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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 23 /* 24 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 25 * Use is subject to license terms. 26 */ 27 28 #pragma ident "%Z%%M% %I% %E% SMI" 29 30 #include <sys/types.h> 31 #include <sys/utsname.h> 32 #include <sys/param.h> 33 #include <sys/systeminfo.h> 34 #include <sys/fm/util.h> 35 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_xprt.h> 63 #include <fmd_ctl.h> 64 65 #include <fmd.h> 66 67 extern const nv_alloc_ops_t fmd_nv_alloc_ops; /* see fmd_nv.c */ 68 69 const char _fmd_version[] = "1.1"; /* daemon version string */ 70 static char _fmd_plat[MAXNAMELEN]; /* native platform string */ 71 static char _fmd_isa[MAXNAMELEN]; /* native instruction set */ 72 static struct utsname _fmd_uts; /* native uname(2) info */ 73 74 /* 75 * Note: the configuration file path is ordered from most common to most host- 76 * specific because new conf files are merged/override previous ones. The 77 * module paths are in the opposite order, from most specific to most common, 78 * because once a module is loaded fmd will not try to load over the same name. 79 */ 80 81 static const char _fmd_conf_path[] = 82 "%r/usr/lib/fm/fmd:" 83 "%r/usr/platform/%m/lib/fm/fmd:" 84 "%r/usr/platform/%i/lib/fm/fmd:" 85 "%r/etc/fm/fmd"; 86 87 static const char _fmd_agent_path[] = 88 "%r/usr/platform/%i/lib/fm/fmd/agents:" 89 "%r/usr/platform/%m/lib/fm/fmd/agents:" 90 "%r/usr/lib/fm/fmd/agents"; 91 92 static const char _fmd_plugin_path[] = 93 "%r/usr/platform/%i/lib/fm/fmd/plugins:" 94 "%r/usr/platform/%m/lib/fm/fmd/plugins:" 95 "%r/usr/lib/fm/fmd/plugins"; 96 97 static const char _fmd_scheme_path[] = 98 "usr/lib/fm/fmd/schemes"; 99 100 static const fmd_conf_mode_t _fmd_cerror_modes[] = { 101 { "unload", "unload offending client module", FMD_CERROR_UNLOAD }, 102 { "stop", "stop daemon for debugger attach", FMD_CERROR_STOP }, 103 { "abort", "abort daemon and force core dump", FMD_CERROR_ABORT }, 104 { NULL, NULL, 0 } 105 }; 106 107 static const fmd_conf_mode_t _fmd_dbout_modes[] = { 108 { "stderr", "send debug messages to stderr", FMD_DBOUT_STDERR }, 109 { "syslog", "send debug messages to syslog", FMD_DBOUT_SYSLOG }, 110 { NULL, NULL, 0 } 111 }; 112 113 static const fmd_conf_mode_t _fmd_debug_modes[] = { 114 { "help", "display debugging modes and exit", FMD_DBG_HELP }, 115 { "mod", "debug module load/unload/locking", FMD_DBG_MOD }, 116 { "disp", "debug dispatch queue processing", FMD_DBG_DISP }, 117 { "xprt", "debug transport-specific routines", FMD_DBG_XPRT }, 118 { "evt", "debug event subsystem routines", FMD_DBG_EVT }, 119 { "log", "debug log subsystem routines", FMD_DBG_LOG }, 120 { "tmr", "debug timer subsystem routines", FMD_DBG_TMR }, 121 { "fmri", "debug fmri subsystem routines", FMD_DBG_FMRI }, 122 { "asru", "debug asru subsystem routines", FMD_DBG_ASRU }, 123 { "case", "debug case subsystem routines", FMD_DBG_CASE }, 124 { "ckpt", "debug checkpoint routines", FMD_DBG_CKPT }, 125 { "rpc", "debug rpc service routines", FMD_DBG_RPC }, 126 { "trace", "display matching trace calls", FMD_DBG_TRACE }, 127 { "all", "enable all available debug modes", FMD_DBG_ALL }, 128 { NULL, NULL, 0 } 129 }; 130 131 static int 132 fmd_cerror_set(fmd_conf_param_t *pp, const char *value) 133 { 134 return (fmd_conf_mode_set(_fmd_cerror_modes, pp, value)); 135 } 136 137 static int 138 fmd_dbout_set(fmd_conf_param_t *pp, const char *value) 139 { 140 return (fmd_conf_mode_set(_fmd_dbout_modes, pp, value)); 141 } 142 143 static int 144 fmd_debug_set(fmd_conf_param_t *pp, const char *value) 145 { 146 int err = fmd_conf_mode_set(_fmd_debug_modes, pp, value); 147 148 if (err == 0) 149 fmd.d_fmd_debug = pp->cp_value.cpv_num; 150 151 return (err); 152 } 153 154 static int 155 fmd_trmode_set(fmd_conf_param_t *pp, const char *value) 156 { 157 fmd_tracebuf_f *func; 158 159 if (strcasecmp(value, "none") == 0) 160 func = fmd_trace_none; 161 else if (strcasecmp(value, "lite") == 0) 162 func = fmd_trace_lite; 163 else if (strcasecmp(value, "full") == 0) 164 func = fmd_trace_full; 165 else 166 return (fmd_set_errno(EFMD_CONF_INVAL)); 167 168 fmd.d_thr_trace = (void (*)())func; 169 pp->cp_value.cpv_ptr = (void *)func; 170 return (0); 171 } 172 173 static void 174 fmd_trmode_get(const fmd_conf_param_t *pp, void *ptr) 175 { 176 *((void **)ptr) = pp->cp_value.cpv_ptr; 177 } 178 179 static int 180 fmd_clkmode_set(fmd_conf_param_t *pp, const char *value) 181 { 182 const fmd_timeops_t *ops; 183 184 if (strcasecmp(value, "native") == 0) 185 ops = &fmd_timeops_native; 186 else if (strcasecmp(value, "simulated") == 0) 187 ops = &fmd_timeops_simulated; 188 else 189 return (fmd_set_errno(EFMD_CONF_INVAL)); 190 191 fmd.d_clockops = ops; 192 pp->cp_value.cpv_ptr = (void *)ops; 193 return (0); 194 } 195 196 static void 197 fmd_clkmode_get(const fmd_conf_param_t *pp, void *ptr) 198 { 199 *((void **)ptr) = pp->cp_value.cpv_ptr; 200 } 201 202 static const fmd_conf_ops_t fmd_cerror_ops = { 203 fmd_cerror_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 204 }; 205 206 static const fmd_conf_ops_t fmd_dbout_ops = { 207 fmd_dbout_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 208 }; 209 210 static const fmd_conf_ops_t fmd_debug_ops = { 211 fmd_debug_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 212 }; 213 214 static const fmd_conf_ops_t fmd_trmode_ops = { 215 fmd_trmode_set, fmd_trmode_get, fmd_conf_notsup, fmd_conf_nop 216 }; 217 218 static const fmd_conf_ops_t fmd_clkmode_ops = { 219 fmd_clkmode_set, fmd_clkmode_get, fmd_conf_notsup, fmd_conf_nop 220 }; 221 222 static const fmd_conf_formal_t _fmd_conf[] = { 223 { "agent.path", &fmd_conf_path, _fmd_agent_path }, /* path for agents */ 224 { "alloc_msecs", &fmd_conf_uint32, "10" }, /* msecs before alloc retry */ 225 { "alloc_tries", &fmd_conf_uint32, "3" }, /* max # of alloc retries */ 226 { "chassis", &fmd_conf_string, NULL }, /* chassis serial number */ 227 { "ckpt.dir", &fmd_conf_string, "var/fm/fmd/ckpt" }, /* ckpt directory path */ 228 { "ckpt.dirmode", &fmd_conf_int32, "0700" }, /* ckpt directory perm mode */ 229 { "ckpt.mode", &fmd_conf_int32, "0400" }, /* ckpt file perm mode */ 230 { "ckpt.restore", &fmd_conf_bool, "true" }, /* restore checkpoints? */ 231 { "ckpt.save", &fmd_conf_bool, "true" }, /* save checkpoints? */ 232 { "ckpt.zero", &fmd_conf_bool, "false" }, /* zero checkpoints on start? */ 233 { "client.buflim", &fmd_conf_size, "10m" }, /* client buffer space limit */ 234 { "client.dbout", &fmd_dbout_ops, NULL }, /* client debug output sinks */ 235 { "client.debug", &fmd_conf_bool, NULL }, /* client debug enable */ 236 { "client.error", &fmd_cerror_ops, "unload" }, /* client error policy */ 237 { "client.memlim", &fmd_conf_size, "10m" }, /* client allocation limit */ 238 { "client.evqlim", &fmd_conf_uint32, "256" }, /* client event queue limit */ 239 { "client.thrlim", &fmd_conf_uint32, "8" }, /* client aux thread limit */ 240 { "client.thrsig", &fmd_conf_signal, "SIGUSR1" }, /* fmd_thr_signal() value */ 241 { "client.tmrlim", &fmd_conf_uint32, "1024" }, /* client pending timer limit */ 242 { "client.xprtlim", &fmd_conf_uint32, "256" }, /* client transport limit */ 243 { "client.xprtlog", &fmd_conf_bool, NULL }, /* client transport logging? */ 244 { "client.xprtqlim", &fmd_conf_uint32, "256" }, /* client transport queue lim */ 245 { "clock", &fmd_clkmode_ops, "native" }, /* clock operation mode */ 246 { "conf_path", &fmd_conf_path, _fmd_conf_path }, /* root config file path */ 247 { "conf_file", &fmd_conf_string, "fmd.conf" }, /* root config file name */ 248 { "core", &fmd_conf_bool, "false" }, /* force core dump on quit */ 249 { "dbout", &fmd_dbout_ops, NULL }, /* daemon debug output sinks */ 250 { "debug", &fmd_debug_ops, NULL }, /* daemon debugging flags */ 251 { "dictdir", &fmd_conf_string, "usr/lib/fm/dict" }, /* default diagcode dir */ 252 { "domain", &fmd_conf_string, NULL }, /* domain id for de auth */ 253 { "fg", &fmd_conf_bool, "false" }, /* run daemon in foreground */ 254 { "gc_interval", &fmd_conf_time, "1d" }, /* garbage collection intvl */ 255 { "ids.avg", &fmd_conf_uint32, "4" }, /* desired idspace chain len */ 256 { "ids.max", &fmd_conf_uint32, "1024" }, /* maximum idspace buckets */ 257 { "isaname", &fmd_conf_string, _fmd_isa }, /* instruction set (uname -p) */ 258 { "log.creator", &fmd_conf_string, "fmd" }, /* exacct log creator string */ 259 { "log.error", &fmd_conf_string, "var/fm/fmd/errlog" }, /* error log path */ 260 { "log.fault", &fmd_conf_string, "var/fm/fmd/fltlog" }, /* fault log path */ 261 { "log.minfree", &fmd_conf_size, "2m" }, /* min log fsys free space */ 262 { "log.rsrc", &fmd_conf_string, "var/fm/fmd/rsrc" }, /* asru log dir path */ 263 { "log.tryrotate", &fmd_conf_uint32, "10" }, /* max log rotation attempts */ 264 { "log.waitrotate", &fmd_conf_time, "200ms" }, /* log rotation retry delay */ 265 { "log.xprt", &fmd_conf_string, "var/fm/fmd/xprt" }, /* transport log dir */ 266 { "machine", &fmd_conf_string, _fmd_uts.machine }, /* machine name (uname -m) */ 267 { "nodiagcode", &fmd_conf_string, "-" }, /* diagcode to use if error */ 268 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */ 269 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */ 270 { "platform", &fmd_conf_string, _fmd_plat }, /* platform string (uname -i) */ 271 { "plugin.close", &fmd_conf_bool, "true" }, /* dlclose plugins on fini */ 272 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */ 273 { "rootdir", &fmd_conf_string, "" }, /* root directory for paths */ 274 { "rpc.adm.path", &fmd_conf_string, NULL }, /* FMD_ADM rendezvous file */ 275 { "rpc.adm.prog", &fmd_conf_uint32, "100169" }, /* FMD_ADM rpc program num */ 276 { "rpc.api.path", &fmd_conf_string, NULL }, /* FMD_API rendezvous file */ 277 { "rpc.api.prog", &fmd_conf_uint32, "100170" }, /* FMD_API rpc program num */ 278 { "rpc.rcvsize", &fmd_conf_size, "128k" }, /* rpc receive buffer size */ 279 { "rpc.sndsize", &fmd_conf_size, "128k" }, /* rpc send buffer size */ 280 { "rsrc.age", &fmd_conf_time, "30d" }, /* max age of old rsrc log */ 281 { "rsrc.zero", &fmd_conf_bool, "false" }, /* zero rsrc cache on start? */ 282 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */ 283 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */ 284 { "self.dict", &fmd_conf_list, "FMD.dict" }, /* self-diag dictionary list */ 285 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */ 286 { "strbuckets", &fmd_conf_uint32, "211" }, /* size of string hashes */ 287 #ifdef DEBUG 288 { "trace.mode", &fmd_trmode_ops, "full" }, /* trace mode: none/lite/full */ 289 #else 290 { "trace.mode", &fmd_trmode_ops, "lite" }, /* trace mode: none/lite/full */ 291 #endif 292 { "trace.recs", &fmd_conf_uint32, "128" }, /* trace records per thread */ 293 { "trace.frames", &fmd_conf_uint32, "16" }, /* max trace rec stack frames */ 294 { "uuidlen", &fmd_conf_uint32, "36" }, /* UUID ASCII string length */ 295 { "xprt.ttl", &fmd_conf_uint8, "1" }, /* default event time-to-live */ 296 }; 297 298 /* 299 * Statistics maintained by fmd itself on behalf of various global subsystems. 300 * NOTE: FMD_TYPE_STRING statistics should not be used here. If they are 301 * required in the future, the FMD_ADM_MODGSTAT service routine must change. 302 */ 303 static fmd_statistics_t _fmd_stats = { 304 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" }, 305 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" }, 306 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" }, 307 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" }, 308 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" }, 309 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" }, 310 }; 311 312 void 313 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf) 314 { 315 fmd_conf_path_t *pap; 316 char file[PATH_MAX]; 317 const char *name; 318 fmd_stat_t *sp; 319 int i; 320 321 (void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat)); 322 (void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa)); 323 (void) uname(&_fmd_uts); 324 325 bzero(dp, sizeof (fmd_t)); 326 327 dp->d_version = _fmd_version; 328 dp->d_pname = fmd_strbasename(arg0); 329 dp->d_pid = getpid(); 330 331 if (pthread_key_create(&dp->d_key, NULL) != 0) 332 fmd_error(EFMD_EXIT, "failed to create pthread key"); 333 334 (void) pthread_mutex_init(&dp->d_xprt_lock, NULL); 335 (void) pthread_mutex_init(&dp->d_err_lock, NULL); 336 (void) pthread_mutex_init(&dp->d_thr_lock, NULL); 337 (void) pthread_mutex_init(&dp->d_mod_lock, NULL); 338 (void) pthread_mutex_init(&dp->d_stats_lock, NULL); 339 (void) pthread_rwlock_init(&dp->d_log_lock, NULL); 340 341 /* 342 * A small number of properties must be set manually before we open 343 * the root configuration file. These include any settings for our 344 * memory allocator and path expansion token values, because these 345 * values are needed by the routines in fmd_conf.c itself. After 346 * the root configuration file is processed, we reset these properties 347 * based upon the latest values from the configuration file. 348 */ 349 dp->d_alloc_msecs = 10; 350 dp->d_alloc_tries = 3; 351 dp->d_str_buckets = 211; 352 353 dp->d_rootdir = root ? root : ""; 354 dp->d_platform = _fmd_plat; 355 dp->d_machine = _fmd_uts.machine; 356 dp->d_isaname = _fmd_isa; 357 358 dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) / 359 sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER); 360 361 if (dp->d_conf == NULL) { 362 fmd_error(EFMD_EXIT, 363 "failed to load required configuration properties\n"); 364 } 365 366 (void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs); 367 (void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries); 368 (void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets); 369 370 (void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform); 371 (void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine); 372 (void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname); 373 374 /* 375 * Manually specified rootdirs override config files, so only update 376 * d_rootdir based on the config files we parsed if no 'root' was set. 377 */ 378 if (root == NULL) 379 (void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir); 380 else 381 (void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir); 382 383 /* 384 * Once the base conf file properties are loaded, lookup the values 385 * of $conf_path and $conf_file and merge in any other conf files. 386 */ 387 (void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap); 388 (void) fmd_conf_getprop(dp->d_conf, "conf_file", &name); 389 390 for (i = 0; i < pap->cpa_argc; i++) { 391 (void) snprintf(file, sizeof (file), 392 "%s/%s", pap->cpa_argv[i], name); 393 if (access(file, F_OK) == 0) 394 fmd_conf_merge(dp->d_conf, file); 395 } 396 397 /* 398 * Update the value of fmd.d_fg based on "fg". We cache this property 399 * because it must be accessed deep within fmd at fmd_verror() time. 400 * Update any other properties that must be cached for performance. 401 */ 402 (void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg); 403 (void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl); 404 405 /* 406 * Initialize our custom libnvpair allocator and create an nvlist for 407 * authority elements corresponding to this instance of the daemon. 408 */ 409 (void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops); 410 dp->d_auth = fmd_protocol_authority(); 411 412 /* 413 * The fmd_module_t for the root module must be created manually. Most 414 * of it remains unused and zero, except for the few things we fill in. 415 */ 416 dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP); 417 dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP); 418 dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod); 419 420 fmd_list_append(&dp->d_mod_list, dp->d_rmod); 421 fmd_module_hold(dp->d_rmod); 422 423 (void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL); 424 (void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL); 425 (void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL); 426 427 dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self()); 428 dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP); 429 dp->d_rmod->mod_ustat = fmd_ustat_create(); 430 431 if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0) 432 fmd_error(EFMD_EXIT, "failed to attach main thread key"); 433 434 if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert( 435 dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) / 436 sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL) 437 fmd_error(EFMD_EXIT, "failed to initialize statistics"); 438 439 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 440 dp->d_rmod->mod_flags |= FMD_MOD_INIT; 441 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 442 443 /* 444 * In addition to inserting the _fmd_stats collection of program-wide 445 * statistics, we also insert a statistic named after each of our 446 * errors and update these counts in fmd_verror() (see fmd_subr.c). 447 */ 448 dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) * 449 (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP); 450 451 for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) { 452 (void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s", 453 strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1); 454 sp->fmds_type = FMD_TYPE_UINT64; 455 } 456 457 (void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, 458 EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL); 459 } 460 461 void 462 fmd_destroy(fmd_t *dp) 463 { 464 fmd_module_t *mp; 465 fmd_case_t *cp; 466 int core; 467 468 (void) fmd_conf_getprop(fmd.d_conf, "core", &core); 469 470 fmd_rpc_fini(); 471 fmd_dr_fini(); 472 473 if (dp->d_xprt_ids != NULL) 474 fmd_xprt_suspend_all(); 475 476 /* 477 * Unload the self-diagnosis module first. This ensures that it does 478 * not get confused as we start unloading other modules, etc. We must 479 * hold the dispq lock as a writer while doing so since it uses d_self. 480 */ 481 if (dp->d_self != NULL) { 482 (void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock); 483 fmd_module_unload(dp->d_self); 484 fmd_module_rele(dp->d_self); 485 dp->d_self = NULL; 486 (void) pthread_rwlock_unlock(&dp->d_disp->dq_lock); 487 } 488 489 /* 490 * Unload modules in reverse order *except* for the root module, which 491 * is first in the list. This allows it to keep its thread and trace. 492 */ 493 for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) { 494 fmd_module_unload(mp); 495 mp = fmd_list_prev(mp); 496 } 497 498 if (dp->d_mod_hash != NULL) { 499 fmd_modhash_destroy(dp->d_mod_hash); 500 dp->d_mod_hash = NULL; 501 } 502 503 /* 504 * Close both log files now that modules are no longer active. We must 505 * set these pointers to NULL in case any subsequent errors occur. 506 */ 507 if (dp->d_errlog != NULL) { 508 fmd_log_rele(dp->d_errlog); 509 dp->d_errlog = NULL; 510 } 511 512 if (dp->d_fltlog != NULL) { 513 fmd_log_rele(dp->d_fltlog); 514 dp->d_fltlog = NULL; 515 } 516 517 /* 518 * Now destroy the resource cache: each ASRU contains a case reference, 519 * which may in turn contain a pointer to a referenced owning module. 520 */ 521 if (dp->d_asrus != NULL) { 522 fmd_asru_hash_destroy(dp->d_asrus); 523 dp->d_asrus = NULL; 524 } 525 526 /* 527 * Now that all data structures that refer to modules are torn down, 528 * no modules should be remaining on the module list except for d_rmod. 529 * If we trip one of these assertions, we're missing a rele somewhere. 530 */ 531 ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod); 532 ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod); 533 534 /* 535 * Now destroy the root module. We clear its thread key first so any 536 * calls to fmd_trace() inside of the module code will be ignored. 537 */ 538 (void) pthread_setspecific(dp->d_key, NULL); 539 fmd_module_lock(dp->d_rmod); 540 541 while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL) 542 fmd_case_discard(cp); 543 544 fmd_module_unlock(dp->d_rmod); 545 fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t)); 546 dp->d_rmod->mod_stats = NULL; 547 548 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 549 dp->d_rmod->mod_flags |= FMD_MOD_FINI; 550 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 551 552 fmd_module_rele(dp->d_rmod); 553 ASSERT(fmd_list_next(&dp->d_mod_list) == NULL); 554 555 /* 556 * Now destroy the remaining global data structures. If 'core' was 557 * set to true, force a core dump so we can check for memory leaks. 558 */ 559 if (dp->d_cases != NULL) 560 fmd_case_hash_destroy(dp->d_cases); 561 if (dp->d_disp != NULL) 562 fmd_dispq_destroy(dp->d_disp); 563 if (dp->d_timers != NULL) 564 fmd_timerq_destroy(dp->d_timers); 565 if (dp->d_schemes != NULL) 566 fmd_scheme_hash_destroy(dp->d_schemes); 567 if (dp->d_xprt_ids != NULL) 568 fmd_idspace_destroy(dp->d_xprt_ids); 569 570 if (dp->d_errstats != NULL) { 571 fmd_free(dp->d_errstats, 572 sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN)); 573 } 574 575 if (dp->d_conf != NULL) 576 fmd_conf_close(dp->d_conf); 577 578 nvlist_free(dp->d_auth); 579 (void) nv_alloc_fini(&dp->d_nva); 580 dp->d_clockops->fto_fini(dp->d_clockptr); 581 582 (void) pthread_key_delete(dp->d_key); 583 bzero(dp, sizeof (fmd_t)); 584 585 if (core) 586 fmd_panic("forcing core dump at user request\n"); 587 } 588 589 /*ARGSUSED*/ 590 static void 591 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt) 592 { 593 hrtime_t delta; 594 595 if (id != 0) { 596 TRACE((FMD_DBG_MOD, "garbage collect start")); 597 fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc); 598 TRACE((FMD_DBG_MOD, "garbage collect end")); 599 600 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 601 fmd_log_update(dp->d_errlog); 602 (void) pthread_rwlock_unlock(&dp->d_log_lock); 603 } 604 605 (void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta); 606 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 607 (fmd_timer_f *)fmd_gc, dp, NULL, delta); 608 } 609 610 /* 611 * Events are committed to the errlog after cases are checkpointed. If fmd 612 * crashes before an event is ever associated with a module, this function will 613 * be called to replay it to all subscribers. If fmd crashes in between the 614 * subscriber checkpointing and committing the event in the error log, the 615 * module will have seen the event and we don't want to replay it. So we look 616 * for the event in all modules and transition it to the proper state. If 617 * it is found, we commit it to the error log and do not replay it. The in- 618 * memory case search used by fmd_module_contains() et al isn't particularly 619 * efficient, but it is faster than doing read i/o's on every case event to 620 * check their status or write i/o's on every event to replay to update states. 621 * We can improve the efficiency of this lookup algorithm later if necessary. 622 */ 623 /*ARGSUSED*/ 624 static void 625 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp) 626 { 627 fmd_module_t *mp; 628 fmd_stat_t *sp; 629 630 (void) pthread_mutex_lock(&dp->d_mod_lock); 631 632 for (mp = fmd_list_next(&dp->d_mod_list); 633 mp != NULL; mp = fmd_list_next(mp)) { 634 if (fmd_module_contains(mp, ep)) { 635 fmd_module_hold(mp); 636 break; 637 } 638 } 639 640 (void) pthread_mutex_unlock(&dp->d_mod_lock); 641 642 if (mp != NULL) { 643 fmd_event_commit(ep); 644 fmd_module_rele(mp); 645 sp = &dp->d_stats->ds_log_partials; 646 } else { 647 fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep)); 648 sp = &dp->d_stats->ds_log_replayed; 649 } 650 651 (void) pthread_mutex_lock(&dp->d_stats_lock); 652 sp->fmds_value.ui64++; 653 (void) pthread_mutex_unlock(&dp->d_stats_lock); 654 } 655 656 void 657 fmd_door_server(void *dip) 658 { 659 fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip); 660 (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 661 (void) door_return(NULL, 0, NULL, 0); 662 } 663 664 /* 665 * Custom door server create callback. Any fmd services that use doors will 666 * require those threads to have their fmd-specific TSD initialized, etc. 667 */ 668 static void 669 fmd_door(door_info_t *dip) 670 { 671 if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL) 672 fmd_panic("failed to create server for door %p", (void *)dip); 673 } 674 675 /* 676 * This signal handler is installed for the client.thrsig signal to be used to 677 * force an auxiliary thread to wake up from a system call and return EINTR in 678 * response to a module's use of fmd_thr_signal(). We also trace the event. 679 */ 680 static void 681 fmd_signal(int sig) 682 { 683 TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig)); 684 } 685 686 void 687 fmd_run(fmd_t *dp, int pfd) 688 { 689 char *nodc_key[] = { FMD_FLT_NODC, NULL }; 690 char nodc_str[128]; 691 struct sigaction act; 692 693 int status = FMD_EXIT_SUCCESS; 694 const char *name; 695 fmd_conf_path_t *pap; 696 fmd_event_t *e; 697 int dbout; 698 699 /* 700 * Cache all the current debug property settings in d_fmd_debug, 701 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout. If a given debug mask 702 * is non-zero and the corresponding dbout mask is zero, set dbout 703 * to a sensible default value based on whether we have daemonized. 704 */ 705 (void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout); 706 707 if (dp->d_fmd_debug != 0 && dbout == 0) 708 dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 709 else 710 dp->d_fmd_dbout = dbout; 711 712 (void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug); 713 (void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout); 714 715 if (dp->d_hdl_debug != 0 && dbout == 0) 716 dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 717 else 718 dp->d_hdl_dbout = dbout; 719 720 /* 721 * Initialize remaining major program data structures such as the 722 * clock, dispatch queues, log files, module hash collections, etc. 723 * This work is done here rather than in fmd_create() to permit the -o 724 * command-line option to modify properties after fmd_create() is done. 725 */ 726 dp->d_clockptr = dp->d_clockops->fto_init(); 727 dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX); 728 fmd_xprt_suspend_all(); 729 730 (void) door_server_create(fmd_door); 731 fmd_dr_init(); 732 733 dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16); 734 dp->d_timers = fmd_timerq_create(); 735 dp->d_disp = fmd_dispq_create(); 736 dp->d_cases = fmd_case_hash_create(); 737 738 /* 739 * The root module's mod_queue is created with limit zero, making it 740 * act like /dev/null; anything inserted here is simply ignored. 741 */ 742 dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod, 743 &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0); 744 745 /* 746 * Once our subsystems that use signals have been set up, install the 747 * signal handler for the fmd_thr_signal() API. Verify that the signal 748 * being used for this purpose doesn't conflict with something else. 749 */ 750 (void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig); 751 752 if (sigaction(dp->d_thr_sig, NULL, &act) != 0) { 753 fmd_error(EFMD_EXIT, "invalid signal selected for " 754 "client.thrsig property: %d\n", dp->d_thr_sig); 755 } 756 757 if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) { 758 fmd_error(EFMD_EXIT, "signal selected for client.thrsig " 759 "property is already in use: %d\n", dp->d_thr_sig); 760 } 761 762 act.sa_handler = fmd_signal; 763 act.sa_flags = 0; 764 765 (void) sigemptyset(&act.sa_mask); 766 (void) sigaction(dp->d_thr_sig, &act, NULL); 767 768 (void) fmd_conf_getprop(dp->d_conf, "schemedir", &name); 769 dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name); 770 771 (void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name); 772 dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name); 773 774 (void) fmd_conf_getprop(dp->d_conf, "log.error", &name); 775 dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR); 776 777 (void) fmd_conf_getprop(dp->d_conf, "log.fault", &name); 778 dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT); 779 780 if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL) 781 fmd_error(EFMD_EXIT, "failed to initialize log files\n"); 782 783 /* 784 * Before loading modules, create an empty control event which will act 785 * as a global barrier for module event processing. Each module we 786 * load successfully will insert it at their head of their event queue, 787 * and then pause inside of fmd_ctl_rele() after dequeuing the event. 788 * This module barrier is required for two reasons: 789 * 790 * (a) During module loading, the restoration of case checkpoints may 791 * result in a list.* event being recreated for which the intended 792 * subscriber has not yet loaded depending on the load order. Such 793 * events could then result in spurious "no subscriber" errors. 794 * 795 * (b) During errlog replay, a sequence of errors from a long time ago 796 * may be replayed, and the module may attempt to install relative 797 * timers associated with one or more of these events. If errlog 798 * replay were "racing" with active module threads, an event E1 799 * that resulted in a relative timer T at time E1 + N nsec could 800 * fire prior to an event E2 being enqueued, even if the relative 801 * time ordering was E1 < E2 < E1 + N, causing mis-diagnosis. 802 */ 803 dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL, 804 FMD_HRT_NOW, NULL, fmd_ctl_init(NULL)); 805 806 fmd_event_hold(e); 807 808 /* 809 * Once all data structures are initialized, we load all of our modules 810 * in order according to class in order to load up any subscriptions. 811 * Once built-in modules are loaded, we detach from our waiting parent. 812 */ 813 dp->d_mod_hash = fmd_modhash_create(); 814 815 if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg) 816 fmd_error(EFMD_EXIT, "failed to initialize fault manager\n"); 817 818 (void) fmd_conf_getprop(dp->d_conf, "self.name", &name); 819 dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name); 820 821 if (dp->d_self != NULL && fmd_module_dc_key2code(dp->d_self, 822 nodc_key, nodc_str, sizeof (nodc_str)) == 0) 823 (void) fmd_conf_setprop(dp->d_conf, "nodiagcode", nodc_str); 824 825 fmd_rpc_init(); 826 dp->d_running = 1; /* we are now officially an active fmd */ 827 828 /* 829 * Now that we're running, if a pipe fd was specified, write an exit 830 * status to it to indicate that our parent process can safely detach. 831 * Then proceed to loading the remaining non-built-in modules. 832 */ 833 if (pfd >= 0) 834 (void) write(pfd, &status, sizeof (status)); 835 836 (void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap); 837 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops); 838 839 (void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap); 840 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops); 841 842 /* 843 * With all modules loaded, replay fault events from the ASRU cache for 844 * any ASRUs that must be retired, replay error events from the errlog 845 * that did not finish processing the last time ran, and then release 846 * the global module barrier by executing a final rele on d_mod_event. 847 */ 848 fmd_asru_hash_refresh(dp->d_asrus); 849 850 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 851 fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp); 852 fmd_log_update(dp->d_errlog); 853 (void) pthread_rwlock_unlock(&dp->d_log_lock); 854 855 dp->d_mod_event = NULL; 856 fmd_event_rele(e); 857 858 /* 859 * Finally, awaken any threads associated with receiving events from 860 * open transports and tell them to proceed with fmd_xprt_recv(). 861 */ 862 fmd_xprt_resume_all(); 863 fmd_gc(dp, 0, 0); 864 865 dp->d_booted = 1; 866 } 867 868 void 869 fmd_help(fmd_t *dp) 870 { 871 const fmd_conf_mode_t *cmp; 872 873 (void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname); 874 875 for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++) 876 (void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc); 877 } 878