xref: /freebsd/sbin/hastd/hastd.c (revision ad30f8e79bd1007cc2476e491bd21b4f5e389e0a)
1 /*-
2  * Copyright (c) 2009-2010 The FreeBSD Foundation
3  * Copyright (c) 2010-2011 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("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/linker.h>
36 #include <sys/module.h>
37 #include <sys/stat.h>
38 #include <sys/wait.h>
39 
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 "event.h"
56 #include "hast.h"
57 #include "hast_proto.h"
58 #include "hastd.h"
59 #include "hooks.h"
60 #include "subr.h"
61 
62 /* Path to configuration file. */
63 const char *cfgpath = HAST_CONFIG;
64 /* Hastd configuration. */
65 static struct hastd_config *cfg;
66 /* Was SIGINT or SIGTERM signal received? */
67 bool sigexit_received = false;
68 /* PID file handle. */
69 struct pidfh *pfh;
70 
71 /* How often check for hooks running for too long. */
72 #define	REPORT_INTERVAL	5
73 
74 static void
75 usage(void)
76 {
77 
78 	errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
79 }
80 
81 static void
82 g_gate_load(void)
83 {
84 
85 	if (modfind("g_gate") == -1) {
86 		/* Not present in kernel, try loading it. */
87 		if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) {
88 			if (errno != EEXIST) {
89 				pjdlog_exit(EX_OSERR,
90 				    "Unable to load geom_gate module");
91 			}
92 		}
93 	}
94 }
95 
96 void
97 descriptors_cleanup(struct hast_resource *res)
98 {
99 	struct hast_resource *tres;
100 
101 	TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) {
102 		if (tres == res) {
103 			PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY ||
104 			    (res->hr_remotein == NULL &&
105 			     res->hr_remoteout == NULL));
106 			continue;
107 		}
108 		if (tres->hr_remotein != NULL)
109 			proto_close(tres->hr_remotein);
110 		if (tres->hr_remoteout != NULL)
111 			proto_close(tres->hr_remoteout);
112 		if (tres->hr_ctrl != NULL)
113 			proto_close(tres->hr_ctrl);
114 		if (tres->hr_event != NULL)
115 			proto_close(tres->hr_event);
116 		if (tres->hr_conn != NULL)
117 			proto_close(tres->hr_conn);
118 	}
119 	if (cfg->hc_controlin != NULL)
120 		proto_close(cfg->hc_controlin);
121 	proto_close(cfg->hc_controlconn);
122 	proto_close(cfg->hc_listenconn);
123 	(void)pidfile_close(pfh);
124 	hook_fini();
125 	pjdlog_fini();
126 }
127 
128 static const char *
129 dtype2str(mode_t mode)
130 {
131 
132 	if (S_ISBLK(mode))
133 		return ("block device");
134 	else if (S_ISCHR(mode))
135 		return ("character device");
136 	else if (S_ISDIR(mode))
137 		return ("directory");
138 	else if (S_ISFIFO(mode))
139 		return ("pipe or FIFO");
140 	else if (S_ISLNK(mode))
141 		return ("symbolic link");
142 	else if (S_ISREG(mode))
143 		return ("regular file");
144 	else if (S_ISSOCK(mode))
145 		return ("socket");
146 	else if (S_ISWHT(mode))
147 		return ("whiteout");
148 	else
149 		return ("unknown");
150 }
151 
152 void
153 descriptors_assert(const struct hast_resource *res, int pjdlogmode)
154 {
155 	char msg[256];
156 	struct stat sb;
157 	long maxfd;
158 	bool isopen;
159 	mode_t mode;
160 	int fd;
161 
162 	/*
163 	 * At this point descriptor to syslog socket is closed, so if we want
164 	 * to log assertion message, we have to first store it in 'msg' local
165 	 * buffer and then open syslog socket and log it.
166 	 */
167 	msg[0] = '\0';
168 
169 	maxfd = sysconf(_SC_OPEN_MAX);
170 	if (maxfd < 0) {
171 		pjdlog_init(pjdlogmode);
172 		pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
173 		    role2str(res->hr_role));
174 		pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed");
175 		pjdlog_fini();
176 		maxfd = 16384;
177 	}
178 	for (fd = 0; fd <= maxfd; fd++) {
179 		if (fstat(fd, &sb) == 0) {
180 			isopen = true;
181 			mode = sb.st_mode;
182 		} else if (errno == EBADF) {
183 			isopen = false;
184 			mode = 0;
185 		} else {
186 			(void)snprintf(msg, sizeof(msg),
187 			    "Unable to fstat descriptor %d: %s", fd,
188 			    strerror(errno));
189 			break;
190 		}
191 		if (fd == STDIN_FILENO || fd == STDOUT_FILENO ||
192 		    fd == STDERR_FILENO) {
193 			if (!isopen) {
194 				(void)snprintf(msg, sizeof(msg),
195 				    "Descriptor %d (%s) is closed, but should be open.",
196 				    fd, (fd == STDIN_FILENO ? "stdin" :
197 				    (fd == STDOUT_FILENO ? "stdout" : "stderr")));
198 				break;
199 			}
200 		} else if (fd == proto_descriptor(res->hr_event)) {
201 			if (!isopen) {
202 				(void)snprintf(msg, sizeof(msg),
203 				    "Descriptor %d (event) is closed, but should be open.",
204 				    fd);
205 				break;
206 			}
207 			if (!S_ISSOCK(mode)) {
208 				(void)snprintf(msg, sizeof(msg),
209 				    "Descriptor %d (event) is %s, but should be %s.",
210 				    fd, dtype2str(mode), dtype2str(S_IFSOCK));
211 				break;
212 			}
213 		} else if (fd == proto_descriptor(res->hr_ctrl)) {
214 			if (!isopen) {
215 				(void)snprintf(msg, sizeof(msg),
216 				    "Descriptor %d (ctrl) is closed, but should be open.",
217 				    fd);
218 				break;
219 			}
220 			if (!S_ISSOCK(mode)) {
221 				(void)snprintf(msg, sizeof(msg),
222 				    "Descriptor %d (ctrl) is %s, but should be %s.",
223 				    fd, dtype2str(mode), dtype2str(S_IFSOCK));
224 				break;
225 			}
226 		} else if (fd == proto_descriptor(res->hr_conn)) {
227 			if (!isopen) {
228 				(void)snprintf(msg, sizeof(msg),
229 				    "Descriptor %d (conn) is closed, but should be open.",
230 				    fd);
231 				break;
232 			}
233 			if (!S_ISSOCK(mode)) {
234 				(void)snprintf(msg, sizeof(msg),
235 				    "Descriptor %d (conn) is %s, but should be %s.",
236 				    fd, dtype2str(mode), dtype2str(S_IFSOCK));
237 				break;
238 			}
239 		} else if (res->hr_role == HAST_ROLE_SECONDARY &&
240 		    fd == proto_descriptor(res->hr_remotein)) {
241 			if (!isopen) {
242 				(void)snprintf(msg, sizeof(msg),
243 				    "Descriptor %d (remote in) is closed, but should be open.",
244 				    fd);
245 				break;
246 			}
247 			if (!S_ISSOCK(mode)) {
248 				(void)snprintf(msg, sizeof(msg),
249 				    "Descriptor %d (remote in) is %s, but should be %s.",
250 				    fd, dtype2str(mode), dtype2str(S_IFSOCK));
251 				break;
252 			}
253 		} else if (res->hr_role == HAST_ROLE_SECONDARY &&
254 		    fd == proto_descriptor(res->hr_remoteout)) {
255 			if (!isopen) {
256 				(void)snprintf(msg, sizeof(msg),
257 				    "Descriptor %d (remote out) is closed, but should be open.",
258 				    fd);
259 				break;
260 			}
261 			if (!S_ISSOCK(mode)) {
262 				(void)snprintf(msg, sizeof(msg),
263 				    "Descriptor %d (remote out) is %s, but should be %s.",
264 				    fd, dtype2str(mode), dtype2str(S_IFSOCK));
265 				break;
266 			}
267 		} else {
268 			if (isopen) {
269 				(void)snprintf(msg, sizeof(msg),
270 				    "Descriptor %d is open (%s), but should be closed.",
271 				    fd, dtype2str(mode));
272 				break;
273 			}
274 		}
275 	}
276 	if (msg[0] != '\0') {
277 		pjdlog_init(pjdlogmode);
278 		pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
279 		    role2str(res->hr_role));
280 		PJDLOG_ABORT("%s", msg);
281 	}
282 }
283 
284 static void
285 child_exit_log(unsigned int pid, int status)
286 {
287 
288 	if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
289 		pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
290 		    pid);
291 	} else if (WIFSIGNALED(status)) {
292 		pjdlog_error("Worker process killed (pid=%u, signal=%d).",
293 		    pid, WTERMSIG(status));
294 	} else {
295 		pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
296 		    pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
297 	}
298 }
299 
300 static void
301 child_exit(void)
302 {
303 	struct hast_resource *res;
304 	int status;
305 	pid_t pid;
306 
307 	while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
308 		/* Find resource related to the process that just exited. */
309 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
310 			if (pid == res->hr_workerpid)
311 				break;
312 		}
313 		if (res == NULL) {
314 			/*
315 			 * This can happen when new connection arrives and we
316 			 * cancel child responsible for the old one or if this
317 			 * was hook which we executed.
318 			 */
319 			hook_check_one(pid, status);
320 			continue;
321 		}
322 		pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
323 		    role2str(res->hr_role));
324 		child_exit_log(pid, status);
325 		child_cleanup(res);
326 		if (res->hr_role == HAST_ROLE_PRIMARY) {
327 			/*
328 			 * Restart child process if it was killed by signal
329 			 * or exited because of temporary problem.
330 			 */
331 			if (WIFSIGNALED(status) ||
332 			    (WIFEXITED(status) &&
333 			     WEXITSTATUS(status) == EX_TEMPFAIL)) {
334 				sleep(1);
335 				pjdlog_info("Restarting worker process.");
336 				hastd_primary(res);
337 			} else {
338 				res->hr_role = HAST_ROLE_INIT;
339 				pjdlog_info("Changing resource role back to %s.",
340 				    role2str(res->hr_role));
341 			}
342 		}
343 		pjdlog_prefix_set("%s", "");
344 	}
345 }
346 
347 static bool
348 resource_needs_restart(const struct hast_resource *res0,
349     const struct hast_resource *res1)
350 {
351 
352 	PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
353 
354 	if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
355 		return (true);
356 	if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
357 		return (true);
358 	if (res0->hr_role == HAST_ROLE_INIT ||
359 	    res0->hr_role == HAST_ROLE_SECONDARY) {
360 		if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
361 			return (true);
362 		if (res0->hr_replication != res1->hr_replication)
363 			return (true);
364 		if (res0->hr_timeout != res1->hr_timeout)
365 			return (true);
366 		if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
367 			return (true);
368 	}
369 	return (false);
370 }
371 
372 static bool
373 resource_needs_reload(const struct hast_resource *res0,
374     const struct hast_resource *res1)
375 {
376 
377 	PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
378 	PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0);
379 	PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
380 
381 	if (res0->hr_role != HAST_ROLE_PRIMARY)
382 		return (false);
383 
384 	if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
385 		return (true);
386 	if (res0->hr_replication != res1->hr_replication)
387 		return (true);
388 	if (res0->hr_timeout != res1->hr_timeout)
389 		return (true);
390 	if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
391 		return (true);
392 	return (false);
393 }
394 
395 static void
396 resource_reload(const struct hast_resource *res)
397 {
398 	struct nv *nvin, *nvout;
399 	int error;
400 
401 	PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
402 
403 	nvout = nv_alloc();
404 	nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd");
405 	nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr");
406 	nv_add_int32(nvout, (int32_t)res->hr_replication, "replication");
407 	nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout");
408 	nv_add_string(nvout, res->hr_exec, "exec");
409 	if (nv_error(nvout) != 0) {
410 		nv_free(nvout);
411 		pjdlog_error("Unable to allocate header for reload message.");
412 		return;
413 	}
414 	if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
415 		pjdlog_errno(LOG_ERR, "Unable to send reload message");
416 		nv_free(nvout);
417 		return;
418 	}
419 	nv_free(nvout);
420 
421 	/* Receive response. */
422 	if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
423 		pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
424 		return;
425 	}
426 	error = nv_get_int16(nvin, "error");
427 	nv_free(nvin);
428 	if (error != 0) {
429 		pjdlog_common(LOG_ERR, 0, error, "Reload failed");
430 		return;
431 	}
432 }
433 
434 static void
435 hastd_reload(void)
436 {
437 	struct hastd_config *newcfg;
438 	struct hast_resource *nres, *cres, *tres;
439 	uint8_t role;
440 
441 	pjdlog_info("Reloading configuration...");
442 
443 	newcfg = yy_config_parse(cfgpath, false);
444 	if (newcfg == NULL)
445 		goto failed;
446 
447 	/*
448 	 * Check if control address has changed.
449 	 */
450 	if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
451 		if (proto_server(newcfg->hc_controladdr,
452 		    &newcfg->hc_controlconn) < 0) {
453 			pjdlog_errno(LOG_ERR,
454 			    "Unable to listen on control address %s",
455 			    newcfg->hc_controladdr);
456 			goto failed;
457 		}
458 	}
459 	/*
460 	 * Check if listen address has changed.
461 	 */
462 	if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) {
463 		if (proto_server(newcfg->hc_listenaddr,
464 		    &newcfg->hc_listenconn) < 0) {
465 			pjdlog_errno(LOG_ERR, "Unable to listen on address %s",
466 			    newcfg->hc_listenaddr);
467 			goto failed;
468 		}
469 	}
470 	/*
471 	 * Only when both control and listen sockets are successfully
472 	 * initialized switch them to new configuration.
473 	 */
474 	if (newcfg->hc_controlconn != NULL) {
475 		pjdlog_info("Control socket changed from %s to %s.",
476 		    cfg->hc_controladdr, newcfg->hc_controladdr);
477 		proto_close(cfg->hc_controlconn);
478 		cfg->hc_controlconn = newcfg->hc_controlconn;
479 		newcfg->hc_controlconn = NULL;
480 		strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
481 		    sizeof(cfg->hc_controladdr));
482 	}
483 	if (newcfg->hc_listenconn != NULL) {
484 		pjdlog_info("Listen socket changed from %s to %s.",
485 		    cfg->hc_listenaddr, newcfg->hc_listenaddr);
486 		proto_close(cfg->hc_listenconn);
487 		cfg->hc_listenconn = newcfg->hc_listenconn;
488 		newcfg->hc_listenconn = NULL;
489 		strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr,
490 		    sizeof(cfg->hc_listenaddr));
491 	}
492 
493 	/*
494 	 * Stop and remove resources that were removed from the configuration.
495 	 */
496 	TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
497 		TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
498 			if (strcmp(cres->hr_name, nres->hr_name) == 0)
499 				break;
500 		}
501 		if (nres == NULL) {
502 			control_set_role(cres, HAST_ROLE_INIT);
503 			TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
504 			pjdlog_info("Resource %s removed.", cres->hr_name);
505 			free(cres);
506 		}
507 	}
508 	/*
509 	 * Move new resources to the current configuration.
510 	 */
511 	TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
512 		TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
513 			if (strcmp(cres->hr_name, nres->hr_name) == 0)
514 				break;
515 		}
516 		if (cres == NULL) {
517 			TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
518 			TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
519 			pjdlog_info("Resource %s added.", nres->hr_name);
520 		}
521 	}
522 	/*
523 	 * Deal with modified resources.
524 	 * Depending on what has changed exactly we might want to perform
525 	 * different actions.
526 	 *
527 	 * We do full resource restart in the following situations:
528 	 * Resource role is INIT or SECONDARY.
529 	 * Resource role is PRIMARY and path to local component or provider
530 	 * name has changed.
531 	 * In case of PRIMARY, the worker process will be killed and restarted,
532 	 * which also means removing /dev/hast/<name> provider and
533 	 * recreating it.
534 	 *
535 	 * We do just reload (send SIGHUP to worker process) if we act as
536 	 * PRIMARY, but only if remote address, replication mode, timeout or
537 	 * execution path has changed. For those, there is no need to restart
538 	 * worker process.
539 	 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
540 	 * replication mode has changed or simply set new timeout if only
541 	 * timeout has changed.
542 	 */
543 	TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
544 		TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
545 			if (strcmp(cres->hr_name, nres->hr_name) == 0)
546 				break;
547 		}
548 		PJDLOG_ASSERT(cres != NULL);
549 		if (resource_needs_restart(cres, nres)) {
550 			pjdlog_info("Resource %s configuration was modified, restarting it.",
551 			    cres->hr_name);
552 			role = cres->hr_role;
553 			control_set_role(cres, HAST_ROLE_INIT);
554 			TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
555 			free(cres);
556 			TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
557 			TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
558 			control_set_role(nres, role);
559 		} else if (resource_needs_reload(cres, nres)) {
560 			pjdlog_info("Resource %s configuration was modified, reloading it.",
561 			    cres->hr_name);
562 			strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
563 			    sizeof(cres->hr_remoteaddr));
564 			cres->hr_replication = nres->hr_replication;
565 			cres->hr_timeout = nres->hr_timeout;
566 			strlcpy(cres->hr_exec, nres->hr_exec,
567 			    sizeof(cres->hr_exec));
568 			if (cres->hr_workerpid != 0)
569 				resource_reload(cres);
570 		}
571 	}
572 
573 	yy_config_free(newcfg);
574 	pjdlog_info("Configuration reloaded successfully.");
575 	return;
576 failed:
577 	if (newcfg != NULL) {
578 		if (newcfg->hc_controlconn != NULL)
579 			proto_close(newcfg->hc_controlconn);
580 		if (newcfg->hc_listenconn != NULL)
581 			proto_close(newcfg->hc_listenconn);
582 		yy_config_free(newcfg);
583 	}
584 	pjdlog_warning("Configuration not reloaded.");
585 }
586 
587 static void
588 terminate_workers(void)
589 {
590 	struct hast_resource *res;
591 
592 	pjdlog_info("Termination signal received, exiting.");
593 	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
594 		if (res->hr_workerpid == 0)
595 			continue;
596 		pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
597 		    res->hr_name, role2str(res->hr_role), res->hr_workerpid);
598 		if (kill(res->hr_workerpid, SIGTERM) == 0)
599 			continue;
600 		pjdlog_errno(LOG_WARNING,
601 		    "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
602 		    res->hr_name, role2str(res->hr_role), res->hr_workerpid);
603 	}
604 }
605 
606 static void
607 listen_accept(void)
608 {
609 	struct hast_resource *res;
610 	struct proto_conn *conn;
611 	struct nv *nvin, *nvout, *nverr;
612 	const char *resname;
613 	const unsigned char *token;
614 	char laddr[256], raddr[256];
615 	size_t size;
616 	pid_t pid;
617 	int status;
618 
619 	proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr));
620 	pjdlog_debug(1, "Accepting connection to %s.", laddr);
621 
622 	if (proto_accept(cfg->hc_listenconn, &conn) < 0) {
623 		pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
624 		return;
625 	}
626 
627 	proto_local_address(conn, laddr, sizeof(laddr));
628 	proto_remote_address(conn, raddr, sizeof(raddr));
629 	pjdlog_info("Connection from %s to %s.", raddr, laddr);
630 
631 	/* Error in setting timeout is not critical, but why should it fail? */
632 	if (proto_timeout(conn, HAST_TIMEOUT) < 0)
633 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
634 
635 	nvin = nvout = nverr = NULL;
636 
637 	/*
638 	 * Before receiving any data see if remote host have access to any
639 	 * resource.
640 	 */
641 	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
642 		if (proto_address_match(conn, res->hr_remoteaddr))
643 			break;
644 	}
645 	if (res == NULL) {
646 		pjdlog_error("Client %s isn't known.", raddr);
647 		goto close;
648 	}
649 	/* Ok, remote host can access at least one resource. */
650 
651 	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
652 		pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
653 		    raddr);
654 		goto close;
655 	}
656 
657 	resname = nv_get_string(nvin, "resource");
658 	if (resname == NULL) {
659 		pjdlog_error("No 'resource' field in the header received from %s.",
660 		    raddr);
661 		goto close;
662 	}
663 	pjdlog_debug(2, "%s: resource=%s", raddr, resname);
664 	token = nv_get_uint8_array(nvin, &size, "token");
665 	/*
666 	 * NULL token means that this is first conection.
667 	 */
668 	if (token != NULL && size != sizeof(res->hr_token)) {
669 		pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
670 		    raddr, sizeof(res->hr_token), size);
671 		goto close;
672 	}
673 
674 	/*
675 	 * From now on we want to send errors to the remote node.
676 	 */
677 	nverr = nv_alloc();
678 
679 	/* Find resource related to this connection. */
680 	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
681 		if (strcmp(resname, res->hr_name) == 0)
682 			break;
683 	}
684 	/* Have we found the resource? */
685 	if (res == NULL) {
686 		pjdlog_error("No resource '%s' as requested by %s.",
687 		    resname, raddr);
688 		nv_add_stringf(nverr, "errmsg", "Resource not configured.");
689 		goto fail;
690 	}
691 
692 	/* Now that we know resource name setup log prefix. */
693 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
694 
695 	/* Does the remote host have access to this resource? */
696 	if (!proto_address_match(conn, res->hr_remoteaddr)) {
697 		pjdlog_error("Client %s has no access to the resource.", raddr);
698 		nv_add_stringf(nverr, "errmsg", "No access to the resource.");
699 		goto fail;
700 	}
701 	/* Is the resource marked as secondary? */
702 	if (res->hr_role != HAST_ROLE_SECONDARY) {
703 		pjdlog_error("We act as %s for the resource and not as %s as requested by %s.",
704 		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
705 		    raddr);
706 		nv_add_stringf(nverr, "errmsg",
707 		    "Remote node acts as %s for the resource and not as %s.",
708 		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
709 		goto fail;
710 	}
711 	/* Does token (if exists) match? */
712 	if (token != NULL && memcmp(token, res->hr_token,
713 	    sizeof(res->hr_token)) != 0) {
714 		pjdlog_error("Token received from %s doesn't match.", raddr);
715 		nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
716 		goto fail;
717 	}
718 	/*
719 	 * If there is no token, but we have half-open connection
720 	 * (only remotein) or full connection (worker process is running)
721 	 * we have to cancel those and accept the new connection.
722 	 */
723 	if (token == NULL) {
724 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
725 		pjdlog_debug(1, "Initial connection from %s.", raddr);
726 		if (res->hr_workerpid != 0) {
727 			PJDLOG_ASSERT(res->hr_remotein == NULL);
728 			pjdlog_debug(1,
729 			    "Worker process exists (pid=%u), stopping it.",
730 			    (unsigned int)res->hr_workerpid);
731 			/* Stop child process. */
732 			if (kill(res->hr_workerpid, SIGINT) < 0) {
733 				pjdlog_errno(LOG_ERR,
734 				    "Unable to stop worker process (pid=%u)",
735 				    (unsigned int)res->hr_workerpid);
736 				/*
737 				 * Other than logging the problem we
738 				 * ignore it - nothing smart to do.
739 				 */
740 			}
741 			/* Wait for it to exit. */
742 			else if ((pid = waitpid(res->hr_workerpid,
743 			    &status, 0)) != res->hr_workerpid) {
744 				/* We can only log the problem. */
745 				pjdlog_errno(LOG_ERR,
746 				    "Waiting for worker process (pid=%u) failed",
747 				    (unsigned int)res->hr_workerpid);
748 			} else {
749 				child_exit_log(res->hr_workerpid, status);
750 			}
751 			child_cleanup(res);
752 		} else if (res->hr_remotein != NULL) {
753 			char oaddr[256];
754 
755 			proto_remote_address(res->hr_remotein, oaddr,
756 			    sizeof(oaddr));
757 			pjdlog_debug(1,
758 			    "Canceling half-open connection from %s on connection from %s.",
759 			    oaddr, raddr);
760 			proto_close(res->hr_remotein);
761 			res->hr_remotein = NULL;
762 		}
763 	}
764 
765 	/*
766 	 * Checks and cleanups are done.
767 	 */
768 
769 	if (token == NULL) {
770 		arc4random_buf(res->hr_token, sizeof(res->hr_token));
771 		nvout = nv_alloc();
772 		nv_add_uint8_array(nvout, res->hr_token,
773 		    sizeof(res->hr_token), "token");
774 		if (nv_error(nvout) != 0) {
775 			pjdlog_common(LOG_ERR, 0, nv_error(nvout),
776 			    "Unable to prepare return header for %s", raddr);
777 			nv_add_stringf(nverr, "errmsg",
778 			    "Remote node was unable to prepare return header: %s.",
779 			    strerror(nv_error(nvout)));
780 			goto fail;
781 		}
782 		if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
783 			int error = errno;
784 
785 			pjdlog_errno(LOG_ERR, "Unable to send response to %s",
786 			    raddr);
787 			nv_add_stringf(nverr, "errmsg",
788 			    "Remote node was unable to send response: %s.",
789 			    strerror(error));
790 			goto fail;
791 		}
792 		res->hr_remotein = conn;
793 		pjdlog_debug(1, "Incoming connection from %s configured.",
794 		    raddr);
795 	} else {
796 		res->hr_remoteout = conn;
797 		pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
798 		hastd_secondary(res, nvin);
799 	}
800 	nv_free(nvin);
801 	nv_free(nvout);
802 	nv_free(nverr);
803 	pjdlog_prefix_set("%s", "");
804 	return;
805 fail:
806 	if (nv_error(nverr) != 0) {
807 		pjdlog_common(LOG_ERR, 0, nv_error(nverr),
808 		    "Unable to prepare error header for %s", raddr);
809 		goto close;
810 	}
811 	if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
812 		pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
813 		goto close;
814 	}
815 close:
816 	if (nvin != NULL)
817 		nv_free(nvin);
818 	if (nvout != NULL)
819 		nv_free(nvout);
820 	if (nverr != NULL)
821 		nv_free(nverr);
822 	proto_close(conn);
823 	pjdlog_prefix_set("%s", "");
824 }
825 
826 static void
827 connection_migrate(struct hast_resource *res)
828 {
829 	struct proto_conn *conn;
830 	int16_t val = 0;
831 
832 	if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
833 		pjdlog_errno(LOG_WARNING,
834 		    "Unable to receive connection command");
835 		return;
836 	}
837 	if (proto_client(res->hr_remoteaddr, &conn) < 0) {
838 		val = errno;
839 		pjdlog_errno(LOG_WARNING,
840 		    "Unable to create outgoing connection to %s",
841 		    res->hr_remoteaddr);
842 		goto out;
843 	}
844 	if (proto_connect(conn, -1) < 0) {
845 		val = errno;
846 		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
847 		    res->hr_remoteaddr);
848 		proto_close(conn);
849 		goto out;
850 	}
851 	val = 0;
852 out:
853 	if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
854 		pjdlog_errno(LOG_WARNING,
855 		    "Unable to send reply to connection request");
856 	}
857 	if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
858 		pjdlog_errno(LOG_WARNING, "Unable to send connection");
859 }
860 
861 static void
862 main_loop(void)
863 {
864 	struct hast_resource *res;
865 	struct timeval seltimeout;
866 	struct timespec sigtimeout;
867 	int fd, maxfd, ret, signo;
868 	sigset_t mask;
869 	fd_set rfds;
870 
871 	seltimeout.tv_sec = REPORT_INTERVAL;
872 	seltimeout.tv_usec = 0;
873 	sigtimeout.tv_sec = 0;
874 	sigtimeout.tv_nsec = 0;
875 
876 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
877 	PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
878 	PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
879 	PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
880 	PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
881 
882 	pjdlog_info("Started successfully, running protocol version %d.",
883 	    HAST_PROTO_VERSION);
884 
885 	for (;;) {
886 		while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
887 			switch (signo) {
888 			case SIGINT:
889 			case SIGTERM:
890 				sigexit_received = true;
891 				terminate_workers();
892 				proto_close(cfg->hc_controlconn);
893 				exit(EX_OK);
894 				break;
895 			case SIGCHLD:
896 				child_exit();
897 				break;
898 			case SIGHUP:
899 				hastd_reload();
900 				break;
901 			default:
902 				PJDLOG_ABORT("Unexpected signal (%d).", signo);
903 			}
904 		}
905 
906 		/* Setup descriptors for select(2). */
907 		FD_ZERO(&rfds);
908 		maxfd = fd = proto_descriptor(cfg->hc_controlconn);
909 		PJDLOG_ASSERT(fd >= 0);
910 		FD_SET(fd, &rfds);
911 		fd = proto_descriptor(cfg->hc_listenconn);
912 		PJDLOG_ASSERT(fd >= 0);
913 		FD_SET(fd, &rfds);
914 		maxfd = fd > maxfd ? fd : maxfd;
915 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
916 			if (res->hr_event == NULL)
917 				continue;
918 			PJDLOG_ASSERT(res->hr_conn != NULL);
919 			fd = proto_descriptor(res->hr_event);
920 			PJDLOG_ASSERT(fd >= 0);
921 			FD_SET(fd, &rfds);
922 			maxfd = fd > maxfd ? fd : maxfd;
923 			if (res->hr_role == HAST_ROLE_PRIMARY) {
924 				/* Only primary workers asks for connections. */
925 				fd = proto_descriptor(res->hr_conn);
926 				PJDLOG_ASSERT(fd >= 0);
927 				FD_SET(fd, &rfds);
928 				maxfd = fd > maxfd ? fd : maxfd;
929 			}
930 		}
931 
932 		PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
933 		ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
934 		if (ret == 0)
935 			hook_check();
936 		else if (ret == -1) {
937 			if (errno == EINTR)
938 				continue;
939 			KEEP_ERRNO((void)pidfile_remove(pfh));
940 			pjdlog_exit(EX_OSERR, "select() failed");
941 		}
942 
943 		if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
944 			control_handle(cfg);
945 		if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds))
946 			listen_accept();
947 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
948 			if (res->hr_event == NULL)
949 				continue;
950 			PJDLOG_ASSERT(res->hr_conn != NULL);
951 			if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
952 				if (event_recv(res) == 0)
953 					continue;
954 				/* The worker process exited? */
955 				proto_close(res->hr_event);
956 				res->hr_event = NULL;
957 				proto_close(res->hr_conn);
958 				res->hr_conn = NULL;
959 				continue;
960 			}
961 			if (res->hr_role == HAST_ROLE_PRIMARY &&
962 			    FD_ISSET(proto_descriptor(res->hr_conn), &rfds)) {
963 				connection_migrate(res);
964 			}
965 		}
966 	}
967 }
968 
969 static void
970 dummy_sighandler(int sig __unused)
971 {
972 	/* Nothing to do. */
973 }
974 
975 int
976 main(int argc, char *argv[])
977 {
978 	const char *pidfile;
979 	pid_t otherpid;
980 	bool foreground;
981 	int debuglevel;
982 	sigset_t mask;
983 
984 	foreground = false;
985 	debuglevel = 0;
986 	pidfile = HASTD_PIDFILE;
987 
988 	for (;;) {
989 		int ch;
990 
991 		ch = getopt(argc, argv, "c:dFhP:");
992 		if (ch == -1)
993 			break;
994 		switch (ch) {
995 		case 'c':
996 			cfgpath = optarg;
997 			break;
998 		case 'd':
999 			debuglevel++;
1000 			break;
1001 		case 'F':
1002 			foreground = true;
1003 			break;
1004 		case 'P':
1005 			pidfile = optarg;
1006 			break;
1007 		case 'h':
1008 		default:
1009 			usage();
1010 		}
1011 	}
1012 	argc -= optind;
1013 	argv += optind;
1014 
1015 	pjdlog_init(PJDLOG_MODE_STD);
1016 	pjdlog_debug_set(debuglevel);
1017 
1018 	g_gate_load();
1019 
1020 	pfh = pidfile_open(pidfile, 0600, &otherpid);
1021 	if (pfh == NULL) {
1022 		if (errno == EEXIST) {
1023 			pjdlog_exitx(EX_TEMPFAIL,
1024 			    "Another hastd is already running, pid: %jd.",
1025 			    (intmax_t)otherpid);
1026 		}
1027 		/* If we cannot create pidfile from other reasons, only warn. */
1028 		pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
1029 	}
1030 
1031 	cfg = yy_config_parse(cfgpath, true);
1032 	PJDLOG_ASSERT(cfg != NULL);
1033 
1034 	/*
1035 	 * Restore default actions for interesting signals in case parent
1036 	 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1037 	 */
1038 	PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1039 	PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1040 	PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1041 	/*
1042 	 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1043 	 * so we can mask it.
1044 	 */
1045 	PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1046 
1047 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1048 	PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1049 	PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1050 	PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1051 	PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1052 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1053 
1054 	/* Listen on control address. */
1055 	if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
1056 		KEEP_ERRNO((void)pidfile_remove(pfh));
1057 		pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1058 		    cfg->hc_controladdr);
1059 	}
1060 	/* Listen for remote connections. */
1061 	if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) {
1062 		KEEP_ERRNO((void)pidfile_remove(pfh));
1063 		pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1064 		    cfg->hc_listenaddr);
1065 	}
1066 
1067 	if (!foreground) {
1068 		if (daemon(0, 0) < 0) {
1069 			KEEP_ERRNO((void)pidfile_remove(pfh));
1070 			pjdlog_exit(EX_OSERR, "Unable to daemonize");
1071 		}
1072 
1073 		/* Start logging to syslog. */
1074 		pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1075 
1076 		/* Write PID to a file. */
1077 		if (pidfile_write(pfh) < 0) {
1078 			pjdlog_errno(LOG_WARNING,
1079 			    "Unable to write PID to a file");
1080 		}
1081 	}
1082 
1083 	hook_init();
1084 
1085 	main_loop();
1086 
1087 	exit(0);
1088 }
1089