hastd.c revision 207348 
1/*-
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * All rights reserved.
4 *
5 * This software was developed by Pawel Jakub Dawidek under sponsorship from
6 * the FreeBSD Foundation.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#include <sys/cdefs.h>
31__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 207348 2010-04-28 22:41:06Z pjd $");
32
33#include <sys/param.h>
34#include <sys/linker.h>
35#include <sys/module.h>
36#include <sys/wait.h>
37
38#include <assert.h>
39#include <err.h>
40#include <errno.h>
41#include <libutil.h>
42#include <signal.h>
43#include <stdbool.h>
44#include <stdio.h>
45#include <stdlib.h>
46#include <string.h>
47#include <sysexits.h>
48#include <unistd.h>
49
50#include <activemap.h>
51#include <pjdlog.h>
52
53#include "control.h"
54#include "hast.h"
55#include "hast_proto.h"
56#include "hastd.h"
57#include "subr.h"
58
59/* Path to configuration file. */
60static const char *cfgpath = HAST_CONFIG;
61/* Hastd configuration. */
62static struct hastd_config *cfg;
63/* Was SIGCHLD signal received? */
64static bool sigchld_received = false;
65/* Was SIGHUP signal received? */
66static bool sighup_received = false;
67/* Was SIGINT or SIGTERM signal received? */
68bool sigexit_received = false;
69/* PID file handle. */
70struct pidfh *pfh;
71
72static void
73usage(void)
74{
75
76	errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
77}
78
79static void
80sighandler(int sig)
81{
82
83	switch (sig) {
84	case SIGCHLD:
85		sigchld_received = true;
86		break;
87	case SIGHUP:
88		sighup_received = true;
89		break;
90	default:
91		assert(!"invalid condition");
92	}
93}
94
95static void
96g_gate_load(void)
97{
98
99	if (modfind("g_gate") == -1) {
100		/* Not present in kernel, try loading it. */
101		if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) {
102			if (errno != EEXIST) {
103				pjdlog_exit(EX_OSERR,
104				    "Unable to load geom_gate module");
105			}
106		}
107	}
108}
109
110static void
111child_exit(void)
112{
113	struct hast_resource *res;
114	int status;
115	pid_t pid;
116
117	while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
118		/* Find resource related to the process that just exited. */
119		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
120			if (pid == res->hr_workerpid)
121				break;
122		}
123		if (res == NULL) {
124			/*
125			 * This can happen when new connection arrives and we
126			 * cancel child responsible for the old one.
127			 */
128			continue;
129		}
130		pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
131		    role2str(res->hr_role));
132		if (WEXITSTATUS(status) == 0) {
133			pjdlog_debug(1,
134			    "Worker process exited gracefully (pid=%u).",
135			    (unsigned int)pid);
136		} else {
137			pjdlog_error("Worker process failed (pid=%u, status=%d).",
138			    (unsigned int)pid, WEXITSTATUS(status));
139		}
140		proto_close(res->hr_ctrl);
141		res->hr_workerpid = 0;
142		if (res->hr_role == HAST_ROLE_PRIMARY) {
143			if (WEXITSTATUS(status) == EX_TEMPFAIL) {
144				sleep(1);
145				pjdlog_info("Restarting worker process.");
146				hastd_primary(res);
147			} else {
148				res->hr_role = HAST_ROLE_INIT;
149				pjdlog_info("Changing resource role back to %s.",
150				    role2str(res->hr_role));
151			}
152		}
153		pjdlog_prefix_set("%s", "");
154	}
155}
156
157static void
158hastd_reload(void)
159{
160
161	/* TODO */
162	pjdlog_warning("Configuration reload is not implemented.");
163}
164
165static void
166listen_accept(void)
167{
168	struct hast_resource *res;
169	struct proto_conn *conn;
170	struct nv *nvin, *nvout, *nverr;
171	const char *resname;
172	const unsigned char *token;
173	char laddr[256], raddr[256];
174	size_t size;
175	pid_t pid;
176	int status;
177
178	proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr));
179	pjdlog_debug(1, "Accepting connection to %s.", laddr);
180
181	if (proto_accept(cfg->hc_listenconn, &conn) < 0) {
182		pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
183		return;
184	}
185
186	proto_local_address(conn, laddr, sizeof(laddr));
187	proto_remote_address(conn, raddr, sizeof(raddr));
188	pjdlog_info("Connection from %s to %s.", laddr, raddr);
189
190	nvin = nvout = nverr = NULL;
191
192	/*
193	 * Before receiving any data see if remote host have access to any
194	 * resource.
195	 */
196	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
197		if (proto_address_match(conn, res->hr_remoteaddr))
198			break;
199	}
200	if (res == NULL) {
201		pjdlog_error("Client %s isn't known.", raddr);
202		goto close;
203	}
204	/* Ok, remote host can access at least one resource. */
205
206	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
207		pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
208		    raddr);
209		goto close;
210	}
211
212	resname = nv_get_string(nvin, "resource");
213	if (resname == NULL) {
214		pjdlog_error("No 'resource' field in the header received from %s.",
215		    raddr);
216		goto close;
217	}
218	pjdlog_debug(2, "%s: resource=%s", raddr, resname);
219	token = nv_get_uint8_array(nvin, &size, "token");
220	/*
221	 * NULL token means that this is first conection.
222	 */
223	if (token != NULL && size != sizeof(res->hr_token)) {
224		pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
225		    raddr, sizeof(res->hr_token), size);
226		goto close;
227	}
228
229	/*
230	 * From now on we want to send errors to the remote node.
231	 */
232	nverr = nv_alloc();
233
234	/* Find resource related to this connection. */
235	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
236		if (strcmp(resname, res->hr_name) == 0)
237			break;
238	}
239	/* Have we found the resource? */
240	if (res == NULL) {
241		pjdlog_error("No resource '%s' as requested by %s.",
242		    resname, raddr);
243		nv_add_stringf(nverr, "errmsg", "Resource not configured.");
244		goto fail;
245	}
246
247	/* Now that we know resource name setup log prefix. */
248	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
249
250	/* Does the remote host have access to this resource? */
251	if (!proto_address_match(conn, res->hr_remoteaddr)) {
252		pjdlog_error("Client %s has no access to the resource.", raddr);
253		nv_add_stringf(nverr, "errmsg", "No access to the resource.");
254		goto fail;
255	}
256	/* Is the resource marked as secondary? */
257	if (res->hr_role != HAST_ROLE_SECONDARY) {
258		pjdlog_error("We act as %s for the resource and not as %s as requested by %s.",
259		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
260		    raddr);
261		nv_add_stringf(nverr, "errmsg",
262		    "Remote node acts as %s for the resource and not as %s.",
263		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
264		goto fail;
265	}
266	/* Does token (if exists) match? */
267	if (token != NULL && memcmp(token, res->hr_token,
268	    sizeof(res->hr_token)) != 0) {
269		pjdlog_error("Token received from %s doesn't match.", raddr);
270		nv_add_stringf(nverr, "errmsg", "Toke doesn't match.");
271		goto fail;
272	}
273	/*
274	 * If there is no token, but we have half-open connection
275	 * (only remotein) or full connection (worker process is running)
276	 * we have to cancel those and accept the new connection.
277	 */
278	if (token == NULL) {
279		assert(res->hr_remoteout == NULL);
280		pjdlog_debug(1, "Initial connection from %s.", raddr);
281		if (res->hr_workerpid != 0) {
282			assert(res->hr_remotein == NULL);
283			pjdlog_debug(1,
284			    "Worker process exists (pid=%u), stopping it.",
285			    (unsigned int)res->hr_workerpid);
286			/* Stop child process. */
287			if (kill(res->hr_workerpid, SIGINT) < 0) {
288				pjdlog_errno(LOG_ERR,
289				    "Unable to stop worker process (pid=%u)",
290				    (unsigned int)res->hr_workerpid);
291				/*
292				 * Other than logging the problem we
293				 * ignore it - nothing smart to do.
294				 */
295			}
296			/* Wait for it to exit. */
297			else if ((pid = waitpid(res->hr_workerpid,
298			    &status, 0)) != res->hr_workerpid) {
299				pjdlog_errno(LOG_ERR,
300				    "Waiting for worker process (pid=%u) failed",
301				    (unsigned int)res->hr_workerpid);
302				/* See above. */
303			} else if (WEXITSTATUS(status) != 0) {
304				pjdlog_error("Worker process (pid=%u) exited ungracefully: status=%d.",
305				    (unsigned int)res->hr_workerpid,
306				    WEXITSTATUS(status));
307				/* See above. */
308			} else {
309				pjdlog_debug(1,
310				    "Worker process (pid=%u) exited gracefully.",
311				    (unsigned int)res->hr_workerpid);
312			}
313			res->hr_workerpid = 0;
314		} else if (res->hr_remotein != NULL) {
315			char oaddr[256];
316
317			proto_remote_address(conn, oaddr, sizeof(oaddr));
318			pjdlog_debug(1,
319			    "Canceling half-open connection from %s on connection from %s.",
320			    oaddr, raddr);
321			proto_close(res->hr_remotein);
322			res->hr_remotein = NULL;
323		}
324	}
325
326	/*
327	 * Checks and cleanups are done.
328	 */
329
330	if (token == NULL) {
331		arc4random_buf(res->hr_token, sizeof(res->hr_token));
332		nvout = nv_alloc();
333		nv_add_uint8_array(nvout, res->hr_token,
334		    sizeof(res->hr_token), "token");
335		if (nv_error(nvout) != 0) {
336			pjdlog_common(LOG_ERR, 0, nv_error(nvout),
337			    "Unable to prepare return header for %s", raddr);
338			nv_add_stringf(nverr, "errmsg",
339			    "Remote node was unable to prepare return header: %s.",
340			    strerror(nv_error(nvout)));
341			goto fail;
342		}
343		if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
344			int error = errno;
345
346			pjdlog_errno(LOG_ERR, "Unable to send response to %s",
347			    raddr);
348			nv_add_stringf(nverr, "errmsg",
349			    "Remote node was unable to send response: %s.",
350			    strerror(error));
351			goto fail;
352		}
353		res->hr_remotein = conn;
354		pjdlog_debug(1, "Incoming connection from %s configured.",
355		    raddr);
356	} else {
357		res->hr_remoteout = conn;
358		pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
359		hastd_secondary(res, nvin);
360	}
361	nv_free(nvin);
362	nv_free(nvout);
363	nv_free(nverr);
364	pjdlog_prefix_set("%s", "");
365	return;
366fail:
367	if (nv_error(nverr) != 0) {
368		pjdlog_common(LOG_ERR, 0, nv_error(nverr),
369		    "Unable to prepare error header for %s", raddr);
370		goto close;
371	}
372	if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
373		pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
374		goto close;
375	}
376close:
377	if (nvin != NULL)
378		nv_free(nvin);
379	if (nvout != NULL)
380		nv_free(nvout);
381	if (nverr != NULL)
382		nv_free(nverr);
383	proto_close(conn);
384	pjdlog_prefix_set("%s", "");
385}
386
387static void
388main_loop(void)
389{
390	fd_set rfds, wfds;
391	int fd, maxfd, ret;
392
393	for (;;) {
394		if (sigchld_received) {
395			sigchld_received = false;
396			child_exit();
397		}
398		if (sighup_received) {
399			sighup_received = false;
400			hastd_reload();
401		}
402
403		maxfd = 0;
404		FD_ZERO(&rfds);
405		FD_ZERO(&wfds);
406
407		/* Setup descriptors for select(2). */
408#define	SETUP_FD(conn)	do {						\
409	fd = proto_descriptor(conn);					\
410	if (fd >= 0) {							\
411		maxfd = fd > maxfd ? fd : maxfd;			\
412		FD_SET(fd, &rfds);					\
413		FD_SET(fd, &wfds);					\
414	}								\
415} while (0)
416		SETUP_FD(cfg->hc_controlconn);
417		SETUP_FD(cfg->hc_listenconn);
418#undef	SETUP_FD
419
420		ret = select(maxfd + 1, &rfds, &wfds, NULL, NULL);
421		if (ret == -1) {
422			if (errno == EINTR)
423				continue;
424			KEEP_ERRNO((void)pidfile_remove(pfh));
425			pjdlog_exit(EX_OSERR, "select() failed");
426		}
427
428#define	ISSET_FD(conn)	\
429	(FD_ISSET((fd = proto_descriptor(conn)), &rfds) || FD_ISSET(fd, &wfds))
430		if (ISSET_FD(cfg->hc_controlconn))
431			control_handle(cfg);
432		if (ISSET_FD(cfg->hc_listenconn))
433			listen_accept();
434#undef	ISSET_FD
435	}
436}
437
438int
439main(int argc, char *argv[])
440{
441	const char *pidfile;
442	pid_t otherpid;
443	bool foreground;
444	int debuglevel;
445
446	g_gate_load();
447
448	foreground = false;
449	debuglevel = 0;
450	pidfile = HASTD_PIDFILE;
451
452	for (;;) {
453		int ch;
454
455		ch = getopt(argc, argv, "c:dFhP:");
456		if (ch == -1)
457			break;
458		switch (ch) {
459		case 'c':
460			cfgpath = optarg;
461			break;
462		case 'd':
463			debuglevel++;
464			break;
465		case 'F':
466			foreground = true;
467			break;
468		case 'P':
469			pidfile = optarg;
470			break;
471		case 'h':
472		default:
473			usage();
474		}
475	}
476	argc -= optind;
477	argv += optind;
478
479	pjdlog_debug_set(debuglevel);
480
481	pfh = pidfile_open(pidfile, 0600, &otherpid);
482	if (pfh == NULL) {
483		if (errno == EEXIST) {
484			pjdlog_exitx(EX_TEMPFAIL,
485			    "Another hastd is already running, pid: %jd.",
486			    (intmax_t)otherpid);
487		}
488		/* If we cannot create pidfile from other reasons, only warn. */
489		pjdlog_errno(LOG_WARNING, "Cannot open or create pidfile");
490	}
491
492	cfg = yy_config_parse(cfgpath);
493	assert(cfg != NULL);
494
495	signal(SIGHUP, sighandler);
496	signal(SIGCHLD, sighandler);
497
498	/* Listen on control address. */
499	if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
500		KEEP_ERRNO((void)pidfile_remove(pfh));
501		pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
502		    cfg->hc_controladdr);
503	}
504	/* Listen for remote connections. */
505	if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) {
506		KEEP_ERRNO((void)pidfile_remove(pfh));
507		pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
508		    cfg->hc_listenaddr);
509	}
510
511	if (!foreground) {
512		if (daemon(0, 0) < 0) {
513			KEEP_ERRNO((void)pidfile_remove(pfh));
514			pjdlog_exit(EX_OSERR, "Unable to daemonize");
515		}
516
517		/* Start logging to syslog. */
518		pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
519
520		/* Write PID to a file. */
521		if (pidfile_write(pfh) < 0) {
522			pjdlog_errno(LOG_WARNING,
523			    "Unable to write PID to a file");
524		}
525	}
526
527	main_loop();
528
529	exit(0);
530}
531