hastd.c revision 222108
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 222108 2011-05-19 23:18:42Z 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	return (false);
391}
392
393static bool
394resource_needs_reload(const struct hast_resource *res0,
395    const struct hast_resource *res1)
396{
397
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);
401
402	if (res0->hr_role != HAST_ROLE_PRIMARY)
403		return (false);
404
405	if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
406		return (true);
407	if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0)
408		return (true);
409	if (res0->hr_replication != res1->hr_replication)
410		return (true);
411	if (res0->hr_checksum != res1->hr_checksum)
412		return (true);
413	if (res0->hr_compression != res1->hr_compression)
414		return (true);
415	if (res0->hr_timeout != res1->hr_timeout)
416		return (true);
417	if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
418		return (true);
419	return (false);
420}
421
422static void
423resource_reload(const struct hast_resource *res)
424{
425	struct nv *nvin, *nvout;
426	int error;
427
428	PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
429
430	nvout = nv_alloc();
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) {
440		nv_free(nvout);
441		pjdlog_error("Unable to allocate header for reload message.");
442		return;
443	}
444	if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) {
445		pjdlog_errno(LOG_ERR, "Unable to send reload message");
446		nv_free(nvout);
447		return;
448	}
449	nv_free(nvout);
450
451	/* Receive response. */
452	if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
453		pjdlog_errno(LOG_ERR, "Unable to receive reload reply");
454		return;
455	}
456	error = nv_get_int16(nvin, "error");
457	nv_free(nvin);
458	if (error != 0) {
459		pjdlog_common(LOG_ERR, 0, error, "Reload failed");
460		return;
461	}
462}
463
464static void
465hastd_reload(void)
466{
467	struct hastd_config *newcfg;
468	struct hast_resource *nres, *cres, *tres;
469	struct hastd_listen *nlst, *clst;
470	unsigned int nlisten;
471	uint8_t role;
472
473	pjdlog_info("Reloading configuration...");
474
475	newcfg = yy_config_parse(cfgpath, false);
476	if (newcfg == NULL)
477		goto failed;
478
479	/*
480	 * Check if control address has changed.
481	 */
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);
488			goto failed;
489		}
490	}
491	/*
492	 * Check if any listen address has changed.
493	 */
494	nlisten = 0;
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)
498				break;
499		}
500		if (clst != NULL && clst->hl_conn != NULL) {
501			pjdlog_info("Keep listening on address %s.",
502			    nlst->hl_addr);
503			nlst->hl_conn = clst->hl_conn;
504			nlisten++;
505		} else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) {
506			pjdlog_info("Listening on new address %s.",
507			    nlst->hl_addr);
508			nlisten++;
509		} else {
510			pjdlog_errno(LOG_WARNING,
511			    "Unable to listen on address %s", nlst->hl_addr);
512		}
513	}
514	if (nlisten == 0) {
515		pjdlog_error("No addresses to listen on.");
516		goto failed;
517	}
518
519	/* No failures from now on. */
520
521	/*
522	 * Switch to new control socket.
523	 */
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));
532	}
533	/*
534	 * Switch to new listen addresses. Close all that were removed.
535	 */
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)
539				break;
540		}
541		if (nlst == NULL && clst->hl_conn != NULL) {
542			proto_close(clst->hl_conn);
543			pjdlog_info("No longer listening on address %s.",
544			    clst->hl_addr);
545		}
546		TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next);
547		free(clst);
548	}
549	TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next);
550
551	/*
552	 * Stop and remove resources that were removed from the configuration.
553	 */
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)
557				break;
558		}
559		if (nres == NULL) {
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);
563			free(cres);
564		}
565	}
566	/*
567	 * Move new resources to the current configuration.
568	 */
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)
572				break;
573		}
574		if (cres == NULL) {
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);
578		}
579	}
580	/*
581	 * Deal with modified resources.
582	 * Depending on what has changed exactly we might want to perform
583	 * different actions.
584	 *
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
588	 * name has changed.
589	 * In case of PRIMARY, the worker process will be killed and restarted,
590	 * which also means removing /dev/hast/<name> provider and
591	 * recreating it.
592	 *
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
596	 * worker process.
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.
600	 */
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)
604				break;
605		}
606		PJDLOG_ASSERT(cres != NULL);
607		if (resource_needs_restart(cres, nres)) {
608			pjdlog_info("Resource %s configuration was modified, restarting it.",
609			    cres->hr_name);
610			role = cres->hr_role;
611			control_set_role(cres, HAST_ROLE_INIT);
612			TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
613			free(cres);
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.",
619			    cres->hr_name);
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);
632		}
633	}
634
635	yy_config_free(newcfg);
636	pjdlog_info("Configuration reloaded successfully.");
637	return;
638failed:
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) {
647						break;
648					}
649				}
650				if (clst == NULL || clst->hl_conn == NULL)
651					proto_close(nlst->hl_conn);
652			}
653			TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next);
654			free(nlst);
655		}
656		yy_config_free(newcfg);
657	}
658	pjdlog_warning("Configuration not reloaded.");
659}
660
661static void
662terminate_workers(void)
663{
664	struct hast_resource *res;
665
666	pjdlog_info("Termination signal received, exiting.");
667	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
668		if (res->hr_workerpid == 0)
669			continue;
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)
673			continue;
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);
677	}
678}
679
680static void
681listen_accept(struct hastd_listen *lst)
682{
683	struct hast_resource *res;
684	struct proto_conn *conn;
685	struct nv *nvin, *nvout, *nverr;
686	const char *resname;
687	const unsigned char *token;
688	char laddr[256], raddr[256];
689	size_t size;
690	pid_t pid;
691	int status;
692
693	proto_local_address(lst->hl_conn, laddr, sizeof(laddr));
694	pjdlog_debug(1, "Accepting connection to %s.", laddr);
695
696	if (proto_accept(lst->hl_conn, &conn) < 0) {
697		pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
698		return;
699	}
700
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);
704
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");
708
709	nvin = nvout = nverr = NULL;
710
711	/*
712	 * Before receiving any data see if remote host have access to any
713	 * resource.
714	 */
715	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
716		if (proto_address_match(conn, res->hr_remoteaddr))
717			break;
718	}
719	if (res == NULL) {
720		pjdlog_error("Client %s isn't known.", raddr);
721		goto close;
722	}
723	/* Ok, remote host can access at least one resource. */
724
725	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
726		pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
727		    raddr);
728		goto close;
729	}
730
731	resname = nv_get_string(nvin, "resource");
732	if (resname == NULL) {
733		pjdlog_error("No 'resource' field in the header received from %s.",
734		    raddr);
735		goto close;
736	}
737	pjdlog_debug(2, "%s: resource=%s", raddr, resname);
738	token = nv_get_uint8_array(nvin, &size, "token");
739	/*
740	 * NULL token means that this is first conection.
741	 */
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);
745		goto close;
746	}
747
748	/*
749	 * From now on we want to send errors to the remote node.
750	 */
751	nverr = nv_alloc();
752
753	/* Find resource related to this connection. */
754	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
755		if (strcmp(resname, res->hr_name) == 0)
756			break;
757	}
758	/* Have we found the resource? */
759	if (res == NULL) {
760		pjdlog_error("No resource '%s' as requested by %s.",
761		    resname, raddr);
762		nv_add_stringf(nverr, "errmsg", "Resource not configured.");
763		goto fail;
764	}
765
766	/* Now that we know resource name setup log prefix. */
767	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
768
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.");
773		goto fail;
774	}
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),
779		    raddr);
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) {
784			/*
785			 * If we act as primary request the other side to wait
786			 * for us a bit, as we might be finishing cleanups.
787			 */
788			nv_add_uint8(nverr, 1, "wait");
789		}
790		goto fail;
791	}
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.");
797		goto fail;
798	}
799	/*
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.
803	 */
804	if (token == NULL) {
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);
809			pjdlog_debug(1,
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);
817				/*
818				 * Other than logging the problem we
819				 * ignore it - nothing smart to do.
820				 */
821			}
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);
829			} else {
830				child_exit_log(res->hr_workerpid, status);
831			}
832			child_cleanup(res);
833		} else if (res->hr_remotein != NULL) {
834			char oaddr[256];
835
836			proto_remote_address(res->hr_remotein, oaddr,
837			    sizeof(oaddr));
838			pjdlog_debug(1,
839			    "Canceling half-open connection from %s on connection from %s.",
840			    oaddr, raddr);
841			proto_close(res->hr_remotein);
842			res->hr_remotein = NULL;
843		}
844	}
845
846	/*
847	 * Checks and cleanups are done.
848	 */
849
850	if (token == NULL) {
851		arc4random_buf(res->hr_token, sizeof(res->hr_token));
852		nvout = nv_alloc();
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)));
861			goto fail;
862		}
863		if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
864			int error = errno;
865
866			pjdlog_errno(LOG_ERR, "Unable to send response to %s",
867			    raddr);
868			nv_add_stringf(nverr, "errmsg",
869			    "Remote node was unable to send response: %s.",
870			    strerror(error));
871			goto fail;
872		}
873		res->hr_remotein = conn;
874		pjdlog_debug(1, "Incoming connection from %s configured.",
875		    raddr);
876	} else {
877		res->hr_remoteout = conn;
878		pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
879		hastd_secondary(res, nvin);
880	}
881	nv_free(nvin);
882	nv_free(nvout);
883	nv_free(nverr);
884	pjdlog_prefix_set("%s", "");
885	return;
886fail:
887	if (nv_error(nverr) != 0) {
888		pjdlog_common(LOG_ERR, 0, nv_error(nverr),
889		    "Unable to prepare error header for %s", raddr);
890		goto close;
891	}
892	if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
893		pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
894		goto close;
895	}
896close:
897	if (nvin != NULL)
898		nv_free(nvin);
899	if (nvout != NULL)
900		nv_free(nvout);
901	if (nverr != NULL)
902		nv_free(nverr);
903	proto_close(conn);
904	pjdlog_prefix_set("%s", "");
905}
906
907static void
908connection_migrate(struct hast_resource *res)
909{
910	struct proto_conn *conn;
911	int16_t val = 0;
912
913	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
914
915	PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
916
917	if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) {
918		pjdlog_errno(LOG_WARNING,
919		    "Unable to receive connection command");
920		return;
921	}
922	if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL,
923	    res->hr_remoteaddr, &conn) < 0) {
924		val = errno;
925		pjdlog_errno(LOG_WARNING,
926		    "Unable to create outgoing connection to %s",
927		    res->hr_remoteaddr);
928		goto out;
929	}
930	if (proto_connect(conn, -1) < 0) {
931		val = errno;
932		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
933		    res->hr_remoteaddr);
934		proto_close(conn);
935		goto out;
936	}
937	val = 0;
938out:
939	if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) {
940		pjdlog_errno(LOG_WARNING,
941		    "Unable to send reply to connection request");
942	}
943	if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0)
944		pjdlog_errno(LOG_WARNING, "Unable to send connection");
945
946	pjdlog_prefix_set("%s", "");
947}
948
949static void
950check_signals(void)
951{
952	struct timespec sigtimeout;
953	sigset_t mask;
954	int signo;
955
956	sigtimeout.tv_sec = 0;
957	sigtimeout.tv_nsec = 0;
958
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);
964
965	while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
966		switch (signo) {
967		case SIGINT:
968		case SIGTERM:
969			sigexit_received = true;
970			terminate_workers();
971			proto_close(cfg->hc_controlconn);
972			exit(EX_OK);
973			break;
974		case SIGCHLD:
975			child_exit();
976			break;
977		case SIGHUP:
978			hastd_reload();
979			break;
980		default:
981			PJDLOG_ABORT("Unexpected signal (%d).", signo);
982		}
983	}
984}
985
986static void
987main_loop(void)
988{
989	struct hast_resource *res;
990	struct hastd_listen *lst;
991	struct timeval seltimeout;
992	int fd, maxfd, ret;
993	time_t lastcheck, now;
994	fd_set rfds;
995
996	lastcheck = time(NULL);
997	seltimeout.tv_sec = REPORT_INTERVAL;
998	seltimeout.tv_usec = 0;
999
1000	for (;;) {
1001		check_signals();
1002
1003		/* Setup descriptors for select(2). */
1004		FD_ZERO(&rfds);
1005		maxfd = fd = proto_descriptor(cfg->hc_controlconn);
1006		PJDLOG_ASSERT(fd >= 0);
1007		FD_SET(fd, &rfds);
1008		TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) {
1009			if (lst->hl_conn == NULL)
1010				continue;
1011			fd = proto_descriptor(lst->hl_conn);
1012			PJDLOG_ASSERT(fd >= 0);
1013			FD_SET(fd, &rfds);
1014			maxfd = fd > maxfd ? fd : maxfd;
1015		}
1016		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1017			if (res->hr_event == NULL)
1018				continue;
1019			fd = proto_descriptor(res->hr_event);
1020			PJDLOG_ASSERT(fd >= 0);
1021			FD_SET(fd, &rfds);
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);
1028				FD_SET(fd, &rfds);
1029				maxfd = fd > maxfd ? fd : maxfd;
1030			} else {
1031				PJDLOG_ASSERT(res->hr_conn == NULL);
1032			}
1033		}
1034
1035		PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE);
1036		ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
1037		now = time(NULL);
1038		if (lastcheck + REPORT_INTERVAL <= now) {
1039			hook_check();
1040			lastcheck = now;
1041		}
1042		if (ret == 0) {
1043			/*
1044			 * select(2) timed out, so there should be no
1045			 * descriptors to check.
1046			 */
1047			continue;
1048		} else if (ret == -1) {
1049			if (errno == EINTR)
1050				continue;
1051			KEEP_ERRNO((void)pidfile_remove(pfh));
1052			pjdlog_exit(EX_OSERR, "select() failed");
1053		}
1054
1055		/*
1056		 * Check for signals before we do anything to update our
1057		 * info about terminated workers in the meantime.
1058		 */
1059		check_signals();
1060
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)
1065				continue;
1066			if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds))
1067				listen_accept(lst);
1068		}
1069		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
1070			if (res->hr_event == NULL)
1071				continue;
1072			if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
1073				if (event_recv(res) == 0)
1074					continue;
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;
1081				}
1082				continue;
1083			}
1084			if (res->hr_role == HAST_ROLE_PRIMARY) {
1085				PJDLOG_ASSERT(res->hr_conn != NULL);
1086				if (FD_ISSET(proto_descriptor(res->hr_conn),
1087				    &rfds)) {
1088					connection_migrate(res);
1089				}
1090			} else {
1091				PJDLOG_ASSERT(res->hr_conn == NULL);
1092			}
1093		}
1094	}
1095}
1096
1097static void
1098dummy_sighandler(int sig __unused)
1099{
1100	/* Nothing to do. */
1101}
1102
1103int
1104main(int argc, char *argv[])
1105{
1106	struct hastd_listen *lst;
1107	const char *pidfile;
1108	pid_t otherpid;
1109	bool foreground;
1110	int debuglevel;
1111	sigset_t mask;
1112
1113	foreground = false;
1114	debuglevel = 0;
1115	pidfile = HASTD_PIDFILE;
1116
1117	for (;;) {
1118		int ch;
1119
1120		ch = getopt(argc, argv, "c:dFhP:");
1121		if (ch == -1)
1122			break;
1123		switch (ch) {
1124		case 'c':
1125			cfgpath = optarg;
1126			break;
1127		case 'd':
1128			debuglevel++;
1129			break;
1130		case 'F':
1131			foreground = true;
1132			break;
1133		case 'P':
1134			pidfile = optarg;
1135			break;
1136		case 'h':
1137		default:
1138			usage();
1139		}
1140	}
1141	argc -= optind;
1142	argv += optind;
1143
1144	pjdlog_init(PJDLOG_MODE_STD);
1145	pjdlog_debug_set(debuglevel);
1146
1147	g_gate_load();
1148
1149	pfh = pidfile_open(pidfile, 0600, &otherpid);
1150	if (pfh == NULL) {
1151		if (errno == EEXIST) {
1152			pjdlog_exitx(EX_TEMPFAIL,
1153			    "Another hastd is already running, pid: %jd.",
1154			    (intmax_t)otherpid);
1155		}
1156		/* If we cannot create pidfile from other reasons, only warn. */
1157		pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
1158	}
1159
1160	cfg = yy_config_parse(cfgpath, true);
1161	PJDLOG_ASSERT(cfg != NULL);
1162
1163	/*
1164	 * Restore default actions for interesting signals in case parent
1165	 * process (like init(8)) decided to ignore some of them (like SIGHUP).
1166	 */
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);
1170	/*
1171	 * Because SIGCHLD is ignored by default, setup dummy handler for it,
1172	 * so we can mask it.
1173	 */
1174	PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR);
1175
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);
1182
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);
1188	}
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",
1194			    lst->hl_addr);
1195		}
1196	}
1197
1198	if (!foreground) {
1199		if (daemon(0, 0) < 0) {
1200			KEEP_ERRNO((void)pidfile_remove(pfh));
1201			pjdlog_exit(EX_OSERR, "Unable to daemonize");
1202		}
1203
1204		/* Start logging to syslog. */
1205		pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
1206
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");
1211		}
1212	}
1213
1214	pjdlog_info("Started successfully, running protocol version %d.",
1215	    HAST_PROTO_VERSION);
1216
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);
1221
1222	hook_init();
1223
1224	main_loop();
1225
1226	exit(0);
1227}
1228