hastd.c revision 211899
1/*-
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
4 * All rights reserved.
5 *
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 211899 2010-08-27 21:28:02Z pjd $");
33
34#include <sys/param.h>
35#include <sys/linker.h>
36#include <sys/module.h>
37#include <sys/wait.h>
38
39#include <assert.h>
40#include <err.h>
41#include <errno.h>
42#include <libutil.h>
43#include <signal.h>
44#include <stdbool.h>
45#include <stdio.h>
46#include <stdlib.h>
47#include <string.h>
48#include <sysexits.h>
49#include <unistd.h>
50
51#include <activemap.h>
52#include <pjdlog.h>
53
54#include "control.h"
55#include "hast.h"
56#include "hast_proto.h"
57#include "hastd.h"
58#include "subr.h"
59
60/* Path to configuration file. */
61const char *cfgpath = HAST_CONFIG;
62/* Hastd configuration. */
63static struct hastd_config *cfg;
64/* Was SIGCHLD signal received? */
65bool sigchld_received = false;
66/* Was SIGHUP signal received? */
67bool sighup_received = false;
68/* Was SIGINT or SIGTERM signal received? */
69bool sigexit_received = false;
70/* PID file handle. */
71struct pidfh *pfh;
72
73static void
74usage(void)
75{
76
77	errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
78}
79
80static void
81sighandler(int sig)
82{
83
84	switch (sig) {
85	case SIGINT:
86	case SIGTERM:
87		sigexit_received = true;
88		break;
89	case SIGCHLD:
90		sigchld_received = true;
91		break;
92	case SIGHUP:
93		sighup_received = true;
94		break;
95	default:
96		assert(!"invalid condition");
97	}
98}
99
100static void
101g_gate_load(void)
102{
103
104	if (modfind("g_gate") == -1) {
105		/* Not present in kernel, try loading it. */
106		if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) {
107			if (errno != EEXIST) {
108				pjdlog_exit(EX_OSERR,
109				    "Unable to load geom_gate module");
110			}
111		}
112	}
113}
114
115static void
116child_exit_log(unsigned int pid, int status)
117{
118
119	if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
120		pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
121		    pid);
122	} else if (WIFSIGNALED(status)) {
123		pjdlog_error("Worker process killed (pid=%u, signal=%d).",
124		    pid, WTERMSIG(status));
125	} else {
126		pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
127		    pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
128	}
129}
130
131static void
132child_exit(void)
133{
134	struct hast_resource *res;
135	int status;
136	pid_t pid;
137
138	while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
139		/* Find resource related to the process that just exited. */
140		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
141			if (pid == res->hr_workerpid)
142				break;
143		}
144		if (res == NULL) {
145			/*
146			 * This can happen when new connection arrives and we
147			 * cancel child responsible for the old one.
148			 */
149			continue;
150		}
151		pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
152		    role2str(res->hr_role));
153		child_exit_log(pid, status);
154		proto_close(res->hr_ctrl);
155		res->hr_workerpid = 0;
156		if (res->hr_role == HAST_ROLE_PRIMARY) {
157			/*
158			 * Restart child process if it was killed by signal
159			 * or exited because of temporary problem.
160			 */
161			if (WIFSIGNALED(status) ||
162			    (WIFEXITED(status) &&
163			     WEXITSTATUS(status) == EX_TEMPFAIL)) {
164				sleep(1);
165				pjdlog_info("Restarting worker process.");
166				hastd_primary(res);
167			} else {
168				res->hr_role = HAST_ROLE_INIT;
169				pjdlog_info("Changing resource role back to %s.",
170				    role2str(res->hr_role));
171			}
172		}
173		pjdlog_prefix_set("%s", "");
174	}
175}
176
177static bool
178resource_needs_restart(const struct hast_resource *res0,
179    const struct hast_resource *res1)
180{
181
182	assert(strcmp(res0->hr_name, res1->hr_name) == 0);
183
184	if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
185		return (true);
186	if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
187		return (true);
188	if (res0->hr_role == HAST_ROLE_INIT ||
189	    res0->hr_role == HAST_ROLE_SECONDARY) {
190		if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
191			return (true);
192		if (res0->hr_replication != res1->hr_replication)
193			return (true);
194		if (res0->hr_timeout != res1->hr_timeout)
195			return (true);
196		if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
197			return (true);
198	}
199	return (false);
200}
201
202static bool
203resource_needs_reload(const struct hast_resource *res0,
204    const struct hast_resource *res1)
205{
206
207	assert(strcmp(res0->hr_name, res1->hr_name) == 0);
208	assert(strcmp(res0->hr_provname, res1->hr_provname) == 0);
209	assert(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
210
211	if (res0->hr_role != HAST_ROLE_PRIMARY)
212		return (false);
213
214	if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
215		return (true);
216	if (res0->hr_replication != res1->hr_replication)
217		return (true);
218	if (res0->hr_timeout != res1->hr_timeout)
219		return (true);
220	if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
221		return (true);
222	return (false);
223}
224
225static void
226hastd_reload(void)
227{
228	struct hastd_config *newcfg;
229	struct hast_resource *nres, *cres, *tres;
230	uint8_t role;
231
232	pjdlog_info("Reloading configuration...");
233
234	newcfg = yy_config_parse(cfgpath, false);
235	if (newcfg == NULL)
236		goto failed;
237
238	/*
239	 * Check if control address has changed.
240	 */
241	if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
242		if (proto_server(newcfg->hc_controladdr,
243		    &newcfg->hc_controlconn) < 0) {
244			pjdlog_errno(LOG_ERR,
245			    "Unable to listen on control address %s",
246			    newcfg->hc_controladdr);
247			goto failed;
248		}
249	}
250	/*
251	 * Check if listen address has changed.
252	 */
253	if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) {
254		if (proto_server(newcfg->hc_listenaddr,
255		    &newcfg->hc_listenconn) < 0) {
256			pjdlog_errno(LOG_ERR, "Unable to listen on address %s",
257			    newcfg->hc_listenaddr);
258			goto failed;
259		}
260	}
261	/*
262	 * Only when both control and listen sockets are successfully
263	 * initialized switch them to new configuration.
264	 */
265	if (newcfg->hc_controlconn != NULL) {
266		pjdlog_info("Control socket changed from %s to %s.",
267		    cfg->hc_controladdr, newcfg->hc_controladdr);
268		proto_close(cfg->hc_controlconn);
269		cfg->hc_controlconn = newcfg->hc_controlconn;
270		newcfg->hc_controlconn = NULL;
271		strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
272		    sizeof(cfg->hc_controladdr));
273	}
274	if (newcfg->hc_listenconn != NULL) {
275		pjdlog_info("Listen socket changed from %s to %s.",
276		    cfg->hc_listenaddr, newcfg->hc_listenaddr);
277		proto_close(cfg->hc_listenconn);
278		cfg->hc_listenconn = newcfg->hc_listenconn;
279		newcfg->hc_listenconn = NULL;
280		strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr,
281		    sizeof(cfg->hc_listenaddr));
282	}
283
284	/*
285	 * Stop and remove resources that were removed from the configuration.
286	 */
287	TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
288		TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
289			if (strcmp(cres->hr_name, nres->hr_name) == 0)
290				break;
291		}
292		if (nres == NULL) {
293			control_set_role(cres, HAST_ROLE_INIT);
294			TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
295			pjdlog_info("Resource %s removed.", cres->hr_name);
296			free(cres);
297		}
298	}
299	/*
300	 * Move new resources to the current configuration.
301	 */
302	TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
303		TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
304			if (strcmp(cres->hr_name, nres->hr_name) == 0)
305				break;
306		}
307		if (cres == NULL) {
308			TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
309			TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
310			pjdlog_info("Resource %s added.", nres->hr_name);
311		}
312	}
313	/*
314	 * Deal with modified resources.
315	 * Depending on what has changed exactly we might want to perform
316	 * different actions.
317	 *
318	 * We do full resource restart in the following situations:
319	 * Resource role is INIT or SECONDARY.
320	 * Resource role is PRIMARY and path to local component or provider
321	 * name has changed.
322	 * In case of PRIMARY, the worker process will be killed and restarted,
323	 * which also means removing /dev/hast/<name> provider and
324	 * recreating it.
325	 *
326	 * We do just reload (send SIGHUP to worker process) if we act as
327	 * PRIMARY, but only remote address, replication mode and timeout
328	 * has changed. For those, there is no need to restart worker process.
329	 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
330	 * replication mode has changed or simply set new timeout if only
331	 * timeout has changed.
332	 */
333	TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
334		TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
335			if (strcmp(cres->hr_name, nres->hr_name) == 0)
336				break;
337		}
338		assert(cres != NULL);
339		if (resource_needs_restart(cres, nres)) {
340			pjdlog_info("Resource %s configuration was modified, restarting it.",
341			    cres->hr_name);
342			role = cres->hr_role;
343			control_set_role(cres, HAST_ROLE_INIT);
344			TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
345			free(cres);
346			TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
347			TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
348			control_set_role(nres, role);
349		} else if (resource_needs_reload(cres, nres)) {
350			pjdlog_info("Resource %s configuration was modified, reloading it.",
351			    cres->hr_name);
352			strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
353			    sizeof(cres->hr_remoteaddr));
354			cres->hr_replication = nres->hr_replication;
355			cres->hr_timeout = nres->hr_timeout;
356			if (cres->hr_workerpid != 0) {
357				if (kill(cres->hr_workerpid, SIGHUP) < 0) {
358					pjdlog_errno(LOG_WARNING,
359					    "Unable to send SIGHUP to worker process %u",
360					    (unsigned int)cres->hr_workerpid);
361				}
362			}
363		}
364	}
365
366	yy_config_free(newcfg);
367	pjdlog_info("Configuration reloaded successfully.");
368	return;
369failed:
370	if (newcfg != NULL) {
371		if (newcfg->hc_controlconn != NULL)
372			proto_close(newcfg->hc_controlconn);
373		if (newcfg->hc_listenconn != NULL)
374			proto_close(newcfg->hc_listenconn);
375		yy_config_free(newcfg);
376	}
377	pjdlog_warning("Configuration not reloaded.");
378}
379
380static void
381terminate_workers(void)
382{
383	struct hast_resource *res;
384
385	pjdlog_info("Termination signal received, exiting.");
386	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
387		if (res->hr_workerpid == 0)
388			continue;
389		pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
390		    res->hr_name, role2str(res->hr_role), res->hr_workerpid);
391		if (kill(res->hr_workerpid, SIGTERM) == 0)
392			continue;
393		pjdlog_errno(LOG_WARNING,
394		    "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
395		    res->hr_name, role2str(res->hr_role), res->hr_workerpid);
396	}
397}
398
399static void
400listen_accept(void)
401{
402	struct hast_resource *res;
403	struct proto_conn *conn;
404	struct nv *nvin, *nvout, *nverr;
405	const char *resname;
406	const unsigned char *token;
407	char laddr[256], raddr[256];
408	size_t size;
409	pid_t pid;
410	int status;
411
412	proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr));
413	pjdlog_debug(1, "Accepting connection to %s.", laddr);
414
415	if (proto_accept(cfg->hc_listenconn, &conn) < 0) {
416		pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
417		return;
418	}
419
420	proto_local_address(conn, laddr, sizeof(laddr));
421	proto_remote_address(conn, raddr, sizeof(raddr));
422	pjdlog_info("Connection from %s to %s.", raddr, laddr);
423
424	/* Error in setting timeout is not critical, but why should it fail? */
425	if (proto_timeout(conn, HAST_TIMEOUT) < 0)
426		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
427
428	nvin = nvout = nverr = NULL;
429
430	/*
431	 * Before receiving any data see if remote host have access to any
432	 * resource.
433	 */
434	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
435		if (proto_address_match(conn, res->hr_remoteaddr))
436			break;
437	}
438	if (res == NULL) {
439		pjdlog_error("Client %s isn't known.", raddr);
440		goto close;
441	}
442	/* Ok, remote host can access at least one resource. */
443
444	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
445		pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
446		    raddr);
447		goto close;
448	}
449
450	resname = nv_get_string(nvin, "resource");
451	if (resname == NULL) {
452		pjdlog_error("No 'resource' field in the header received from %s.",
453		    raddr);
454		goto close;
455	}
456	pjdlog_debug(2, "%s: resource=%s", raddr, resname);
457	token = nv_get_uint8_array(nvin, &size, "token");
458	/*
459	 * NULL token means that this is first conection.
460	 */
461	if (token != NULL && size != sizeof(res->hr_token)) {
462		pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
463		    raddr, sizeof(res->hr_token), size);
464		goto close;
465	}
466
467	/*
468	 * From now on we want to send errors to the remote node.
469	 */
470	nverr = nv_alloc();
471
472	/* Find resource related to this connection. */
473	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
474		if (strcmp(resname, res->hr_name) == 0)
475			break;
476	}
477	/* Have we found the resource? */
478	if (res == NULL) {
479		pjdlog_error("No resource '%s' as requested by %s.",
480		    resname, raddr);
481		nv_add_stringf(nverr, "errmsg", "Resource not configured.");
482		goto fail;
483	}
484
485	/* Now that we know resource name setup log prefix. */
486	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
487
488	/* Does the remote host have access to this resource? */
489	if (!proto_address_match(conn, res->hr_remoteaddr)) {
490		pjdlog_error("Client %s has no access to the resource.", raddr);
491		nv_add_stringf(nverr, "errmsg", "No access to the resource.");
492		goto fail;
493	}
494	/* Is the resource marked as secondary? */
495	if (res->hr_role != HAST_ROLE_SECONDARY) {
496		pjdlog_error("We act as %s for the resource and not as %s as requested by %s.",
497		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
498		    raddr);
499		nv_add_stringf(nverr, "errmsg",
500		    "Remote node acts as %s for the resource and not as %s.",
501		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
502		goto fail;
503	}
504	/* Does token (if exists) match? */
505	if (token != NULL && memcmp(token, res->hr_token,
506	    sizeof(res->hr_token)) != 0) {
507		pjdlog_error("Token received from %s doesn't match.", raddr);
508		nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
509		goto fail;
510	}
511	/*
512	 * If there is no token, but we have half-open connection
513	 * (only remotein) or full connection (worker process is running)
514	 * we have to cancel those and accept the new connection.
515	 */
516	if (token == NULL) {
517		assert(res->hr_remoteout == NULL);
518		pjdlog_debug(1, "Initial connection from %s.", raddr);
519		if (res->hr_workerpid != 0) {
520			assert(res->hr_remotein == NULL);
521			pjdlog_debug(1,
522			    "Worker process exists (pid=%u), stopping it.",
523			    (unsigned int)res->hr_workerpid);
524			/* Stop child process. */
525			if (kill(res->hr_workerpid, SIGINT) < 0) {
526				pjdlog_errno(LOG_ERR,
527				    "Unable to stop worker process (pid=%u)",
528				    (unsigned int)res->hr_workerpid);
529				/*
530				 * Other than logging the problem we
531				 * ignore it - nothing smart to do.
532				 */
533			}
534			/* Wait for it to exit. */
535			else if ((pid = waitpid(res->hr_workerpid,
536			    &status, 0)) != res->hr_workerpid) {
537				/* We can only log the problem. */
538				pjdlog_errno(LOG_ERR,
539				    "Waiting for worker process (pid=%u) failed",
540				    (unsigned int)res->hr_workerpid);
541			} else {
542				child_exit_log(res->hr_workerpid, status);
543			}
544			res->hr_workerpid = 0;
545		} else if (res->hr_remotein != NULL) {
546			char oaddr[256];
547
548			proto_remote_address(conn, oaddr, sizeof(oaddr));
549			pjdlog_debug(1,
550			    "Canceling half-open connection from %s on connection from %s.",
551			    oaddr, raddr);
552			proto_close(res->hr_remotein);
553			res->hr_remotein = NULL;
554		}
555	}
556
557	/*
558	 * Checks and cleanups are done.
559	 */
560
561	if (token == NULL) {
562		arc4random_buf(res->hr_token, sizeof(res->hr_token));
563		nvout = nv_alloc();
564		nv_add_uint8_array(nvout, res->hr_token,
565		    sizeof(res->hr_token), "token");
566		if (nv_error(nvout) != 0) {
567			pjdlog_common(LOG_ERR, 0, nv_error(nvout),
568			    "Unable to prepare return header for %s", raddr);
569			nv_add_stringf(nverr, "errmsg",
570			    "Remote node was unable to prepare return header: %s.",
571			    strerror(nv_error(nvout)));
572			goto fail;
573		}
574		if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
575			int error = errno;
576
577			pjdlog_errno(LOG_ERR, "Unable to send response to %s",
578			    raddr);
579			nv_add_stringf(nverr, "errmsg",
580			    "Remote node was unable to send response: %s.",
581			    strerror(error));
582			goto fail;
583		}
584		res->hr_remotein = conn;
585		pjdlog_debug(1, "Incoming connection from %s configured.",
586		    raddr);
587	} else {
588		res->hr_remoteout = conn;
589		pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
590		hastd_secondary(res, nvin);
591	}
592	nv_free(nvin);
593	nv_free(nvout);
594	nv_free(nverr);
595	pjdlog_prefix_set("%s", "");
596	return;
597fail:
598	if (nv_error(nverr) != 0) {
599		pjdlog_common(LOG_ERR, 0, nv_error(nverr),
600		    "Unable to prepare error header for %s", raddr);
601		goto close;
602	}
603	if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
604		pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
605		goto close;
606	}
607close:
608	if (nvin != NULL)
609		nv_free(nvin);
610	if (nvout != NULL)
611		nv_free(nvout);
612	if (nverr != NULL)
613		nv_free(nverr);
614	proto_close(conn);
615	pjdlog_prefix_set("%s", "");
616}
617
618static void
619main_loop(void)
620{
621	fd_set rfds, wfds;
622	int cfd, lfd, maxfd, ret;
623
624	for (;;) {
625		if (sigexit_received) {
626			sigexit_received = false;
627			terminate_workers();
628			exit(EX_OK);
629		}
630		if (sigchld_received) {
631			sigchld_received = false;
632			child_exit();
633		}
634		if (sighup_received) {
635			sighup_received = false;
636			hastd_reload();
637		}
638
639		cfd = proto_descriptor(cfg->hc_controlconn);
640		lfd = proto_descriptor(cfg->hc_listenconn);
641		maxfd = cfd > lfd ? cfd : lfd;
642
643		/* Setup descriptors for select(2). */
644		FD_ZERO(&rfds);
645		FD_SET(cfd, &rfds);
646		FD_SET(lfd, &rfds);
647		FD_ZERO(&wfds);
648		FD_SET(cfd, &wfds);
649		FD_SET(lfd, &wfds);
650
651		ret = select(maxfd + 1, &rfds, &wfds, NULL, NULL);
652		if (ret == -1) {
653			if (errno == EINTR)
654				continue;
655			KEEP_ERRNO((void)pidfile_remove(pfh));
656			pjdlog_exit(EX_OSERR, "select() failed");
657		}
658
659		if (FD_ISSET(cfd, &rfds) || FD_ISSET(cfd, &wfds))
660			control_handle(cfg);
661		if (FD_ISSET(lfd, &rfds) || FD_ISSET(lfd, &wfds))
662			listen_accept();
663	}
664}
665
666int
667main(int argc, char *argv[])
668{
669	const char *pidfile;
670	pid_t otherpid;
671	bool foreground;
672	int debuglevel;
673
674	g_gate_load();
675
676	foreground = false;
677	debuglevel = 0;
678	pidfile = HASTD_PIDFILE;
679
680	for (;;) {
681		int ch;
682
683		ch = getopt(argc, argv, "c:dFhP:");
684		if (ch == -1)
685			break;
686		switch (ch) {
687		case 'c':
688			cfgpath = optarg;
689			break;
690		case 'd':
691			debuglevel++;
692			break;
693		case 'F':
694			foreground = true;
695			break;
696		case 'P':
697			pidfile = optarg;
698			break;
699		case 'h':
700		default:
701			usage();
702		}
703	}
704	argc -= optind;
705	argv += optind;
706
707	pjdlog_debug_set(debuglevel);
708
709	pfh = pidfile_open(pidfile, 0600, &otherpid);
710	if (pfh == NULL) {
711		if (errno == EEXIST) {
712			pjdlog_exitx(EX_TEMPFAIL,
713			    "Another hastd is already running, pid: %jd.",
714			    (intmax_t)otherpid);
715		}
716		/* If we cannot create pidfile from other reasons, only warn. */
717		pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
718	}
719
720	cfg = yy_config_parse(cfgpath, true);
721	assert(cfg != NULL);
722
723	signal(SIGINT, sighandler);
724	signal(SIGTERM, sighandler);
725	signal(SIGHUP, sighandler);
726	signal(SIGCHLD, sighandler);
727
728	/* Listen on control address. */
729	if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
730		KEEP_ERRNO((void)pidfile_remove(pfh));
731		pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
732		    cfg->hc_controladdr);
733	}
734	/* Listen for remote connections. */
735	if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) {
736		KEEP_ERRNO((void)pidfile_remove(pfh));
737		pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
738		    cfg->hc_listenaddr);
739	}
740
741	if (!foreground) {
742		if (daemon(0, 0) < 0) {
743			KEEP_ERRNO((void)pidfile_remove(pfh));
744			pjdlog_exit(EX_OSERR, "Unable to daemonize");
745		}
746
747		/* Start logging to syslog. */
748		pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
749
750		/* Write PID to a file. */
751		if (pidfile_write(pfh) < 0) {
752			pjdlog_errno(LOG_WARNING,
753			    "Unable to write PID to a file");
754		}
755	}
756
757	main_loop();
758
759	exit(0);
760}
761