xref: /titanic_51/usr/src/cmd/fm/fmd/common/fmd.c (revision fd9e7635fa85e33de5aff912b955d797589f6f87)
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 2008 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 { "repaircode", &fmd_conf_string, "-" },	/* diagcode for list.repaired */
275 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */
276 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */
277 { "platform", &fmd_conf_string, _fmd_plat },	/* platform string (uname -i) */
278 { "plugin.close", &fmd_conf_bool, "true" },	/* dlclose plugins on fini */
279 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */
280 { "product", &fmd_conf_string, _fmd_prod },	/* product name string */
281 { "rootdir", &fmd_conf_string, "" },		/* root directory for paths */
282 { "rpc.adm.path", &fmd_conf_string, NULL },	/* FMD_ADM rendezvous file */
283 { "rpc.adm.prog", &fmd_conf_uint32, "100169" },	/* FMD_ADM rpc program num */
284 { "rpc.api.path", &fmd_conf_string, NULL },	/* FMD_API rendezvous file */
285 { "rpc.api.prog", &fmd_conf_uint32, "100170" },	/* FMD_API rpc program num */
286 { "rpc.rcvsize", &fmd_conf_size, "128k" },	/* rpc receive buffer size */
287 { "rpc.sndsize", &fmd_conf_size, "128k" },	/* rpc send buffer size */
288 { "rsrc.age", &fmd_conf_time, "30d" },		/* max age of old rsrc log */
289 { "rsrc.zero", &fmd_conf_bool, "false" },	/* zero rsrc cache on start? */
290 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */
291 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */
292 { "self.dict", &fmd_conf_list, "FMD.dict" },	/* self-diag dictionary list */
293 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */
294 { "strbuckets", &fmd_conf_uint32, "211" },	/* size of string hashes */
295 #ifdef DEBUG
296 { "trace.mode", &fmd_trmode_ops, "full" },	/* trace mode: none/lite/full */
297 #else
298 { "trace.mode", &fmd_trmode_ops, "lite" },	/* trace mode: none/lite/full */
299 #endif
300 { "trace.recs", &fmd_conf_uint32, "128" },	/* trace records per thread */
301 { "trace.frames", &fmd_conf_uint32, "16" },	/* max trace rec stack frames */
302 { "uuidlen", &fmd_conf_uint32, "36" },		/* UUID ASCII string length */
303 { "xprt.ttl", &fmd_conf_uint8, "1" },		/* default event time-to-live */
304 };
305 
306 /*
307  * Statistics maintained by fmd itself on behalf of various global subsystems.
308  * NOTE: FMD_TYPE_STRING statistics should not be used here.  If they are
309  * required in the future, the FMD_ADM_MODGSTAT service routine must change.
310  */
311 static fmd_statistics_t _fmd_stats = {
312 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" },
313 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" },
314 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" },
315 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" },
316 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" },
317 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" },
318 { "topo.gen", FMD_TYPE_UINT64, "topology snapshot generation" },
319 { "topo.drgen", FMD_TYPE_UINT64, "current topology DR generation number" },
320 };
321 
322 void
323 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf)
324 {
325 	fmd_conf_path_t *pap;
326 	char file[PATH_MAX];
327 	const char *name;
328 	fmd_stat_t *sp;
329 	int i;
330 
331 	smbios_hdl_t *shp;
332 	smbios_system_t s1;
333 	smbios_info_t s2;
334 	id_t id;
335 
336 	di_prom_handle_t promh = DI_PROM_HANDLE_NIL;
337 	di_node_t rooth = DI_NODE_NIL;
338 	char *bufp;
339 
340 	(void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat));
341 	(void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa));
342 	(void) uname(&_fmd_uts);
343 
344 	if ((shp = smbios_open(NULL, SMB_VERSION, 0, NULL)) != NULL) {
345 		if ((id = smbios_info_system(shp, &s1)) != SMB_ERR &&
346 		    smbios_info_common(shp, id, &s2) != SMB_ERR) {
347 			(void) strlcpy(_fmd_prod, s2.smbi_product, MAXNAMELEN);
348 			(void) strlcpy(_fmd_csn, s2.smbi_serial, MAXNAMELEN);
349 		}
350 		smbios_close(shp);
351 	} else if ((rooth = di_init("/", DINFOPROP)) != DI_NODE_NIL &&
352 	    (promh = di_prom_init()) != DI_PROM_HANDLE_NIL) {
353 		if (di_prom_prop_lookup_bytes(promh, rooth, "chassis-sn",
354 		    (unsigned char **)&bufp) != -1) {
355 			(void) strlcpy(_fmd_csn, bufp, MAXNAMELEN);
356 		}
357 	}
358 
359 	if (promh != DI_PROM_HANDLE_NIL)
360 		di_prom_fini(promh);
361 	if (rooth != DI_NODE_NIL)
362 		di_fini(rooth);
363 
364 	bzero(dp, sizeof (fmd_t));
365 
366 	dp->d_version = _fmd_version;
367 	dp->d_pname = fmd_strbasename(arg0);
368 	dp->d_pid = getpid();
369 
370 	if (pthread_key_create(&dp->d_key, NULL) != 0)
371 		fmd_error(EFMD_EXIT, "failed to create pthread key");
372 
373 	(void) pthread_mutex_init(&dp->d_xprt_lock, NULL);
374 	(void) pthread_mutex_init(&dp->d_err_lock, NULL);
375 	(void) pthread_mutex_init(&dp->d_thr_lock, NULL);
376 	(void) pthread_mutex_init(&dp->d_mod_lock, NULL);
377 	(void) pthread_mutex_init(&dp->d_stats_lock, NULL);
378 	(void) pthread_mutex_init(&dp->d_topo_lock, NULL);
379 	(void) pthread_rwlock_init(&dp->d_log_lock, NULL);
380 	(void) pthread_mutex_init(&dp->d_fmd_lock, NULL);
381 	(void) pthread_cond_init(&dp->d_fmd_cv, NULL);
382 
383 	/*
384 	 * A small number of properties must be set manually before we open
385 	 * the root configuration file.  These include any settings for our
386 	 * memory allocator and path expansion token values, because these
387 	 * values are needed by the routines in fmd_conf.c itself.  After
388 	 * the root configuration file is processed, we reset these properties
389 	 * based upon the latest values from the configuration file.
390 	 */
391 	dp->d_alloc_msecs = 10;
392 	dp->d_alloc_tries = 3;
393 	dp->d_str_buckets = 211;
394 
395 	dp->d_rootdir = root ? root : "";
396 	dp->d_platform = _fmd_plat;
397 	dp->d_machine = _fmd_uts.machine;
398 	dp->d_isaname = _fmd_isa;
399 
400 	dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) /
401 	    sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER);
402 
403 	if (dp->d_conf == NULL) {
404 		fmd_error(EFMD_EXIT,
405 		    "failed to load required configuration properties\n");
406 	}
407 
408 	(void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs);
409 	(void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries);
410 	(void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets);
411 
412 	(void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform);
413 	(void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine);
414 	(void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname);
415 
416 	/*
417 	 * Manually specified rootdirs override config files, so only update
418 	 * d_rootdir based on the config files we parsed if no 'root' was set.
419 	 */
420 	if (root == NULL)
421 		(void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir);
422 	else
423 		(void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir);
424 
425 	/*
426 	 * Once the base conf file properties are loaded, lookup the values
427 	 * of $conf_path and $conf_file and merge in any other conf files.
428 	 */
429 	(void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap);
430 	(void) fmd_conf_getprop(dp->d_conf, "conf_file", &name);
431 
432 	for (i = 0; i < pap->cpa_argc; i++) {
433 		(void) snprintf(file, sizeof (file),
434 		    "%s/%s", pap->cpa_argv[i], name);
435 		if (access(file, F_OK) == 0)
436 			fmd_conf_merge(dp->d_conf, file);
437 	}
438 
439 	/*
440 	 * Update the value of fmd.d_fg based on "fg".  We cache this property
441 	 * because it must be accessed deep within fmd at fmd_verror() time.
442 	 * Update any other properties that must be cached for performance.
443 	 */
444 	(void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg);
445 	(void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl);
446 
447 	/*
448 	 * Initialize our custom libnvpair allocator and create an nvlist for
449 	 * authority elements corresponding to this instance of the daemon.
450 	 */
451 	(void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops);
452 	dp->d_auth = fmd_protocol_authority();
453 
454 	/*
455 	 * The fmd_module_t for the root module must be created manually.  Most
456 	 * of it remains unused and zero, except for the few things we fill in.
457 	 */
458 	dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP);
459 	dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP);
460 	dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod);
461 
462 	fmd_list_append(&dp->d_mod_list, dp->d_rmod);
463 	fmd_module_hold(dp->d_rmod);
464 
465 	(void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL);
466 	(void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL);
467 	(void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL);
468 
469 	dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self());
470 	dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP);
471 	dp->d_rmod->mod_ustat = fmd_ustat_create();
472 
473 	if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0)
474 		fmd_error(EFMD_EXIT, "failed to attach main thread key");
475 
476 	if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert(
477 	    dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) /
478 	    sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL)
479 		fmd_error(EFMD_EXIT, "failed to initialize statistics");
480 
481 	(void) pthread_mutex_lock(&dp->d_rmod->mod_lock);
482 	dp->d_rmod->mod_flags |= FMD_MOD_INIT;
483 	(void) pthread_mutex_unlock(&dp->d_rmod->mod_lock);
484 
485 	/*
486 	 * In addition to inserting the _fmd_stats collection of program-wide
487 	 * statistics, we also insert a statistic named after each of our
488 	 * errors and update these counts in fmd_verror() (see fmd_subr.c).
489 	 */
490 	dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) *
491 	    (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP);
492 
493 	for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) {
494 		(void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s",
495 		    strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1);
496 		sp->fmds_type = FMD_TYPE_UINT64;
497 	}
498 
499 	(void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC,
500 	    EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL);
501 }
502 
503 void
504 fmd_destroy(fmd_t *dp)
505 {
506 	fmd_module_t *mp;
507 	fmd_case_t *cp;
508 	int core;
509 
510 	(void) fmd_conf_getprop(fmd.d_conf, "core", &core);
511 
512 	fmd_rpc_fini();
513 	fmd_dr_fini();
514 
515 	if (dp->d_xprt_ids != NULL)
516 		fmd_xprt_suspend_all();
517 
518 	/*
519 	 * Unload the self-diagnosis module first.  This ensures that it does
520 	 * not get confused as we start unloading other modules, etc.  We must
521 	 * hold the dispq lock as a writer while doing so since it uses d_self.
522 	 */
523 	if (dp->d_self != NULL) {
524 		fmd_module_t *self;
525 
526 		(void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock);
527 		self = dp->d_self;
528 		dp->d_self = NULL;
529 		(void) pthread_rwlock_unlock(&dp->d_disp->dq_lock);
530 
531 		fmd_module_unload(self);
532 		fmd_module_rele(self);
533 	}
534 
535 	/*
536 	 * Unload modules in reverse order *except* for the root module, which
537 	 * is first in the list.  This allows it to keep its thread and trace.
538 	 */
539 	for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) {
540 		fmd_module_unload(mp);
541 		mp = fmd_list_prev(mp);
542 	}
543 
544 	if (dp->d_mod_hash != NULL) {
545 		fmd_modhash_destroy(dp->d_mod_hash);
546 		dp->d_mod_hash = NULL;
547 	}
548 
549 	/*
550 	 * Close both log files now that modules are no longer active.  We must
551 	 * set these pointers to NULL in case any subsequent errors occur.
552 	 */
553 	if (dp->d_errlog != NULL) {
554 		fmd_log_rele(dp->d_errlog);
555 		dp->d_errlog = NULL;
556 	}
557 
558 	if (dp->d_fltlog != NULL) {
559 		fmd_log_rele(dp->d_fltlog);
560 		dp->d_fltlog = NULL;
561 	}
562 
563 	/*
564 	 * Now destroy the resource cache: each ASRU contains a case reference,
565 	 * which may in turn contain a pointer to a referenced owning module.
566 	 */
567 	if (dp->d_asrus != NULL) {
568 		fmd_asru_hash_destroy(dp->d_asrus);
569 		dp->d_asrus = NULL;
570 	}
571 
572 	/*
573 	 * Now that all data structures that refer to modules are torn down,
574 	 * no modules should be remaining on the module list except for d_rmod.
575 	 * If we trip one of these assertions, we're missing a rele somewhere.
576 	 */
577 	ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod);
578 	ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod);
579 
580 	/*
581 	 * Now destroy the root module.  We clear its thread key first so any
582 	 * calls to fmd_trace() inside of the module code will be ignored.
583 	 */
584 	(void) pthread_setspecific(dp->d_key, NULL);
585 	fmd_module_lock(dp->d_rmod);
586 
587 	while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL)
588 		fmd_case_discard(cp);
589 
590 	fmd_module_unlock(dp->d_rmod);
591 	fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t));
592 	dp->d_rmod->mod_stats = NULL;
593 
594 	(void) pthread_mutex_lock(&dp->d_rmod->mod_lock);
595 	dp->d_rmod->mod_flags |= FMD_MOD_FINI;
596 	(void) pthread_mutex_unlock(&dp->d_rmod->mod_lock);
597 
598 	fmd_module_rele(dp->d_rmod);
599 	ASSERT(fmd_list_next(&dp->d_mod_list) == NULL);
600 
601 	/*
602 	 * Now destroy the remaining global data structures.  If 'core' was
603 	 * set to true, force a core dump so we can check for memory leaks.
604 	 */
605 	if (dp->d_cases != NULL)
606 		fmd_case_hash_destroy(dp->d_cases);
607 	if (dp->d_disp != NULL)
608 		fmd_dispq_destroy(dp->d_disp);
609 	if (dp->d_timers != NULL)
610 		fmd_timerq_destroy(dp->d_timers);
611 	if (dp->d_schemes != NULL)
612 		fmd_scheme_hash_destroy(dp->d_schemes);
613 	if (dp->d_xprt_ids != NULL)
614 		fmd_idspace_destroy(dp->d_xprt_ids);
615 
616 	if (dp->d_errstats != NULL) {
617 		fmd_free(dp->d_errstats,
618 		    sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN));
619 	}
620 
621 	if (dp->d_conf != NULL)
622 		fmd_conf_close(dp->d_conf);
623 
624 	fmd_topo_fini();
625 
626 	nvlist_free(dp->d_auth);
627 	(void) nv_alloc_fini(&dp->d_nva);
628 	dp->d_clockops->fto_fini(dp->d_clockptr);
629 
630 	(void) pthread_key_delete(dp->d_key);
631 	bzero(dp, sizeof (fmd_t));
632 
633 	if (core)
634 		fmd_panic("forcing core dump at user request\n");
635 }
636 
637 /*ARGSUSED*/
638 static void
639 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt)
640 {
641 	hrtime_t delta;
642 
643 	if (id != 0) {
644 		TRACE((FMD_DBG_MOD, "garbage collect start"));
645 		fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc);
646 		TRACE((FMD_DBG_MOD, "garbage collect end"));
647 
648 		(void) pthread_rwlock_rdlock(&dp->d_log_lock);
649 		fmd_log_update(dp->d_errlog);
650 		(void) pthread_rwlock_unlock(&dp->d_log_lock);
651 	}
652 
653 	(void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta);
654 	(void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids,
655 	    (fmd_timer_f *)fmd_gc, dp, NULL, delta);
656 }
657 
658 /*ARGSUSED*/
659 static void
660 fmd_clear_aged_rsrcs(fmd_t *dp, id_t id, hrtime_t hrt)
661 {
662 	hrtime_t delta;
663 
664 	fmd_asru_clear_aged_rsrcs();
665 	(void) fmd_conf_getprop(dp->d_conf, "rsrc.age", &delta);
666 	(void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids,
667 	    (fmd_timer_f *)fmd_clear_aged_rsrcs, dp, NULL, delta/10);
668 }
669 
670 /*
671  * Events are committed to the errlog after cases are checkpointed.  If fmd
672  * crashes before an event is ever associated with a module, this function will
673  * be called to replay it to all subscribers.  If fmd crashes in between the
674  * subscriber checkpointing and committing the event in the error log, the
675  * module will have seen the event and we don't want to replay it.  So we look
676  * for the event in all modules and transition it to the proper state.  If
677  * it is found, we commit it to the error log and do not replay it.  The in-
678  * memory case search used by fmd_module_contains() et al isn't particularly
679  * efficient, but it is faster than doing read i/o's on every case event to
680  * check their status or write i/o's on every event to replay to update states.
681  * We can improve the efficiency of this lookup algorithm later if necessary.
682  */
683 /*ARGSUSED*/
684 static void
685 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp)
686 {
687 	fmd_module_t *mp;
688 	fmd_stat_t *sp;
689 
690 	(void) pthread_mutex_lock(&dp->d_mod_lock);
691 
692 	for (mp = fmd_list_next(&dp->d_mod_list);
693 	    mp != NULL; mp = fmd_list_next(mp)) {
694 		if (fmd_module_contains(mp, ep)) {
695 			fmd_module_hold(mp);
696 			break;
697 		}
698 	}
699 
700 	(void) pthread_mutex_unlock(&dp->d_mod_lock);
701 
702 	if (mp != NULL) {
703 		fmd_event_commit(ep);
704 		fmd_module_rele(mp);
705 		sp = &dp->d_stats->ds_log_partials;
706 	} else {
707 		fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep));
708 		sp = &dp->d_stats->ds_log_replayed;
709 	}
710 
711 	(void) pthread_mutex_lock(&dp->d_stats_lock);
712 	sp->fmds_value.ui64++;
713 	(void) pthread_mutex_unlock(&dp->d_stats_lock);
714 }
715 
716 void
717 fmd_door_server(void *dip)
718 {
719 	fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip);
720 	(void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL);
721 	(void) door_return(NULL, 0, NULL, 0);
722 }
723 
724 /*
725  * Custom door server create callback.  Any fmd services that use doors will
726  * require those threads to have their fmd-specific TSD initialized, etc.
727  */
728 static void
729 fmd_door(door_info_t *dip)
730 {
731 	if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL)
732 		fmd_panic("failed to create server for door %p", (void *)dip);
733 }
734 
735 /*
736  * This signal handler is installed for the client.thrsig signal to be used to
737  * force an auxiliary thread to wake up from a system call and return EINTR in
738  * response to a module's use of fmd_thr_signal().  We also trace the event.
739  */
740 static void
741 fmd_signal(int sig)
742 {
743 	TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig));
744 }
745 
746 void
747 fmd_run(fmd_t *dp, int pfd)
748 {
749 	char *nodc_key[] = { FMD_FLT_NODC, NULL };
750 	char *repair_key[] = { FM_LIST_REPAIRED_CLASS, NULL };
751 	char code_str[128];
752 	struct sigaction act;
753 
754 	int status = FMD_EXIT_SUCCESS;
755 	const char *name;
756 	fmd_conf_path_t *pap;
757 	fmd_event_t *e;
758 	int dbout;
759 
760 	/*
761 	 * Cache all the current debug property settings in d_fmd_debug,
762 	 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout.  If a given debug mask
763 	 * is non-zero and the corresponding dbout mask is zero, set dbout
764 	 * to a sensible default value based on whether we have daemonized.
765 	 */
766 	(void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout);
767 
768 	if (dp->d_fmd_debug != 0 && dbout == 0)
769 		dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG;
770 	else
771 		dp->d_fmd_dbout = dbout;
772 
773 	(void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug);
774 	(void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout);
775 
776 	if (dp->d_hdl_debug != 0 && dbout == 0)
777 		dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG;
778 	else
779 		dp->d_hdl_dbout = dbout;
780 
781 	/*
782 	 * Initialize remaining major program data structures such as the
783 	 * clock, dispatch queues, log files, module hash collections, etc.
784 	 * This work is done here rather than in fmd_create() to permit the -o
785 	 * command-line option to modify properties after fmd_create() is done.
786 	 */
787 	name = dp->d_rootdir != NULL &&
788 	    *dp->d_rootdir != '\0' ? dp->d_rootdir : NULL;
789 
790 	/*
791 	 * The clock must be initialized before fmd_topo_init() because
792 	 * fmd_topo_update() calls fmd_time_gethrtime().
793 	 */
794 	dp->d_clockptr = dp->d_clockops->fto_init();
795 
796 	fmd_topo_init();
797 
798 	dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX);
799 	fmd_xprt_suspend_all();
800 
801 	(void) door_server_create(fmd_door);
802 
803 	dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16);
804 	dp->d_timers = fmd_timerq_create();
805 	dp->d_disp = fmd_dispq_create();
806 	dp->d_cases = fmd_case_hash_create();
807 
808 	/*
809 	 * The root module's mod_queue is created with limit zero, making it
810 	 * act like /dev/null; anything inserted here is simply ignored.
811 	 */
812 	dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod,
813 	    &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0);
814 
815 	/*
816 	 * Once our subsystems that use signals have been set up, install the
817 	 * signal handler for the fmd_thr_signal() API.  Verify that the signal
818 	 * being used for this purpose doesn't conflict with something else.
819 	 */
820 	(void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig);
821 
822 	if (sigaction(dp->d_thr_sig, NULL, &act) != 0) {
823 		fmd_error(EFMD_EXIT, "invalid signal selected for "
824 		    "client.thrsig property: %d\n", dp->d_thr_sig);
825 	}
826 
827 	if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) {
828 		fmd_error(EFMD_EXIT, "signal selected for client.thrsig "
829 		    "property is already in use: %d\n", dp->d_thr_sig);
830 	}
831 
832 	act.sa_handler = fmd_signal;
833 	act.sa_flags = 0;
834 
835 	(void) sigemptyset(&act.sa_mask);
836 	(void) sigaction(dp->d_thr_sig, &act, NULL);
837 
838 	(void) fmd_conf_getprop(dp->d_conf, "schemedir", &name);
839 	dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name);
840 
841 	(void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name);
842 	dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name);
843 
844 	(void) fmd_conf_getprop(dp->d_conf, "log.error", &name);
845 	dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR);
846 
847 	(void) fmd_conf_getprop(dp->d_conf, "log.fault", &name);
848 	dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT);
849 
850 	if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL)
851 		fmd_error(EFMD_EXIT, "failed to initialize log files\n");
852 
853 	/*
854 	 * Before loading modules, create an empty control event which will act
855 	 * as a global barrier for module event processing.  Each module we
856 	 * load successfully will insert it at their head of their event queue,
857 	 * and then pause inside of fmd_ctl_rele() after dequeuing the event.
858 	 * This module barrier is required for two reasons:
859 	 *
860 	 * (a) During module loading, the restoration of case checkpoints may
861 	 *    result in a list.* event being recreated for which the intended
862 	 *    subscriber has not yet loaded depending on the load order. Such
863 	 *    events could then result in spurious "no subscriber" errors.
864 	 *
865 	 * (b) During errlog replay, a sequence of errors from a long time ago
866 	 *    may be replayed, and the module may attempt to install relative
867 	 *    timers associated with one or more of these events.  If errlog
868 	 *    replay were "racing" with active module threads, an event E1
869 	 *    that resulted in a relative timer T at time E1 + N nsec could
870 	 *    fire prior to an event E2 being enqueued, even if the relative
871 	 *    time ordering was E1 < E2 < E1 + N, causing mis-diagnosis.
872 	 */
873 	dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL,
874 	    FMD_HRT_NOW, NULL, fmd_ctl_init(NULL));
875 
876 	fmd_event_hold(e);
877 
878 	/*
879 	 * Once all data structures are initialized, we load all of our modules
880 	 * in order according to class in order to load up any subscriptions.
881 	 * Once built-in modules are loaded, we detach from our waiting parent.
882 	 */
883 	dp->d_mod_hash = fmd_modhash_create();
884 
885 	if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg)
886 		fmd_error(EFMD_EXIT, "failed to initialize fault manager\n");
887 
888 	(void) fmd_conf_getprop(dp->d_conf, "self.name", &name);
889 	dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name);
890 
891 	if (dp->d_self != NULL) {
892 		if (fmd_module_dc_key2code(dp->d_self, nodc_key, code_str,
893 		    sizeof (code_str)) == 0)
894 			(void) fmd_conf_setprop(dp->d_conf, "nodiagcode",
895 			    code_str);
896 		if (fmd_module_dc_key2code(dp->d_self, repair_key, code_str,
897 		    sizeof (code_str)) == 0)
898 			(void) fmd_conf_setprop(dp->d_conf, "repaircode",
899 			    code_str);
900 	}
901 
902 	fmd_rpc_init();
903 	dp->d_running = 1; /* we are now officially an active fmd */
904 
905 	/*
906 	 * Now that we're running, if a pipe fd was specified, write an exit
907 	 * status to it to indicate that our parent process can safely detach.
908 	 * Then proceed to loading the remaining non-built-in modules.
909 	 */
910 	if (pfd >= 0)
911 		(void) write(pfd, &status, sizeof (status));
912 
913 	/*
914 	 * Before loading all modules, repopulate the ASRU cache from its
915 	 * persistent repository on disk.  Then during module loading, the
916 	 * restoration of checkpoint files will reparent any active cases.
917 	 */
918 	fmd_asru_hash_refresh(dp->d_asrus);
919 
920 	(void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap);
921 	fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops, ".so");
922 
923 	(void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap);
924 	fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops, NULL);
925 
926 	/*
927 	 * Subscribe to sysevents after all modules have been loaded.
928 	 */
929 	fmd_dr_init();
930 
931 	/*
932 	 * With all modules loaded, replay fault events from the ASRU cache for
933 	 * any ASRUs that must be retired, replay error events from the errlog
934 	 * that did not finish processing the last time ran, and then release
935 	 * the global module barrier by executing a final rele on d_mod_event.
936 	 */
937 	fmd_asru_hash_replay(dp->d_asrus);
938 
939 	(void) pthread_rwlock_rdlock(&dp->d_log_lock);
940 	fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp);
941 	fmd_log_update(dp->d_errlog);
942 	(void) pthread_rwlock_unlock(&dp->d_log_lock);
943 
944 	dp->d_mod_event = NULL;
945 	fmd_event_rele(e);
946 
947 	/*
948 	 * Finally, awaken any threads associated with receiving events from
949 	 * open transports and tell them to proceed with fmd_xprt_recv().
950 	 */
951 	fmd_xprt_resume_all();
952 	fmd_gc(dp, 0, 0);
953 	fmd_clear_aged_rsrcs(dp, 0, 0);
954 
955 	(void) pthread_mutex_lock(&dp->d_fmd_lock);
956 	dp->d_booted = 1;
957 	(void) pthread_cond_broadcast(&dp->d_fmd_cv);
958 	(void) pthread_mutex_unlock(&dp->d_fmd_lock);
959 }
960 
961 void
962 fmd_help(fmd_t *dp)
963 {
964 	const fmd_conf_mode_t *cmp;
965 
966 	(void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname);
967 
968 	for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++)
969 		(void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc);
970 }
971