1/*-
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
4 * All rights reserved.
5 *
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 225830 2011-09-28 13:08:51Z pjd $");
33
34#include <sys/param.h>
35#include <sys/linker.h>
36#include <sys/module.h>
37#include <sys/stat.h>
38#include <sys/wait.h>
39
40#include <err.h>
41#include <errno.h>
42#include <libutil.h>
43#include <signal.h>
44#include <stdbool.h>
45#include <stdio.h>
46#include <stdlib.h>
47#include <string.h>
48#include <sysexits.h>
49#include <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. */
64const char *cfgpath = HAST_CONFIG;
65/* Hastd configuration. */
66static struct hastd_config *cfg;
67/* Was SIGINT or SIGTERM signal received? */
68bool sigexit_received = false;
69/* PID file handle. */
70struct pidfh *pfh;
71
72/* How often check for hooks running for too long. */
73#define REPORT_INTERVAL 5
74
75static void
76usage(void)
77{
78
79 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
80}
81
82static void
83g_gate_load(void)
84{
85
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) {
90 pjdlog_exit(EX_OSERR,
91 "Unable to load geom_gate module");
92 }
93 }
94 }
95}
96
97void
98descriptors_cleanup(struct hast_resource *res)
99{
100 struct hast_resource *tres;
101 struct hastd_listen *lst;
102
103 TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) {
104 if (tres == res) {
105 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY ||
106 (res->hr_remotein == NULL &&
107 res->hr_remoteout == NULL));
108 continue;
109 }
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);
120 }
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);
127 }
128 (void)pidfile_close(pfh);
129 hook_fini();
130 pjdlog_fini();
131}
132
133static const char *
134dtype2str(mode_t mode)
135{
136
137 if (S_ISBLK(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))
150 return ("socket");
151 else if (S_ISWHT(mode))
152 return ("whiteout");
153 else
154 return ("unknown");
155}
156
157void
158descriptors_assert(const struct hast_resource *res, int pjdlogmode)
159{
160 char msg[256];
161 struct stat sb;
162 long maxfd;
163 bool isopen;
164 mode_t mode;
165 int fd;
166
167 /*
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.
171 */
172 msg[0] = '\0';
173
174 maxfd = sysconf(_SC_OPEN_MAX);
175 if (maxfd < 0) {
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");
180 pjdlog_fini();
181 maxfd = 16384;
182 }
183 for (fd = 0; fd <= maxfd; fd++) {
184 if (fstat(fd, &sb) == 0) {
185 isopen = true;
186 mode = sb.st_mode;
187 } else if (errno == EBADF) {
188 isopen = false;
189 mode = 0;
190 } else {
191 (void)snprintf(msg, sizeof(msg),
192 "Unable to fstat descriptor %d: %s", fd,
193 strerror(errno));
194 break;
195 }
196 if (fd == STDIN_FILENO || fd == STDOUT_FILENO ||
197 fd == STDERR_FILENO) {
198 if (!isopen) {
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")));
203 break;
204 }
205 } else if (fd == proto_descriptor(res->hr_event)) {
206 if (!isopen) {
207 (void)snprintf(msg, sizeof(msg),
208 "Descriptor %d (event) is closed, but should be open.",
209 fd);
210 break;
211 }
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));
216 break;
217 }
218 } else if (fd == proto_descriptor(res->hr_ctrl)) {
219 if (!isopen) {
220 (void)snprintf(msg, sizeof(msg),
221 "Descriptor %d (ctrl) is closed, but should be open.",
222 fd);
223 break;
224 }
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));
229 break;
230 }
231 } else if (res->hr_role == HAST_ROLE_PRIMARY &&
232 fd == proto_descriptor(res->hr_conn)) {
233 if (!isopen) {
234 (void)snprintf(msg, sizeof(msg),
235 "Descriptor %d (conn) is closed, but should be open.",
236 fd);
237 break;
238 }
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));
243 break;
244 }
245 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
246 res->hr_conn != NULL &&
247 fd == proto_descriptor(res->hr_conn)) {
248 if (isopen) {
249 (void)snprintf(msg, sizeof(msg),
250 "Descriptor %d (conn) is open, but should be closed.",
251 fd);
252 break;
253 }
254 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
255 fd == proto_descriptor(res->hr_remotein)) {
256 if (!isopen) {
257 (void)snprintf(msg, sizeof(msg),
258 "Descriptor %d (remote in) is closed, but should be open.",
259 fd);
260 break;
261 }
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));
266 break;
267 }
268 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
269 fd == proto_descriptor(res->hr_remoteout)) {
270 if (!isopen) {
271 (void)snprintf(msg, sizeof(msg),
272 "Descriptor %d (remote out) is closed, but should be open.",
273 fd);
274 break;
275 }
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));
280 break;
281 }
282 } else {
283 if (isopen) {
284 (void)snprintf(msg, sizeof(msg),
285 "Descriptor %d is open (%s), but should be closed.",
286 fd, dtype2str(mode));
287 break;
288 }
289 }
290 }
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);
296 }
297}
298
299static void
300child_exit_log(unsigned int pid, int status)
301{
302
303 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
304 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
305 pid);
306 } else if (WIFSIGNALED(status)) {
307 pjdlog_error("Worker process killed (pid=%u, signal=%d).",
308 pid, WTERMSIG(status));
309 } else {
310 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
311 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
312 }
313}
314
315static void
316child_exit(void)
317{
318 struct hast_resource *res;
319 int status;
320 pid_t pid;
321
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)
326 break;
327 }
328 if (res == NULL) {
329 /*
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.
333 */
334 hook_check_one(pid, status);
335 continue;
336 }
337 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
338 role2str(res->hr_role));
339 child_exit_log(pid, status);
340 child_cleanup(res);
341 if (res->hr_role == HAST_ROLE_PRIMARY) {
342 /*
343 * Restart child process if it was killed by signal
344 * or exited because of temporary problem.
345 */
346 if (WIFSIGNALED(status) ||
347 (WIFEXITED(status) &&
348 WEXITSTATUS(status) == EX_TEMPFAIL)) {
349 sleep(1);
350 pjdlog_info("Restarting worker process.");
351 hastd_primary(res);
352 } else {
353 res->hr_role = HAST_ROLE_INIT;
354 pjdlog_info("Changing resource role back to %s.",
355 role2str(res->hr_role));
356 }
357 }
358 pjdlog_prefix_set("%s", "");
359 }
360}
361
362static bool
363resource_needs_restart(const struct hast_resource *res0,
364 const struct hast_resource *res1)
365{
366
367 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
368
369 if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
370 return (true);
371 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
372 return (true);
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)
376 return (true);
377 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
378 return (true);
379 if (res0->hr_replication != res1->hr_replication)
380 return (true);
381 if (res0->hr_checksum != res1->hr_checksum)
382 return (true);
383 if (res0->hr_compression != res1->hr_compression)
384 return (true);
385 if (res0->hr_timeout != res1->hr_timeout)
386 return (true);
387 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
388 return (true);
389 /*
390 * When metaflush has changed we don't really need restart,
391 * but it is just easier this way.
392 */
393 if (res0->hr_metaflush != res1->hr_metaflush)
394 return (true);
395 }
396 return (false);
397}
398
399static bool
400resource_needs_reload(const struct hast_resource *res0,
401 const struct hast_resource *res1)
402{
403
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);
407
408 if (res0->hr_role != HAST_ROLE_PRIMARY)
409 return (false);
410
411 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
412 return (true);
413 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
414 return (true);
415 if (res0->hr_replication != res1->hr_replication)
416 return (true);
417 if (res0->hr_checksum != res1->hr_checksum)
418 return (true);
419 if (res0->hr_compression != res1->hr_compression)
420 return (true);
421 if (res0->hr_timeout != res1->hr_timeout)
422 return (true);
423 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
424 return (true);
425 if (res0->hr_metaflush != res1->hr_metaflush)
426 return (true);
427 return (false);
428}
429
430static void
431resource_reload(const struct hast_resource *res)
432{
433 struct nv *nvin, *nvout;
434 int error;
435
436 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
437
438 nvout = nv_alloc();
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) {
449 nv_free(nvout);
450 pjdlog_error("Unable to allocate header for reload message.");
451 return;
452 }
453 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
454 pjdlog_errno(LOG_ERR, "Unable to send reload message");
455 nv_free(nvout);
456 return;
457 }
458 nv_free(nvout);
459
460 /* Receive response. */
461 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
462 pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
463 return;
464 }
465 error = nv_get_int16(nvin, "error");
466 nv_free(nvin);
467 if (error != 0) {
468 pjdlog_common(LOG_ERR, 0, error, "Reload failed");
469 return;
470 }
471}
472
473static void
474hastd_reload(void)
475{
476 struct hastd_config *newcfg;
477 struct hast_resource *nres, *cres, *tres;
478 struct hastd_listen *nlst, *clst;
479 unsigned int nlisten;
480 uint8_t role;
481
482 pjdlog_info("Reloading configuration...");
483
484 newcfg = yy_config_parse(cfgpath, false);
485 if (newcfg == NULL)
486 goto failed;
487
488 /*
489 * Check if control address has changed.
490 */
491 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
492 if (proto_server(newcfg->hc_controladdr,
493 &newcfg->hc_controlconn) < 0) {
494 pjdlog_errno(LOG_ERR,
495 "Unable to listen on control address %s",
496 newcfg->hc_controladdr);
497 goto failed;
498 }
499 }
500 /*
501 * Check if any listen address has changed.
502 */
503 nlisten = 0;
504 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
505 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
506 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
507 break;
508 }
509 if (clst != NULL && clst->hl_conn != NULL) {
510 pjdlog_info("Keep listening on address %s.",
511 nlst->hl_addr);
512 nlst->hl_conn = clst->hl_conn;
513 nlisten++;
514 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
515 pjdlog_info("Listening on new address %s.",
516 nlst->hl_addr);
517 nlisten++;
518 } else {
519 pjdlog_errno(LOG_WARNING,
520 "Unable to listen on address %s", nlst->hl_addr);
521 }
522 }
523 if (nlisten == 0) {
524 pjdlog_error("No addresses to listen on.");
525 goto failed;
526 }
527
528 /* No failures from now on. */
529
530 /*
531 * Switch to new control socket.
532 */
533 if (newcfg->hc_controlconn != NULL) {
534 pjdlog_info("Control socket changed from %s to %s.",
535 cfg->hc_controladdr, newcfg->hc_controladdr);
536 proto_close(cfg->hc_controlconn);
537 cfg->hc_controlconn = newcfg->hc_controlconn;
538 newcfg->hc_controlconn = NULL;
539 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
540 sizeof(cfg->hc_controladdr));
541 }
542 /*
543 * Switch to new listen addresses. Close all that were removed.
544 */
545 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
546 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
547 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
548 break;
549 }
550 if (nlst == NULL && clst->hl_conn != NULL) {
551 proto_close(clst->hl_conn);
552 pjdlog_info("No longer listening on address %s.",
553 clst->hl_addr);
554 }
555 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
556 free(clst);
557 }
558 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
559
560 /*
561 * Stop and remove resources that were removed from the configuration.
562 */
563 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
564 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
565 if (strcmp(cres->hr_name, nres->hr_name) == 0)
566 break;
567 }
568 if (nres == NULL) {
569 control_set_role(cres, HAST_ROLE_INIT);
570 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
571 pjdlog_info("Resource %s removed.", cres->hr_name);
572 free(cres);
573 }
574 }
575 /*
576 * Move new resources to the current configuration.
577 */
578 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
579 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
580 if (strcmp(cres->hr_name, nres->hr_name) == 0)
581 break;
582 }
583 if (cres == NULL) {
584 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
585 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
586 pjdlog_info("Resource %s added.", nres->hr_name);
587 }
588 }
589 /*
590 * Deal with modified resources.
591 * Depending on what has changed exactly we might want to perform
592 * different actions.
593 *
594 * We do full resource restart in the following situations:
595 * Resource role is INIT or SECONDARY.
596 * Resource role is PRIMARY and path to local component or provider
597 * name has changed.
598 * In case of PRIMARY, the worker process will be killed and restarted,
599 * which also means removing /dev/hast/<name> provider and
600 * recreating it.
601 *
602 * We do just reload (send SIGHUP to worker process) if we act as
603 * PRIMARY, but only if remote address, source address, replication
604 * mode, timeout, execution path or metaflush has changed.
605 * For those, there is no need to restart worker process.
606 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
607 * source address has changed or it will set new timeout if only timeout
608 * has changed or it will update metaflush if only metaflush has
609 * changed.
610 */
611 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
612 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
613 if (strcmp(cres->hr_name, nres->hr_name) == 0)
614 break;
615 }
616 PJDLOG_ASSERT(cres != NULL);
617 if (resource_needs_restart(cres, nres)) {
618 pjdlog_info("Resource %s configuration was modified, restarting it.",
619 cres->hr_name);
620 role = cres->hr_role;
621 control_set_role(cres, HAST_ROLE_INIT);
622 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
623 free(cres);
624 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
625 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
626 control_set_role(nres, role);
627 } else if (resource_needs_reload(cres, nres)) {
628 pjdlog_info("Resource %s configuration was modified, reloading it.",
629 cres->hr_name);
630 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
631 sizeof(cres->hr_remoteaddr));
632 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr,
633 sizeof(cres->hr_sourceaddr));
634 cres->hr_replication = nres->hr_replication;
635 cres->hr_checksum = nres->hr_checksum;
636 cres->hr_compression = nres->hr_compression;
637 cres->hr_timeout = nres->hr_timeout;
638 strlcpy(cres->hr_exec, nres->hr_exec,
639 sizeof(cres->hr_exec));
640 cres->hr_metaflush = nres->hr_metaflush;
641 if (cres->hr_workerpid != 0)
642 resource_reload(cres);
643 }
644 }
645
646 yy_config_free(newcfg);
647 pjdlog_info("Configuration reloaded successfully.");
648 return;
649failed:
650 if (newcfg != NULL) {
651 if (newcfg->hc_controlconn != NULL)
652 proto_close(newcfg->hc_controlconn);
653 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) {
654 if (nlst->hl_conn != NULL) {
655 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
656 if (strcmp(nlst->hl_addr,
657 clst->hl_addr) == 0) {
658 break;
659 }
660 }
661 if (clst == NULL || clst->hl_conn == NULL)
662 proto_close(nlst->hl_conn);
663 }
664 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
665 free(nlst);
666 }
667 yy_config_free(newcfg);
668 }
669 pjdlog_warning("Configuration not reloaded.");
670}
671
672static void
673terminate_workers(void)
674{
675 struct hast_resource *res;
676
677 pjdlog_info("Termination signal received, exiting.");
678 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
679 if (res->hr_workerpid == 0)
680 continue;
681 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
682 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
683 if (kill(res->hr_workerpid, SIGTERM) == 0)
684 continue;
685 pjdlog_errno(LOG_WARNING,
686 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
687 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
688 }
689}
690
691static void
692listen_accept(struct hastd_listen *lst)
693{
694 struct hast_resource *res;
695 struct proto_conn *conn;
696 struct nv *nvin, *nvout, *nverr;
697 const char *resname;
698 const unsigned char *token;
699 char laddr[256], raddr[256];
700 size_t size;
701 pid_t pid;
702 int status;
703
704 proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
705 pjdlog_debug(1, "Accepting connection to %s.", laddr);
706
707 if (proto_accept(lst->hl_conn, &conn) < 0) {
708 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
709 return;
710 }
711
712 proto_local_address(conn, laddr, sizeof(laddr));
713 proto_remote_address(conn, raddr, sizeof(raddr));
714 pjdlog_info("Connection from %s to %s.", raddr, laddr);
715
716 /* Error in setting timeout is not critical, but why should it fail? */
717 if (proto_timeout(conn, HAST_TIMEOUT) < 0)
718 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
719
720 nvin = nvout = nverr = NULL;
721
722 /*
723 * Before receiving any data see if remote host have access to any
724 * resource.
725 */
726 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
727 if (proto_address_match(conn, res->hr_remoteaddr))
728 break;
729 }
730 if (res == NULL) {
731 pjdlog_error("Client %s isn't known.", raddr);
732 goto close;
733 }
734 /* Ok, remote host can access at least one resource. */
735
736 if (hast_proto_recv_hdr(conn, &nvin) < 0) {
737 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
738 raddr);
739 goto close;
740 }
741
742 resname = nv_get_string(nvin, "resource");
743 if (resname == NULL) {
744 pjdlog_error("No 'resource' field in the header received from %s.",
745 raddr);
746 goto close;
747 }
748 pjdlog_debug(2, "%s: resource=%s", raddr, resname);
749 token = nv_get_uint8_array(nvin, &size, "token");
750 /*
751 * NULL token means that this is first conection.
752 */
753 if (token != NULL && size != sizeof(res->hr_token)) {
754 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
755 raddr, sizeof(res->hr_token), size);
756 goto close;
757 }
758
759 /*
760 * From now on we want to send errors to the remote node.
761 */
762 nverr = nv_alloc();
763
764 /* Find resource related to this connection. */
765 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
766 if (strcmp(resname, res->hr_name) == 0)
767 break;
768 }
769 /* Have we found the resource? */
770 if (res == NULL) {
771 pjdlog_error("No resource '%s' as requested by %s.",
772 resname, raddr);
773 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
774 goto fail;
775 }
776
777 /* Now that we know resource name setup log prefix. */
778 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
779
780 /* Does the remote host have access to this resource? */
781 if (!proto_address_match(conn, res->hr_remoteaddr)) {
782 pjdlog_error("Client %s has no access to the resource.", raddr);
783 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
784 goto fail;
785 }
786 /* Is the resource marked as secondary? */
787 if (res->hr_role != HAST_ROLE_SECONDARY) {
788 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.",
789 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
790 raddr);
791 nv_add_stringf(nverr, "errmsg",
792 "Remote node acts as %s for the resource and not as %s.",
793 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
794 if (res->hr_role == HAST_ROLE_PRIMARY) {
795 /*
796 * If we act as primary request the other side to wait
797 * for us a bit, as we might be finishing cleanups.
798 */
799 nv_add_uint8(nverr, 1, "wait");
800 }
801 goto fail;
802 }
803 /* Does token (if exists) match? */
804 if (token != NULL && memcmp(token, res->hr_token,
805 sizeof(res->hr_token)) != 0) {
806 pjdlog_error("Token received from %s doesn't match.", raddr);
807 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
808 goto fail;
809 }
810 /*
811 * If there is no token, but we have half-open connection
812 * (only remotein) or full connection (worker process is running)
813 * we have to cancel those and accept the new connection.
814 */
815 if (token == NULL) {
816 PJDLOG_ASSERT(res->hr_remoteout == NULL);
817 pjdlog_debug(1, "Initial connection from %s.", raddr);
818 if (res->hr_workerpid != 0) {
819 PJDLOG_ASSERT(res->hr_remotein == NULL);
820 pjdlog_debug(1,
821 "Worker process exists (pid=%u), stopping it.",
822 (unsigned int)res->hr_workerpid);
823 /* Stop child process. */
824 if (kill(res->hr_workerpid, SIGINT) < 0) {
825 pjdlog_errno(LOG_ERR,
826 "Unable to stop worker process (pid=%u)",
827 (unsigned int)res->hr_workerpid);
828 /*
829 * Other than logging the problem we
830 * ignore it - nothing smart to do.
831 */
832 }
833 /* Wait for it to exit. */
834 else if ((pid = waitpid(res->hr_workerpid,
835 &status, 0)) != res->hr_workerpid) {
836 /* We can only log the problem. */
837 pjdlog_errno(LOG_ERR,
838 "Waiting for worker process (pid=%u) failed",
839 (unsigned int)res->hr_workerpid);
840 } else {
841 child_exit_log(res->hr_workerpid, status);
842 }
843 child_cleanup(res);
844 } else if (res->hr_remotein != NULL) {
845 char oaddr[256];
846
847 proto_remote_address(res->hr_remotein, oaddr,
848 sizeof(oaddr));
849 pjdlog_debug(1,
850 "Canceling half-open connection from %s on connection from %s.",
851 oaddr, raddr);
852 proto_close(res->hr_remotein);
853 res->hr_remotein = NULL;
854 }
855 }
856
857 /*
858 * Checks and cleanups are done.
859 */
860
861 if (token == NULL) {
862 arc4random_buf(res->hr_token, sizeof(res->hr_token));
863 nvout = nv_alloc();
864 nv_add_uint8_array(nvout, res->hr_token,
865 sizeof(res->hr_token), "token");
866 if (nv_error(nvout) != 0) {
867 pjdlog_common(LOG_ERR, 0, nv_error(nvout),
868 "Unable to prepare return header for %s", raddr);
869 nv_add_stringf(nverr, "errmsg",
870 "Remote node was unable to prepare return header: %s.",
871 strerror(nv_error(nvout)));
872 goto fail;
873 }
874 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
875 int error = errno;
876
877 pjdlog_errno(LOG_ERR, "Unable to send response to %s",
878 raddr);
879 nv_add_stringf(nverr, "errmsg",
880 "Remote node was unable to send response: %s.",
881 strerror(error));
882 goto fail;
883 }
884 res->hr_remotein = conn;
885 pjdlog_debug(1, "Incoming connection from %s configured.",
886 raddr);
887 } else {
888 res->hr_remoteout = conn;
889 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
890 hastd_secondary(res, nvin);
891 }
892 nv_free(nvin);
893 nv_free(nvout);
894 nv_free(nverr);
895 pjdlog_prefix_set("%s", "");
896 return;
897fail:
898 if (nv_error(nverr) != 0) {
899 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
900 "Unable to prepare error header for %s", raddr);
901 goto close;
902 }
903 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
904 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
905 goto close;
906 }
907close:
908 if (nvin != NULL)
909 nv_free(nvin);
910 if (nvout != NULL)
911 nv_free(nvout);
912 if (nverr != NULL)
913 nv_free(nverr);
914 proto_close(conn);
915 pjdlog_prefix_set("%s", "");
916}
917
918static void
919connection_migrate(struct hast_resource *res)
920{
921 struct proto_conn *conn;
922 int16_t val = 0;
923
924 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
925
926 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
927
928 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
929 pjdlog_errno(LOG_WARNING,
930 "Unable to receive connection command");
931 return;
932 }
933 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
934 res->hr_remoteaddr, &conn) < 0) {
935 val = errno;
936 pjdlog_errno(LOG_WARNING,
937 "Unable to create outgoing connection to %s",
938 res->hr_remoteaddr);
939 goto out;
940 }
941 if (proto_connect(conn, -1) < 0) {
942 val = errno;
943 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
944 res->hr_remoteaddr);
945 proto_close(conn);
946 goto out;
947 }
948 val = 0;
949out:
950 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
951 pjdlog_errno(LOG_WARNING,
952 "Unable to send reply to connection request");
953 }
954 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
955 pjdlog_errno(LOG_WARNING, "Unable to send connection");
956
957 pjdlog_prefix_set("%s", "");
958}
959
960static void
961check_signals(void)
962{
963 struct timespec sigtimeout;
964 sigset_t mask;
965 int signo;
966
967 sigtimeout.tv_sec = 0;
968 sigtimeout.tv_nsec = 0;
969
970 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
971 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
972 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
973 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
974 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
975
976 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
977 switch (signo) {
978 case SIGINT:
979 case SIGTERM:
980 sigexit_received = true;
981 terminate_workers();
982 proto_close(cfg->hc_controlconn);
983 exit(EX_OK);
984 break;
985 case SIGCHLD:
986 child_exit();
987 break;
988 case SIGHUP:
989 hastd_reload();
990 break;
991 default:
992 PJDLOG_ABORT("Unexpected signal (%d).", signo);
993 }
994 }
995}
996
997static void
998main_loop(void)
999{
1000 struct hast_resource *res;
1001 struct hastd_listen *lst;
1002 struct timeval seltimeout;
1003 int fd, maxfd, ret;
1004 time_t lastcheck, now;
1005 fd_set rfds;
1006
1007 lastcheck = time(NULL);
1008 seltimeout.tv_sec = REPORT_INTERVAL;
1009 seltimeout.tv_usec = 0;
1010
1011 for (;;) {
1012 check_signals();
1013
1014 /* Setup descriptors for select(2). */
1015 FD_ZERO(&rfds);
1016 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1017 PJDLOG_ASSERT(fd >= 0);
1018 FD_SET(fd, &rfds);
1019 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1020 if (lst->hl_conn == NULL)
1021 continue;
1022 fd = proto_descriptor(lst->hl_conn);
1023 PJDLOG_ASSERT(fd >= 0);
1024 FD_SET(fd, &rfds);
1025 maxfd = fd > maxfd ? fd : maxfd;
1026 }
1027 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1028 if (res->hr_event == NULL)
1029 continue;
1030 fd = proto_descriptor(res->hr_event);
1031 PJDLOG_ASSERT(fd >= 0);
1032 FD_SET(fd, &rfds);
1033 maxfd = fd > maxfd ? fd : maxfd;
1034 if (res->hr_role == HAST_ROLE_PRIMARY) {
1035 /* Only primary workers asks for connections. */
1036 PJDLOG_ASSERT(res->hr_conn != NULL);
1037 fd = proto_descriptor(res->hr_conn);
1038 PJDLOG_ASSERT(fd >= 0);
1039 FD_SET(fd, &rfds);
1040 maxfd = fd > maxfd ? fd : maxfd;
1041 } else {
1042 PJDLOG_ASSERT(res->hr_conn == NULL);
1043 }
1044 }
1045
1046 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1047 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1048 now = time(NULL);
1049 if (lastcheck + REPORT_INTERVAL <= now) {
1050 hook_check();
1051 lastcheck = now;
1052 }
1053 if (ret == 0) {
1054 /*
1055 * select(2) timed out, so there should be no
1056 * descriptors to check.
1057 */
1058 continue;
1059 } else if (ret == -1) {
1060 if (errno == EINTR)
1061 continue;
1062 KEEP_ERRNO((void)pidfile_remove(pfh));
1063 pjdlog_exit(EX_OSERR, "select() failed");
1064 }
1065
1066 /*
1067 * Check for signals before we do anything to update our
1068 * info about terminated workers in the meantime.
1069 */
1070 check_signals();
1071
1072 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
1073 control_handle(cfg);
1074 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1075 if (lst->hl_conn == NULL)
1076 continue;
1077 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1078 listen_accept(lst);
1079 }
1080 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1081 if (res->hr_event == NULL)
1082 continue;
1083 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1084 if (event_recv(res) == 0)
1085 continue;
1086 /* The worker process exited? */
1087 proto_close(res->hr_event);
1088 res->hr_event = NULL;
1089 if (res->hr_conn != NULL) {
1090 proto_close(res->hr_conn);
1091 res->hr_conn = NULL;
1092 }
1093 continue;
1094 }
1095 if (res->hr_role == HAST_ROLE_PRIMARY) {
1096 PJDLOG_ASSERT(res->hr_conn != NULL);
1097 if (FD_ISSET(proto_descriptor(res->hr_conn),
1098 &rfds)) {
1099 connection_migrate(res);
1100 }
1101 } else {
1102 PJDLOG_ASSERT(res->hr_conn == NULL);
1103 }
1104 }
1105 }
1106}
1107
1108static void
1109dummy_sighandler(int sig __unused)
1110{
1111 /* Nothing to do. */
1112}
1113
1114int
1115main(int argc, char *argv[])
1116{
1117 struct hastd_listen *lst;
1118 const char *pidfile;
1119 pid_t otherpid;
1120 bool foreground;
1121 int debuglevel;
1122 sigset_t mask;
1123
1124 foreground = false;
1125 debuglevel = 0;
1126 pidfile = HASTD_PIDFILE;
1127
1128 for (;;) {
1129 int ch;
1130
1131 ch = getopt(argc, argv, "c:dFhP:");
1132 if (ch == -1)
1133 break;
1134 switch (ch) {
1135 case 'c':
1136 cfgpath = optarg;
1137 break;
1138 case 'd':
1139 debuglevel++;
1140 break;
1141 case 'F':
1142 foreground = true;
1143 break;
1144 case 'P':
1145 pidfile = optarg;
1146 break;
1147 case 'h':
1148 default:
1149 usage();
1150 }
1151 }
1152 argc -= optind;
1153 argv += optind;
1154
1155 pjdlog_init(PJDLOG_MODE_STD);
1156 pjdlog_debug_set(debuglevel);
1157
1158 g_gate_load();
1159
1160 pfh = pidfile_open(pidfile, 0600, &otherpid);
1161 if (pfh == NULL) {
1162 if (errno == EEXIST) {
1163 pjdlog_exitx(EX_TEMPFAIL,
1164 "Another hastd is already running, pid: %jd.",
1165 (intmax_t)otherpid);
1166 }
1167 /* If we cannot create pidfile from other reasons, only warn. */
1168 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
1169 }
1170
1171 cfg = yy_config_parse(cfgpath, true);
1172 PJDLOG_ASSERT(cfg != NULL);
1173
1174 /*
1175 * Restore default actions for interesting signals in case parent
1176 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1177 */
1178 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1179 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1180 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1181 /*
1182 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1183 * so we can mask it.
1184 */
1185 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1186
1187 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1188 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1189 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1190 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1191 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1192 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1193
1194 /* Listen on control address. */
1195 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
1196 KEEP_ERRNO((void)pidfile_remove(pfh));
1197 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1198 cfg->hc_controladdr);
1199 }
1200 /* Listen for remote connections. */
1201 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1202 if (proto_server(lst->hl_addr, &lst->hl_conn) < 0) {
1203 KEEP_ERRNO((void)pidfile_remove(pfh));
1204 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1205 lst->hl_addr);
1206 }
1207 }
1208
1209 if (!foreground) {
1210 if (daemon(0, 0) < 0) {
1211 KEEP_ERRNO((void)pidfile_remove(pfh));
1212 pjdlog_exit(EX_OSERR, "Unable to daemonize");
1213 }
1214
1215 /* Start logging to syslog. */
1216 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1217
1218 /* Write PID to a file. */
1219 if (pidfile_write(pfh) < 0) {
1220 pjdlog_errno(LOG_WARNING,
1221 "Unable to write PID to a file");
1222 }
1223 }
1224
1225 pjdlog_info("Started successfully, running protocol version %d.",
1226 HAST_PROTO_VERSION);
1227
1228 pjdlog_debug(1, "Listening on control address %s.",
1229 cfg->hc_controladdr);
1230 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next)
1231 pjdlog_info("Listening on address %s.", lst->hl_addr);
1232
1233 hook_init();
1234
1235 main_loop();
1236
1237 exit(0);
1238}
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
4 * All rights reserved.
5 *
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 225830 2011-09-28 13:08:51Z pjd $");
33
34#include <sys/param.h>
35#include <sys/linker.h>
36#include <sys/module.h>
37#include <sys/stat.h>
38#include <sys/wait.h>
39
40#include <err.h>
41#include <errno.h>
42#include <libutil.h>
43#include <signal.h>
44#include <stdbool.h>
45#include <stdio.h>
46#include <stdlib.h>
47#include <string.h>
48#include <sysexits.h>
49#include <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. */
64const char *cfgpath = HAST_CONFIG;
65/* Hastd configuration. */
66static struct hastd_config *cfg;
67/* Was SIGINT or SIGTERM signal received? */
68bool sigexit_received = false;
69/* PID file handle. */
70struct pidfh *pfh;
71
72/* How often check for hooks running for too long. */
73#define REPORT_INTERVAL 5
74
75static void
76usage(void)
77{
78
79 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
80}
81
82static void
83g_gate_load(void)
84{
85
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) {
90 pjdlog_exit(EX_OSERR,
91 "Unable to load geom_gate module");
92 }
93 }
94 }
95}
96
97void
98descriptors_cleanup(struct hast_resource *res)
99{
100 struct hast_resource *tres;
101 struct hastd_listen *lst;
102
103 TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) {
104 if (tres == res) {
105 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY ||
106 (res->hr_remotein == NULL &&
107 res->hr_remoteout == NULL));
108 continue;
109 }
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);
120 }
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);
127 }
128 (void)pidfile_close(pfh);
129 hook_fini();
130 pjdlog_fini();
131}
132
133static const char *
134dtype2str(mode_t mode)
135{
136
137 if (S_ISBLK(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))
150 return ("socket");
151 else if (S_ISWHT(mode))
152 return ("whiteout");
153 else
154 return ("unknown");
155}
156
157void
158descriptors_assert(const struct hast_resource *res, int pjdlogmode)
159{
160 char msg[256];
161 struct stat sb;
162 long maxfd;
163 bool isopen;
164 mode_t mode;
165 int fd;
166
167 /*
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.
171 */
172 msg[0] = '\0';
173
174 maxfd = sysconf(_SC_OPEN_MAX);
175 if (maxfd < 0) {
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");
180 pjdlog_fini();
181 maxfd = 16384;
182 }
183 for (fd = 0; fd <= maxfd; fd++) {
184 if (fstat(fd, &sb) == 0) {
185 isopen = true;
186 mode = sb.st_mode;
187 } else if (errno == EBADF) {
188 isopen = false;
189 mode = 0;
190 } else {
191 (void)snprintf(msg, sizeof(msg),
192 "Unable to fstat descriptor %d: %s", fd,
193 strerror(errno));
194 break;
195 }
196 if (fd == STDIN_FILENO || fd == STDOUT_FILENO ||
197 fd == STDERR_FILENO) {
198 if (!isopen) {
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")));
203 break;
204 }
205 } else if (fd == proto_descriptor(res->hr_event)) {
206 if (!isopen) {
207 (void)snprintf(msg, sizeof(msg),
208 "Descriptor %d (event) is closed, but should be open.",
209 fd);
210 break;
211 }
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));
216 break;
217 }
218 } else if (fd == proto_descriptor(res->hr_ctrl)) {
219 if (!isopen) {
220 (void)snprintf(msg, sizeof(msg),
221 "Descriptor %d (ctrl) is closed, but should be open.",
222 fd);
223 break;
224 }
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));
229 break;
230 }
231 } else if (res->hr_role == HAST_ROLE_PRIMARY &&
232 fd == proto_descriptor(res->hr_conn)) {
233 if (!isopen) {
234 (void)snprintf(msg, sizeof(msg),
235 "Descriptor %d (conn) is closed, but should be open.",
236 fd);
237 break;
238 }
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));
243 break;
244 }
245 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
246 res->hr_conn != NULL &&
247 fd == proto_descriptor(res->hr_conn)) {
248 if (isopen) {
249 (void)snprintf(msg, sizeof(msg),
250 "Descriptor %d (conn) is open, but should be closed.",
251 fd);
252 break;
253 }
254 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
255 fd == proto_descriptor(res->hr_remotein)) {
256 if (!isopen) {
257 (void)snprintf(msg, sizeof(msg),
258 "Descriptor %d (remote in) is closed, but should be open.",
259 fd);
260 break;
261 }
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));
266 break;
267 }
268 } else if (res->hr_role == HAST_ROLE_SECONDARY &&
269 fd == proto_descriptor(res->hr_remoteout)) {
270 if (!isopen) {
271 (void)snprintf(msg, sizeof(msg),
272 "Descriptor %d (remote out) is closed, but should be open.",
273 fd);
274 break;
275 }
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));
280 break;
281 }
282 } else {
283 if (isopen) {
284 (void)snprintf(msg, sizeof(msg),
285 "Descriptor %d is open (%s), but should be closed.",
286 fd, dtype2str(mode));
287 break;
288 }
289 }
290 }
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);
296 }
297}
298
299static void
300child_exit_log(unsigned int pid, int status)
301{
302
303 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
304 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
305 pid);
306 } else if (WIFSIGNALED(status)) {
307 pjdlog_error("Worker process killed (pid=%u, signal=%d).",
308 pid, WTERMSIG(status));
309 } else {
310 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
311 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
312 }
313}
314
315static void
316child_exit(void)
317{
318 struct hast_resource *res;
319 int status;
320 pid_t pid;
321
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)
326 break;
327 }
328 if (res == NULL) {
329 /*
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.
333 */
334 hook_check_one(pid, status);
335 continue;
336 }
337 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
338 role2str(res->hr_role));
339 child_exit_log(pid, status);
340 child_cleanup(res);
341 if (res->hr_role == HAST_ROLE_PRIMARY) {
342 /*
343 * Restart child process if it was killed by signal
344 * or exited because of temporary problem.
345 */
346 if (WIFSIGNALED(status) ||
347 (WIFEXITED(status) &&
348 WEXITSTATUS(status) == EX_TEMPFAIL)) {
349 sleep(1);
350 pjdlog_info("Restarting worker process.");
351 hastd_primary(res);
352 } else {
353 res->hr_role = HAST_ROLE_INIT;
354 pjdlog_info("Changing resource role back to %s.",
355 role2str(res->hr_role));
356 }
357 }
358 pjdlog_prefix_set("%s", "");
359 }
360}
361
362static bool
363resource_needs_restart(const struct hast_resource *res0,
364 const struct hast_resource *res1)
365{
366
367 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0);
368
369 if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
370 return (true);
371 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
372 return (true);
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)
376 return (true);
377 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
378 return (true);
379 if (res0->hr_replication != res1->hr_replication)
380 return (true);
381 if (res0->hr_checksum != res1->hr_checksum)
382 return (true);
383 if (res0->hr_compression != res1->hr_compression)
384 return (true);
385 if (res0->hr_timeout != res1->hr_timeout)
386 return (true);
387 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
388 return (true);
389 /*
390 * When metaflush has changed we don't really need restart,
391 * but it is just easier this way.
392 */
393 if (res0->hr_metaflush != res1->hr_metaflush)
394 return (true);
395 }
396 return (false);
397}
398
399static bool
400resource_needs_reload(const struct hast_resource *res0,
401 const struct hast_resource *res1)
402{
403
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);
407
408 if (res0->hr_role != HAST_ROLE_PRIMARY)
409 return (false);
410
411 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
412 return (true);
413 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
414 return (true);
415 if (res0->hr_replication != res1->hr_replication)
416 return (true);
417 if (res0->hr_checksum != res1->hr_checksum)
418 return (true);
419 if (res0->hr_compression != res1->hr_compression)
420 return (true);
421 if (res0->hr_timeout != res1->hr_timeout)
422 return (true);
423 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
424 return (true);
425 if (res0->hr_metaflush != res1->hr_metaflush)
426 return (true);
427 return (false);
428}
429
430static void
431resource_reload(const struct hast_resource *res)
432{
433 struct nv *nvin, *nvout;
434 int error;
435
436 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
437
438 nvout = nv_alloc();
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) {
449 nv_free(nvout);
450 pjdlog_error("Unable to allocate header for reload message.");
451 return;
452 }
453 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
454 pjdlog_errno(LOG_ERR, "Unable to send reload message");
455 nv_free(nvout);
456 return;
457 }
458 nv_free(nvout);
459
460 /* Receive response. */
461 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
462 pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
463 return;
464 }
465 error = nv_get_int16(nvin, "error");
466 nv_free(nvin);
467 if (error != 0) {
468 pjdlog_common(LOG_ERR, 0, error, "Reload failed");
469 return;
470 }
471}
472
473static void
474hastd_reload(void)
475{
476 struct hastd_config *newcfg;
477 struct hast_resource *nres, *cres, *tres;
478 struct hastd_listen *nlst, *clst;
479 unsigned int nlisten;
480 uint8_t role;
481
482 pjdlog_info("Reloading configuration...");
483
484 newcfg = yy_config_parse(cfgpath, false);
485 if (newcfg == NULL)
486 goto failed;
487
488 /*
489 * Check if control address has changed.
490 */
491 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
492 if (proto_server(newcfg->hc_controladdr,
493 &newcfg->hc_controlconn) < 0) {
494 pjdlog_errno(LOG_ERR,
495 "Unable to listen on control address %s",
496 newcfg->hc_controladdr);
497 goto failed;
498 }
499 }
500 /*
501 * Check if any listen address has changed.
502 */
503 nlisten = 0;
504 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
505 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
506 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
507 break;
508 }
509 if (clst != NULL && clst->hl_conn != NULL) {
510 pjdlog_info("Keep listening on address %s.",
511 nlst->hl_addr);
512 nlst->hl_conn = clst->hl_conn;
513 nlisten++;
514 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
515 pjdlog_info("Listening on new address %s.",
516 nlst->hl_addr);
517 nlisten++;
518 } else {
519 pjdlog_errno(LOG_WARNING,
520 "Unable to listen on address %s", nlst->hl_addr);
521 }
522 }
523 if (nlisten == 0) {
524 pjdlog_error("No addresses to listen on.");
525 goto failed;
526 }
527
528 /* No failures from now on. */
529
530 /*
531 * Switch to new control socket.
532 */
533 if (newcfg->hc_controlconn != NULL) {
534 pjdlog_info("Control socket changed from %s to %s.",
535 cfg->hc_controladdr, newcfg->hc_controladdr);
536 proto_close(cfg->hc_controlconn);
537 cfg->hc_controlconn = newcfg->hc_controlconn;
538 newcfg->hc_controlconn = NULL;
539 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
540 sizeof(cfg->hc_controladdr));
541 }
542 /*
543 * Switch to new listen addresses. Close all that were removed.
544 */
545 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) {
546 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) {
547 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0)
548 break;
549 }
550 if (nlst == NULL && clst->hl_conn != NULL) {
551 proto_close(clst->hl_conn);
552 pjdlog_info("No longer listening on address %s.",
553 clst->hl_addr);
554 }
555 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
556 free(clst);
557 }
558 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
559
560 /*
561 * Stop and remove resources that were removed from the configuration.
562 */
563 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
564 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
565 if (strcmp(cres->hr_name, nres->hr_name) == 0)
566 break;
567 }
568 if (nres == NULL) {
569 control_set_role(cres, HAST_ROLE_INIT);
570 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
571 pjdlog_info("Resource %s removed.", cres->hr_name);
572 free(cres);
573 }
574 }
575 /*
576 * Move new resources to the current configuration.
577 */
578 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
579 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
580 if (strcmp(cres->hr_name, nres->hr_name) == 0)
581 break;
582 }
583 if (cres == NULL) {
584 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
585 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
586 pjdlog_info("Resource %s added.", nres->hr_name);
587 }
588 }
589 /*
590 * Deal with modified resources.
591 * Depending on what has changed exactly we might want to perform
592 * different actions.
593 *
594 * We do full resource restart in the following situations:
595 * Resource role is INIT or SECONDARY.
596 * Resource role is PRIMARY and path to local component or provider
597 * name has changed.
598 * In case of PRIMARY, the worker process will be killed and restarted,
599 * which also means removing /dev/hast/<name> provider and
600 * recreating it.
601 *
602 * We do just reload (send SIGHUP to worker process) if we act as
603 * PRIMARY, but only if remote address, source address, replication
604 * mode, timeout, execution path or metaflush has changed.
605 * For those, there is no need to restart worker process.
606 * If PRIMARY receives SIGHUP, it will reconnect if remote address or
607 * source address has changed or it will set new timeout if only timeout
608 * has changed or it will update metaflush if only metaflush has
609 * changed.
610 */
611 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
612 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
613 if (strcmp(cres->hr_name, nres->hr_name) == 0)
614 break;
615 }
616 PJDLOG_ASSERT(cres != NULL);
617 if (resource_needs_restart(cres, nres)) {
618 pjdlog_info("Resource %s configuration was modified, restarting it.",
619 cres->hr_name);
620 role = cres->hr_role;
621 control_set_role(cres, HAST_ROLE_INIT);
622 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
623 free(cres);
624 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
625 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
626 control_set_role(nres, role);
627 } else if (resource_needs_reload(cres, nres)) {
628 pjdlog_info("Resource %s configuration was modified, reloading it.",
629 cres->hr_name);
630 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
631 sizeof(cres->hr_remoteaddr));
632 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr,
633 sizeof(cres->hr_sourceaddr));
634 cres->hr_replication = nres->hr_replication;
635 cres->hr_checksum = nres->hr_checksum;
636 cres->hr_compression = nres->hr_compression;
637 cres->hr_timeout = nres->hr_timeout;
638 strlcpy(cres->hr_exec, nres->hr_exec,
639 sizeof(cres->hr_exec));
640 cres->hr_metaflush = nres->hr_metaflush;
641 if (cres->hr_workerpid != 0)
642 resource_reload(cres);
643 }
644 }
645
646 yy_config_free(newcfg);
647 pjdlog_info("Configuration reloaded successfully.");
648 return;
649failed:
650 if (newcfg != NULL) {
651 if (newcfg->hc_controlconn != NULL)
652 proto_close(newcfg->hc_controlconn);
653 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) {
654 if (nlst->hl_conn != NULL) {
655 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) {
656 if (strcmp(nlst->hl_addr,
657 clst->hl_addr) == 0) {
658 break;
659 }
660 }
661 if (clst == NULL || clst->hl_conn == NULL)
662 proto_close(nlst->hl_conn);
663 }
664 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
665 free(nlst);
666 }
667 yy_config_free(newcfg);
668 }
669 pjdlog_warning("Configuration not reloaded.");
670}
671
672static void
673terminate_workers(void)
674{
675 struct hast_resource *res;
676
677 pjdlog_info("Termination signal received, exiting.");
678 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
679 if (res->hr_workerpid == 0)
680 continue;
681 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
682 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
683 if (kill(res->hr_workerpid, SIGTERM) == 0)
684 continue;
685 pjdlog_errno(LOG_WARNING,
686 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
687 res->hr_name, role2str(res->hr_role), res->hr_workerpid);
688 }
689}
690
691static void
692listen_accept(struct hastd_listen *lst)
693{
694 struct hast_resource *res;
695 struct proto_conn *conn;
696 struct nv *nvin, *nvout, *nverr;
697 const char *resname;
698 const unsigned char *token;
699 char laddr[256], raddr[256];
700 size_t size;
701 pid_t pid;
702 int status;
703
704 proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
705 pjdlog_debug(1, "Accepting connection to %s.", laddr);
706
707 if (proto_accept(lst->hl_conn, &conn) < 0) {
708 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
709 return;
710 }
711
712 proto_local_address(conn, laddr, sizeof(laddr));
713 proto_remote_address(conn, raddr, sizeof(raddr));
714 pjdlog_info("Connection from %s to %s.", raddr, laddr);
715
716 /* Error in setting timeout is not critical, but why should it fail? */
717 if (proto_timeout(conn, HAST_TIMEOUT) < 0)
718 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
719
720 nvin = nvout = nverr = NULL;
721
722 /*
723 * Before receiving any data see if remote host have access to any
724 * resource.
725 */
726 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
727 if (proto_address_match(conn, res->hr_remoteaddr))
728 break;
729 }
730 if (res == NULL) {
731 pjdlog_error("Client %s isn't known.", raddr);
732 goto close;
733 }
734 /* Ok, remote host can access at least one resource. */
735
736 if (hast_proto_recv_hdr(conn, &nvin) < 0) {
737 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
738 raddr);
739 goto close;
740 }
741
742 resname = nv_get_string(nvin, "resource");
743 if (resname == NULL) {
744 pjdlog_error("No 'resource' field in the header received from %s.",
745 raddr);
746 goto close;
747 }
748 pjdlog_debug(2, "%s: resource=%s", raddr, resname);
749 token = nv_get_uint8_array(nvin, &size, "token");
750 /*
751 * NULL token means that this is first conection.
752 */
753 if (token != NULL && size != sizeof(res->hr_token)) {
754 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
755 raddr, sizeof(res->hr_token), size);
756 goto close;
757 }
758
759 /*
760 * From now on we want to send errors to the remote node.
761 */
762 nverr = nv_alloc();
763
764 /* Find resource related to this connection. */
765 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
766 if (strcmp(resname, res->hr_name) == 0)
767 break;
768 }
769 /* Have we found the resource? */
770 if (res == NULL) {
771 pjdlog_error("No resource '%s' as requested by %s.",
772 resname, raddr);
773 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
774 goto fail;
775 }
776
777 /* Now that we know resource name setup log prefix. */
778 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
779
780 /* Does the remote host have access to this resource? */
781 if (!proto_address_match(conn, res->hr_remoteaddr)) {
782 pjdlog_error("Client %s has no access to the resource.", raddr);
783 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
784 goto fail;
785 }
786 /* Is the resource marked as secondary? */
787 if (res->hr_role != HAST_ROLE_SECONDARY) {
788 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.",
789 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
790 raddr);
791 nv_add_stringf(nverr, "errmsg",
792 "Remote node acts as %s for the resource and not as %s.",
793 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
794 if (res->hr_role == HAST_ROLE_PRIMARY) {
795 /*
796 * If we act as primary request the other side to wait
797 * for us a bit, as we might be finishing cleanups.
798 */
799 nv_add_uint8(nverr, 1, "wait");
800 }
801 goto fail;
802 }
803 /* Does token (if exists) match? */
804 if (token != NULL && memcmp(token, res->hr_token,
805 sizeof(res->hr_token)) != 0) {
806 pjdlog_error("Token received from %s doesn't match.", raddr);
807 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
808 goto fail;
809 }
810 /*
811 * If there is no token, but we have half-open connection
812 * (only remotein) or full connection (worker process is running)
813 * we have to cancel those and accept the new connection.
814 */
815 if (token == NULL) {
816 PJDLOG_ASSERT(res->hr_remoteout == NULL);
817 pjdlog_debug(1, "Initial connection from %s.", raddr);
818 if (res->hr_workerpid != 0) {
819 PJDLOG_ASSERT(res->hr_remotein == NULL);
820 pjdlog_debug(1,
821 "Worker process exists (pid=%u), stopping it.",
822 (unsigned int)res->hr_workerpid);
823 /* Stop child process. */
824 if (kill(res->hr_workerpid, SIGINT) < 0) {
825 pjdlog_errno(LOG_ERR,
826 "Unable to stop worker process (pid=%u)",
827 (unsigned int)res->hr_workerpid);
828 /*
829 * Other than logging the problem we
830 * ignore it - nothing smart to do.
831 */
832 }
833 /* Wait for it to exit. */
834 else if ((pid = waitpid(res->hr_workerpid,
835 &status, 0)) != res->hr_workerpid) {
836 /* We can only log the problem. */
837 pjdlog_errno(LOG_ERR,
838 "Waiting for worker process (pid=%u) failed",
839 (unsigned int)res->hr_workerpid);
840 } else {
841 child_exit_log(res->hr_workerpid, status);
842 }
843 child_cleanup(res);
844 } else if (res->hr_remotein != NULL) {
845 char oaddr[256];
846
847 proto_remote_address(res->hr_remotein, oaddr,
848 sizeof(oaddr));
849 pjdlog_debug(1,
850 "Canceling half-open connection from %s on connection from %s.",
851 oaddr, raddr);
852 proto_close(res->hr_remotein);
853 res->hr_remotein = NULL;
854 }
855 }
856
857 /*
858 * Checks and cleanups are done.
859 */
860
861 if (token == NULL) {
862 arc4random_buf(res->hr_token, sizeof(res->hr_token));
863 nvout = nv_alloc();
864 nv_add_uint8_array(nvout, res->hr_token,
865 sizeof(res->hr_token), "token");
866 if (nv_error(nvout) != 0) {
867 pjdlog_common(LOG_ERR, 0, nv_error(nvout),
868 "Unable to prepare return header for %s", raddr);
869 nv_add_stringf(nverr, "errmsg",
870 "Remote node was unable to prepare return header: %s.",
871 strerror(nv_error(nvout)));
872 goto fail;
873 }
874 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
875 int error = errno;
876
877 pjdlog_errno(LOG_ERR, "Unable to send response to %s",
878 raddr);
879 nv_add_stringf(nverr, "errmsg",
880 "Remote node was unable to send response: %s.",
881 strerror(error));
882 goto fail;
883 }
884 res->hr_remotein = conn;
885 pjdlog_debug(1, "Incoming connection from %s configured.",
886 raddr);
887 } else {
888 res->hr_remoteout = conn;
889 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
890 hastd_secondary(res, nvin);
891 }
892 nv_free(nvin);
893 nv_free(nvout);
894 nv_free(nverr);
895 pjdlog_prefix_set("%s", "");
896 return;
897fail:
898 if (nv_error(nverr) != 0) {
899 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
900 "Unable to prepare error header for %s", raddr);
901 goto close;
902 }
903 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
904 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
905 goto close;
906 }
907close:
908 if (nvin != NULL)
909 nv_free(nvin);
910 if (nvout != NULL)
911 nv_free(nvout);
912 if (nverr != NULL)
913 nv_free(nverr);
914 proto_close(conn);
915 pjdlog_prefix_set("%s", "");
916}
917
918static void
919connection_migrate(struct hast_resource *res)
920{
921 struct proto_conn *conn;
922 int16_t val = 0;
923
924 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
925
926 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
927
928 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
929 pjdlog_errno(LOG_WARNING,
930 "Unable to receive connection command");
931 return;
932 }
933 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
934 res->hr_remoteaddr, &conn) < 0) {
935 val = errno;
936 pjdlog_errno(LOG_WARNING,
937 "Unable to create outgoing connection to %s",
938 res->hr_remoteaddr);
939 goto out;
940 }
941 if (proto_connect(conn, -1) < 0) {
942 val = errno;
943 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
944 res->hr_remoteaddr);
945 proto_close(conn);
946 goto out;
947 }
948 val = 0;
949out:
950 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
951 pjdlog_errno(LOG_WARNING,
952 "Unable to send reply to connection request");
953 }
954 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
955 pjdlog_errno(LOG_WARNING, "Unable to send connection");
956
957 pjdlog_prefix_set("%s", "");
958}
959
960static void
961check_signals(void)
962{
963 struct timespec sigtimeout;
964 sigset_t mask;
965 int signo;
966
967 sigtimeout.tv_sec = 0;
968 sigtimeout.tv_nsec = 0;
969
970 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
971 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
972 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
973 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
974 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
975
976 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
977 switch (signo) {
978 case SIGINT:
979 case SIGTERM:
980 sigexit_received = true;
981 terminate_workers();
982 proto_close(cfg->hc_controlconn);
983 exit(EX_OK);
984 break;
985 case SIGCHLD:
986 child_exit();
987 break;
988 case SIGHUP:
989 hastd_reload();
990 break;
991 default:
992 PJDLOG_ABORT("Unexpected signal (%d).", signo);
993 }
994 }
995}
996
997static void
998main_loop(void)
999{
1000 struct hast_resource *res;
1001 struct hastd_listen *lst;
1002 struct timeval seltimeout;
1003 int fd, maxfd, ret;
1004 time_t lastcheck, now;
1005 fd_set rfds;
1006
1007 lastcheck = time(NULL);
1008 seltimeout.tv_sec = REPORT_INTERVAL;
1009 seltimeout.tv_usec = 0;
1010
1011 for (;;) {
1012 check_signals();
1013
1014 /* Setup descriptors for select(2). */
1015 FD_ZERO(&rfds);
1016 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1017 PJDLOG_ASSERT(fd >= 0);
1018 FD_SET(fd, &rfds);
1019 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1020 if (lst->hl_conn == NULL)
1021 continue;
1022 fd = proto_descriptor(lst->hl_conn);
1023 PJDLOG_ASSERT(fd >= 0);
1024 FD_SET(fd, &rfds);
1025 maxfd = fd > maxfd ? fd : maxfd;
1026 }
1027 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1028 if (res->hr_event == NULL)
1029 continue;
1030 fd = proto_descriptor(res->hr_event);
1031 PJDLOG_ASSERT(fd >= 0);
1032 FD_SET(fd, &rfds);
1033 maxfd = fd > maxfd ? fd : maxfd;
1034 if (res->hr_role == HAST_ROLE_PRIMARY) {
1035 /* Only primary workers asks for connections. */
1036 PJDLOG_ASSERT(res->hr_conn != NULL);
1037 fd = proto_descriptor(res->hr_conn);
1038 PJDLOG_ASSERT(fd >= 0);
1039 FD_SET(fd, &rfds);
1040 maxfd = fd > maxfd ? fd : maxfd;
1041 } else {
1042 PJDLOG_ASSERT(res->hr_conn == NULL);
1043 }
1044 }
1045
1046 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1047 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1048 now = time(NULL);
1049 if (lastcheck + REPORT_INTERVAL <= now) {
1050 hook_check();
1051 lastcheck = now;
1052 }
1053 if (ret == 0) {
1054 /*
1055 * select(2) timed out, so there should be no
1056 * descriptors to check.
1057 */
1058 continue;
1059 } else if (ret == -1) {
1060 if (errno == EINTR)
1061 continue;
1062 KEEP_ERRNO((void)pidfile_remove(pfh));
1063 pjdlog_exit(EX_OSERR, "select() failed");
1064 }
1065
1066 /*
1067 * Check for signals before we do anything to update our
1068 * info about terminated workers in the meantime.
1069 */
1070 check_signals();
1071
1072 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
1073 control_handle(cfg);
1074 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1075 if (lst->hl_conn == NULL)
1076 continue;
1077 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1078 listen_accept(lst);
1079 }
1080 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1081 if (res->hr_event == NULL)
1082 continue;
1083 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1084 if (event_recv(res) == 0)
1085 continue;
1086 /* The worker process exited? */
1087 proto_close(res->hr_event);
1088 res->hr_event = NULL;
1089 if (res->hr_conn != NULL) {
1090 proto_close(res->hr_conn);
1091 res->hr_conn = NULL;
1092 }
1093 continue;
1094 }
1095 if (res->hr_role == HAST_ROLE_PRIMARY) {
1096 PJDLOG_ASSERT(res->hr_conn != NULL);
1097 if (FD_ISSET(proto_descriptor(res->hr_conn),
1098 &rfds)) {
1099 connection_migrate(res);
1100 }
1101 } else {
1102 PJDLOG_ASSERT(res->hr_conn == NULL);
1103 }
1104 }
1105 }
1106}
1107
1108static void
1109dummy_sighandler(int sig __unused)
1110{
1111 /* Nothing to do. */
1112}
1113
1114int
1115main(int argc, char *argv[])
1116{
1117 struct hastd_listen *lst;
1118 const char *pidfile;
1119 pid_t otherpid;
1120 bool foreground;
1121 int debuglevel;
1122 sigset_t mask;
1123
1124 foreground = false;
1125 debuglevel = 0;
1126 pidfile = HASTD_PIDFILE;
1127
1128 for (;;) {
1129 int ch;
1130
1131 ch = getopt(argc, argv, "c:dFhP:");
1132 if (ch == -1)
1133 break;
1134 switch (ch) {
1135 case 'c':
1136 cfgpath = optarg;
1137 break;
1138 case 'd':
1139 debuglevel++;
1140 break;
1141 case 'F':
1142 foreground = true;
1143 break;
1144 case 'P':
1145 pidfile = optarg;
1146 break;
1147 case 'h':
1148 default:
1149 usage();
1150 }
1151 }
1152 argc -= optind;
1153 argv += optind;
1154
1155 pjdlog_init(PJDLOG_MODE_STD);
1156 pjdlog_debug_set(debuglevel);
1157
1158 g_gate_load();
1159
1160 pfh = pidfile_open(pidfile, 0600, &otherpid);
1161 if (pfh == NULL) {
1162 if (errno == EEXIST) {
1163 pjdlog_exitx(EX_TEMPFAIL,
1164 "Another hastd is already running, pid: %jd.",
1165 (intmax_t)otherpid);
1166 }
1167 /* If we cannot create pidfile from other reasons, only warn. */
1168 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
1169 }
1170
1171 cfg = yy_config_parse(cfgpath, true);
1172 PJDLOG_ASSERT(cfg != NULL);
1173
1174 /*
1175 * Restore default actions for interesting signals in case parent
1176 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1177 */
1178 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR);
1179 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR);
1180 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR);
1181 /*
1182 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1183 * so we can mask it.
1184 */
1185 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1186
1187 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
1188 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
1189 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
1190 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
1191 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
1192 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1193
1194 /* Listen on control address. */
1195 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
1196 KEEP_ERRNO((void)pidfile_remove(pfh));
1197 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
1198 cfg->hc_controladdr);
1199 }
1200 /* Listen for remote connections. */
1201 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1202 if (proto_server(lst->hl_addr, &lst->hl_conn) < 0) {
1203 KEEP_ERRNO((void)pidfile_remove(pfh));
1204 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
1205 lst->hl_addr);
1206 }
1207 }
1208
1209 if (!foreground) {
1210 if (daemon(0, 0) < 0) {
1211 KEEP_ERRNO((void)pidfile_remove(pfh));
1212 pjdlog_exit(EX_OSERR, "Unable to daemonize");
1213 }
1214
1215 /* Start logging to syslog. */
1216 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1217
1218 /* Write PID to a file. */
1219 if (pidfile_write(pfh) < 0) {
1220 pjdlog_errno(LOG_WARNING,
1221 "Unable to write PID to a file");
1222 }
1223 }
1224
1225 pjdlog_info("Started successfully, running protocol version %d.",
1226 HAST_PROTO_VERSION);
1227
1228 pjdlog_debug(1, "Listening on control address %s.",
1229 cfg->hc_controladdr);
1230 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next)
1231 pjdlog_info("Listening on address %s.", lst->hl_addr);
1232
1233 hook_init();
1234
1235 main_loop();
1236
1237 exit(0);
1238}