2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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.
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
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/linker.h>
36 #include <sys/module.h>
52 #include <activemap.h>
58 #include "hast_proto.h"
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 /* PID file handle. */
72 /* How often check for hooks running for too long. */
73 #define REPORT_INTERVAL 5
79 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
86 if (modfind("g_gate") == -1) {
87 /* Not present in kernel, try loading it. */
88 if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) {
89 if (errno != EEXIST) {
91 "Unable to load geom_gate module");
98 descriptors_cleanup(struct hast_resource *res)
100 struct hast_resource *tres;
101 struct hastd_listen *lst;
103 TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) {
105 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY ||
106 (res->hr_remotein == NULL &&
107 res->hr_remoteout == NULL));
110 if (tres->hr_remotein != NULL)
111 proto_close(tres->hr_remotein);
112 if (tres->hr_remoteout != NULL)
113 proto_close(tres->hr_remoteout);
114 if (tres->hr_ctrl != NULL)
115 proto_close(tres->hr_ctrl);
116 if (tres->hr_event != NULL)
117 proto_close(tres->hr_event);
118 if (tres->hr_conn != NULL)
119 proto_close(tres->hr_conn);
121 if (cfg->hc_controlin != NULL)
122 proto_close(cfg->hc_controlin);
123 proto_close(cfg->hc_controlconn);
124 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
125 if (lst->hl_conn != NULL)
126 proto_close(lst->hl_conn);
128 (void)pidfile_close(pfh);
134 dtype2str(mode_t mode)
138 return ("block device");
139 else if (S_ISCHR(mode))
140 return ("character device");
141 else if (S_ISDIR(mode))
142 return ("directory");
143 else if (S_ISFIFO(mode))
144 return ("pipe or FIFO");
145 else if (S_ISLNK(mode))
146 return ("symbolic link");
147 else if (S_ISREG(mode))
148 return ("regular file");
149 else if (S_ISSOCK(mode))
151 else if (S_ISWHT(mode))
158 descriptors_assert(const struct hast_resource *res, int pjdlogmode)
168 * At this point descriptor to syslog socket is closed, so if we want
169 * to log assertion message, we have to first store it in 'msg' local
170 * buffer and then open syslog socket and log it.
174 maxfd = sysconf(_SC_OPEN_MAX);
176 pjdlog_init(pjdlogmode);
177 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
178 role2str(res->hr_role));
179 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed");
183 for (fd = 0; fd <= maxfd; fd++) {
184 if (fstat(fd, &sb) == 0) {
187 } else if (errno == EBADF) {
191 (void)snprintf(msg, sizeof(msg),
192 "Unable to fstat descriptor %d: %s", fd,
196 if (fd == STDIN_FILENO || fd == STDOUT_FILENO ||
197 fd == STDERR_FILENO) {
199 (void)snprintf(msg, sizeof(msg),
200 "Descriptor %d (%s) is closed, but should be open.",
201 fd, (fd == STDIN_FILENO ? "stdin" :
202 (fd == STDOUT_FILENO ? "stdout" : "stderr")));
205 } else if (fd == proto_descriptor(res->hr_event)) {
207 (void)snprintf(msg, sizeof(msg),
208 "Descriptor %d (event) is closed, but should be open.",
212 if (!S_ISSOCK(mode)) {
213 (void)snprintf(msg, sizeof(msg),
214 "Descriptor %d (event) is %s, but should be %s.",
215 fd, dtype2str(mode), dtype2str(S_IFSOCK));
218 } else if (fd == proto_descriptor(res->hr_ctrl)) {
220 (void)snprintf(msg, sizeof(msg),
221 "Descriptor %d (ctrl) is closed, but should be open.",
225 if (!S_ISSOCK(mode)) {
226 (void)snprintf(msg, sizeof(msg),
227 "Descriptor %d (ctrl) is %s, but should be %s.",
228 fd, dtype2str(mode), dtype2str(S_IFSOCK));
231 } else if (res->hr_role == HAST_ROLE_PRIMARY &&
232 fd == proto_descriptor(res->hr_conn)) {
234 (void)snprintf(msg, sizeof(msg),
235 "Descriptor %d (conn) is closed, but should be open.",
239 if (!S_ISSOCK(mode)) {
240 (void)snprintf(msg, sizeof(msg),
241 "Descriptor %d (conn) is %s, but should be %s.",
242 fd, dtype2str(mode), dtype2str(S_IFSOCK));
245 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
246 res->hr_conn != NULL &&
247 fd == proto_descriptor(res->hr_conn)) {
249 (void)snprintf(msg, sizeof(msg),
250 "Descriptor %d (conn) is open, but should be closed.",
254 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
255 fd == proto_descriptor(res->hr_remotein)) {
257 (void)snprintf(msg, sizeof(msg),
258 "Descriptor %d (remote in) is closed, but should be open.",
262 if (!S_ISSOCK(mode)) {
263 (void)snprintf(msg, sizeof(msg),
264 "Descriptor %d (remote in) is %s, but should be %s.",
265 fd, dtype2str(mode), dtype2str(S_IFSOCK));
268 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
269 fd == proto_descriptor(res->hr_remoteout)) {
271 (void)snprintf(msg, sizeof(msg),
272 "Descriptor %d (remote out) is closed, but should be open.",
276 if (!S_ISSOCK(mode)) {
277 (void)snprintf(msg, sizeof(msg),
278 "Descriptor %d (remote out) is %s, but should be %s.",
279 fd, dtype2str(mode), dtype2str(S_IFSOCK));
284 (void)snprintf(msg, sizeof(msg),
285 "Descriptor %d is open (%s), but should be closed.",
286 fd, dtype2str(mode));
291 if (msg[0] != '\0') {
292 pjdlog_init(pjdlogmode);
293 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
294 role2str(res->hr_role));
295 PJDLOG_ABORT("%s", msg);
300 child_exit_log(unsigned int pid, int status)
303 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
304 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
306 } else if (WIFSIGNALED(status)) {
307 pjdlog_error("Worker process killed (pid=%u, signal=%d).",
308 pid, WTERMSIG(status));
310 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
311 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
318 struct hast_resource *res;
322 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
323 /* Find resource related to the process that just exited. */
324 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
325 if (pid == res->hr_workerpid)
330 * This can happen when new connection arrives and we
331 * cancel child responsible for the old one or if this
332 * was hook which we executed.
334 hook_check_one(pid, status);
337 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
338 role2str(res->hr_role));
339 child_exit_log(pid, status);
341 if (res->hr_role == HAST_ROLE_PRIMARY) {
343 * Restart child process if it was killed by signal
344 * or exited because of temporary problem.
346 if (WIFSIGNALED(status) ||
347 (WIFEXITED(status) &&
348 WEXITSTATUS(status) == EX_TEMPFAIL)) {
350 pjdlog_info("Restarting worker process.");
353 res->hr_role = HAST_ROLE_INIT;
354 pjdlog_info("Changing resource role back to %s.",
355 role2str(res->hr_role));
358 pjdlog_prefix_set("%s", "");
363 resource_needs_restart(const struct hast_resource *res0,
364 const struct hast_resource *res1)
367 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
369 if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
371 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
373 if (res0->hr_role == HAST_ROLE_INIT ||
374 res0->hr_role == HAST_ROLE_SECONDARY) {
375 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
377 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
379 if (res0->hr_replication != res1->hr_replication)
381 if (res0->hr_checksum != res1->hr_checksum)
383 if (res0->hr_compression != res1->hr_compression)
385 if (res0->hr_timeout != res1->hr_timeout)
387 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
390 * When metaflush has changed we don't really need restart,
391 * but it is just easier this way.
393 if (res0->hr_metaflush != res1->hr_metaflush)
400 resource_needs_reload(const struct hast_resource *res0,
401 const struct hast_resource *res1)
404 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
405 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0);
406 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
408 if (res0->hr_role != HAST_ROLE_PRIMARY)
411 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
413 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
415 if (res0->hr_replication != res1->hr_replication)
417 if (res0->hr_checksum != res1->hr_checksum)
419 if (res0->hr_compression != res1->hr_compression)
421 if (res0->hr_timeout != res1->hr_timeout)
423 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
425 if (res0->hr_metaflush != res1->hr_metaflush)
431 resource_reload(const struct hast_resource *res)
433 struct nv *nvin, *nvout;
436 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
439 nv_add_uint8(nvout, CONTROL_RELOAD, "cmd");
440 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr");
441 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr");
442 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication");
443 nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum");
444 nv_add_int32(nvout, (int32_t)res->hr_compression, "compression");
445 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout");
446 nv_add_string(nvout, res->hr_exec, "exec");
447 nv_add_int32(nvout, (int32_t)res->hr_metaflush, "metaflush");
448 if (nv_error(nvout) != 0) {
450 pjdlog_error("Unable to allocate header for reload message.");
453 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
454 pjdlog_errno(LOG_ERR, "Unable to send reload message");
460 /* Receive response. */
461 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
462 pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
465 error = nv_get_int16(nvin, "error");
468 pjdlog_common(LOG_ERR, 0, error, "Reload failed");
476 struct hastd_config *newcfg;
477 struct hast_resource *nres, *cres, *tres;
478 struct hastd_listen *nlst, *clst;
479 struct pidfh *newpfh;
480 unsigned int nlisten;
484 pjdlog_info("Reloading configuration...");
488 newcfg = yy_config_parse(cfgpath, false);
493 * Check if control address has changed.
495 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
496 if (proto_server(newcfg->hc_controladdr,
497 &newcfg->hc_controlconn) < 0) {
498 pjdlog_errno(LOG_ERR,
499 "Unable to listen on control address %s",
500 newcfg->hc_controladdr);
505 * Check if any listen address has changed.
508 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
509 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
510 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
513 if (clst != NULL && clst->hl_conn != NULL) {
514 pjdlog_info("Keep listening on address %s.",
516 nlst->hl_conn = clst->hl_conn;
518 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
519 pjdlog_info("Listening on new address %s.",
523 pjdlog_errno(LOG_WARNING,
524 "Unable to listen on address %s", nlst->hl_addr);
528 pjdlog_error("No addresses to listen on.");
532 * Check if pidfile's path has changed.
534 if (strcmp(cfg->hc_pidfile, newcfg->hc_pidfile) != 0) {
535 newpfh = pidfile_open(newcfg->hc_pidfile, 0600, &otherpid);
536 if (newpfh == NULL) {
537 if (errno == EEXIST) {
538 pjdlog_errno(LOG_WARNING,
539 "Another hastd is already running, pidfile: %s, pid: %jd.",
540 newcfg->hc_pidfile, (intmax_t)otherpid);
542 pjdlog_errno(LOG_WARNING,
543 "Unable to open or create pidfile %s",
546 } else if (pidfile_write(newpfh) < 0) {
547 /* Write PID to a file. */
548 pjdlog_errno(LOG_WARNING,
549 "Unable to write PID to file %s",
552 pjdlog_debug(1, "PID stored in %s.",
557 /* No failures from now on. */
560 * Switch to new control socket.
562 if (newcfg->hc_controlconn != NULL) {
563 pjdlog_info("Control socket changed from %s to %s.",
564 cfg->hc_controladdr, newcfg->hc_controladdr);
565 proto_close(cfg->hc_controlconn);
566 cfg->hc_controlconn = newcfg->hc_controlconn;
567 newcfg->hc_controlconn = NULL;
568 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
569 sizeof(cfg->hc_controladdr));
572 * Switch to new pidfile.
574 (void)pidfile_remove(pfh);
576 (void)strlcpy(cfg->hc_pidfile, newcfg->hc_pidfile,
577 sizeof(cfg->hc_pidfile));
579 * Switch to new listen addresses. Close all that were removed.
581 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
582 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
583 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
586 if (nlst == NULL && clst->hl_conn != NULL) {
587 proto_close(clst->hl_conn);
588 pjdlog_info("No longer listening on address %s.",
591 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
594 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
597 * Stop and remove resources that were removed from the configuration.
599 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
600 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
601 if (strcmp(cres->hr_name, nres->hr_name) == 0)
605 control_set_role(cres, HAST_ROLE_INIT);
606 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
607 pjdlog_info("Resource %s removed.", cres->hr_name);
612 * Move new resources to the current configuration.
614 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
615 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
616 if (strcmp(cres->hr_name, nres->hr_name) == 0)
620 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
621 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
622 pjdlog_info("Resource %s added.", nres->hr_name);
626 * Deal with modified resources.
627 * Depending on what has changed exactly we might want to perform
630 * We do full resource restart in the following situations:
631 * Resource role is INIT or SECONDARY.
632 * Resource role is PRIMARY and path to local component or provider
634 * In case of PRIMARY, the worker process will be killed and restarted,
635 * which also means removing /dev/hast/<name> provider and
638 * We do just reload (send SIGHUP to worker process) if we act as
639 * PRIMARY, but only if remote address, source address, replication
640 * mode, timeout, execution path or metaflush has changed.
641 * For those, there is no need to restart worker process.
642 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
643 * source address has changed or it will set new timeout if only timeout
644 * has changed or it will update metaflush if only metaflush has
647 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
648 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
649 if (strcmp(cres->hr_name, nres->hr_name) == 0)
652 PJDLOG_ASSERT(cres != NULL);
653 if (resource_needs_restart(cres, nres)) {
654 pjdlog_info("Resource %s configuration was modified, restarting it.",
656 role = cres->hr_role;
657 control_set_role(cres, HAST_ROLE_INIT);
658 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
660 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
661 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
662 control_set_role(nres, role);
663 } else if (resource_needs_reload(cres, nres)) {
664 pjdlog_info("Resource %s configuration was modified, reloading it.",
666 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
667 sizeof(cres->hr_remoteaddr));
668 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr,
669 sizeof(cres->hr_sourceaddr));
670 cres->hr_replication = nres->hr_replication;
671 cres->hr_checksum = nres->hr_checksum;
672 cres->hr_compression = nres->hr_compression;
673 cres->hr_timeout = nres->hr_timeout;
674 strlcpy(cres->hr_exec, nres->hr_exec,
675 sizeof(cres->hr_exec));
676 cres->hr_metaflush = nres->hr_metaflush;
677 if (cres->hr_workerpid != 0)
678 resource_reload(cres);
682 yy_config_free(newcfg);
683 pjdlog_info("Configuration reloaded successfully.");
686 if (newcfg != NULL) {
687 if (newcfg->hc_controlconn != NULL)
688 proto_close(newcfg->hc_controlconn);
689 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) {
690 if (nlst->hl_conn != NULL) {
691 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
692 if (strcmp(nlst->hl_addr,
693 clst->hl_addr) == 0) {
697 if (clst == NULL || clst->hl_conn == NULL)
698 proto_close(nlst->hl_conn);
700 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
703 yy_config_free(newcfg);
706 (void)pidfile_remove(newpfh);
707 pjdlog_warning("Configuration not reloaded.");
711 terminate_workers(void)
713 struct hast_resource *res;
715 pjdlog_info("Termination signal received, exiting.");
716 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
717 if (res->hr_workerpid == 0)
719 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
720 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
721 if (kill(res->hr_workerpid, SIGTERM) == 0)
723 pjdlog_errno(LOG_WARNING,
724 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
725 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
730 listen_accept(struct hastd_listen *lst)
732 struct hast_resource *res;
733 struct proto_conn *conn;
734 struct nv *nvin, *nvout, *nverr;
736 const unsigned char *token;
737 char laddr[256], raddr[256];
742 proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
743 pjdlog_debug(1, "Accepting connection to %s.", laddr);
745 if (proto_accept(lst->hl_conn, &conn) < 0) {
746 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
750 proto_local_address(conn, laddr, sizeof(laddr));
751 proto_remote_address(conn, raddr, sizeof(raddr));
752 pjdlog_info("Connection from %s to %s.", raddr, laddr);
754 /* Error in setting timeout is not critical, but why should it fail? */
755 if (proto_timeout(conn, HAST_TIMEOUT) < 0)
756 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
758 nvin = nvout = nverr = NULL;
761 * Before receiving any data see if remote host have access to any
764 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
765 if (proto_address_match(conn, res->hr_remoteaddr))
769 pjdlog_error("Client %s isn't known.", raddr);
772 /* Ok, remote host can access at least one resource. */
774 if (hast_proto_recv_hdr(conn, &nvin) < 0) {
775 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
780 resname = nv_get_string(nvin, "resource");
781 if (resname == NULL) {
782 pjdlog_error("No 'resource' field in the header received from %s.",
786 pjdlog_debug(2, "%s: resource=%s", raddr, resname);
787 token = nv_get_uint8_array(nvin, &size, "token");
789 * NULL token means that this is first conection.
791 if (token != NULL && size != sizeof(res->hr_token)) {
792 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
793 raddr, sizeof(res->hr_token), size);
798 * From now on we want to send errors to the remote node.
802 /* Find resource related to this connection. */
803 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
804 if (strcmp(resname, res->hr_name) == 0)
807 /* Have we found the resource? */
809 pjdlog_error("No resource '%s' as requested by %s.",
811 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
815 /* Now that we know resource name setup log prefix. */
816 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
818 /* Does the remote host have access to this resource? */
819 if (!proto_address_match(conn, res->hr_remoteaddr)) {
820 pjdlog_error("Client %s has no access to the resource.", raddr);
821 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
824 /* Is the resource marked as secondary? */
825 if (res->hr_role != HAST_ROLE_SECONDARY) {
826 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.",
827 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
829 nv_add_stringf(nverr, "errmsg",
830 "Remote node acts as %s for the resource and not as %s.",
831 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
832 if (res->hr_role == HAST_ROLE_PRIMARY) {
834 * If we act as primary request the other side to wait
835 * for us a bit, as we might be finishing cleanups.
837 nv_add_uint8(nverr, 1, "wait");
841 /* Does token (if exists) match? */
842 if (token != NULL && memcmp(token, res->hr_token,
843 sizeof(res->hr_token)) != 0) {
844 pjdlog_error("Token received from %s doesn't match.", raddr);
845 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
849 * If there is no token, but we have half-open connection
850 * (only remotein) or full connection (worker process is running)
851 * we have to cancel those and accept the new connection.
854 PJDLOG_ASSERT(res->hr_remoteout == NULL);
855 pjdlog_debug(1, "Initial connection from %s.", raddr);
856 if (res->hr_workerpid != 0) {
857 PJDLOG_ASSERT(res->hr_remotein == NULL);
859 "Worker process exists (pid=%u), stopping it.",
860 (unsigned int)res->hr_workerpid);
861 /* Stop child process. */
862 if (kill(res->hr_workerpid, SIGINT) < 0) {
863 pjdlog_errno(LOG_ERR,
864 "Unable to stop worker process (pid=%u)",
865 (unsigned int)res->hr_workerpid);
867 * Other than logging the problem we
868 * ignore it - nothing smart to do.
871 /* Wait for it to exit. */
872 else if ((pid = waitpid(res->hr_workerpid,
873 &status, 0)) != res->hr_workerpid) {
874 /* We can only log the problem. */
875 pjdlog_errno(LOG_ERR,
876 "Waiting for worker process (pid=%u) failed",
877 (unsigned int)res->hr_workerpid);
879 child_exit_log(res->hr_workerpid, status);
882 } else if (res->hr_remotein != NULL) {
885 proto_remote_address(res->hr_remotein, oaddr,
888 "Canceling half-open connection from %s on connection from %s.",
890 proto_close(res->hr_remotein);
891 res->hr_remotein = NULL;
896 * Checks and cleanups are done.
900 arc4random_buf(res->hr_token, sizeof(res->hr_token));
902 nv_add_uint8_array(nvout, res->hr_token,
903 sizeof(res->hr_token), "token");
904 if (nv_error(nvout) != 0) {
905 pjdlog_common(LOG_ERR, 0, nv_error(nvout),
906 "Unable to prepare return header for %s", raddr);
907 nv_add_stringf(nverr, "errmsg",
908 "Remote node was unable to prepare return header: %s.",
909 strerror(nv_error(nvout)));
912 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
915 pjdlog_errno(LOG_ERR, "Unable to send response to %s",
917 nv_add_stringf(nverr, "errmsg",
918 "Remote node was unable to send response: %s.",
922 res->hr_remotein = conn;
923 pjdlog_debug(1, "Incoming connection from %s configured.",
926 res->hr_remoteout = conn;
927 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
928 hastd_secondary(res, nvin);
933 pjdlog_prefix_set("%s", "");
936 if (nv_error(nverr) != 0) {
937 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
938 "Unable to prepare error header for %s", raddr);
941 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
942 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
953 pjdlog_prefix_set("%s", "");
957 connection_migrate(struct hast_resource *res)
959 struct proto_conn *conn;
962 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
964 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
966 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
967 pjdlog_errno(LOG_WARNING,
968 "Unable to receive connection command");
971 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
972 res->hr_remoteaddr, &conn) < 0) {
974 pjdlog_errno(LOG_WARNING,
975 "Unable to create outgoing connection to %s",
979 if (proto_connect(conn, -1) < 0) {
981 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
988 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
989 pjdlog_errno(LOG_WARNING,
990 "Unable to send reply to connection request");
992 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
993 pjdlog_errno(LOG_WARNING, "Unable to send connection");
995 pjdlog_prefix_set("%s", "");
1001 struct timespec sigtimeout;
1005 sigtimeout.tv_sec = 0;
1006 sigtimeout.tv_nsec = 0;
1008 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1009 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1010 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1011 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1012 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1014 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
1018 sigexit_received = true;
1019 terminate_workers();
1020 proto_close(cfg->hc_controlconn);
1030 PJDLOG_ABORT("Unexpected signal (%d).", signo);
1038 struct hast_resource *res;
1039 struct hastd_listen *lst;
1040 struct timeval seltimeout;
1042 time_t lastcheck, now;
1045 lastcheck = time(NULL);
1046 seltimeout.tv_sec = REPORT_INTERVAL;
1047 seltimeout.tv_usec = 0;
1052 /* Setup descriptors for select(2). */
1054 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1055 PJDLOG_ASSERT(fd >= 0);
1057 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1058 if (lst->hl_conn == NULL)
1060 fd = proto_descriptor(lst->hl_conn);
1061 PJDLOG_ASSERT(fd >= 0);
1063 maxfd = fd > maxfd ? fd : maxfd;
1065 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1066 if (res->hr_event == NULL)
1068 fd = proto_descriptor(res->hr_event);
1069 PJDLOG_ASSERT(fd >= 0);
1071 maxfd = fd > maxfd ? fd : maxfd;
1072 if (res->hr_role == HAST_ROLE_PRIMARY) {
1073 /* Only primary workers asks for connections. */
1074 PJDLOG_ASSERT(res->hr_conn != NULL);
1075 fd = proto_descriptor(res->hr_conn);
1076 PJDLOG_ASSERT(fd >= 0);
1078 maxfd = fd > maxfd ? fd : maxfd;
1080 PJDLOG_ASSERT(res->hr_conn == NULL);
1084 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1085 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1087 if (lastcheck + REPORT_INTERVAL <= now) {
1093 * select(2) timed out, so there should be no
1094 * descriptors to check.
1097 } else if (ret == -1) {
1100 KEEP_ERRNO((void)pidfile_remove(pfh));
1101 pjdlog_exit(EX_OSERR, "select() failed");
1105 * Check for signals before we do anything to update our
1106 * info about terminated workers in the meantime.
1110 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
1111 control_handle(cfg);
1112 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1113 if (lst->hl_conn == NULL)
1115 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1118 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1119 if (res->hr_event == NULL)
1121 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1122 if (event_recv(res) == 0)
1124 /* The worker process exited? */
1125 proto_close(res->hr_event);
1126 res->hr_event = NULL;
1127 if (res->hr_conn != NULL) {
1128 proto_close(res->hr_conn);
1129 res->hr_conn = NULL;
1133 if (res->hr_role == HAST_ROLE_PRIMARY) {
1134 PJDLOG_ASSERT(res->hr_conn != NULL);
1135 if (FD_ISSET(proto_descriptor(res->hr_conn),
1137 connection_migrate(res);
1140 PJDLOG_ASSERT(res->hr_conn == NULL);
1147 dummy_sighandler(int sig __unused)
1149 /* Nothing to do. */
1153 main(int argc, char *argv[])
1155 struct hastd_listen *lst;
1156 const char *pidfile;
1169 ch = getopt(argc, argv, "c:dFhP:");
1193 pjdlog_init(PJDLOG_MODE_STD);
1194 pjdlog_debug_set(debuglevel);
1199 * When path to the configuration file is relative, obtain full path,
1200 * so we can always find the file, even after daemonizing and changing
1201 * working directory to /.
1203 if (cfgpath[0] != '/') {
1204 const char *newcfgpath;
1206 newcfgpath = realpath(cfgpath, NULL);
1207 if (newcfgpath == NULL) {
1208 pjdlog_exit(EX_CONFIG,
1209 "Unable to obtain full path of %s", cfgpath);
1211 cfgpath = newcfgpath;
1214 cfg = yy_config_parse(cfgpath, true);
1215 PJDLOG_ASSERT(cfg != NULL);
1217 if (pidfile != NULL) {
1218 if (strlcpy(cfg->hc_pidfile, pidfile,
1219 sizeof(cfg->hc_pidfile)) >= sizeof(cfg->hc_pidfile)) {
1220 pjdlog_exitx(EX_CONFIG, "Pidfile path is too long.");
1223 pfh = pidfile_open(cfg->hc_pidfile, 0600, &otherpid);
1225 if (errno == EEXIST) {
1226 pjdlog_exitx(EX_TEMPFAIL,
1227 "Another hastd is already running, pidfile: %s, pid: %jd.",
1228 cfg->hc_pidfile, (intmax_t)otherpid);
1230 /* If we cannot create pidfile for other reasons, only warn. */
1231 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile %s",
1236 * Restore default actions for interesting signals in case parent
1237 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1239 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1240 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1241 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1243 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1244 * so we can mask it.
1246 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1248 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1249 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1250 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1251 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1252 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1253 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1255 /* Listen on control address. */
1256 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
1257 KEEP_ERRNO((void)pidfile_remove(pfh));
1258 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1259 cfg->hc_controladdr);
1261 /* Listen for remote connections. */
1262 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1263 if (proto_server(lst->hl_addr, &lst->hl_conn) < 0) {
1264 KEEP_ERRNO((void)pidfile_remove(pfh));
1265 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1271 if (daemon(0, 0) < 0) {
1272 KEEP_ERRNO((void)pidfile_remove(pfh));
1273 pjdlog_exit(EX_OSERR, "Unable to daemonize");
1276 /* Start logging to syslog. */
1277 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1279 /* Write PID to a file. */
1280 if (pidfile_write(pfh) < 0) {
1281 pjdlog_errno(LOG_WARNING,
1282 "Unable to write PID to a file %s",
1285 pjdlog_debug(1, "PID stored in %s.", cfg->hc_pidfile);
1289 pjdlog_info("Started successfully, running protocol version %d.",
1290 HAST_PROTO_VERSION);
1292 pjdlog_debug(1, "Listening on control address %s.",
1293 cfg->hc_controladdr);
1294 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next)
1295 pjdlog_info("Listening on address %s.", lst->hl_addr);