1 /* $OpenBSD: session.c,v 1.415 2021/05/16 09:09:11 claudio Exp $ */ 2 3 /* 4 * Copyright (c) 2003, 2004, 2005 Henning Brauer <henning@openbsd.org> 5 * Copyright (c) 2017 Peter van Dijk <peter.van.dijk@powerdns.com> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include <sys/types.h> 21 22 #include <sys/mman.h> 23 #include <sys/socket.h> 24 #include <sys/time.h> 25 #include <sys/resource.h> 26 #include <sys/un.h> 27 #include <netinet/in.h> 28 #include <netinet/ip.h> 29 #include <netinet/tcp.h> 30 #include <arpa/inet.h> 31 #include <limits.h> 32 33 #include <err.h> 34 #include <errno.h> 35 #include <fcntl.h> 36 #include <ifaddrs.h> 37 #include <poll.h> 38 #include <pwd.h> 39 #include <signal.h> 40 #include <stdio.h> 41 #include <stdlib.h> 42 #include <string.h> 43 #include <syslog.h> 44 #include <unistd.h> 45 46 #include "bgpd.h" 47 #include "session.h" 48 #include "log.h" 49 50 #define PFD_PIPE_MAIN 0 51 #define PFD_PIPE_ROUTE 1 52 #define PFD_PIPE_ROUTE_CTL 2 53 #define PFD_SOCK_CTL 3 54 #define PFD_SOCK_RCTL 4 55 #define PFD_LISTENERS_START 5 56 57 void session_sighdlr(int); 58 int setup_listeners(u_int *); 59 void init_peer(struct peer *); 60 void start_timer_holdtime(struct peer *); 61 void start_timer_keepalive(struct peer *); 62 void session_close_connection(struct peer *); 63 void change_state(struct peer *, enum session_state, enum session_events); 64 int session_setup_socket(struct peer *); 65 void session_accept(int); 66 int session_connect(struct peer *); 67 void session_tcp_established(struct peer *); 68 void session_capa_ann_none(struct peer *); 69 int session_capa_add(struct ibuf *, u_int8_t, u_int8_t); 70 int session_capa_add_mp(struct ibuf *, u_int8_t); 71 int session_capa_add_gr(struct peer *, struct ibuf *, u_int8_t); 72 struct bgp_msg *session_newmsg(enum msg_type, u_int16_t); 73 int session_sendmsg(struct bgp_msg *, struct peer *); 74 void session_open(struct peer *); 75 void session_keepalive(struct peer *); 76 void session_update(u_int32_t, void *, size_t); 77 void session_notification(struct peer *, u_int8_t, u_int8_t, void *, 78 ssize_t); 79 void session_rrefresh(struct peer *, u_int8_t); 80 int session_graceful_restart(struct peer *); 81 int session_graceful_stop(struct peer *); 82 int session_dispatch_msg(struct pollfd *, struct peer *); 83 void session_process_msg(struct peer *); 84 int parse_header(struct peer *, u_char *, u_int16_t *, u_int8_t *); 85 int parse_open(struct peer *); 86 int parse_update(struct peer *); 87 int parse_refresh(struct peer *); 88 int parse_notification(struct peer *); 89 int parse_capabilities(struct peer *, u_char *, u_int16_t, u_int32_t *); 90 int capa_neg_calc(struct peer *); 91 void session_dispatch_imsg(struct imsgbuf *, int, u_int *); 92 void session_up(struct peer *); 93 void session_down(struct peer *); 94 int imsg_rde(int, u_int32_t, void *, u_int16_t); 95 void session_demote(struct peer *, int); 96 void merge_peers(struct bgpd_config *, struct bgpd_config *); 97 98 int la_cmp(struct listen_addr *, struct listen_addr *); 99 void session_template_clone(struct peer *, struct sockaddr *, 100 u_int32_t, u_int32_t); 101 int session_match_mask(struct peer *, struct bgpd_addr *); 102 103 static struct bgpd_config *conf, *nconf; 104 static struct imsgbuf *ibuf_rde; 105 static struct imsgbuf *ibuf_rde_ctl; 106 static struct imsgbuf *ibuf_main; 107 108 struct bgpd_sysdep sysdep; 109 volatile sig_atomic_t session_quit; 110 int pending_reconf; 111 int csock = -1, rcsock = -1; 112 u_int peer_cnt; 113 114 struct mrt_head mrthead; 115 time_t pauseaccept; 116 117 static inline int 118 peer_compare(const struct peer *a, const struct peer *b) 119 { 120 return a->conf.id - b->conf.id; 121 } 122 123 RB_GENERATE(peer_head, peer, entry, peer_compare); 124 125 void 126 session_sighdlr(int sig) 127 { 128 switch (sig) { 129 case SIGINT: 130 case SIGTERM: 131 session_quit = 1; 132 break; 133 } 134 } 135 136 int 137 setup_listeners(u_int *la_cnt) 138 { 139 int ttl = 255; 140 struct listen_addr *la; 141 u_int cnt = 0; 142 143 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 144 la->reconf = RECONF_NONE; 145 cnt++; 146 147 if (la->flags & LISTENER_LISTENING) 148 continue; 149 150 if (la->fd == -1) { 151 log_warn("cannot establish listener on %s: invalid fd", 152 log_sockaddr((struct sockaddr *)&la->sa, 153 la->sa_len)); 154 continue; 155 } 156 157 if (tcp_md5_prep_listener(la, &conf->peers) == -1) 158 fatal("tcp_md5_prep_listener"); 159 160 /* set ttl to 255 so that ttl-security works */ 161 if (la->sa.ss_family == AF_INET && setsockopt(la->fd, 162 IPPROTO_IP, IP_TTL, &ttl, sizeof(ttl)) == -1) { 163 log_warn("setup_listeners setsockopt TTL"); 164 continue; 165 } 166 if (la->sa.ss_family == AF_INET6 && setsockopt(la->fd, 167 IPPROTO_IPV6, IPV6_UNICAST_HOPS, &ttl, sizeof(ttl)) == -1) { 168 log_warn("setup_listeners setsockopt hoplimit"); 169 continue; 170 } 171 172 if (listen(la->fd, MAX_BACKLOG)) { 173 close(la->fd); 174 fatal("listen"); 175 } 176 177 la->flags |= LISTENER_LISTENING; 178 179 log_info("listening on %s", 180 log_sockaddr((struct sockaddr *)&la->sa, la->sa_len)); 181 } 182 183 *la_cnt = cnt; 184 185 return (0); 186 } 187 188 void 189 session_main(int debug, int verbose) 190 { 191 int timeout; 192 unsigned int i, j, idx_peers, idx_listeners, idx_mrts; 193 u_int pfd_elms = 0, peer_l_elms = 0, mrt_l_elms = 0; 194 u_int listener_cnt, ctl_cnt, mrt_cnt; 195 u_int new_cnt; 196 struct passwd *pw; 197 struct peer *p, **peer_l = NULL, *next; 198 struct mrt *m, *xm, **mrt_l = NULL; 199 struct pollfd *pfd = NULL; 200 struct listen_addr *la; 201 void *newp; 202 time_t now; 203 short events; 204 205 log_init(debug, LOG_DAEMON); 206 log_setverbose(verbose); 207 208 log_procinit(log_procnames[PROC_SE]); 209 210 if ((pw = getpwnam(BGPD_USER)) == NULL) 211 fatal(NULL); 212 213 if (chroot(pw->pw_dir) == -1) 214 fatal("chroot"); 215 if (chdir("/") == -1) 216 fatal("chdir(\"/\")"); 217 218 setproctitle("session engine"); 219 220 if (setgroups(1, &pw->pw_gid) || 221 setresgid(pw->pw_gid, pw->pw_gid, pw->pw_gid) || 222 setresuid(pw->pw_uid, pw->pw_uid, pw->pw_uid)) 223 fatal("can't drop privileges"); 224 225 if (pledge("stdio inet recvfd", NULL) == -1) 226 fatal("pledge"); 227 228 signal(SIGTERM, session_sighdlr); 229 signal(SIGINT, session_sighdlr); 230 signal(SIGPIPE, SIG_IGN); 231 signal(SIGHUP, SIG_IGN); 232 signal(SIGALRM, SIG_IGN); 233 signal(SIGUSR1, SIG_IGN); 234 235 if ((ibuf_main = malloc(sizeof(struct imsgbuf))) == NULL) 236 fatal(NULL); 237 imsg_init(ibuf_main, 3); 238 239 LIST_INIT(&mrthead); 240 listener_cnt = 0; 241 peer_cnt = 0; 242 ctl_cnt = 0; 243 244 conf = new_config(); 245 log_info("session engine ready"); 246 247 while (session_quit == 0) { 248 /* check for peers to be initialized or deleted */ 249 if (!pending_reconf) { 250 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 251 /* cloned peer that idled out? */ 252 if (p->template && (p->state == STATE_IDLE || 253 p->state == STATE_ACTIVE) && 254 getmonotime() - p->stats.last_updown >= 255 INTERVAL_HOLD_CLONED) 256 p->reconf_action = RECONF_DELETE; 257 258 /* new peer that needs init? */ 259 if (p->state == STATE_NONE) 260 init_peer(p); 261 262 /* reinit due? */ 263 if (p->reconf_action == RECONF_REINIT) { 264 session_stop(p, ERR_CEASE_ADMIN_RESET); 265 if (!p->conf.down) 266 timer_set(&p->timers, 267 Timer_IdleHold, 0); 268 } 269 270 /* deletion due? */ 271 if (p->reconf_action == RECONF_DELETE) { 272 if (p->demoted) 273 session_demote(p, -1); 274 p->conf.demote_group[0] = 0; 275 session_stop(p, ERR_CEASE_PEER_UNCONF); 276 timer_remove_all(&p->timers); 277 tcp_md5_del_listener(conf, p); 278 log_peer_warnx(&p->conf, "removed"); 279 RB_REMOVE(peer_head, &conf->peers, p); 280 free(p); 281 peer_cnt--; 282 continue; 283 } 284 p->reconf_action = RECONF_NONE; 285 } 286 } 287 288 if (peer_cnt > peer_l_elms) { 289 if ((newp = reallocarray(peer_l, peer_cnt, 290 sizeof(struct peer *))) == NULL) { 291 /* panic for now */ 292 log_warn("could not resize peer_l from %u -> %u" 293 " entries", peer_l_elms, peer_cnt); 294 fatalx("exiting"); 295 } 296 peer_l = newp; 297 peer_l_elms = peer_cnt; 298 } 299 300 mrt_cnt = 0; 301 for (m = LIST_FIRST(&mrthead); m != NULL; m = xm) { 302 xm = LIST_NEXT(m, entry); 303 if (m->state == MRT_STATE_REMOVE) { 304 mrt_clean(m); 305 LIST_REMOVE(m, entry); 306 free(m); 307 continue; 308 } 309 if (m->wbuf.queued) 310 mrt_cnt++; 311 } 312 313 if (mrt_cnt > mrt_l_elms) { 314 if ((newp = reallocarray(mrt_l, mrt_cnt, 315 sizeof(struct mrt *))) == NULL) { 316 /* panic for now */ 317 log_warn("could not resize mrt_l from %u -> %u" 318 " entries", mrt_l_elms, mrt_cnt); 319 fatalx("exiting"); 320 } 321 mrt_l = newp; 322 mrt_l_elms = mrt_cnt; 323 } 324 325 new_cnt = PFD_LISTENERS_START + listener_cnt + peer_cnt + 326 ctl_cnt + mrt_cnt; 327 if (new_cnt > pfd_elms) { 328 if ((newp = reallocarray(pfd, new_cnt, 329 sizeof(struct pollfd))) == NULL) { 330 /* panic for now */ 331 log_warn("could not resize pfd from %u -> %u" 332 " entries", pfd_elms, new_cnt); 333 fatalx("exiting"); 334 } 335 pfd = newp; 336 pfd_elms = new_cnt; 337 } 338 339 bzero(pfd, sizeof(struct pollfd) * pfd_elms); 340 341 set_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main); 342 set_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde); 343 set_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl); 344 345 if (pauseaccept == 0) { 346 pfd[PFD_SOCK_CTL].fd = csock; 347 pfd[PFD_SOCK_CTL].events = POLLIN; 348 pfd[PFD_SOCK_RCTL].fd = rcsock; 349 pfd[PFD_SOCK_RCTL].events = POLLIN; 350 } else { 351 pfd[PFD_SOCK_CTL].fd = -1; 352 pfd[PFD_SOCK_RCTL].fd = -1; 353 } 354 355 i = PFD_LISTENERS_START; 356 TAILQ_FOREACH(la, conf->listen_addrs, entry) { 357 if (pauseaccept == 0) { 358 pfd[i].fd = la->fd; 359 pfd[i].events = POLLIN; 360 } else 361 pfd[i].fd = -1; 362 i++; 363 } 364 idx_listeners = i; 365 timeout = 240; /* loop every 240s at least */ 366 367 now = getmonotime(); 368 RB_FOREACH(p, peer_head, &conf->peers) { 369 time_t nextaction; 370 struct timer *pt; 371 372 /* check timers */ 373 if ((pt = timer_nextisdue(&p->timers, now)) != NULL) { 374 switch (pt->type) { 375 case Timer_Hold: 376 bgp_fsm(p, EVNT_TIMER_HOLDTIME); 377 break; 378 case Timer_SendHold: 379 bgp_fsm(p, EVNT_TIMER_SENDHOLD); 380 break; 381 case Timer_ConnectRetry: 382 bgp_fsm(p, EVNT_TIMER_CONNRETRY); 383 break; 384 case Timer_Keepalive: 385 bgp_fsm(p, EVNT_TIMER_KEEPALIVE); 386 break; 387 case Timer_IdleHold: 388 bgp_fsm(p, EVNT_START); 389 break; 390 case Timer_IdleHoldReset: 391 p->IdleHoldTime = 392 INTERVAL_IDLE_HOLD_INITIAL; 393 p->errcnt = 0; 394 timer_stop(&p->timers, 395 Timer_IdleHoldReset); 396 break; 397 case Timer_CarpUndemote: 398 timer_stop(&p->timers, 399 Timer_CarpUndemote); 400 if (p->demoted && 401 p->state == STATE_ESTABLISHED) 402 session_demote(p, -1); 403 break; 404 case Timer_RestartTimeout: 405 timer_stop(&p->timers, 406 Timer_RestartTimeout); 407 session_graceful_stop(p); 408 break; 409 default: 410 fatalx("King Bula lost in time"); 411 } 412 } 413 if ((nextaction = timer_nextduein(&p->timers, 414 now)) != -1 && nextaction < timeout) 415 timeout = nextaction; 416 417 /* are we waiting for a write? */ 418 events = POLLIN; 419 if (p->wbuf.queued > 0 || p->state == STATE_CONNECT) 420 events |= POLLOUT; 421 /* is there still work to do? */ 422 if (p->rpending && p->rbuf && p->rbuf->wpos) 423 timeout = 0; 424 425 /* poll events */ 426 if (p->fd != -1 && events != 0) { 427 pfd[i].fd = p->fd; 428 pfd[i].events = events; 429 peer_l[i - idx_listeners] = p; 430 i++; 431 } 432 } 433 434 idx_peers = i; 435 436 LIST_FOREACH(m, &mrthead, entry) 437 if (m->wbuf.queued) { 438 pfd[i].fd = m->wbuf.fd; 439 pfd[i].events = POLLOUT; 440 mrt_l[i - idx_peers] = m; 441 i++; 442 } 443 444 idx_mrts = i; 445 446 i += control_fill_pfds(pfd + i, pfd_elms -i); 447 448 if (i > pfd_elms) 449 fatalx("poll pfd overflow"); 450 451 if (pauseaccept && timeout > 1) 452 timeout = 1; 453 if (timeout < 0) 454 timeout = 0; 455 if (poll(pfd, i, timeout * 1000) == -1) 456 if (errno != EINTR) 457 fatal("poll error"); 458 459 /* 460 * If we previously saw fd exhaustion, we stop accept() 461 * for 1 second to throttle the accept() loop. 462 */ 463 if (pauseaccept && getmonotime() > pauseaccept + 1) 464 pauseaccept = 0; 465 466 if (handle_pollfd(&pfd[PFD_PIPE_MAIN], ibuf_main) == -1) { 467 log_warnx("SE: Lost connection to parent"); 468 session_quit = 1; 469 continue; 470 } else 471 session_dispatch_imsg(ibuf_main, PFD_PIPE_MAIN, 472 &listener_cnt); 473 474 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE], ibuf_rde) == -1) { 475 log_warnx("SE: Lost connection to RDE"); 476 msgbuf_clear(&ibuf_rde->w); 477 free(ibuf_rde); 478 ibuf_rde = NULL; 479 } else 480 session_dispatch_imsg(ibuf_rde, PFD_PIPE_ROUTE, 481 &listener_cnt); 482 483 if (handle_pollfd(&pfd[PFD_PIPE_ROUTE_CTL], ibuf_rde_ctl) == 484 -1) { 485 log_warnx("SE: Lost connection to RDE control"); 486 msgbuf_clear(&ibuf_rde_ctl->w); 487 free(ibuf_rde_ctl); 488 ibuf_rde_ctl = NULL; 489 } else 490 session_dispatch_imsg(ibuf_rde_ctl, PFD_PIPE_ROUTE_CTL, 491 &listener_cnt); 492 493 if (pfd[PFD_SOCK_CTL].revents & POLLIN) 494 ctl_cnt += control_accept(csock, 0); 495 496 if (pfd[PFD_SOCK_RCTL].revents & POLLIN) 497 ctl_cnt += control_accept(rcsock, 1); 498 499 for (j = PFD_LISTENERS_START; j < idx_listeners; j++) 500 if (pfd[j].revents & POLLIN) 501 session_accept(pfd[j].fd); 502 503 for (; j < idx_peers; j++) 504 session_dispatch_msg(&pfd[j], 505 peer_l[j - idx_listeners]); 506 507 RB_FOREACH(p, peer_head, &conf->peers) 508 if (p->rbuf && p->rbuf->wpos) 509 session_process_msg(p); 510 511 for (; j < idx_mrts; j++) 512 if (pfd[j].revents & POLLOUT) 513 mrt_write(mrt_l[j - idx_peers]); 514 515 for (; j < i; j++) 516 ctl_cnt -= control_dispatch_msg(&pfd[j], &conf->peers); 517 } 518 519 RB_FOREACH_SAFE(p, peer_head, &conf->peers, next) { 520 RB_REMOVE(peer_head, &conf->peers, p); 521 strlcpy(p->conf.reason, 522 "bgpd shutting down", 523 sizeof(p->conf.reason)); 524 session_stop(p, ERR_CEASE_ADMIN_DOWN); 525 timer_remove_all(&p->timers); 526 free(p); 527 } 528 529 while ((m = LIST_FIRST(&mrthead)) != NULL) { 530 mrt_clean(m); 531 LIST_REMOVE(m, entry); 532 free(m); 533 } 534 535 free_config(conf); 536 free(peer_l); 537 free(mrt_l); 538 free(pfd); 539 540 /* close pipes */ 541 if (ibuf_rde) { 542 msgbuf_write(&ibuf_rde->w); 543 msgbuf_clear(&ibuf_rde->w); 544 close(ibuf_rde->fd); 545 free(ibuf_rde); 546 } 547 if (ibuf_rde_ctl) { 548 msgbuf_clear(&ibuf_rde_ctl->w); 549 close(ibuf_rde_ctl->fd); 550 free(ibuf_rde_ctl); 551 } 552 msgbuf_write(&ibuf_main->w); 553 msgbuf_clear(&ibuf_main->w); 554 close(ibuf_main->fd); 555 free(ibuf_main); 556 557 control_shutdown(csock); 558 control_shutdown(rcsock); 559 log_info("session engine exiting"); 560 exit(0); 561 } 562 563 void 564 init_peer(struct peer *p) 565 { 566 TAILQ_INIT(&p->timers); 567 p->fd = p->wbuf.fd = -1; 568 569 if (p->conf.if_depend[0]) 570 imsg_compose(ibuf_main, IMSG_IFINFO, 0, 0, -1, 571 p->conf.if_depend, sizeof(p->conf.if_depend)); 572 else 573 p->depend_ok = 1; 574 575 peer_cnt++; 576 577 change_state(p, STATE_IDLE, EVNT_NONE); 578 if (p->conf.down) 579 timer_stop(&p->timers, Timer_IdleHold); /* no autostart */ 580 else 581 timer_set(&p->timers, Timer_IdleHold, 0); /* start ASAP */ 582 583 /* 584 * on startup, demote if requested. 585 * do not handle new peers. they must reach ESTABLISHED beforehands. 586 * peers added at runtime have reconf_action set to RECONF_REINIT. 587 */ 588 if (p->reconf_action != RECONF_REINIT && p->conf.demote_group[0]) 589 session_demote(p, +1); 590 } 591 592 void 593 bgp_fsm(struct peer *peer, enum session_events event) 594 { 595 switch (peer->state) { 596 case STATE_NONE: 597 /* nothing */ 598 break; 599 case STATE_IDLE: 600 switch (event) { 601 case EVNT_START: 602 timer_stop(&peer->timers, Timer_Hold); 603 timer_stop(&peer->timers, Timer_SendHold); 604 timer_stop(&peer->timers, Timer_Keepalive); 605 timer_stop(&peer->timers, Timer_IdleHold); 606 607 /* allocate read buffer */ 608 peer->rbuf = calloc(1, sizeof(struct ibuf_read)); 609 if (peer->rbuf == NULL) 610 fatal(NULL); 611 612 /* init write buffer */ 613 msgbuf_init(&peer->wbuf); 614 615 peer->stats.last_sent_errcode = 0; 616 peer->stats.last_sent_suberr = 0; 617 peer->stats.last_rcvd_errcode = 0; 618 peer->stats.last_rcvd_suberr = 0; 619 620 if (!peer->depend_ok) 621 timer_stop(&peer->timers, Timer_ConnectRetry); 622 else if (peer->passive || peer->conf.passive || 623 peer->conf.template) { 624 change_state(peer, STATE_ACTIVE, event); 625 timer_stop(&peer->timers, Timer_ConnectRetry); 626 } else { 627 change_state(peer, STATE_CONNECT, event); 628 timer_set(&peer->timers, Timer_ConnectRetry, 629 conf->connectretry); 630 session_connect(peer); 631 } 632 peer->passive = 0; 633 break; 634 default: 635 /* ignore */ 636 break; 637 } 638 break; 639 case STATE_CONNECT: 640 switch (event) { 641 case EVNT_START: 642 /* ignore */ 643 break; 644 case EVNT_CON_OPEN: 645 session_tcp_established(peer); 646 session_open(peer); 647 timer_stop(&peer->timers, Timer_ConnectRetry); 648 peer->holdtime = INTERVAL_HOLD_INITIAL; 649 start_timer_holdtime(peer); 650 change_state(peer, STATE_OPENSENT, event); 651 break; 652 case EVNT_CON_OPENFAIL: 653 timer_set(&peer->timers, Timer_ConnectRetry, 654 conf->connectretry); 655 session_close_connection(peer); 656 change_state(peer, STATE_ACTIVE, event); 657 break; 658 case EVNT_TIMER_CONNRETRY: 659 timer_set(&peer->timers, Timer_ConnectRetry, 660 conf->connectretry); 661 session_connect(peer); 662 break; 663 default: 664 change_state(peer, STATE_IDLE, event); 665 break; 666 } 667 break; 668 case STATE_ACTIVE: 669 switch (event) { 670 case EVNT_START: 671 /* ignore */ 672 break; 673 case EVNT_CON_OPEN: 674 session_tcp_established(peer); 675 session_open(peer); 676 timer_stop(&peer->timers, Timer_ConnectRetry); 677 peer->holdtime = INTERVAL_HOLD_INITIAL; 678 start_timer_holdtime(peer); 679 change_state(peer, STATE_OPENSENT, event); 680 break; 681 case EVNT_CON_OPENFAIL: 682 timer_set(&peer->timers, Timer_ConnectRetry, 683 conf->connectretry); 684 session_close_connection(peer); 685 change_state(peer, STATE_ACTIVE, event); 686 break; 687 case EVNT_TIMER_CONNRETRY: 688 timer_set(&peer->timers, Timer_ConnectRetry, 689 peer->holdtime); 690 change_state(peer, STATE_CONNECT, event); 691 session_connect(peer); 692 break; 693 default: 694 change_state(peer, STATE_IDLE, event); 695 break; 696 } 697 break; 698 case STATE_OPENSENT: 699 switch (event) { 700 case EVNT_START: 701 /* ignore */ 702 break; 703 case EVNT_STOP: 704 change_state(peer, STATE_IDLE, event); 705 break; 706 case EVNT_CON_CLOSED: 707 session_close_connection(peer); 708 timer_set(&peer->timers, Timer_ConnectRetry, 709 conf->connectretry); 710 change_state(peer, STATE_ACTIVE, event); 711 break; 712 case EVNT_CON_FATAL: 713 change_state(peer, STATE_IDLE, event); 714 break; 715 case EVNT_TIMER_HOLDTIME: 716 case EVNT_TIMER_SENDHOLD: 717 session_notification(peer, ERR_HOLDTIMEREXPIRED, 718 0, NULL, 0); 719 change_state(peer, STATE_IDLE, event); 720 break; 721 case EVNT_RCVD_OPEN: 722 /* parse_open calls change_state itself on failure */ 723 if (parse_open(peer)) 724 break; 725 session_keepalive(peer); 726 change_state(peer, STATE_OPENCONFIRM, event); 727 break; 728 case EVNT_RCVD_NOTIFICATION: 729 if (parse_notification(peer)) { 730 change_state(peer, STATE_IDLE, event); 731 /* don't punish, capa negotiation */ 732 timer_set(&peer->timers, Timer_IdleHold, 0); 733 peer->IdleHoldTime /= 2; 734 } else 735 change_state(peer, STATE_IDLE, event); 736 break; 737 default: 738 session_notification(peer, 739 ERR_FSM, ERR_FSM_UNEX_OPENSENT, NULL, 0); 740 change_state(peer, STATE_IDLE, event); 741 break; 742 } 743 break; 744 case STATE_OPENCONFIRM: 745 switch (event) { 746 case EVNT_START: 747 /* ignore */ 748 break; 749 case EVNT_STOP: 750 change_state(peer, STATE_IDLE, event); 751 break; 752 case EVNT_CON_CLOSED: 753 case EVNT_CON_FATAL: 754 change_state(peer, STATE_IDLE, event); 755 break; 756 case EVNT_TIMER_HOLDTIME: 757 case EVNT_TIMER_SENDHOLD: 758 session_notification(peer, ERR_HOLDTIMEREXPIRED, 759 0, NULL, 0); 760 change_state(peer, STATE_IDLE, event); 761 break; 762 case EVNT_TIMER_KEEPALIVE: 763 session_keepalive(peer); 764 break; 765 case EVNT_RCVD_KEEPALIVE: 766 start_timer_holdtime(peer); 767 change_state(peer, STATE_ESTABLISHED, event); 768 break; 769 case EVNT_RCVD_NOTIFICATION: 770 parse_notification(peer); 771 change_state(peer, STATE_IDLE, event); 772 break; 773 default: 774 session_notification(peer, 775 ERR_FSM, ERR_FSM_UNEX_OPENCONFIRM, NULL, 0); 776 change_state(peer, STATE_IDLE, event); 777 break; 778 } 779 break; 780 case STATE_ESTABLISHED: 781 switch (event) { 782 case EVNT_START: 783 /* ignore */ 784 break; 785 case EVNT_STOP: 786 change_state(peer, STATE_IDLE, event); 787 break; 788 case EVNT_CON_CLOSED: 789 case EVNT_CON_FATAL: 790 change_state(peer, STATE_IDLE, event); 791 break; 792 case EVNT_TIMER_HOLDTIME: 793 case EVNT_TIMER_SENDHOLD: 794 session_notification(peer, ERR_HOLDTIMEREXPIRED, 795 0, NULL, 0); 796 change_state(peer, STATE_IDLE, event); 797 break; 798 case EVNT_TIMER_KEEPALIVE: 799 session_keepalive(peer); 800 break; 801 case EVNT_RCVD_KEEPALIVE: 802 start_timer_holdtime(peer); 803 break; 804 case EVNT_RCVD_UPDATE: 805 start_timer_holdtime(peer); 806 if (parse_update(peer)) 807 change_state(peer, STATE_IDLE, event); 808 else 809 start_timer_holdtime(peer); 810 break; 811 case EVNT_RCVD_NOTIFICATION: 812 parse_notification(peer); 813 change_state(peer, STATE_IDLE, event); 814 break; 815 default: 816 session_notification(peer, 817 ERR_FSM, ERR_FSM_UNEX_ESTABLISHED, NULL, 0); 818 change_state(peer, STATE_IDLE, event); 819 break; 820 } 821 break; 822 } 823 } 824 825 void 826 start_timer_holdtime(struct peer *peer) 827 { 828 if (peer->holdtime > 0) 829 timer_set(&peer->timers, Timer_Hold, peer->holdtime); 830 else 831 timer_stop(&peer->timers, Timer_Hold); 832 } 833 834 void 835 start_timer_keepalive(struct peer *peer) 836 { 837 if (peer->holdtime > 0) 838 timer_set(&peer->timers, Timer_Keepalive, peer->holdtime / 3); 839 else 840 timer_stop(&peer->timers, Timer_Keepalive); 841 } 842 843 void 844 session_close_connection(struct peer *peer) 845 { 846 if (peer->fd != -1) { 847 close(peer->fd); 848 pauseaccept = 0; 849 } 850 peer->fd = peer->wbuf.fd = -1; 851 } 852 853 void 854 change_state(struct peer *peer, enum session_state state, 855 enum session_events event) 856 { 857 struct mrt *mrt; 858 859 switch (state) { 860 case STATE_IDLE: 861 /* carp demotion first. new peers handled in init_peer */ 862 if (peer->state == STATE_ESTABLISHED && 863 peer->conf.demote_group[0] && !peer->demoted) 864 session_demote(peer, +1); 865 866 /* 867 * try to write out what's buffered (maybe a notification), 868 * don't bother if it fails 869 */ 870 if (peer->state >= STATE_OPENSENT && peer->wbuf.queued) 871 msgbuf_write(&peer->wbuf); 872 873 /* 874 * we must start the timer for the next EVNT_START 875 * if we are coming here due to an error and the 876 * session was not established successfully before, the 877 * starttimerinterval needs to be exponentially increased 878 */ 879 if (peer->IdleHoldTime == 0) 880 peer->IdleHoldTime = INTERVAL_IDLE_HOLD_INITIAL; 881 peer->holdtime = INTERVAL_HOLD_INITIAL; 882 timer_stop(&peer->timers, Timer_ConnectRetry); 883 timer_stop(&peer->timers, Timer_Keepalive); 884 timer_stop(&peer->timers, Timer_Hold); 885 timer_stop(&peer->timers, Timer_SendHold); 886 timer_stop(&peer->timers, Timer_IdleHold); 887 timer_stop(&peer->timers, Timer_IdleHoldReset); 888 session_close_connection(peer); 889 msgbuf_clear(&peer->wbuf); 890 free(peer->rbuf); 891 peer->rbuf = NULL; 892 peer->rpending = 0; 893 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 894 if (!peer->template) 895 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 896 peer->conf.id, 0, -1, NULL, 0); 897 898 if (event != EVNT_STOP) { 899 timer_set(&peer->timers, Timer_IdleHold, 900 peer->IdleHoldTime); 901 if (event != EVNT_NONE && 902 peer->IdleHoldTime < MAX_IDLE_HOLD/2) 903 peer->IdleHoldTime *= 2; 904 } 905 if (peer->state == STATE_ESTABLISHED) { 906 if (peer->capa.neg.grestart.restart == 2 && 907 (event == EVNT_CON_CLOSED || 908 event == EVNT_CON_FATAL)) { 909 /* don't punish graceful restart */ 910 timer_set(&peer->timers, Timer_IdleHold, 0); 911 peer->IdleHoldTime /= 2; 912 session_graceful_restart(peer); 913 } else 914 session_down(peer); 915 } 916 if (peer->state == STATE_NONE || 917 peer->state == STATE_ESTABLISHED) { 918 /* initialize capability negotiation structures */ 919 memcpy(&peer->capa.ann, &peer->conf.capabilities, 920 sizeof(peer->capa.ann)); 921 if (!peer->conf.announce_capa) 922 session_capa_ann_none(peer); 923 } 924 break; 925 case STATE_CONNECT: 926 if (peer->state == STATE_ESTABLISHED && 927 peer->capa.neg.grestart.restart == 2) { 928 /* do the graceful restart dance */ 929 session_graceful_restart(peer); 930 peer->holdtime = INTERVAL_HOLD_INITIAL; 931 timer_stop(&peer->timers, Timer_ConnectRetry); 932 timer_stop(&peer->timers, Timer_Keepalive); 933 timer_stop(&peer->timers, Timer_Hold); 934 timer_stop(&peer->timers, Timer_SendHold); 935 timer_stop(&peer->timers, Timer_IdleHold); 936 timer_stop(&peer->timers, Timer_IdleHoldReset); 937 session_close_connection(peer); 938 msgbuf_clear(&peer->wbuf); 939 bzero(&peer->capa.peer, sizeof(peer->capa.peer)); 940 } 941 break; 942 case STATE_ACTIVE: 943 if (!peer->template) 944 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 945 peer->conf.id, 0, -1, NULL, 0); 946 break; 947 case STATE_OPENSENT: 948 break; 949 case STATE_OPENCONFIRM: 950 break; 951 case STATE_ESTABLISHED: 952 timer_set(&peer->timers, Timer_IdleHoldReset, 953 peer->IdleHoldTime); 954 if (peer->demoted) 955 timer_set(&peer->timers, Timer_CarpUndemote, 956 INTERVAL_HOLD_DEMOTED); 957 session_up(peer); 958 break; 959 default: /* something seriously fucked */ 960 break; 961 } 962 963 log_statechange(peer, state, event); 964 LIST_FOREACH(mrt, &mrthead, entry) { 965 if (!(mrt->type == MRT_ALL_IN || mrt->type == MRT_ALL_OUT)) 966 continue; 967 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 968 mrt->peer_id == peer->conf.id || (mrt->group_id != 0 && 969 mrt->group_id == peer->conf.groupid)) 970 mrt_dump_state(mrt, peer->state, state, peer); 971 } 972 peer->prev_state = peer->state; 973 peer->state = state; 974 } 975 976 void 977 session_accept(int listenfd) 978 { 979 int connfd; 980 socklen_t len; 981 struct sockaddr_storage cliaddr; 982 struct peer *p = NULL; 983 984 len = sizeof(cliaddr); 985 if ((connfd = accept4(listenfd, 986 (struct sockaddr *)&cliaddr, &len, 987 SOCK_CLOEXEC | SOCK_NONBLOCK)) == -1) { 988 if (errno == ENFILE || errno == EMFILE) 989 pauseaccept = getmonotime(); 990 else if (errno != EWOULDBLOCK && errno != EINTR && 991 errno != ECONNABORTED) 992 log_warn("accept"); 993 return; 994 } 995 996 p = getpeerbyip(conf, (struct sockaddr *)&cliaddr); 997 998 if (p != NULL && p->state == STATE_IDLE && p->errcnt < 2) { 999 if (timer_running(&p->timers, Timer_IdleHold, NULL)) { 1000 /* fast reconnect after clear */ 1001 p->passive = 1; 1002 bgp_fsm(p, EVNT_START); 1003 } 1004 } 1005 1006 if (p != NULL && 1007 (p->state == STATE_CONNECT || p->state == STATE_ACTIVE)) { 1008 if (p->fd != -1) { 1009 if (p->state == STATE_CONNECT) 1010 session_close_connection(p); 1011 else { 1012 close(connfd); 1013 return; 1014 } 1015 } 1016 1017 open: 1018 if (p->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1019 log_peer_warnx(&p->conf, 1020 "ipsec or md5sig configured but not available"); 1021 close(connfd); 1022 return; 1023 } 1024 1025 if (tcp_md5_check(connfd, p) == -1) { 1026 close(connfd); 1027 return; 1028 } 1029 p->fd = p->wbuf.fd = connfd; 1030 if (session_setup_socket(p)) { 1031 close(connfd); 1032 return; 1033 } 1034 bgp_fsm(p, EVNT_CON_OPEN); 1035 return; 1036 } else if (p != NULL && p->state == STATE_ESTABLISHED && 1037 p->capa.neg.grestart.restart == 2) { 1038 /* first do the graceful restart dance */ 1039 change_state(p, STATE_CONNECT, EVNT_CON_CLOSED); 1040 /* then do part of the open dance */ 1041 goto open; 1042 } else { 1043 log_conn_attempt(p, (struct sockaddr *)&cliaddr, len); 1044 close(connfd); 1045 } 1046 } 1047 1048 int 1049 session_connect(struct peer *peer) 1050 { 1051 struct sockaddr *sa; 1052 struct bgpd_addr *bind_addr = NULL; 1053 socklen_t sa_len; 1054 1055 /* 1056 * we do not need the overcomplicated collision detection RFC 1771 1057 * describes; we simply make sure there is only ever one concurrent 1058 * tcp connection per peer. 1059 */ 1060 if (peer->fd != -1) 1061 return (-1); 1062 1063 if ((peer->fd = socket(aid2af(peer->conf.remote_addr.aid), 1064 SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, IPPROTO_TCP)) == -1) { 1065 log_peer_warn(&peer->conf, "session_connect socket"); 1066 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1067 return (-1); 1068 } 1069 1070 if (peer->conf.auth.method != AUTH_NONE && sysdep.no_pfkey) { 1071 log_peer_warnx(&peer->conf, 1072 "ipsec or md5sig configured but not available"); 1073 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1074 return (-1); 1075 } 1076 1077 tcp_md5_set(peer->fd, peer); 1078 peer->wbuf.fd = peer->fd; 1079 1080 /* if local-address is set we need to bind() */ 1081 switch (peer->conf.remote_addr.aid) { 1082 case AID_INET: 1083 bind_addr = &peer->conf.local_addr_v4; 1084 break; 1085 case AID_INET6: 1086 bind_addr = &peer->conf.local_addr_v6; 1087 break; 1088 } 1089 if ((sa = addr2sa(bind_addr, 0, &sa_len)) != NULL) { 1090 if (bind(peer->fd, sa, sa_len) == -1) { 1091 log_peer_warn(&peer->conf, "session_connect bind"); 1092 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1093 return (-1); 1094 } 1095 } 1096 1097 if (session_setup_socket(peer)) { 1098 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1099 return (-1); 1100 } 1101 1102 sa = addr2sa(&peer->conf.remote_addr, BGP_PORT, &sa_len); 1103 if (connect(peer->fd, sa, sa_len) == -1) { 1104 if (errno != EINPROGRESS) { 1105 if (errno != peer->lasterr) 1106 log_peer_warn(&peer->conf, "connect"); 1107 peer->lasterr = errno; 1108 bgp_fsm(peer, EVNT_CON_OPENFAIL); 1109 return (-1); 1110 } 1111 } else 1112 bgp_fsm(peer, EVNT_CON_OPEN); 1113 1114 return (0); 1115 } 1116 1117 int 1118 session_setup_socket(struct peer *p) 1119 { 1120 int ttl = p->conf.distance; 1121 int pre = IPTOS_PREC_INTERNETCONTROL; 1122 int nodelay = 1; 1123 int bsize; 1124 1125 switch (p->conf.remote_addr.aid) { 1126 case AID_INET: 1127 /* set precedence, see RFC 1771 appendix 5 */ 1128 if (setsockopt(p->fd, IPPROTO_IP, IP_TOS, &pre, sizeof(pre)) == 1129 -1) { 1130 log_peer_warn(&p->conf, 1131 "session_setup_socket setsockopt TOS"); 1132 return (-1); 1133 } 1134 1135 if (p->conf.ebgp) { 1136 /* 1137 * set TTL to foreign router's distance 1138 * 1=direct n=multihop with ttlsec, we always use 255 1139 */ 1140 if (p->conf.ttlsec) { 1141 ttl = 256 - p->conf.distance; 1142 if (setsockopt(p->fd, IPPROTO_IP, IP_MINTTL, 1143 &ttl, sizeof(ttl)) == -1) { 1144 log_peer_warn(&p->conf, 1145 "session_setup_socket: " 1146 "setsockopt MINTTL"); 1147 return (-1); 1148 } 1149 ttl = 255; 1150 } 1151 1152 if (setsockopt(p->fd, IPPROTO_IP, IP_TTL, &ttl, 1153 sizeof(ttl)) == -1) { 1154 log_peer_warn(&p->conf, 1155 "session_setup_socket setsockopt TTL"); 1156 return (-1); 1157 } 1158 } 1159 break; 1160 case AID_INET6: 1161 if (p->conf.ebgp) { 1162 /* 1163 * set hoplimit to foreign router's distance 1164 * 1=direct n=multihop with ttlsec, we always use 255 1165 */ 1166 if (p->conf.ttlsec) { 1167 ttl = 256 - p->conf.distance; 1168 if (setsockopt(p->fd, IPPROTO_IPV6, 1169 IPV6_MINHOPCOUNT, &ttl, sizeof(ttl)) 1170 == -1) { 1171 log_peer_warn(&p->conf, 1172 "session_setup_socket: " 1173 "setsockopt MINHOPCOUNT"); 1174 return (-1); 1175 } 1176 ttl = 255; 1177 } 1178 if (setsockopt(p->fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, 1179 &ttl, sizeof(ttl)) == -1) { 1180 log_peer_warn(&p->conf, 1181 "session_setup_socket setsockopt hoplimit"); 1182 return (-1); 1183 } 1184 } 1185 break; 1186 } 1187 1188 /* set TCP_NODELAY */ 1189 if (setsockopt(p->fd, IPPROTO_TCP, TCP_NODELAY, &nodelay, 1190 sizeof(nodelay)) == -1) { 1191 log_peer_warn(&p->conf, 1192 "session_setup_socket setsockopt TCP_NODELAY"); 1193 return (-1); 1194 } 1195 1196 /* only increase bufsize (and thus window) if md5 or ipsec is in use */ 1197 if (p->conf.auth.method != AUTH_NONE) { 1198 /* try to increase bufsize. no biggie if it fails */ 1199 bsize = 65535; 1200 while (bsize > 8192 && 1201 setsockopt(p->fd, SOL_SOCKET, SO_RCVBUF, &bsize, 1202 sizeof(bsize)) == -1 && errno != EINVAL) 1203 bsize /= 2; 1204 bsize = 65535; 1205 while (bsize > 8192 && 1206 setsockopt(p->fd, SOL_SOCKET, SO_SNDBUF, &bsize, 1207 sizeof(bsize)) == -1 && errno != EINVAL) 1208 bsize /= 2; 1209 } 1210 1211 return (0); 1212 } 1213 1214 /* compare two sockaddrs by converting them into bgpd_addr */ 1215 static int 1216 sa_cmp(struct sockaddr *a, struct sockaddr *b) 1217 { 1218 struct bgpd_addr ba, bb; 1219 1220 sa2addr(a, &ba, NULL); 1221 sa2addr(b, &bb, NULL); 1222 1223 return (memcmp(&ba, &bb, sizeof(ba)) == 0); 1224 } 1225 1226 static void 1227 get_alternate_addr(struct sockaddr *sa, struct bgpd_addr *alt) 1228 { 1229 struct ifaddrs *ifap, *ifa, *match; 1230 1231 if (getifaddrs(&ifap) == -1) 1232 fatal("getifaddrs"); 1233 1234 for (match = ifap; match != NULL; match = match->ifa_next) 1235 if (match->ifa_addr != NULL && 1236 sa_cmp(sa, match->ifa_addr) == 0) 1237 break; 1238 1239 if (match == NULL) { 1240 log_warnx("%s: local address not found", __func__); 1241 return; 1242 } 1243 1244 switch (sa->sa_family) { 1245 case AF_INET6: 1246 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1247 if (ifa->ifa_addr != NULL && 1248 ifa->ifa_addr->sa_family == AF_INET && 1249 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1250 sa2addr(ifa->ifa_addr, alt, NULL); 1251 break; 1252 } 1253 } 1254 break; 1255 case AF_INET: 1256 for (ifa = ifap; ifa != NULL; ifa = ifa->ifa_next) { 1257 if (ifa->ifa_addr != NULL && 1258 ifa->ifa_addr->sa_family == AF_INET6 && 1259 strcmp(ifa->ifa_name, match->ifa_name) == 0) { 1260 struct sockaddr_in6 *s = 1261 (struct sockaddr_in6 *)ifa->ifa_addr; 1262 1263 /* only accept global scope addresses */ 1264 if (IN6_IS_ADDR_LINKLOCAL(&s->sin6_addr) || 1265 IN6_IS_ADDR_SITELOCAL(&s->sin6_addr)) 1266 continue; 1267 sa2addr(ifa->ifa_addr, alt, NULL); 1268 break; 1269 } 1270 } 1271 break; 1272 default: 1273 log_warnx("%s: unsupported address family %d", __func__, 1274 sa->sa_family); 1275 break; 1276 } 1277 1278 freeifaddrs(ifap); 1279 } 1280 1281 void 1282 session_tcp_established(struct peer *peer) 1283 { 1284 struct sockaddr_storage ss; 1285 socklen_t len; 1286 1287 len = sizeof(ss); 1288 if (getsockname(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1289 log_warn("getsockname"); 1290 sa2addr((struct sockaddr *)&ss, &peer->local, &peer->local_port); 1291 get_alternate_addr((struct sockaddr *)&ss, &peer->local_alt); 1292 len = sizeof(ss); 1293 if (getpeername(peer->fd, (struct sockaddr *)&ss, &len) == -1) 1294 log_warn("getpeername"); 1295 sa2addr((struct sockaddr *)&ss, &peer->remote, &peer->remote_port); 1296 } 1297 1298 void 1299 session_capa_ann_none(struct peer *peer) 1300 { 1301 bzero(&peer->capa.ann, sizeof(peer->capa.ann)); 1302 } 1303 1304 int 1305 session_capa_add(struct ibuf *opb, u_int8_t capa_code, u_int8_t capa_len) 1306 { 1307 int errs = 0; 1308 1309 errs += ibuf_add(opb, &capa_code, sizeof(capa_code)); 1310 errs += ibuf_add(opb, &capa_len, sizeof(capa_len)); 1311 return (errs); 1312 } 1313 1314 int 1315 session_capa_add_mp(struct ibuf *buf, u_int8_t aid) 1316 { 1317 u_int8_t safi, pad = 0; 1318 u_int16_t afi; 1319 int errs = 0; 1320 1321 if (aid2afi(aid, &afi, &safi) == -1) 1322 fatalx("session_capa_add_mp: bad afi/safi pair"); 1323 afi = htons(afi); 1324 errs += ibuf_add(buf, &afi, sizeof(afi)); 1325 errs += ibuf_add(buf, &pad, sizeof(pad)); 1326 errs += ibuf_add(buf, &safi, sizeof(safi)); 1327 1328 return (errs); 1329 } 1330 1331 int 1332 session_capa_add_gr(struct peer *p, struct ibuf *b, u_int8_t aid) 1333 { 1334 u_int errs = 0; 1335 u_int16_t afi; 1336 u_int8_t flags, safi; 1337 1338 if (aid2afi(aid, &afi, &safi)) { 1339 log_warn("session_capa_add_gr: bad AID"); 1340 return (1); 1341 } 1342 if (p->capa.neg.grestart.flags[aid] & CAPA_GR_RESTARTING) 1343 flags = CAPA_GR_F_FLAG; 1344 else 1345 flags = 0; 1346 1347 afi = htons(afi); 1348 errs += ibuf_add(b, &afi, sizeof(afi)); 1349 errs += ibuf_add(b, &safi, sizeof(safi)); 1350 errs += ibuf_add(b, &flags, sizeof(flags)); 1351 1352 return (errs); 1353 } 1354 1355 struct bgp_msg * 1356 session_newmsg(enum msg_type msgtype, u_int16_t len) 1357 { 1358 struct bgp_msg *msg; 1359 struct msg_header hdr; 1360 struct ibuf *buf; 1361 int errs = 0; 1362 1363 memset(&hdr.marker, 0xff, sizeof(hdr.marker)); 1364 hdr.len = htons(len); 1365 hdr.type = msgtype; 1366 1367 if ((buf = ibuf_open(len)) == NULL) 1368 return (NULL); 1369 1370 errs += ibuf_add(buf, &hdr.marker, sizeof(hdr.marker)); 1371 errs += ibuf_add(buf, &hdr.len, sizeof(hdr.len)); 1372 errs += ibuf_add(buf, &hdr.type, sizeof(hdr.type)); 1373 1374 if (errs || (msg = calloc(1, sizeof(*msg))) == NULL) { 1375 ibuf_free(buf); 1376 return (NULL); 1377 } 1378 1379 msg->buf = buf; 1380 msg->type = msgtype; 1381 msg->len = len; 1382 1383 return (msg); 1384 } 1385 1386 int 1387 session_sendmsg(struct bgp_msg *msg, struct peer *p) 1388 { 1389 struct mrt *mrt; 1390 1391 LIST_FOREACH(mrt, &mrthead, entry) { 1392 if (!(mrt->type == MRT_ALL_OUT || (msg->type == UPDATE && 1393 mrt->type == MRT_UPDATE_OUT))) 1394 continue; 1395 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1396 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1397 mrt->group_id == p->conf.groupid)) 1398 mrt_dump_bgp_msg(mrt, msg->buf->buf, msg->len, p); 1399 } 1400 1401 ibuf_close(&p->wbuf, msg->buf); 1402 if (!p->throttled && p->wbuf.queued > SESS_MSG_HIGH_MARK) { 1403 if (imsg_rde(IMSG_XOFF, p->conf.id, NULL, 0) == -1) 1404 log_peer_warn(&p->conf, "imsg_compose XOFF"); 1405 else 1406 p->throttled = 1; 1407 } 1408 1409 free(msg); 1410 return (0); 1411 } 1412 1413 void 1414 session_open(struct peer *p) 1415 { 1416 struct bgp_msg *buf; 1417 struct ibuf *opb; 1418 struct msg_open msg; 1419 u_int16_t len; 1420 u_int8_t i, op_type, optparamlen = 0; 1421 int errs = 0; 1422 int mpcapa = 0; 1423 1424 1425 if ((opb = ibuf_dynamic(0, UCHAR_MAX - sizeof(op_type) - 1426 sizeof(optparamlen))) == NULL) { 1427 bgp_fsm(p, EVNT_CON_FATAL); 1428 return; 1429 } 1430 1431 /* multiprotocol extensions, RFC 4760 */ 1432 for (i = 0; i < AID_MAX; i++) 1433 if (p->capa.ann.mp[i]) { /* 4 bytes data */ 1434 errs += session_capa_add(opb, CAPA_MP, 4); 1435 errs += session_capa_add_mp(opb, i); 1436 mpcapa++; 1437 } 1438 1439 /* route refresh, RFC 2918 */ 1440 if (p->capa.ann.refresh) /* no data */ 1441 errs += session_capa_add(opb, CAPA_REFRESH, 0); 1442 1443 /* graceful restart and End-of-RIB marker, RFC 4724 */ 1444 if (p->capa.ann.grestart.restart) { 1445 int rst = 0; 1446 u_int16_t hdr; 1447 u_int8_t grlen; 1448 1449 if (mpcapa) { 1450 grlen = 2 + 4 * mpcapa; 1451 for (i = 0; i < AID_MAX; i++) { 1452 if (p->capa.neg.grestart.flags[i] & 1453 CAPA_GR_RESTARTING) 1454 rst++; 1455 } 1456 } else { /* AID_INET */ 1457 grlen = 2 + 4; 1458 if (p->capa.neg.grestart.flags[AID_INET] & 1459 CAPA_GR_RESTARTING) 1460 rst++; 1461 } 1462 1463 hdr = conf->holdtime; /* default timeout */ 1464 /* if client does graceful restart don't set R flag */ 1465 if (!rst) 1466 hdr |= CAPA_GR_R_FLAG; 1467 hdr = htons(hdr); 1468 1469 errs += session_capa_add(opb, CAPA_RESTART, grlen); 1470 errs += ibuf_add(opb, &hdr, sizeof(hdr)); 1471 1472 if (mpcapa) { 1473 for (i = 0; i < AID_MAX; i++) { 1474 if (p->capa.ann.mp[i]) { 1475 errs += session_capa_add_gr(p, opb, i); 1476 } 1477 } 1478 } else { /* AID_INET */ 1479 errs += session_capa_add_gr(p, opb, AID_INET); 1480 } 1481 } 1482 1483 /* 4-bytes AS numbers, draft-ietf-idr-as4bytes-13 */ 1484 if (p->capa.ann.as4byte) { /* 4 bytes data */ 1485 u_int32_t nas; 1486 1487 nas = htonl(p->conf.local_as); 1488 errs += session_capa_add(opb, CAPA_AS4BYTE, sizeof(nas)); 1489 errs += ibuf_add(opb, &nas, sizeof(nas)); 1490 } 1491 1492 if (ibuf_size(opb)) 1493 optparamlen = ibuf_size(opb) + sizeof(op_type) + 1494 sizeof(optparamlen); 1495 1496 len = MSGSIZE_OPEN_MIN + optparamlen; 1497 if (errs || (buf = session_newmsg(OPEN, len)) == NULL) { 1498 ibuf_free(opb); 1499 bgp_fsm(p, EVNT_CON_FATAL); 1500 return; 1501 } 1502 1503 msg.version = 4; 1504 msg.myas = htons(p->conf.local_short_as); 1505 if (p->conf.holdtime) 1506 msg.holdtime = htons(p->conf.holdtime); 1507 else 1508 msg.holdtime = htons(conf->holdtime); 1509 msg.bgpid = conf->bgpid; /* is already in network byte order */ 1510 msg.optparamlen = optparamlen; 1511 1512 errs += ibuf_add(buf->buf, &msg.version, sizeof(msg.version)); 1513 errs += ibuf_add(buf->buf, &msg.myas, sizeof(msg.myas)); 1514 errs += ibuf_add(buf->buf, &msg.holdtime, sizeof(msg.holdtime)); 1515 errs += ibuf_add(buf->buf, &msg.bgpid, sizeof(msg.bgpid)); 1516 errs += ibuf_add(buf->buf, &msg.optparamlen, sizeof(msg.optparamlen)); 1517 1518 if (optparamlen) { 1519 op_type = OPT_PARAM_CAPABILITIES; 1520 optparamlen = ibuf_size(opb); 1521 errs += ibuf_add(buf->buf, &op_type, sizeof(op_type)); 1522 errs += ibuf_add(buf->buf, &optparamlen, sizeof(optparamlen)); 1523 errs += ibuf_add(buf->buf, opb->buf, ibuf_size(opb)); 1524 } 1525 1526 ibuf_free(opb); 1527 1528 if (errs) { 1529 ibuf_free(buf->buf); 1530 free(buf); 1531 bgp_fsm(p, EVNT_CON_FATAL); 1532 return; 1533 } 1534 1535 if (session_sendmsg(buf, p) == -1) { 1536 bgp_fsm(p, EVNT_CON_FATAL); 1537 return; 1538 } 1539 1540 p->stats.msg_sent_open++; 1541 } 1542 1543 void 1544 session_keepalive(struct peer *p) 1545 { 1546 struct bgp_msg *buf; 1547 1548 if ((buf = session_newmsg(KEEPALIVE, MSGSIZE_KEEPALIVE)) == NULL || 1549 session_sendmsg(buf, p) == -1) { 1550 bgp_fsm(p, EVNT_CON_FATAL); 1551 return; 1552 } 1553 1554 start_timer_keepalive(p); 1555 p->stats.msg_sent_keepalive++; 1556 } 1557 1558 void 1559 session_update(u_int32_t peerid, void *data, size_t datalen) 1560 { 1561 struct peer *p; 1562 struct bgp_msg *buf; 1563 1564 if ((p = getpeerbyid(conf, peerid)) == NULL) { 1565 log_warnx("no such peer: id=%u", peerid); 1566 return; 1567 } 1568 1569 if (p->state != STATE_ESTABLISHED) 1570 return; 1571 1572 if ((buf = session_newmsg(UPDATE, MSGSIZE_HEADER + datalen)) == NULL) { 1573 bgp_fsm(p, EVNT_CON_FATAL); 1574 return; 1575 } 1576 1577 if (ibuf_add(buf->buf, data, datalen)) { 1578 ibuf_free(buf->buf); 1579 free(buf); 1580 bgp_fsm(p, EVNT_CON_FATAL); 1581 return; 1582 } 1583 1584 if (session_sendmsg(buf, p) == -1) { 1585 bgp_fsm(p, EVNT_CON_FATAL); 1586 return; 1587 } 1588 1589 start_timer_keepalive(p); 1590 p->stats.msg_sent_update++; 1591 } 1592 1593 void 1594 session_notification(struct peer *p, u_int8_t errcode, u_int8_t subcode, 1595 void *data, ssize_t datalen) 1596 { 1597 struct bgp_msg *buf; 1598 int errs = 0; 1599 1600 if (p->stats.last_sent_errcode) /* some notification already sent */ 1601 return; 1602 1603 log_notification(p, errcode, subcode, data, datalen, "sending"); 1604 1605 if ((buf = session_newmsg(NOTIFICATION, 1606 MSGSIZE_NOTIFICATION_MIN + datalen)) == NULL) { 1607 bgp_fsm(p, EVNT_CON_FATAL); 1608 return; 1609 } 1610 1611 errs += ibuf_add(buf->buf, &errcode, sizeof(errcode)); 1612 errs += ibuf_add(buf->buf, &subcode, sizeof(subcode)); 1613 1614 if (datalen > 0) 1615 errs += ibuf_add(buf->buf, data, datalen); 1616 1617 if (errs) { 1618 ibuf_free(buf->buf); 1619 free(buf); 1620 bgp_fsm(p, EVNT_CON_FATAL); 1621 return; 1622 } 1623 1624 if (session_sendmsg(buf, p) == -1) { 1625 bgp_fsm(p, EVNT_CON_FATAL); 1626 return; 1627 } 1628 1629 p->stats.msg_sent_notification++; 1630 p->stats.last_sent_errcode = errcode; 1631 p->stats.last_sent_suberr = subcode; 1632 } 1633 1634 int 1635 session_neighbor_rrefresh(struct peer *p) 1636 { 1637 u_int8_t i; 1638 1639 if (!p->capa.neg.refresh) 1640 return (-1); 1641 1642 for (i = 0; i < AID_MAX; i++) { 1643 if (p->capa.peer.mp[i] != 0) 1644 session_rrefresh(p, i); 1645 } 1646 1647 return (0); 1648 } 1649 1650 void 1651 session_rrefresh(struct peer *p, u_int8_t aid) 1652 { 1653 struct bgp_msg *buf; 1654 int errs = 0; 1655 u_int16_t afi; 1656 u_int8_t safi, null8 = 0; 1657 1658 if (aid2afi(aid, &afi, &safi) == -1) 1659 fatalx("session_rrefresh: bad afi/safi pair"); 1660 1661 if ((buf = session_newmsg(RREFRESH, MSGSIZE_RREFRESH)) == NULL) { 1662 bgp_fsm(p, EVNT_CON_FATAL); 1663 return; 1664 } 1665 1666 afi = htons(afi); 1667 errs += ibuf_add(buf->buf, &afi, sizeof(afi)); 1668 errs += ibuf_add(buf->buf, &null8, sizeof(null8)); 1669 errs += ibuf_add(buf->buf, &safi, sizeof(safi)); 1670 1671 if (errs) { 1672 ibuf_free(buf->buf); 1673 free(buf); 1674 bgp_fsm(p, EVNT_CON_FATAL); 1675 return; 1676 } 1677 1678 if (session_sendmsg(buf, p) == -1) { 1679 bgp_fsm(p, EVNT_CON_FATAL); 1680 return; 1681 } 1682 1683 p->stats.msg_sent_rrefresh++; 1684 } 1685 1686 int 1687 session_graceful_restart(struct peer *p) 1688 { 1689 u_int8_t i; 1690 1691 timer_set(&p->timers, Timer_RestartTimeout, 1692 p->capa.neg.grestart.timeout); 1693 1694 for (i = 0; i < AID_MAX; i++) { 1695 if (p->capa.neg.grestart.flags[i] & CAPA_GR_PRESENT) { 1696 if (imsg_rde(IMSG_SESSION_STALE, p->conf.id, 1697 &i, sizeof(i)) == -1) 1698 return (-1); 1699 log_peer_warnx(&p->conf, 1700 "graceful restart of %s, keeping routes", 1701 aid2str(i)); 1702 p->capa.neg.grestart.flags[i] |= CAPA_GR_RESTARTING; 1703 } else if (p->capa.neg.mp[i]) { 1704 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1705 &i, sizeof(i)) == -1) 1706 return (-1); 1707 log_peer_warnx(&p->conf, 1708 "graceful restart of %s, flushing routes", 1709 aid2str(i)); 1710 } 1711 } 1712 return (0); 1713 } 1714 1715 int 1716 session_graceful_stop(struct peer *p) 1717 { 1718 u_int8_t i; 1719 1720 for (i = 0; i < AID_MAX; i++) { 1721 /* 1722 * Only flush if the peer is restarting and the timeout fired. 1723 * In all other cases the session was already flushed when the 1724 * session went down or when the new open message was parsed. 1725 */ 1726 if (p->capa.neg.grestart.flags[i] & CAPA_GR_RESTARTING) { 1727 log_peer_warnx(&p->conf, "graceful restart of %s, " 1728 "time-out, flushing", aid2str(i)); 1729 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 1730 &i, sizeof(i)) == -1) 1731 return (-1); 1732 } 1733 p->capa.neg.grestart.flags[i] &= ~CAPA_GR_RESTARTING; 1734 } 1735 return (0); 1736 } 1737 1738 int 1739 session_dispatch_msg(struct pollfd *pfd, struct peer *p) 1740 { 1741 ssize_t n; 1742 socklen_t len; 1743 int error; 1744 1745 if (p->state == STATE_CONNECT) { 1746 if (pfd->revents & POLLOUT) { 1747 if (pfd->revents & POLLIN) { 1748 /* error occurred */ 1749 len = sizeof(error); 1750 if (getsockopt(pfd->fd, SOL_SOCKET, SO_ERROR, 1751 &error, &len) == -1 || error) { 1752 if (error) 1753 errno = error; 1754 if (errno != p->lasterr) { 1755 log_peer_warn(&p->conf, 1756 "socket error"); 1757 p->lasterr = errno; 1758 } 1759 bgp_fsm(p, EVNT_CON_OPENFAIL); 1760 return (1); 1761 } 1762 } 1763 bgp_fsm(p, EVNT_CON_OPEN); 1764 return (1); 1765 } 1766 if (pfd->revents & POLLHUP) { 1767 bgp_fsm(p, EVNT_CON_OPENFAIL); 1768 return (1); 1769 } 1770 if (pfd->revents & (POLLERR|POLLNVAL)) { 1771 bgp_fsm(p, EVNT_CON_FATAL); 1772 return (1); 1773 } 1774 return (0); 1775 } 1776 1777 if (pfd->revents & POLLHUP) { 1778 bgp_fsm(p, EVNT_CON_CLOSED); 1779 return (1); 1780 } 1781 if (pfd->revents & (POLLERR|POLLNVAL)) { 1782 bgp_fsm(p, EVNT_CON_FATAL); 1783 return (1); 1784 } 1785 1786 if (pfd->revents & POLLOUT && p->wbuf.queued) { 1787 if ((error = msgbuf_write(&p->wbuf)) <= 0 && errno != EAGAIN) { 1788 if (error == 0) 1789 log_peer_warnx(&p->conf, "Connection closed"); 1790 else if (error == -1) 1791 log_peer_warn(&p->conf, "write error"); 1792 bgp_fsm(p, EVNT_CON_FATAL); 1793 return (1); 1794 } 1795 p->stats.last_write = getmonotime(); 1796 if (p->holdtime > 0) 1797 timer_set(&p->timers, Timer_SendHold, 1798 p->holdtime < INTERVAL_HOLD ? INTERVAL_HOLD : 1799 p->holdtime); 1800 if (p->throttled && p->wbuf.queued < SESS_MSG_LOW_MARK) { 1801 if (imsg_rde(IMSG_XON, p->conf.id, NULL, 0) == -1) 1802 log_peer_warn(&p->conf, "imsg_compose XON"); 1803 else 1804 p->throttled = 0; 1805 } 1806 if (!(pfd->revents & POLLIN)) 1807 return (1); 1808 } 1809 1810 if (p->rbuf && pfd->revents & POLLIN) { 1811 if ((n = read(p->fd, p->rbuf->buf + p->rbuf->wpos, 1812 sizeof(p->rbuf->buf) - p->rbuf->wpos)) == -1) { 1813 if (errno != EINTR && errno != EAGAIN) { 1814 log_peer_warn(&p->conf, "read error"); 1815 bgp_fsm(p, EVNT_CON_FATAL); 1816 } 1817 return (1); 1818 } 1819 if (n == 0) { /* connection closed */ 1820 bgp_fsm(p, EVNT_CON_CLOSED); 1821 return (1); 1822 } 1823 1824 p->rbuf->wpos += n; 1825 p->stats.last_read = getmonotime(); 1826 return (1); 1827 } 1828 return (0); 1829 } 1830 1831 void 1832 session_process_msg(struct peer *p) 1833 { 1834 struct mrt *mrt; 1835 ssize_t rpos, av, left; 1836 int processed = 0; 1837 u_int16_t msglen; 1838 u_int8_t msgtype; 1839 1840 rpos = 0; 1841 av = p->rbuf->wpos; 1842 p->rpending = 0; 1843 1844 /* 1845 * session might drop to IDLE -> buffers deallocated 1846 * we MUST check rbuf != NULL before use 1847 */ 1848 for (;;) { 1849 if (p->rbuf == NULL) 1850 return; 1851 if (rpos + MSGSIZE_HEADER > av) 1852 break; 1853 if (parse_header(p, p->rbuf->buf + rpos, &msglen, 1854 &msgtype) == -1) 1855 return; 1856 if (rpos + msglen > av) 1857 break; 1858 p->rbuf->rptr = p->rbuf->buf + rpos; 1859 1860 /* dump to MRT as soon as we have a full packet */ 1861 LIST_FOREACH(mrt, &mrthead, entry) { 1862 if (!(mrt->type == MRT_ALL_IN || (msgtype == UPDATE && 1863 mrt->type == MRT_UPDATE_IN))) 1864 continue; 1865 if ((mrt->peer_id == 0 && mrt->group_id == 0) || 1866 mrt->peer_id == p->conf.id || (mrt->group_id != 0 && 1867 mrt->group_id == p->conf.groupid)) 1868 mrt_dump_bgp_msg(mrt, p->rbuf->rptr, msglen, p); 1869 } 1870 1871 switch (msgtype) { 1872 case OPEN: 1873 bgp_fsm(p, EVNT_RCVD_OPEN); 1874 p->stats.msg_rcvd_open++; 1875 break; 1876 case UPDATE: 1877 bgp_fsm(p, EVNT_RCVD_UPDATE); 1878 p->stats.msg_rcvd_update++; 1879 break; 1880 case NOTIFICATION: 1881 bgp_fsm(p, EVNT_RCVD_NOTIFICATION); 1882 p->stats.msg_rcvd_notification++; 1883 break; 1884 case KEEPALIVE: 1885 bgp_fsm(p, EVNT_RCVD_KEEPALIVE); 1886 p->stats.msg_rcvd_keepalive++; 1887 break; 1888 case RREFRESH: 1889 parse_refresh(p); 1890 p->stats.msg_rcvd_rrefresh++; 1891 break; 1892 default: /* cannot happen */ 1893 session_notification(p, ERR_HEADER, ERR_HDR_TYPE, 1894 &msgtype, 1); 1895 log_warnx("received message with unknown type %u", 1896 msgtype); 1897 bgp_fsm(p, EVNT_CON_FATAL); 1898 } 1899 rpos += msglen; 1900 if (++processed > MSG_PROCESS_LIMIT) { 1901 p->rpending = 1; 1902 break; 1903 } 1904 } 1905 1906 if (rpos < av) { 1907 left = av - rpos; 1908 memmove(&p->rbuf->buf, p->rbuf->buf + rpos, left); 1909 p->rbuf->wpos = left; 1910 } else 1911 p->rbuf->wpos = 0; 1912 } 1913 1914 int 1915 parse_header(struct peer *peer, u_char *data, u_int16_t *len, u_int8_t *type) 1916 { 1917 u_char *p; 1918 u_int16_t olen; 1919 static const u_int8_t marker[MSGSIZE_HEADER_MARKER] = { 0xff, 0xff, 1920 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 1921 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; 1922 1923 /* caller MUST make sure we are getting 19 bytes! */ 1924 p = data; 1925 if (memcmp(p, marker, sizeof(marker))) { 1926 log_peer_warnx(&peer->conf, "sync error"); 1927 session_notification(peer, ERR_HEADER, ERR_HDR_SYNC, NULL, 0); 1928 bgp_fsm(peer, EVNT_CON_FATAL); 1929 return (-1); 1930 } 1931 p += MSGSIZE_HEADER_MARKER; 1932 1933 memcpy(&olen, p, 2); 1934 *len = ntohs(olen); 1935 p += 2; 1936 memcpy(type, p, 1); 1937 1938 if (*len < MSGSIZE_HEADER || *len > MAX_PKTSIZE) { 1939 log_peer_warnx(&peer->conf, 1940 "received message: illegal length: %u byte", *len); 1941 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1942 &olen, sizeof(olen)); 1943 bgp_fsm(peer, EVNT_CON_FATAL); 1944 return (-1); 1945 } 1946 1947 switch (*type) { 1948 case OPEN: 1949 if (*len < MSGSIZE_OPEN_MIN) { 1950 log_peer_warnx(&peer->conf, 1951 "received OPEN: illegal len: %u byte", *len); 1952 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1953 &olen, sizeof(olen)); 1954 bgp_fsm(peer, EVNT_CON_FATAL); 1955 return (-1); 1956 } 1957 break; 1958 case NOTIFICATION: 1959 if (*len < MSGSIZE_NOTIFICATION_MIN) { 1960 log_peer_warnx(&peer->conf, 1961 "received NOTIFICATION: illegal len: %u byte", 1962 *len); 1963 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1964 &olen, sizeof(olen)); 1965 bgp_fsm(peer, EVNT_CON_FATAL); 1966 return (-1); 1967 } 1968 break; 1969 case UPDATE: 1970 if (*len < MSGSIZE_UPDATE_MIN) { 1971 log_peer_warnx(&peer->conf, 1972 "received UPDATE: illegal len: %u byte", *len); 1973 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1974 &olen, sizeof(olen)); 1975 bgp_fsm(peer, EVNT_CON_FATAL); 1976 return (-1); 1977 } 1978 break; 1979 case KEEPALIVE: 1980 if (*len != MSGSIZE_KEEPALIVE) { 1981 log_peer_warnx(&peer->conf, 1982 "received KEEPALIVE: illegal len: %u byte", *len); 1983 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1984 &olen, sizeof(olen)); 1985 bgp_fsm(peer, EVNT_CON_FATAL); 1986 return (-1); 1987 } 1988 break; 1989 case RREFRESH: 1990 if (*len != MSGSIZE_RREFRESH) { 1991 log_peer_warnx(&peer->conf, 1992 "received RREFRESH: illegal len: %u byte", *len); 1993 session_notification(peer, ERR_HEADER, ERR_HDR_LEN, 1994 &olen, sizeof(olen)); 1995 bgp_fsm(peer, EVNT_CON_FATAL); 1996 return (-1); 1997 } 1998 break; 1999 default: 2000 log_peer_warnx(&peer->conf, 2001 "received msg with unknown type %u", *type); 2002 session_notification(peer, ERR_HEADER, ERR_HDR_TYPE, 2003 type, 1); 2004 bgp_fsm(peer, EVNT_CON_FATAL); 2005 return (-1); 2006 } 2007 return (0); 2008 } 2009 2010 int 2011 parse_open(struct peer *peer) 2012 { 2013 u_char *p, *op_val; 2014 u_int8_t version, rversion; 2015 u_int16_t short_as, msglen; 2016 u_int16_t holdtime, oholdtime, myholdtime; 2017 u_int32_t as, bgpid; 2018 u_int8_t optparamlen, plen; 2019 u_int8_t op_type, op_len; 2020 2021 p = peer->rbuf->rptr; 2022 p += MSGSIZE_HEADER_MARKER; 2023 memcpy(&msglen, p, sizeof(msglen)); 2024 msglen = ntohs(msglen); 2025 2026 p = peer->rbuf->rptr; 2027 p += MSGSIZE_HEADER; /* header is already checked */ 2028 2029 memcpy(&version, p, sizeof(version)); 2030 p += sizeof(version); 2031 2032 if (version != BGP_VERSION) { 2033 log_peer_warnx(&peer->conf, 2034 "peer wants unrecognized version %u", version); 2035 if (version > BGP_VERSION) 2036 rversion = version - BGP_VERSION; 2037 else 2038 rversion = BGP_VERSION; 2039 session_notification(peer, ERR_OPEN, ERR_OPEN_VERSION, 2040 &rversion, sizeof(rversion)); 2041 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2042 return (-1); 2043 } 2044 2045 memcpy(&short_as, p, sizeof(short_as)); 2046 p += sizeof(short_as); 2047 as = peer->short_as = ntohs(short_as); 2048 if (as == 0) { 2049 log_peer_warnx(&peer->conf, 2050 "peer requests unacceptable AS %u", as); 2051 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, 2052 NULL, 0); 2053 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2054 return (-1); 2055 } 2056 2057 memcpy(&oholdtime, p, sizeof(oholdtime)); 2058 p += sizeof(oholdtime); 2059 2060 holdtime = ntohs(oholdtime); 2061 if (holdtime && holdtime < peer->conf.min_holdtime) { 2062 log_peer_warnx(&peer->conf, 2063 "peer requests unacceptable holdtime %u", holdtime); 2064 session_notification(peer, ERR_OPEN, ERR_OPEN_HOLDTIME, 2065 NULL, 0); 2066 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2067 return (-1); 2068 } 2069 2070 myholdtime = peer->conf.holdtime; 2071 if (!myholdtime) 2072 myholdtime = conf->holdtime; 2073 if (holdtime < myholdtime) 2074 peer->holdtime = holdtime; 2075 else 2076 peer->holdtime = myholdtime; 2077 2078 memcpy(&bgpid, p, sizeof(bgpid)); 2079 p += sizeof(bgpid); 2080 2081 /* check bgpid for validity - just disallow 0 */ 2082 if (ntohl(bgpid) == 0) { 2083 log_peer_warnx(&peer->conf, "peer BGPID %u unacceptable", 2084 ntohl(bgpid)); 2085 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2086 NULL, 0); 2087 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2088 return (-1); 2089 } 2090 peer->remote_bgpid = bgpid; 2091 2092 memcpy(&optparamlen, p, sizeof(optparamlen)); 2093 p += sizeof(optparamlen); 2094 2095 if (optparamlen != msglen - MSGSIZE_OPEN_MIN) { 2096 log_peer_warnx(&peer->conf, 2097 "corrupt OPEN message received: length mismatch"); 2098 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2099 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2100 return (-1); 2101 } 2102 2103 plen = optparamlen; 2104 while (plen > 0) { 2105 if (plen < 2) { 2106 log_peer_warnx(&peer->conf, 2107 "corrupt OPEN message received, len wrong"); 2108 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2109 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2110 return (-1); 2111 } 2112 memcpy(&op_type, p, sizeof(op_type)); 2113 p += sizeof(op_type); 2114 plen -= sizeof(op_type); 2115 memcpy(&op_len, p, sizeof(op_len)); 2116 p += sizeof(op_len); 2117 plen -= sizeof(op_len); 2118 if (op_len > 0) { 2119 if (plen < op_len) { 2120 log_peer_warnx(&peer->conf, 2121 "corrupt OPEN message received, len wrong"); 2122 session_notification(peer, ERR_OPEN, 0, 2123 NULL, 0); 2124 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2125 return (-1); 2126 } 2127 op_val = p; 2128 p += op_len; 2129 plen -= op_len; 2130 } else 2131 op_val = NULL; 2132 2133 switch (op_type) { 2134 case OPT_PARAM_CAPABILITIES: /* RFC 3392 */ 2135 if (parse_capabilities(peer, op_val, op_len, 2136 &as) == -1) { 2137 session_notification(peer, ERR_OPEN, 0, 2138 NULL, 0); 2139 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2140 return (-1); 2141 } 2142 break; 2143 case OPT_PARAM_AUTH: /* deprecated */ 2144 default: 2145 /* 2146 * unsupported type 2147 * the RFCs tell us to leave the data section empty 2148 * and notify the peer with ERR_OPEN, ERR_OPEN_OPT. 2149 * How the peer should know _which_ optional parameter 2150 * we don't support is beyond me. 2151 */ 2152 log_peer_warnx(&peer->conf, 2153 "received OPEN message with unsupported optional " 2154 "parameter: type %u", op_type); 2155 session_notification(peer, ERR_OPEN, ERR_OPEN_OPT, 2156 NULL, 0); 2157 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2158 /* no punish */ 2159 timer_set(&peer->timers, Timer_IdleHold, 0); 2160 peer->IdleHoldTime /= 2; 2161 return (-1); 2162 } 2163 } 2164 2165 /* if remote-as is zero and it's a cloned neighbor, accept any */ 2166 if (peer->template && !peer->conf.remote_as && as != AS_TRANS) { 2167 peer->conf.remote_as = as; 2168 peer->conf.ebgp = (peer->conf.remote_as != peer->conf.local_as); 2169 if (!peer->conf.ebgp) 2170 /* force enforce_as off for iBGP sessions */ 2171 peer->conf.enforce_as = ENFORCE_AS_OFF; 2172 } 2173 2174 if (peer->conf.remote_as != as) { 2175 log_peer_warnx(&peer->conf, "peer sent wrong AS %s", 2176 log_as(as)); 2177 session_notification(peer, ERR_OPEN, ERR_OPEN_AS, NULL, 0); 2178 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2179 return (-1); 2180 } 2181 2182 /* on iBGP sessions check for bgpid collision */ 2183 if (!peer->conf.ebgp && peer->remote_bgpid == conf->bgpid) { 2184 log_peer_warnx(&peer->conf, "peer BGPID %u conflicts with ours", 2185 ntohl(bgpid)); 2186 session_notification(peer, ERR_OPEN, ERR_OPEN_BGPID, 2187 NULL, 0); 2188 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2189 return (-1); 2190 } 2191 2192 if (capa_neg_calc(peer) == -1) { 2193 log_peer_warnx(&peer->conf, 2194 "capability negotiation calculation failed"); 2195 session_notification(peer, ERR_OPEN, 0, NULL, 0); 2196 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2197 return (-1); 2198 } 2199 2200 return (0); 2201 } 2202 2203 int 2204 parse_update(struct peer *peer) 2205 { 2206 u_char *p; 2207 u_int16_t datalen; 2208 2209 /* 2210 * we pass the message verbatim to the rde. 2211 * in case of errors the whole session is reset with a 2212 * notification anyway, we only need to know the peer 2213 */ 2214 p = peer->rbuf->rptr; 2215 p += MSGSIZE_HEADER_MARKER; 2216 memcpy(&datalen, p, sizeof(datalen)); 2217 datalen = ntohs(datalen); 2218 2219 p = peer->rbuf->rptr; 2220 p += MSGSIZE_HEADER; /* header is already checked */ 2221 datalen -= MSGSIZE_HEADER; 2222 2223 if (imsg_rde(IMSG_UPDATE, peer->conf.id, p, datalen) == -1) 2224 return (-1); 2225 2226 return (0); 2227 } 2228 2229 int 2230 parse_refresh(struct peer *peer) 2231 { 2232 u_char *p; 2233 u_int16_t afi; 2234 u_int8_t aid, safi; 2235 2236 p = peer->rbuf->rptr; 2237 p += MSGSIZE_HEADER; /* header is already checked */ 2238 2239 /* 2240 * We could check if we actually announced the capability but 2241 * as long as the message is correctly encoded we don't care. 2242 */ 2243 2244 /* afi, 2 byte */ 2245 memcpy(&afi, p, sizeof(afi)); 2246 afi = ntohs(afi); 2247 p += 2; 2248 /* reserved, 1 byte */ 2249 p += 1; 2250 /* safi, 1 byte */ 2251 memcpy(&safi, p, sizeof(safi)); 2252 2253 /* afi/safi unchecked - unrecognized values will be ignored anyway */ 2254 if (afi2aid(afi, safi, &aid) == -1) { 2255 log_peer_warnx(&peer->conf, "peer sent bad refresh, " 2256 "invalid afi/safi pair"); 2257 return (0); 2258 } 2259 2260 if (!peer->capa.neg.refresh) { 2261 log_peer_warnx(&peer->conf, "peer sent unexpected refresh"); 2262 return (0); 2263 } 2264 2265 if (imsg_rde(IMSG_REFRESH, peer->conf.id, &aid, sizeof(aid)) == -1) 2266 return (-1); 2267 2268 return (0); 2269 } 2270 2271 int 2272 parse_notification(struct peer *peer) 2273 { 2274 u_char *p; 2275 u_int16_t datalen; 2276 u_int8_t errcode; 2277 u_int8_t subcode; 2278 u_int8_t capa_code; 2279 u_int8_t capa_len; 2280 size_t reason_len; 2281 u_int8_t i; 2282 2283 /* just log */ 2284 p = peer->rbuf->rptr; 2285 p += MSGSIZE_HEADER_MARKER; 2286 memcpy(&datalen, p, sizeof(datalen)); 2287 datalen = ntohs(datalen); 2288 2289 p = peer->rbuf->rptr; 2290 p += MSGSIZE_HEADER; /* header is already checked */ 2291 datalen -= MSGSIZE_HEADER; 2292 2293 memcpy(&errcode, p, sizeof(errcode)); 2294 p += sizeof(errcode); 2295 datalen -= sizeof(errcode); 2296 2297 memcpy(&subcode, p, sizeof(subcode)); 2298 p += sizeof(subcode); 2299 datalen -= sizeof(subcode); 2300 2301 log_notification(peer, errcode, subcode, p, datalen, "received"); 2302 peer->errcnt++; 2303 peer->stats.last_rcvd_errcode = errcode; 2304 peer->stats.last_rcvd_suberr = subcode; 2305 2306 if (errcode == ERR_OPEN && subcode == ERR_OPEN_CAPA) { 2307 if (datalen == 0) { /* zebra likes to send those.. humbug */ 2308 log_peer_warnx(&peer->conf, "received \"unsupported " 2309 "capability\" notification without data part, " 2310 "disabling capability announcements altogether"); 2311 session_capa_ann_none(peer); 2312 } 2313 2314 while (datalen > 0) { 2315 if (datalen < 2) { 2316 log_peer_warnx(&peer->conf, 2317 "parse_notification: " 2318 "expect len >= 2, len is %u", datalen); 2319 return (-1); 2320 } 2321 memcpy(&capa_code, p, sizeof(capa_code)); 2322 p += sizeof(capa_code); 2323 datalen -= sizeof(capa_code); 2324 memcpy(&capa_len, p, sizeof(capa_len)); 2325 p += sizeof(capa_len); 2326 datalen -= sizeof(capa_len); 2327 if (datalen < capa_len) { 2328 log_peer_warnx(&peer->conf, 2329 "parse_notification: capa_len %u exceeds " 2330 "remaining msg length %u", capa_len, 2331 datalen); 2332 return (-1); 2333 } 2334 p += capa_len; 2335 datalen -= capa_len; 2336 switch (capa_code) { 2337 case CAPA_MP: 2338 for (i = 0; i < AID_MAX; i++) 2339 peer->capa.ann.mp[i] = 0; 2340 log_peer_warnx(&peer->conf, 2341 "disabling multiprotocol capability"); 2342 break; 2343 case CAPA_REFRESH: 2344 peer->capa.ann.refresh = 0; 2345 log_peer_warnx(&peer->conf, 2346 "disabling route refresh capability"); 2347 break; 2348 case CAPA_RESTART: 2349 peer->capa.ann.grestart.restart = 0; 2350 log_peer_warnx(&peer->conf, 2351 "disabling restart capability"); 2352 break; 2353 case CAPA_AS4BYTE: 2354 peer->capa.ann.as4byte = 0; 2355 log_peer_warnx(&peer->conf, 2356 "disabling 4-byte AS num capability"); 2357 break; 2358 default: /* should not happen... */ 2359 log_peer_warnx(&peer->conf, "received " 2360 "\"unsupported capability\" notification " 2361 "for unknown capability %u, disabling " 2362 "capability announcements altogether", 2363 capa_code); 2364 session_capa_ann_none(peer); 2365 break; 2366 } 2367 } 2368 2369 return (1); 2370 } 2371 2372 if (errcode == ERR_OPEN && subcode == ERR_OPEN_OPT) { 2373 session_capa_ann_none(peer); 2374 return (1); 2375 } 2376 2377 if (errcode == ERR_CEASE && 2378 (subcode == ERR_CEASE_ADMIN_DOWN || 2379 subcode == ERR_CEASE_ADMIN_RESET)) { 2380 if (datalen > 1) { 2381 reason_len = *p++; 2382 datalen--; 2383 if (datalen < reason_len) { 2384 log_peer_warnx(&peer->conf, 2385 "received truncated shutdown reason"); 2386 return (0); 2387 } 2388 if (reason_len > REASON_LEN - 1) { 2389 log_peer_warnx(&peer->conf, 2390 "received overly long shutdown reason"); 2391 return (0); 2392 } 2393 memcpy(peer->stats.last_reason, p, reason_len); 2394 peer->stats.last_reason[reason_len] = '\0'; 2395 log_peer_warnx(&peer->conf, 2396 "received shutdown reason: \"%s\"", 2397 log_reason(peer->stats.last_reason)); 2398 p += reason_len; 2399 datalen -= reason_len; 2400 } 2401 } 2402 2403 return (0); 2404 } 2405 2406 int 2407 parse_capabilities(struct peer *peer, u_char *d, u_int16_t dlen, u_int32_t *as) 2408 { 2409 u_char *capa_val; 2410 u_int32_t remote_as; 2411 u_int16_t len; 2412 u_int16_t afi; 2413 u_int16_t gr_header; 2414 u_int8_t safi; 2415 u_int8_t aid; 2416 u_int8_t gr_flags; 2417 u_int8_t capa_code; 2418 u_int8_t capa_len; 2419 u_int8_t i; 2420 2421 len = dlen; 2422 while (len > 0) { 2423 if (len < 2) { 2424 log_peer_warnx(&peer->conf, "Bad capabilities attr " 2425 "length: %u, too short", len); 2426 return (-1); 2427 } 2428 memcpy(&capa_code, d, sizeof(capa_code)); 2429 d += sizeof(capa_code); 2430 len -= sizeof(capa_code); 2431 memcpy(&capa_len, d, sizeof(capa_len)); 2432 d += sizeof(capa_len); 2433 len -= sizeof(capa_len); 2434 if (capa_len > 0) { 2435 if (len < capa_len) { 2436 log_peer_warnx(&peer->conf, 2437 "Bad capabilities attr length: " 2438 "len %u smaller than capa_len %u", 2439 len, capa_len); 2440 return (-1); 2441 } 2442 capa_val = d; 2443 d += capa_len; 2444 len -= capa_len; 2445 } else 2446 capa_val = NULL; 2447 2448 switch (capa_code) { 2449 case CAPA_MP: /* RFC 4760 */ 2450 if (capa_len != 4) { 2451 log_peer_warnx(&peer->conf, 2452 "Bad multi protocol capability length: " 2453 "%u", capa_len); 2454 break; 2455 } 2456 memcpy(&afi, capa_val, sizeof(afi)); 2457 afi = ntohs(afi); 2458 memcpy(&safi, capa_val + 3, sizeof(safi)); 2459 if (afi2aid(afi, safi, &aid) == -1) { 2460 log_peer_warnx(&peer->conf, 2461 "Received multi protocol capability: " 2462 " unknown AFI %u, safi %u pair", 2463 afi, safi); 2464 break; 2465 } 2466 peer->capa.peer.mp[aid] = 1; 2467 break; 2468 case CAPA_REFRESH: 2469 peer->capa.peer.refresh = 1; 2470 break; 2471 case CAPA_RESTART: 2472 if (capa_len == 2) { 2473 /* peer only supports EoR marker */ 2474 peer->capa.peer.grestart.restart = 1; 2475 peer->capa.peer.grestart.timeout = 0; 2476 break; 2477 } else if (capa_len % 4 != 2) { 2478 log_peer_warnx(&peer->conf, 2479 "Bad graceful restart capability length: " 2480 "%u", capa_len); 2481 peer->capa.peer.grestart.restart = 0; 2482 peer->capa.peer.grestart.timeout = 0; 2483 break; 2484 } 2485 2486 memcpy(&gr_header, capa_val, sizeof(gr_header)); 2487 gr_header = ntohs(gr_header); 2488 peer->capa.peer.grestart.timeout = 2489 gr_header & CAPA_GR_TIMEMASK; 2490 if (peer->capa.peer.grestart.timeout == 0) { 2491 log_peer_warnx(&peer->conf, "Received " 2492 "graceful restart timeout is zero"); 2493 peer->capa.peer.grestart.restart = 0; 2494 break; 2495 } 2496 2497 for (i = 2; i <= capa_len - 4; i += 4) { 2498 memcpy(&afi, capa_val + i, sizeof(afi)); 2499 afi = ntohs(afi); 2500 memcpy(&safi, capa_val + i + 2, sizeof(safi)); 2501 if (afi2aid(afi, safi, &aid) == -1) { 2502 log_peer_warnx(&peer->conf, 2503 "Received graceful restart capa: " 2504 " unknown AFI %u, safi %u pair", 2505 afi, safi); 2506 continue; 2507 } 2508 memcpy(&gr_flags, capa_val + i + 3, 2509 sizeof(gr_flags)); 2510 peer->capa.peer.grestart.flags[aid] |= 2511 CAPA_GR_PRESENT; 2512 if (gr_flags & CAPA_GR_F_FLAG) 2513 peer->capa.peer.grestart.flags[aid] |= 2514 CAPA_GR_FORWARD; 2515 if (gr_header & CAPA_GR_R_FLAG) 2516 peer->capa.peer.grestart.flags[aid] |= 2517 CAPA_GR_RESTART; 2518 peer->capa.peer.grestart.restart = 2; 2519 } 2520 break; 2521 case CAPA_AS4BYTE: 2522 if (capa_len != 4) { 2523 log_peer_warnx(&peer->conf, 2524 "Bad AS4BYTE capability length: " 2525 "%u", capa_len); 2526 peer->capa.peer.as4byte = 0; 2527 break; 2528 } 2529 memcpy(&remote_as, capa_val, sizeof(remote_as)); 2530 *as = ntohl(remote_as); 2531 if (*as == 0) { 2532 log_peer_warnx(&peer->conf, 2533 "peer requests unacceptable AS %u", *as); 2534 session_notification(peer, ERR_OPEN, 2535 ERR_OPEN_AS, NULL, 0); 2536 change_state(peer, STATE_IDLE, EVNT_RCVD_OPEN); 2537 return (-1); 2538 } 2539 peer->capa.peer.as4byte = 1; 2540 break; 2541 default: 2542 break; 2543 } 2544 } 2545 2546 return (0); 2547 } 2548 2549 int 2550 capa_neg_calc(struct peer *p) 2551 { 2552 u_int8_t i, hasmp = 0; 2553 2554 /* a capability is accepted only if both sides announced it */ 2555 2556 p->capa.neg.refresh = 2557 (p->capa.ann.refresh && p->capa.peer.refresh) != 0; 2558 2559 p->capa.neg.as4byte = 2560 (p->capa.ann.as4byte && p->capa.peer.as4byte) != 0; 2561 2562 /* MP: both side must agree on the AFI,SAFI pair */ 2563 for (i = 0; i < AID_MAX; i++) { 2564 if (p->capa.ann.mp[i] && p->capa.peer.mp[i]) 2565 p->capa.neg.mp[i] = 1; 2566 else 2567 p->capa.neg.mp[i] = 0; 2568 if (p->capa.ann.mp[i]) 2569 hasmp = 1; 2570 } 2571 /* if no MP capability present default to IPv4 unicast mode */ 2572 if (!hasmp) 2573 p->capa.neg.mp[AID_INET] = 1; 2574 2575 /* 2576 * graceful restart: the peer capabilities are of interest here. 2577 * It is necessary to compare the new values with the previous ones 2578 * and act acordingly. AFI/SAFI that are not part in the MP capability 2579 * are treated as not being present. 2580 * Also make sure that a flush happens if the session stopped 2581 * supporting graceful restart. 2582 */ 2583 2584 for (i = 0; i < AID_MAX; i++) { 2585 int8_t negflags; 2586 2587 /* disable GR if the AFI/SAFI is not present */ 2588 if (p->capa.ann.grestart.restart == 0 || 2589 (p->capa.peer.grestart.flags[i] & CAPA_GR_PRESENT && 2590 p->capa.neg.mp[i] == 0)) 2591 p->capa.peer.grestart.flags[i] = 0; /* disable */ 2592 /* look at current GR state and decide what to do */ 2593 negflags = p->capa.neg.grestart.flags[i]; 2594 p->capa.neg.grestart.flags[i] = p->capa.peer.grestart.flags[i]; 2595 if (negflags & CAPA_GR_RESTARTING) { 2596 if (!(p->capa.peer.grestart.flags[i] & 2597 CAPA_GR_FORWARD)) { 2598 if (imsg_rde(IMSG_SESSION_FLUSH, p->conf.id, 2599 &i, sizeof(i)) == -1) 2600 return (-1); 2601 log_peer_warnx(&p->conf, "graceful restart of " 2602 "%s, not restarted, flushing", aid2str(i)); 2603 } else 2604 p->capa.neg.grestart.flags[i] |= 2605 CAPA_GR_RESTARTING; 2606 } 2607 } 2608 p->capa.neg.grestart.timeout = p->capa.peer.grestart.timeout; 2609 p->capa.neg.grestart.restart = p->capa.peer.grestart.restart; 2610 if (p->capa.ann.grestart.restart == 0) 2611 p->capa.neg.grestart.restart = 0; 2612 2613 return (0); 2614 } 2615 2616 void 2617 session_dispatch_imsg(struct imsgbuf *ibuf, int idx, u_int *listener_cnt) 2618 { 2619 struct imsg imsg; 2620 struct mrt xmrt; 2621 struct mrt *mrt; 2622 struct imsgbuf *i; 2623 struct peer *p; 2624 struct listen_addr *la, *nla; 2625 struct kif *kif; 2626 u_char *data; 2627 int n, fd, depend_ok, restricted; 2628 u_int16_t t; 2629 u_int8_t aid, errcode, subcode; 2630 2631 while (ibuf) { 2632 if ((n = imsg_get(ibuf, &imsg)) == -1) 2633 fatal("session_dispatch_imsg: imsg_get error"); 2634 2635 if (n == 0) 2636 break; 2637 2638 switch (imsg.hdr.type) { 2639 case IMSG_SOCKET_CONN: 2640 case IMSG_SOCKET_CONN_CTL: 2641 if (idx != PFD_PIPE_MAIN) 2642 fatalx("reconf request not from parent"); 2643 if ((fd = imsg.fd) == -1) { 2644 log_warnx("expected to receive imsg fd to " 2645 "RDE but didn't receive any"); 2646 break; 2647 } 2648 if ((i = malloc(sizeof(struct imsgbuf))) == NULL) 2649 fatal(NULL); 2650 imsg_init(i, fd); 2651 if (imsg.hdr.type == IMSG_SOCKET_CONN) { 2652 if (ibuf_rde) { 2653 log_warnx("Unexpected imsg connection " 2654 "to RDE received"); 2655 msgbuf_clear(&ibuf_rde->w); 2656 free(ibuf_rde); 2657 } 2658 ibuf_rde = i; 2659 } else { 2660 if (ibuf_rde_ctl) { 2661 log_warnx("Unexpected imsg ctl " 2662 "connection to RDE received"); 2663 msgbuf_clear(&ibuf_rde_ctl->w); 2664 free(ibuf_rde_ctl); 2665 } 2666 ibuf_rde_ctl = i; 2667 } 2668 break; 2669 case IMSG_RECONF_CONF: 2670 if (idx != PFD_PIPE_MAIN) 2671 fatalx("reconf request not from parent"); 2672 nconf = new_config(); 2673 2674 copy_config(nconf, imsg.data); 2675 pending_reconf = 1; 2676 break; 2677 case IMSG_RECONF_PEER: 2678 if (idx != PFD_PIPE_MAIN) 2679 fatalx("reconf request not from parent"); 2680 if ((p = calloc(1, sizeof(struct peer))) == NULL) 2681 fatal("new_peer"); 2682 memcpy(&p->conf, imsg.data, sizeof(struct peer_config)); 2683 p->state = p->prev_state = STATE_NONE; 2684 p->reconf_action = RECONF_REINIT; 2685 if (RB_INSERT(peer_head, &nconf->peers, p) != NULL) 2686 fatalx("%s: peer tree is corrupt", __func__); 2687 break; 2688 case IMSG_RECONF_LISTENER: 2689 if (idx != PFD_PIPE_MAIN) 2690 fatalx("reconf request not from parent"); 2691 if (nconf == NULL) 2692 fatalx("IMSG_RECONF_LISTENER but no config"); 2693 nla = imsg.data; 2694 TAILQ_FOREACH(la, conf->listen_addrs, entry) 2695 if (!la_cmp(la, nla)) 2696 break; 2697 2698 if (la == NULL) { 2699 if (nla->reconf != RECONF_REINIT) 2700 fatalx("king bula sez: " 2701 "expected REINIT"); 2702 2703 if ((nla->fd = imsg.fd) == -1) 2704 log_warnx("expected to receive fd for " 2705 "%s but didn't receive any", 2706 log_sockaddr((struct sockaddr *) 2707 &nla->sa, nla->sa_len)); 2708 2709 la = calloc(1, sizeof(struct listen_addr)); 2710 if (la == NULL) 2711 fatal(NULL); 2712 memcpy(&la->sa, &nla->sa, sizeof(la->sa)); 2713 la->flags = nla->flags; 2714 la->fd = nla->fd; 2715 la->reconf = RECONF_REINIT; 2716 TAILQ_INSERT_TAIL(nconf->listen_addrs, la, 2717 entry); 2718 } else { 2719 if (nla->reconf != RECONF_KEEP) 2720 fatalx("king bula sez: expected KEEP"); 2721 la->reconf = RECONF_KEEP; 2722 } 2723 2724 break; 2725 case IMSG_RECONF_CTRL: 2726 if (idx != PFD_PIPE_MAIN) 2727 fatalx("reconf request not from parent"); 2728 if (imsg.hdr.len != IMSG_HEADER_SIZE + 2729 sizeof(restricted)) 2730 fatalx("IFINFO imsg with wrong len"); 2731 memcpy(&restricted, imsg.data, sizeof(restricted)); 2732 if (imsg.fd == -1) { 2733 log_warnx("expected to receive fd for control " 2734 "socket but didn't receive any"); 2735 break; 2736 } 2737 if (restricted) { 2738 control_shutdown(rcsock); 2739 rcsock = imsg.fd; 2740 } else { 2741 control_shutdown(csock); 2742 csock = imsg.fd; 2743 } 2744 break; 2745 case IMSG_RECONF_DRAIN: 2746 switch (idx) { 2747 case PFD_PIPE_ROUTE: 2748 if (nconf != NULL) 2749 fatalx("got unexpected %s from RDE", 2750 "IMSG_RECONF_DONE"); 2751 imsg_compose(ibuf_main, IMSG_RECONF_DONE, 0, 0, 2752 -1, NULL, 0); 2753 break; 2754 case PFD_PIPE_MAIN: 2755 if (nconf == NULL) 2756 fatalx("got unexpected %s from parent", 2757 "IMSG_RECONF_DONE"); 2758 imsg_compose(ibuf_main, IMSG_RECONF_DRAIN, 0, 0, 2759 -1, NULL, 0); 2760 break; 2761 default: 2762 fatalx("reconf request not from parent or RDE"); 2763 } 2764 break; 2765 case IMSG_RECONF_DONE: 2766 if (idx != PFD_PIPE_MAIN) 2767 fatalx("reconf request not from parent"); 2768 if (nconf == NULL) 2769 fatalx("got IMSG_RECONF_DONE but no config"); 2770 copy_config(conf, nconf); 2771 merge_peers(conf, nconf); 2772 2773 /* delete old listeners */ 2774 for (la = TAILQ_FIRST(conf->listen_addrs); la != NULL; 2775 la = nla) { 2776 nla = TAILQ_NEXT(la, entry); 2777 if (la->reconf == RECONF_NONE) { 2778 log_info("not listening on %s any more", 2779 log_sockaddr((struct sockaddr *) 2780 &la->sa, la->sa_len)); 2781 TAILQ_REMOVE(conf->listen_addrs, la, 2782 entry); 2783 close(la->fd); 2784 free(la); 2785 } 2786 } 2787 2788 /* add new listeners */ 2789 TAILQ_CONCAT(conf->listen_addrs, nconf->listen_addrs, 2790 entry); 2791 2792 setup_listeners(listener_cnt); 2793 free_config(nconf); 2794 nconf = NULL; 2795 pending_reconf = 0; 2796 log_info("SE reconfigured"); 2797 /* 2798 * IMSG_RECONF_DONE is sent when the RDE drained 2799 * the peer config sent in merge_peers(). 2800 */ 2801 break; 2802 case IMSG_IFINFO: 2803 if (idx != PFD_PIPE_MAIN) 2804 fatalx("IFINFO message not from parent"); 2805 if (imsg.hdr.len != IMSG_HEADER_SIZE + 2806 sizeof(struct kif)) 2807 fatalx("IFINFO imsg with wrong len"); 2808 kif = imsg.data; 2809 depend_ok = kif->depend_state; 2810 2811 RB_FOREACH(p, peer_head, &conf->peers) 2812 if (!strcmp(p->conf.if_depend, kif->ifname)) { 2813 if (depend_ok && !p->depend_ok) { 2814 p->depend_ok = depend_ok; 2815 bgp_fsm(p, EVNT_START); 2816 } else if (!depend_ok && p->depend_ok) { 2817 p->depend_ok = depend_ok; 2818 session_stop(p, 2819 ERR_CEASE_OTHER_CHANGE); 2820 } 2821 } 2822 break; 2823 case IMSG_MRT_OPEN: 2824 case IMSG_MRT_REOPEN: 2825 if (imsg.hdr.len > IMSG_HEADER_SIZE + 2826 sizeof(struct mrt)) { 2827 log_warnx("wrong imsg len"); 2828 break; 2829 } 2830 2831 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 2832 if ((xmrt.wbuf.fd = imsg.fd) == -1) 2833 log_warnx("expected to receive fd for mrt dump " 2834 "but didn't receive any"); 2835 2836 mrt = mrt_get(&mrthead, &xmrt); 2837 if (mrt == NULL) { 2838 /* new dump */ 2839 mrt = calloc(1, sizeof(struct mrt)); 2840 if (mrt == NULL) 2841 fatal("session_dispatch_imsg"); 2842 memcpy(mrt, &xmrt, sizeof(struct mrt)); 2843 TAILQ_INIT(&mrt->wbuf.bufs); 2844 LIST_INSERT_HEAD(&mrthead, mrt, entry); 2845 } else { 2846 /* old dump reopened */ 2847 close(mrt->wbuf.fd); 2848 mrt->wbuf.fd = xmrt.wbuf.fd; 2849 } 2850 break; 2851 case IMSG_MRT_CLOSE: 2852 if (imsg.hdr.len > IMSG_HEADER_SIZE + 2853 sizeof(struct mrt)) { 2854 log_warnx("wrong imsg len"); 2855 break; 2856 } 2857 2858 memcpy(&xmrt, imsg.data, sizeof(struct mrt)); 2859 mrt = mrt_get(&mrthead, &xmrt); 2860 if (mrt != NULL) 2861 mrt_done(mrt); 2862 break; 2863 case IMSG_CTL_KROUTE: 2864 case IMSG_CTL_KROUTE_ADDR: 2865 case IMSG_CTL_SHOW_NEXTHOP: 2866 case IMSG_CTL_SHOW_INTERFACE: 2867 case IMSG_CTL_SHOW_FIB_TABLES: 2868 case IMSG_CTL_SHOW_RTR: 2869 case IMSG_CTL_SHOW_TIMER: 2870 if (idx != PFD_PIPE_MAIN) 2871 fatalx("ctl kroute request not from parent"); 2872 control_imsg_relay(&imsg); 2873 break; 2874 case IMSG_CTL_SHOW_RIB: 2875 case IMSG_CTL_SHOW_RIB_PREFIX: 2876 case IMSG_CTL_SHOW_RIB_COMMUNITIES: 2877 case IMSG_CTL_SHOW_RIB_ATTR: 2878 case IMSG_CTL_SHOW_RIB_MEM: 2879 case IMSG_CTL_SHOW_RIB_HASH: 2880 case IMSG_CTL_SHOW_NETWORK: 2881 case IMSG_CTL_SHOW_NEIGHBOR: 2882 case IMSG_CTL_SHOW_SET: 2883 if (idx != PFD_PIPE_ROUTE_CTL) 2884 fatalx("ctl rib request not from RDE"); 2885 control_imsg_relay(&imsg); 2886 break; 2887 case IMSG_CTL_END: 2888 case IMSG_CTL_RESULT: 2889 control_imsg_relay(&imsg); 2890 break; 2891 case IMSG_UPDATE: 2892 if (idx != PFD_PIPE_ROUTE) 2893 fatalx("update request not from RDE"); 2894 if (imsg.hdr.len > IMSG_HEADER_SIZE + 2895 MAX_PKTSIZE - MSGSIZE_HEADER || 2896 imsg.hdr.len < IMSG_HEADER_SIZE + 2897 MSGSIZE_UPDATE_MIN - MSGSIZE_HEADER) 2898 log_warnx("RDE sent invalid update"); 2899 else 2900 session_update(imsg.hdr.peerid, imsg.data, 2901 imsg.hdr.len - IMSG_HEADER_SIZE); 2902 break; 2903 case IMSG_UPDATE_ERR: 2904 if (idx != PFD_PIPE_ROUTE) 2905 fatalx("update request not from RDE"); 2906 if (imsg.hdr.len < IMSG_HEADER_SIZE + 2) { 2907 log_warnx("RDE sent invalid notification"); 2908 break; 2909 } 2910 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 2911 log_warnx("no such peer: id=%u", 2912 imsg.hdr.peerid); 2913 break; 2914 } 2915 data = imsg.data; 2916 errcode = *data++; 2917 subcode = *data++; 2918 2919 if (imsg.hdr.len == IMSG_HEADER_SIZE + 2) 2920 data = NULL; 2921 2922 session_notification(p, errcode, subcode, 2923 data, imsg.hdr.len - IMSG_HEADER_SIZE - 2); 2924 switch (errcode) { 2925 case ERR_CEASE: 2926 switch (subcode) { 2927 case ERR_CEASE_MAX_PREFIX: 2928 case ERR_CEASE_MAX_SENT_PREFIX: 2929 t = p->conf.max_out_prefix_restart; 2930 if (subcode == ERR_CEASE_MAX_PREFIX) 2931 t = p->conf.max_prefix_restart; 2932 2933 bgp_fsm(p, EVNT_STOP); 2934 if (t) 2935 timer_set(&p->timers, 2936 Timer_IdleHold, 60 * t); 2937 break; 2938 default: 2939 bgp_fsm(p, EVNT_CON_FATAL); 2940 break; 2941 } 2942 break; 2943 default: 2944 bgp_fsm(p, EVNT_CON_FATAL); 2945 break; 2946 } 2947 break; 2948 case IMSG_SESSION_RESTARTED: 2949 if (idx != PFD_PIPE_ROUTE) 2950 fatalx("update request not from RDE"); 2951 if (imsg.hdr.len < IMSG_HEADER_SIZE + sizeof(aid)) { 2952 log_warnx("RDE sent invalid restart msg"); 2953 break; 2954 } 2955 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 2956 log_warnx("no such peer: id=%u", 2957 imsg.hdr.peerid); 2958 break; 2959 } 2960 memcpy(&aid, imsg.data, sizeof(aid)); 2961 if (aid >= AID_MAX) 2962 fatalx("IMSG_SESSION_RESTARTED: bad AID"); 2963 if (p->capa.neg.grestart.flags[aid] & 2964 CAPA_GR_RESTARTING) { 2965 log_peer_warnx(&p->conf, 2966 "graceful restart of %s finished", 2967 aid2str(aid)); 2968 p->capa.neg.grestart.flags[aid] &= 2969 ~CAPA_GR_RESTARTING; 2970 timer_stop(&p->timers, Timer_RestartTimeout); 2971 2972 /* signal back to RDE to cleanup stale routes */ 2973 if (imsg_rde(IMSG_SESSION_RESTARTED, 2974 imsg.hdr.peerid, &aid, sizeof(aid)) == -1) 2975 fatal("imsg_compose: " 2976 "IMSG_SESSION_RESTARTED"); 2977 } 2978 break; 2979 case IMSG_SESSION_DOWN: 2980 if (idx != PFD_PIPE_ROUTE) 2981 fatalx("update request not from RDE"); 2982 if ((p = getpeerbyid(conf, imsg.hdr.peerid)) == NULL) { 2983 log_warnx("no such peer: id=%u", 2984 imsg.hdr.peerid); 2985 break; 2986 } 2987 session_stop(p, ERR_CEASE_ADMIN_DOWN); 2988 break; 2989 default: 2990 break; 2991 } 2992 imsg_free(&imsg); 2993 } 2994 } 2995 2996 int 2997 la_cmp(struct listen_addr *a, struct listen_addr *b) 2998 { 2999 struct sockaddr_in *in_a, *in_b; 3000 struct sockaddr_in6 *in6_a, *in6_b; 3001 3002 if (a->sa.ss_family != b->sa.ss_family) 3003 return (1); 3004 3005 switch (a->sa.ss_family) { 3006 case AF_INET: 3007 in_a = (struct sockaddr_in *)&a->sa; 3008 in_b = (struct sockaddr_in *)&b->sa; 3009 if (in_a->sin_addr.s_addr != in_b->sin_addr.s_addr) 3010 return (1); 3011 if (in_a->sin_port != in_b->sin_port) 3012 return (1); 3013 break; 3014 case AF_INET6: 3015 in6_a = (struct sockaddr_in6 *)&a->sa; 3016 in6_b = (struct sockaddr_in6 *)&b->sa; 3017 if (bcmp(&in6_a->sin6_addr, &in6_b->sin6_addr, 3018 sizeof(struct in6_addr))) 3019 return (1); 3020 if (in6_a->sin6_port != in6_b->sin6_port) 3021 return (1); 3022 break; 3023 default: 3024 fatal("king bula sez: unknown address family"); 3025 /* NOTREACHED */ 3026 } 3027 3028 return (0); 3029 } 3030 3031 struct peer * 3032 getpeerbydesc(struct bgpd_config *c, const char *descr) 3033 { 3034 struct peer *p, *res = NULL; 3035 int match = 0; 3036 3037 RB_FOREACH(p, peer_head, &c->peers) 3038 if (!strcmp(p->conf.descr, descr)) { 3039 res = p; 3040 match++; 3041 } 3042 3043 if (match > 1) 3044 log_info("neighbor description \"%s\" not unique, request " 3045 "aborted", descr); 3046 3047 if (match == 1) 3048 return (res); 3049 else 3050 return (NULL); 3051 } 3052 3053 struct peer * 3054 getpeerbyip(struct bgpd_config *c, struct sockaddr *ip) 3055 { 3056 struct bgpd_addr addr; 3057 struct peer *p, *newpeer, *loose = NULL; 3058 u_int32_t id; 3059 3060 sa2addr(ip, &addr, NULL); 3061 3062 /* we might want a more effective way to find peers by IP */ 3063 RB_FOREACH(p, peer_head, &c->peers) 3064 if (!p->conf.template && 3065 !memcmp(&addr, &p->conf.remote_addr, sizeof(addr))) 3066 return (p); 3067 3068 /* try template matching */ 3069 RB_FOREACH(p, peer_head, &c->peers) 3070 if (p->conf.template && 3071 p->conf.remote_addr.aid == addr.aid && 3072 session_match_mask(p, &addr)) 3073 if (loose == NULL || loose->conf.remote_masklen < 3074 p->conf.remote_masklen) 3075 loose = p; 3076 3077 if (loose != NULL) { 3078 /* clone */ 3079 if ((newpeer = malloc(sizeof(struct peer))) == NULL) 3080 fatal(NULL); 3081 memcpy(newpeer, loose, sizeof(struct peer)); 3082 for (id = PEER_ID_DYN_MAX; id > PEER_ID_STATIC_MAX; id--) { 3083 if (getpeerbyid(c, id) == NULL) /* we found a free id */ 3084 break; 3085 } 3086 newpeer->template = loose; 3087 session_template_clone(newpeer, ip, id, 0); 3088 newpeer->state = newpeer->prev_state = STATE_NONE; 3089 newpeer->reconf_action = RECONF_KEEP; 3090 newpeer->rbuf = NULL; 3091 newpeer->rpending = 0; 3092 init_peer(newpeer); 3093 bgp_fsm(newpeer, EVNT_START); 3094 if (RB_INSERT(peer_head, &c->peers, newpeer) != NULL) 3095 fatalx("%s: peer tree is corrupt", __func__); 3096 return (newpeer); 3097 } 3098 3099 return (NULL); 3100 } 3101 3102 struct peer * 3103 getpeerbyid(struct bgpd_config *c, u_int32_t peerid) 3104 { 3105 static struct peer lookup; 3106 3107 lookup.conf.id = peerid; 3108 3109 return RB_FIND(peer_head, &c->peers, &lookup); 3110 } 3111 3112 int 3113 peer_matched(struct peer *p, struct ctl_neighbor *n) 3114 { 3115 char *s; 3116 3117 if (n && n->addr.aid) { 3118 if (memcmp(&p->conf.remote_addr, &n->addr, 3119 sizeof(p->conf.remote_addr))) 3120 return 0; 3121 } else if (n && n->descr[0]) { 3122 s = n->is_group ? p->conf.group : p->conf.descr; 3123 if (strcmp(s, n->descr)) 3124 return 0; 3125 } 3126 return 1; 3127 } 3128 3129 void 3130 session_template_clone(struct peer *p, struct sockaddr *ip, u_int32_t id, 3131 u_int32_t as) 3132 { 3133 struct bgpd_addr remote_addr; 3134 3135 if (ip) 3136 sa2addr(ip, &remote_addr, NULL); 3137 else 3138 memcpy(&remote_addr, &p->conf.remote_addr, sizeof(remote_addr)); 3139 3140 memcpy(&p->conf, &p->template->conf, sizeof(struct peer_config)); 3141 3142 p->conf.id = id; 3143 3144 if (as) { 3145 p->conf.remote_as = as; 3146 p->conf.ebgp = (p->conf.remote_as != p->conf.local_as); 3147 if (!p->conf.ebgp) 3148 /* force enforce_as off for iBGP sessions */ 3149 p->conf.enforce_as = ENFORCE_AS_OFF; 3150 } 3151 3152 memcpy(&p->conf.remote_addr, &remote_addr, sizeof(remote_addr)); 3153 switch (p->conf.remote_addr.aid) { 3154 case AID_INET: 3155 p->conf.remote_masklen = 32; 3156 break; 3157 case AID_INET6: 3158 p->conf.remote_masklen = 128; 3159 break; 3160 } 3161 p->conf.template = 0; 3162 } 3163 3164 int 3165 session_match_mask(struct peer *p, struct bgpd_addr *a) 3166 { 3167 struct in_addr v4masked; 3168 struct in6_addr v6masked; 3169 3170 switch (p->conf.remote_addr.aid) { 3171 case AID_INET: 3172 inet4applymask(&v4masked, &a->v4, p->conf.remote_masklen); 3173 if (p->conf.remote_addr.v4.s_addr == v4masked.s_addr) 3174 return (1); 3175 return (0); 3176 case AID_INET6: 3177 inet6applymask(&v6masked, &a->v6, p->conf.remote_masklen); 3178 3179 if (memcmp(&v6masked, &p->conf.remote_addr.v6, 3180 sizeof(v6masked)) == 0) 3181 return (1); 3182 return (0); 3183 } 3184 return (0); 3185 } 3186 3187 void 3188 session_down(struct peer *peer) 3189 { 3190 bzero(&peer->capa.neg, sizeof(peer->capa.neg)); 3191 peer->stats.last_updown = getmonotime(); 3192 /* 3193 * session_down is called in the exit code path so check 3194 * if the RDE is still around, if not there is no need to 3195 * send the message. 3196 */ 3197 if (ibuf_rde == NULL) 3198 return; 3199 if (imsg_rde(IMSG_SESSION_DOWN, peer->conf.id, NULL, 0) == -1) 3200 fatalx("imsg_compose error"); 3201 } 3202 3203 void 3204 session_up(struct peer *p) 3205 { 3206 struct session_up sup; 3207 3208 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3209 &p->conf, sizeof(p->conf)) == -1) 3210 fatalx("imsg_compose error"); 3211 3212 if (p->local.aid == AID_INET) { 3213 sup.local_v4_addr = p->local; 3214 sup.local_v6_addr = p->local_alt; 3215 } else { 3216 sup.local_v6_addr = p->local; 3217 sup.local_v4_addr = p->local_alt; 3218 } 3219 sup.remote_addr = p->remote; 3220 3221 sup.remote_bgpid = p->remote_bgpid; 3222 sup.short_as = p->short_as; 3223 memcpy(&sup.capa, &p->capa.neg, sizeof(sup.capa)); 3224 p->stats.last_updown = getmonotime(); 3225 if (imsg_rde(IMSG_SESSION_UP, p->conf.id, &sup, sizeof(sup)) == -1) 3226 fatalx("imsg_compose error"); 3227 } 3228 3229 int 3230 imsg_ctl_parent(int type, u_int32_t peerid, pid_t pid, void *data, 3231 u_int16_t datalen) 3232 { 3233 return (imsg_compose(ibuf_main, type, peerid, pid, -1, data, datalen)); 3234 } 3235 3236 int 3237 imsg_ctl_rde(int type, pid_t pid, void *data, u_int16_t datalen) 3238 { 3239 if (ibuf_rde_ctl == NULL) { 3240 log_warnx("Can't send message %u to RDE, ctl pipe closed", 3241 type); 3242 return (0); 3243 } 3244 /* 3245 * Use control socket to talk to RDE to bypass the queue of the 3246 * regular imsg socket. 3247 */ 3248 return (imsg_compose(ibuf_rde_ctl, type, 0, pid, -1, data, datalen)); 3249 } 3250 3251 int 3252 imsg_rde(int type, uint32_t peerid, void *data, u_int16_t datalen) 3253 { 3254 if (ibuf_rde == NULL) { 3255 log_warnx("Can't send message %u to RDE, pipe closed", type); 3256 return (0); 3257 } 3258 3259 return (imsg_compose(ibuf_rde, type, peerid, 0, -1, data, datalen)); 3260 } 3261 3262 void 3263 session_demote(struct peer *p, int level) 3264 { 3265 struct demote_msg msg; 3266 3267 strlcpy(msg.demote_group, p->conf.demote_group, 3268 sizeof(msg.demote_group)); 3269 msg.level = level; 3270 if (imsg_compose(ibuf_main, IMSG_DEMOTE, p->conf.id, 0, -1, 3271 &msg, sizeof(msg)) == -1) 3272 fatalx("imsg_compose error"); 3273 3274 p->demoted += level; 3275 } 3276 3277 void 3278 session_stop(struct peer *peer, u_int8_t subcode) 3279 { 3280 char data[REASON_LEN]; 3281 size_t datalen; 3282 size_t reason_len; 3283 char *communication; 3284 3285 datalen = 0; 3286 communication = peer->conf.reason; 3287 3288 if ((subcode == ERR_CEASE_ADMIN_DOWN || 3289 subcode == ERR_CEASE_ADMIN_RESET) 3290 && communication && *communication) { 3291 reason_len = strlen(communication); 3292 if (reason_len > REASON_LEN - 1) { 3293 log_peer_warnx(&peer->conf, 3294 "trying to send overly long shutdown reason"); 3295 } else { 3296 data[0] = reason_len; 3297 datalen = reason_len + sizeof(data[0]); 3298 memcpy(data + 1, communication, reason_len); 3299 } 3300 } 3301 switch (peer->state) { 3302 case STATE_OPENSENT: 3303 case STATE_OPENCONFIRM: 3304 case STATE_ESTABLISHED: 3305 session_notification(peer, ERR_CEASE, subcode, data, datalen); 3306 break; 3307 default: 3308 /* session not open, no need to send notification */ 3309 break; 3310 } 3311 bgp_fsm(peer, EVNT_STOP); 3312 } 3313 3314 void 3315 merge_peers(struct bgpd_config *c, struct bgpd_config *nc) 3316 { 3317 struct peer *p, *np, *next; 3318 3319 RB_FOREACH(p, peer_head, &c->peers) { 3320 /* templates are handled specially */ 3321 if (p->template != NULL) 3322 continue; 3323 np = getpeerbyid(nc, p->conf.id); 3324 if (np == NULL) { 3325 p->reconf_action = RECONF_DELETE; 3326 continue; 3327 } 3328 3329 /* peer no longer uses TCP MD5SIG so deconfigure */ 3330 if (p->conf.auth.method == AUTH_MD5SIG && 3331 np->conf.auth.method != AUTH_MD5SIG) 3332 tcp_md5_del_listener(c, p); 3333 else if (np->conf.auth.method == AUTH_MD5SIG) 3334 tcp_md5_add_listener(c, np); 3335 3336 memcpy(&p->conf, &np->conf, sizeof(p->conf)); 3337 RB_REMOVE(peer_head, &nc->peers, np); 3338 free(np); 3339 3340 p->reconf_action = RECONF_KEEP; 3341 3342 /* had demotion, is demoted, demote removed? */ 3343 if (p->demoted && !p->conf.demote_group[0]) 3344 session_demote(p, -1); 3345 3346 /* if session is not open then refresh pfkey data */ 3347 if (p->state < STATE_OPENSENT && !p->template) 3348 imsg_compose(ibuf_main, IMSG_PFKEY_RELOAD, 3349 p->conf.id, 0, -1, NULL, 0); 3350 3351 /* sync the RDE in case we keep the peer */ 3352 if (imsg_rde(IMSG_SESSION_ADD, p->conf.id, 3353 &p->conf, sizeof(struct peer_config)) == -1) 3354 fatalx("imsg_compose error"); 3355 3356 /* apply the config to all clones of a template */ 3357 if (p->conf.template) { 3358 struct peer *xp; 3359 RB_FOREACH(xp, peer_head, &c->peers) { 3360 if (xp->template != p) 3361 continue; 3362 session_template_clone(xp, NULL, xp->conf.id, 3363 xp->conf.remote_as); 3364 if (imsg_rde(IMSG_SESSION_ADD, xp->conf.id, 3365 &xp->conf, sizeof(xp->conf)) == -1) 3366 fatalx("imsg_compose error"); 3367 } 3368 } 3369 } 3370 3371 if (imsg_rde(IMSG_RECONF_DRAIN, 0, NULL, 0) == -1) 3372 fatalx("imsg_compose error"); 3373 3374 /* pfkeys of new peers already loaded by the parent process */ 3375 RB_FOREACH_SAFE(np, peer_head, &nc->peers, next) { 3376 RB_REMOVE(peer_head, &nc->peers, np); 3377 if (RB_INSERT(peer_head, &c->peers, np) != NULL) 3378 fatalx("%s: peer tree is corrupt", __func__); 3379 if (np->conf.auth.method == AUTH_MD5SIG) 3380 tcp_md5_add_listener(c, np); 3381 } 3382 } 3383