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 #include <fm/libtopo.h> 36 37 #include <smbios.h> 38 #include <limits.h> 39 #include <unistd.h> 40 #include <signal.h> 41 #include <stdlib.h> 42 #include <stdio.h> 43 #include <door.h> 44 45 #include <fmd_conf.h> 46 #include <fmd_dispq.h> 47 #include <fmd_timerq.h> 48 #include <fmd_subr.h> 49 #include <fmd_error.h> 50 #include <fmd_module.h> 51 #include <fmd_thread.h> 52 #include <fmd_alloc.h> 53 #include <fmd_string.h> 54 #include <fmd_builtin.h> 55 #include <fmd_ustat.h> 56 #include <fmd_protocol.h> 57 #include <fmd_scheme.h> 58 #include <fmd_asru.h> 59 #include <fmd_case.h> 60 #include <fmd_log.h> 61 #include <fmd_idspace.h> 62 #include <fmd_rpc.h> 63 #include <fmd_dr.h> 64 #include <fmd_xprt.h> 65 #include <fmd_ctl.h> 66 67 #include <fmd.h> 68 69 extern const nv_alloc_ops_t fmd_nv_alloc_ops; /* see fmd_nv.c */ 70 71 const char _fmd_version[] = "1.1"; /* daemon version string */ 72 static char _fmd_plat[MAXNAMELEN]; /* native platform string */ 73 static char _fmd_isa[MAXNAMELEN]; /* native instruction set */ 74 static struct utsname _fmd_uts; /* native uname(2) info */ 75 static char _fmd_csn[MAXNAMELEN]; /* chassis serial number */ 76 static char _fmd_prod[MAXNAMELEN]; /* product name string */ 77 78 /* 79 * Note: the configuration file path is ordered from most common to most host- 80 * specific because new conf files are merged/override previous ones. The 81 * module paths are in the opposite order, from most specific to most common, 82 * because once a module is loaded fmd will not try to load over the same name. 83 */ 84 85 static const char _fmd_conf_path[] = 86 "%r/usr/lib/fm/fmd:" 87 "%r/usr/platform/%m/lib/fm/fmd:" 88 "%r/usr/platform/%i/lib/fm/fmd:" 89 "%r/etc/fm/fmd"; 90 91 static const char _fmd_agent_path[] = 92 "%r/usr/platform/%i/lib/fm/fmd/agents:" 93 "%r/usr/platform/%m/lib/fm/fmd/agents:" 94 "%r/usr/lib/fm/fmd/agents"; 95 96 static const char _fmd_plugin_path[] = 97 "%r/usr/platform/%i/lib/fm/fmd/plugins:" 98 "%r/usr/platform/%m/lib/fm/fmd/plugins:" 99 "%r/usr/lib/fm/fmd/plugins"; 100 101 static const char _fmd_scheme_path[] = 102 "usr/lib/fm/fmd/schemes"; 103 104 static const fmd_conf_mode_t _fmd_cerror_modes[] = { 105 { "unload", "unload offending client module", FMD_CERROR_UNLOAD }, 106 { "stop", "stop daemon for debugger attach", FMD_CERROR_STOP }, 107 { "abort", "abort daemon and force core dump", FMD_CERROR_ABORT }, 108 { NULL, NULL, 0 } 109 }; 110 111 static const fmd_conf_mode_t _fmd_dbout_modes[] = { 112 { "stderr", "send debug messages to stderr", FMD_DBOUT_STDERR }, 113 { "syslog", "send debug messages to syslog", FMD_DBOUT_SYSLOG }, 114 { NULL, NULL, 0 } 115 }; 116 117 static const fmd_conf_mode_t _fmd_debug_modes[] = { 118 { "help", "display debugging modes and exit", FMD_DBG_HELP }, 119 { "mod", "debug module load/unload/locking", FMD_DBG_MOD }, 120 { "disp", "debug dispatch queue processing", FMD_DBG_DISP }, 121 { "xprt", "debug transport-specific routines", FMD_DBG_XPRT }, 122 { "evt", "debug event subsystem routines", FMD_DBG_EVT }, 123 { "log", "debug log subsystem routines", FMD_DBG_LOG }, 124 { "tmr", "debug timer subsystem routines", FMD_DBG_TMR }, 125 { "fmri", "debug fmri subsystem routines", FMD_DBG_FMRI }, 126 { "asru", "debug asru subsystem routines", FMD_DBG_ASRU }, 127 { "case", "debug case subsystem routines", FMD_DBG_CASE }, 128 { "ckpt", "debug checkpoint routines", FMD_DBG_CKPT }, 129 { "rpc", "debug rpc service routines", FMD_DBG_RPC }, 130 { "trace", "display matching trace calls", FMD_DBG_TRACE }, 131 { "all", "enable all available debug modes", FMD_DBG_ALL }, 132 { NULL, NULL, 0 } 133 }; 134 135 static int 136 fmd_cerror_set(fmd_conf_param_t *pp, const char *value) 137 { 138 return (fmd_conf_mode_set(_fmd_cerror_modes, pp, value)); 139 } 140 141 static int 142 fmd_dbout_set(fmd_conf_param_t *pp, const char *value) 143 { 144 return (fmd_conf_mode_set(_fmd_dbout_modes, pp, value)); 145 } 146 147 static int 148 fmd_debug_set(fmd_conf_param_t *pp, const char *value) 149 { 150 int err = fmd_conf_mode_set(_fmd_debug_modes, pp, value); 151 152 if (err == 0) 153 fmd.d_fmd_debug = pp->cp_value.cpv_num; 154 155 return (err); 156 } 157 158 static int 159 fmd_trmode_set(fmd_conf_param_t *pp, const char *value) 160 { 161 fmd_tracebuf_f *func; 162 163 if (strcasecmp(value, "none") == 0) 164 func = fmd_trace_none; 165 else if (strcasecmp(value, "lite") == 0) 166 func = fmd_trace_lite; 167 else if (strcasecmp(value, "full") == 0) 168 func = fmd_trace_full; 169 else 170 return (fmd_set_errno(EFMD_CONF_INVAL)); 171 172 fmd.d_thr_trace = (void (*)())func; 173 pp->cp_value.cpv_ptr = (void *)func; 174 return (0); 175 } 176 177 static void 178 fmd_trmode_get(const fmd_conf_param_t *pp, void *ptr) 179 { 180 *((void **)ptr) = pp->cp_value.cpv_ptr; 181 } 182 183 static int 184 fmd_clkmode_set(fmd_conf_param_t *pp, const char *value) 185 { 186 const fmd_timeops_t *ops; 187 188 if (strcasecmp(value, "native") == 0) 189 ops = &fmd_timeops_native; 190 else if (strcasecmp(value, "simulated") == 0) 191 ops = &fmd_timeops_simulated; 192 else 193 return (fmd_set_errno(EFMD_CONF_INVAL)); 194 195 fmd.d_clockops = ops; 196 pp->cp_value.cpv_ptr = (void *)ops; 197 return (0); 198 } 199 200 static void 201 fmd_clkmode_get(const fmd_conf_param_t *pp, void *ptr) 202 { 203 *((void **)ptr) = pp->cp_value.cpv_ptr; 204 } 205 206 static const fmd_conf_ops_t fmd_cerror_ops = { 207 fmd_cerror_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 208 }; 209 210 static const fmd_conf_ops_t fmd_dbout_ops = { 211 fmd_dbout_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 212 }; 213 214 static const fmd_conf_ops_t fmd_debug_ops = { 215 fmd_debug_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 216 }; 217 218 static const fmd_conf_ops_t fmd_trmode_ops = { 219 fmd_trmode_set, fmd_trmode_get, fmd_conf_notsup, fmd_conf_nop 220 }; 221 222 static const fmd_conf_ops_t fmd_clkmode_ops = { 223 fmd_clkmode_set, fmd_clkmode_get, fmd_conf_notsup, fmd_conf_nop 224 }; 225 226 static const fmd_conf_formal_t _fmd_conf[] = { 227 { "agent.path", &fmd_conf_path, _fmd_agent_path }, /* path for agents */ 228 { "alloc_msecs", &fmd_conf_uint32, "10" }, /* msecs before alloc retry */ 229 { "alloc_tries", &fmd_conf_uint32, "3" }, /* max # of alloc retries */ 230 { "chassis", &fmd_conf_string, _fmd_csn }, /* chassis serial number */ 231 { "ckpt.dir", &fmd_conf_string, "var/fm/fmd/ckpt" }, /* ckpt directory path */ 232 { "ckpt.dirmode", &fmd_conf_int32, "0700" }, /* ckpt directory perm mode */ 233 { "ckpt.mode", &fmd_conf_int32, "0400" }, /* ckpt file perm mode */ 234 { "ckpt.restore", &fmd_conf_bool, "true" }, /* restore checkpoints? */ 235 { "ckpt.save", &fmd_conf_bool, "true" }, /* save checkpoints? */ 236 { "ckpt.zero", &fmd_conf_bool, "false" }, /* zero checkpoints on start? */ 237 { "client.buflim", &fmd_conf_size, "10m" }, /* client buffer space limit */ 238 { "client.dbout", &fmd_dbout_ops, NULL }, /* client debug output sinks */ 239 { "client.debug", &fmd_conf_bool, NULL }, /* client debug enable */ 240 { "client.error", &fmd_cerror_ops, "unload" }, /* client error policy */ 241 { "client.memlim", &fmd_conf_size, "10m" }, /* client allocation limit */ 242 { "client.evqlim", &fmd_conf_uint32, "256" }, /* client event queue limit */ 243 { "client.thrlim", &fmd_conf_uint32, "8" }, /* client aux thread limit */ 244 { "client.thrsig", &fmd_conf_signal, "SIGUSR1" }, /* fmd_thr_signal() value */ 245 { "client.tmrlim", &fmd_conf_uint32, "1024" }, /* client pending timer limit */ 246 { "client.xprtlim", &fmd_conf_uint32, "256" }, /* client transport limit */ 247 { "client.xprtlog", &fmd_conf_bool, NULL }, /* client transport logging? */ 248 { "client.xprtqlim", &fmd_conf_uint32, "256" }, /* client transport queue lim */ 249 { "clock", &fmd_clkmode_ops, "native" }, /* clock operation mode */ 250 { "conf_path", &fmd_conf_path, _fmd_conf_path }, /* root config file path */ 251 { "conf_file", &fmd_conf_string, "fmd.conf" }, /* root config file name */ 252 { "core", &fmd_conf_bool, "false" }, /* force core dump on quit */ 253 { "dbout", &fmd_dbout_ops, NULL }, /* daemon debug output sinks */ 254 { "debug", &fmd_debug_ops, NULL }, /* daemon debugging flags */ 255 { "dictdir", &fmd_conf_string, "usr/lib/fm/dict" }, /* default diagcode dir */ 256 { "domain", &fmd_conf_string, NULL }, /* domain id for de auth */ 257 { "fg", &fmd_conf_bool, "false" }, /* run daemon in foreground */ 258 { "gc_interval", &fmd_conf_time, "1d" }, /* garbage collection intvl */ 259 { "ids.avg", &fmd_conf_uint32, "4" }, /* desired idspace chain len */ 260 { "ids.max", &fmd_conf_uint32, "1024" }, /* maximum idspace buckets */ 261 { "isaname", &fmd_conf_string, _fmd_isa }, /* instruction set (uname -p) */ 262 { "log.creator", &fmd_conf_string, "fmd" }, /* exacct log creator string */ 263 { "log.error", &fmd_conf_string, "var/fm/fmd/errlog" }, /* error log path */ 264 { "log.fault", &fmd_conf_string, "var/fm/fmd/fltlog" }, /* fault log path */ 265 { "log.minfree", &fmd_conf_size, "2m" }, /* min log fsys free space */ 266 { "log.rsrc", &fmd_conf_string, "var/fm/fmd/rsrc" }, /* asru log dir path */ 267 { "log.tryrotate", &fmd_conf_uint32, "10" }, /* max log rotation attempts */ 268 { "log.waitrotate", &fmd_conf_time, "200ms" }, /* log rotation retry delay */ 269 { "log.xprt", &fmd_conf_string, "var/fm/fmd/xprt" }, /* transport log dir */ 270 { "machine", &fmd_conf_string, _fmd_uts.machine }, /* machine name (uname -m) */ 271 { "nodiagcode", &fmd_conf_string, "-" }, /* diagcode to use if error */ 272 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */ 273 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */ 274 { "platform", &fmd_conf_string, _fmd_plat }, /* platform string (uname -i) */ 275 { "plugin.close", &fmd_conf_bool, "true" }, /* dlclose plugins on fini */ 276 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */ 277 { "product", &fmd_conf_string, _fmd_prod }, /* product name string */ 278 { "rootdir", &fmd_conf_string, "" }, /* root directory for paths */ 279 { "rpc.adm.path", &fmd_conf_string, NULL }, /* FMD_ADM rendezvous file */ 280 { "rpc.adm.prog", &fmd_conf_uint32, "100169" }, /* FMD_ADM rpc program num */ 281 { "rpc.api.path", &fmd_conf_string, NULL }, /* FMD_API rendezvous file */ 282 { "rpc.api.prog", &fmd_conf_uint32, "100170" }, /* FMD_API rpc program num */ 283 { "rpc.rcvsize", &fmd_conf_size, "128k" }, /* rpc receive buffer size */ 284 { "rpc.sndsize", &fmd_conf_size, "128k" }, /* rpc send buffer size */ 285 { "rsrc.age", &fmd_conf_time, "30d" }, /* max age of old rsrc log */ 286 { "rsrc.zero", &fmd_conf_bool, "false" }, /* zero rsrc cache on start? */ 287 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */ 288 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */ 289 { "self.dict", &fmd_conf_list, "FMD.dict" }, /* self-diag dictionary list */ 290 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */ 291 { "strbuckets", &fmd_conf_uint32, "211" }, /* size of string hashes */ 292 #ifdef DEBUG 293 { "trace.mode", &fmd_trmode_ops, "full" }, /* trace mode: none/lite/full */ 294 #else 295 { "trace.mode", &fmd_trmode_ops, "lite" }, /* trace mode: none/lite/full */ 296 #endif 297 { "trace.recs", &fmd_conf_uint32, "128" }, /* trace records per thread */ 298 { "trace.frames", &fmd_conf_uint32, "16" }, /* max trace rec stack frames */ 299 { "uuidlen", &fmd_conf_uint32, "36" }, /* UUID ASCII string length */ 300 { "xprt.ttl", &fmd_conf_uint8, "1" }, /* default event time-to-live */ 301 }; 302 303 /* 304 * Statistics maintained by fmd itself on behalf of various global subsystems. 305 * NOTE: FMD_TYPE_STRING statistics should not be used here. If they are 306 * required in the future, the FMD_ADM_MODGSTAT service routine must change. 307 */ 308 static fmd_statistics_t _fmd_stats = { 309 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" }, 310 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" }, 311 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" }, 312 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" }, 313 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" }, 314 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" }, 315 }; 316 317 void 318 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf) 319 { 320 fmd_conf_path_t *pap; 321 char file[PATH_MAX]; 322 const char *name; 323 fmd_stat_t *sp; 324 int i; 325 326 smbios_hdl_t *shp; 327 smbios_system_t s1; 328 smbios_info_t s2; 329 id_t id; 330 331 (void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat)); 332 (void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa)); 333 (void) uname(&_fmd_uts); 334 335 if ((shp = smbios_open(NULL, SMB_VERSION, 0, NULL)) != NULL) { 336 if ((id = smbios_info_system(shp, &s1)) != SMB_ERR && 337 smbios_info_common(shp, id, &s2) != SMB_ERR) { 338 (void) strlcpy(_fmd_prod, s2.smbi_product, MAXNAMELEN); 339 (void) strlcpy(_fmd_csn, s2.smbi_serial, MAXNAMELEN); 340 } 341 smbios_close(shp); 342 } 343 344 bzero(dp, sizeof (fmd_t)); 345 346 dp->d_version = _fmd_version; 347 dp->d_pname = fmd_strbasename(arg0); 348 dp->d_pid = getpid(); 349 350 if (pthread_key_create(&dp->d_key, NULL) != 0) 351 fmd_error(EFMD_EXIT, "failed to create pthread key"); 352 353 (void) pthread_mutex_init(&dp->d_xprt_lock, NULL); 354 (void) pthread_mutex_init(&dp->d_err_lock, NULL); 355 (void) pthread_mutex_init(&dp->d_thr_lock, NULL); 356 (void) pthread_mutex_init(&dp->d_mod_lock, NULL); 357 (void) pthread_mutex_init(&dp->d_stats_lock, NULL); 358 (void) pthread_rwlock_init(&dp->d_log_lock, NULL); 359 360 /* 361 * A small number of properties must be set manually before we open 362 * the root configuration file. These include any settings for our 363 * memory allocator and path expansion token values, because these 364 * values are needed by the routines in fmd_conf.c itself. After 365 * the root configuration file is processed, we reset these properties 366 * based upon the latest values from the configuration file. 367 */ 368 dp->d_alloc_msecs = 10; 369 dp->d_alloc_tries = 3; 370 dp->d_str_buckets = 211; 371 372 dp->d_rootdir = root ? root : ""; 373 dp->d_platform = _fmd_plat; 374 dp->d_machine = _fmd_uts.machine; 375 dp->d_isaname = _fmd_isa; 376 377 dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) / 378 sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER); 379 380 if (dp->d_conf == NULL) { 381 fmd_error(EFMD_EXIT, 382 "failed to load required configuration properties\n"); 383 } 384 385 (void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs); 386 (void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries); 387 (void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets); 388 389 (void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform); 390 (void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine); 391 (void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname); 392 393 /* 394 * Manually specified rootdirs override config files, so only update 395 * d_rootdir based on the config files we parsed if no 'root' was set. 396 */ 397 if (root == NULL) 398 (void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir); 399 else 400 (void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir); 401 402 /* 403 * Once the base conf file properties are loaded, lookup the values 404 * of $conf_path and $conf_file and merge in any other conf files. 405 */ 406 (void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap); 407 (void) fmd_conf_getprop(dp->d_conf, "conf_file", &name); 408 409 for (i = 0; i < pap->cpa_argc; i++) { 410 (void) snprintf(file, sizeof (file), 411 "%s/%s", pap->cpa_argv[i], name); 412 if (access(file, F_OK) == 0) 413 fmd_conf_merge(dp->d_conf, file); 414 } 415 416 /* 417 * Update the value of fmd.d_fg based on "fg". We cache this property 418 * because it must be accessed deep within fmd at fmd_verror() time. 419 * Update any other properties that must be cached for performance. 420 */ 421 (void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg); 422 (void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl); 423 424 /* 425 * Initialize our custom libnvpair allocator and create an nvlist for 426 * authority elements corresponding to this instance of the daemon. 427 */ 428 (void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops); 429 dp->d_auth = fmd_protocol_authority(); 430 431 /* 432 * The fmd_module_t for the root module must be created manually. Most 433 * of it remains unused and zero, except for the few things we fill in. 434 */ 435 dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP); 436 dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP); 437 dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod); 438 439 fmd_list_append(&dp->d_mod_list, dp->d_rmod); 440 fmd_module_hold(dp->d_rmod); 441 442 (void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL); 443 (void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL); 444 (void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL); 445 446 dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self()); 447 dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP); 448 dp->d_rmod->mod_ustat = fmd_ustat_create(); 449 450 if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0) 451 fmd_error(EFMD_EXIT, "failed to attach main thread key"); 452 453 if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert( 454 dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) / 455 sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL) 456 fmd_error(EFMD_EXIT, "failed to initialize statistics"); 457 458 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 459 dp->d_rmod->mod_flags |= FMD_MOD_INIT; 460 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 461 462 /* 463 * In addition to inserting the _fmd_stats collection of program-wide 464 * statistics, we also insert a statistic named after each of our 465 * errors and update these counts in fmd_verror() (see fmd_subr.c). 466 */ 467 dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) * 468 (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP); 469 470 for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) { 471 (void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s", 472 strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1); 473 sp->fmds_type = FMD_TYPE_UINT64; 474 } 475 476 (void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, 477 EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL); 478 } 479 480 void 481 fmd_destroy(fmd_t *dp) 482 { 483 fmd_module_t *mp; 484 fmd_case_t *cp; 485 int core; 486 487 (void) fmd_conf_getprop(fmd.d_conf, "core", &core); 488 489 fmd_rpc_fini(); 490 fmd_dr_fini(); 491 492 if (dp->d_xprt_ids != NULL) 493 fmd_xprt_suspend_all(); 494 495 /* 496 * Unload the self-diagnosis module first. This ensures that it does 497 * not get confused as we start unloading other modules, etc. We must 498 * hold the dispq lock as a writer while doing so since it uses d_self. 499 */ 500 if (dp->d_self != NULL) { 501 (void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock); 502 fmd_module_unload(dp->d_self); 503 fmd_module_rele(dp->d_self); 504 dp->d_self = NULL; 505 (void) pthread_rwlock_unlock(&dp->d_disp->dq_lock); 506 } 507 508 /* 509 * Unload modules in reverse order *except* for the root module, which 510 * is first in the list. This allows it to keep its thread and trace. 511 */ 512 for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) { 513 fmd_module_unload(mp); 514 mp = fmd_list_prev(mp); 515 } 516 517 if (dp->d_mod_hash != NULL) { 518 fmd_modhash_destroy(dp->d_mod_hash); 519 dp->d_mod_hash = NULL; 520 } 521 522 /* 523 * Close both log files now that modules are no longer active. We must 524 * set these pointers to NULL in case any subsequent errors occur. 525 */ 526 if (dp->d_errlog != NULL) { 527 fmd_log_rele(dp->d_errlog); 528 dp->d_errlog = NULL; 529 } 530 531 if (dp->d_fltlog != NULL) { 532 fmd_log_rele(dp->d_fltlog); 533 dp->d_fltlog = NULL; 534 } 535 536 /* 537 * Now destroy the resource cache: each ASRU contains a case reference, 538 * which may in turn contain a pointer to a referenced owning module. 539 */ 540 if (dp->d_asrus != NULL) { 541 fmd_asru_hash_destroy(dp->d_asrus); 542 dp->d_asrus = NULL; 543 } 544 545 /* 546 * Now that all data structures that refer to modules are torn down, 547 * no modules should be remaining on the module list except for d_rmod. 548 * If we trip one of these assertions, we're missing a rele somewhere. 549 */ 550 ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod); 551 ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod); 552 553 /* 554 * Now destroy the root module. We clear its thread key first so any 555 * calls to fmd_trace() inside of the module code will be ignored. 556 */ 557 (void) pthread_setspecific(dp->d_key, NULL); 558 fmd_module_lock(dp->d_rmod); 559 560 while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL) 561 fmd_case_discard(cp); 562 563 fmd_module_unlock(dp->d_rmod); 564 fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t)); 565 dp->d_rmod->mod_stats = NULL; 566 567 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 568 dp->d_rmod->mod_flags |= FMD_MOD_FINI; 569 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 570 571 fmd_module_rele(dp->d_rmod); 572 ASSERT(fmd_list_next(&dp->d_mod_list) == NULL); 573 574 /* 575 * Now destroy the remaining global data structures. If 'core' was 576 * set to true, force a core dump so we can check for memory leaks. 577 */ 578 if (dp->d_cases != NULL) 579 fmd_case_hash_destroy(dp->d_cases); 580 if (dp->d_disp != NULL) 581 fmd_dispq_destroy(dp->d_disp); 582 if (dp->d_timers != NULL) 583 fmd_timerq_destroy(dp->d_timers); 584 if (dp->d_schemes != NULL) 585 fmd_scheme_hash_destroy(dp->d_schemes); 586 if (dp->d_xprt_ids != NULL) 587 fmd_idspace_destroy(dp->d_xprt_ids); 588 589 if (dp->d_errstats != NULL) { 590 fmd_free(dp->d_errstats, 591 sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN)); 592 } 593 594 if (dp->d_conf != NULL) 595 fmd_conf_close(dp->d_conf); 596 597 if (dp->d_topo != NULL) 598 topo_close(dp->d_topo); 599 600 nvlist_free(dp->d_auth); 601 (void) nv_alloc_fini(&dp->d_nva); 602 dp->d_clockops->fto_fini(dp->d_clockptr); 603 604 (void) pthread_key_delete(dp->d_key); 605 bzero(dp, sizeof (fmd_t)); 606 607 if (core) 608 fmd_panic("forcing core dump at user request\n"); 609 } 610 611 /*ARGSUSED*/ 612 static void 613 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt) 614 { 615 hrtime_t delta; 616 617 if (id != 0) { 618 TRACE((FMD_DBG_MOD, "garbage collect start")); 619 fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc); 620 TRACE((FMD_DBG_MOD, "garbage collect end")); 621 622 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 623 fmd_log_update(dp->d_errlog); 624 (void) pthread_rwlock_unlock(&dp->d_log_lock); 625 } 626 627 (void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta); 628 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 629 (fmd_timer_f *)fmd_gc, dp, NULL, delta); 630 } 631 632 /* 633 * Events are committed to the errlog after cases are checkpointed. If fmd 634 * crashes before an event is ever associated with a module, this function will 635 * be called to replay it to all subscribers. If fmd crashes in between the 636 * subscriber checkpointing and committing the event in the error log, the 637 * module will have seen the event and we don't want to replay it. So we look 638 * for the event in all modules and transition it to the proper state. If 639 * it is found, we commit it to the error log and do not replay it. The in- 640 * memory case search used by fmd_module_contains() et al isn't particularly 641 * efficient, but it is faster than doing read i/o's on every case event to 642 * check their status or write i/o's on every event to replay to update states. 643 * We can improve the efficiency of this lookup algorithm later if necessary. 644 */ 645 /*ARGSUSED*/ 646 static void 647 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp) 648 { 649 fmd_module_t *mp; 650 fmd_stat_t *sp; 651 652 (void) pthread_mutex_lock(&dp->d_mod_lock); 653 654 for (mp = fmd_list_next(&dp->d_mod_list); 655 mp != NULL; mp = fmd_list_next(mp)) { 656 if (fmd_module_contains(mp, ep)) { 657 fmd_module_hold(mp); 658 break; 659 } 660 } 661 662 (void) pthread_mutex_unlock(&dp->d_mod_lock); 663 664 if (mp != NULL) { 665 fmd_event_commit(ep); 666 fmd_module_rele(mp); 667 sp = &dp->d_stats->ds_log_partials; 668 } else { 669 fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep)); 670 sp = &dp->d_stats->ds_log_replayed; 671 } 672 673 (void) pthread_mutex_lock(&dp->d_stats_lock); 674 sp->fmds_value.ui64++; 675 (void) pthread_mutex_unlock(&dp->d_stats_lock); 676 } 677 678 void 679 fmd_door_server(void *dip) 680 { 681 fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip); 682 (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 683 (void) door_return(NULL, 0, NULL, 0); 684 } 685 686 /* 687 * Custom door server create callback. Any fmd services that use doors will 688 * require those threads to have their fmd-specific TSD initialized, etc. 689 */ 690 static void 691 fmd_door(door_info_t *dip) 692 { 693 if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL) 694 fmd_panic("failed to create server for door %p", (void *)dip); 695 } 696 697 /* 698 * This signal handler is installed for the client.thrsig signal to be used to 699 * force an auxiliary thread to wake up from a system call and return EINTR in 700 * response to a module's use of fmd_thr_signal(). We also trace the event. 701 */ 702 static void 703 fmd_signal(int sig) 704 { 705 TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig)); 706 } 707 708 void 709 fmd_run(fmd_t *dp, int pfd) 710 { 711 char *nodc_key[] = { FMD_FLT_NODC, NULL }; 712 char nodc_str[128]; 713 struct sigaction act; 714 715 int status = FMD_EXIT_SUCCESS; 716 const char *name; 717 fmd_conf_path_t *pap; 718 fmd_event_t *e; 719 int dbout, err; 720 721 /* 722 * Cache all the current debug property settings in d_fmd_debug, 723 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout. If a given debug mask 724 * is non-zero and the corresponding dbout mask is zero, set dbout 725 * to a sensible default value based on whether we have daemonized. 726 */ 727 (void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout); 728 729 if (dp->d_fmd_debug != 0 && dbout == 0) 730 dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 731 else 732 dp->d_fmd_dbout = dbout; 733 734 (void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug); 735 (void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout); 736 737 if (dp->d_hdl_debug != 0 && dbout == 0) 738 dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 739 else 740 dp->d_hdl_dbout = dbout; 741 742 /* 743 * Initialize remaining major program data structures such as the 744 * clock, dispatch queues, log files, module hash collections, etc. 745 * This work is done here rather than in fmd_create() to permit the -o 746 * command-line option to modify properties after fmd_create() is done. 747 */ 748 name = dp->d_rootdir != NULL && 749 *dp->d_rootdir != '\0' ? dp->d_rootdir : NULL; 750 751 if ((dp->d_topo = topo_open(TOPO_VERSION, name, &err)) == NULL) { 752 fmd_error(EFMD_EXIT, "failed to initialize " 753 "topology library: %s\n", topo_strerror(err)); 754 } 755 756 dp->d_clockptr = dp->d_clockops->fto_init(); 757 dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX); 758 fmd_xprt_suspend_all(); 759 760 (void) door_server_create(fmd_door); 761 fmd_dr_init(); 762 763 dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16); 764 dp->d_timers = fmd_timerq_create(); 765 dp->d_disp = fmd_dispq_create(); 766 dp->d_cases = fmd_case_hash_create(); 767 768 /* 769 * The root module's mod_queue is created with limit zero, making it 770 * act like /dev/null; anything inserted here is simply ignored. 771 */ 772 dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod, 773 &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0); 774 775 /* 776 * Once our subsystems that use signals have been set up, install the 777 * signal handler for the fmd_thr_signal() API. Verify that the signal 778 * being used for this purpose doesn't conflict with something else. 779 */ 780 (void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig); 781 782 if (sigaction(dp->d_thr_sig, NULL, &act) != 0) { 783 fmd_error(EFMD_EXIT, "invalid signal selected for " 784 "client.thrsig property: %d\n", dp->d_thr_sig); 785 } 786 787 if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) { 788 fmd_error(EFMD_EXIT, "signal selected for client.thrsig " 789 "property is already in use: %d\n", dp->d_thr_sig); 790 } 791 792 act.sa_handler = fmd_signal; 793 act.sa_flags = 0; 794 795 (void) sigemptyset(&act.sa_mask); 796 (void) sigaction(dp->d_thr_sig, &act, NULL); 797 798 (void) fmd_conf_getprop(dp->d_conf, "schemedir", &name); 799 dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name); 800 801 (void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name); 802 dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name); 803 804 (void) fmd_conf_getprop(dp->d_conf, "log.error", &name); 805 dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR); 806 807 (void) fmd_conf_getprop(dp->d_conf, "log.fault", &name); 808 dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT); 809 810 if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL) 811 fmd_error(EFMD_EXIT, "failed to initialize log files\n"); 812 813 /* 814 * Before loading modules, create an empty control event which will act 815 * as a global barrier for module event processing. Each module we 816 * load successfully will insert it at their head of their event queue, 817 * and then pause inside of fmd_ctl_rele() after dequeuing the event. 818 * This module barrier is required for two reasons: 819 * 820 * (a) During module loading, the restoration of case checkpoints may 821 * result in a list.* event being recreated for which the intended 822 * subscriber has not yet loaded depending on the load order. Such 823 * events could then result in spurious "no subscriber" errors. 824 * 825 * (b) During errlog replay, a sequence of errors from a long time ago 826 * may be replayed, and the module may attempt to install relative 827 * timers associated with one or more of these events. If errlog 828 * replay were "racing" with active module threads, an event E1 829 * that resulted in a relative timer T at time E1 + N nsec could 830 * fire prior to an event E2 being enqueued, even if the relative 831 * time ordering was E1 < E2 < E1 + N, causing mis-diagnosis. 832 */ 833 dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL, 834 FMD_HRT_NOW, NULL, fmd_ctl_init(NULL)); 835 836 fmd_event_hold(e); 837 838 /* 839 * Once all data structures are initialized, we load all of our modules 840 * in order according to class in order to load up any subscriptions. 841 * Once built-in modules are loaded, we detach from our waiting parent. 842 */ 843 dp->d_mod_hash = fmd_modhash_create(); 844 845 if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg) 846 fmd_error(EFMD_EXIT, "failed to initialize fault manager\n"); 847 848 (void) fmd_conf_getprop(dp->d_conf, "self.name", &name); 849 dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name); 850 851 if (dp->d_self != NULL && fmd_module_dc_key2code(dp->d_self, 852 nodc_key, nodc_str, sizeof (nodc_str)) == 0) 853 (void) fmd_conf_setprop(dp->d_conf, "nodiagcode", nodc_str); 854 855 fmd_rpc_init(); 856 dp->d_running = 1; /* we are now officially an active fmd */ 857 858 /* 859 * Now that we're running, if a pipe fd was specified, write an exit 860 * status to it to indicate that our parent process can safely detach. 861 * Then proceed to loading the remaining non-built-in modules. 862 */ 863 if (pfd >= 0) 864 (void) write(pfd, &status, sizeof (status)); 865 866 /* 867 * Before loading all modules, repopulate the ASRU cache from its 868 * persistent repository on disk. Then during module loading, the 869 * restoration of checkpoint files will reparent any active cases. 870 */ 871 fmd_asru_hash_refresh(dp->d_asrus); 872 873 (void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap); 874 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops, ".so"); 875 876 (void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap); 877 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops, NULL); 878 879 /* 880 * With all modules loaded, replay fault events from the ASRU cache for 881 * any ASRUs that must be retired, replay error events from the errlog 882 * that did not finish processing the last time ran, and then release 883 * the global module barrier by executing a final rele on d_mod_event. 884 */ 885 fmd_asru_hash_replay(dp->d_asrus); 886 887 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 888 fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp); 889 fmd_log_update(dp->d_errlog); 890 (void) pthread_rwlock_unlock(&dp->d_log_lock); 891 892 dp->d_mod_event = NULL; 893 fmd_event_rele(e); 894 895 /* 896 * Finally, awaken any threads associated with receiving events from 897 * open transports and tell them to proceed with fmd_xprt_recv(). 898 */ 899 fmd_xprt_resume_all(); 900 fmd_gc(dp, 0, 0); 901 902 dp->d_booted = 1; 903 } 904 905 void 906 fmd_help(fmd_t *dp) 907 { 908 const fmd_conf_mode_t *cmp; 909 910 (void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname); 911 912 for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++) 913 (void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc); 914 } 915