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)
394 resource_needs_reload(const struct hast_resource *res0,
395 const struct hast_resource *res1)
398 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
399 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0);
400 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
402 if (res0->hr_role != HAST_ROLE_PRIMARY)
405 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
407 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
409 if (res0->hr_replication != res1->hr_replication)
411 if (res0->hr_checksum != res1->hr_checksum)
413 if (res0->hr_compression != res1->hr_compression)
415 if (res0->hr_timeout != res1->hr_timeout)
417 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
423 resource_reload(const struct hast_resource *res)
425 struct nv *nvin, *nvout;
428 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
431 nv_add_uint8(nvout, CONTROL_RELOAD, "cmd");
432 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr");
433 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr");
434 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication");
435 nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum");
436 nv_add_int32(nvout, (int32_t)res->hr_compression, "compression");
437 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout");
438 nv_add_string(nvout, res->hr_exec, "exec");
439 if (nv_error(nvout) != 0) {
441 pjdlog_error("Unable to allocate header for reload message.");
444 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
445 pjdlog_errno(LOG_ERR, "Unable to send reload message");
451 /* Receive response. */
452 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
453 pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
456 error = nv_get_int16(nvin, "error");
459 pjdlog_common(LOG_ERR, 0, error, "Reload failed");
467 struct hastd_config *newcfg;
468 struct hast_resource *nres, *cres, *tres;
469 struct hastd_listen *nlst, *clst;
470 unsigned int nlisten;
473 pjdlog_info("Reloading configuration...");
475 newcfg = yy_config_parse(cfgpath, false);
480 * Check if control address has changed.
482 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
483 if (proto_server(newcfg->hc_controladdr,
484 &newcfg->hc_controlconn) < 0) {
485 pjdlog_errno(LOG_ERR,
486 "Unable to listen on control address %s",
487 newcfg->hc_controladdr);
492 * Check if any listen address has changed.
495 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
496 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
497 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
500 if (clst != NULL && clst->hl_conn != NULL) {
501 pjdlog_info("Keep listening on address %s.",
503 nlst->hl_conn = clst->hl_conn;
505 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
506 pjdlog_info("Listening on new address %s.",
510 pjdlog_errno(LOG_WARNING,
511 "Unable to listen on address %s", nlst->hl_addr);
515 pjdlog_error("No addresses to listen on.");
519 /* No failures from now on. */
522 * Switch to new control socket.
524 if (newcfg->hc_controlconn != NULL) {
525 pjdlog_info("Control socket changed from %s to %s.",
526 cfg->hc_controladdr, newcfg->hc_controladdr);
527 proto_close(cfg->hc_controlconn);
528 cfg->hc_controlconn = newcfg->hc_controlconn;
529 newcfg->hc_controlconn = NULL;
530 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
531 sizeof(cfg->hc_controladdr));
534 * Switch to new listen addresses. Close all that were removed.
536 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
537 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
538 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
541 if (nlst == NULL && clst->hl_conn != NULL) {
542 proto_close(clst->hl_conn);
543 pjdlog_info("No longer listening on address %s.",
546 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
549 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
552 * Stop and remove resources that were removed from the configuration.
554 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
555 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
556 if (strcmp(cres->hr_name, nres->hr_name) == 0)
560 control_set_role(cres, HAST_ROLE_INIT);
561 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
562 pjdlog_info("Resource %s removed.", cres->hr_name);
567 * Move new resources to the current configuration.
569 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
570 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
571 if (strcmp(cres->hr_name, nres->hr_name) == 0)
575 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
576 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
577 pjdlog_info("Resource %s added.", nres->hr_name);
581 * Deal with modified resources.
582 * Depending on what has changed exactly we might want to perform
585 * We do full resource restart in the following situations:
586 * Resource role is INIT or SECONDARY.
587 * Resource role is PRIMARY and path to local component or provider
589 * In case of PRIMARY, the worker process will be killed and restarted,
590 * which also means removing /dev/hast/<name> provider and
593 * We do just reload (send SIGHUP to worker process) if we act as
594 * PRIMARY, but only if remote address, replication mode, timeout or
595 * execution path has changed. For those, there is no need to restart
597 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
598 * replication mode has changed or simply set new timeout if only
599 * timeout has changed.
601 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
602 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
603 if (strcmp(cres->hr_name, nres->hr_name) == 0)
606 PJDLOG_ASSERT(cres != NULL);
607 if (resource_needs_restart(cres, nres)) {
608 pjdlog_info("Resource %s configuration was modified, restarting it.",
610 role = cres->hr_role;
611 control_set_role(cres, HAST_ROLE_INIT);
612 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
614 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
615 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
616 control_set_role(nres, role);
617 } else if (resource_needs_reload(cres, nres)) {
618 pjdlog_info("Resource %s configuration was modified, reloading it.",
620 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
621 sizeof(cres->hr_remoteaddr));
622 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr,
623 sizeof(cres->hr_sourceaddr));
624 cres->hr_replication = nres->hr_replication;
625 cres->hr_checksum = nres->hr_checksum;
626 cres->hr_compression = nres->hr_compression;
627 cres->hr_timeout = nres->hr_timeout;
628 strlcpy(cres->hr_exec, nres->hr_exec,
629 sizeof(cres->hr_exec));
630 if (cres->hr_workerpid != 0)
631 resource_reload(cres);
635 yy_config_free(newcfg);
636 pjdlog_info("Configuration reloaded successfully.");
639 if (newcfg != NULL) {
640 if (newcfg->hc_controlconn != NULL)
641 proto_close(newcfg->hc_controlconn);
642 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) {
643 if (nlst->hl_conn != NULL) {
644 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
645 if (strcmp(nlst->hl_addr,
646 clst->hl_addr) == 0) {
650 if (clst == NULL || clst->hl_conn == NULL)
651 proto_close(nlst->hl_conn);
653 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
656 yy_config_free(newcfg);
658 pjdlog_warning("Configuration not reloaded.");
662 terminate_workers(void)
664 struct hast_resource *res;
666 pjdlog_info("Termination signal received, exiting.");
667 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
668 if (res->hr_workerpid == 0)
670 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
671 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
672 if (kill(res->hr_workerpid, SIGTERM) == 0)
674 pjdlog_errno(LOG_WARNING,
675 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
676 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
681 listen_accept(struct hastd_listen *lst)
683 struct hast_resource *res;
684 struct proto_conn *conn;
685 struct nv *nvin, *nvout, *nverr;
687 const unsigned char *token;
688 char laddr[256], raddr[256];
693 proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
694 pjdlog_debug(1, "Accepting connection to %s.", laddr);
696 if (proto_accept(lst->hl_conn, &conn) < 0) {
697 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
701 proto_local_address(conn, laddr, sizeof(laddr));
702 proto_remote_address(conn, raddr, sizeof(raddr));
703 pjdlog_info("Connection from %s to %s.", raddr, laddr);
705 /* Error in setting timeout is not critical, but why should it fail? */
706 if (proto_timeout(conn, HAST_TIMEOUT) < 0)
707 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
709 nvin = nvout = nverr = NULL;
712 * Before receiving any data see if remote host have access to any
715 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
716 if (proto_address_match(conn, res->hr_remoteaddr))
720 pjdlog_error("Client %s isn't known.", raddr);
723 /* Ok, remote host can access at least one resource. */
725 if (hast_proto_recv_hdr(conn, &nvin) < 0) {
726 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
731 resname = nv_get_string(nvin, "resource");
732 if (resname == NULL) {
733 pjdlog_error("No 'resource' field in the header received from %s.",
737 pjdlog_debug(2, "%s: resource=%s", raddr, resname);
738 token = nv_get_uint8_array(nvin, &size, "token");
740 * NULL token means that this is first conection.
742 if (token != NULL && size != sizeof(res->hr_token)) {
743 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
744 raddr, sizeof(res->hr_token), size);
749 * From now on we want to send errors to the remote node.
753 /* Find resource related to this connection. */
754 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
755 if (strcmp(resname, res->hr_name) == 0)
758 /* Have we found the resource? */
760 pjdlog_error("No resource '%s' as requested by %s.",
762 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
766 /* Now that we know resource name setup log prefix. */
767 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
769 /* Does the remote host have access to this resource? */
770 if (!proto_address_match(conn, res->hr_remoteaddr)) {
771 pjdlog_error("Client %s has no access to the resource.", raddr);
772 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
775 /* Is the resource marked as secondary? */
776 if (res->hr_role != HAST_ROLE_SECONDARY) {
777 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.",
778 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
780 nv_add_stringf(nverr, "errmsg",
781 "Remote node acts as %s for the resource and not as %s.",
782 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
783 if (res->hr_role == HAST_ROLE_PRIMARY) {
785 * If we act as primary request the other side to wait
786 * for us a bit, as we might be finishing cleanups.
788 nv_add_uint8(nverr, 1, "wait");
792 /* Does token (if exists) match? */
793 if (token != NULL && memcmp(token, res->hr_token,
794 sizeof(res->hr_token)) != 0) {
795 pjdlog_error("Token received from %s doesn't match.", raddr);
796 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
800 * If there is no token, but we have half-open connection
801 * (only remotein) or full connection (worker process is running)
802 * we have to cancel those and accept the new connection.
805 PJDLOG_ASSERT(res->hr_remoteout == NULL);
806 pjdlog_debug(1, "Initial connection from %s.", raddr);
807 if (res->hr_workerpid != 0) {
808 PJDLOG_ASSERT(res->hr_remotein == NULL);
810 "Worker process exists (pid=%u), stopping it.",
811 (unsigned int)res->hr_workerpid);
812 /* Stop child process. */
813 if (kill(res->hr_workerpid, SIGINT) < 0) {
814 pjdlog_errno(LOG_ERR,
815 "Unable to stop worker process (pid=%u)",
816 (unsigned int)res->hr_workerpid);
818 * Other than logging the problem we
819 * ignore it - nothing smart to do.
822 /* Wait for it to exit. */
823 else if ((pid = waitpid(res->hr_workerpid,
824 &status, 0)) != res->hr_workerpid) {
825 /* We can only log the problem. */
826 pjdlog_errno(LOG_ERR,
827 "Waiting for worker process (pid=%u) failed",
828 (unsigned int)res->hr_workerpid);
830 child_exit_log(res->hr_workerpid, status);
833 } else if (res->hr_remotein != NULL) {
836 proto_remote_address(res->hr_remotein, oaddr,
839 "Canceling half-open connection from %s on connection from %s.",
841 proto_close(res->hr_remotein);
842 res->hr_remotein = NULL;
847 * Checks and cleanups are done.
851 arc4random_buf(res->hr_token, sizeof(res->hr_token));
853 nv_add_uint8_array(nvout, res->hr_token,
854 sizeof(res->hr_token), "token");
855 if (nv_error(nvout) != 0) {
856 pjdlog_common(LOG_ERR, 0, nv_error(nvout),
857 "Unable to prepare return header for %s", raddr);
858 nv_add_stringf(nverr, "errmsg",
859 "Remote node was unable to prepare return header: %s.",
860 strerror(nv_error(nvout)));
863 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
866 pjdlog_errno(LOG_ERR, "Unable to send response to %s",
868 nv_add_stringf(nverr, "errmsg",
869 "Remote node was unable to send response: %s.",
873 res->hr_remotein = conn;
874 pjdlog_debug(1, "Incoming connection from %s configured.",
877 res->hr_remoteout = conn;
878 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
879 hastd_secondary(res, nvin);
884 pjdlog_prefix_set("%s", "");
887 if (nv_error(nverr) != 0) {
888 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
889 "Unable to prepare error header for %s", raddr);
892 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
893 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
904 pjdlog_prefix_set("%s", "");
908 connection_migrate(struct hast_resource *res)
910 struct proto_conn *conn;
913 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
915 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
917 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
918 pjdlog_errno(LOG_WARNING,
919 "Unable to receive connection command");
922 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
923 res->hr_remoteaddr, &conn) < 0) {
925 pjdlog_errno(LOG_WARNING,
926 "Unable to create outgoing connection to %s",
930 if (proto_connect(conn, -1) < 0) {
932 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
939 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
940 pjdlog_errno(LOG_WARNING,
941 "Unable to send reply to connection request");
943 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
944 pjdlog_errno(LOG_WARNING, "Unable to send connection");
946 pjdlog_prefix_set("%s", "");
952 struct timespec sigtimeout;
956 sigtimeout.tv_sec = 0;
957 sigtimeout.tv_nsec = 0;
959 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
960 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
961 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
962 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
963 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
965 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
969 sigexit_received = true;
971 proto_close(cfg->hc_controlconn);
981 PJDLOG_ABORT("Unexpected signal (%d).", signo);
989 struct hast_resource *res;
990 struct hastd_listen *lst;
991 struct timeval seltimeout;
993 time_t lastcheck, now;
996 lastcheck = time(NULL);
997 seltimeout.tv_sec = REPORT_INTERVAL;
998 seltimeout.tv_usec = 0;
1003 /* Setup descriptors for select(2). */
1005 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1006 PJDLOG_ASSERT(fd >= 0);
1008 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1009 if (lst->hl_conn == NULL)
1011 fd = proto_descriptor(lst->hl_conn);
1012 PJDLOG_ASSERT(fd >= 0);
1014 maxfd = fd > maxfd ? fd : maxfd;
1016 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1017 if (res->hr_event == NULL)
1019 fd = proto_descriptor(res->hr_event);
1020 PJDLOG_ASSERT(fd >= 0);
1022 maxfd = fd > maxfd ? fd : maxfd;
1023 if (res->hr_role == HAST_ROLE_PRIMARY) {
1024 /* Only primary workers asks for connections. */
1025 PJDLOG_ASSERT(res->hr_conn != NULL);
1026 fd = proto_descriptor(res->hr_conn);
1027 PJDLOG_ASSERT(fd >= 0);
1029 maxfd = fd > maxfd ? fd : maxfd;
1031 PJDLOG_ASSERT(res->hr_conn == NULL);
1035 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1036 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1038 if (lastcheck + REPORT_INTERVAL <= now) {
1044 * select(2) timed out, so there should be no
1045 * descriptors to check.
1048 } else if (ret == -1) {
1051 KEEP_ERRNO((void)pidfile_remove(pfh));
1052 pjdlog_exit(EX_OSERR, "select() failed");
1056 * Check for signals before we do anything to update our
1057 * info about terminated workers in the meantime.
1061 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
1062 control_handle(cfg);
1063 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1064 if (lst->hl_conn == NULL)
1066 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1069 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1070 if (res->hr_event == NULL)
1072 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1073 if (event_recv(res) == 0)
1075 /* The worker process exited? */
1076 proto_close(res->hr_event);
1077 res->hr_event = NULL;
1078 if (res->hr_conn != NULL) {
1079 proto_close(res->hr_conn);
1080 res->hr_conn = NULL;
1084 if (res->hr_role == HAST_ROLE_PRIMARY) {
1085 PJDLOG_ASSERT(res->hr_conn != NULL);
1086 if (FD_ISSET(proto_descriptor(res->hr_conn),
1088 connection_migrate(res);
1091 PJDLOG_ASSERT(res->hr_conn == NULL);
1098 dummy_sighandler(int sig __unused)
1100 /* Nothing to do. */
1104 main(int argc, char *argv[])
1106 struct hastd_listen *lst;
1107 const char *pidfile;
1115 pidfile = HASTD_PIDFILE;
1120 ch = getopt(argc, argv, "c:dFhP:");
1144 pjdlog_init(PJDLOG_MODE_STD);
1145 pjdlog_debug_set(debuglevel);
1149 pfh = pidfile_open(pidfile, 0600, &otherpid);
1151 if (errno == EEXIST) {
1152 pjdlog_exitx(EX_TEMPFAIL,
1153 "Another hastd is already running, pid: %jd.",
1154 (intmax_t)otherpid);
1156 /* If we cannot create pidfile from other reasons, only warn. */
1157 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
1160 cfg = yy_config_parse(cfgpath, true);
1161 PJDLOG_ASSERT(cfg != NULL);
1164 * Restore default actions for interesting signals in case parent
1165 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1167 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1168 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1169 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1171 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1172 * so we can mask it.
1174 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1176 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1177 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1178 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1179 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1180 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1181 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1183 /* Listen on control address. */
1184 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
1185 KEEP_ERRNO((void)pidfile_remove(pfh));
1186 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1187 cfg->hc_controladdr);
1189 /* Listen for remote connections. */
1190 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1191 if (proto_server(lst->hl_addr, &lst->hl_conn) < 0) {
1192 KEEP_ERRNO((void)pidfile_remove(pfh));
1193 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1199 if (daemon(0, 0) < 0) {
1200 KEEP_ERRNO((void)pidfile_remove(pfh));
1201 pjdlog_exit(EX_OSERR, "Unable to daemonize");
1204 /* Start logging to syslog. */
1205 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1207 /* Write PID to a file. */
1208 if (pidfile_write(pfh) < 0) {
1209 pjdlog_errno(LOG_WARNING,
1210 "Unable to write PID to a file");
1214 pjdlog_info("Started successfully, running protocol version %d.",
1215 HAST_PROTO_VERSION);
1217 pjdlog_debug(1, "Listening on control address %s.",
1218 cfg->hc_controladdr);
1219 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next)
1220 pjdlog_info("Listening on address %s.", lst->hl_addr);