bgpd.c revision 1.244 
1/*	$OpenBSD: bgpd.c,v 1.244 2022/06/05 12:43:13 claudio Exp $ */
2
3/*
4 * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
5 *
6 * Permission to use, copy, modify, and distribute this software for any
7 * purpose with or without fee is hereby granted, provided that the above
8 * copyright notice and this permission notice appear in all copies.
9 *
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17 */
18
19#include <sys/types.h>
20#include <sys/socket.h>
21#include <sys/wait.h>
22#include <netinet/in.h>
23#include <arpa/inet.h>
24#include <err.h>
25#include <errno.h>
26#include <fcntl.h>
27#include <poll.h>
28#include <pwd.h>
29#include <signal.h>
30#include <stdio.h>
31#include <stdlib.h>
32#include <string.h>
33#include <syslog.h>
34#include <unistd.h>
35
36#include "bgpd.h"
37#include "session.h"
38#include "log.h"
39#include "version.h"
40
41void		sighdlr(int);
42__dead void	usage(void);
43int		main(int, char *[]);
44pid_t		start_child(enum bgpd_process, char *, int, int, int);
45int		send_filterset(struct imsgbuf *, struct filter_set_head *);
46int		reconfigure(char *, struct bgpd_config *);
47int		send_config(struct bgpd_config *);
48int		dispatch_imsg(struct imsgbuf *, int, struct bgpd_config *);
49int		control_setup(struct bgpd_config *);
50static void	getsockpair(int [2]);
51int		imsg_send_sockets(struct imsgbuf *, struct imsgbuf *,
52		    struct imsgbuf *);
53void		bgpd_rtr_connect(struct rtr_config *);
54void		bgpd_rtr_connect_done(int, struct bgpd_config *);
55
56int			 cflags;
57volatile sig_atomic_t	 mrtdump;
58volatile sig_atomic_t	 quit;
59volatile sig_atomic_t	 reconfig;
60pid_t			 reconfpid;
61int			 reconfpending;
62struct imsgbuf		*ibuf_se;
63struct imsgbuf		*ibuf_rde;
64struct imsgbuf		*ibuf_rtr;
65struct rib_names	 ribnames = SIMPLEQ_HEAD_INITIALIZER(ribnames);
66char			*cname;
67char			*rcname;
68
69struct connect_elm {
70	TAILQ_ENTRY(connect_elm)	entry;
71	uint32_t			id;
72	int				fd;
73};
74
75TAILQ_HEAD( ,connect_elm)	connect_queue = \
76				    TAILQ_HEAD_INITIALIZER(connect_queue);
77u_int				connect_cnt;
78#define MAX_CONNECT_CNT		32
79
80void
81sighdlr(int sig)
82{
83	switch (sig) {
84	case SIGTERM:
85	case SIGINT:
86		quit = 1;
87		break;
88	case SIGHUP:
89		reconfig = 1;
90		break;
91	case SIGALRM:
92	case SIGUSR1:
93		mrtdump = 1;
94		break;
95	}
96}
97
98__dead void
99usage(void)
100{
101	extern char *__progname;
102
103	fprintf(stderr, "usage: %s [-cdnvV] [-D macro=value] [-f file]\n",
104	    __progname);
105	exit(1);
106}
107
108#define PFD_PIPE_SESSION	0
109#define PFD_PIPE_RDE		1
110#define PFD_PIPE_RTR		2
111#define PFD_SOCK_ROUTE		3
112#define PFD_SOCK_PFKEY		4
113#define PFD_CONNECT_START	5
114#define MAX_TIMEOUT		3600
115
116int	 cmd_opts;
117
118int
119main(int argc, char *argv[])
120{
121	struct bgpd_config	*conf;
122	enum bgpd_process	 proc = PROC_MAIN;
123	struct rde_rib		*rr;
124	struct peer		*p;
125	struct pollfd		*pfd = NULL;
126	struct connect_elm	*ce;
127	time_t			 timeout;
128	pid_t			 se_pid = 0, rde_pid = 0, rtr_pid = 0, pid;
129	char			*conffile;
130	char			*saved_argv0;
131	u_int			 pfd_elms = 0, npfd, i;
132	int			 debug = 0;
133	int			 rfd, keyfd;
134	int			 ch, status;
135	int			 pipe_m2s[2];
136	int			 pipe_m2r[2];
137	int			 pipe_m2roa[2];
138
139	conffile = CONFFILE;
140
141	log_init(1, LOG_DAEMON);	/* log to stderr until daemonized */
142	log_procinit(log_procnames[PROC_MAIN]);
143	log_setverbose(1);
144
145	saved_argv0 = argv[0];
146	if (saved_argv0 == NULL)
147		saved_argv0 = "bgpd";
148
149	while ((ch = getopt(argc, argv, "cdD:f:nRSTvV")) != -1) {
150		switch (ch) {
151		case 'c':
152			cmd_opts |= BGPD_OPT_FORCE_DEMOTE;
153			break;
154		case 'd':
155			debug = 1;
156			break;
157		case 'D':
158			if (cmdline_symset(optarg) < 0)
159				log_warnx("could not parse macro definition %s",
160				    optarg);
161			break;
162		case 'f':
163			conffile = optarg;
164			break;
165		case 'n':
166			cmd_opts |= BGPD_OPT_NOACTION;
167			break;
168		case 'v':
169			if (cmd_opts & BGPD_OPT_VERBOSE)
170				cmd_opts |= BGPD_OPT_VERBOSE2;
171			cmd_opts |= BGPD_OPT_VERBOSE;
172			break;
173		case 'R':
174			proc = PROC_RDE;
175			break;
176		case 'S':
177			proc = PROC_SE;
178			break;
179		case 'T':
180			proc = PROC_RTR;
181			break;
182		case 'V':
183			fprintf(stderr, "OpenBGPD %s\n", BGPD_VERSION);
184			return 0;
185		default:
186			usage();
187			/* NOTREACHED */
188		}
189	}
190
191	argc -= optind;
192	argv += optind;
193	if (argc > 0)
194		usage();
195
196	if (cmd_opts & BGPD_OPT_NOACTION) {
197		if ((conf = parse_config(conffile, NULL, NULL)) == NULL)
198			exit(1);
199
200		if (cmd_opts & BGPD_OPT_VERBOSE)
201			print_config(conf, &ribnames);
202		else
203			fprintf(stderr, "configuration OK\n");
204
205		while ((rr = SIMPLEQ_FIRST(&ribnames)) != NULL) {
206			SIMPLEQ_REMOVE_HEAD(&ribnames, entry);
207			free(rr);
208		}
209		free_config(conf);
210		exit(0);
211	}
212
213	switch (proc) {
214	case PROC_MAIN:
215		break;
216	case PROC_RDE:
217		rde_main(debug, cmd_opts & BGPD_OPT_VERBOSE);
218		/* NOTREACHED */
219	case PROC_SE:
220		session_main(debug, cmd_opts & BGPD_OPT_VERBOSE);
221		/* NOTREACHED */
222	case PROC_RTR:
223		rtr_main(debug, cmd_opts & BGPD_OPT_VERBOSE);
224		/* NOTREACHED */
225	}
226
227	if (geteuid())
228		errx(1, "need root privileges");
229
230	if (getpwnam(BGPD_USER) == NULL)
231		errx(1, "unknown user %s", BGPD_USER);
232
233	if ((conf = parse_config(conffile, NULL, NULL)) == NULL) {
234		log_warnx("config file %s has errors", conffile);
235		exit(1);
236	}
237
238	if (prepare_listeners(conf) == -1)
239		exit(1);
240
241	log_init(debug, LOG_DAEMON);
242	log_setverbose(cmd_opts & BGPD_OPT_VERBOSE);
243
244	if (!debug)
245		daemon(1, 0);
246
247	log_info("startup");
248
249	getsockpair(pipe_m2s);
250	getsockpair(pipe_m2r);
251	getsockpair(pipe_m2roa);
252
253	/* fork children */
254	rde_pid = start_child(PROC_RDE, saved_argv0, pipe_m2r[1], debug,
255	    cmd_opts & BGPD_OPT_VERBOSE);
256	se_pid = start_child(PROC_SE, saved_argv0, pipe_m2s[1], debug,
257	    cmd_opts & BGPD_OPT_VERBOSE);
258	rtr_pid = start_child(PROC_RTR, saved_argv0, pipe_m2roa[1], debug,
259	    cmd_opts & BGPD_OPT_VERBOSE);
260
261	signal(SIGTERM, sighdlr);
262	signal(SIGINT, sighdlr);
263	signal(SIGHUP, sighdlr);
264	signal(SIGALRM, sighdlr);
265	signal(SIGUSR1, sighdlr);
266	signal(SIGPIPE, SIG_IGN);
267
268	if ((ibuf_se = malloc(sizeof(struct imsgbuf))) == NULL ||
269	    (ibuf_rde = malloc(sizeof(struct imsgbuf))) == NULL ||
270	    (ibuf_rtr = malloc(sizeof(struct imsgbuf))) == NULL)
271		fatal(NULL);
272	imsg_init(ibuf_se, pipe_m2s[0]);
273	imsg_init(ibuf_rde, pipe_m2r[0]);
274	imsg_init(ibuf_rtr, pipe_m2roa[0]);
275	mrt_init(ibuf_rde, ibuf_se);
276	if (kr_init(&rfd, conf->fib_priority) == -1)
277		quit = 1;
278	keyfd = pfkey_init();
279
280	/*
281	 * rpath, read config file
282	 * cpath, unlink control socket
283	 * fattr, chmod on control socket
284	 * wpath, needed if we are doing mrt dumps
285	 *
286	 * pledge placed here because kr_init() does a setsockopt on the
287	 * routing socket thats not allowed at all.
288	 */
289#if 0
290	/*
291	 * disabled because we do ioctls on /dev/pf and SIOCSIFGATTR
292	 * this needs some redesign of bgpd to be fixed.
293	 */
294BROKEN	if (pledge("stdio rpath wpath cpath fattr unix route recvfd sendfd",
295	    NULL) == -1)
296		fatal("pledge");
297#endif
298
299	if (imsg_send_sockets(ibuf_se, ibuf_rde, ibuf_rtr))
300		fatal("could not establish imsg links");
301	/* control setup needs to happen late since it sends imsgs */
302	if (control_setup(conf) == -1)
303		quit = 1;
304	if (send_config(conf) != 0)
305		quit = 1;
306	if (pftable_clear_all() != 0)
307		quit = 1;
308
309	while (quit == 0) {
310		if (pfd_elms < PFD_CONNECT_START + connect_cnt) {
311			struct pollfd *newp;
312
313			if ((newp = reallocarray(pfd,
314			    PFD_CONNECT_START + connect_cnt,
315			    sizeof(struct pollfd))) == NULL) {
316				log_warn("could not resize pfd from %u -> %u"
317				    " entries", pfd_elms, PFD_CONNECT_START +
318				    connect_cnt);
319				fatalx("exiting");
320			}
321			pfd = newp;
322			pfd_elms = PFD_CONNECT_START + connect_cnt;
323		}
324		bzero(pfd, sizeof(struct pollfd) * pfd_elms);
325
326		timeout = mrt_timeout(conf->mrt);
327
328		pfd[PFD_SOCK_ROUTE].fd = rfd;
329		pfd[PFD_SOCK_ROUTE].events = POLLIN;
330
331		pfd[PFD_SOCK_PFKEY].fd = keyfd;
332		pfd[PFD_SOCK_PFKEY].events = POLLIN;
333
334		set_pollfd(&pfd[PFD_PIPE_SESSION], ibuf_se);
335		set_pollfd(&pfd[PFD_PIPE_RDE], ibuf_rde);
336		set_pollfd(&pfd[PFD_PIPE_RTR], ibuf_rtr);
337
338		npfd = PFD_CONNECT_START;
339		TAILQ_FOREACH(ce, &connect_queue, entry) {
340			pfd[npfd].fd = ce->fd;
341			pfd[npfd++].events = POLLOUT;
342			if (npfd > pfd_elms)
343				fatalx("polli pfd overflow");
344		}
345
346		if (timeout < 0 || timeout > MAX_TIMEOUT)
347			timeout = MAX_TIMEOUT;
348		if (poll(pfd, npfd, timeout * 1000) == -1) {
349			if (errno != EINTR) {
350				log_warn("poll error");
351				quit = 1;
352			}
353			goto next_loop;
354		}
355
356		if (handle_pollfd(&pfd[PFD_PIPE_SESSION], ibuf_se) == -1) {
357			log_warnx("main: Lost connection to SE");
358			msgbuf_clear(&ibuf_se->w);
359			free(ibuf_se);
360			ibuf_se = NULL;
361			quit = 1;
362		} else {
363			if (dispatch_imsg(ibuf_se, PFD_PIPE_SESSION, conf) ==
364			    -1)
365				quit = 1;
366		}
367
368		if (handle_pollfd(&pfd[PFD_PIPE_RDE], ibuf_rde) == -1) {
369			log_warnx("main: Lost connection to RDE");
370			msgbuf_clear(&ibuf_rde->w);
371			free(ibuf_rde);
372			ibuf_rde = NULL;
373			quit = 1;
374		} else {
375			if (dispatch_imsg(ibuf_rde, PFD_PIPE_RDE, conf) == -1)
376				quit = 1;
377		}
378
379		if (handle_pollfd(&pfd[PFD_PIPE_RTR], ibuf_rtr) == -1) {
380			log_warnx("main: Lost connection to RTR");
381			msgbuf_clear(&ibuf_rtr->w);
382			free(ibuf_rtr);
383			ibuf_rtr = NULL;
384			quit = 1;
385		} else {
386			if (dispatch_imsg(ibuf_rtr, PFD_PIPE_RTR, conf) == -1)
387				quit = 1;
388		}
389
390		if (pfd[PFD_SOCK_ROUTE].revents & POLLIN) {
391			if (kr_dispatch_msg(conf->default_tableid) == -1)
392				quit = 1;
393		}
394
395		if (pfd[PFD_SOCK_PFKEY].revents & POLLIN) {
396			if (pfkey_read(keyfd, NULL) == -1) {
397				log_warnx("pfkey_read failed, exiting...");
398				quit = 1;
399			}
400		}
401
402		for (i = PFD_CONNECT_START; i < npfd; i++)
403			if (pfd[i].revents != 0)
404				bgpd_rtr_connect_done(pfd[i].fd, conf);
405
406 next_loop:
407		if (reconfig) {
408			u_int	error;
409
410			reconfig = 0;
411			switch (reconfigure(conffile, conf)) {
412			case -1:	/* fatal error */
413				quit = 1;
414				break;
415			case 0:		/* all OK */
416				error = 0;
417				break;
418			case 2:
419				log_info("previous reload still running");
420				error = CTL_RES_PENDING;
421				break;
422			default:	/* parse error */
423				log_warnx("config file %s has errors, "
424				    "not reloading", conffile);
425				error = CTL_RES_PARSE_ERROR;
426				break;
427			}
428			if (reconfpid != 0) {
429				send_imsg_session(IMSG_CTL_RESULT, reconfpid,
430				    &error, sizeof(error));
431				reconfpid = 0;
432			}
433		}
434
435		if (mrtdump) {
436			mrtdump = 0;
437			mrt_handler(conf->mrt);
438		}
439	}
440
441	/* close pipes */
442	if (ibuf_se) {
443		msgbuf_clear(&ibuf_se->w);
444		close(ibuf_se->fd);
445		free(ibuf_se);
446		ibuf_se = NULL;
447	}
448	if (ibuf_rde) {
449		msgbuf_clear(&ibuf_rde->w);
450		close(ibuf_rde->fd);
451		free(ibuf_rde);
452		ibuf_rde = NULL;
453	}
454	if (ibuf_rtr) {
455		msgbuf_clear(&ibuf_rtr->w);
456		close(ibuf_rtr->fd);
457		free(ibuf_rtr);
458		ibuf_rtr = NULL;
459	}
460
461	/* cleanup kernel data structures */
462	carp_demote_shutdown();
463	kr_shutdown(conf->default_tableid);
464	pftable_clear_all();
465
466	RB_FOREACH(p, peer_head, &conf->peers)
467		pfkey_remove(p);
468
469	while ((rr = SIMPLEQ_FIRST(&ribnames)) != NULL) {
470		SIMPLEQ_REMOVE_HEAD(&ribnames, entry);
471		free(rr);
472	}
473	free_config(conf);
474
475	log_debug("waiting for children to terminate");
476	do {
477		pid = wait(&status);
478		if (pid == -1) {
479			if (errno != EINTR && errno != ECHILD)
480				fatal("wait");
481		} else if (WIFSIGNALED(status)) {
482			char *name = "unknown process";
483			if (pid == rde_pid)
484				name = "route decision engine";
485			else if (pid == se_pid)
486				name = "session engine";
487			else if (pid == rtr_pid)
488				name = "rtr engine";
489			log_warnx("%s terminated; signal %d", name,
490				WTERMSIG(status));
491		}
492	} while (pid != -1 || (pid == -1 && errno == EINTR));
493
494	free(rcname);
495	free(cname);
496
497	log_info("terminating");
498	return (0);
499}
500
501pid_t
502start_child(enum bgpd_process p, char *argv0, int fd, int debug, int verbose)
503{
504	char *argv[5];
505	int argc = 0;
506	pid_t pid;
507
508	switch (pid = fork()) {
509	case -1:
510		fatal("cannot fork");
511	case 0:
512		break;
513	default:
514		close(fd);
515		return (pid);
516	}
517
518	if (fd != 3) {
519		if (dup2(fd, 3) == -1)
520			fatal("cannot setup imsg fd");
521	} else if (fcntl(fd, F_SETFD, 0) == -1)
522		fatal("cannot setup imsg fd");
523
524	argv[argc++] = argv0;
525	switch (p) {
526	case PROC_MAIN:
527		fatalx("Can not start main process");
528	case PROC_RDE:
529		argv[argc++] = "-R";
530		break;
531	case PROC_SE:
532		argv[argc++] = "-S";
533		break;
534	case PROC_RTR:
535		argv[argc++] = "-T";
536		break;
537	}
538	if (debug)
539		argv[argc++] = "-d";
540	if (verbose)
541		argv[argc++] = "-v";
542	argv[argc++] = NULL;
543
544	execvp(argv0, argv);
545	fatal("execvp");
546}
547
548int
549send_filterset(struct imsgbuf *i, struct filter_set_head *set)
550{
551	struct filter_set	*s;
552
553	TAILQ_FOREACH(s, set, entry)
554		if (imsg_compose(i, IMSG_FILTER_SET, 0, 0, -1, s,
555		    sizeof(struct filter_set)) == -1)
556			return (-1);
557	return (0);
558}
559
560int
561reconfigure(char *conffile, struct bgpd_config *conf)
562{
563	struct bgpd_config	*new_conf;
564
565	if (reconfpending)
566		return (2);
567
568	log_info("rereading config");
569	if ((new_conf = parse_config(conffile, &conf->peers,
570	    &conf->rtrs)) == NULL)
571		return (1);
572
573	merge_config(conf, new_conf);
574
575	if (prepare_listeners(conf) == -1) {
576		return (1);
577	}
578
579	if (control_setup(conf) == -1) {
580		return (1);
581	}
582
583	return send_config(conf);
584}
585
586int
587send_config(struct bgpd_config *conf)
588{
589	struct peer		*p;
590	struct filter_rule	*r;
591	struct listen_addr	*la;
592	struct rde_rib		*rr;
593	struct l3vpn		*vpn;
594	struct as_set		*aset;
595	struct prefixset	*ps;
596	struct prefixset_item	*psi, *npsi;
597	struct roa		*roa, *nroa;
598	struct rtr_config	*rtr;
599
600	reconfpending = 3;	/* one per child */
601
602	expand_networks(conf);
603
604	cflags = conf->flags;
605
606	/* start reconfiguration */
607	if (imsg_compose(ibuf_se, IMSG_RECONF_CONF, 0, 0, -1,
608	    conf, sizeof(*conf)) == -1)
609		return (-1);
610	if (imsg_compose(ibuf_rde, IMSG_RECONF_CONF, 0, 0, -1,
611	    conf, sizeof(*conf)) == -1)
612		return (-1);
613	if (imsg_compose(ibuf_rtr, IMSG_RECONF_CONF, 0, 0, -1,
614	    conf, sizeof(*conf)) == -1)
615		return (-1);
616
617	TAILQ_FOREACH(la, conf->listen_addrs, entry) {
618		if (imsg_compose(ibuf_se, IMSG_RECONF_LISTENER, 0, 0, la->fd,
619		    la, sizeof(*la)) == -1)
620			return (-1);
621		la->fd = -1;
622	}
623
624	/* adjust fib syncing on reload */
625	ktable_preload();
626
627	/* RIBs for the RDE */
628	while ((rr = SIMPLEQ_FIRST(&ribnames))) {
629		SIMPLEQ_REMOVE_HEAD(&ribnames, entry);
630		if (ktable_update(rr->rtableid, rr->name, rr->flags) == -1) {
631			log_warnx("failed to load routing table %d",
632			    rr->rtableid);
633			return (-1);
634		}
635		if (imsg_compose(ibuf_rde, IMSG_RECONF_RIB, 0, 0, -1,
636		    rr, sizeof(*rr)) == -1)
637			return (-1);
638		free(rr);
639	}
640
641	/* send peer list to the SE */
642	RB_FOREACH(p, peer_head, &conf->peers) {
643		if (imsg_compose(ibuf_se, IMSG_RECONF_PEER, p->conf.id, 0, -1,
644		    &p->conf, sizeof(p->conf)) == -1)
645			return (-1);
646
647		if (p->reconf_action == RECONF_REINIT)
648			if (pfkey_establish(p) == -1)
649				log_peer_warnx(&p->conf, "pfkey setup failed");
650	}
651
652	/* networks go via kroute to the RDE */
653	kr_net_reload(conf->default_tableid, 0, &conf->networks);
654
655	/* prefixsets for filters in the RDE */
656	while ((ps = SIMPLEQ_FIRST(&conf->prefixsets)) != NULL) {
657		SIMPLEQ_REMOVE_HEAD(&conf->prefixsets, entry);
658		if (imsg_compose(ibuf_rde, IMSG_RECONF_PREFIX_SET, 0, 0, -1,
659		    ps->name, sizeof(ps->name)) == -1)
660			return (-1);
661		RB_FOREACH_SAFE(psi, prefixset_tree, &ps->psitems, npsi) {
662			RB_REMOVE(prefixset_tree, &ps->psitems, psi);
663			if (imsg_compose(ibuf_rde, IMSG_RECONF_PREFIX_SET_ITEM,
664			    0, 0, -1, psi, sizeof(*psi)) == -1)
665				return (-1);
666			free(psi);
667		}
668		free(ps);
669	}
670
671	/* originsets for filters in the RDE */
672	while ((ps = SIMPLEQ_FIRST(&conf->originsets)) != NULL) {
673		SIMPLEQ_REMOVE_HEAD(&conf->originsets, entry);
674		if (imsg_compose(ibuf_rde, IMSG_RECONF_ORIGIN_SET, 0, 0, -1,
675		    ps->name, sizeof(ps->name)) == -1)
676			return (-1);
677		RB_FOREACH_SAFE(roa, roa_tree, &ps->roaitems, nroa) {
678			RB_REMOVE(roa_tree, &ps->roaitems, roa);
679			if (imsg_compose(ibuf_rde, IMSG_RECONF_ROA_ITEM, 0, 0,
680			    -1, roa, sizeof(*roa)) == -1)
681				return (-1);
682			free(roa);
683		}
684		free(ps);
685	}
686
687	/* roa table and rtr config are sent to the RTR engine */
688	RB_FOREACH_SAFE(roa, roa_tree, &conf->roa, nroa) {
689		RB_REMOVE(roa_tree, &conf->roa, roa);
690		if (imsg_compose(ibuf_rtr, IMSG_RECONF_ROA_ITEM, 0, 0,
691		    -1, roa, sizeof(*roa)) == -1)
692			return (-1);
693		free(roa);
694	}
695	SIMPLEQ_FOREACH(rtr, &conf->rtrs, entry) {
696		if (imsg_compose(ibuf_rtr, IMSG_RECONF_RTR_CONFIG, rtr->id,
697		    0, -1, rtr->descr, sizeof(rtr->descr)) == -1)
698			return (-1);
699	}
700
701	/* as-sets for filters in the RDE */
702	while ((aset = SIMPLEQ_FIRST(&conf->as_sets)) != NULL) {
703		struct ibuf *wbuf;
704		uint32_t *as;
705		size_t i, l, n;
706
707		SIMPLEQ_REMOVE_HEAD(&conf->as_sets, entry);
708
709		as = set_get(aset->set, &n);
710		if ((wbuf = imsg_create(ibuf_rde, IMSG_RECONF_AS_SET, 0, 0,
711		    sizeof(n) + sizeof(aset->name))) == NULL)
712			return -1;
713		if (imsg_add(wbuf, &n, sizeof(n)) == -1 ||
714		    imsg_add(wbuf, aset->name, sizeof(aset->name)) == -1)
715			return -1;
716		imsg_close(ibuf_rde, wbuf);
717
718		for (i = 0; i < n; i += l) {
719			l = (n - i > 1024 ? 1024 : n - i);
720			if (imsg_compose(ibuf_rde, IMSG_RECONF_AS_SET_ITEMS,
721			    0, 0, -1, as + i, l * sizeof(*as)) == -1)
722				return -1;
723		}
724
725		if (imsg_compose(ibuf_rde, IMSG_RECONF_AS_SET_DONE, 0, 0, -1,
726		    NULL, 0) == -1)
727			return -1;
728
729		set_free(aset->set);
730		free(aset);
731	}
732
733	/* filters for the RDE */
734	while ((r = TAILQ_FIRST(conf->filters)) != NULL) {
735		TAILQ_REMOVE(conf->filters, r, entry);
736		if (send_filterset(ibuf_rde, &r->set) == -1)
737			return (-1);
738		if (imsg_compose(ibuf_rde, IMSG_RECONF_FILTER, 0, 0, -1,
739		    r, sizeof(struct filter_rule)) == -1)
740			return (-1);
741		filterset_free(&r->set);
742		free(r);
743	}
744
745	while ((vpn = SIMPLEQ_FIRST(&conf->l3vpns)) != NULL) {
746		SIMPLEQ_REMOVE_HEAD(&conf->l3vpns, entry);
747		if (ktable_update(vpn->rtableid, vpn->descr, vpn->flags) ==
748		    -1) {
749			log_warnx("failed to load routing table %d",
750			    vpn->rtableid);
751			return (-1);
752		}
753		/* networks go via kroute to the RDE */
754		kr_net_reload(vpn->rtableid, vpn->rd, &vpn->net_l);
755
756		if (imsg_compose(ibuf_rde, IMSG_RECONF_VPN, 0, 0, -1,
757		    vpn, sizeof(*vpn)) == -1)
758			return (-1);
759
760		/* export targets */
761		if (send_filterset(ibuf_rde, &vpn->export) == -1)
762			return (-1);
763		if (imsg_compose(ibuf_rde, IMSG_RECONF_VPN_EXPORT, 0, 0,
764		    -1, NULL, 0) == -1)
765			return (-1);
766		filterset_free(&vpn->export);
767
768		/* import targets */
769		if (send_filterset(ibuf_rde, &vpn->import) == -1)
770			return (-1);
771		if (imsg_compose(ibuf_rde, IMSG_RECONF_VPN_IMPORT, 0, 0,
772		    -1, NULL, 0) == -1)
773			return (-1);
774		filterset_free(&vpn->import);
775
776		if (imsg_compose(ibuf_rde, IMSG_RECONF_VPN_DONE, 0, 0,
777		    -1, NULL, 0) == -1)
778			return (-1);
779
780		free(vpn);
781	}
782
783	/* send a drain message to know when all messages where processed */
784	if (imsg_compose(ibuf_se, IMSG_RECONF_DRAIN, 0, 0, -1, NULL, 0) == -1)
785		return (-1);
786	if (imsg_compose(ibuf_rde, IMSG_RECONF_DRAIN, 0, 0, -1, NULL, 0) == -1)
787		return (-1);
788	if (imsg_compose(ibuf_rtr, IMSG_RECONF_DRAIN, 0, 0, -1, NULL, 0) == -1)
789		return (-1);
790
791	/* mrt changes can be sent out of bound */
792	mrt_reconfigure(conf->mrt);
793	return (0);
794}
795
796int
797dispatch_imsg(struct imsgbuf *ibuf, int idx, struct bgpd_config *conf)
798{
799	struct imsg		 imsg;
800	struct peer		*p;
801	struct rtr_config	*r;
802	ssize_t			 n;
803	int			 rv, verbose;
804
805	rv = 0;
806	while (ibuf) {
807		if ((n = imsg_get(ibuf, &imsg)) == -1)
808			return (-1);
809
810		if (n == 0)
811			break;
812
813		switch (imsg.hdr.type) {
814		case IMSG_KROUTE_CHANGE:
815			if (idx != PFD_PIPE_RDE)
816				log_warnx("route request not from RDE");
817			else if (imsg.hdr.len != IMSG_HEADER_SIZE +
818			    sizeof(struct kroute_full))
819				log_warnx("wrong imsg len");
820			else if (kr_change(imsg.hdr.peerid, imsg.data))
821				rv = -1;
822			break;
823		case IMSG_KROUTE_DELETE:
824			if (idx != PFD_PIPE_RDE)
825				log_warnx("route request not from RDE");
826			else if (imsg.hdr.len != IMSG_HEADER_SIZE +
827			    sizeof(struct kroute_full))
828				log_warnx("wrong imsg len");
829			else if (kr_delete(imsg.hdr.peerid, imsg.data))
830				rv = -1;
831			break;
832		case IMSG_KROUTE_FLUSH:
833			if (idx != PFD_PIPE_RDE)
834				log_warnx("route request not from RDE");
835			else if (imsg.hdr.len != IMSG_HEADER_SIZE)
836				log_warnx("wrong imsg len");
837			else if (kr_flush(imsg.hdr.peerid))
838				rv = -1;
839			break;
840		case IMSG_NEXTHOP_ADD:
841			if (idx != PFD_PIPE_RDE)
842				log_warnx("nexthop request not from RDE");
843			else if (imsg.hdr.len != IMSG_HEADER_SIZE +
844			    sizeof(struct bgpd_addr))
845				log_warnx("wrong imsg len");
846			else if (kr_nexthop_add(imsg.hdr.peerid, imsg.data,
847			    conf) == -1)
848				rv = -1;
849			break;
850		case IMSG_NEXTHOP_REMOVE:
851			if (idx != PFD_PIPE_RDE)
852				log_warnx("nexthop request not from RDE");
853			else if (imsg.hdr.len != IMSG_HEADER_SIZE +
854			    sizeof(struct bgpd_addr))
855				log_warnx("wrong imsg len");
856			else
857				kr_nexthop_delete(imsg.hdr.peerid, imsg.data,
858				    conf);
859			break;
860		case IMSG_PFTABLE_ADD:
861			if (idx != PFD_PIPE_RDE)
862				log_warnx("pftable request not from RDE");
863			else
864				if (imsg.hdr.len != IMSG_HEADER_SIZE +
865				    sizeof(struct pftable_msg))
866					log_warnx("wrong imsg len");
867				else if (pftable_addr_add(imsg.data) != 0)
868					rv = -1;
869			break;
870		case IMSG_PFTABLE_REMOVE:
871			if (idx != PFD_PIPE_RDE)
872				log_warnx("pftable request not from RDE");
873			else
874				if (imsg.hdr.len != IMSG_HEADER_SIZE +
875				    sizeof(struct pftable_msg))
876					log_warnx("wrong imsg len");
877				else if (pftable_addr_remove(imsg.data) != 0)
878					rv = -1;
879			break;
880		case IMSG_PFTABLE_COMMIT:
881			if (idx != PFD_PIPE_RDE)
882				log_warnx("pftable request not from RDE");
883			else if (imsg.hdr.len != IMSG_HEADER_SIZE)
884				log_warnx("wrong imsg len");
885			else if (pftable_commit() != 0)
886				rv = -1;
887			break;
888		case IMSG_PFKEY_RELOAD:
889			if (idx != PFD_PIPE_SESSION) {
890				log_warnx("pfkey reload request not from SE");
891				break;
892			}
893			p = getpeerbyid(conf, imsg.hdr.peerid);
894			if (p != NULL) {
895				if (pfkey_establish(p) == -1)
896					log_peer_warnx(&p->conf,
897					    "pfkey setup failed");
898			}
899			break;
900		case IMSG_CTL_RELOAD:
901			if (idx != PFD_PIPE_SESSION)
902				log_warnx("reload request not from SE");
903			else {
904				reconfig = 1;
905				reconfpid = imsg.hdr.pid;
906				if (imsg.hdr.len == IMSG_HEADER_SIZE +
907				    REASON_LEN && ((char *)imsg.data)[0])
908					log_info("reload due to: %s",
909					    log_reason(imsg.data));
910			}
911			break;
912		case IMSG_CTL_FIB_COUPLE:
913			if (idx != PFD_PIPE_SESSION)
914				log_warnx("couple request not from SE");
915			else
916				kr_fib_couple(imsg.hdr.peerid);
917			break;
918		case IMSG_CTL_FIB_DECOUPLE:
919			if (idx != PFD_PIPE_SESSION)
920				log_warnx("decouple request not from SE");
921			else
922				kr_fib_decouple(imsg.hdr.peerid);
923			break;
924		case IMSG_CTL_KROUTE:
925		case IMSG_CTL_KROUTE_ADDR:
926		case IMSG_CTL_SHOW_NEXTHOP:
927		case IMSG_CTL_SHOW_INTERFACE:
928		case IMSG_CTL_SHOW_FIB_TABLES:
929			if (idx != PFD_PIPE_SESSION)
930				log_warnx("kroute request not from SE");
931			else
932				kr_show_route(&imsg);
933			break;
934		case IMSG_IFINFO:
935			if (idx != PFD_PIPE_SESSION)
936				log_warnx("IFINFO request not from SE");
937			else if (imsg.hdr.len != IMSG_HEADER_SIZE + IFNAMSIZ)
938				log_warnx("IFINFO request with wrong len");
939			else
940				kr_ifinfo(imsg.data);
941			break;
942		case IMSG_DEMOTE:
943			if (idx != PFD_PIPE_SESSION)
944				log_warnx("demote request not from SE");
945			else if (imsg.hdr.len != IMSG_HEADER_SIZE +
946			    sizeof(struct demote_msg))
947				log_warnx("DEMOTE request with wrong len");
948			else {
949				struct demote_msg	*msg;
950
951				msg = imsg.data;
952				carp_demote_set(msg->demote_group, msg->level);
953			}
954			break;
955		case IMSG_CTL_LOG_VERBOSE:
956			/* already checked by SE */
957			memcpy(&verbose, imsg.data, sizeof(verbose));
958			log_setverbose(verbose);
959			break;
960		case IMSG_RECONF_DONE:
961			if (reconfpending == 0) {
962				log_warnx("unexpected RECONF_DONE received");
963				break;
964			}
965			if (idx == PFD_PIPE_SESSION) {
966				imsg_compose(ibuf_rtr, IMSG_RECONF_DONE, 0,
967				    0, -1, NULL, 0);
968			} else if (idx == PFD_PIPE_RTR) {
969				imsg_compose(ibuf_rde, IMSG_RECONF_DONE, 0,
970				    0, -1, NULL, 0);
971
972				/* finally fix kroute information */
973				ktable_postload();
974
975				/* redistribute list needs to be reloaded too */
976				kr_reload();
977			}
978			reconfpending--;
979			break;
980		case IMSG_RECONF_DRAIN:
981			if (reconfpending == 0) {
982				log_warnx("unexpected RECONF_DRAIN received");
983				break;
984			}
985			reconfpending--;
986			if (reconfpending == 0) {
987				/*
988				 * SE goes first to bring templated neighbors
989				 * in sync.
990				 */
991				imsg_compose(ibuf_se, IMSG_RECONF_DONE, 0,
992				    0, -1, NULL, 0);
993				reconfpending = 3; /* expecting 2 DONE msg */
994			}
995			break;
996		case IMSG_SOCKET_CONN:
997			if (idx != PFD_PIPE_RTR) {
998				log_warnx("connect request not from RTR");
999			} else {
1000				SIMPLEQ_FOREACH(r, &conf->rtrs, entry) {
1001					if (imsg.hdr.peerid == r->id)
1002						break;
1003				}
1004				if (r == NULL)
1005					log_warnx("unknown rtr id %d",
1006					    imsg.hdr.peerid);
1007				else
1008					bgpd_rtr_connect(r);
1009			}
1010			break;
1011		case IMSG_CTL_SHOW_RTR:
1012			if (idx == PFD_PIPE_SESSION) {
1013				SIMPLEQ_FOREACH(r, &conf->rtrs, entry) {
1014					imsg_compose(ibuf_rtr, imsg.hdr.type,
1015					    r->id, imsg.hdr.pid, -1, NULL, 0);
1016				}
1017				imsg_compose(ibuf_rtr, IMSG_CTL_END,
1018				    0, imsg.hdr.pid, -1, NULL, 0);
1019			} else if (imsg.hdr.len != IMSG_HEADER_SIZE +
1020			    sizeof(struct ctl_show_rtr)) {
1021				log_warnx("IMSG_CTL_SHOW_RTR with wrong len");
1022			} else if (idx == PFD_PIPE_RTR) {
1023				SIMPLEQ_FOREACH(r, &conf->rtrs, entry) {
1024					if (imsg.hdr.peerid == r->id)
1025						break;
1026				}
1027				if (r != NULL) {
1028					struct ctl_show_rtr *msg;
1029					msg = imsg.data;
1030					strlcpy(msg->descr, r->descr,
1031					    sizeof(msg->descr));
1032					msg->local_addr = r->local_addr;
1033					msg->remote_addr = r->remote_addr;
1034					msg->remote_port = r->remote_port;
1035
1036					imsg_compose(ibuf_se, imsg.hdr.type,
1037					    imsg.hdr.peerid, imsg.hdr.pid,
1038					    -1, imsg.data,
1039					    imsg.hdr.len - IMSG_HEADER_SIZE);
1040				}
1041			}
1042			break;
1043		case IMSG_CTL_END:
1044		case IMSG_CTL_SHOW_TIMER:
1045			if (idx != PFD_PIPE_RTR) {
1046				log_warnx("connect request not from RTR");
1047				break;
1048			}
1049			imsg_compose(ibuf_se, imsg.hdr.type, imsg.hdr.peerid,
1050			    imsg.hdr.pid, -1, imsg.data,
1051			    imsg.hdr.len - IMSG_HEADER_SIZE);
1052			break;
1053		default:
1054			break;
1055		}
1056		imsg_free(&imsg);
1057		if (rv != 0)
1058			return (rv);
1059	}
1060	return (0);
1061}
1062
1063void
1064send_nexthop_update(struct kroute_nexthop *msg)
1065{
1066	char	*gw = NULL;
1067
1068	if (msg->gateway.aid)
1069		if (asprintf(&gw, ": via %s",
1070		    log_addr(&msg->gateway)) == -1) {
1071			log_warn("send_nexthop_update");
1072			quit = 1;
1073		}
1074
1075	log_debug("nexthop %s now %s%s%s", log_addr(&msg->nexthop),
1076	    msg->valid ? "valid" : "invalid",
1077	    msg->connected ? ": directly connected" : "",
1078	    msg->gateway.aid ? gw : "");
1079
1080	free(gw);
1081
1082	if (imsg_compose(ibuf_rde, IMSG_NEXTHOP_UPDATE, 0, 0, -1,
1083	    msg, sizeof(struct kroute_nexthop)) == -1)
1084		quit = 1;
1085}
1086
1087void
1088send_imsg_session(int type, pid_t pid, void *data, uint16_t datalen)
1089{
1090	imsg_compose(ibuf_se, type, 0, pid, -1, data, datalen);
1091}
1092
1093int
1094send_network(int type, struct network_config *net, struct filter_set_head *h)
1095{
1096	if (quit)
1097		return (0);
1098	if (imsg_compose(ibuf_rde, type, 0, 0, -1, net,
1099	    sizeof(struct network_config)) == -1)
1100		return (-1);
1101	/* networks that get deleted don't need to send the filter set */
1102	if (type == IMSG_NETWORK_REMOVE)
1103		return (0);
1104	if (send_filterset(ibuf_rde, h) == -1)
1105		return (-1);
1106	if (imsg_compose(ibuf_rde, IMSG_NETWORK_DONE, 0, 0, -1, NULL, 0) == -1)
1107		return (-1);
1108
1109	return (0);
1110}
1111
1112int
1113bgpd_filternexthop(struct kroute *kr, struct kroute6 *kr6)
1114{
1115	/* kernel routes are never filtered */
1116	if (kr && kr->flags & F_KERNEL && kr->prefixlen != 0)
1117		return (0);
1118	if (kr6 && kr6->flags & F_KERNEL && kr6->prefixlen != 0)
1119		return (0);
1120
1121	if (cflags & BGPD_FLAG_NEXTHOP_BGP) {
1122		if (kr && kr->flags & F_BGPD_INSERTED)
1123			return (0);
1124		if (kr6 && kr6->flags & F_BGPD_INSERTED)
1125			return (0);
1126	}
1127
1128	if (cflags & BGPD_FLAG_NEXTHOP_DEFAULT) {
1129		if (kr && kr->prefixlen == 0)
1130			return (0);
1131		if (kr6 && kr6->prefixlen == 0)
1132			return (0);
1133	}
1134
1135	return (1);
1136}
1137
1138int
1139control_setup(struct bgpd_config *conf)
1140{
1141	int fd, restricted;
1142
1143	/* control socket is outside chroot */
1144	if (!cname || strcmp(cname, conf->csock)) {
1145		if (cname) {
1146			free(cname);
1147		}
1148		if ((cname = strdup(conf->csock)) == NULL)
1149			fatal("strdup");
1150		if (control_check(cname) == -1)
1151			return (-1);
1152		if ((fd = control_init(0, cname)) == -1)
1153			fatalx("control socket setup failed");
1154		if (control_listen(fd) == -1)
1155			fatalx("control socket setup failed");
1156		restricted = 0;
1157		if (imsg_compose(ibuf_se, IMSG_RECONF_CTRL, 0, 0, fd,
1158		    &restricted, sizeof(restricted)) == -1)
1159			return (-1);
1160	}
1161	if (!conf->rcsock) {
1162		/* remove restricted socket */
1163		free(rcname);
1164		rcname = NULL;
1165	} else if (!rcname || strcmp(rcname, conf->rcsock)) {
1166		if (rcname) {
1167			free(rcname);
1168		}
1169		if ((rcname = strdup(conf->rcsock)) == NULL)
1170			fatal("strdup");
1171		if (control_check(rcname) == -1)
1172			return (-1);
1173		if ((fd = control_init(1, rcname)) == -1)
1174			fatalx("control socket setup failed");
1175		if (control_listen(fd) == -1)
1176			fatalx("control socket setup failed");
1177		restricted = 1;
1178		if (imsg_compose(ibuf_se, IMSG_RECONF_CTRL, 0, 0, fd,
1179		    &restricted, sizeof(restricted)) == -1)
1180			return (-1);
1181	}
1182	return (0);
1183}
1184
1185void
1186set_pollfd(struct pollfd *pfd, struct imsgbuf *i)
1187{
1188	if (i == NULL || i->fd == -1) {
1189		pfd->fd = -1;
1190		return;
1191	}
1192	pfd->fd = i->fd;
1193	pfd->events = POLLIN;
1194	if (i->w.queued > 0)
1195		pfd->events |= POLLOUT;
1196}
1197
1198int
1199handle_pollfd(struct pollfd *pfd, struct imsgbuf *i)
1200{
1201	ssize_t n;
1202
1203	if (i == NULL)
1204		return (0);
1205
1206	if (pfd->revents & POLLOUT)
1207		if (msgbuf_write(&i->w) <= 0 && errno != EAGAIN) {
1208			log_warn("imsg write error");
1209			close(i->fd);
1210			i->fd = -1;
1211			return (-1);
1212		}
1213
1214	if (pfd->revents & POLLIN) {
1215		if ((n = imsg_read(i)) == -1 && errno != EAGAIN) {
1216			log_warn("imsg read error");
1217			close(i->fd);
1218			i->fd = -1;
1219			return (-1);
1220		}
1221		if (n == 0) {
1222			log_warnx("peer closed imsg connection");
1223			close(i->fd);
1224			i->fd = -1;
1225			return (-1);
1226		}
1227	}
1228	return (0);
1229}
1230
1231static void
1232getsockpair(int pipe[2])
1233{
1234	int bsize, i;
1235
1236	if (socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK,
1237	    PF_UNSPEC, pipe) == -1)
1238		fatal("socketpair");
1239
1240	for (i = 0; i < 2; i++) {
1241		for (bsize = MAX_SOCK_BUF; bsize >= 16 * 1024; bsize /= 2) {
1242			if (setsockopt(pipe[i], SOL_SOCKET, SO_RCVBUF,
1243			    &bsize, sizeof(bsize)) == -1) {
1244				if (errno != ENOBUFS)
1245					fatal("setsockopt(SO_RCVBUF, %d)",
1246					    bsize);
1247				log_warn("setsockopt(SO_RCVBUF, %d)", bsize);
1248				continue;
1249			}
1250			break;
1251		}
1252	}
1253	for (i = 0; i < 2; i++) {
1254		for (bsize = MAX_SOCK_BUF; bsize >= 16 * 1024; bsize /= 2) {
1255			if (setsockopt(pipe[i], SOL_SOCKET, SO_SNDBUF,
1256			    &bsize, sizeof(bsize)) == -1) {
1257				if (errno != ENOBUFS)
1258					fatal("setsockopt(SO_SNDBUF, %d)",
1259					    bsize);
1260				log_warn("setsockopt(SO_SNDBUF, %d)", bsize);
1261				continue;
1262			}
1263			break;
1264		}
1265	}
1266}
1267
1268int
1269imsg_send_sockets(struct imsgbuf *se, struct imsgbuf *rde, struct imsgbuf *roa)
1270{
1271	int pipe_s2r[2];
1272	int pipe_s2r_ctl[2];
1273	int pipe_r2r[2];
1274
1275	getsockpair(pipe_s2r);
1276	getsockpair(pipe_s2r_ctl);
1277	getsockpair(pipe_r2r);
1278
1279	if (imsg_compose(se, IMSG_SOCKET_CONN, 0, 0, pipe_s2r[0],
1280	    NULL, 0) == -1)
1281		return (-1);
1282	if (imsg_compose(rde, IMSG_SOCKET_CONN, 0, 0, pipe_s2r[1],
1283	    NULL, 0) == -1)
1284		return (-1);
1285
1286	if (imsg_compose(se, IMSG_SOCKET_CONN_CTL, 0, 0, pipe_s2r_ctl[0],
1287	    NULL, 0) == -1)
1288		return (-1);
1289	if (imsg_compose(rde, IMSG_SOCKET_CONN_CTL, 0, 0, pipe_s2r_ctl[1],
1290	    NULL, 0) == -1)
1291		return (-1);
1292
1293	if (imsg_compose(roa, IMSG_SOCKET_CONN_RTR, 0, 0, pipe_r2r[0],
1294	    NULL, 0) == -1)
1295		return (-1);
1296	if (imsg_compose(rde, IMSG_SOCKET_CONN_RTR, 0, 0, pipe_r2r[1],
1297	    NULL, 0) == -1)
1298		return (-1);
1299
1300	return (0);
1301}
1302
1303void
1304bgpd_rtr_connect(struct rtr_config *r)
1305{
1306	struct connect_elm *ce;
1307	struct sockaddr *sa;
1308	socklen_t len;
1309
1310	if (connect_cnt >= MAX_CONNECT_CNT) {
1311		log_warnx("rtr %s: too many concurrent connection requests",
1312		    r->descr);
1313		return;
1314	}
1315
1316	if ((ce = calloc(1, sizeof(*ce))) == NULL) {
1317		log_warn("rtr %s", r->descr);
1318		return;
1319	}
1320
1321	ce->id = r->id;
1322	ce->fd = socket(aid2af(r->remote_addr.aid),
1323	     SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, IPPROTO_TCP);
1324	if (ce->fd == -1) {
1325		log_warn("rtr %s", r->descr);
1326		free(ce);
1327		return;
1328	}
1329
1330	if ((sa = addr2sa(&r->local_addr, 0, &len)) != NULL) {
1331		if (bind(ce->fd, sa, len) == -1) {
1332			log_warn("rtr %s: bind to %s", r->descr,
1333			    log_addr(&r->local_addr));
1334			close(ce->fd);
1335			free(ce);
1336			return;
1337		}
1338	}
1339
1340	sa = addr2sa(&r->remote_addr, r->remote_port, &len);
1341	if (connect(ce->fd, sa, len) == -1) {
1342		if (errno != EINPROGRESS) {
1343			log_warn("rtr %s: connect to %s:%u", r->descr,
1344			    log_addr(&r->remote_addr), r->remote_port);
1345			close(ce->fd);
1346			free(ce);
1347			return;
1348		}
1349		TAILQ_INSERT_TAIL(&connect_queue, ce, entry);
1350		connect_cnt++;
1351		return;
1352	}
1353
1354	imsg_compose(ibuf_rtr, IMSG_SOCKET_CONN, ce->id, 0, ce->fd, NULL, 0);
1355	free(ce);
1356}
1357
1358void
1359bgpd_rtr_connect_done(int fd, struct bgpd_config *conf)
1360{
1361	struct rtr_config *r;
1362	struct connect_elm *ce;
1363	int error = 0;
1364	socklen_t len;
1365
1366	TAILQ_FOREACH(ce, &connect_queue, entry) {
1367		if (ce->fd == fd)
1368			break;
1369	}
1370	if (ce == NULL)
1371		fatalx("connect entry not found");
1372
1373	TAILQ_REMOVE(&connect_queue, ce, entry);
1374	connect_cnt--;
1375
1376	SIMPLEQ_FOREACH(r, &conf->rtrs, entry) {
1377		if (ce->id == r->id)
1378			break;
1379	}
1380	if (r == NULL) {
1381		log_warnx("rtr id %d no longer exists", ce->id);
1382		goto fail;
1383	}
1384
1385	len = sizeof(error);
1386	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) == -1) {
1387		log_warn("rtr %s: getsockopt SO_ERROR", r->descr);
1388		goto fail;
1389	}
1390
1391	if (error != 0) {
1392		errno = error;
1393		log_warn("rtr %s: connect to %s:%u", r->descr,
1394		    log_addr(&r->remote_addr), r->remote_port);
1395		goto fail;
1396	}
1397
1398	imsg_compose(ibuf_rtr, IMSG_SOCKET_CONN, ce->id, 0, ce->fd, NULL, 0);
1399	free(ce);
1400	return;
1401
1402fail:
1403	close(fd);
1404	free(ce);
1405}
1406