hastd.c revision 207371 
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 207371 2010-04-29 15:36:32Z 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	/* Error in setting timeout is not critical, but why should it fail? */
191	if (proto_timeout(conn, HAST_TIMEOUT) < 0)
192		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
193
194	nvin = nvout = nverr = NULL;
195
196	/*
197	 * Before receiving any data see if remote host have access to any
198	 * resource.
199	 */
200	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
201		if (proto_address_match(conn, res->hr_remoteaddr))
202			break;
203	}
204	if (res == NULL) {
205		pjdlog_error("Client %s isn't known.", raddr);
206		goto close;
207	}
208	/* Ok, remote host can access at least one resource. */
209
210	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
211		pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
212		    raddr);
213		goto close;
214	}
215
216	resname = nv_get_string(nvin, "resource");
217	if (resname == NULL) {
218		pjdlog_error("No 'resource' field in the header received from %s.",
219		    raddr);
220		goto close;
221	}
222	pjdlog_debug(2, "%s: resource=%s", raddr, resname);
223	token = nv_get_uint8_array(nvin, &size, "token");
224	/*
225	 * NULL token means that this is first conection.
226	 */
227	if (token != NULL && size != sizeof(res->hr_token)) {
228		pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
229		    raddr, sizeof(res->hr_token), size);
230		goto close;
231	}
232
233	/*
234	 * From now on we want to send errors to the remote node.
235	 */
236	nverr = nv_alloc();
237
238	/* Find resource related to this connection. */
239	TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
240		if (strcmp(resname, res->hr_name) == 0)
241			break;
242	}
243	/* Have we found the resource? */
244	if (res == NULL) {
245		pjdlog_error("No resource '%s' as requested by %s.",
246		    resname, raddr);
247		nv_add_stringf(nverr, "errmsg", "Resource not configured.");
248		goto fail;
249	}
250
251	/* Now that we know resource name setup log prefix. */
252	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
253
254	/* Does the remote host have access to this resource? */
255	if (!proto_address_match(conn, res->hr_remoteaddr)) {
256		pjdlog_error("Client %s has no access to the resource.", raddr);
257		nv_add_stringf(nverr, "errmsg", "No access to the resource.");
258		goto fail;
259	}
260	/* Is the resource marked as secondary? */
261	if (res->hr_role != HAST_ROLE_SECONDARY) {
262		pjdlog_error("We act as %s for the resource and not as %s as requested by %s.",
263		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
264		    raddr);
265		nv_add_stringf(nverr, "errmsg",
266		    "Remote node acts as %s for the resource and not as %s.",
267		    role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
268		goto fail;
269	}
270	/* Does token (if exists) match? */
271	if (token != NULL && memcmp(token, res->hr_token,
272	    sizeof(res->hr_token)) != 0) {
273		pjdlog_error("Token received from %s doesn't match.", raddr);
274		nv_add_stringf(nverr, "errmsg", "Toke doesn't match.");
275		goto fail;
276	}
277	/*
278	 * If there is no token, but we have half-open connection
279	 * (only remotein) or full connection (worker process is running)
280	 * we have to cancel those and accept the new connection.
281	 */
282	if (token == NULL) {
283		assert(res->hr_remoteout == NULL);
284		pjdlog_debug(1, "Initial connection from %s.", raddr);
285		if (res->hr_workerpid != 0) {
286			assert(res->hr_remotein == NULL);
287			pjdlog_debug(1,
288			    "Worker process exists (pid=%u), stopping it.",
289			    (unsigned int)res->hr_workerpid);
290			/* Stop child process. */
291			if (kill(res->hr_workerpid, SIGINT) < 0) {
292				pjdlog_errno(LOG_ERR,
293				    "Unable to stop worker process (pid=%u)",
294				    (unsigned int)res->hr_workerpid);
295				/*
296				 * Other than logging the problem we
297				 * ignore it - nothing smart to do.
298				 */
299			}
300			/* Wait for it to exit. */
301			else if ((pid = waitpid(res->hr_workerpid,
302			    &status, 0)) != res->hr_workerpid) {
303				pjdlog_errno(LOG_ERR,
304				    "Waiting for worker process (pid=%u) failed",
305				    (unsigned int)res->hr_workerpid);
306				/* See above. */
307			} else if (WEXITSTATUS(status) != 0) {
308				pjdlog_error("Worker process (pid=%u) exited ungracefully: status=%d.",
309				    (unsigned int)res->hr_workerpid,
310				    WEXITSTATUS(status));
311				/* See above. */
312			} else {
313				pjdlog_debug(1,
314				    "Worker process (pid=%u) exited gracefully.",
315				    (unsigned int)res->hr_workerpid);
316			}
317			res->hr_workerpid = 0;
318		} else if (res->hr_remotein != NULL) {
319			char oaddr[256];
320
321			proto_remote_address(conn, oaddr, sizeof(oaddr));
322			pjdlog_debug(1,
323			    "Canceling half-open connection from %s on connection from %s.",
324			    oaddr, raddr);
325			proto_close(res->hr_remotein);
326			res->hr_remotein = NULL;
327		}
328	}
329
330	/*
331	 * Checks and cleanups are done.
332	 */
333
334	if (token == NULL) {
335		arc4random_buf(res->hr_token, sizeof(res->hr_token));
336		nvout = nv_alloc();
337		nv_add_uint8_array(nvout, res->hr_token,
338		    sizeof(res->hr_token), "token");
339		if (nv_error(nvout) != 0) {
340			pjdlog_common(LOG_ERR, 0, nv_error(nvout),
341			    "Unable to prepare return header for %s", raddr);
342			nv_add_stringf(nverr, "errmsg",
343			    "Remote node was unable to prepare return header: %s.",
344			    strerror(nv_error(nvout)));
345			goto fail;
346		}
347		if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
348			int error = errno;
349
350			pjdlog_errno(LOG_ERR, "Unable to send response to %s",
351			    raddr);
352			nv_add_stringf(nverr, "errmsg",
353			    "Remote node was unable to send response: %s.",
354			    strerror(error));
355			goto fail;
356		}
357		res->hr_remotein = conn;
358		pjdlog_debug(1, "Incoming connection from %s configured.",
359		    raddr);
360	} else {
361		res->hr_remoteout = conn;
362		pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
363		hastd_secondary(res, nvin);
364	}
365	nv_free(nvin);
366	nv_free(nvout);
367	nv_free(nverr);
368	pjdlog_prefix_set("%s", "");
369	return;
370fail:
371	if (nv_error(nverr) != 0) {
372		pjdlog_common(LOG_ERR, 0, nv_error(nverr),
373		    "Unable to prepare error header for %s", raddr);
374		goto close;
375	}
376	if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
377		pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
378		goto close;
379	}
380close:
381	if (nvin != NULL)
382		nv_free(nvin);
383	if (nvout != NULL)
384		nv_free(nvout);
385	if (nverr != NULL)
386		nv_free(nverr);
387	proto_close(conn);
388	pjdlog_prefix_set("%s", "");
389}
390
391static void
392main_loop(void)
393{
394	fd_set rfds, wfds;
395	int fd, maxfd, ret;
396
397	for (;;) {
398		if (sigchld_received) {
399			sigchld_received = false;
400			child_exit();
401		}
402		if (sighup_received) {
403			sighup_received = false;
404			hastd_reload();
405		}
406
407		maxfd = 0;
408		FD_ZERO(&rfds);
409		FD_ZERO(&wfds);
410
411		/* Setup descriptors for select(2). */
412#define	SETUP_FD(conn)	do {						\
413	fd = proto_descriptor(conn);					\
414	if (fd >= 0) {							\
415		maxfd = fd > maxfd ? fd : maxfd;			\
416		FD_SET(fd, &rfds);					\
417		FD_SET(fd, &wfds);					\
418	}								\
419} while (0)
420		SETUP_FD(cfg->hc_controlconn);
421		SETUP_FD(cfg->hc_listenconn);
422#undef	SETUP_FD
423
424		ret = select(maxfd + 1, &rfds, &wfds, NULL, NULL);
425		if (ret == -1) {
426			if (errno == EINTR)
427				continue;
428			KEEP_ERRNO((void)pidfile_remove(pfh));
429			pjdlog_exit(EX_OSERR, "select() failed");
430		}
431
432#define	ISSET_FD(conn)	\
433	(FD_ISSET((fd = proto_descriptor(conn)), &rfds) || FD_ISSET(fd, &wfds))
434		if (ISSET_FD(cfg->hc_controlconn))
435			control_handle(cfg);
436		if (ISSET_FD(cfg->hc_listenconn))
437			listen_accept();
438#undef	ISSET_FD
439	}
440}
441
442int
443main(int argc, char *argv[])
444{
445	const char *pidfile;
446	pid_t otherpid;
447	bool foreground;
448	int debuglevel;
449
450	g_gate_load();
451
452	foreground = false;
453	debuglevel = 0;
454	pidfile = HASTD_PIDFILE;
455
456	for (;;) {
457		int ch;
458
459		ch = getopt(argc, argv, "c:dFhP:");
460		if (ch == -1)
461			break;
462		switch (ch) {
463		case 'c':
464			cfgpath = optarg;
465			break;
466		case 'd':
467			debuglevel++;
468			break;
469		case 'F':
470			foreground = true;
471			break;
472		case 'P':
473			pidfile = optarg;
474			break;
475		case 'h':
476		default:
477			usage();
478		}
479	}
480	argc -= optind;
481	argv += optind;
482
483	pjdlog_debug_set(debuglevel);
484
485	pfh = pidfile_open(pidfile, 0600, &otherpid);
486	if (pfh == NULL) {
487		if (errno == EEXIST) {
488			pjdlog_exitx(EX_TEMPFAIL,
489			    "Another hastd is already running, pid: %jd.",
490			    (intmax_t)otherpid);
491		}
492		/* If we cannot create pidfile from other reasons, only warn. */
493		pjdlog_errno(LOG_WARNING, "Cannot open or create pidfile");
494	}
495
496	cfg = yy_config_parse(cfgpath);
497	assert(cfg != NULL);
498
499	signal(SIGHUP, sighandler);
500	signal(SIGCHLD, sighandler);
501
502	/* Listen on control address. */
503	if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
504		KEEP_ERRNO((void)pidfile_remove(pfh));
505		pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
506		    cfg->hc_controladdr);
507	}
508	/* Listen for remote connections. */
509	if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) {
510		KEEP_ERRNO((void)pidfile_remove(pfh));
511		pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
512		    cfg->hc_listenaddr);
513	}
514
515	if (!foreground) {
516		if (daemon(0, 0) < 0) {
517			KEEP_ERRNO((void)pidfile_remove(pfh));
518			pjdlog_exit(EX_OSERR, "Unable to daemonize");
519		}
520
521		/* Start logging to syslog. */
522		pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
523
524		/* Write PID to a file. */
525		if (pidfile_write(pfh) < 0) {
526			pjdlog_errno(LOG_WARNING,
527			    "Unable to write PID to a file");
528		}
529	}
530
531	main_loop();
532
533	exit(0);
534}
535