULL) { 366 switch (pt->type) { 367 case Timer_Hold: 368 bgp_fsm(p, EVNT_TIMER_HOLDTIME); 369 break; 370 case Timer_SendHold: 371 bgp_fsm(p, EVNT_TIMER_SENDHOLD); 372 break; 373 case Timer_ConnectRetry: 374 bgp_fsm(p, EVNT_TIMER_CONNRETRY); 375 break; 376 case Timer_Keepalive: 377 bgp_fsm(p, EVNT_TIMER_KEEPALIVE); 378 break; 379 case Timer_IdleHold: 380 bgp_fsm(p, EVNT_START); 381 break; 382 case Timer_IdleHoldReset: 383 p->IdleHoldTime = 384 INTERVAL_IDLE_HOLD_INITIAL; 385 p->errcnt = 0; 386 timer_stop(&p->timers, 387 Timer_IdleHoldReset); 388 break; 389 case Timer_CarpUndemote: 390 timer_stop(&p->timers, 391 Timer_CarpUndemote); 392 if (p->demoted && 393 p->state == STATE_ESTABLISHED) 394 session_demote(p, -1); 395 break; 396 case Timer_RestartTimeout: 397 timer_stop(&p->timers, 398 Timer_RestartTimeout); 399 session_graceful_stop(p); 400 break; 401 default: 402 fatalx("King Bula lost in time"); 403 } 404 } 405 if ((nextaction = timer_nextduein(&p->timers, 406 now)) != -1 && nextaction < timeout) 407 timeout = nextaction; 408 409 /* are we waiting for a write? */ 410 events = POLLIN; 411 if (p->wbuf.queued > 0 || p->state == STATE_CONNECT) 412 events |= POLLOUT; 413 /* is there still work to do? */ 414 if (p->rpending && p->rbuf && p->rbuf->wpos) 415 timeout = 0; 416 417 /* poll events */ 418 if (p->fd != -1 && events != 0) { 419 pfd[i].fd = p->fd; 420 pfd[i].events = events; 421 peer_l[i - idx_listeners] = p; 422 i++; 423 } 424 } 425 426 idx_peers = i; 427 428 LIST_FOREACH(m, &mrthead, entry) 429 if (m->wbuf.queued) { 430 pfd[i].fd = m->wbuf.fd; 431 pfd[i].events = POLLOUT; 432 mrt_l[i - idx_peers] = m; 433 i++; 434 } 435 436 idx_mrts = i; 437 438 i += control_fill_pfds(pfd + i, pfd_elms -i); 439 440 if (i > pfd_elms) 441 fatalx("poll pfd overflow"); 442 443 if (pauseaccept && timeout > 1) 444 timeout = 1; 445 if (timeout < 0) 446 timeout = 0; 447 if (poll(pfd, i, timeout * 1000) == -1) { 448 if (errno == EINTR) 449 continue; 450 fatal("poll error"); 451 } 452 453 /* 454 * If we previously saw fd exhaustion, we stop accept() 455 * for 1 second to throttle the accept() loop. 456 */ 457 if (pauseaccept && getmonotime() > pauseaccept + 1) 458 pauseaccept = 0; 459 460 if (handle_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main) == -1) { 461 log_warnx("SE: Lost connection to parent"); 462 session_quit = 1; 463 continue; 464 } else 465 session_dispatch_imsg(ibuf_main, PFD_PIPE_MAIN, 466 &listener_cnt); 467 468 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde) == -1) { 469 log_warnx("SE: Lost connection to RDE"); 470 msgbuf_clear(&ibuf_rde->w); 471 free(ibuf_rde); 472 ibuf_rde = NULL; 473 } else 474 session_dispatch_imsg(ibuf_rde, PFD_PIPE_ROUTE, 475 &listener_cnt); 476 477 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl) == 478 -1) { 479 log_warnx("SE: Lost connection to RDE control"); 480 msgbuf_clear(&ibuf_rde_ctl->w); 481 free(ibuf_rde_ctl); 482 ibuf_rde_ctl = NULL; 483 } else 484 session_dispatch_imsg(ibuf_rde_ctl, PFD_PIPE_ROUTE_CTL, 485 &listener_cnt); 486 487 if (pfd[PFD_SOCK_CTL].revents & POLLIN) 488 ctl_cnt += control_accept(csock, 0); 489 490 if (pfd[PFD_SOCK_RCTL].revents & POLLIN) 491 ctl_cnt += control_accept(rcsock, 1); 492 493 for (j = PFD_LISTENERS_START; j < idx_listeners; j++) 494 if (pfd[j].revents & POLLIN) 495 session_accept(pfd[j].fd); 496 497 for (; j < idx_peers; j++) 498 session_dispatch_msg(&pfd[j], 499 peer_l[j - idx_listeners]); 500 501 RB_FOREACH(p, peer_head, &conf->peers) 502 if (p->rbuf && p->rbuf->wpos) 503 session_process_msg(p); 504 505 for (; j < idx_mrts; j++) 506 if (pfd[j].revents & POLLOUT) 507 mrt_write(mrt_l[j - idx_peers]); 508 509 for (; j < i; j++) 510 ctl_cnt -= control_dispatch_msg(&pfd[j], &conf->peers); 511 } 512 513 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 514 RB_REMOVE(peer_head, &conf->peers, p); 515 strlcpy(p->conf.reason, 516 "bgpd shutting down", 517 sizeof(p->conf.reason)); 518 session_stop(p, ERR_CEASE_ADMIN_DOWN); 519 timer_remove_all(&p->timers); 520 free(p); 521 } 522 523 while ((m = LIST_FIRST(&mrthead)) != NULL) { 524 mrt_clean(m); 525 LIST_REMOVE(m, entry); 526 free(m); 527 } 528 529 free_config(conf); 530 free(peer_l); 531 free(mrt_l); 532 free(pfd); 533 534 /* close pipes */ 535 if (ibuf_rde) { 536 msgbuf_write(&ibuf_rde->w); 537 msgbuf_clear(&ibuf_rde->w); 538 close(ibuf_rde->fd); 539 free(ibuf_rde); 540 } 541 if (ibuf_rde_ctl) { 542 msgbuf_clear(&ibuf_rde_ctl->w); 543 close(ibuf_rde_ctl->fd); 544 free(ibuf_rde_ctl); 545 } 546 msgbuf_write(&ibuf_main->w); 547 msgbuf_clear(&ibuf_main->w); 548 close(ibuf_main->fd); 549 free(ibuf_main); 550 551 control_shutdown(csock); 552 control_shutdown(rcsock); 553 log_info("session engine exiting"); 554 exit(0); 555 } 556 557 void 558 init_peer(struct peer *p) 559 { 560 TAILQ_INIT(&p->timers); 561 p->fd = p->wbuf.fd = -1; 562 563 if (p->conf.if_depend[0]) 564 imsg_compose(ibuf_main, IMSG_SESSION_DEPENDON, 0, 0, -1, 565 p->conf.if_depend, sizeof(p->conf.if_depend)); 566 else 567 p->depend_ok = 1; 568 569 peer_cnt++; 570 571 change_state(p, STATE_IDLE, EVNT_NONE); 572 if (p->conf.down) 573 timer_stop(&p->timers, Timer_IdleHold); /* no autostart */ 574 else 575 timer_set(&p->timers, Timer_IdleHold, SESSION_CLEAR_DELAY); 576 577 p->stats.last_updown = getmonotime(); 578 579 /* 580 * on startup, demote if requested. 581 * do not handle new peers. they must reach ESTABLISHED beforehand. 582 * peers added at runtime have reconf_action set to RECONF_REINIT. 583 */ 584 if (p->reconf_action != RECONF_REINIT && p->conf.demote_group[0]) 585 session_demote(p, +1); 586 } 587 588 void 589 bgp_fsm(struct peer *peer, enum session_events event) 590 { 591 switch (peer->state) { 592 case STATE_NONE: 593 /* nothing */ 594 break; 595 case STATE_IDLE: 596 switch (event) { 597 case EVNT_START: 598 timer_stop(&peer->timers, Timer_Hold); 599 timer_stop(&peer->timers, Timer_SendHold); 600 timer_stop(&peer->timers, Timer_Keepalive); 601 timer_stop(&peer->timers, Timer_IdleHold); 602 603 /* allocate read buffer */ 604 peer->rbuf = calloc(1, sizeof(struct ibuf_read)); 605 if (peer->rbuf == NULL) 606 fatal(NULL); 607 608 /* init write buffer */ 609 msgbuf_init(&peer->wbuf); 610 611 peer->stats.last_sent_errcode = 0; 612 peer->stats.last_sent_suberr = 0; 613 peer->stats.last_rcvd_errcode = 0; 614 peer->stats.last_rcvd_suberr = 0; 615 616 if (!peer->depend_ok) 617 timer_stop(&peer->timers, Timer_ConnectRetry); 618 else if (peer->passive || peer->conf.passive || 619 peer->conf.template) { 620 change_state(peer, STATE_ACTIVE, event); 621 timer_stop(&peer->timers, Timer_ConnectRetry); 622 } else { 623 change_state(peer, STATE_CONNECT, event); 624 timer_set(&peer->timers, Timer_ConnectRetry, 625 conf->connectretry); 626 session_connect(peer); 627 } 628 peer->passive = 0; 629 break; 630 default: 631 /* ignore */ 632 break; 633 } 634 break; 635 case STATE_CONNECT: 636 switch (event) { 637 case EVNT_START: 638 /* ignore */ 639 break; 640 case EVNT_CON_OPEN: 641 session_tcp_established(peer); 642 session_open(peer); 643 timer_stop(&peer->timers, Timer_ConnectRetry); 644 peer->holdtime = INTERVAL_HOLD_INITIAL; 645 start_timer_holdtime(peer); 646 change_state(peer, STATE_OPENSENT, event); 647 break; 648 case EVNT_CON_OPENFAIL: 649 timer_set(&peer->timers, Timer_ConnectRetry, 650 conf->connectretry); 651 session_close_connection(peer); 652 change_state(peer, STATE_ACTIVE, event); 653 break; 654 case EVNT_TIMER_CONNRETRY: 655 timer_set(&peer->timers, Timer_ConnectRetry, 656 conf->connectretry); 657 session_connect(peer); 658 break; 659 default: 660 change_state(peer, STATE_IDLE, event); 661 break; 662 } 663 break; 664 case STATE_ACTIVE: 665 switch (event) { 666 case EVNT_START: 667 /* ignore */ 668 break; 669 case EVNT_CON_OPEN: 670 session_tcp_established(peer); 671 session_open(peer); 672 timer_stop(&peer->timers, Timer_ConnectRetry); 673 peer->holdtime = INTERVAL_HOLD_INITIAL; 674 start_timer_holdtime(peer); 675 change_state(peer, STATE_OPENSENT, event); 676 break; 677 case EVNT_CON_OPENFAIL: 678 timer_set(&peer->timers, Timer_ConnectRetry, 679 conf->connectretry); 680 session_close_connection(peer); 681 change_state(peer, STATE_ACTIVE, event); 682 break; 683 case EVNT_TIMER_CONNRETRY: 684 timer_set(&peer->timers, Timer_ConnectRetry, 685 peer->holdtime); 686 change_state(peer, STATE_CONNECT, event); 687 session_connect(peer); 688 break; 689 default: 690 change_state(peer, STATE_IDLE, event); 691 break; 692 } 693 break; 694 case STATE_OPENSENT: 695 switch (event) { 696 case EVNT_START: 697 /* ignore */ 698 break; 699 case EVNT_STOP: 700 change_state(peer, STATE_IDLE, event); 701 break; 702 case EVNT_CON_CLOSED: 703 session_close_connection(peer); 704 timer_set(&peer->timers, Timer_ConnectRetry, 705 conf->connectretry); 706 change_state(peer, STATE_ACTIVE, event); 707 break; 708 case EVNT_CON_FATAL: 709 change_state(peer, STATE_IDLE, event); 710 break; 711 case EVNT_TIMER_HOLDTIME: 712 case EVNT_TIMER_SENDHOLD: 713 session_notification(peer, ERR_HOLDTIMEREXPIRED, 714 0, NULL, 0); 715 change_state(peer, STATE_IDLE, event); 716 break; 717 case EVNT_RCVD_OPEN: 718 /* parse_open calls change_state itself on failure */ 719 if (parse_open(peer)) 720 break; 721 session_keepalive(peer); 722 change_state(peer, STATE_OPENCONFIRM, event); 723 break; 724 case EVNT_RCVD_NOTIFICATION: 725 if (parse_notification(peer)) { 726 change_state(peer, STATE_IDLE, event); 727 /* don't punish, capa negotiation */ 728 timer_set(&peer->timers, Timer_IdleHold, 0); 729 peer->IdleHoldTime /= 2; 730 } else 731 change_state(peer, STATE_IDLE, event); 732 break; 733 default: 734 session_notification(peer, 735 ERR_FSM, ERR_FSM_UNEX_OPENSENT, NULL, 0); 736 change_state(peer, STATE_IDLE, event); 737 break; 738 } 739 break; 740 case STATE_OPENCONFIRM: 741 switch (event) { 742 case EVNT_START: 743 /* ignore */ 744 break; 745 case EVNT_STOP: 746 change_state(peer, STATE_IDLE, event); 747 break; 748 case EVNT_CON_CLOSED: 749 case EVNT_CON_FATAL: 750 change_state(peer, STATE_IDLE, event); 751 break; 752 case EVNT_TIMER_HOLDTIME: 753 case EVNT_TIMER_SENDHOLD: 754 session_notification(peer, ERR_HOLDTIMEREXPIRED, 755 0, NULL, 0); 756 change_state(peer, STATE_IDLE, event); 757 break; 758 case EVNT_TIMER_KEEPALIVE: 759 session_keepalive(peer); 760 break; 761 case EVNT_RCVD_KEEPALIVE: 762 start_timer_holdtime(peer); 763 change_state(peer, STATE_ESTABLISHED, event); 764 break; 765 case EVNT_RCVD_NOTIFICATION: 766 parse_notification(peer); 767 change_state(peer, STATE_IDLE, event); 768 break; 769 default: 770 session_notification(peer, 771 ERR_FSM, ERR_FSM_UNEX_OPENCONFIRM, NULL, 0); 772 change_state(peer, STATE_IDLE, event); 773 break; 774 } 775 break; 776 case STATE_ESTABLISHED: 777 switch (event) { 778 case EVNT_START: 779 /* ignore */ 780 break; 781 case EVNT_STOP: 782 change_state(peer, STATE_IDLE, event); 783 break; 784 case EVNT_CON_CLOSED: 785 case EVNT_CON_FATAL: 786 change_state(peer, STATE_IDLE, event); 787 break; 788 case EVNT_TIMER_HOLDTIME: 789 case EVNT_TIMER_SENDHOLD: 790 session_notification(peer, ERR_HOLDTIMEREXPIRED, 791 0, NULL, 0); 792 change_state(peer, STATE_IDLE, event); 793 break; 794 case EVNT_TIMER_KEEPALIVE: 795 session_keepalive(peer); 796 break; 797 case EVNT_RCVD_KEEPALIVE: 798 start_timer_holdtime(peer); 799 break; 800 case EVNT_RCVD_UPDATE: 801 start_timer_holdtime(peer); 802 if (parse_update(peer)) 803 change_state(peer, STATE_IDLE, event); 804 else 805 start_timer_holdtime(peer); 806 break; 807 case EVNT_RCVD_NOTIFICATION: 808 parse_notification(peer); 809 change_state(peer, STATE_IDLE, event); 810 break; 811 default: 812 session_notification(peer, 813 ERR_FSM, ERR_FSM_UNEX_ESTABLISHED, NULL, 0); 814 change_state(peer, STATE_IDLE, event); 815 break; 816 } 817 break; 818 } 819 } 820 821 void 822 start_timer_holdtime(struct peer *peer) 823 { 824 if (peer->holdtime > 0) 825 timer_set(&peer->timers, Timer_Hold, peer->holdtime); 826 else 827 timer_stop(&peer->timers, Timer_Hold); 828 } 829 830 void 831 start_timer_keepalive(struct peer *peer) 832 { 833 if (peer->holdtime > 0) 834 timer_set(&peer->timers, Timer_Keepalive, peer->holdtime / 3); 835 else 836 timer_stop(&peer->timers, Timer_Keepalive); 837 } 838 839 void 840 session_close_connection(struct peer *peer) 841 { 842 if (peer->fd != -1) { 843 close(peer->fd); 844 pauseaccept = 0; 845 } 846 peer->fd = peer->wbuf.fd = -1; 847 } 848 849 void 850 change_state(struct peer *peer, enum session_state state, 851 enum session_events event) 852 { 853 struct mrt *mrt; 854 855 switch (state) { 856 case STATE_IDLE: 857 /* carp demotion first. new peers handled in init_peer */ 858 if (peer->state == STATE_ESTABLISHED && 859 peer->conf.demote_group[0] && !peer->demoted) 860 session_demote(peer, +1); 861 862 /* 863 * try to write out what's buffered (maybe a notification), 864 * don't bother if it fails 865 */ 866 if (peer->state >= STATE_OPENSENT && peer->wbuf.queued) 867 msgbuf_write(&peer->wbuf); 868 869 /* 870 * we must start the timer for the next EVNT_START 871 * if we are coming here due to an error and the 872 * session was not established successfully before, the 873 * starttimerinterval needs to be exponentially increased 874 */ 875 if (peer->IdleHoldTime == 0) 876 peer->IdleHoldTime = INTERVAL_IDLE_HOLD_INITIAL; 877 peer->holdtime = INTERVAL_HOLD_INITIAL; 878 timer_stop(&peer->timers, Timer_ConnectRetry); 879 timer_stop(&peer->timers, Timer_Keepalive); 880 timer_stop(&peer->timers, Timer_Hold); 881 timer_stop(&peer->timers, Timer_SendHold); 882 timer_stop(&peer->timers, Timer_IdleHold); 883 timer_stop(&peer->timers, Timer_IdleHoldReset); 884 session_close_connection(peer); 885 msgbuf_clear(&peer->wbuf); 886 free(peer->rbuf); 887 peer->rbuf = NULL; 888 peer->rpending = 0; 889 memset(&peer->capa.peer, 0, sizeof(peer->capa.peer)); 890 if (!peer->template) 891 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 892 peer->conf.id, 0, -1, NULL, 0); 893 894 if (event != EVNT_STOP) { 895 timer_set(&peer->timers, Timer_IdleHold, 896 peer->IdleHoldTime); 897 if (event != EVNT_NONE && 898 peer->IdleHoldTime < MAX_IDLE_HOLD/2) 899 peer->IdleHoldTime *= 2; 900 } 901 if (peer->state == STATE_ESTABLISHED) { 902 if (peer->capa.neg.grestart.restart == 2 && 903 (event == EVNT_CON_CLOSED || 904 event == EVNT_CON_FATAL)) { 905 /* don't punish graceful restart */ 906 timer_set(&peer->timers, Timer_IdleHold, 0); 907 peer->IdleHoldTime /= 2; 908 session_graceful_restart(peer); 909 } else 910 session_down(peer); 911 } 912 if (peer->state == STATE_NONE || 913 peer->state == STATE_ESTABLISHED) { 914 /* initialize capability negotiation structures */ 915 memcpy(&peer->capa.ann, &peer->conf.capabilities, 916 sizeof(peer->capa.ann)); 917 if (!peer->conf.announce_capa) 918 session_capa_ann_none(peer); 919 } 920 break; 921 case STATE_CONNECT: 922 if (peer->state == STATE_ESTABLISHED && 923 peer->capa.neg.grestart.restart == 2) { 924 /* do the graceful restart dance */ 925 session_graceful_restart(peer); 926 peer->holdtime = INTERVAL_HOLD_INITIAL; 927 timer_stop(&peer->timers, Timer_ConnectRetry); 928 timer_stop(&peer->timers, Timer_Keepalive); 929 timer_stop(&peer->timers, Timer_Hold); 930 timer_stop(&peer->timers, Timer_SendHold); 931 timer_stop(&peer->timers, Timer_IdleHold); 932 timer_stop(&peer->timers, Timer_IdleHoldReset); 933 session_close_connection(peer); 934 msgbuf_clear(&peer->wbuf); 935 memset(&peer->capa.peer, 0, sizeof(peer->capa.peer)); 936 } 937 break; 938 case STATE_ACTIVE: 939 if (!peer->template) 940 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 941 peer->conf.id, 0, -1, NULL, 0); 942 break; 943 case STATE_OPENSENT: 944 break; 945 case STATE_OPENCONFIRM: 946 break; 947 case STATE_ESTABLISHED: 948 timer_set(&peer->timers, Timer_IdleHoldReset, 949 peer->IdleHoldTime); 950 if (peer->demoted) 951 timer_set(&peer->timers, Timer_CarpUndemote, 952 INTERVAL_HOLD_DEMOTED); 953 session_up(peer); 954 break; 955 default: /* something seriously fucked */ 956 break; 957 } 958 959 log_statechange(peer, state, event); 960 LIST_FOREACH(mrt, &mrthead, entry) { 961 if (!(mrt->type == MRT_ALL_IN || mrt->type == MRT_ALL_OUT)) 962 continue; 963 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 964 mrt->peer_id == peer->conf.id || (mrt->group_id != 0 && 965 mrt->group_id == peer->conf.groupid)) 966 mrt_dump_state(mrt, peer->state, state, peer); 967 } 968 peer->prev_state = peer->state; 969 peer->state = state; 970 } 971 972 void 973 session_accept(int listenfd) 974 { 975 int connfd; 976 socklen_t len; 977 struct sockaddr_storage cliaddr; 978 struct peer *p = NULL; 979 980 len = sizeof(cliaddr); 981 if ((connfd = accept4(listenfd, 982 (struct sockaddr *)&cliaddr, &len, 983 SOCK_CLOEXEC | SOCK_NONBLOCK)) == -1) { 984 if (errno == ENFILE || errno == EMFILE) 985 pauseaccept = getmonotime(); 986 else if (errno != EWOULDBLOCK && errno != EINTR && 987 errno != ECONNABORTED) 988 log_warn("accept"); 989 return; 990 } 991 992 p = getpeerbyip(conf, (struct sockaddr *)&cliaddr); 993 994 if (p != NULL && p->state == STATE_IDLE && p->errcnt < 2) { 995 if (timer_running(&p->timers, Timer_IdleHold, NULL)) { 996 /* fast reconnect after clear */ 997 p->passive = 1; 998 bgp_fsm(p, EVNT_START); 999 } 1000 } 1001 1002 if (p != NULL && 1003 (p->state == STATE_CONNECT || p->state == STATE_ACTIVE)) { 1004 if (p->fd != -1) { 1005 if (p->state == STATE_CONNECT) 1006 session_close_connection(p); 1007 else { 1008 close(connfd); 1009 return; 1010 } 1011 } 1012 1013 open: 1014 if (p->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1015 log_peer_warnx(&p->conf, 1016 "ipsec or md5sig configured but not available"); 1017 close(connfd); 1018 return; 1019 } 1020 1021 if (tcp_md5_check(connfd, p) == -1) { 1022 close(connfd); 1023 return; 1024 } 1025 p->fd = p->wbuf.fd = connfd; 1026 if (session_setup_socket(p)) { 1027 close(connfd); 1028 return; 1029 } 1030 bgp_fsm(p, EVNT_CON_OPEN); 1031 return; 1032 } else if (p != NULL && p->state == STATE_ESTABLISHED && 1033 p->capa.neg.grestart.restart == 2) { 1034 /* first do the graceful restart dance */ 1035 change_state(p, STATE_CONNECT, EVNT_CON_CLOSED); 1036 /* then do part of the open dance */ 1037 goto open; 1038 } else { 1039 log_conn_attempt(p, (struct sockaddr *)&cliaddr, len); 1040 close(connfd); 1041 } 1042 } 1043 1044 int 1045 session_connect(struct peer *peer) 1046 { 1047 struct sockaddr *sa; 1048 struct bgpd_addr *bind_addr = NULL; 1049 socklen_t sa_len; 1050 1051 /* 1052 * we do not need the overcomplicated collision detection RFC 1771 1053 * describes; we simply make sure there is only ever one concurrent 1054 * tcp connection per peer. 1055 */ 1056 if (peer->fd != -1) 1057 return (-1); 1058 1059 if ((peer->fd = socket(aid2af(peer->conf.remote_addr.aid), 1060 SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, IPPROTO_TCP)) == -1) { 1061 log_peer_warn(&peer->conf, "session_connect socket"); 1062 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1063 return (-1); 1064 } 1065 1066 if (peer->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1067 log_peer_warnx(&peer->conf, 1068 "ipsec or md5sig configured but not available"); 1069 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1070 return (-1); 1071 } 1072 1073 tcp_md5_set(peer->fd, peer); 1074 peer->wbuf.fd = peer->fd; 1075 1076 /* if local-address is set we need to bind() */ 1077 switch (peer->conf.remote_addr.aid) { 1078 case AID_INET: 1079 bind_addr = &peer->conf.local_addr_v4; 1080 break; 1081 case AID_INET6: 1082 bind_addr = &peer->conf.local_addr_v6; 1083 break; 1084 } 1085 if ((sa = addr2sa(bind_addr, 0, &sa_len)) != NULL) { 1086 if (bind(peer->fd, sa, sa_len) == -1) { 1087 log_peer_warn(&peer->conf, "session_connect bind"); 1088 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1089 return (-1); 1090 } 1091 } 1092 1093 if (session_setup_socket(peer)) { 1094 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1095 return (-1); 1096 } 1097 1098 sa = addr2sa(&peer->conf.remote_addr, peer->conf.remote_port, &sa_len); 1099 if (connect(peer->fd, sa, sa_len) == -1) { 1100 if (errno != EINPROGRESS) { 1101 if (errno != peer->lasterr) 1102 log_peer_warn(&peer->conf, "connect"); 1103 peer->lasterr = errno; 1104 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1105 return (-1); 1106 } 1107 } else 1108 bgp_fsm(peer, EVNT_CON_OPEN); 1109 1110 return (0); 1111 } 1112 1113 int 1114 session_setup_socket(struct peer *p) 1115 { 1116 int ttl = p->conf.distance; 1117 int pre = IPTOS_PREC_INTERNETCONTROL; 1118 int nodelay = 1; 1119 int bsize; 1120 1121 switch (p->conf.remote_addr.aid) { 1122 case AID_INET: 1123 /* set precedence, see RFC 1771 appendix 5 */ 1124 if (setsockopt(p->fd, IPPROTO_IP, IP_TOS, &pre, sizeof(pre)) == 1125 -1) { 1126 log_peer_warn(&p->conf, 1127 "session_setup_socket setsockopt TOS"); 1128 return (-1); 1129 } 1130 1131 if (p->conf.ebgp) { 1132 /* 1133 * set TTL to foreign router's distance 1134 * 1=direct n=multihop with ttlsec, we always use 255 1135 */ 1136 if (p->conf.ttlsec) { 1137 ttl = 256 - p->conf.distance; 1138 if (setsockopt(p->fd, IPPROTO_IP, IP_MINTTL, 1139 &ttl, sizeof(ttl)) == -1) { 1140 log_peer_warn(&p->conf, 1141 "session_setup_socket: " 1142 "setsockopt MINTTL"); 1143 return (-1); 1144 } 1145 ttl = 255; 1146 } 1147 1148 if (setsockopt(p->fd, IPPROTO_IP, IP_TTL, &ttl, 1149 sizeof(ttl)) == -1) { 1150 log_peer_warn(&p->conf, 1151 "session_setup_socket setsockopt TTL"); 1152 return (-1); 1153 } 1154 } 1155 break; 1156 case AID_INET6: 1157 if (p->conf.ebgp) { 1158 /* 1159 * set hoplimit to foreign router's distance 1160 * 1=direct n=multihop with ttlsec, we always use 255 1161 */ 1162 if (p->conf.ttlsec) { 1163 ttl = 256 - p->conf.distance; 1164 if (setsockopt(p->fd, IPPROTO_IPV6, 1165 IPV6_MINHOPCOUNT, &ttl, sizeof(ttl)) 1166 == -1) { 1167 log_peer_warn(&p->conf, 1168 "session_setup_socket: " 1169 "setsockopt MINHOPCOUNT"); 1170 return (-1); 1171 } 1172 ttl = 255; 1173 } 1174 if (setsockopt(p->fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, 1175 &ttl, sizeof(ttl)) == -1) { 1176 log_peer_warn(&p->conf, 1177 "session_setup_socket setsockopt hoplimit"); 1178 return (-1); 1179 } 1180 } 1181 break; 1182 } 1183 1184 /* set TCP_NODELAY */ 1185 if (setsockopt(p->fd, IPPROTO_TCP, TCP_NODELAY, &nodelay, 1186 sizeof(nodelay)) == -1) { 1187 log_peer_warn(&p->conf, 1188 "session_setup_socket setsockopt TCP_NODELAY"); 1189 return (-1); 1190 } 1191 1192 /* limit bufsize. no biggie if it fails */ 1193 bsize = 65535; 1194 while (bsize > 8192 && setsockopt(p->fd, SOL_SOCKET, SO_RCVBUF, 1195 &bsize, sizeof(bsize)) == -1 && errno != EINVAL) 1196 bsize /= 2; 1197 bsize = 65535; 1198 while (bsize > 8192 && setsockopt(p->fd, SOL_SOCKET, SO_SNDBUF, 1199 &bsize, sizeof(bsize)) == -1 && errno != EINVAL) 1200 bsize /= 2; 1201 1202 return (0); 1203 } 1204 1205 /* compare two sockaddrs by converting them into bgpd_addr */ 1206 static int 1207 sa_cmp(struct sockaddr *a, struct sockaddr *b) 1208 { 1209 struct bgpd_addr ba, bb; 1210 1211 sa2addr(a, &ba, NULL); 1212 sa2addr(b, &bb, NULL); 1213 1214 return (memcmp(&ba, &bb, sizeof(ba)) == 0); 1215 } 1216 1217 static void 1218 get_alternate_addr(struct sockaddr *sa, struct bgpd_addr *alt) 1219 { 1220 struct ifaddrs *ifap, *ifa, *match; 1221 1222 if (getifaddrs(&ifap) == -1) 1223 fatal("getifaddrs"); 1224 1225 for (match = ifap; match != NULL; match = match->ifa_next) 1226 if (match->ifa_addr != NULL && 1227 sa_cmp(sa, match->ifa_addr) == 0) 1228 break; 1229 1230 if (match == NULL) { 1231 log_warnx("%s: local address not found", __func__); 1232 return; 1233 } 1234 1235 switch (sa->sa_family) { 1236 case AF_INET6: 1237 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1238 if (ifa->ifa_addr != NULL && 1239 ifa->ifa_addr->sa_family == AF_INET && 1240 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1241 sa2addr(ifa->ifa_addr, alt, NULL); 1242 break; 1243 } 1244 } 1245 break; 1246 case AF_INET: 1247 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1248 if (ifa->ifa_addr != NULL && 1249 ifa->ifa_addr->sa_family == AF_INET6 && 1250 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1251 struct sockaddr_in6 *s = 1252 (struct sockaddr_in6 *)ifa->ifa_addr; 1253 1254 /* only accept global scope addresses */ 1255 if (IN6_IS_ADDR_LINKLOCAL(&s->sin6_addr) || 1256 IN6_IS_ADDR_SITELOCAL(&s->sin6_addr)) 1257 continue; 1258 sa2addr(ifa->ifa_addr, alt, NULL); 1259 break; 1260 } 1261 } 1262 break; 1263 default: 1264 log_warnx("%s: unsupported address family %d", __func__, 1265 sa->sa_family); 1266 break; 1267 } 1268 1269 freeifaddrs(ifap); 1270 } 1271 1272 void 1273 session_tcp_established(struct peer *peer) 1274 { 1275 struct sockaddr_storage ss; 1276 socklen_t len; 1277 1278 len = sizeof(ss); 1279 if (getsockname(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1280 log_warn("getsockname"); 1281 sa2addr((struct sockaddr *)&ss, &peer->local, &peer->local_port); 1282 get_alternate_addr((struct sockaddr *)&ss, &peer->local_alt); 1283 len = sizeof(ss); 1284 if (getpeername(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1285 log_warn("getpeername"); 1286 sa2addr((struct sockaddr *)&ss, &peer->remote, &peer->remote_port); 1287 } 1288 1289 void 1290 session_capa_ann_none(struct peer *peer) 1291 { 1292 memset(&peer->capa.ann, 0, sizeof(peer->capa.ann)); 1293 } 1294 1295 int 1296 session_capa_add(struct ibuf *opb, uint8_t capa_code, uint8_t capa_len) 1297 { 1298 int errs = 0; 1299 1300 errs += ibuf_add(opb, &capa_code, sizeof(capa_code)); 1301 errs += ibuf_add(opb, &capa_len, sizeof(capa_len)); 1302 return (errs); 1303 } 1304 1305 int 1306 session_capa_add_mp(struct ibuf *buf, uint8_t aid) 1307 { 1308 uint8_t safi, pad = 0; 1309 uint16_t afi; 1310 int errs = 0; 1311 1312 if (aid2afi(aid, &afi, &safi) == -1) 1313 fatalx("session_capa_add_mp: bad afi/safi pair"); 1314 afi = htons(afi); 1315 errs += ibuf_add(buf, &afi, sizeof(afi)); 1316 errs += ibuf_add(buf, &pad, sizeof(pad)); 1317 errs += ibuf_add(buf, &safi, sizeof(safi)); 1318 1319 return (errs); 1320 } 1321 1322 int 1323 session_capa_add_afi(struct peer *p, struct ibuf *b, uint8_t aid, 1324 uint8_t flags) 1325 { 1326 u_int errs = 0; 1327 uint16_t afi; 1328 uint8_t safi; 1329 1330 if (aid2afi(aid, &afi, &safi)) { 1331 log_warn("session_capa_add_afi: bad AID"); 1332 return (1); 1333 } 1334 1335 afi = htons(afi); 1336 errs += ibuf_add(b, &afi, sizeof(afi)); 1337 errs += ibuf_add(b, &safi, sizeof(safi)); 1338 errs += ibuf_add(b, &flags, sizeof(flags)); 1339 1340 return (errs); 1341 } 1342 1343 struct bgp_msg * 1344 session_newmsg(enum msg_type msgtype, uint16_t len) 1345 { 1346 struct bgp_msg *msg; 1347 struct msg_header hdr; 1348 struct ibuf *buf; 1349 int errs = 0; 1350 1351 memset(&hdr.marker, 0xff, sizeof(hdr.marker)); 1352 hdr.len = htons(len); 1353 hdr.type = msgtype; 1354 1355 if ((buf = ibuf_open(len)) == NULL) 1356 return (NULL); 1357 1358 errs += ibuf_add(buf, &hdr.marker, sizeof(hdr.marker)); 1359 errs += ibuf_add(buf, &hdr.len, sizeof(hdr.len)); 1360 errs += ibuf_add(buf, &hdr.type, sizeof(hdr.type)); 1361 1362 if (errs || (msg = calloc(1, sizeof(*msg))) == NULL) { 1363 ibuf_free(buf); 1364 return (NULL); 1365 } 1366 1367 msg->buf = buf; 1368 msg->type = msgtype; 1369 msg->len = len; 1370 1371 return (msg); 1372 } 1373 1374 int 1375 session_sendmsg(struct bgp_msg *msg, struct peer *p) 1376 { 1377 struct mrt *mrt; 1378 1379 LIST_FOREACH(mrt, &mrthead, entry) { 1380 if (!(mrt->type == MRT_ALL_OUT || (msg->type == UPDATE && 1381 mrt->type == MRT_UPDATE_OUT))) 1382 continue; 1383 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1384 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1385 mrt->group_id == p->conf.groupid)) 1386 mrt_dump_bgp_msg(mrt, msg->buf->buf, msg->len, p, 1387 msg->type); 1388 } 1389 1390 ibuf_close(&p->wbuf, msg->buf); 1391 if (!p->throttled && p->wbuf.queued > SESS_MSG_HIGH_MARK) { 1392 if (imsg_rde(IMSG_XOFF, p->conf.id, NULL, 0) == -1) 1393 log_peer_warn(&p->conf, "imsg_compose XOFF"); 1394 else 1395 p->throttled = 1; 1396 } 1397 1398 free(msg); 1399 return (0); 1400 } 1401 1402 /* 1403 * Translate between internal roles and the value expected by RFC 9234. 1404 */ 1405 static uint8_t 1406 role2capa(enum role role) 1407 { 1408 switch (role) { 1409 case ROLE_CUSTOMER: 1410 return CAPA_ROLE_CUSTOMER; 1411 case ROLE_PROVIDER: 1412 return CAPA_ROLE_PROVIDER; 1413 case ROLE_RS: 1414 return CAPA_ROLE_RS; 1415 case ROLE_RS_CLIENT: 1416 return CAPA_ROLE_RS_CLIENT; 1417 case ROLE_PEER: 1418 return CAPA_ROLE_PEER; 1419 default: 1420 fatalx("Unsupported role for role capability"); 1421 } 1422 } 1423 1424 static enum role 1425 capa2role(uint8_t val) 1426 { 1427 switch (val) { 1428 case CAPA_ROLE_PROVIDER: 1429 return ROLE_PROVIDER; 1430 case CAPA_ROLE_RS: 1431 return ROLE_RS; 1432 case CAPA_ROLE_RS_CLIENT: 1433 return ROLE_RS_CLIENT; 1434 case CAPA_ROLE_CUSTOMER: 1435 return ROLE_CUSTOMER; 1436 case CAPA_ROLE_PEER: 1437 return ROLE_PEER; 1438 default: 1439 return ROLE_NONE; 1440 } 1441 } 1442 1443 void 1444 session_open(struct peer *p) 1445 { 1446 struct bgp_msg *buf; 1447 struct ibuf *opb; 1448 struct msg_open msg; 1449 uint16_t len, optparamlen = 0; 1450 uint8_t i, op_type; 1451 int errs = 0, extlen = 0; 1452 int mpcapa = 0; 1453 1454 1455 if ((opb = ibuf_dynamic(0, UINT16_MAX - 3)) == NULL) { 1456 bgp_fsm(p, EVNT_CON_FATAL); 1457 return; 1458 } 1459 1460 /* multiprotocol extensions, RFC 4760 */ 1461 for (i = 0; i < AID_MAX; i++) 1462 if (p->capa.ann.mp[i]) { /* 4 bytes data */ 1463 errs += session_capa_add(opb, CAPA_MP, 4); 1464 errs += session_capa_add_mp(opb, i); 1465 mpcapa++; 1466 } 1467 1468 /* route refresh, RFC 2918 */ 1469 if (p->capa.ann.refresh) /* no data */ 1470 errs += session_capa_add(opb, CAPA_REFRESH, 0); 1471 1472 /* BGP open policy, RFC 9234, only for ebgp sessions */ 1473 if (p->conf.ebgp && p->capa.ann.policy && 1474 p->conf.role != ROLE_NONE && 1475 (p->capa.ann.mp[AID_INET] || p->capa.ann.mp[AID_INET6] || 1476 mpcapa == 0)) { 1477 uint8_t val; 1478 val = role2capa(p->conf.role); 1479 errs += session_capa_add(opb, CAPA_ROLE, 1); 1480 errs += ibuf_add(opb, &val, 1); 1481 } 1482 1483 /* graceful restart and End-of-RIB marker, RFC 4724 */ 1484 if (p->capa.ann.grestart.restart) { 1485 int rst = 0; 1486 uint16_t hdr = 0; 1487 1488 for (i = 0; i < AID_MAX; i++) { 1489 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) 1490 rst++; 1491 } 1492 1493 /* Only set the R-flag if no graceful restart is ongoing */ 1494 if (!rst) 1495 hdr |= CAPA_GR_R_FLAG; 1496 hdr = htons(hdr); 1497 1498 errs += session_capa_add(opb, CAPA_RESTART, sizeof(hdr)); 1499 errs += ibuf_add(opb, &hdr, sizeof(hdr)); 1500 } 1501 1502 /* 4-bytes AS numbers, RFC6793 */ 1503 if (p->capa.ann.as4byte) { /* 4 bytes data */ 1504 uint32_t nas; 1505 1506 nas = htonl(p->conf.local_as); 1507 errs += session_capa_add(opb, CAPA_AS4BYTE, sizeof(nas)); 1508 errs += ibuf_add(opb, &nas, sizeof(nas)); 1509 } 1510 1511 /* advertisement of multiple paths, RFC7911 */ 1512 if (p->capa.ann.add_path[0]) { /* variable */ 1513 uint8_t aplen; 1514 1515 if (mpcapa) 1516 aplen = 4 * mpcapa; 1517 else /* AID_INET */ 1518 aplen = 4; 1519 errs += session_capa_add(opb, CAPA_ADD_PATH, aplen); 1520 if (mpcapa) { 1521 for (i = AID_MIN; i < AID_MAX; i++) { 1522 if (p->capa.ann.mp[i]) { 1523 errs += session_capa_add_afi(p, opb, 1524 i, p->capa.ann.add_path[i]); 1525 } 1526 } 1527 } else { /* AID_INET */ 1528 errs += session_capa_add_afi(p, opb, AID_INET, 1529 p->capa.ann.add_path[AID_INET]); 1530 } 1531 } 1532 1533 /* enhanced route-refresh, RFC7313 */ 1534 if (p->capa.ann.enhanced_rr) /* no data */ 1535 errs += session_capa_add(opb, CAPA_ENHANCED_RR, 0); 1536 1537 optparamlen = ibuf_size(opb); 1538 if (optparamlen == 0) { 1539 /* nothing */ 1540 } else if (optparamlen + 2 >= 255) { 1541 /* RFC9072: 2 byte length instead of 1 + 3 byte extra header */ 1542 optparamlen += sizeof(op_type) + 2 + 3; 1543 msg.optparamlen = 255; 1544 extlen = 1; 1545 } else { 1546 optparamlen += sizeof(op_type) + 1; 1547 msg.optparamlen = optparamlen; 1548 } 1549 1550 len = MSGSIZE_OPEN_MIN + optparamlen; 1551 if (errs || (buf = session_newmsg(OPEN, len)) == NULL) { 1552 ibuf_free(opb); 1553 bgp_fsm(p, EVNT_CON_FATAL); 1554 return; 1555 } 1556 1557 msg.version = 4; 1558 msg.myas = htons(p->conf.local_short_as); 1559 if (p->conf.holdtime) 1560 msg.holdtime = htons(p->conf.holdtime); 1561 else 1562 msg.holdtime = htons(conf->holdtime); 1563 msg.bgpid = conf->bgpid; /* is already in network byte order */ 1564 1565 errs += ibuf_add(buf->buf, &msg.version, sizeof(msg.version)); 1566 errs += ibuf_add(buf->buf, &msg.myas, sizeof(msg.myas)); 1567 errs += ibuf_add(buf->buf, &msg.holdtime, sizeof(msg.holdtime)); 1568 errs += ibuf_add(buf->buf, &msg.bgpid, sizeof(msg.bgpid)); 1569 errs += ibuf_add(buf->buf, &msg.optparamlen, 1); 1570 1571 if (extlen) { 1572 /* write RFC9072 extra header */ 1573 uint16_t op_extlen = htons(optparamlen - 3); 1574 op_type = OPT_PARAM_EXT_LEN; 1575 errs += ibuf_add(buf->buf, &op_type, 1); 1576 errs += ibuf_add(buf->buf, &op_extlen, 2); 1577 } 1578 1579 if (optparamlen) { 1580 op_type = OPT_PARAM_CAPABILITIES; 1581 errs += ibuf_add(buf->buf, &op_type, sizeof(op_type)); 1582 1583 optparamlen = ibuf_size(opb); 1584 if (extlen) { 1585 /* RFC9072: 2-byte extended length */ 1586 uint16_t op_extlen = htons(optparamlen); 1587 errs += ibuf_add(buf->buf, &op_extlen, 2); 1588 } else { 1589 uint8_t op_len = optparamlen; 1590 errs += ibuf_add(buf->buf, &op_len, 1); 1591 } 1592 errs += ibuf_add(buf->buf, opb->buf, ibuf_size(opb)); 1593 } 1594 1595 ibuf_free(opb); 1596 1597 if (errs) { 1598 ibuf_free(buf->buf); 1599 free(buf); 1600 bgp_fsm(p, EVNT_CON_FATAL); 1601 return; 1602 } 1603 1604 if (session_sendmsg(buf, p) == -1) { 1605 bgp_fsm(p, EVNT_CON_FATAL); 1606 return; 1607 } 1608 1609 p->stats.msg_sent_open++; 1610 } 1611 1612 void 1613 session_keepalive(struct peer *p) 1614 { 1615 struct bgp_msg *buf; 1616 1617 if ((buf = session_newmsg(KEEPALIVE, MSGSIZE_KEEPALIVE)) == NULL || 1618 session_sendmsg(buf, p) == -1) { 1619 bgp_fsm(p, EVNT_CON_FATAL); 1620 return; 1621 } 1622 1623 start_timer_keepalive(p); 1624 p->stats.msg_sent_keepalive++; 1625 } 1626 1627 void 1628 session_update(uint32_t peerid, void *data, size_t datalen) 1629 { 1630 struct peer *p; 1631 struct bgp_msg *buf; 1632 1633 if ((p = getpeerbyid(conf, peerid)) == NULL) { 1634 log_warnx("no such peer: id=%u", peerid); 1635 return; 1636 } 1637 1638 if (p->state != STATE_ESTABLISHED) 1639 return; 1640 1641 if ((buf = session_newmsg(UPDATE, MSGSIZE_HEADER + datalen)) == NULL) { 1642 bgp_fsm(p, EVNT_CON_FATAL); 1643 return; 1644 } 1645 1646 if (ibuf_add(buf->buf, data, datalen)) { 1647 ibuf_free(buf->buf); 1648 free(buf); 1649 bgp_fsm(p, EVNT_CON_FATAL); 1650 return; 1651 } 1652 1653 if (session_sendmsg(buf, p) == -1) { 1654 bgp_fsm(p, EVNT_CON_FATAL); 1655 return; 1656 } 1657 1658 start_timer_keepalive(p); 1659 p->stats.msg_sent_update++; 1660 } 1661 1662 void 1663 session_notification(struct peer *p, uint8_t errcode, uint8_t subcode, 1664 void *data, ssize_t datalen) 1665 { 1666 struct bgp_msg *buf; 1667 int errs = 0; 1668 1669 if (p->stats.last_sent_errcode) /* some notification already sent */ 1670 return; 1671 1672 log_notification(p, errcode, subcode, data, datalen, "sending"); 1673 1674 /* cap to maximum size */ 1675 if (datalen > MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN) { 1676 log_peer_warnx(&p->conf, 1677 "oversized notification, data trunkated"); 1678 datalen = MAX_PKTSIZE - MSGSIZE_NOTIFICATION_MIN; 1679 } 1680 1681 if ((buf = session_newmsg(NOTIFICATION, 1682 MSGSIZE_NOTIFICATION_MIN + datalen)) == NULL) { 1683 bgp_fsm(p, EVNT_CON_FATAL); 1684 return; 1685 } 1686 1687 errs += ibuf_add(buf->buf, &errcode, sizeof(errcode)); 1688 errs += ibuf_add(buf->buf, &subcode, sizeof(subcode)); 1689 1690 if (datalen > 0) 1691 errs += ibuf_add(buf->buf, data, datalen); 1692 1693 if (errs) { 1694 ibuf_free(buf->buf); 1695 free(buf); 1696 bgp_fsm(p, EVNT_CON_FATAL); 1697 return; 1698 } 1699 1700 if (session_sendmsg(buf, p) == -1) { 1701 bgp_fsm(p, EVNT_CON_FATAL); 1702 return; 1703 } 1704 1705 p->stats.msg_sent_notification++; 1706 p->stats.last_sent_errcode = errcode; 1707 p->stats.last_sent_suberr = subcode; 1708 } 1709 1710 int 1711 session_neighbor_rrefresh(struct peer *p) 1712 { 1713 uint8_t i; 1714 1715 if (!(p->capa.neg.refresh || p->capa.neg.enhanced_rr)) 1716 return (-1); 1717 1718 for (i = 0; i < AID_MAX; i++) { 1719 if (p->capa.neg.mp[i] != 0) 1720 session_rrefresh(p, i, ROUTE_REFRESH_REQUEST); 1721 } 1722 1723 return (0); 1724 } 1725 1726 void 1727 session_rrefresh(struct peer *p, uint8_t aid, uint8_t subtype) 1728 { 1729 struct bgp_msg *buf; 1730 int errs = 0; 1731 uint16_t afi; 1732 uint8_t safi; 1733 1734 switch (subtype) { 1735 case ROUTE_REFRESH_REQUEST: 1736 p->stats.refresh_sent_req++; 1737 break; 1738 case ROUTE_REFRESH_BEGIN_RR: 1739 case ROUTE_REFRESH_END_RR: 1740 /* requires enhanced route refresh */ 1741 if (!p->capa.neg.enhanced_rr) 1742 return; 1743 if (subtype == ROUTE_REFRESH_BEGIN_RR) 1744 p->stats.refresh_sent_borr++; 1745 else 1746 p->stats.refresh_sent_eorr++; 1747 break; 1748 default: 1749 fatalx("session_rrefresh: bad subtype %d", subtype); 1750 } 1751 1752 if (aid2afi(aid, &afi, &safi) == -1) 1753 fatalx("session_rrefresh: bad afi/safi pair"); 1754 1755 if ((buf = session_newmsg(RREFRESH, MSGSIZE_RREFRESH)) == NULL) { 1756 bgp_fsm(p, EVNT_CON_FATAL); 1757 return; 1758 } 1759 1760 afi = htons(afi); 1761 errs += ibuf_add(buf->buf, &afi, sizeof(afi)); 1762 errs += ibuf_add(buf->buf, &subtype, sizeof(subtype)); 1763 errs += ibuf_add(buf->buf, &safi, sizeof(safi)); 1764 1765 if (errs) { 1766 ibuf_free(buf->buf); 1767 free(buf); 1768 bgp_fsm(p, EVNT_CON_FATAL); 1769 return; 1770 } 1771 1772 if (session_sendmsg(buf, p) == -1) { 1773 bgp_fsm(p, EVNT_CON_FATAL); 1774 return; 1775 } 1776 1777 p->stats.msg_sent_rrefresh++; 1778 } 1779 1780 int 1781 session_graceful_restart(struct peer *p) 1782 { 1783 uint8_t i; 1784 1785 timer_set(&p->timers, Timer_RestartTimeout, 1786 p->capa.neg.grestart.timeout); 1787 1788 for (i = 0; i < AID_MAX; i++) { 1789 if (p->capa.neg.grestart.flags[i] & CAPA_GR_PRESENT) { 1790 if (imsg_rde(IMSG_SESSION_STALE, p->conf.id, 1791 &i, sizeof(i)) == -1) 1792 return (-1); 1793 log_peer_warnx(&p->conf, 1794 "graceful restart of %s, keeping routes", 1795 aid2str(i)); 1796 p->capa.neg.grestart.flags[i] |= CAPA_GR_RESTARTING; 1797 } else if (p->capa.neg.mp[i]) { 1798 if (imsg_rde(IMSG_SESSION_NOGRACE, p->conf.id, 1799 &i, sizeof(i)) == -1) 1800 return (-1); 1801 log_peer_warnx(&p->conf, 1802 "graceful restart of %s, flushing routes", 1803 aid2str(i)); 1804 } 1805 } 1806 return (0); 1807 } 1808 1809 int 1810 session_graceful_stop(struct peer *p) 1811 { 1812 uint8_t i; 1813 1814 for (i = 0; i < AID_MAX; i++) { 1815 /* 1816 * Only flush if the peer is restarting and the timeout fired. 1817 * In all other cases the session was already flushed when the 1818 * session went down or when the new open message was parsed. 1819 */ 1820 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) { 1821 log_peer_warnx(&p->conf, "graceful restart of %s, " 1822 "time-out, flushing", aid2str(i)); 1823 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1824 &i, sizeof(i)) == -1) 1825 return (-1); 1826 } 1827 p->capa.neg.grestart.flags[i] &= ~CAPA_GR_RESTARTING; 1828 } 1829 return (0); 1830 } 1831 1832 int 1833 session_dispatch_msg(struct pollfd *pfd, struct peer *p) 1834 { 1835 ssize_t n; 1836 socklen_t len; 1837 int error; 1838 1839 if (p->state == STATE_CONNECT) { 1840 if (pfd->revents & POLLOUT) { 1841 if (pfd->revents & POLLIN) { 1842 /* error occurred */ 1843 len = sizeof(error); 1844 if (getsockopt(pfd->fd, SOL_SOCKET, SO_ERROR, 1845 &error, &len) == -1 || error) { 1846 if (error) 1847 errno = error; 1848 if (errno != p->lasterr) { 1849 log_peer_warn(&p->conf, 1850 "socket error"); 1851 p->lasterr = errno; 1852 } 1853 bgp_fsm(p, EVNT_CON_OPENFAIL); 1854 return (1); 1855 } 1856 } 1857 bgp_fsm(p, EVNT_CON_OPEN); 1858 return (1); 1859 } 1860 if (pfd->revents & POLLHUP) { 1861 bgp_fsm(p, EVNT_CON_OPENFAIL); 1862 return (1); 1863 } 1864 if (pfd->revents & (POLLERR|POLLNVAL)) { 1865 bgp_fsm(p, EVNT_CON_FATAL); 1866 return (1); 1867 } 1868 return (0); 1869 } 1870 1871 if (pfd->revents & POLLHUP) { 1872 bgp_fsm(p, EVNT_CON_CLOSED); 1873 return (1); 1874 } 1875 if (pfd->revents & (POLLERR|POLLNVAL)) { 1876 bgp_fsm(p, EVNT_CON_FATAL); 1877 return (1); 1878 } 1879 1880 if (pfd->revents & POLLOUT && p->wbuf.queued) { 1881 if ((error = msgbuf_write(&p->wbuf)) <= 0 && errno != EAGAIN) { 1882 if (error == 0) 1883 log_peer_warnx(&p->conf, "Connection closed"); 1884 else if (error == -1) 1885 log_peer_warn(&p->conf, "write error"); 1886 bgp_fsm(p, EVNT_CON_FATAL); 1887 return (1); 1888 } 1889 p->stats.last_write = getmonotime(); 1890 if (p->holdtime > 0) 1891 timer_set(&p->timers, Timer_SendHold, 1892 p->holdtime < INTERVAL_HOLD ? INTERVAL_HOLD : 1893 p->holdtime); 1894 if (p->throttled && p->wbuf.queued < SESS_MSG_LOW_MARK) { 1895 if (imsg_rde(IMSG_XON, p->conf.id, NULL, 0) == -1) 1896 log_peer_warn(&p->conf, "imsg_compose XON"); 1897 else 1898 p->throttled = 0; 1899 } 1900 if (!(pfd->revents & POLLIN)) 1901 return (1); 1902 } 1903 1904 if (p->rbuf && pfd->revents & POLLIN) { 1905 if ((n = read(p->fd, p->rbuf->buf + p->rbuf->wpos, 1906 sizeof(p->rbuf->buf) - p->rbuf->wpos)) == -1) { 1907 if (errno != EINTR && errno != EAGAIN) { 1908 log_peer_warn(&p->conf, "read error"); 1909 bgp_fsm(p, EVNT_CON_FATAL); 1910 } 1911 return (1); 1912 } 1913 if (n == 0) { /* connection closed */ 1914 bgp_fsm(p, EVNT_CON_CLOSED); 1915 return (1); 1916 } 1917 1918 p->rbuf->wpos += n; 1919 p->stats.last_read = getmonotime(); 1920 return (1); 1921 } 1922 return (0); 1923 } 1924 1925 void 1926 session_process_msg(struct peer *p) 1927 { 1928 struct mrt *mrt; 1929 ssize_t rpos, av, left; 1930 int processed = 0; 1931 uint16_t msglen; 1932 uint8_t msgtype; 1933 1934 rpos = 0; 1935 av = p->rbuf->wpos; 1936 p->rpending = 0; 1937 1938 /* 1939 * session might drop to IDLE -> buffers deallocated 1940 * we MUST check rbuf != NULL before use 1941 */ 1942 for (;;) { 1943 if (p->rbuf == NULL) 1944 return; 1945 if (rpos + MSGSIZE_HEADER > av) 1946 break; 1947 if (parse_header(p, p->rbuf->buf + rpos, &msglen, 1948 &msgtype) == -1) 1949 return; 1950 if (rpos + msglen > av) 1951 break; 1952 p->rbuf->rptr = p->rbuf->buf + rpos; 1953 1954 /* dump to MRT as soon as we have a full packet */ 1955 LIST_FOREACH(mrt, &mrthead, entry) { 1956 if (!(mrt->type == MRT_ALL_IN || (msgtype == UPDATE && 1957 mrt->type == MRT_UPDATE_IN))) 1958 continue; 1959 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1960 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1961 mrt->group_id == p->conf.groupid)) 1962 mrt_dump_bgp_msg(mrt, p->rbuf->rptr, msglen, p, 1963 msgtype); 1964 } 1965 1966 switch (msgtype) { 1967 case OPEN: 1968 bgp_fsm(p, EVNT_RCVD_OPEN); 1969 p->stats.msg_rcvd_open++; 1970 break; 1971 case UPDATE: 1972 bgp_fsm(p, EVNT_RCVD_UPDATE); 1973 p->stats.msg_rcvd_update++; 1974 break; 1975 case NOTIFICATION: 1976 bgp_fsm(p, EVNT_RCVD_NOTIFICATION); 1977 p->stats.msg_rcvd_notification++; 1978 break; 1979 case KEEPALIVE: 1980 bgp_fsm(p, EVNT_RCVD_KEEPALIVE); 1981 p->stats.msg_rcvd_keepalive++; 1982 break; 1983 case RREFRESH: 1984 parse_rrefresh(p); 1985 p->stats.msg_rcvd_rrefresh++; 1986 break; 1987 default: /* cannot happen */ 1988 session_notification(p, ERR_HEADER, ERR_HDR_TYPE, 1989 &msgtype, 1); 1990 log_warnx("received message with unknown type %u", 1991 msgtype); 1992 bgp_fsm(p, EVNT_CON_FATAL); 1993 } 1994 rpos += msglen; 1995 if (++processed > MSG_PROCESS_LIMIT) { 1996 p->rpending = 1; 1997 break; 1998 } 1999 } 2000 2001 if (p->rbuf == NULL) 2002 return; 2003 if (rpos < av) { 2004 left = av - rpos; 2005 memmove(&p->rbuf->buf, p->rbuf->buf + rpos, left); 2006 p->rbuf->wpos = left; 2007 } else 2008 p->rbuf->wpos = 0; 2009 } 2010 2011 int 2012 parse_header(struct peer *peer, u_char *data, uint16_t *len, uint8_t *type) 2013 { 2014 u_char *p; 2015 uint16_t olen; 2016 static const uint8_t marker[MSGSIZE_HEADER_MARKER] = { 0xff, 0xff, 2017 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 2018 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; 2019 2020 /* caller MUST make sure we are getting 19 bytes! */ 2021 p = data; 2022 if (memcmp(p, marker, sizeof(marker))) { 2023 log_peer_warnx(&peer->conf, "sync error"); 2024 session_notification(peer, ERR_HEADER, ERR_HDR_SYNC, NULL, 0); 2025 bgp_fsm(peer, EVNT_CON_FATAL); 2026 return (-1); 2027 } 2028 p += MSGSIZE_HEADER_MARKER; 2029 2030 memcpy(&olen, p, 2); 2031 *len = ntohs(olen); 2032 p += 2; 2033 memcpy(type, p, 1); 2034 2035 if (*len < MSGSIZE_HEADER || *len > MAX_PKTSIZE) { 2036 log_peer_warnx(&peer->conf, 2037 "received message: illegal length: %u byte", *len); 2038 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2039 &olen, sizeof(olen)); 2040 bgp_fsm(peer, EVNT_CON_FATAL); 2041 return (-1); 2042 } 2043 2044 switch (*type) { 2045 case OPEN: 2046 if (*len < MSGSIZE_OPEN_MIN) { 2047 log_peer_warnx(&peer->conf, 2048 "received OPEN: illegal len: %u byte", *len); 2049 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2050 &olen, sizeof(olen)); 2051 bgp_fsm(peer, EVNT_CON_FATAL); 2052 return (-1); 2053 } 2054 break; 2055 case NOTIFICATION: 2056 if (*len < MSGSIZE_NOTIFICATION_MIN) { 2057 log_peer_warnx(&peer->conf, 2058 "received NOTIFICATION: illegal len: %u byte", 2059 *len); 2060 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2061 &olen, sizeof(olen)); 2062 bgp_fsm(peer, EVNT_CON_FATAL); 2063 return (-1); 2064 } 2065 break; 2066 case UPDATE: 2067 if (*len < MSGSIZE_UPDATE_MIN) { 2068 log_peer_warnx(&peer->conf, 2069 "received UPDATE: illegal len: %u byte", *len); 2070 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2071 &olen, sizeof(olen)); 2072 bgp_fsm(peer, EVNT_CON_FATAL); 2073 return (-1); 2074 } 2075 break; 2076 case KEEPALIVE: 2077 if (*len != MSGSIZE_KEEPALIVE) { 2078 log_peer_warnx(&peer->conf, 2079 "received KEEPALIVE: illegal len: %u byte", *len); 2080 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2081 &olen, sizeof(olen)); 2082 bgp_fsm(peer, EVNT_CON_FATAL); 2083 return (-1); 2084 } 2085 break; 2086 case RREFRESH: 2087 if (*len < MSGSIZE_RREFRESH_MIN) { 2088 log_peer_warnx(&peer->conf, 2089 "received RREFRESH: illegal len: %u byte", *len); 2090 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 2091 &olen, sizeof(olen)); 2092 bgp_fsm(peer, EVNT_CON_FATAL); 2093 return (-1); 2094 } 2095 break; 2096 default: 2097 log_peer_warnx(&peer->conf, 2098 "received msg with unknown type %u", *type); 2099 session_notification(peer, ERR_HEADER, ERR_HDR_TYPE, 2100 type, 1); 2101 bgp_fsm(peer, EVNT_CON_FATAL); 2102 return (-1); 2103 } 2104 return (0); 2105 } 2106 2107 int 2108 parse_open(struct peer *peer) 2109 { 2110 u_char *p, *op_val; 2111 uint8_t version, rversion; 2112 uint16_t short_as, msglen; 2113 uint16_t holdtime, oholdtime, myholdtime; 2114 uint32_t as, bgpid; 2115 uint16_t optparamlen, extlen, plen, op_len; 2116 uint8_t op_type, suberr = 0; 2117 2118 p = peer->rbuf->rptr; 2119 p += MSGSIZE_HEADER_MARKER; 2120 memcpy(&msglen, p, sizeof(msglen)); 2121 msglen = ntohs(msglen); 2122 2123 p = peer->rbuf->rptr; 2124 p += MSGSIZE_HEADER; /* header is already checked */ 2125 2126 memcpy(&version, p, sizeof(version)); 2127 p += sizeof(version); 2128 2129 if (version != BGP_VERSION) { 2130 log_peer_warnx(&peer->conf, 2131 "peer wants unrecognized version %u", version); 2132 if (version > BGP_VERSION) 2133 rversion = version - BGP_VERSION; 2134 else 2135 rversion = BGP_VERSION; 2136 session_notification(peer, ERR_OPEN, ERR_OPEN_VERSION, 2137 &rversion, sizeof(rversion)); 2138 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2139 return (-1); 2140 } 2141 2142 memcpy(&short_as, p, sizeof(short_as)); 2143 p += sizeof(short_as); 2144 as = peer->short_as = ntohs(short_as); 2145 if (as == 0) { 2146 log_peer_warnx(&peer->conf, 2147 "peer requests unacceptable AS %u", as); 2148 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, 2149 NULL, 0); 2150 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2151 return (-1); 2152 } 2153 2154 memcpy(&oholdtime, p, sizeof(oholdtime)); 2155 p += sizeof(oholdtime); 2156 2157 holdtime = ntohs(oholdtime); 2158 if (holdtime && holdtime < peer->conf.min_holdtime) { 2159 log_peer_warnx(&peer->conf, 2160 "peer requests unacceptable holdtime %u", holdtime); 2161 session_notification(peer, ERR_OPEN, ERR_OPEN_HOLDTIME, 2162 NULL, 0); 2163 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2164 return (-1); 2165 } 2166 2167 myholdtime = peer->conf.holdtime; 2168 if (!myholdtime) 2169 myholdtime = conf->holdtime; 2170 if (holdtime < myholdtime) 2171 peer->holdtime = holdtime; 2172 else 2173 peer->holdtime = myholdtime; 2174 2175 memcpy(&bgpid, p, sizeof(bgpid)); 2176 p += sizeof(bgpid); 2177 2178 /* check bgpid for validity - just disallow 0 */ 2179 if (ntohl(bgpid) == 0) { 2180 log_peer_warnx(&peer->conf, "peer BGPID %u unacceptable", 2181 ntohl(bgpid)); 2182 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2183 NULL, 0); 2184 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2185 return (-1); 2186 } 2187 peer->remote_bgpid = bgpid; 2188 2189 extlen = 0; 2190 optparamlen = *p++; 2191 2192 if (optparamlen == 0) { 2193 if (msglen != MSGSIZE_OPEN_MIN) { 2194 bad_len: 2195 log_peer_warnx(&peer->conf, 2196 "corrupt OPEN message received: length mismatch"); 2197 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2198 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2199 return (-1); 2200 } 2201 } else { 2202 if (msglen < MSGSIZE_OPEN_MIN + 1) 2203 goto bad_len; 2204 2205 op_type = *p; 2206 if (op_type == OPT_PARAM_EXT_LEN) { 2207 p++; 2208 memcpy(&optparamlen, p, sizeof(optparamlen)); 2209 optparamlen = ntohs(optparamlen); 2210 p += sizeof(optparamlen); 2211 extlen = 1; 2212 } 2213 2214 /* RFC9020 encoding has 3 extra bytes */ 2215 if (optparamlen + 3 * extlen != msglen - MSGSIZE_OPEN_MIN) 2216 goto bad_len; 2217 } 2218 2219 plen = optparamlen; 2220 while (plen > 0) { 2221 if (plen < 2 + extlen) 2222 goto bad_len; 2223 2224 memcpy(&op_type, p, sizeof(op_type)); 2225 p += sizeof(op_type); 2226 plen -= sizeof(op_type); 2227 if (!extlen) { 2228 op_len = *p++; 2229 plen--; 2230 } else { 2231 memcpy(&op_len, p, sizeof(op_len)); 2232 op_len = ntohs(op_len); 2233 p += sizeof(op_len); 2234 plen -= sizeof(op_len); 2235 } 2236 if (op_len > 0) { 2237 if (plen < op_len) 2238 goto bad_len; 2239 op_val = p; 2240 p += op_len; 2241 plen -= op_len; 2242 } else 2243 op_val = NULL; 2244 2245 switch (op_type) { 2246 case OPT_PARAM_CAPABILITIES: /* RFC 3392 */ 2247 if (parse_capabilities(peer, op_val, op_len, 2248 &as) == -1) { 2249 session_notification(peer, ERR_OPEN, 0, 2250 NULL, 0); 2251 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2252 return (-1); 2253 } 2254 break; 2255 case OPT_PARAM_AUTH: /* deprecated */ 2256 default: 2257 /* 2258 * unsupported type 2259 * the RFCs tell us to leave the data section empty 2260 * and notify the peer with ERR_OPEN, ERR_OPEN_OPT. 2261 * How the peer should know _which_ optional parameter 2262 * we don't support is beyond me. 2263 */ 2264 log_peer_warnx(&peer->conf, 2265 "received OPEN message with unsupported optional " 2266 "parameter: type %u", op_type); 2267 session_notification(peer, ERR_OPEN, ERR_OPEN_OPT, 2268 NULL, 0); 2269 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2270 /* no punish */ 2271 timer_set(&peer->timers, Timer_IdleHold, 0); 2272 peer->IdleHoldTime /= 2; 2273 return (-1); 2274 } 2275 } 2276 2277 /* if remote-as is zero and it's a cloned neighbor, accept any */ 2278 if (peer->template && !peer->conf.remote_as && as != AS_TRANS) { 2279 peer->conf.remote_as = as; 2280 peer->conf.ebgp = (peer->conf.remote_as != peer->conf.local_as); 2281 if (!peer->conf.ebgp) 2282 /* force enforce_as off for iBGP sessions */ 2283 peer->conf.enforce_as = ENFORCE_AS_OFF; 2284 } 2285 2286 if (peer->conf.remote_as != as) { 2287 log_peer_warnx(&peer->conf, "peer sent wrong AS %s", 2288 log_as(as)); 2289 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, NULL, 0); 2290 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2291 return (-1); 2292 } 2293 2294 /* on iBGP sessions check for bgpid collision */ 2295 if (!peer->conf.ebgp && peer->remote_bgpid == conf->bgpid) { 2296 log_peer_warnx(&peer->conf, "peer BGPID %u conflicts with ours", 2297 ntohl(bgpid)); 2298 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2299 NULL, 0); 2300 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2301 return (-1); 2302 } 2303 2304 if (capa_neg_calc(peer, &suberr) == -1) { 2305 session_notification(peer, ERR_OPEN, suberr, NULL, 0); 2306 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2307 return (-1); 2308 } 2309 2310 return (0); 2311 } 2312 2313 int 2314 parse_update(struct peer *peer) 2315 { 2316 u_char *p; 2317 uint16_t datalen; 2318 2319 /* 2320 * we pass the message verbatim to the rde. 2321 * in case of errors the whole session is reset with a 2322 * notification anyway, we only need to know the peer 2323 */ 2324 p = peer->rbuf->rptr; 2325 p += MSGSIZE_HEADER_MARKER; 2326 memcpy(&datalen, p, sizeof(datalen)); 2327 datalen = ntohs(datalen); 2328 2329 p = peer->rbuf->rptr; 2330 p += MSGSIZE_HEADER; /* header is already checked */ 2331 datalen -= MSGSIZE_HEADER; 2332 2333 if (imsg_rde(IMSG_UPDATE, peer->conf.id, p, datalen) == -1) 2334 return (-1); 2335 2336 return (0); 2337 } 2338 2339 int 2340 parse_rrefresh(struct peer *peer) 2341 { 2342 struct route_refresh rr; 2343 uint16_t afi, datalen; 2344 uint8_t aid, safi, subtype; 2345 u_char *p; 2346 2347 p = peer->rbuf->rptr; 2348 p += MSGSIZE_HEADER_MARKER; 2349 memcpy(&datalen, p, sizeof(datalen)); 2350 datalen = ntohs(datalen); 2351 2352 p = peer->rbuf->rptr; 2353 p += MSGSIZE_HEADER; /* header is already checked */ 2354 2355 /* 2356 * We could check if we actually announced the capability but 2357 * as long as the message is correctly encoded we don't care. 2358 */ 2359 2360 /* afi, 2 byte */ 2361 memcpy(&afi, p, sizeof(afi)); 2362 afi = ntohs(afi); 2363 p += 2; 2364 /* subtype, 1 byte */ 2365 subtype = *p; 2366 p += 1; 2367 /* safi, 1 byte */ 2368 safi = *p; 2369 2370 /* check subtype if peer announced enhanced route refresh */ 2371 if (peer->capa.neg.enhanced_rr) { 2372 switch (subtype) { 2373 case ROUTE_REFRESH_REQUEST: 2374 /* no ORF support, so no oversized RREFRESH msgs */ 2375 if (datalen != MSGSIZE_RREFRESH) { 2376 log_peer_warnx(&peer->conf, 2377 "received RREFRESH: illegal len: %u byte", 2378 datalen); 2379 datalen = htons(datalen); 2380 session_notification(peer, ERR_HEADER, 2381 ERR_HDR_LEN, &datalen, sizeof(datalen)); 2382 bgp_fsm(peer, EVNT_CON_FATAL); 2383 return (-1); 2384 } 2385 peer->stats.refresh_rcvd_req++; 2386 break; 2387 case ROUTE_REFRESH_BEGIN_RR: 2388 case ROUTE_REFRESH_END_RR: 2389 /* special handling for RFC7313 */ 2390 if (datalen != MSGSIZE_RREFRESH) { 2391 log_peer_warnx(&peer->conf, 2392 "received RREFRESH: illegal len: %u byte", 2393 datalen); 2394 p = peer->rbuf->rptr; 2395 p += MSGSIZE_HEADER; 2396 datalen -= MSGSIZE_HEADER; 2397 session_notification(peer, ERR_RREFRESH, 2398 ERR_RR_INV_LEN, p, datalen); 2399 bgp_fsm(peer, EVNT_CON_FATAL); 2400 return (-1); 2401 } 2402 if (subtype == ROUTE_REFRESH_BEGIN_RR) 2403 peer->stats.refresh_rcvd_borr++; 2404 else 2405 peer->stats.refresh_rcvd_eorr++; 2406 break; 2407 default: 2408 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2409 "bad subtype %d", subtype); 2410 return (0); 2411 } 2412 } else { 2413 /* force subtype to default */ 2414 subtype = ROUTE_REFRESH_REQUEST; 2415 peer->stats.refresh_rcvd_req++; 2416 } 2417 2418 /* afi/safi unchecked - unrecognized values will be ignored anyway */ 2419 if (afi2aid(afi, safi, &aid) == -1) { 2420 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2421 "invalid afi/safi pair"); 2422 return (0); 2423 } 2424 2425 if (!peer->capa.neg.refresh && !peer->capa.neg.enhanced_rr) { 2426 log_peer_warnx(&peer->conf, "peer sent unexpected refresh"); 2427 return (0); 2428 } 2429 2430 rr.aid = aid; 2431 rr.subtype = subtype; 2432 2433 if (imsg_rde(IMSG_REFRESH, peer->conf.id, &rr, sizeof(rr)) == -1) 2434 return (-1); 2435 2436 return (0); 2437 } 2438 2439 int 2440 parse_notification(struct peer *peer) 2441 { 2442 u_char *p; 2443 uint16_t datalen; 2444 uint8_t errcode; 2445 uint8_t subcode; 2446 uint8_t capa_code; 2447 uint8_t capa_len; 2448 size_t reason_len; 2449 uint8_t i; 2450 2451 /* just log */ 2452 p = peer->rbuf->rptr; 2453 p += MSGSIZE_HEADER_MARKER; 2454 memcpy(&datalen, p, sizeof(datalen)); 2455 datalen = ntohs(datalen); 2456 2457 p = peer->rbuf->rptr; 2458 p += MSGSIZE_HEADER; /* header is already checked */ 2459 datalen -= MSGSIZE_HEADER; 2460 2461 memcpy(&errcode, p, sizeof(errcode)); 2462 p += sizeof(errcode); 2463 datalen -= sizeof(errcode); 2464 2465 memcpy(&subcode, p, sizeof(subcode)); 2466 p += sizeof(subcode); 2467 datalen -= sizeof(subcode); 2468 2469 log_notification(peer, errcode, subcode, p, datalen, "received"); 2470 peer->errcnt++; 2471 peer->stats.last_rcvd_errcode = errcode; 2472 peer->stats.last_rcvd_suberr = subcode; 2473 2474 if (errcode == ERR_OPEN && subcode == ERR_OPEN_CAPA) { 2475 if (datalen == 0) { /* zebra likes to send those.. humbug */ 2476 log_peer_warnx(&peer->conf, "received \"unsupported " 2477 "capability\" notification without data part, " 2478 "disabling capability announcements altogether"); 2479 session_capa_ann_none(peer); 2480 } 2481 2482 while (datalen > 0) { 2483 if (datalen < 2) { 2484 log_peer_warnx(&peer->conf, 2485 "parse_notification: " 2486 "expect len >= 2, len is %u", datalen); 2487 return (-1); 2488 } 2489 memcpy(&capa_code, p, sizeof(capa_code)); 2490 p += sizeof(capa_code); 2491 datalen -= sizeof(capa_code); 2492 memcpy(&capa_len, p, sizeof(capa_len)); 2493 p += sizeof(capa_len); 2494 datalen -= sizeof(capa_len); 2495 if (datalen < capa_len) { 2496 log_peer_warnx(&peer->conf, 2497 "parse_notification: capa_len %u exceeds " 2498 "remaining msg length %u", capa_len, 2499 datalen); 2500 return (-1); 2501 } 2502 p += capa_len; 2503 datalen -= capa_len; 2504 switch (capa_code) { 2505 case CAPA_MP: 2506 for (i = 0; i < AID_MAX; i++) 2507 peer->capa.ann.mp[i] = 0; 2508 log_peer_warnx(&peer->conf, 2509 "disabling multiprotocol capability"); 2510 break; 2511 case CAPA_REFRESH: 2512 peer->capa.ann.refresh = 0; 2513 log_peer_warnx(&peer->conf, 2514 "disabling route refresh capability"); 2515 break; 2516 case CAPA_ROLE: 2517 if (peer->capa.ann.policy == 1) { 2518 peer->capa.ann.policy = 0; 2519 log_peer_warnx(&peer->conf, 2520 "disabling role capability"); 2521 } else { 2522 log_peer_warnx(&peer->conf, 2523 "role capability enforced, " 2524 "not disabling"); 2525 } 2526 break; 2527 case CAPA_RESTART: 2528 peer->capa.ann.grestart.restart = 0; 2529 log_peer_warnx(&peer->conf, 2530 "disabling restart capability"); 2531 break; 2532 case CAPA_AS4BYTE: 2533 peer->capa.ann.as4byte = 0; 2534 log_peer_warnx(&peer->conf, 2535 "disabling 4-byte AS num capability"); 2536 break; 2537 case CAPA_ADD_PATH: 2538 memset(peer->capa.ann.add_path, 0, 2539 sizeof(peer->capa.ann.add_path)); 2540 log_peer_warnx(&peer->conf, 2541 "disabling ADD-PATH capability"); 2542 break; 2543 case CAPA_ENHANCED_RR: 2544 peer->capa.ann.enhanced_rr = 0; 2545 log_peer_warnx(&peer->conf, 2546 "disabling enhanced route refresh " 2547 "capability"); 2548 break; 2549 default: /* should not happen... */ 2550 log_peer_warnx(&peer->conf, "received " 2551 "\"unsupported capability\" notification " 2552 "for unknown capability %u, disabling " 2553 "capability announcements altogether", 2554 capa_code); 2555 session_capa_ann_none(peer); 2556 break; 2557 } 2558 } 2559 2560 return (1); 2561 } 2562 2563 if (errcode == ERR_OPEN && subcode == ERR_OPEN_OPT) { 2564 session_capa_ann_none(peer); 2565 return (1); 2566 } 2567 2568 if (errcode == ERR_CEASE && 2569 (subcode == ERR_CEASE_ADMIN_DOWN || 2570 subcode == ERR_CEASE_ADMIN_RESET)) { 2571 if (datalen > 1) { 2572 reason_len = *p++; 2573 datalen--; 2574 if (datalen < reason_len) { 2575 log_peer_warnx(&peer->conf, 2576 "received truncated shutdown reason"); 2577 return (0); 2578 } 2579 if (reason_len > REASON_LEN - 1) { 2580 log_peer_warnx(&peer->conf, 2581 "received overly long shutdown reason"); 2582 return (0); 2583 } 2584 memcpy(peer->stats.last_reason, p, reason_len); 2585 peer->stats.last_reason[reason_len] = '\0'; 2586 log_peer_warnx(&peer->conf, 2587 "received shutdown reason: \"%s\"", 2588 log_reason(peer->stats.last_reason)); 2589 p += reason_len; 2590 datalen -= reason_len; 2591 } 2592 } 2593 2594 return (0); 2595 } 2596 2597 int 2598 parse_capabilities(struct peer *peer, u_char *d, uint16_t dlen, uint32_t *as) 2599 { 2600 u_char *capa_val; 2601 uint32_t remote_as; 2602 uint16_t len; 2603 uint16_t afi; 2604 uint16_t gr_header; 2605 uint8_t safi; 2606 uint8_t aid; 2607 uint8_t flags; 2608 uint8_t capa_code; 2609 uint8_t capa_len; 2610 uint8_t i; 2611 2612 len = dlen; 2613 while (len > 0) { 2614 if (len < 2) { 2615 log_peer_warnx(&peer->conf, "Bad capabilities attr " 2616 "length: %u, too short", len); 2617 return (-1); 2618 } 2619 memcpy(&capa_code, d, sizeof(capa_code)); 2620 d += sizeof(capa_code); 2621 len -= sizeof(capa_code); 2622 memcpy(&capa_len, d, sizeof(capa_len)); 2623 d += sizeof(capa_len); 2624 len -= sizeof(capa_len); 2625 if (capa_len > 0) { 2626 if (len < capa_len) { 2627 log_peer_warnx(&peer->conf, 2628 "Bad capabilities attr length: " 2629 "len %u smaller than capa_len %u", 2630 len, capa_len); 2631 return (-1); 2632 } 2633 capa_val = d; 2634 d += capa_len; 2635 len -= capa_len; 2636 } else 2637 capa_val = NULL; 2638 2639 switch (capa_code) { 2640 case CAPA_MP: /* RFC 4760 */ 2641 if (capa_len != 4) { 2642 log_peer_warnx(&peer->conf, 2643 "Bad multi protocol capability length: " 2644 "%u", capa_len); 2645 break; 2646 } 2647 memcpy(&afi, capa_val, sizeof(afi)); 2648 afi = ntohs(afi); 2649 memcpy(&safi, capa_val + 3, sizeof(safi)); 2650 if (afi2aid(afi, safi, &aid) == -1) { 2651 log_peer_warnx(&peer->conf, 2652 "Received multi protocol capability: " 2653 " unknown AFI %u, safi %u pair", 2654 afi, safi); 2655 break; 2656 } 2657 peer->capa.peer.mp[aid] = 1; 2658 break; 2659 case CAPA_REFRESH: 2660 peer->capa.peer.refresh = 1; 2661 break; 2662 case CAPA_ROLE: 2663 if (capa_len != 1) { 2664 log_peer_warnx(&peer->conf, 2665 "Bad role capability length: %u", capa_len); 2666 break; 2667 } 2668 if (!peer->conf.ebgp) { 2669 log_peer_warnx(&peer->conf, 2670 "Received role capability on iBGP session"); 2671 break; 2672 } 2673 peer->capa.peer.policy = 1; 2674 peer->remote_role = capa2role(*capa_val); 2675 break; 2676 case CAPA_RESTART: 2677 if (capa_len == 2) { 2678 /* peer only supports EoR marker */ 2679 peer->capa.peer.grestart.restart = 1; 2680 peer->capa.peer.grestart.timeout = 0; 2681 break; 2682 } else if (capa_len % 4 != 2) { 2683 log_peer_warnx(&peer->conf, 2684 "Bad graceful restart capability length: " 2685 "%u", capa_len); 2686 peer->capa.peer.grestart.restart = 0; 2687 peer->capa.peer.grestart.timeout = 0; 2688 break; 2689 } 2690 2691 memcpy(&gr_header, capa_val, sizeof(gr_header)); 2692 gr_header = ntohs(gr_header); 2693 peer->capa.peer.grestart.timeout = 2694 gr_header & CAPA_GR_TIMEMASK; 2695 if (peer->capa.peer.grestart.timeout == 0) { 2696 log_peer_warnx(&peer->conf, "Received " 2697 "graceful restart timeout is zero"); 2698 peer->capa.peer.grestart.restart = 0; 2699 break; 2700 } 2701 2702 for (i = 2; i <= capa_len - 4; i += 4) { 2703 memcpy(&afi, capa_val + i, sizeof(afi)); 2704 afi = ntohs(afi); 2705 safi = capa_val[i + 2]; 2706 flags = capa_val[i + 3]; 2707 if (afi2aid(afi, safi, &aid) == -1) { 2708 log_peer_warnx(&peer->conf, 2709 "Received graceful restart capa: " 2710 " unknown AFI %u, safi %u pair", 2711 afi, safi); 2712 continue; 2713 } 2714 peer->capa.peer.grestart.flags[aid] |= 2715 CAPA_GR_PRESENT; 2716 if (flags & CAPA_GR_F_FLAG) 2717 peer->capa.peer.grestart.flags[aid] |= 2718 CAPA_GR_FORWARD; 2719 if (gr_header & CAPA_GR_R_FLAG) 2720 peer->capa.peer.grestart.flags[aid] |= 2721 CAPA_GR_RESTART; 2722 peer->capa.peer.grestart.restart = 2; 2723 } 2724 break; 2725 case CAPA_AS4BYTE: 2726 if (capa_len != 4) { 2727 log_peer_warnx(&peer->conf, 2728 "Bad AS4BYTE capability length: " 2729 "%u", capa_len); 2730 peer->capa.peer.as4byte = 0; 2731 break; 2732 } 2733 memcpy(&remote_as, capa_val, sizeof(remote_as)); 2734 *as = ntohl(remote_as); 2735 if (*as == 0) { 2736 log_peer_warnx(&peer->conf, 2737 "peer requests unacceptable AS %u", *as); 2738 session_notification(peer, ERR_OPEN, 2739 ERR_OPEN_AS, NULL, 0); 2740 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2741 return (-1); 2742 } 2743 peer->capa.peer.as4byte = 1; 2744 break; 2745 case CAPA_ADD_PATH: 2746 if (capa_len % 4 != 0) { 2747 log_peer_warnx(&peer->conf, 2748 "Bad ADD-PATH capability length: " 2749 "%u", capa_len); 2750 memset(peer->capa.peer.add_path, 0, 2751 sizeof(peer->capa.peer.add_path)); 2752 break; 2753 } 2754 for (i = 0; i <= capa_len - 4; i += 4) { 2755 memcpy(&afi, capa_val + i, sizeof(afi)); 2756 afi = ntohs(afi); 2757 safi = capa_val[i + 2]; 2758 flags = capa_val[i + 3]; 2759 if (afi2aid(afi, safi, &aid) == -1) { 2760 log_peer_warnx(&peer->conf, 2761 "Received ADD-PATH capa: " 2762 " unknown AFI %u, safi %u pair", 2763 afi, safi); 2764 memset(peer->capa.peer.add_path, 0, 2765 sizeof(peer->capa.peer.add_path)); 2766 break; 2767 } 2768 if (flags & ~CAPA_AP_BIDIR) { 2769 log_peer_warnx(&peer->conf, 2770 "Received ADD-PATH capa: " 2771 " bad flags %x", flags); 2772 memset(peer->capa.peer.add_path, 0, 2773 sizeof(peer->capa.peer.add_path)); 2774 break; 2775 } 2776 peer->capa.peer.add_path[aid] = flags; 2777 } 2778 break; 2779 case CAPA_ENHANCED_RR: 2780 peer->capa.peer.enhanced_rr = 1; 2781 break; 2782 default: 2783 break; 2784 } 2785 } 2786 2787 return (0); 2788 } 2789 2790 int 2791 capa_neg_calc(struct peer *p, uint8_t *suberr) 2792 { 2793 uint8_t i, hasmp = 0; 2794 2795 /* a capability is accepted only if both sides announced it */ 2796 2797 p->capa.neg.refresh = 2798 (p->capa.ann.refresh && p->capa.peer.refresh) != 0; 2799 p->capa.neg.enhanced_rr = 2800 (p->capa.ann.enhanced_rr && p->capa.peer.enhanced_rr) != 0; 2801 2802 p->capa.neg.as4byte = 2803 (p->capa.ann.as4byte && p->capa.peer.as4byte) != 0; 2804 2805 /* MP: both side must agree on the AFI,SAFI pair */ 2806 for (i = 0; i < AID_MAX; i++) { 2807 if (p->capa.ann.mp[i] && p->capa.peer.mp[i]) 2808 p->capa.neg.mp[i] = 1; 2809 else 2810 p->capa.neg.mp[i] = 0; 2811 if (p->capa.ann.mp[i]) 2812 hasmp = 1; 2813 } 2814 /* if no MP capability present default to IPv4 unicast mode */ 2815 if (!hasmp) 2816 p->capa.neg.mp[AID_INET] = 1; 2817 2818 /* 2819 * graceful restart: the peer capabilities are of interest here. 2820 * It is necessary to compare the new values with the previous ones 2821 * and act accordingly. AFI/SAFI that are not part in the MP capability 2822 * are treated as not being present. 2823 * Also make sure that a flush happens if the session stopped 2824 * supporting graceful restart. 2825 */ 2826 2827 for (i = 0; i < AID_MAX; i++) { 2828 int8_t negflags; 2829 2830 /* disable GR if the AFI/SAFI is not present */ 2831 if ((p->capa.peer.grestart.flags[i] & CAPA_GR_PRESENT && 2832 p->capa.neg.mp[i] == 0)) 2833 p->capa.peer.grestart.flags[i] = 0; /* disable */ 2834 /* look at current GR state and decide what to do */ 2835 negflags = p->capa.neg.grestart.flags[i]; 2836 p->capa.neg.grestart.flags[i] = p->capa.peer.grestart.flags[i]; 2837 if (negflags & CAPA_GR_RESTARTING) { 2838 if (p->capa.ann.grestart.restart != 0 && 2839 p->capa.peer.grestart.flags[i] & CAPA_GR_FORWARD) { 2840 p->capa.neg.grestart.flags[i] |= 2841 CAPA_GR_RESTARTING; 2842 } else { 2843 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 2844 &i, sizeof(i)) == -1) { 2845 log_peer_warnx(&p->conf, 2846 "imsg send failed"); 2847 return (-1); 2848 } 2849 log_peer_warnx(&p->conf, "graceful restart of " 2850 "%s, not restarted, flushing", aid2str(i)); 2851 } 2852 } 2853 } 2854 p->capa.neg.grestart.timeout = p->capa.peer.grestart.timeout; 2855 p->capa.neg.grestart.restart = p->capa.peer.grestart.restart; 2856 if (p->capa.ann.grestart.restart == 0) 2857 p->capa.neg.grestart.restart = 0; 2858 2859 2860 /* 2861 * ADD-PATH: set only those bits where both sides agree. 2862 * For this compare our send bit with the recv bit from the peer 2863 * and vice versa. 2864 * The flags are stored from this systems view point. 2865 */ 2866 memset(p->capa.neg.add_path, 0, sizeof(p->capa.neg.add_path)); 2867 if (p->capa.ann.add_path[0]) { 2868 for (i = AID_MIN; i < AID_MAX; i++) { 2869 if ((p->capa.ann.add_path[i] & CAPA_AP_RECV) && 2870 (p->capa.peer.add_path[i] & CAPA_AP_SEND)) { 2871 p->capa.neg.add_path[i] |= CAPA_AP_RECV; 2872 p->capa.neg.add_path[0] |= CAPA_AP_RECV; 2873 } 2874 if ((p->capa.ann.add_path[i] & CAPA_AP_SEND) && 2875 (p->capa.peer.add_path[i] & CAPA_AP_RECV)) { 2876 p->capa.neg.add_path[i] |= CAPA_AP_SEND; 2877 p->capa.neg.add_path[0] |= CAPA_AP_SEND; 2878 } 2879 } 2880 } 2881 2882 /* 2883 * Open policy: check that the policy is sensible. 2884 * 2885 * Make sure that the roles match and set the negotiated capability 2886 * to the role of the peer. So the RDE can inject the OTC attribute. 2887 * See RFC 9234, section 4.2. 2888 * These checks should only happen on ebgp sessions. 2889 */ 2890 if (p->capa.ann.policy != 0 && p->capa.peer.policy != 0 && 2891 p->conf.ebgp) { 2892 switch (p->conf.role) { 2893 case ROLE_PROVIDER: 2894 if (p->remote_role != ROLE_CUSTOMER) 2895 goto fail; 2896 break; 2897 case ROLE_RS: 2898 if (p->remote_role != ROLE_RS_CLIENT) 2899 goto fail; 2900 break; 2901 case ROLE_RS_CLIENT: 2902 if (p->remote_role != ROLE_RS) 2903 goto fail; 2904 break; 2905 case ROLE_CUSTOMER: 2906 if (p->remote_role != ROLE_PROVIDER) 2907 goto fail; 2908 break; 2909 case ROLE_PEER: 2910 if (p->remote_role != ROLE_PEER) 2911 goto fail; 2912 break; 2913 default: 2914 fail: 2915 log_peer_warnx(&p->conf, "open policy role mismatch: " 2916 "our role %s, their role %s", 2917 log_policy(p->conf.role), 2918 log_policy(p->remote_role)); 2919 *suberr = ERR_OPEN_ROLE; 2920 return (-1); 2921 } 2922 p->capa.neg.policy = 1; 2923 } else if (p->capa.ann.policy == 2 && p->conf.ebgp) { 2924 /* enforce presence of open policy role capability */ 2925 log_peer_warnx(&p->conf, "open policy role enforced but " 2926 "not present"); 2927 *suberr = ERR_OPEN_ROLE; 2928 return (-1); 2929 } 2930 2931 return (0); 2932 } 2933 2934 void 2935 session_dispatch_imsg(struct imsgbuf *ibuf, int idx, u_int *listener_cnt) 2936 { 2937 struct imsg imsg; 2938 struct mrt xmrt; 2939 struct route_refresh rr; 2940 struct mrt *mrt; 2941 struct imsgbuf *i; 2942 struct peer *p; 2943 struct listen_addr *la, *nla; 2944 struct session_dependon *sdon; 2945 u_char *data; 2946 int n, fd, depend_ok, restricted; 2947 uint16_t t; 2948 uint8_t aid, errcode, subcode; 2949 2950 while (ibuf) { 2951 if ((n = imsg_get(ibuf, &imsg)) == -1) 2952 fatal("session_dispatch_imsg: imsg_get error"); 2953 2954 if (n == 0) 2955 break; 2956 2957 switch (imsg.hdr.type) { 2958 case IMSG_SOCKET_CONN: 2959 case IMSG_SOCKET_CONN_CTL: 2960 if (idx != PFD_PIPE_MAIN) 2961 fatalx("reconf request not from parent"); 2962 if ((fd = imsg.fd) == -1) { 2963 log_warnx("expected to receive imsg fd to " 2964 "RDE but didn't receive any"); 2965 break; 2966 } 2967 if ((i = malloc(sizeof(struct imsgbuf))) == NULL) 2968 fatal(NULL); 2969 imsg_init(i, fd); 2970 if (imsg.hdr.type == IMSG_SOCKET_CONN) { 2971 if (ibuf_rde) { 2972 log_warnx("Unexpected imsg connection " 2973 "to RDE received"); 2974 msgbuf_clear(&ibuf_rde->w); 2975 free(ibuf_rde); 2976 } 2977 ibuf_rde = i; 2978 } else { 2979 if (ibuf_rde_ctl) { 2980 log_warnx("Unexpected imsg ctl " 2981 "connection to RDE received"); 2982 msgbuf_clear(&ibuf_rde_ctl->w); 2983 free(ibuf_rde_ctl); 2984 } 2985 ibuf_rde_ctl = i; 2986 } 2987 break; 2988 case IMSG_RECONF_CONF: 2989 if (idx != PFD_PIPE_MAIN) 2990 fatalx("reconf request not from parent"); 2991 nconf = new_config(); 2992 2993 copy_config(nconf, imsg.data); 2994 pending_reconf = 1; 2995 break; 2996 case IMSG_RECONF_PEER: 2997 if (idx != PFD_PIPE_MAIN) 2998 fatalx("reconf request not from parent"); 2999 if ((p = calloc(1, sizeof(struct peer))) == NULL) 3000 fatal("new_peer"); 3001 memcpy(&p->conf, imsg.data, sizeof(struct peer_config)); 3002 p->state = p->prev_state = STATE_NONE; 3003 p->reconf_action = RECONF_REINIT; 3004 if (RB_INSERT(peer_head, &nconf->peers, p) != NULL) 3005 fatalx("%s: peer tree is corrupt", __func__); 3006 break; 3007 case IMSG_RECONF_LISTENER: 3008 if (idx != PFD_PIPE_MAIN) 3009 fatalx("reconf request not from parent"); 3010 if (nconf == NULL) 3011 fatalx("IMSG_RECONF_LISTENER but no config"); 3012 nla = imsg.data; 3013 TAILQ_FOREACH(la, conf->listen_addrs, entry) 3014 if (!la_cmp(la, nla)) 3015 break; 3016 3017 if (la == NULL) { 3018 if (nla->reconf != RECONF_REINIT) 3019 fatalx("king bula sez: " 3020 "expected REINIT"); 3021 3022 if ((nla->fd = imsg.fd) == -1) 3023 log_warnx("expected to receive fd for " 3024 "%s but didn't receive any", 3025 log_sockaddr((struct sockaddr *) 3026 &nla->sa, nla->sa_len)); 3027 3028 la = calloc(1, sizeof(struct listen_addr)); 3029 if (la == NULL) 3030 fatal(NULL); 3031 memcpy(&la->sa, &nla->sa, sizeof(la->sa)); 3032 la->flags = nla->flags; 3033 la->fd = nla->fd; 3034 la->reconf = RECONF_REINIT; 3035 TAILQ_INSERT_TAIL(nconf->listen_addrs, la, 3036 entry); 3037 } else { 3038 if (nla->reconf != RECONF_KEEP) 3039 fatalx("king bula sez: expected KEEP"); 3040 la->reconf = RECONF_KEEP; 3041 } 3042 3043 break; 3044 case IMSG_RECONF_CTRL: 3045 if (idx != PFD_PIPE_MAIN) 3046 fatalx("reconf request not from parent"); 3047 if (imsg.hdr.len != IMSG_HEADER_SIZE + 3048 sizeof(restricted)) 3049 fatalx("RECONF_CTRL imsg with wrong len"); 3050 memcpy(&restricted, imsg.data, sizeof(restricted)); 3051 if (imsg.fd == -1) { 3052 log_warnx("expected to receive fd for control " 3053 "socket but didn't receive any"); 3054 break; 3055 } 3056 if (restricted) { 3057 control_shutdown(rcsock); 3058 rcsock = imsg.fd; 3059 } else { 3060 control_shutdown(csock); 3061 csock = imsg.fd; 3062 } 3063 break; 3064 case IMSG_RECONF_DRAIN: 3065 switch (idx) { 3066 case PFD_PIPE_ROUTE: 3067 if (nconf != NULL) 3068 fatalx("got unexpected %s from RDE", 3069 "IMSG_RECONF_DONE"); 3070 imsg_compose(ibuf_main, IMSG_RECONF_DONE, 0, 0, 3071 -1, NULL, 0); 3072 break; 3073 case PFD_PIPE_MAIN: 3074 if (nconf == NULL) 3075 fatalx("got unexpected %s from parent", 3076 "IMSG_RECONF_DONE"); 3077 imsg_compose(ibuf_main, IMSG_RECONF_DRAIN, 0, 0, 3078 -1, NULL, 0); 3079 break; 3080 default: 3081 fatalx("reconf request not from parent or RDE"); 3082 } 3083 break; 3084 case IMSG_RECONF_DONE: 3085 if (idx != PFD_PIPE_MAIN) 3086 fatalx("reconf request not from parent"); 3087 if (nconf == NULL) 3088 fatalx("got IMSG_RECONF_DONE but no config"); 3089 copy_config(conf, nconf); 3090 merge_peers(conf, nconf); 3091 3092 /* delete old listeners */ 3093 for (la = TAILQ_FIRST(conf->listen_addrs); la != NULL; 3094 la = nla) { 3095 nla = TAILQ_NEXT(la, entry); 3096 if (la->reconf == RECONF_NONE) { 3097 log_info("not listening on %s any more", 3098 log_sockaddr((struct sockaddr *) 3099 &la->sa, la->sa_len)); 3100 TAILQ_REMOVE(conf->listen_addrs, la, 3101 entry); 3102 close(la->fd); 3103 free(la); 3104 } 3105 } 3106 3107 /* add new listeners */ 3108 TAILQ_CONCAT(conf->listen_addrs, nconf->listen_addrs, 3109 entry); 3110 3111 setup_listeners(listener_cnt); 3112 free_config(nconf); 3113 nconf = NULL; 3114 pending_reconf = 0; 3115 log_info("SE reconfigured"); 3116 /* 3117 * IMSG_RECONF_DONE is sent when the RDE drained 3118 * the peer config sent in merge_peers(). 3119 */ 3120 break; 3121 case IMSG_SESSION_DEPENDON: 3122 if (idx != PFD_PIPE_MAIN) 3123 fatalx("IFINFO message not from parent"); 3124 if (imsg.hdr.len != IMSG_HEADER_SIZE + 3125 sizeof(struct session_dependon)) 3126 fatalx("DEPENDON imsg with wrong len"); 3127 sdon = imsg.data; 3128 depend_ok = sdon->depend_state; 3129 3130 RB_FOREACH(p, peer_head, &conf->peers) 3131 if (!strcmp(p->conf.if_depend, sdon->ifname)) { 3132 if (depend_ok && !p->depend_ok) { 3133 p->depend_ok = depend_ok; 3134 bgp_fsm(p, EVNT_START); 3135 } else if (!depend_ok && p->depend_ok) { 3136 p->depend_ok = depend_ok; 3137 session_stop(p, 3138 ERR_CEASE_OTHER_CHANGE); 3139 } 3140 } 3141 break; 3142 case IMSG_MRT_OPEN: 3143 case IMSG_MRT_REOPEN: 3144 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3145 sizeof(struct mrt)) { 3146 log_warnx("wrong imsg len"); 3147 break; 3148 } 3149 3150 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 3151 if ((xmrt.wbuf.fd = imsg.fd) == -1) 3152 log_warnx("expected to receive fd for mrt dump " 3153 "but didn't receive any"); 3154 3155 mrt = mrt_get(&mrthead, &xmrt); 3156 if (mrt == NULL) { 3157 /* new dump */ 3158 mrt = calloc(1, sizeof(struct mrt)); 3159 if (mrt == NULL) 3160 fatal("session_dispatch_imsg"); 3161 memcpy(mrt, &xmrt, sizeof(struct mrt)); 3162 TAILQ_INIT(&mrt->wbuf.bufs); 3163 LIST_INSERT_HEAD(&mrthead, mrt, entry); 3164 } else { 3165 /* old dump reopened */ 3166 close(mrt->wbuf.fd); 3167 mrt->wbuf.fd = xmrt.wbuf.fd; 3168 } 3169 break; 3170 case IMSG_MRT_CLOSE: 3171 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3172 sizeof(struct mrt)) { 3173 log_warnx("wrong imsg len"); 3174 break; 3175 } 3176 3177 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 3178 mrt = mrt_get(&mrthead, &xmrt); 3179 if (mrt != NULL) 3180 mrt_done(mrt); 3181 break; 3182 case IMSG_CTL_KROUTE: 3183 case IMSG_CTL_KROUTE_ADDR: 3184 case IMSG_CTL_SHOW_NEXTHOP: 3185 case IMSG_CTL_SHOW_INTERFACE: 3186 case IMSG_CTL_SHOW_FIB_TABLES: 3187 case IMSG_CTL_SHOW_RTR: 3188 case IMSG_CTL_SHOW_TIMER: 3189 if (idx != PFD_PIPE_MAIN) 3190 fatalx("ctl kroute request not from parent"); 3191 control_imsg_relay(&imsg, NULL); 3192 break; 3193 case IMSG_CTL_SHOW_NEIGHBOR: 3194 if (idx != PFD_PIPE_ROUTE_CTL) 3195 fatalx("ctl rib request not from RDE"); 3196 p = getpeerbyid(conf, imsg.hdr.peerid); 3197 control_imsg_relay(&imsg, p); 3198 break; 3199 case IMSG_CTL_SHOW_RIB: 3200 case IMSG_CTL_SHOW_RIB_PREFIX: 3201 case IMSG_CTL_SHOW_RIB_COMMUNITIES: 3202 case IMSG_CTL_SHOW_RIB_ATTR: 3203 case IMSG_CTL_SHOW_RIB_MEM: 3204 case IMSG_CTL_SHOW_NETWORK: 3205 case IMSG_CTL_SHOW_FLOWSPEC: 3206 case IMSG_CTL_SHOW_SET: 3207 if (idx != PFD_PIPE_ROUTE_CTL) 3208 fatalx("ctl rib request not from RDE"); 3209 control_imsg_relay(&imsg, NULL); 3210 break; 3211 case IMSG_CTL_END: 3212 case IMSG_CTL_RESULT: 3213 control_imsg_relay(&imsg, NULL); 3214 break; 3215 case IMSG_UPDATE: 3216 if (idx != PFD_PIPE_ROUTE) 3217 fatalx("update request not from RDE"); 3218 if (imsg.hdr.len > IMSG_HEADER_SIZE + 3219 MAX_PKTSIZE - MSGSIZE_HEADER || 3220 imsg.hdr.len < IMSG_HEADER_SIZE + 3221 MSGSIZE_UPDATE_MIN - MSGSIZE_HEADER) 3222 log_warnx("RDE sent invalid update"); 3223 else 3224 session_update(imsg.hdr.peerid, imsg.data, 3225 imsg.hdr.len - IMSG_HEADER_SIZE); 3226 break; 3227 case IMSG_UPDATE_ERR: 3228 if (idx != PFD_PIPE_ROUTE) 3229 fatalx("update request not from RDE"); 3230 if (imsg.hdr.len < IMSG_HEADER_SIZE + 2) { 3231 log_warnx("RDE sent invalid notification"); 3232 break; 3233 } 3234 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3235 log_warnx("no such peer: id=%u", 3236 imsg.hdr.peerid); 3237 break; 3238 } 3239 data = imsg.data; 3240 errcode = *data++; 3241 subcode = *data++; 3242 3243 if (imsg.hdr.len == IMSG_HEADER_SIZE + 2) 3244 data = NULL; 3245 3246 session_notification(p, errcode, subcode, 3247 data, imsg.hdr.len - IMSG_HEADER_SIZE - 2); 3248 switch (errcode) { 3249 case ERR_CEASE: 3250 switch (subcode) { 3251 case ERR_CEASE_MAX_PREFIX: 3252 case ERR_CEASE_MAX_SENT_PREFIX: 3253 t = p->conf.max_out_prefix_restart; 3254 if (subcode == ERR_CEASE_MAX_PREFIX) 3255 t = p->conf.max_prefix_restart; 3256 3257 bgp_fsm(p, EVNT_STOP); 3258 if (t) 3259 timer_set(&p->timers, 3260 Timer_IdleHold, 60 * t); 3261 break; 3262 default: 3263 bgp_fsm(p, EVNT_CON_FATAL); 3264 break; 3265 } 3266 break; 3267 default: 3268 bgp_fsm(p, EVNT_CON_FATAL); 3269 break; 3270 } 3271 break; 3272 case IMSG_REFRESH: 3273 if (idx != PFD_PIPE_ROUTE) 3274 fatalx("route refresh request not from RDE"); 3275 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(rr)) { 3276 log_warnx("RDE sent invalid refresh msg"); 3277 break; 3278 } 3279 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3280 log_warnx("no such peer: id=%u", 3281 imsg.hdr.peerid); 3282 break; 3283 } 3284 memcpy(&rr, imsg.data, sizeof(rr)); 3285 if (rr.aid >= AID_MAX) 3286 fatalx("IMSG_REFRESH: bad AID"); 3287 session_rrefresh(p, rr.aid, rr.subtype); 3288 break; 3289 case IMSG_SESSION_RESTARTED: 3290 if (idx != PFD_PIPE_ROUTE) 3291 fatalx("update request not from RDE"); 3292 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(aid)) { 3293 log_warnx("RDE sent invalid restart msg"); 3294 break; 3295 } 3296 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3297 log_warnx("no such peer: id=%u", 3298 imsg.hdr.peerid); 3299 break; 3300 } 3301 memcpy(&aid, imsg.data, sizeof(aid)); 3302 if (aid >= AID_MAX) 3303 fatalx("IMSG_SESSION_RESTARTED: bad AID"); 3304 if (p->capa.neg.grestart.flags[aid] & 3305 CAPA_GR_RESTARTING) { 3306 log_peer_warnx(&p->conf, 3307 "graceful restart of %s finished", 3308 aid2str(aid)); 3309 p->capa.neg.grestart.flags[aid] &= 3310 ~CAPA_GR_RESTARTING; 3311 timer_stop(&p->timers, Timer_RestartTimeout); 3312 3313 /* signal back to RDE to cleanup stale routes */ 3314 if (imsg_rde(IMSG_SESSION_RESTARTED, 3315 imsg.hdr.peerid, &aid, sizeof(aid)) == -1) 3316 fatal("imsg_compose: " 3317 "IMSG_SESSION_RESTARTED"); 3318 } 3319 break; 3320 case IMSG_SESSION_DOWN: 3321 if (idx != PFD_PIPE_ROUTE) 3322 fatalx("update request not from RDE"); 3323 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 3324 log_warnx("no such peer: id=%u", 3325 imsg.hdr.peerid); 3326 break; 3327 } 3328 session_stop(p, ERR_CEASE_ADMIN_DOWN); 3329 break; 3330 default: 3331 break; 3332 } 3333 imsg_free(&imsg); 3334 } 3335 } 3336 3337 int 3338 la_cmp(struct listen_addr *a, struct listen_addr *b) 3339 { 3340 struct sockaddr_in *in_a, *in_b; 3341 struct sockaddr_in6 *in6_a, *in6_b; 3342 3343 if (a->sa.ss_family != b->sa.ss_family) 3344 return (1); 3345 3346 switch (a->sa.ss_family) { 3347 case AF_INET: 3348 in_a = (struct sockaddr_in *)&a->sa; 3349 in_b = (struct sockaddr_in *)&b->sa; 3350 if (in_a->sin_addr.s_addr != in_b->sin_addr.s_addr) 3351 return (1); 3352 if (in_a->sin_port != in_b->sin_port) 3353 return (1); 3354 break; 3355 case AF_INET6: 3356 in6_a = (struct sockaddr_in6 *)&a->sa; 3357 in6_b = (struct sockaddr_in6 *)&b->sa; 3358 if (memcmp(&in6_a->sin6_addr, &in6_b->sin6_addr, 3359 sizeof(struct in6_addr))) 3360 return (1); 3361 if (in6_a->sin6_port != in6_b->sin6_port) 3362 return (1); 3363 break; 3364 default: 3365 fatal("king bula sez: unknown address family"); 3366 /* NOTREACHED */ 3367 } 3368 3369 return (0); 3370 } 3371 3372 struct peer * 3373 getpeerbydesc(struct bgpd_config *c, const char *descr) 3374 { 3375 struct peer *p, *res = NULL; 3376 int match = 0; 3377 3378 RB_FOREACH(p, peer_head, &c->peers) 3379 if (!strcmp(p->conf.descr, descr)) { 3380 res = p; 3381 match++; 3382 } 3383 3384 if (match > 1) 3385 log_info("neighbor description \"%s\" not unique, request " 3386 "aborted", descr); 3387 3388 if (match == 1) 3389 return (res); 3390 else 3391 return (NULL); 3392 } 3393 3394 struct peer * 3395 getpeerbyip(struct bgpd_config *c, struct sockaddr *ip) 3396 { 3397 struct bgpd_addr addr; 3398 struct peer *p, *newpeer, *loose = NULL; 3399 uint32_t id; 3400 3401 sa2addr(ip, &addr, NULL); 3402 3403 /* we might want a more effective way to find peers by IP */ 3404 RB_FOREACH(p, peer_head, &c->peers) 3405 if (!p->conf.template && 3406 !memcmp(&addr, &p->conf.remote_addr, sizeof(addr))) 3407 return (p); 3408 3409 /* try template matching */ 3410 RB_FOREACH(p, peer_head, &c->peers) 3411 if (p->conf.template && 3412 p->conf.remote_addr.aid == addr.aid && 3413 session_match_mask(p, &addr)) 3414 if (loose == NULL || loose->conf.remote_masklen < 3415 p->conf.remote_masklen) 3416 loose = p; 3417 3418 if (loose != NULL) { 3419 /* clone */ 3420 if ((newpeer = malloc(sizeof(struct peer))) == NULL) 3421 fatal(NULL); 3422 memcpy(newpeer, loose, sizeof(struct peer)); 3423 for (id = PEER_ID_DYN_MAX; id > PEER_ID_STATIC_MAX; id--) { 3424 if (getpeerbyid(c, id) == NULL) /* we found a free id */ 3425 break; 3426 } 3427 newpeer->template = loose; 3428 session_template_clone(newpeer, ip, id, 0); 3429 newpeer->state = newpeer->prev_state = STATE_NONE; 3430 newpeer->reconf_action = RECONF_KEEP; 3431 newpeer->rbuf = NULL; 3432 newpeer->rpending = 0; 3433 init_peer(newpeer); 3434 bgp_fsm(newpeer, EVNT_START); 3435 if (RB_INSERT(peer_head, &c->peers, newpeer) != NULL) 3436 fatalx("%s: peer tree is corrupt", __func__); 3437 return (newpeer); 3438 } 3439 3440 return (NULL); 3441 } 3442 3443 struct peer * 3444 getpeerbyid(struct bgpd_config *c, uint32_t peerid) 3445 { 3446 static struct peer lookup; 3447 3448 lookup.conf.id = peerid; 3449 3450 return RB_FIND(peer_head, &c->peers, &lookup); 3451 } 3452 3453 int 3454 peer_matched(struct peer *p, struct ctl_neighbor *n) 3455 { 3456 char *s; 3457 3458 if (n && n->addr.aid) { 3459 if (memcmp(&p->conf.remote_addr, &n->addr, 3460 sizeof(p->conf.remote_addr))) 3461 return 0; 3462 } else if (n && n->descr[0]) { 3463 s = n->is_group ? p->conf.group : p->conf.descr; 3464 if (strcmp(s, n->descr)) 3465 return 0; 3466 } 3467 return 1; 3468 } 3469 3470 void 3471 session_template_clone(struct peer *p, struct sockaddr *ip, uint32_t id, 3472 uint32_t as) 3473 { 3474 struct bgpd_addr remote_addr; 3475 3476 if (ip) 3477 sa2addr(ip, &remote_addr, NULL); 3478 else 3479 memcpy(&remote_addr, &p->conf.remote_addr, sizeof(remote_addr)); 3480 3481 memcpy(&p->conf, &p->template->conf, sizeof(struct peer_config)); 3482 3483 p->conf.id = id; 3484 3485 if (as) { 3486 p->conf.remote_as = as; 3487 p->conf.ebgp = (p->conf.remote_as != p->conf.local_as); 3488 if (!p->conf.ebgp) 3489 /* force enforce_as off for iBGP sessions */ 3490 p->conf.enforce_as = ENFORCE_AS_OFF; 3491 } 3492 3493 memcpy(&p->conf.remote_addr, &remote_addr, sizeof(remote_addr)); 3494 switch (p->conf.remote_addr.aid) { 3495 case AID_INET: 3496 p->conf.remote_masklen = 32; 3497 break; 3498 case AID_INET6: 3499 p->conf.remote_masklen = 128; 3500 break; 3501 } 3502 p->conf.template = 0; 3503 } 3504 3505 int 3506 session_match_mask(struct peer *p, struct bgpd_addr *a) 3507 { 3508 struct bgpd_addr masked; 3509 3510 applymask(&masked, a, p->conf.remote_masklen); 3511 if (memcmp(&masked, &p->conf.remote_addr, sizeof(masked)) == 0) 3512 return (1); 3513 return (0); 3514 } 3515 3516 void 3517 session_down(struct peer *peer) 3518 { 3519 memset(&peer->capa.neg, 0, sizeof(peer->capa.neg)); 3520 peer->stats.last_updown = getmonotime(); 3521 /* 3522 * session_down is called in the exit code path so check 3523 * if the RDE is still around, if not there is no need to 3524 * send the message. 3525 */ 3526 if (ibuf_rde == NULL) 3527 return; 3528 if (imsg_rde(IMSG_SESSION_DOWN, peer->conf.id, NULL, 0) == -1) 3529 fatalx("imsg_compose error"); 3530 } 3531 3532 void 3533 session_up(struct peer *p) 3534 { 3535 struct session_up sup; 3536 3537 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3538 &p->conf, sizeof(p->conf)) == -1) 3539 fatalx("imsg_compose error"); 3540 3541 if (p->local.aid == AID_INET) { 3542 sup.local_v4_addr = p->local; 3543 sup.local_v6_addr = p->local_alt; 3544 } else { 3545 sup.local_v6_addr = p->local; 3546 sup.local_v4_addr = p->local_alt; 3547 } 3548 sup.remote_addr = p->remote; 3549 3550 sup.remote_bgpid = p->remote_bgpid; 3551 sup.short_as = p->short_as; 3552 memcpy(&sup.capa, &p->capa.neg, sizeof(sup.capa)); 3553 p->stats.last_updown = getmonotime(); 3554 if (imsg_rde(IMSG_SESSION_UP, p->conf.id, &sup, sizeof(sup)) == -1) 3555 fatalx("imsg_compose error"); 3556 } 3557 3558 int 3559 imsg_ctl_parent(int type, uint32_t peerid, pid_t pid, void *data, 3560 uint16_t datalen) 3561 { 3562 return imsg_compose(ibuf_main, type, peerid, pid, -1, data, datalen); 3563 } 3564 3565 int 3566 imsg_ctl_rde(int type, uint32_t peerid, pid_t pid, void *data, uint16_t datalen) 3567 { 3568 if (ibuf_rde_ctl == NULL) 3569 return (0); 3570 3571 /* 3572 * Use control socket to talk to RDE to bypass the queue of the 3573 * regular imsg socket. 3574 */ 3575 return imsg_compose(ibuf_rde_ctl, type, peerid, pid, -1, data, datalen); 3576 } 3577 3578 int 3579 imsg_rde(int type, uint32_t peerid, void *data, uint16_t datalen) 3580 { 3581 if (ibuf_rde == NULL) 3582 return (0); 3583 3584 return imsg_compose(ibuf_rde, type, peerid, 0, -1, data, datalen); 3585 } 3586 3587 void 3588 session_demote(struct peer *p, int level) 3589 { 3590 struct demote_msg msg; 3591 3592 strlcpy(msg.demote_group, p->conf.demote_group, 3593 sizeof(msg.demote_group)); 3594 msg.level = level; 3595 if (imsg_compose(ibuf_main, IMSG_DEMOTE, p->conf.id, 0, -1, 3596 &msg, sizeof(msg)) == -1) 3597 fatalx("imsg_compose error"); 3598 3599 p->demoted += level; 3600 } 3601 3602 void 3603 session_stop(struct peer *peer, uint8_t subcode) 3604 { 3605 char data[REASON_LEN]; 3606 size_t datalen; 3607 size_t reason_len; 3608 char *communication; 3609 3610 datalen = 0; 3611 communication = peer->conf.reason; 3612 3613 if ((subcode == ERR_CEASE_ADMIN_DOWN || 3614 subcode == ERR_CEASE_ADMIN_RESET) 3615 && communication && *communication) { 3616 reason_len = strlen(communication); 3617 if (reason_len > REASON_LEN - 1) { 3618 log_peer_warnx(&peer->conf, 3619 "trying to send overly long shutdown reason"); 3620 } else { 3621 data[0] = reason_len; 3622 datalen = reason_len + sizeof(data[0]); 3623 memcpy(data + 1, communication, reason_len); 3624 } 3625 } 3626 switch (peer->state) { 3627 case STATE_OPENSENT: 3628 case STATE_OPENCONFIRM: 3629 case STATE_ESTABLISHED: 3630 session_notification(peer, ERR_CEASE, subcode, data, datalen); 3631 break; 3632 default: 3633 /* session not open, no need to send notification */ 3634 break; 3635 } 3636 bgp_fsm(peer, EVNT_STOP); 3637 } 3638 3639 void 3640 merge_peers(struct bgpd_config *c, struct bgpd_config *nc) 3641 { 3642 struct peer *p, *np, *next; 3643 3644 RB_FOREACH(p, peer_head, &c->peers) { 3645 /* templates are handled specially */ 3646 if (p->template != NULL) 3647 continue; 3648 np = getpeerbyid(nc, p->conf.id); 3649 if (np == NULL) { 3650 p->reconf_action = RECONF_DELETE; 3651 continue; 3652 } 3653 3654 /* peer no longer uses TCP MD5SIG so deconfigure */ 3655 if (p->conf.auth.method == AUTH_MD5SIG && 3656 np->conf.auth.method != AUTH_MD5SIG) 3657 tcp_md5_del_listener(c, p); 3658 else if (np->conf.auth.method == AUTH_MD5SIG) 3659 tcp_md5_add_listener(c, np); 3660 3661 memcpy(&p->conf, &np->conf, sizeof(p->conf)); 3662 RB_REMOVE(peer_head, &nc->peers, np); 3663 free(np); 3664 3665 p->reconf_action = RECONF_KEEP; 3666 3667 /* had demotion, is demoted, demote removed? */ 3668 if (p->demoted && !p->conf.demote_group[0]) 3669 session_demote(p, -1); 3670 3671 /* if session is not open then refresh pfkey data */ 3672 if (p->state < STATE_OPENSENT && !p->template) 3673 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 3674 p->conf.id, 0, -1, NULL, 0); 3675 3676 /* sync the RDE in case we keep the peer */ 3677 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3678 &p->conf, sizeof(struct peer_config)) == -1) 3679 fatalx("imsg_compose error"); 3680 3681 /* apply the config to all clones of a template */ 3682 if (p->conf.template) { 3683 struct peer *xp; 3684 RB_FOREACH(xp, peer_head, &c->peers) { 3685 if (xp->template != p) 3686 continue; 3687 session_template_clone(xp, NULL, xp->conf.id, 3688 xp->conf.remote_as); 3689 if (imsg_rde(IMSG_SESSION_ADD, xp->conf.id, 3690 &xp->conf, sizeof(xp->conf)) == -1) 3691 fatalx("imsg_compose error"); 3692 } 3693 } 3694 } 3695 3696 if (imsg_rde(IMSG_RECONF_DRAIN, 0, NULL, 0) == -1) 3697 fatalx("imsg_compose error"); 3698 3699 /* pfkeys of new peers already loaded by the parent process */ 3700 RB_FOREACH_SAFE(np, peer_head, &nc->peers, next) { 3701 RB_REMOVE(peer_head, &nc->peers, np); 3702 if (RB_INSERT(peer_head, &c->peers, np) != NULL) 3703 fatalx("%s: peer tree is corrupt", __func__); 3704 if (np->conf.auth.method == AUTH_MD5SIG) 3705 tcp_md5_add_listener(c, np); 3706 } 3707 } 3708