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