1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2009-2010 The FreeBSD Foundation
5 * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
6 * All rights reserved.
7 *
8 * This software was developed by Pawel Jakub Dawidek under sponsorship from
9 * the FreeBSD Foundation.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 #include <sys/cdefs.h>
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 <time.h>
50 #include <unistd.h>
51
52 #include <activemap.h>
53 #include <pjdlog.h>
54
55 #include "control.h"
56 #include "event.h"
57 #include "hast.h"
58 #include "hast_proto.h"
59 #include "hastd.h"
60 #include "hooks.h"
61 #include "subr.h"
62
63 /* Path to configuration file. */
64 const char *cfgpath = HAST_CONFIG;
65 /* Hastd configuration. */
66 static struct hastd_config *cfg;
67 /* Was SIGINT or SIGTERM signal received? */
68 bool sigexit_received = false;
69 /* Path to pidfile. */
70 static const char *pidfile;
71 /* Pidfile handle. */
72 struct pidfh *pfh;
73 /* Do we run in foreground? */
74 static bool foreground;
75
76 /* How often check for hooks running for too long. */
77 #define REPORT_INTERVAL 5
78
79 static void
usage(void)80 usage(void)
81 {
82
83 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
84 }
85
86 static void
g_gate_load(void)87 g_gate_load(void)
88 {
89
90 if (modfind("g_gate") == -1) {
91 /* Not present in kernel, try loading it. */
92 if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) {
93 if (errno != EEXIST) {
94 pjdlog_exit(EX_OSERR,
95 "Unable to load geom_gate module");
96 }
97 }
98 }
99 }
100
101 void
descriptors_cleanup(struct hast_resource * res)102 descriptors_cleanup(struct hast_resource *res)
103 {
104 struct hast_resource *tres, *tmres;
105 struct hastd_listen *lst;
106
107 TAILQ_FOREACH_SAFE(tres, &cfg->hc_resources, hr_next, tmres) {
108 if (tres == res) {
109 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY ||
110 (res->hr_remotein == NULL &&
111 res->hr_remoteout == NULL));
112 continue;
113 }
114 if (tres->hr_remotein != NULL)
115 proto_close(tres->hr_remotein);
116 if (tres->hr_remoteout != NULL)
117 proto_close(tres->hr_remoteout);
118 if (tres->hr_ctrl != NULL)
119 proto_close(tres->hr_ctrl);
120 if (tres->hr_event != NULL)
121 proto_close(tres->hr_event);
122 if (tres->hr_conn != NULL)
123 proto_close(tres->hr_conn);
124 TAILQ_REMOVE(&cfg->hc_resources, tres, hr_next);
125 free(tres);
126 }
127 if (cfg->hc_controlin != NULL)
128 proto_close(cfg->hc_controlin);
129 proto_close(cfg->hc_controlconn);
130 while ((lst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
131 TAILQ_REMOVE(&cfg->hc_listen, lst, hl_next);
132 if (lst->hl_conn != NULL)
133 proto_close(lst->hl_conn);
134 free(lst);
135 }
136 (void)pidfile_close(pfh);
137 hook_fini();
138 pjdlog_fini();
139 }
140
141 static const char *
dtype2str(mode_t mode)142 dtype2str(mode_t mode)
143 {
144
145 if (S_ISBLK(mode))
146 return ("block device");
147 else if (S_ISCHR(mode))
148 return ("character device");
149 else if (S_ISDIR(mode))
150 return ("directory");
151 else if (S_ISFIFO(mode))
152 return ("pipe or FIFO");
153 else if (S_ISLNK(mode))
154 return ("symbolic link");
155 else if (S_ISREG(mode))
156 return ("regular file");
157 else if (S_ISSOCK(mode))
158 return ("socket");
159 else if (S_ISWHT(mode))
160 return ("whiteout");
161 else
162 return ("unknown");
163 }
164
165 void
descriptors_assert(const struct hast_resource * res,int pjdlogmode)166 descriptors_assert(const struct hast_resource *res, int pjdlogmode)
167 {
168 char msg[256];
169 struct stat sb;
170 long maxfd;
171 bool isopen;
172 mode_t mode;
173 int fd;
174
175 /*
176 * At this point descriptor to syslog socket is closed, so if we want
177 * to log assertion message, we have to first store it in 'msg' local
178 * buffer and then open syslog socket and log it.
179 */
180 msg[0] = '\0';
181
182 maxfd = sysconf(_SC_OPEN_MAX);
183 if (maxfd == -1) {
184 pjdlog_init(pjdlogmode);
185 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
186 role2str(res->hr_role));
187 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed");
188 pjdlog_fini();
189 maxfd = 16384;
190 }
191 for (fd = 0; fd <= maxfd; fd++) {
192 if (fstat(fd, &sb) == 0) {
193 isopen = true;
194 mode = sb.st_mode;
195 } else if (errno == EBADF) {
196 isopen = false;
197 mode = 0;
198 } else {
199 (void)snprintf(msg, sizeof(msg),
200 "Unable to fstat descriptor %d: %s", fd,
201 strerror(errno));
202 break;
203 }
204 if (fd == STDIN_FILENO || fd == STDOUT_FILENO ||
205 fd == STDERR_FILENO) {
206 if (!isopen) {
207 (void)snprintf(msg, sizeof(msg),
208 "Descriptor %d (%s) is closed, but should be open.",
209 fd, (fd == STDIN_FILENO ? "stdin" :
210 (fd == STDOUT_FILENO ? "stdout" : "stderr")));
211 break;
212 }
213 } else if (fd == proto_descriptor(res->hr_event)) {
214 if (!isopen) {
215 (void)snprintf(msg, sizeof(msg),
216 "Descriptor %d (event) 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 (event) is %s, but should be %s.",
223 fd, dtype2str(mode), dtype2str(S_IFSOCK));
224 break;
225 }
226 } else if (fd == proto_descriptor(res->hr_ctrl)) {
227 if (!isopen) {
228 (void)snprintf(msg, sizeof(msg),
229 "Descriptor %d (ctrl) 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 (ctrl) is %s, but should be %s.",
236 fd, dtype2str(mode), dtype2str(S_IFSOCK));
237 break;
238 }
239 } else if (res->hr_role == HAST_ROLE_PRIMARY &&
240 fd == proto_descriptor(res->hr_conn)) {
241 if (!isopen) {
242 (void)snprintf(msg, sizeof(msg),
243 "Descriptor %d (conn) 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 (conn) 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 res->hr_conn != NULL &&
255 fd == proto_descriptor(res->hr_conn)) {
256 if (isopen) {
257 (void)snprintf(msg, sizeof(msg),
258 "Descriptor %d (conn) is open, but should be closed.",
259 fd);
260 break;
261 }
262 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
263 fd == proto_descriptor(res->hr_remotein)) {
264 if (!isopen) {
265 (void)snprintf(msg, sizeof(msg),
266 "Descriptor %d (remote in) is closed, but should be open.",
267 fd);
268 break;
269 }
270 if (!S_ISSOCK(mode)) {
271 (void)snprintf(msg, sizeof(msg),
272 "Descriptor %d (remote in) is %s, but should be %s.",
273 fd, dtype2str(mode), dtype2str(S_IFSOCK));
274 break;
275 }
276 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
277 fd == proto_descriptor(res->hr_remoteout)) {
278 if (!isopen) {
279 (void)snprintf(msg, sizeof(msg),
280 "Descriptor %d (remote out) is closed, but should be open.",
281 fd);
282 break;
283 }
284 if (!S_ISSOCK(mode)) {
285 (void)snprintf(msg, sizeof(msg),
286 "Descriptor %d (remote out) is %s, but should be %s.",
287 fd, dtype2str(mode), dtype2str(S_IFSOCK));
288 break;
289 }
290 } else {
291 if (isopen) {
292 (void)snprintf(msg, sizeof(msg),
293 "Descriptor %d is open (%s), but should be closed.",
294 fd, dtype2str(mode));
295 break;
296 }
297 }
298 }
299 if (msg[0] != '\0') {
300 pjdlog_init(pjdlogmode);
301 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
302 role2str(res->hr_role));
303 PJDLOG_ABORT("%s", msg);
304 }
305 }
306
307 static void
child_exit_log(unsigned int pid,int status)308 child_exit_log(unsigned int pid, int status)
309 {
310
311 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
312 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
313 pid);
314 } else if (WIFSIGNALED(status)) {
315 pjdlog_error("Worker process killed (pid=%u, signal=%d).",
316 pid, WTERMSIG(status));
317 } else {
318 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
319 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
320 }
321 }
322
323 static void
child_exit(void)324 child_exit(void)
325 {
326 struct hast_resource *res;
327 int status;
328 pid_t pid;
329
330 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
331 /* Find resource related to the process that just exited. */
332 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
333 if (pid == res->hr_workerpid)
334 break;
335 }
336 if (res == NULL) {
337 /*
338 * This can happen when new connection arrives and we
339 * cancel child responsible for the old one or if this
340 * was hook which we executed.
341 */
342 hook_check_one(pid, status);
343 continue;
344 }
345 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
346 role2str(res->hr_role));
347 child_exit_log(pid, status);
348 child_cleanup(res);
349 if (res->hr_role == HAST_ROLE_PRIMARY) {
350 /*
351 * Restart child process if it was killed by signal
352 * or exited because of temporary problem.
353 */
354 if (WIFSIGNALED(status) ||
355 (WIFEXITED(status) &&
356 WEXITSTATUS(status) == EX_TEMPFAIL)) {
357 sleep(1);
358 pjdlog_info("Restarting worker process.");
359 hastd_primary(res);
360 } else {
361 res->hr_role = HAST_ROLE_INIT;
362 pjdlog_info("Changing resource role back to %s.",
363 role2str(res->hr_role));
364 }
365 }
366 pjdlog_prefix_set("%s", "");
367 }
368 }
369
370 static bool
resource_needs_restart(const struct hast_resource * res0,const struct hast_resource * res1)371 resource_needs_restart(const struct hast_resource *res0,
372 const struct hast_resource *res1)
373 {
374
375 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
376
377 if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
378 return (true);
379 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
380 return (true);
381 if (res0->hr_role == HAST_ROLE_INIT ||
382 res0->hr_role == HAST_ROLE_SECONDARY) {
383 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
384 return (true);
385 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
386 return (true);
387 if (res0->hr_replication != res1->hr_replication)
388 return (true);
389 if (res0->hr_checksum != res1->hr_checksum)
390 return (true);
391 if (res0->hr_compression != res1->hr_compression)
392 return (true);
393 if (res0->hr_timeout != res1->hr_timeout)
394 return (true);
395 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
396 return (true);
397 /*
398 * When metaflush has changed we don't really need restart,
399 * but it is just easier this way.
400 */
401 if (res0->hr_metaflush != res1->hr_metaflush)
402 return (true);
403 }
404 return (false);
405 }
406
407 static bool
resource_needs_reload(const struct hast_resource * res0,const struct hast_resource * res1)408 resource_needs_reload(const struct hast_resource *res0,
409 const struct hast_resource *res1)
410 {
411
412 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
413 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0);
414 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
415
416 if (res0->hr_role != HAST_ROLE_PRIMARY)
417 return (false);
418
419 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
420 return (true);
421 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
422 return (true);
423 if (res0->hr_replication != res1->hr_replication)
424 return (true);
425 if (res0->hr_checksum != res1->hr_checksum)
426 return (true);
427 if (res0->hr_compression != res1->hr_compression)
428 return (true);
429 if (res0->hr_timeout != res1->hr_timeout)
430 return (true);
431 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
432 return (true);
433 if (res0->hr_metaflush != res1->hr_metaflush)
434 return (true);
435 return (false);
436 }
437
438 static void
resource_reload(const struct hast_resource * res)439 resource_reload(const struct hast_resource *res)
440 {
441 struct nv *nvin, *nvout;
442 int error;
443
444 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
445
446 nvout = nv_alloc();
447 nv_add_uint8(nvout, CONTROL_RELOAD, "cmd");
448 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr");
449 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr");
450 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication");
451 nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum");
452 nv_add_int32(nvout, (int32_t)res->hr_compression, "compression");
453 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout");
454 nv_add_string(nvout, res->hr_exec, "exec");
455 nv_add_int32(nvout, (int32_t)res->hr_metaflush, "metaflush");
456 if (nv_error(nvout) != 0) {
457 nv_free(nvout);
458 pjdlog_error("Unable to allocate header for reload message.");
459 return;
460 }
461 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) == -1) {
462 pjdlog_errno(LOG_ERR, "Unable to send reload message");
463 nv_free(nvout);
464 return;
465 }
466 nv_free(nvout);
467
468 /* Receive response. */
469 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) == -1) {
470 pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
471 return;
472 }
473 error = nv_get_int16(nvin, "error");
474 nv_free(nvin);
475 if (error != 0) {
476 pjdlog_common(LOG_ERR, 0, error, "Reload failed");
477 return;
478 }
479 }
480
481 static void
hastd_reload(void)482 hastd_reload(void)
483 {
484 struct hastd_config *newcfg;
485 struct hast_resource *nres, *cres, *tres;
486 struct hastd_listen *nlst, *clst;
487 struct pidfh *newpfh;
488 unsigned int nlisten;
489 uint8_t role;
490 pid_t otherpid;
491
492 pjdlog_info("Reloading configuration...");
493
494 newpfh = NULL;
495
496 newcfg = yy_config_parse(cfgpath, false);
497 if (newcfg == NULL)
498 goto failed;
499
500 /*
501 * Check if control address has changed.
502 */
503 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
504 if (proto_server(newcfg->hc_controladdr,
505 &newcfg->hc_controlconn) == -1) {
506 pjdlog_errno(LOG_ERR,
507 "Unable to listen on control address %s",
508 newcfg->hc_controladdr);
509 goto failed;
510 }
511 }
512 /*
513 * Check if any listen address has changed.
514 */
515 nlisten = 0;
516 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
517 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
518 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
519 break;
520 }
521 if (clst != NULL && clst->hl_conn != NULL) {
522 pjdlog_info("Keep listening on address %s.",
523 nlst->hl_addr);
524 nlst->hl_conn = clst->hl_conn;
525 nlisten++;
526 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
527 pjdlog_info("Listening on new address %s.",
528 nlst->hl_addr);
529 nlisten++;
530 } else {
531 pjdlog_errno(LOG_WARNING,
532 "Unable to listen on address %s", nlst->hl_addr);
533 }
534 }
535 if (nlisten == 0) {
536 pjdlog_error("No addresses to listen on.");
537 goto failed;
538 }
539 /*
540 * Check if pidfile's path has changed.
541 */
542 if (!foreground && pidfile == NULL &&
543 strcmp(cfg->hc_pidfile, newcfg->hc_pidfile) != 0) {
544 newpfh = pidfile_open(newcfg->hc_pidfile, 0600, &otherpid);
545 if (newpfh == NULL) {
546 if (errno == EEXIST) {
547 pjdlog_errno(LOG_WARNING,
548 "Another hastd is already running, pidfile: %s, pid: %jd.",
549 newcfg->hc_pidfile, (intmax_t)otherpid);
550 } else {
551 pjdlog_errno(LOG_WARNING,
552 "Unable to open or create pidfile %s",
553 newcfg->hc_pidfile);
554 }
555 } else if (pidfile_write(newpfh) == -1) {
556 /* Write PID to a file. */
557 pjdlog_errno(LOG_WARNING,
558 "Unable to write PID to file %s",
559 newcfg->hc_pidfile);
560 } else {
561 pjdlog_debug(1, "PID stored in %s.",
562 newcfg->hc_pidfile);
563 }
564 }
565
566 /* No failures from now on. */
567
568 /*
569 * Switch to new control socket.
570 */
571 if (newcfg->hc_controlconn != NULL) {
572 pjdlog_info("Control socket changed from %s to %s.",
573 cfg->hc_controladdr, newcfg->hc_controladdr);
574 proto_close(cfg->hc_controlconn);
575 cfg->hc_controlconn = newcfg->hc_controlconn;
576 newcfg->hc_controlconn = NULL;
577 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
578 sizeof(cfg->hc_controladdr));
579 }
580 /*
581 * Switch to new pidfile.
582 */
583 if (newpfh != NULL) {
584 pjdlog_info("Pidfile changed from %s to %s.", cfg->hc_pidfile,
585 newcfg->hc_pidfile);
586 (void)pidfile_remove(pfh);
587 pfh = newpfh;
588 (void)strlcpy(cfg->hc_pidfile, newcfg->hc_pidfile,
589 sizeof(cfg->hc_pidfile));
590 }
591 /*
592 * Switch to new listen addresses. Close all that were removed.
593 */
594 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
595 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
596 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
597 break;
598 }
599 if (nlst == NULL && clst->hl_conn != NULL) {
600 proto_close(clst->hl_conn);
601 pjdlog_info("No longer listening on address %s.",
602 clst->hl_addr);
603 }
604 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
605 free(clst);
606 }
607 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
608
609 /*
610 * Stop and remove resources that were removed from the configuration.
611 */
612 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
613 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
614 if (strcmp(cres->hr_name, nres->hr_name) == 0)
615 break;
616 }
617 if (nres == NULL) {
618 control_set_role(cres, HAST_ROLE_INIT);
619 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
620 pjdlog_info("Resource %s removed.", cres->hr_name);
621 free(cres);
622 }
623 }
624 /*
625 * Move new resources to the current configuration.
626 */
627 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
628 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
629 if (strcmp(cres->hr_name, nres->hr_name) == 0)
630 break;
631 }
632 if (cres == NULL) {
633 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
634 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
635 pjdlog_info("Resource %s added.", nres->hr_name);
636 }
637 }
638 /*
639 * Deal with modified resources.
640 * Depending on what has changed exactly we might want to perform
641 * different actions.
642 *
643 * We do full resource restart in the following situations:
644 * Resource role is INIT or SECONDARY.
645 * Resource role is PRIMARY and path to local component or provider
646 * name has changed.
647 * In case of PRIMARY, the worker process will be killed and restarted,
648 * which also means removing /dev/hast/<name> provider and
649 * recreating it.
650 *
651 * We do just reload (send SIGHUP to worker process) if we act as
652 * PRIMARY, but only if remote address, source address, replication
653 * mode, timeout, execution path or metaflush has changed.
654 * For those, there is no need to restart worker process.
655 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
656 * source address has changed or it will set new timeout if only timeout
657 * has changed or it will update metaflush if only metaflush has
658 * changed.
659 */
660 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
661 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
662 if (strcmp(cres->hr_name, nres->hr_name) == 0)
663 break;
664 }
665 PJDLOG_ASSERT(cres != NULL);
666 if (resource_needs_restart(cres, nres)) {
667 pjdlog_info("Resource %s configuration was modified, restarting it.",
668 cres->hr_name);
669 role = cres->hr_role;
670 control_set_role(cres, HAST_ROLE_INIT);
671 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
672 free(cres);
673 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
674 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
675 control_set_role(nres, role);
676 } else if (resource_needs_reload(cres, nres)) {
677 pjdlog_info("Resource %s configuration was modified, reloading it.",
678 cres->hr_name);
679 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
680 sizeof(cres->hr_remoteaddr));
681 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr,
682 sizeof(cres->hr_sourceaddr));
683 cres->hr_replication = nres->hr_replication;
684 cres->hr_checksum = nres->hr_checksum;
685 cres->hr_compression = nres->hr_compression;
686 cres->hr_timeout = nres->hr_timeout;
687 strlcpy(cres->hr_exec, nres->hr_exec,
688 sizeof(cres->hr_exec));
689 cres->hr_metaflush = nres->hr_metaflush;
690 if (cres->hr_workerpid != 0)
691 resource_reload(cres);
692 }
693 }
694
695 yy_config_free(newcfg);
696 pjdlog_info("Configuration reloaded successfully.");
697 return;
698 failed:
699 if (newcfg != NULL) {
700 if (newcfg->hc_controlconn != NULL)
701 proto_close(newcfg->hc_controlconn);
702 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) {
703 if (nlst->hl_conn != NULL) {
704 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
705 if (strcmp(nlst->hl_addr,
706 clst->hl_addr) == 0) {
707 break;
708 }
709 }
710 if (clst == NULL || clst->hl_conn == NULL)
711 proto_close(nlst->hl_conn);
712 }
713 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
714 free(nlst);
715 }
716 yy_config_free(newcfg);
717 }
718 if (newpfh != NULL)
719 (void)pidfile_remove(newpfh);
720 pjdlog_warning("Configuration not reloaded.");
721 }
722
723 static void
terminate_workers(void)724 terminate_workers(void)
725 {
726 struct hast_resource *res;
727
728 pjdlog_info("Termination signal received, exiting.");
729 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
730 if (res->hr_workerpid == 0)
731 continue;
732 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
733 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
734 if (kill(res->hr_workerpid, SIGTERM) == 0)
735 continue;
736 pjdlog_errno(LOG_WARNING,
737 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
738 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
739 }
740 }
741
742 static void
listen_accept(struct hastd_listen * lst)743 listen_accept(struct hastd_listen *lst)
744 {
745 struct hast_resource *res;
746 struct proto_conn *conn;
747 struct nv *nvin, *nvout, *nverr;
748 const char *resname;
749 const unsigned char *token;
750 char laddr[256], raddr[256];
751 uint8_t version;
752 size_t size;
753 pid_t pid;
754 int status;
755
756 proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
757 pjdlog_debug(1, "Accepting connection to %s.", laddr);
758
759 if (proto_accept(lst->hl_conn, &conn) == -1) {
760 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
761 return;
762 }
763
764 proto_local_address(conn, laddr, sizeof(laddr));
765 proto_remote_address(conn, raddr, sizeof(raddr));
766 pjdlog_info("Connection from %s to %s.", raddr, laddr);
767
768 /* Error in setting timeout is not critical, but why should it fail? */
769 if (proto_timeout(conn, HAST_TIMEOUT) == -1)
770 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
771
772 nvin = nvout = nverr = NULL;
773
774 /*
775 * Before receiving any data see if remote host have access to any
776 * resource.
777 */
778 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
779 if (proto_address_match(conn, res->hr_remoteaddr))
780 break;
781 }
782 if (res == NULL) {
783 pjdlog_error("Client %s isn't known.", raddr);
784 goto close;
785 }
786 /* Ok, remote host can access at least one resource. */
787
788 if (hast_proto_recv_hdr(conn, &nvin) == -1) {
789 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
790 raddr);
791 goto close;
792 }
793
794 resname = nv_get_string(nvin, "resource");
795 if (resname == NULL) {
796 pjdlog_error("No 'resource' field in the header received from %s.",
797 raddr);
798 goto close;
799 }
800 pjdlog_debug(2, "%s: resource=%s", raddr, resname);
801 version = nv_get_uint8(nvin, "version");
802 pjdlog_debug(2, "%s: version=%hhu", raddr, version);
803 if (version == 0) {
804 /*
805 * If no version is sent, it means this is protocol version 1.
806 */
807 version = 1;
808 }
809 token = nv_get_uint8_array(nvin, &size, "token");
810 /*
811 * NULL token means that this is first connection.
812 */
813 if (token != NULL && size != sizeof(res->hr_token)) {
814 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
815 raddr, sizeof(res->hr_token), size);
816 goto close;
817 }
818
819 /*
820 * From now on we want to send errors to the remote node.
821 */
822 nverr = nv_alloc();
823
824 /* Find resource related to this connection. */
825 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
826 if (strcmp(resname, res->hr_name) == 0)
827 break;
828 }
829 /* Have we found the resource? */
830 if (res == NULL) {
831 pjdlog_error("No resource '%s' as requested by %s.",
832 resname, raddr);
833 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
834 goto fail;
835 }
836
837 /* Now that we know resource name setup log prefix. */
838 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
839
840 /* Does the remote host have access to this resource? */
841 if (!proto_address_match(conn, res->hr_remoteaddr)) {
842 pjdlog_error("Client %s has no access to the resource.", raddr);
843 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
844 goto fail;
845 }
846 /* Is the resource marked as secondary? */
847 if (res->hr_role != HAST_ROLE_SECONDARY) {
848 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.",
849 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
850 raddr);
851 nv_add_stringf(nverr, "errmsg",
852 "Remote node acts as %s for the resource and not as %s.",
853 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
854 if (res->hr_role == HAST_ROLE_PRIMARY) {
855 /*
856 * If we act as primary request the other side to wait
857 * for us a bit, as we might be finishing cleanups.
858 */
859 nv_add_uint8(nverr, 1, "wait");
860 }
861 goto fail;
862 }
863 /* Does token (if exists) match? */
864 if (token != NULL && memcmp(token, res->hr_token,
865 sizeof(res->hr_token)) != 0) {
866 pjdlog_error("Token received from %s doesn't match.", raddr);
867 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
868 goto fail;
869 }
870 /*
871 * If there is no token, but we have half-open connection
872 * (only remotein) or full connection (worker process is running)
873 * we have to cancel those and accept the new connection.
874 */
875 if (token == NULL) {
876 PJDLOG_ASSERT(res->hr_remoteout == NULL);
877 pjdlog_debug(1, "Initial connection from %s.", raddr);
878 if (res->hr_workerpid != 0) {
879 PJDLOG_ASSERT(res->hr_remotein == NULL);
880 pjdlog_debug(1,
881 "Worker process exists (pid=%u), stopping it.",
882 (unsigned int)res->hr_workerpid);
883 /* Stop child process. */
884 if (kill(res->hr_workerpid, SIGINT) == -1) {
885 pjdlog_errno(LOG_ERR,
886 "Unable to stop worker process (pid=%u)",
887 (unsigned int)res->hr_workerpid);
888 /*
889 * Other than logging the problem we
890 * ignore it - nothing smart to do.
891 */
892 }
893 /* Wait for it to exit. */
894 else if ((pid = waitpid(res->hr_workerpid,
895 &status, 0)) != res->hr_workerpid) {
896 /* We can only log the problem. */
897 pjdlog_errno(LOG_ERR,
898 "Waiting for worker process (pid=%u) failed",
899 (unsigned int)res->hr_workerpid);
900 } else {
901 child_exit_log(res->hr_workerpid, status);
902 }
903 child_cleanup(res);
904 } else if (res->hr_remotein != NULL) {
905 char oaddr[256];
906
907 proto_remote_address(res->hr_remotein, oaddr,
908 sizeof(oaddr));
909 pjdlog_debug(1,
910 "Canceling half-open connection from %s on connection from %s.",
911 oaddr, raddr);
912 proto_close(res->hr_remotein);
913 res->hr_remotein = NULL;
914 }
915 }
916
917 /*
918 * Checks and cleanups are done.
919 */
920
921 if (token == NULL) {
922 if (version > HAST_PROTO_VERSION) {
923 pjdlog_info("Remote protocol version %hhu is not supported, falling back to version %hhu.",
924 version, (unsigned char)HAST_PROTO_VERSION);
925 version = HAST_PROTO_VERSION;
926 }
927 pjdlog_debug(1, "Negotiated protocol version %hhu.", version);
928 res->hr_version = version;
929 arc4random_buf(res->hr_token, sizeof(res->hr_token));
930 nvout = nv_alloc();
931 nv_add_uint8(nvout, version, "version");
932 nv_add_uint8_array(nvout, res->hr_token,
933 sizeof(res->hr_token), "token");
934 if (nv_error(nvout) != 0) {
935 pjdlog_common(LOG_ERR, 0, nv_error(nvout),
936 "Unable to prepare return header for %s", raddr);
937 nv_add_stringf(nverr, "errmsg",
938 "Remote node was unable to prepare return header: %s.",
939 strerror(nv_error(nvout)));
940 goto fail;
941 }
942 if (hast_proto_send(res, conn, nvout, NULL, 0) == -1) {
943 int error = errno;
944
945 pjdlog_errno(LOG_ERR, "Unable to send response to %s",
946 raddr);
947 nv_add_stringf(nverr, "errmsg",
948 "Remote node was unable to send response: %s.",
949 strerror(error));
950 goto fail;
951 }
952 res->hr_remotein = conn;
953 pjdlog_debug(1, "Incoming connection from %s configured.",
954 raddr);
955 } else {
956 res->hr_remoteout = conn;
957 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
958 hastd_secondary(res, nvin);
959 }
960 nv_free(nvin);
961 nv_free(nvout);
962 nv_free(nverr);
963 pjdlog_prefix_set("%s", "");
964 return;
965 fail:
966 if (nv_error(nverr) != 0) {
967 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
968 "Unable to prepare error header for %s", raddr);
969 goto close;
970 }
971 if (hast_proto_send(NULL, conn, nverr, NULL, 0) == -1) {
972 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
973 goto close;
974 }
975 close:
976 if (nvin != NULL)
977 nv_free(nvin);
978 if (nvout != NULL)
979 nv_free(nvout);
980 if (nverr != NULL)
981 nv_free(nverr);
982 proto_close(conn);
983 pjdlog_prefix_set("%s", "");
984 }
985
986 static void
connection_migrate(struct hast_resource * res)987 connection_migrate(struct hast_resource *res)
988 {
989 struct proto_conn *conn;
990 int16_t val = 0;
991
992 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
993
994 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
995
996 if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) {
997 pjdlog_errno(LOG_WARNING,
998 "Unable to receive connection command");
999 return;
1000 }
1001 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
1002 res->hr_remoteaddr, &conn) == -1) {
1003 val = errno;
1004 pjdlog_errno(LOG_WARNING,
1005 "Unable to create outgoing connection to %s",
1006 res->hr_remoteaddr);
1007 goto out;
1008 }
1009 if (proto_connect(conn, -1) == -1) {
1010 val = errno;
1011 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
1012 res->hr_remoteaddr);
1013 proto_close(conn);
1014 goto out;
1015 }
1016 val = 0;
1017 out:
1018 if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) {
1019 pjdlog_errno(LOG_WARNING,
1020 "Unable to send reply to connection request");
1021 }
1022 if (val == 0 && proto_connection_send(res->hr_conn, conn) == -1)
1023 pjdlog_errno(LOG_WARNING, "Unable to send connection");
1024
1025 pjdlog_prefix_set("%s", "");
1026 }
1027
1028 static void
check_signals(void)1029 check_signals(void)
1030 {
1031 struct timespec sigtimeout;
1032 sigset_t mask;
1033 int signo;
1034
1035 sigtimeout.tv_sec = 0;
1036 sigtimeout.tv_nsec = 0;
1037
1038 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1039 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1040 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1041 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1042 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1043
1044 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
1045 switch (signo) {
1046 case SIGINT:
1047 case SIGTERM:
1048 sigexit_received = true;
1049 terminate_workers();
1050 proto_close(cfg->hc_controlconn);
1051 exit(EX_OK);
1052 break;
1053 case SIGCHLD:
1054 child_exit();
1055 break;
1056 case SIGHUP:
1057 hastd_reload();
1058 break;
1059 default:
1060 PJDLOG_ABORT("Unexpected signal (%d).", signo);
1061 }
1062 }
1063 }
1064
1065 static void
main_loop(void)1066 main_loop(void)
1067 {
1068 struct hast_resource *res;
1069 struct hastd_listen *lst;
1070 struct timeval seltimeout;
1071 int fd, maxfd, ret;
1072 time_t lastcheck, now;
1073 fd_set rfds;
1074
1075 lastcheck = time(NULL);
1076 seltimeout.tv_sec = REPORT_INTERVAL;
1077 seltimeout.tv_usec = 0;
1078
1079 for (;;) {
1080 check_signals();
1081
1082 /* Setup descriptors for select(2). */
1083 FD_ZERO(&rfds);
1084 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1085 PJDLOG_ASSERT(fd >= 0);
1086 FD_SET(fd, &rfds);
1087 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1088 if (lst->hl_conn == NULL)
1089 continue;
1090 fd = proto_descriptor(lst->hl_conn);
1091 PJDLOG_ASSERT(fd >= 0);
1092 FD_SET(fd, &rfds);
1093 maxfd = MAX(fd, maxfd);
1094 }
1095 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1096 if (res->hr_event == NULL)
1097 continue;
1098 fd = proto_descriptor(res->hr_event);
1099 PJDLOG_ASSERT(fd >= 0);
1100 FD_SET(fd, &rfds);
1101 maxfd = MAX(fd, maxfd);
1102 if (res->hr_role == HAST_ROLE_PRIMARY) {
1103 /* Only primary workers asks for connections. */
1104 PJDLOG_ASSERT(res->hr_conn != NULL);
1105 fd = proto_descriptor(res->hr_conn);
1106 PJDLOG_ASSERT(fd >= 0);
1107 FD_SET(fd, &rfds);
1108 maxfd = MAX(fd, maxfd);
1109 } else {
1110 PJDLOG_ASSERT(res->hr_conn == NULL);
1111 }
1112 }
1113
1114 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1115 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1116 now = time(NULL);
1117 if (lastcheck + REPORT_INTERVAL <= now) {
1118 hook_check();
1119 lastcheck = now;
1120 }
1121 if (ret == 0) {
1122 /*
1123 * select(2) timed out, so there should be no
1124 * descriptors to check.
1125 */
1126 continue;
1127 } else if (ret == -1) {
1128 if (errno == EINTR)
1129 continue;
1130 KEEP_ERRNO((void)pidfile_remove(pfh));
1131 pjdlog_exit(EX_OSERR, "select() failed");
1132 }
1133
1134 /*
1135 * Check for signals before we do anything to update our
1136 * info about terminated workers in the meantime.
1137 */
1138 check_signals();
1139
1140 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
1141 control_handle(cfg);
1142 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1143 if (lst->hl_conn == NULL)
1144 continue;
1145 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1146 listen_accept(lst);
1147 }
1148 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1149 if (res->hr_event == NULL)
1150 continue;
1151 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1152 if (event_recv(res) == 0)
1153 continue;
1154 /* The worker process exited? */
1155 proto_close(res->hr_event);
1156 res->hr_event = NULL;
1157 if (res->hr_conn != NULL) {
1158 proto_close(res->hr_conn);
1159 res->hr_conn = NULL;
1160 }
1161 continue;
1162 }
1163 if (res->hr_role == HAST_ROLE_PRIMARY) {
1164 PJDLOG_ASSERT(res->hr_conn != NULL);
1165 if (FD_ISSET(proto_descriptor(res->hr_conn),
1166 &rfds)) {
1167 connection_migrate(res);
1168 }
1169 } else {
1170 PJDLOG_ASSERT(res->hr_conn == NULL);
1171 }
1172 }
1173 }
1174 }
1175
1176 static void
dummy_sighandler(int sig __unused)1177 dummy_sighandler(int sig __unused)
1178 {
1179 /* Nothing to do. */
1180 }
1181
1182 int
main(int argc,char * argv[])1183 main(int argc, char *argv[])
1184 {
1185 struct hastd_listen *lst;
1186 pid_t otherpid;
1187 int debuglevel;
1188 sigset_t mask;
1189
1190 foreground = false;
1191 debuglevel = 0;
1192
1193 for (;;) {
1194 int ch;
1195
1196 ch = getopt(argc, argv, "c:dFhP:");
1197 if (ch == -1)
1198 break;
1199 switch (ch) {
1200 case 'c':
1201 cfgpath = optarg;
1202 break;
1203 case 'd':
1204 debuglevel++;
1205 break;
1206 case 'F':
1207 foreground = true;
1208 break;
1209 case 'P':
1210 pidfile = optarg;
1211 break;
1212 case 'h':
1213 default:
1214 usage();
1215 }
1216 }
1217 argc -= optind;
1218 argv += optind;
1219
1220 pjdlog_init(PJDLOG_MODE_STD);
1221 pjdlog_debug_set(debuglevel);
1222
1223 closefrom(MAX(MAX(STDIN_FILENO, STDOUT_FILENO), STDERR_FILENO) + 1);
1224 g_gate_load();
1225
1226 /*
1227 * When path to the configuration file is relative, obtain full path,
1228 * so we can always find the file, even after daemonizing and changing
1229 * working directory to /.
1230 */
1231 if (cfgpath[0] != '/') {
1232 const char *newcfgpath;
1233
1234 newcfgpath = realpath(cfgpath, NULL);
1235 if (newcfgpath == NULL) {
1236 pjdlog_exit(EX_CONFIG,
1237 "Unable to obtain full path of %s", cfgpath);
1238 }
1239 cfgpath = newcfgpath;
1240 }
1241
1242 cfg = yy_config_parse(cfgpath, true);
1243 PJDLOG_ASSERT(cfg != NULL);
1244
1245 if (pidfile != NULL) {
1246 if (strlcpy(cfg->hc_pidfile, pidfile,
1247 sizeof(cfg->hc_pidfile)) >= sizeof(cfg->hc_pidfile)) {
1248 pjdlog_exitx(EX_CONFIG, "Pidfile path is too long.");
1249 }
1250 }
1251
1252 if (pidfile != NULL || !foreground) {
1253 pfh = pidfile_open(cfg->hc_pidfile, 0600, &otherpid);
1254 if (pfh == NULL) {
1255 if (errno == EEXIST) {
1256 pjdlog_exitx(EX_TEMPFAIL,
1257 "Another hastd is already running, pidfile: %s, pid: %jd.",
1258 cfg->hc_pidfile, (intmax_t)otherpid);
1259 }
1260 /*
1261 * If we cannot create pidfile for other reasons,
1262 * only warn.
1263 */
1264 pjdlog_errno(LOG_WARNING,
1265 "Unable to open or create pidfile %s",
1266 cfg->hc_pidfile);
1267 }
1268 }
1269
1270 /*
1271 * Restore default actions for interesting signals in case parent
1272 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1273 */
1274 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1275 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1276 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1277 /*
1278 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1279 * so we can mask it.
1280 */
1281 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1282
1283 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1284 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1285 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1286 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1287 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1288 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1289
1290 /* Listen on control address. */
1291 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) == -1) {
1292 KEEP_ERRNO((void)pidfile_remove(pfh));
1293 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1294 cfg->hc_controladdr);
1295 }
1296 /* Listen for remote connections. */
1297 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1298 if (proto_server(lst->hl_addr, &lst->hl_conn) == -1) {
1299 KEEP_ERRNO((void)pidfile_remove(pfh));
1300 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1301 lst->hl_addr);
1302 }
1303 }
1304
1305 if (!foreground) {
1306 if (daemon(0, 0) == -1) {
1307 KEEP_ERRNO((void)pidfile_remove(pfh));
1308 pjdlog_exit(EX_OSERR, "Unable to daemonize");
1309 }
1310
1311 /* Start logging to syslog. */
1312 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1313 }
1314 if (pidfile != NULL || !foreground) {
1315 /* Write PID to a file. */
1316 if (pidfile_write(pfh) == -1) {
1317 pjdlog_errno(LOG_WARNING,
1318 "Unable to write PID to a file %s",
1319 cfg->hc_pidfile);
1320 } else {
1321 pjdlog_debug(1, "PID stored in %s.", cfg->hc_pidfile);
1322 }
1323 }
1324
1325 pjdlog_info("Started successfully, running protocol version %d.",
1326 HAST_PROTO_VERSION);
1327
1328 pjdlog_debug(1, "Listening on control address %s.",
1329 cfg->hc_controladdr);
1330 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next)
1331 pjdlog_info("Listening on address %s.", lst->hl_addr);
1332
1333 hook_init();
1334
1335 main_loop();
1336
1337 exit(0);
1338 }
1339