xref: /titanic_41/usr/src/cmd/fm/fmd/common/fmd.c (revision b02e9a2d4d2071d770e5aa9ae8f83f2bbe1f2ced)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/utsname.h>
31 #include <sys/param.h>
32 #include <sys/systeminfo.h>
33 #include <sys/fm/util.h>
34 
35 #include <smbios.h>
36 #include <limits.h>
37 #include <unistd.h>
38 #include <signal.h>
39 #include <stdlib.h>
40 #include <stdio.h>
41 #include <door.h>
42 
43 #include <fmd_conf.h>
44 #include <fmd_dispq.h>
45 #include <fmd_timerq.h>
46 #include <fmd_subr.h>
47 #include <fmd_error.h>
48 #include <fmd_module.h>
49 #include <fmd_thread.h>
50 #include <fmd_alloc.h>
51 #include <fmd_string.h>
52 #include <fmd_builtin.h>
53 #include <fmd_ustat.h>
54 #include <fmd_protocol.h>
55 #include <fmd_scheme.h>
56 #include <fmd_asru.h>
57 #include <fmd_case.h>
58 #include <fmd_log.h>
59 #include <fmd_idspace.h>
60 #include <fmd_rpc.h>
61 #include <fmd_dr.h>
62 #include <fmd_topo.h>
63 #include <fmd_xprt.h>
64 #include <fmd_ctl.h>
65 #include <sys/openpromio.h>
66 #include <libdevinfo.h>
67 
68 #include <fmd.h>
69 
70 extern const nv_alloc_ops_t fmd_nv_alloc_ops;	/* see fmd_nv.c */
71 
72 const char _fmd_version[] = "1.2";		/* daemon version string */
73 static char _fmd_plat[MAXNAMELEN];		/* native platform string */
74 static char _fmd_isa[MAXNAMELEN];		/* native instruction set */
75 static struct utsname _fmd_uts;			/* native uname(2) info */
76 static char _fmd_csn[MAXNAMELEN];		/* chassis serial number */
77 static char _fmd_prod[MAXNAMELEN];		/* product name string */
78 
79 /*
80  * Note: the configuration file path is ordered from most common to most host-
81  * specific because new conf files are merged/override previous ones.  The
82  * module paths are in the opposite order, from most specific to most common,
83  * because once a module is loaded fmd will not try to load over the same name.
84  */
85 
86 static const char _fmd_conf_path[] =
87 	"%r/usr/lib/fm/fmd:"
88 	"%r/usr/platform/%m/lib/fm/fmd:"
89 	"%r/usr/platform/%i/lib/fm/fmd:"
90 	"%r/etc/fm/fmd";
91 
92 static const char _fmd_agent_path[] =
93 	"%r/usr/platform/%i/lib/fm/fmd/agents:"
94 	"%r/usr/platform/%m/lib/fm/fmd/agents:"
95 	"%r/usr/lib/fm/fmd/agents";
96 
97 static const char _fmd_plugin_path[] =
98 	"%r/usr/platform/%i/lib/fm/fmd/plugins:"
99 	"%r/usr/platform/%m/lib/fm/fmd/plugins:"
100 	"%r/usr/lib/fm/fmd/plugins";
101 
102 static const char _fmd_scheme_path[] =
103 	"usr/lib/fm/fmd/schemes";
104 
105 static const fmd_conf_mode_t _fmd_cerror_modes[] = {
106 	{ "unload", "unload offending client module", FMD_CERROR_UNLOAD },
107 	{ "stop", "stop daemon for debugger attach", FMD_CERROR_STOP },
108 	{ "abort", "abort daemon and force core dump", FMD_CERROR_ABORT },
109 	{ NULL, NULL, 0 }
110 };
111 
112 static const fmd_conf_mode_t _fmd_dbout_modes[] = {
113 	{ "stderr", "send debug messages to stderr", FMD_DBOUT_STDERR },
114 	{ "syslog", "send debug messages to syslog", FMD_DBOUT_SYSLOG },
115 	{ NULL, NULL, 0 }
116 };
117 
118 static const fmd_conf_mode_t _fmd_debug_modes[] = {
119 	{ "help", "display debugging modes and exit", FMD_DBG_HELP },
120 	{ "mod", "debug module load/unload/locking", FMD_DBG_MOD },
121 	{ "disp", "debug dispatch queue processing", FMD_DBG_DISP },
122 	{ "xprt", "debug transport-specific routines", FMD_DBG_XPRT },
123 	{ "evt", "debug event subsystem routines", FMD_DBG_EVT },
124 	{ "log", "debug log subsystem routines", FMD_DBG_LOG },
125 	{ "tmr", "debug timer subsystem routines", FMD_DBG_TMR },
126 	{ "fmri", "debug fmri subsystem routines", FMD_DBG_FMRI },
127 	{ "asru", "debug asru subsystem routines", FMD_DBG_ASRU },
128 	{ "case", "debug case subsystem routines", FMD_DBG_CASE },
129 	{ "ckpt", "debug checkpoint routines", FMD_DBG_CKPT },
130 	{ "rpc", "debug rpc service routines", FMD_DBG_RPC },
131 	{ "trace", "display matching trace calls", FMD_DBG_TRACE },
132 	{ "all", "enable all available debug modes", FMD_DBG_ALL },
133 	{ NULL, NULL, 0 }
134 };
135 
136 static int
137 fmd_cerror_set(fmd_conf_param_t *pp, const char *value)
138 {
139 	return (fmd_conf_mode_set(_fmd_cerror_modes, pp, value));
140 }
141 
142 static int
143 fmd_dbout_set(fmd_conf_param_t *pp, const char *value)
144 {
145 	return (fmd_conf_mode_set(_fmd_dbout_modes, pp, value));
146 }
147 
148 static int
149 fmd_debug_set(fmd_conf_param_t *pp, const char *value)
150 {
151 	int err = fmd_conf_mode_set(_fmd_debug_modes, pp, value);
152 
153 	if (err == 0)
154 		fmd.d_fmd_debug = pp->cp_value.cpv_num;
155 
156 	return (err);
157 }
158 
159 static int
160 fmd_trmode_set(fmd_conf_param_t *pp, const char *value)
161 {
162 	fmd_tracebuf_f *func;
163 
164 	if (strcasecmp(value, "none") == 0)
165 		func = fmd_trace_none;
166 	else if (strcasecmp(value, "lite") == 0)
167 		func = fmd_trace_lite;
168 	else if (strcasecmp(value, "full") == 0)
169 		func = fmd_trace_full;
170 	else
171 		return (fmd_set_errno(EFMD_CONF_INVAL));
172 
173 	fmd.d_thr_trace = (void (*)())func;
174 	pp->cp_value.cpv_ptr = (void *)func;
175 	return (0);
176 }
177 
178 static void
179 fmd_trmode_get(const fmd_conf_param_t *pp, void *ptr)
180 {
181 	*((void **)ptr) = pp->cp_value.cpv_ptr;
182 }
183 
184 static int
185 fmd_clkmode_set(fmd_conf_param_t *pp, const char *value)
186 {
187 	const fmd_timeops_t *ops;
188 
189 	if (strcasecmp(value, "native") == 0)
190 		ops = &fmd_timeops_native;
191 	else if (strcasecmp(value, "simulated") == 0)
192 		ops = &fmd_timeops_simulated;
193 	else
194 		return (fmd_set_errno(EFMD_CONF_INVAL));
195 
196 	fmd.d_clockops = ops;
197 	pp->cp_value.cpv_ptr = (void *)ops;
198 	return (0);
199 }
200 
201 static void
202 fmd_clkmode_get(const fmd_conf_param_t *pp, void *ptr)
203 {
204 	*((void **)ptr) = pp->cp_value.cpv_ptr;
205 }
206 
207 static const fmd_conf_ops_t fmd_cerror_ops = {
208 	fmd_cerror_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop
209 };
210 
211 static const fmd_conf_ops_t fmd_dbout_ops = {
212 	fmd_dbout_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop
213 };
214 
215 static const fmd_conf_ops_t fmd_debug_ops = {
216 	fmd_debug_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop
217 };
218 
219 static const fmd_conf_ops_t fmd_trmode_ops = {
220 	fmd_trmode_set, fmd_trmode_get, fmd_conf_notsup, fmd_conf_nop
221 };
222 
223 static const fmd_conf_ops_t fmd_clkmode_ops = {
224 	fmd_clkmode_set, fmd_clkmode_get, fmd_conf_notsup, fmd_conf_nop
225 };
226 
227 static const fmd_conf_formal_t _fmd_conf[] = {
228 { "agent.path", &fmd_conf_path, _fmd_agent_path }, /* path for agents */
229 { "alloc_msecs", &fmd_conf_uint32, "10" },	/* msecs before alloc retry */
230 { "alloc_tries", &fmd_conf_uint32, "3" },	/* max # of alloc retries */
231 { "chassis", &fmd_conf_string, _fmd_csn },	/* chassis serial number */
232 { "ckpt.dir", &fmd_conf_string, "var/fm/fmd/ckpt" }, /* ckpt directory path */
233 { "ckpt.dirmode", &fmd_conf_int32, "0700" },	/* ckpt directory perm mode */
234 { "ckpt.mode", &fmd_conf_int32, "0400" },	/* ckpt file perm mode */
235 { "ckpt.restore", &fmd_conf_bool, "true" },	/* restore checkpoints? */
236 { "ckpt.save", &fmd_conf_bool, "true" },	/* save checkpoints? */
237 { "ckpt.zero", &fmd_conf_bool, "false" },	/* zero checkpoints on start? */
238 { "client.buflim", &fmd_conf_size, "10m" },	/* client buffer space limit */
239 { "client.dbout", &fmd_dbout_ops, NULL },	/* client debug output sinks */
240 { "client.debug", &fmd_conf_bool, NULL },	/* client debug enable */
241 { "client.error", &fmd_cerror_ops, "unload" },	/* client error policy */
242 { "client.memlim", &fmd_conf_size, "10m" },	/* client allocation limit */
243 { "client.evqlim", &fmd_conf_uint32, "256" },	/* client event queue limit */
244 { "client.thrlim", &fmd_conf_uint32, "8" },	/* client aux thread limit */
245 { "client.thrsig", &fmd_conf_signal, "SIGUSR1" }, /* fmd_thr_signal() value */
246 { "client.tmrlim", &fmd_conf_uint32, "1024" },	/* client pending timer limit */
247 { "client.xprtlim", &fmd_conf_uint32, "256" },	/* client transport limit */
248 { "client.xprtlog", &fmd_conf_bool, NULL },	/* client transport logging? */
249 { "client.xprtqlim", &fmd_conf_uint32, "1024" }, /* client transport queue li */
250 { "clock", &fmd_clkmode_ops, "native" },	/* clock operation mode */
251 { "conf_path", &fmd_conf_path, _fmd_conf_path }, /* root config file path */
252 { "conf_file", &fmd_conf_string, "fmd.conf" },	/* root config file name */
253 { "core", &fmd_conf_bool, "false" },		/* force core dump on quit */
254 { "dbout", &fmd_dbout_ops, NULL },		/* daemon debug output sinks */
255 { "debug", &fmd_debug_ops, NULL },		/* daemon debugging flags */
256 { "dictdir", &fmd_conf_string, "usr/lib/fm/dict" }, /* default diagcode dir */
257 { "domain", &fmd_conf_string, NULL },		/* domain id for de auth */
258 { "fakenotpresent", &fmd_conf_bool, "false" },	/* simulate rsrc not present */
259 { "fg", &fmd_conf_bool, "false" },		/* run daemon in foreground */
260 { "gc_interval", &fmd_conf_time, "1d" },	/* garbage collection intvl */
261 { "ids.avg", &fmd_conf_uint32, "4" },		/* desired idspace chain len */
262 { "ids.max", &fmd_conf_uint32, "1024" },	/* maximum idspace buckets */
263 { "isaname", &fmd_conf_string, _fmd_isa },	/* instruction set (uname -p) */
264 { "log.creator", &fmd_conf_string, "fmd" },	/* exacct log creator string */
265 { "log.error", &fmd_conf_string, "var/fm/fmd/errlog" }, /* error log path */
266 { "log.fault", &fmd_conf_string, "var/fm/fmd/fltlog" }, /* fault log path */
267 { "log.minfree", &fmd_conf_size, "2m" },	/* min log fsys free space */
268 { "log.rsrc", &fmd_conf_string, "var/fm/fmd/rsrc" }, /* asru log dir path */
269 { "log.tryrotate", &fmd_conf_uint32, "10" },	/* max log rotation attempts */
270 { "log.waitrotate", &fmd_conf_time, "200ms" },	/* log rotation retry delay */
271 { "log.xprt", &fmd_conf_string, "var/fm/fmd/xprt" }, /* transport log dir */
272 { "machine", &fmd_conf_string, _fmd_uts.machine }, /* machine name (uname -m) */
273 { "nodiagcode", &fmd_conf_string, "-" },	/* diagcode to use if error */
274 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */
275 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */
276 { "platform", &fmd_conf_string, _fmd_plat },	/* platform string (uname -i) */
277 { "plugin.close", &fmd_conf_bool, "true" },	/* dlclose plugins on fini */
278 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */
279 { "product", &fmd_conf_string, _fmd_prod },	/* product name string */
280 { "rootdir", &fmd_conf_string, "" },		/* root directory for paths */
281 { "rpc.adm.path", &fmd_conf_string, NULL },	/* FMD_ADM rendezvous file */
282 { "rpc.adm.prog", &fmd_conf_uint32, "100169" },	/* FMD_ADM rpc program num */
283 { "rpc.api.path", &fmd_conf_string, NULL },	/* FMD_API rendezvous file */
284 { "rpc.api.prog", &fmd_conf_uint32, "100170" },	/* FMD_API rpc program num */
285 { "rpc.rcvsize", &fmd_conf_size, "128k" },	/* rpc receive buffer size */
286 { "rpc.sndsize", &fmd_conf_size, "128k" },	/* rpc send buffer size */
287 { "rsrc.age", &fmd_conf_time, "30d" },		/* max age of old rsrc log */
288 { "rsrc.zero", &fmd_conf_bool, "false" },	/* zero rsrc cache on start? */
289 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */
290 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */
291 { "self.dict", &fmd_conf_list, "FMD.dict" },	/* self-diag dictionary list */
292 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */
293 { "strbuckets", &fmd_conf_uint32, "211" },	/* size of string hashes */
294 #ifdef DEBUG
295 { "trace.mode", &fmd_trmode_ops, "full" },	/* trace mode: none/lite/full */
296 #else
297 { "trace.mode", &fmd_trmode_ops, "lite" },	/* trace mode: none/lite/full */
298 #endif
299 { "trace.recs", &fmd_conf_uint32, "128" },	/* trace records per thread */
300 { "trace.frames", &fmd_conf_uint32, "16" },	/* max trace rec stack frames */
301 { "uuidlen", &fmd_conf_uint32, "36" },		/* UUID ASCII string length */
302 { "xprt.ttl", &fmd_conf_uint8, "1" },		/* default event time-to-live */
303 };
304 
305 /*
306  * Statistics maintained by fmd itself on behalf of various global subsystems.
307  * NOTE: FMD_TYPE_STRING statistics should not be used here.  If they are
308  * required in the future, the FMD_ADM_MODGSTAT service routine must change.
309  */
310 static fmd_statistics_t _fmd_stats = {
311 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" },
312 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" },
313 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" },
314 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" },
315 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" },
316 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" },
317 { "topo.gen", FMD_TYPE_UINT64, "topology snapshot generation" },
318 { "topo.drgen", FMD_TYPE_UINT64, "current topology DR generation number" },
319 };
320 
321 void
322 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf)
323 {
324 	fmd_conf_path_t *pap;
325 	char file[PATH_MAX];
326 	const char *name;
327 	fmd_stat_t *sp;
328 	int i;
329 
330 	smbios_hdl_t *shp;
331 	smbios_system_t s1;
332 	smbios_info_t s2;
333 	id_t id;
334 
335 	di_prom_handle_t promh = DI_PROM_HANDLE_NIL;
336 	di_node_t rooth = DI_NODE_NIL;
337 	char *bufp;
338 
339 	(void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat));
340 	(void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa));
341 	(void) uname(&_fmd_uts);
342 
343 	if ((shp = smbios_open(NULL, SMB_VERSION, 0, NULL)) != NULL) {
344 		if ((id = smbios_info_system(shp, &s1)) != SMB_ERR &&
345 		    smbios_info_common(shp, id, &s2) != SMB_ERR) {
346 			(void) strlcpy(_fmd_prod, s2.smbi_product, MAXNAMELEN);
347 			(void) strlcpy(_fmd_csn, s2.smbi_serial, MAXNAMELEN);
348 		}
349 		smbios_close(shp);
350 	} else if ((rooth = di_init("/", DINFOPROP)) != DI_NODE_NIL &&
351 	    (promh = di_prom_init()) != DI_PROM_HANDLE_NIL) {
352 		if (di_prom_prop_lookup_bytes(promh, rooth, "chassis-sn",
353 		    (unsigned char **)&bufp) != -1) {
354 			(void) strlcpy(_fmd_csn, bufp, MAXNAMELEN);
355 		}
356 	}
357 
358 	if (promh != DI_PROM_HANDLE_NIL)
359 		di_prom_fini(promh);
360 	if (rooth != DI_NODE_NIL)
361 		di_fini(rooth);
362 
363 	bzero(dp, sizeof (fmd_t));
364 
365 	dp->d_version = _fmd_version;
366 	dp->d_pname = fmd_strbasename(arg0);
367 	dp->d_pid = getpid();
368 
369 	if (pthread_key_create(&dp->d_key, NULL) != 0)
370 		fmd_error(EFMD_EXIT, "failed to create pthread key");
371 
372 	(void) pthread_mutex_init(&dp->d_xprt_lock, NULL);
373 	(void) pthread_mutex_init(&dp->d_err_lock, NULL);
374 	(void) pthread_mutex_init(&dp->d_thr_lock, NULL);
375 	(void) pthread_mutex_init(&dp->d_mod_lock, NULL);
376 	(void) pthread_mutex_init(&dp->d_stats_lock, NULL);
377 	(void) pthread_mutex_init(&dp->d_topo_lock, NULL);
378 	(void) pthread_rwlock_init(&dp->d_log_lock, NULL);
379 
380 	/*
381 	 * A small number of properties must be set manually before we open
382 	 * the root configuration file.  These include any settings for our
383 	 * memory allocator and path expansion token values, because these
384 	 * values are needed by the routines in fmd_conf.c itself.  After
385 	 * the root configuration file is processed, we reset these properties
386 	 * based upon the latest values from the configuration file.
387 	 */
388 	dp->d_alloc_msecs = 10;
389 	dp->d_alloc_tries = 3;
390 	dp->d_str_buckets = 211;
391 
392 	dp->d_rootdir = root ? root : "";
393 	dp->d_platform = _fmd_plat;
394 	dp->d_machine = _fmd_uts.machine;
395 	dp->d_isaname = _fmd_isa;
396 
397 	dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) /
398 	    sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER);
399 
400 	if (dp->d_conf == NULL) {
401 		fmd_error(EFMD_EXIT,
402 		    "failed to load required configuration properties\n");
403 	}
404 
405 	(void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs);
406 	(void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries);
407 	(void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets);
408 
409 	(void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform);
410 	(void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine);
411 	(void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname);
412 
413 	/*
414 	 * Manually specified rootdirs override config files, so only update
415 	 * d_rootdir based on the config files we parsed if no 'root' was set.
416 	 */
417 	if (root == NULL)
418 		(void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir);
419 	else
420 		(void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir);
421 
422 	/*
423 	 * Once the base conf file properties are loaded, lookup the values
424 	 * of $conf_path and $conf_file and merge in any other conf files.
425 	 */
426 	(void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap);
427 	(void) fmd_conf_getprop(dp->d_conf, "conf_file", &name);
428 
429 	for (i = 0; i < pap->cpa_argc; i++) {
430 		(void) snprintf(file, sizeof (file),
431 		    "%s/%s", pap->cpa_argv[i], name);
432 		if (access(file, F_OK) == 0)
433 			fmd_conf_merge(dp->d_conf, file);
434 	}
435 
436 	/*
437 	 * Update the value of fmd.d_fg based on "fg".  We cache this property
438 	 * because it must be accessed deep within fmd at fmd_verror() time.
439 	 * Update any other properties that must be cached for performance.
440 	 */
441 	(void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg);
442 	(void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl);
443 
444 	/*
445 	 * Initialize our custom libnvpair allocator and create an nvlist for
446 	 * authority elements corresponding to this instance of the daemon.
447 	 */
448 	(void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops);
449 	dp->d_auth = fmd_protocol_authority();
450 
451 	/*
452 	 * The fmd_module_t for the root module must be created manually.  Most
453 	 * of it remains unused and zero, except for the few things we fill in.
454 	 */
455 	dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP);
456 	dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP);
457 	dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod);
458 
459 	fmd_list_append(&dp->d_mod_list, dp->d_rmod);
460 	fmd_module_hold(dp->d_rmod);
461 
462 	(void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL);
463 	(void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL);
464 	(void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL);
465 
466 	dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self());
467 	dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP);
468 	dp->d_rmod->mod_ustat = fmd_ustat_create();
469 
470 	if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0)
471 		fmd_error(EFMD_EXIT, "failed to attach main thread key");
472 
473 	if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert(
474 	    dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) /
475 	    sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL)
476 		fmd_error(EFMD_EXIT, "failed to initialize statistics");
477 
478 	(void) pthread_mutex_lock(&dp->d_rmod->mod_lock);
479 	dp->d_rmod->mod_flags |= FMD_MOD_INIT;
480 	(void) pthread_mutex_unlock(&dp->d_rmod->mod_lock);
481 
482 	/*
483 	 * In addition to inserting the _fmd_stats collection of program-wide
484 	 * statistics, we also insert a statistic named after each of our
485 	 * errors and update these counts in fmd_verror() (see fmd_subr.c).
486 	 */
487 	dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) *
488 	    (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP);
489 
490 	for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) {
491 		(void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s",
492 		    strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1);
493 		sp->fmds_type = FMD_TYPE_UINT64;
494 	}
495 
496 	(void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC,
497 	    EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL);
498 }
499 
500 void
501 fmd_destroy(fmd_t *dp)
502 {
503 	fmd_module_t *mp;
504 	fmd_case_t *cp;
505 	int core;
506 
507 	(void) fmd_conf_getprop(fmd.d_conf, "core", &core);
508 
509 	fmd_rpc_fini();
510 	fmd_dr_fini();
511 
512 	if (dp->d_xprt_ids != NULL)
513 		fmd_xprt_suspend_all();
514 
515 	/*
516 	 * Unload the self-diagnosis module first.  This ensures that it does
517 	 * not get confused as we start unloading other modules, etc.  We must
518 	 * hold the dispq lock as a writer while doing so since it uses d_self.
519 	 */
520 	if (dp->d_self != NULL) {
521 		fmd_module_t *self;
522 
523 		(void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock);
524 		self = dp->d_self;
525 		dp->d_self = NULL;
526 		(void) pthread_rwlock_unlock(&dp->d_disp->dq_lock);
527 
528 		fmd_module_unload(self);
529 		fmd_module_rele(self);
530 	}
531 
532 	/*
533 	 * Unload modules in reverse order *except* for the root module, which
534 	 * is first in the list.  This allows it to keep its thread and trace.
535 	 */
536 	for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) {
537 		fmd_module_unload(mp);
538 		mp = fmd_list_prev(mp);
539 	}
540 
541 	if (dp->d_mod_hash != NULL) {
542 		fmd_modhash_destroy(dp->d_mod_hash);
543 		dp->d_mod_hash = NULL;
544 	}
545 
546 	/*
547 	 * Close both log files now that modules are no longer active.  We must
548 	 * set these pointers to NULL in case any subsequent errors occur.
549 	 */
550 	if (dp->d_errlog != NULL) {
551 		fmd_log_rele(dp->d_errlog);
552 		dp->d_errlog = NULL;
553 	}
554 
555 	if (dp->d_fltlog != NULL) {
556 		fmd_log_rele(dp->d_fltlog);
557 		dp->d_fltlog = NULL;
558 	}
559 
560 	/*
561 	 * Now destroy the resource cache: each ASRU contains a case reference,
562 	 * which may in turn contain a pointer to a referenced owning module.
563 	 */
564 	if (dp->d_asrus != NULL) {
565 		fmd_asru_hash_destroy(dp->d_asrus);
566 		dp->d_asrus = NULL;
567 	}
568 
569 	/*
570 	 * Now that all data structures that refer to modules are torn down,
571 	 * no modules should be remaining on the module list except for d_rmod.
572 	 * If we trip one of these assertions, we're missing a rele somewhere.
573 	 */
574 	ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod);
575 	ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod);
576 
577 	/*
578 	 * Now destroy the root module.  We clear its thread key first so any
579 	 * calls to fmd_trace() inside of the module code will be ignored.
580 	 */
581 	(void) pthread_setspecific(dp->d_key, NULL);
582 	fmd_module_lock(dp->d_rmod);
583 
584 	while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL)
585 		fmd_case_discard(cp);
586 
587 	fmd_module_unlock(dp->d_rmod);
588 	fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t));
589 	dp->d_rmod->mod_stats = NULL;
590 
591 	(void) pthread_mutex_lock(&dp->d_rmod->mod_lock);
592 	dp->d_rmod->mod_flags |= FMD_MOD_FINI;
593 	(void) pthread_mutex_unlock(&dp->d_rmod->mod_lock);
594 
595 	fmd_module_rele(dp->d_rmod);
596 	ASSERT(fmd_list_next(&dp->d_mod_list) == NULL);
597 
598 	/*
599 	 * Now destroy the remaining global data structures.  If 'core' was
600 	 * set to true, force a core dump so we can check for memory leaks.
601 	 */
602 	if (dp->d_cases != NULL)
603 		fmd_case_hash_destroy(dp->d_cases);
604 	if (dp->d_disp != NULL)
605 		fmd_dispq_destroy(dp->d_disp);
606 	if (dp->d_timers != NULL)
607 		fmd_timerq_destroy(dp->d_timers);
608 	if (dp->d_schemes != NULL)
609 		fmd_scheme_hash_destroy(dp->d_schemes);
610 	if (dp->d_xprt_ids != NULL)
611 		fmd_idspace_destroy(dp->d_xprt_ids);
612 
613 	if (dp->d_errstats != NULL) {
614 		fmd_free(dp->d_errstats,
615 		    sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN));
616 	}
617 
618 	if (dp->d_conf != NULL)
619 		fmd_conf_close(dp->d_conf);
620 
621 	fmd_topo_fini();
622 
623 	nvlist_free(dp->d_auth);
624 	(void) nv_alloc_fini(&dp->d_nva);
625 	dp->d_clockops->fto_fini(dp->d_clockptr);
626 
627 	(void) pthread_key_delete(dp->d_key);
628 	bzero(dp, sizeof (fmd_t));
629 
630 	if (core)
631 		fmd_panic("forcing core dump at user request\n");
632 }
633 
634 /*ARGSUSED*/
635 static void
636 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt)
637 {
638 	hrtime_t delta;
639 
640 	if (id != 0) {
641 		TRACE((FMD_DBG_MOD, "garbage collect start"));
642 		fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc);
643 		TRACE((FMD_DBG_MOD, "garbage collect end"));
644 
645 		(void) pthread_rwlock_rdlock(&dp->d_log_lock);
646 		fmd_log_update(dp->d_errlog);
647 		(void) pthread_rwlock_unlock(&dp->d_log_lock);
648 	}
649 
650 	(void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta);
651 	(void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids,
652 	    (fmd_timer_f *)fmd_gc, dp, NULL, delta);
653 }
654 
655 /*
656  * Events are committed to the errlog after cases are checkpointed.  If fmd
657  * crashes before an event is ever associated with a module, this function will
658  * be called to replay it to all subscribers.  If fmd crashes in between the
659  * subscriber checkpointing and committing the event in the error log, the
660  * module will have seen the event and we don't want to replay it.  So we look
661  * for the event in all modules and transition it to the proper state.  If
662  * it is found, we commit it to the error log and do not replay it.  The in-
663  * memory case search used by fmd_module_contains() et al isn't particularly
664  * efficient, but it is faster than doing read i/o's on every case event to
665  * check their status or write i/o's on every event to replay to update states.
666  * We can improve the efficiency of this lookup algorithm later if necessary.
667  */
668 /*ARGSUSED*/
669 static void
670 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp)
671 {
672 	fmd_module_t *mp;
673 	fmd_stat_t *sp;
674 
675 	(void) pthread_mutex_lock(&dp->d_mod_lock);
676 
677 	for (mp = fmd_list_next(&dp->d_mod_list);
678 	    mp != NULL; mp = fmd_list_next(mp)) {
679 		if (fmd_module_contains(mp, ep)) {
680 			fmd_module_hold(mp);
681 			break;
682 		}
683 	}
684 
685 	(void) pthread_mutex_unlock(&dp->d_mod_lock);
686 
687 	if (mp != NULL) {
688 		fmd_event_commit(ep);
689 		fmd_module_rele(mp);
690 		sp = &dp->d_stats->ds_log_partials;
691 	} else {
692 		fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep));
693 		sp = &dp->d_stats->ds_log_replayed;
694 	}
695 
696 	(void) pthread_mutex_lock(&dp->d_stats_lock);
697 	sp->fmds_value.ui64++;
698 	(void) pthread_mutex_unlock(&dp->d_stats_lock);
699 }
700 
701 void
702 fmd_door_server(void *dip)
703 {
704 	fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip);
705 	(void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL);
706 	(void) door_return(NULL, 0, NULL, 0);
707 }
708 
709 /*
710  * Custom door server create callback.  Any fmd services that use doors will
711  * require those threads to have their fmd-specific TSD initialized, etc.
712  */
713 static void
714 fmd_door(door_info_t *dip)
715 {
716 	if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL)
717 		fmd_panic("failed to create server for door %p", (void *)dip);
718 }
719 
720 /*
721  * This signal handler is installed for the client.thrsig signal to be used to
722  * force an auxiliary thread to wake up from a system call and return EINTR in
723  * response to a module's use of fmd_thr_signal().  We also trace the event.
724  */
725 static void
726 fmd_signal(int sig)
727 {
728 	TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig));
729 }
730 
731 void
732 fmd_run(fmd_t *dp, int pfd)
733 {
734 	char *nodc_key[] = { FMD_FLT_NODC, NULL };
735 	char nodc_str[128];
736 	struct sigaction act;
737 
738 	int status = FMD_EXIT_SUCCESS;
739 	const char *name;
740 	fmd_conf_path_t *pap;
741 	fmd_event_t *e;
742 	int dbout;
743 
744 	/*
745 	 * Cache all the current debug property settings in d_fmd_debug,
746 	 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout.  If a given debug mask
747 	 * is non-zero and the corresponding dbout mask is zero, set dbout
748 	 * to a sensible default value based on whether we have daemonized.
749 	 */
750 	(void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout);
751 
752 	if (dp->d_fmd_debug != 0 && dbout == 0)
753 		dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG;
754 	else
755 		dp->d_fmd_dbout = dbout;
756 
757 	(void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug);
758 	(void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout);
759 
760 	if (dp->d_hdl_debug != 0 && dbout == 0)
761 		dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG;
762 	else
763 		dp->d_hdl_dbout = dbout;
764 
765 	/*
766 	 * Initialize remaining major program data structures such as the
767 	 * clock, dispatch queues, log files, module hash collections, etc.
768 	 * This work is done here rather than in fmd_create() to permit the -o
769 	 * command-line option to modify properties after fmd_create() is done.
770 	 */
771 	name = dp->d_rootdir != NULL &&
772 	    *dp->d_rootdir != '\0' ? dp->d_rootdir : NULL;
773 
774 	/*
775 	 * The clock must be initialized before fmd_topo_init() because
776 	 * fmd_topo_update() calls fmd_time_gethrtime().
777 	 */
778 	dp->d_clockptr = dp->d_clockops->fto_init();
779 
780 	fmd_topo_init();
781 
782 	dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX);
783 	fmd_xprt_suspend_all();
784 
785 	(void) door_server_create(fmd_door);
786 	fmd_dr_init();
787 
788 	dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16);
789 	dp->d_timers = fmd_timerq_create();
790 	dp->d_disp = fmd_dispq_create();
791 	dp->d_cases = fmd_case_hash_create();
792 
793 	/*
794 	 * The root module's mod_queue is created with limit zero, making it
795 	 * act like /dev/null; anything inserted here is simply ignored.
796 	 */
797 	dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod,
798 	    &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0);
799 
800 	/*
801 	 * Once our subsystems that use signals have been set up, install the
802 	 * signal handler for the fmd_thr_signal() API.  Verify that the signal
803 	 * being used for this purpose doesn't conflict with something else.
804 	 */
805 	(void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig);
806 
807 	if (sigaction(dp->d_thr_sig, NULL, &act) != 0) {
808 		fmd_error(EFMD_EXIT, "invalid signal selected for "
809 		    "client.thrsig property: %d\n", dp->d_thr_sig);
810 	}
811 
812 	if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) {
813 		fmd_error(EFMD_EXIT, "signal selected for client.thrsig "
814 		    "property is already in use: %d\n", dp->d_thr_sig);
815 	}
816 
817 	act.sa_handler = fmd_signal;
818 	act.sa_flags = 0;
819 
820 	(void) sigemptyset(&act.sa_mask);
821 	(void) sigaction(dp->d_thr_sig, &act, NULL);
822 
823 	(void) fmd_conf_getprop(dp->d_conf, "schemedir", &name);
824 	dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name);
825 
826 	(void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name);
827 	dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name);
828 
829 	(void) fmd_conf_getprop(dp->d_conf, "log.error", &name);
830 	dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR);
831 
832 	(void) fmd_conf_getprop(dp->d_conf, "log.fault", &name);
833 	dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT);
834 
835 	if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL)
836 		fmd_error(EFMD_EXIT, "failed to initialize log files\n");
837 
838 	/*
839 	 * Before loading modules, create an empty control event which will act
840 	 * as a global barrier for module event processing.  Each module we
841 	 * load successfully will insert it at their head of their event queue,
842 	 * and then pause inside of fmd_ctl_rele() after dequeuing the event.
843 	 * This module barrier is required for two reasons:
844 	 *
845 	 * (a) During module loading, the restoration of case checkpoints may
846 	 *    result in a list.* event being recreated for which the intended
847 	 *    subscriber has not yet loaded depending on the load order. Such
848 	 *    events could then result in spurious "no subscriber" errors.
849 	 *
850 	 * (b) During errlog replay, a sequence of errors from a long time ago
851 	 *    may be replayed, and the module may attempt to install relative
852 	 *    timers associated with one or more of these events.  If errlog
853 	 *    replay were "racing" with active module threads, an event E1
854 	 *    that resulted in a relative timer T at time E1 + N nsec could
855 	 *    fire prior to an event E2 being enqueued, even if the relative
856 	 *    time ordering was E1 < E2 < E1 + N, causing mis-diagnosis.
857 	 */
858 	dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL,
859 	    FMD_HRT_NOW, NULL, fmd_ctl_init(NULL));
860 
861 	fmd_event_hold(e);
862 
863 	/*
864 	 * Once all data structures are initialized, we load all of our modules
865 	 * in order according to class in order to load up any subscriptions.
866 	 * Once built-in modules are loaded, we detach from our waiting parent.
867 	 */
868 	dp->d_mod_hash = fmd_modhash_create();
869 
870 	if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg)
871 		fmd_error(EFMD_EXIT, "failed to initialize fault manager\n");
872 
873 	(void) fmd_conf_getprop(dp->d_conf, "self.name", &name);
874 	dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name);
875 
876 	if (dp->d_self != NULL && fmd_module_dc_key2code(dp->d_self,
877 	    nodc_key, nodc_str, sizeof (nodc_str)) == 0)
878 		(void) fmd_conf_setprop(dp->d_conf, "nodiagcode", nodc_str);
879 
880 	fmd_rpc_init();
881 	dp->d_running = 1; /* we are now officially an active fmd */
882 
883 	/*
884 	 * Now that we're running, if a pipe fd was specified, write an exit
885 	 * status to it to indicate that our parent process can safely detach.
886 	 * Then proceed to loading the remaining non-built-in modules.
887 	 */
888 	if (pfd >= 0)
889 		(void) write(pfd, &status, sizeof (status));
890 
891 	/*
892 	 * Before loading all modules, repopulate the ASRU cache from its
893 	 * persistent repository on disk.  Then during module loading, the
894 	 * restoration of checkpoint files will reparent any active cases.
895 	 */
896 	fmd_asru_hash_refresh(dp->d_asrus);
897 
898 	(void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap);
899 	fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops, ".so");
900 
901 	(void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap);
902 	fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops, NULL);
903 
904 	/*
905 	 * With all modules loaded, replay fault events from the ASRU cache for
906 	 * any ASRUs that must be retired, replay error events from the errlog
907 	 * that did not finish processing the last time ran, and then release
908 	 * the global module barrier by executing a final rele on d_mod_event.
909 	 */
910 	fmd_asru_hash_replay(dp->d_asrus);
911 
912 	(void) pthread_rwlock_rdlock(&dp->d_log_lock);
913 	fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp);
914 	fmd_log_update(dp->d_errlog);
915 	(void) pthread_rwlock_unlock(&dp->d_log_lock);
916 
917 	dp->d_mod_event = NULL;
918 	fmd_event_rele(e);
919 
920 	/*
921 	 * Finally, awaken any threads associated with receiving events from
922 	 * open transports and tell them to proceed with fmd_xprt_recv().
923 	 */
924 	fmd_xprt_resume_all();
925 	fmd_gc(dp, 0, 0);
926 
927 	dp->d_booted = 1;
928 }
929 
930 void
931 fmd_help(fmd_t *dp)
932 {
933 	const fmd_conf_mode_t *cmp;
934 
935 	(void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname);
936 
937 	for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++)
938 		(void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc);
939 }
940