1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/types.h> 30 #include <sys/utsname.h> 31 #include <sys/param.h> 32 #include <sys/systeminfo.h> 33 #include <sys/fm/util.h> 34 35 #include <smbios.h> 36 #include <limits.h> 37 #include <unistd.h> 38 #include <signal.h> 39 #include <stdlib.h> 40 #include <stdio.h> 41 #include <door.h> 42 43 #include <fmd_conf.h> 44 #include <fmd_dispq.h> 45 #include <fmd_timerq.h> 46 #include <fmd_subr.h> 47 #include <fmd_error.h> 48 #include <fmd_module.h> 49 #include <fmd_thread.h> 50 #include <fmd_alloc.h> 51 #include <fmd_string.h> 52 #include <fmd_builtin.h> 53 #include <fmd_ustat.h> 54 #include <fmd_protocol.h> 55 #include <fmd_scheme.h> 56 #include <fmd_asru.h> 57 #include <fmd_case.h> 58 #include <fmd_log.h> 59 #include <fmd_idspace.h> 60 #include <fmd_rpc.h> 61 #include <fmd_dr.h> 62 #include <fmd_topo.h> 63 #include <fmd_xprt.h> 64 #include <fmd_ctl.h> 65 #include <sys/openpromio.h> 66 #include <libdevinfo.h> 67 68 #include <fmd.h> 69 70 extern const nv_alloc_ops_t fmd_nv_alloc_ops; /* see fmd_nv.c */ 71 72 const char _fmd_version[] = "1.2"; /* daemon version string */ 73 static char _fmd_plat[MAXNAMELEN]; /* native platform string */ 74 static char _fmd_isa[MAXNAMELEN]; /* native instruction set */ 75 static struct utsname _fmd_uts; /* native uname(2) info */ 76 static char _fmd_csn[MAXNAMELEN]; /* chassis serial number */ 77 static char _fmd_prod[MAXNAMELEN]; /* product name string */ 78 79 /* 80 * Note: the configuration file path is ordered from most common to most host- 81 * specific because new conf files are merged/override previous ones. The 82 * module paths are in the opposite order, from most specific to most common, 83 * because once a module is loaded fmd will not try to load over the same name. 84 */ 85 86 static const char _fmd_conf_path[] = 87 "%r/usr/lib/fm/fmd:" 88 "%r/usr/platform/%m/lib/fm/fmd:" 89 "%r/usr/platform/%i/lib/fm/fmd:" 90 "%r/etc/fm/fmd"; 91 92 static const char _fmd_agent_path[] = 93 "%r/usr/platform/%i/lib/fm/fmd/agents:" 94 "%r/usr/platform/%m/lib/fm/fmd/agents:" 95 "%r/usr/lib/fm/fmd/agents"; 96 97 static const char _fmd_plugin_path[] = 98 "%r/usr/platform/%i/lib/fm/fmd/plugins:" 99 "%r/usr/platform/%m/lib/fm/fmd/plugins:" 100 "%r/usr/lib/fm/fmd/plugins"; 101 102 static const char _fmd_scheme_path[] = 103 "usr/lib/fm/fmd/schemes"; 104 105 static const fmd_conf_mode_t _fmd_cerror_modes[] = { 106 { "unload", "unload offending client module", FMD_CERROR_UNLOAD }, 107 { "stop", "stop daemon for debugger attach", FMD_CERROR_STOP }, 108 { "abort", "abort daemon and force core dump", FMD_CERROR_ABORT }, 109 { NULL, NULL, 0 } 110 }; 111 112 static const fmd_conf_mode_t _fmd_dbout_modes[] = { 113 { "stderr", "send debug messages to stderr", FMD_DBOUT_STDERR }, 114 { "syslog", "send debug messages to syslog", FMD_DBOUT_SYSLOG }, 115 { NULL, NULL, 0 } 116 }; 117 118 static const fmd_conf_mode_t _fmd_debug_modes[] = { 119 { "help", "display debugging modes and exit", FMD_DBG_HELP }, 120 { "mod", "debug module load/unload/locking", FMD_DBG_MOD }, 121 { "disp", "debug dispatch queue processing", FMD_DBG_DISP }, 122 { "xprt", "debug transport-specific routines", FMD_DBG_XPRT }, 123 { "evt", "debug event subsystem routines", FMD_DBG_EVT }, 124 { "log", "debug log subsystem routines", FMD_DBG_LOG }, 125 { "tmr", "debug timer subsystem routines", FMD_DBG_TMR }, 126 { "fmri", "debug fmri subsystem routines", FMD_DBG_FMRI }, 127 { "asru", "debug asru subsystem routines", FMD_DBG_ASRU }, 128 { "case", "debug case subsystem routines", FMD_DBG_CASE }, 129 { "ckpt", "debug checkpoint routines", FMD_DBG_CKPT }, 130 { "rpc", "debug rpc service routines", FMD_DBG_RPC }, 131 { "trace", "display matching trace calls", FMD_DBG_TRACE }, 132 { "all", "enable all available debug modes", FMD_DBG_ALL }, 133 { NULL, NULL, 0 } 134 }; 135 136 static int 137 fmd_cerror_set(fmd_conf_param_t *pp, const char *value) 138 { 139 return (fmd_conf_mode_set(_fmd_cerror_modes, pp, value)); 140 } 141 142 static int 143 fmd_dbout_set(fmd_conf_param_t *pp, const char *value) 144 { 145 return (fmd_conf_mode_set(_fmd_dbout_modes, pp, value)); 146 } 147 148 static int 149 fmd_debug_set(fmd_conf_param_t *pp, const char *value) 150 { 151 int err = fmd_conf_mode_set(_fmd_debug_modes, pp, value); 152 153 if (err == 0) 154 fmd.d_fmd_debug = pp->cp_value.cpv_num; 155 156 return (err); 157 } 158 159 static int 160 fmd_trmode_set(fmd_conf_param_t *pp, const char *value) 161 { 162 fmd_tracebuf_f *func; 163 164 if (strcasecmp(value, "none") == 0) 165 func = fmd_trace_none; 166 else if (strcasecmp(value, "lite") == 0) 167 func = fmd_trace_lite; 168 else if (strcasecmp(value, "full") == 0) 169 func = fmd_trace_full; 170 else 171 return (fmd_set_errno(EFMD_CONF_INVAL)); 172 173 fmd.d_thr_trace = (void (*)())func; 174 pp->cp_value.cpv_ptr = (void *)func; 175 return (0); 176 } 177 178 static void 179 fmd_trmode_get(const fmd_conf_param_t *pp, void *ptr) 180 { 181 *((void **)ptr) = pp->cp_value.cpv_ptr; 182 } 183 184 static int 185 fmd_clkmode_set(fmd_conf_param_t *pp, const char *value) 186 { 187 const fmd_timeops_t *ops; 188 189 if (strcasecmp(value, "native") == 0) 190 ops = &fmd_timeops_native; 191 else if (strcasecmp(value, "simulated") == 0) 192 ops = &fmd_timeops_simulated; 193 else 194 return (fmd_set_errno(EFMD_CONF_INVAL)); 195 196 fmd.d_clockops = ops; 197 pp->cp_value.cpv_ptr = (void *)ops; 198 return (0); 199 } 200 201 static void 202 fmd_clkmode_get(const fmd_conf_param_t *pp, void *ptr) 203 { 204 *((void **)ptr) = pp->cp_value.cpv_ptr; 205 } 206 207 static const fmd_conf_ops_t fmd_cerror_ops = { 208 fmd_cerror_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 209 }; 210 211 static const fmd_conf_ops_t fmd_dbout_ops = { 212 fmd_dbout_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 213 }; 214 215 static const fmd_conf_ops_t fmd_debug_ops = { 216 fmd_debug_set, fmd_conf_mode_get, fmd_conf_notsup, fmd_conf_nop 217 }; 218 219 static const fmd_conf_ops_t fmd_trmode_ops = { 220 fmd_trmode_set, fmd_trmode_get, fmd_conf_notsup, fmd_conf_nop 221 }; 222 223 static const fmd_conf_ops_t fmd_clkmode_ops = { 224 fmd_clkmode_set, fmd_clkmode_get, fmd_conf_notsup, fmd_conf_nop 225 }; 226 227 static const fmd_conf_formal_t _fmd_conf[] = { 228 { "agent.path", &fmd_conf_path, _fmd_agent_path }, /* path for agents */ 229 { "alloc_msecs", &fmd_conf_uint32, "10" }, /* msecs before alloc retry */ 230 { "alloc_tries", &fmd_conf_uint32, "3" }, /* max # of alloc retries */ 231 { "chassis", &fmd_conf_string, _fmd_csn }, /* chassis serial number */ 232 { "ckpt.dir", &fmd_conf_string, "var/fm/fmd/ckpt" }, /* ckpt directory path */ 233 { "ckpt.dirmode", &fmd_conf_int32, "0700" }, /* ckpt directory perm mode */ 234 { "ckpt.mode", &fmd_conf_int32, "0400" }, /* ckpt file perm mode */ 235 { "ckpt.restore", &fmd_conf_bool, "true" }, /* restore checkpoints? */ 236 { "ckpt.save", &fmd_conf_bool, "true" }, /* save checkpoints? */ 237 { "ckpt.zero", &fmd_conf_bool, "false" }, /* zero checkpoints on start? */ 238 { "client.buflim", &fmd_conf_size, "10m" }, /* client buffer space limit */ 239 { "client.dbout", &fmd_dbout_ops, NULL }, /* client debug output sinks */ 240 { "client.debug", &fmd_conf_bool, NULL }, /* client debug enable */ 241 { "client.error", &fmd_cerror_ops, "unload" }, /* client error policy */ 242 { "client.memlim", &fmd_conf_size, "10m" }, /* client allocation limit */ 243 { "client.evqlim", &fmd_conf_uint32, "256" }, /* client event queue limit */ 244 { "client.thrlim", &fmd_conf_uint32, "8" }, /* client aux thread limit */ 245 { "client.thrsig", &fmd_conf_signal, "SIGUSR1" }, /* fmd_thr_signal() value */ 246 { "client.tmrlim", &fmd_conf_uint32, "1024" }, /* client pending timer limit */ 247 { "client.xprtlim", &fmd_conf_uint32, "256" }, /* client transport limit */ 248 { "client.xprtlog", &fmd_conf_bool, NULL }, /* client transport logging? */ 249 { "client.xprtqlim", &fmd_conf_uint32, "1024" }, /* client transport queue li */ 250 { "clock", &fmd_clkmode_ops, "native" }, /* clock operation mode */ 251 { "conf_path", &fmd_conf_path, _fmd_conf_path }, /* root config file path */ 252 { "conf_file", &fmd_conf_string, "fmd.conf" }, /* root config file name */ 253 { "core", &fmd_conf_bool, "false" }, /* force core dump on quit */ 254 { "dbout", &fmd_dbout_ops, NULL }, /* daemon debug output sinks */ 255 { "debug", &fmd_debug_ops, NULL }, /* daemon debugging flags */ 256 { "dictdir", &fmd_conf_string, "usr/lib/fm/dict" }, /* default diagcode dir */ 257 { "domain", &fmd_conf_string, NULL }, /* domain id for de auth */ 258 { "fakenotpresent", &fmd_conf_bool, "false" }, /* simulate rsrc not present */ 259 { "fg", &fmd_conf_bool, "false" }, /* run daemon in foreground */ 260 { "gc_interval", &fmd_conf_time, "1d" }, /* garbage collection intvl */ 261 { "ids.avg", &fmd_conf_uint32, "4" }, /* desired idspace chain len */ 262 { "ids.max", &fmd_conf_uint32, "1024" }, /* maximum idspace buckets */ 263 { "isaname", &fmd_conf_string, _fmd_isa }, /* instruction set (uname -p) */ 264 { "log.creator", &fmd_conf_string, "fmd" }, /* exacct log creator string */ 265 { "log.error", &fmd_conf_string, "var/fm/fmd/errlog" }, /* error log path */ 266 { "log.fault", &fmd_conf_string, "var/fm/fmd/fltlog" }, /* fault log path */ 267 { "log.minfree", &fmd_conf_size, "2m" }, /* min log fsys free space */ 268 { "log.rsrc", &fmd_conf_string, "var/fm/fmd/rsrc" }, /* asru log dir path */ 269 { "log.tryrotate", &fmd_conf_uint32, "10" }, /* max log rotation attempts */ 270 { "log.waitrotate", &fmd_conf_time, "200ms" }, /* log rotation retry delay */ 271 { "log.xprt", &fmd_conf_string, "var/fm/fmd/xprt" }, /* transport log dir */ 272 { "machine", &fmd_conf_string, _fmd_uts.machine }, /* machine name (uname -m) */ 273 { "nodiagcode", &fmd_conf_string, "-" }, /* diagcode to use if error */ 274 { "repaircode", &fmd_conf_string, "-" }, /* diagcode for list.repaired */ 275 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */ 276 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */ 277 { "platform", &fmd_conf_string, _fmd_plat }, /* platform string (uname -i) */ 278 { "plugin.close", &fmd_conf_bool, "true" }, /* dlclose plugins on fini */ 279 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */ 280 { "product", &fmd_conf_string, _fmd_prod }, /* product name string */ 281 { "rootdir", &fmd_conf_string, "" }, /* root directory for paths */ 282 { "rpc.adm.path", &fmd_conf_string, NULL }, /* FMD_ADM rendezvous file */ 283 { "rpc.adm.prog", &fmd_conf_uint32, "100169" }, /* FMD_ADM rpc program num */ 284 { "rpc.api.path", &fmd_conf_string, NULL }, /* FMD_API rendezvous file */ 285 { "rpc.api.prog", &fmd_conf_uint32, "100170" }, /* FMD_API rpc program num */ 286 { "rpc.rcvsize", &fmd_conf_size, "128k" }, /* rpc receive buffer size */ 287 { "rpc.sndsize", &fmd_conf_size, "128k" }, /* rpc send buffer size */ 288 { "rsrc.age", &fmd_conf_time, "30d" }, /* max age of old rsrc log */ 289 { "rsrc.zero", &fmd_conf_bool, "false" }, /* zero rsrc cache on start? */ 290 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */ 291 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */ 292 { "self.dict", &fmd_conf_list, "FMD.dict" }, /* self-diag dictionary list */ 293 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */ 294 { "strbuckets", &fmd_conf_uint32, "211" }, /* size of string hashes */ 295 #ifdef DEBUG 296 { "trace.mode", &fmd_trmode_ops, "full" }, /* trace mode: none/lite/full */ 297 #else 298 { "trace.mode", &fmd_trmode_ops, "lite" }, /* trace mode: none/lite/full */ 299 #endif 300 { "trace.recs", &fmd_conf_uint32, "128" }, /* trace records per thread */ 301 { "trace.frames", &fmd_conf_uint32, "16" }, /* max trace rec stack frames */ 302 { "uuidlen", &fmd_conf_uint32, "36" }, /* UUID ASCII string length */ 303 { "xprt.ttl", &fmd_conf_uint8, "1" }, /* default event time-to-live */ 304 }; 305 306 /* 307 * Statistics maintained by fmd itself on behalf of various global subsystems. 308 * NOTE: FMD_TYPE_STRING statistics should not be used here. If they are 309 * required in the future, the FMD_ADM_MODGSTAT service routine must change. 310 */ 311 static fmd_statistics_t _fmd_stats = { 312 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" }, 313 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" }, 314 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" }, 315 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" }, 316 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" }, 317 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" }, 318 { "topo.gen", FMD_TYPE_UINT64, "topology snapshot generation" }, 319 { "topo.drgen", FMD_TYPE_UINT64, "current topology DR generation number" }, 320 }; 321 322 void 323 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf) 324 { 325 fmd_conf_path_t *pap; 326 char file[PATH_MAX]; 327 const char *name; 328 fmd_stat_t *sp; 329 int i; 330 331 smbios_hdl_t *shp; 332 smbios_system_t s1; 333 smbios_info_t s2; 334 id_t id; 335 336 di_prom_handle_t promh = DI_PROM_HANDLE_NIL; 337 di_node_t rooth = DI_NODE_NIL; 338 char *bufp; 339 340 (void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat)); 341 (void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa)); 342 (void) uname(&_fmd_uts); 343 344 if ((shp = smbios_open(NULL, SMB_VERSION, 0, NULL)) != NULL) { 345 if ((id = smbios_info_system(shp, &s1)) != SMB_ERR && 346 smbios_info_common(shp, id, &s2) != SMB_ERR) { 347 (void) strlcpy(_fmd_prod, s2.smbi_product, MAXNAMELEN); 348 (void) strlcpy(_fmd_csn, s2.smbi_serial, MAXNAMELEN); 349 } 350 smbios_close(shp); 351 } else if ((rooth = di_init("/", DINFOPROP)) != DI_NODE_NIL && 352 (promh = di_prom_init()) != DI_PROM_HANDLE_NIL) { 353 if (di_prom_prop_lookup_bytes(promh, rooth, "chassis-sn", 354 (unsigned char **)&bufp) != -1) { 355 (void) strlcpy(_fmd_csn, bufp, MAXNAMELEN); 356 } 357 } 358 359 if (promh != DI_PROM_HANDLE_NIL) 360 di_prom_fini(promh); 361 if (rooth != DI_NODE_NIL) 362 di_fini(rooth); 363 364 bzero(dp, sizeof (fmd_t)); 365 366 dp->d_version = _fmd_version; 367 dp->d_pname = fmd_strbasename(arg0); 368 dp->d_pid = getpid(); 369 370 if (pthread_key_create(&dp->d_key, NULL) != 0) 371 fmd_error(EFMD_EXIT, "failed to create pthread key"); 372 373 (void) pthread_mutex_init(&dp->d_xprt_lock, NULL); 374 (void) pthread_mutex_init(&dp->d_err_lock, NULL); 375 (void) pthread_mutex_init(&dp->d_thr_lock, NULL); 376 (void) pthread_mutex_init(&dp->d_mod_lock, NULL); 377 (void) pthread_mutex_init(&dp->d_stats_lock, NULL); 378 (void) pthread_mutex_init(&dp->d_topo_lock, NULL); 379 (void) pthread_rwlock_init(&dp->d_log_lock, NULL); 380 381 /* 382 * A small number of properties must be set manually before we open 383 * the root configuration file. These include any settings for our 384 * memory allocator and path expansion token values, because these 385 * values are needed by the routines in fmd_conf.c itself. After 386 * the root configuration file is processed, we reset these properties 387 * based upon the latest values from the configuration file. 388 */ 389 dp->d_alloc_msecs = 10; 390 dp->d_alloc_tries = 3; 391 dp->d_str_buckets = 211; 392 393 dp->d_rootdir = root ? root : ""; 394 dp->d_platform = _fmd_plat; 395 dp->d_machine = _fmd_uts.machine; 396 dp->d_isaname = _fmd_isa; 397 398 dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) / 399 sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER); 400 401 if (dp->d_conf == NULL) { 402 fmd_error(EFMD_EXIT, 403 "failed to load required configuration properties\n"); 404 } 405 406 (void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs); 407 (void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries); 408 (void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets); 409 410 (void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform); 411 (void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine); 412 (void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname); 413 414 /* 415 * Manually specified rootdirs override config files, so only update 416 * d_rootdir based on the config files we parsed if no 'root' was set. 417 */ 418 if (root == NULL) 419 (void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir); 420 else 421 (void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir); 422 423 /* 424 * Once the base conf file properties are loaded, lookup the values 425 * of $conf_path and $conf_file and merge in any other conf files. 426 */ 427 (void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap); 428 (void) fmd_conf_getprop(dp->d_conf, "conf_file", &name); 429 430 for (i = 0; i < pap->cpa_argc; i++) { 431 (void) snprintf(file, sizeof (file), 432 "%s/%s", pap->cpa_argv[i], name); 433 if (access(file, F_OK) == 0) 434 fmd_conf_merge(dp->d_conf, file); 435 } 436 437 /* 438 * Update the value of fmd.d_fg based on "fg". We cache this property 439 * because it must be accessed deep within fmd at fmd_verror() time. 440 * Update any other properties that must be cached for performance. 441 */ 442 (void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg); 443 (void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl); 444 445 /* 446 * Initialize our custom libnvpair allocator and create an nvlist for 447 * authority elements corresponding to this instance of the daemon. 448 */ 449 (void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops); 450 dp->d_auth = fmd_protocol_authority(); 451 452 /* 453 * The fmd_module_t for the root module must be created manually. Most 454 * of it remains unused and zero, except for the few things we fill in. 455 */ 456 dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP); 457 dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP); 458 dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod); 459 460 fmd_list_append(&dp->d_mod_list, dp->d_rmod); 461 fmd_module_hold(dp->d_rmod); 462 463 (void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL); 464 (void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL); 465 (void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL); 466 467 dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self()); 468 dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP); 469 dp->d_rmod->mod_ustat = fmd_ustat_create(); 470 471 if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0) 472 fmd_error(EFMD_EXIT, "failed to attach main thread key"); 473 474 if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert( 475 dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) / 476 sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL) 477 fmd_error(EFMD_EXIT, "failed to initialize statistics"); 478 479 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 480 dp->d_rmod->mod_flags |= FMD_MOD_INIT; 481 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 482 483 /* 484 * In addition to inserting the _fmd_stats collection of program-wide 485 * statistics, we also insert a statistic named after each of our 486 * errors and update these counts in fmd_verror() (see fmd_subr.c). 487 */ 488 dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) * 489 (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP); 490 491 for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) { 492 (void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s", 493 strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1); 494 sp->fmds_type = FMD_TYPE_UINT64; 495 } 496 497 (void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, 498 EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL); 499 } 500 501 void 502 fmd_destroy(fmd_t *dp) 503 { 504 fmd_module_t *mp; 505 fmd_case_t *cp; 506 int core; 507 508 (void) fmd_conf_getprop(fmd.d_conf, "core", &core); 509 510 fmd_rpc_fini(); 511 fmd_dr_fini(); 512 513 if (dp->d_xprt_ids != NULL) 514 fmd_xprt_suspend_all(); 515 516 /* 517 * Unload the self-diagnosis module first. This ensures that it does 518 * not get confused as we start unloading other modules, etc. We must 519 * hold the dispq lock as a writer while doing so since it uses d_self. 520 */ 521 if (dp->d_self != NULL) { 522 fmd_module_t *self; 523 524 (void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock); 525 self = dp->d_self; 526 dp->d_self = NULL; 527 (void) pthread_rwlock_unlock(&dp->d_disp->dq_lock); 528 529 fmd_module_unload(self); 530 fmd_module_rele(self); 531 } 532 533 /* 534 * Unload modules in reverse order *except* for the root module, which 535 * is first in the list. This allows it to keep its thread and trace. 536 */ 537 for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) { 538 fmd_module_unload(mp); 539 mp = fmd_list_prev(mp); 540 } 541 542 if (dp->d_mod_hash != NULL) { 543 fmd_modhash_destroy(dp->d_mod_hash); 544 dp->d_mod_hash = NULL; 545 } 546 547 /* 548 * Close both log files now that modules are no longer active. We must 549 * set these pointers to NULL in case any subsequent errors occur. 550 */ 551 if (dp->d_errlog != NULL) { 552 fmd_log_rele(dp->d_errlog); 553 dp->d_errlog = NULL; 554 } 555 556 if (dp->d_fltlog != NULL) { 557 fmd_log_rele(dp->d_fltlog); 558 dp->d_fltlog = NULL; 559 } 560 561 /* 562 * Now destroy the resource cache: each ASRU contains a case reference, 563 * which may in turn contain a pointer to a referenced owning module. 564 */ 565 if (dp->d_asrus != NULL) { 566 fmd_asru_hash_destroy(dp->d_asrus); 567 dp->d_asrus = NULL; 568 } 569 570 /* 571 * Now that all data structures that refer to modules are torn down, 572 * no modules should be remaining on the module list except for d_rmod. 573 * If we trip one of these assertions, we're missing a rele somewhere. 574 */ 575 ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod); 576 ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod); 577 578 /* 579 * Now destroy the root module. We clear its thread key first so any 580 * calls to fmd_trace() inside of the module code will be ignored. 581 */ 582 (void) pthread_setspecific(dp->d_key, NULL); 583 fmd_module_lock(dp->d_rmod); 584 585 while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL) 586 fmd_case_discard(cp); 587 588 fmd_module_unlock(dp->d_rmod); 589 fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t)); 590 dp->d_rmod->mod_stats = NULL; 591 592 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 593 dp->d_rmod->mod_flags |= FMD_MOD_FINI; 594 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 595 596 fmd_module_rele(dp->d_rmod); 597 ASSERT(fmd_list_next(&dp->d_mod_list) == NULL); 598 599 /* 600 * Now destroy the remaining global data structures. If 'core' was 601 * set to true, force a core dump so we can check for memory leaks. 602 */ 603 if (dp->d_cases != NULL) 604 fmd_case_hash_destroy(dp->d_cases); 605 if (dp->d_disp != NULL) 606 fmd_dispq_destroy(dp->d_disp); 607 if (dp->d_timers != NULL) 608 fmd_timerq_destroy(dp->d_timers); 609 if (dp->d_schemes != NULL) 610 fmd_scheme_hash_destroy(dp->d_schemes); 611 if (dp->d_xprt_ids != NULL) 612 fmd_idspace_destroy(dp->d_xprt_ids); 613 614 if (dp->d_errstats != NULL) { 615 fmd_free(dp->d_errstats, 616 sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN)); 617 } 618 619 if (dp->d_conf != NULL) 620 fmd_conf_close(dp->d_conf); 621 622 fmd_topo_fini(); 623 624 nvlist_free(dp->d_auth); 625 (void) nv_alloc_fini(&dp->d_nva); 626 dp->d_clockops->fto_fini(dp->d_clockptr); 627 628 (void) pthread_key_delete(dp->d_key); 629 bzero(dp, sizeof (fmd_t)); 630 631 if (core) 632 fmd_panic("forcing core dump at user request\n"); 633 } 634 635 /*ARGSUSED*/ 636 static void 637 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt) 638 { 639 hrtime_t delta; 640 641 if (id != 0) { 642 TRACE((FMD_DBG_MOD, "garbage collect start")); 643 fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc); 644 TRACE((FMD_DBG_MOD, "garbage collect end")); 645 646 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 647 fmd_log_update(dp->d_errlog); 648 (void) pthread_rwlock_unlock(&dp->d_log_lock); 649 } 650 651 (void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta); 652 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 653 (fmd_timer_f *)fmd_gc, dp, NULL, delta); 654 } 655 656 /*ARGSUSED*/ 657 static void 658 fmd_clear_aged_rsrcs(fmd_t *dp, id_t id, hrtime_t hrt) 659 { 660 hrtime_t delta; 661 662 fmd_asru_clear_aged_rsrcs(); 663 (void) fmd_conf_getprop(dp->d_conf, "rsrc.age", &delta); 664 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 665 (fmd_timer_f *)fmd_clear_aged_rsrcs, dp, NULL, delta/10); 666 } 667 668 /* 669 * Events are committed to the errlog after cases are checkpointed. If fmd 670 * crashes before an event is ever associated with a module, this function will 671 * be called to replay it to all subscribers. If fmd crashes in between the 672 * subscriber checkpointing and committing the event in the error log, the 673 * module will have seen the event and we don't want to replay it. So we look 674 * for the event in all modules and transition it to the proper state. If 675 * it is found, we commit it to the error log and do not replay it. The in- 676 * memory case search used by fmd_module_contains() et al isn't particularly 677 * efficient, but it is faster than doing read i/o's on every case event to 678 * check their status or write i/o's on every event to replay to update states. 679 * We can improve the efficiency of this lookup algorithm later if necessary. 680 */ 681 /*ARGSUSED*/ 682 static void 683 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp) 684 { 685 fmd_module_t *mp; 686 fmd_stat_t *sp; 687 688 (void) pthread_mutex_lock(&dp->d_mod_lock); 689 690 for (mp = fmd_list_next(&dp->d_mod_list); 691 mp != NULL; mp = fmd_list_next(mp)) { 692 if (fmd_module_contains(mp, ep)) { 693 fmd_module_hold(mp); 694 break; 695 } 696 } 697 698 (void) pthread_mutex_unlock(&dp->d_mod_lock); 699 700 if (mp != NULL) { 701 fmd_event_commit(ep); 702 fmd_module_rele(mp); 703 sp = &dp->d_stats->ds_log_partials; 704 } else { 705 fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep)); 706 sp = &dp->d_stats->ds_log_replayed; 707 } 708 709 (void) pthread_mutex_lock(&dp->d_stats_lock); 710 sp->fmds_value.ui64++; 711 (void) pthread_mutex_unlock(&dp->d_stats_lock); 712 } 713 714 void 715 fmd_door_server(void *dip) 716 { 717 fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip); 718 (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 719 (void) door_return(NULL, 0, NULL, 0); 720 } 721 722 /* 723 * Custom door server create callback. Any fmd services that use doors will 724 * require those threads to have their fmd-specific TSD initialized, etc. 725 */ 726 static void 727 fmd_door(door_info_t *dip) 728 { 729 if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL) 730 fmd_panic("failed to create server for door %p", (void *)dip); 731 } 732 733 /* 734 * This signal handler is installed for the client.thrsig signal to be used to 735 * force an auxiliary thread to wake up from a system call and return EINTR in 736 * response to a module's use of fmd_thr_signal(). We also trace the event. 737 */ 738 static void 739 fmd_signal(int sig) 740 { 741 TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig)); 742 } 743 744 void 745 fmd_run(fmd_t *dp, int pfd) 746 { 747 char *nodc_key[] = { FMD_FLT_NODC, NULL }; 748 char *repair_key[] = { FM_LIST_REPAIRED_CLASS, NULL }; 749 char code_str[128]; 750 struct sigaction act; 751 752 int status = FMD_EXIT_SUCCESS; 753 const char *name; 754 fmd_conf_path_t *pap; 755 fmd_event_t *e; 756 int dbout; 757 758 /* 759 * Cache all the current debug property settings in d_fmd_debug, 760 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout. If a given debug mask 761 * is non-zero and the corresponding dbout mask is zero, set dbout 762 * to a sensible default value based on whether we have daemonized. 763 */ 764 (void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout); 765 766 if (dp->d_fmd_debug != 0 && dbout == 0) 767 dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 768 else 769 dp->d_fmd_dbout = dbout; 770 771 (void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug); 772 (void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout); 773 774 if (dp->d_hdl_debug != 0 && dbout == 0) 775 dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 776 else 777 dp->d_hdl_dbout = dbout; 778 779 /* 780 * Initialize remaining major program data structures such as the 781 * clock, dispatch queues, log files, module hash collections, etc. 782 * This work is done here rather than in fmd_create() to permit the -o 783 * command-line option to modify properties after fmd_create() is done. 784 */ 785 name = dp->d_rootdir != NULL && 786 *dp->d_rootdir != '\0' ? dp->d_rootdir : NULL; 787 788 /* 789 * The clock must be initialized before fmd_topo_init() because 790 * fmd_topo_update() calls fmd_time_gethrtime(). 791 */ 792 dp->d_clockptr = dp->d_clockops->fto_init(); 793 794 fmd_topo_init(); 795 796 dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX); 797 fmd_xprt_suspend_all(); 798 799 (void) door_server_create(fmd_door); 800 801 dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16); 802 dp->d_timers = fmd_timerq_create(); 803 dp->d_disp = fmd_dispq_create(); 804 dp->d_cases = fmd_case_hash_create(); 805 806 /* 807 * The root module's mod_queue is created with limit zero, making it 808 * act like /dev/null; anything inserted here is simply ignored. 809 */ 810 dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod, 811 &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0); 812 813 /* 814 * Once our subsystems that use signals have been set up, install the 815 * signal handler for the fmd_thr_signal() API. Verify that the signal 816 * being used for this purpose doesn't conflict with something else. 817 */ 818 (void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig); 819 820 if (sigaction(dp->d_thr_sig, NULL, &act) != 0) { 821 fmd_error(EFMD_EXIT, "invalid signal selected for " 822 "client.thrsig property: %d\n", dp->d_thr_sig); 823 } 824 825 if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) { 826 fmd_error(EFMD_EXIT, "signal selected for client.thrsig " 827 "property is already in use: %d\n", dp->d_thr_sig); 828 } 829 830 act.sa_handler = fmd_signal; 831 act.sa_flags = 0; 832 833 (void) sigemptyset(&act.sa_mask); 834 (void) sigaction(dp->d_thr_sig, &act, NULL); 835 836 (void) fmd_conf_getprop(dp->d_conf, "schemedir", &name); 837 dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name); 838 839 (void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name); 840 dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name); 841 842 (void) fmd_conf_getprop(dp->d_conf, "log.error", &name); 843 dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR); 844 845 (void) fmd_conf_getprop(dp->d_conf, "log.fault", &name); 846 dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT); 847 848 if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL) 849 fmd_error(EFMD_EXIT, "failed to initialize log files\n"); 850 851 /* 852 * Before loading modules, create an empty control event which will act 853 * as a global barrier for module event processing. Each module we 854 * load successfully will insert it at their head of their event queue, 855 * and then pause inside of fmd_ctl_rele() after dequeuing the event. 856 * This module barrier is required for two reasons: 857 * 858 * (a) During module loading, the restoration of case checkpoints may 859 * result in a list.* event being recreated for which the intended 860 * subscriber has not yet loaded depending on the load order. Such 861 * events could then result in spurious "no subscriber" errors. 862 * 863 * (b) During errlog replay, a sequence of errors from a long time ago 864 * may be replayed, and the module may attempt to install relative 865 * timers associated with one or more of these events. If errlog 866 * replay were "racing" with active module threads, an event E1 867 * that resulted in a relative timer T at time E1 + N nsec could 868 * fire prior to an event E2 being enqueued, even if the relative 869 * time ordering was E1 < E2 < E1 + N, causing mis-diagnosis. 870 */ 871 dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL, 872 FMD_HRT_NOW, NULL, fmd_ctl_init(NULL)); 873 874 fmd_event_hold(e); 875 876 /* 877 * Once all data structures are initialized, we load all of our modules 878 * in order according to class in order to load up any subscriptions. 879 * Once built-in modules are loaded, we detach from our waiting parent. 880 */ 881 dp->d_mod_hash = fmd_modhash_create(); 882 883 if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg) 884 fmd_error(EFMD_EXIT, "failed to initialize fault manager\n"); 885 886 (void) fmd_conf_getprop(dp->d_conf, "self.name", &name); 887 dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name); 888 889 if (dp->d_self != NULL) { 890 if (fmd_module_dc_key2code(dp->d_self, nodc_key, code_str, 891 sizeof (code_str)) == 0) 892 (void) fmd_conf_setprop(dp->d_conf, "nodiagcode", 893 code_str); 894 if (fmd_module_dc_key2code(dp->d_self, repair_key, code_str, 895 sizeof (code_str)) == 0) 896 (void) fmd_conf_setprop(dp->d_conf, "repaircode", 897 code_str); 898 } 899 900 fmd_rpc_init(); 901 dp->d_running = 1; /* we are now officially an active fmd */ 902 903 /* 904 * Now that we're running, if a pipe fd was specified, write an exit 905 * status to it to indicate that our parent process can safely detach. 906 * Then proceed to loading the remaining non-built-in modules. 907 */ 908 if (pfd >= 0) 909 (void) write(pfd, &status, sizeof (status)); 910 911 /* 912 * Before loading all modules, repopulate the ASRU cache from its 913 * persistent repository on disk. Then during module loading, the 914 * restoration of checkpoint files will reparent any active cases. 915 */ 916 fmd_asru_hash_refresh(dp->d_asrus); 917 918 (void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap); 919 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops, ".so"); 920 921 (void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap); 922 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops, NULL); 923 924 /* 925 * Subscribe to sysevents after all modules have been loaded. 926 */ 927 fmd_dr_init(); 928 929 /* 930 * With all modules loaded, replay fault events from the ASRU cache for 931 * any ASRUs that must be retired, replay error events from the errlog 932 * that did not finish processing the last time ran, and then release 933 * the global module barrier by executing a final rele on d_mod_event. 934 */ 935 fmd_asru_hash_replay(dp->d_asrus); 936 937 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 938 fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp); 939 fmd_log_update(dp->d_errlog); 940 (void) pthread_rwlock_unlock(&dp->d_log_lock); 941 942 dp->d_mod_event = NULL; 943 fmd_event_rele(e); 944 945 /* 946 * Finally, awaken any threads associated with receiving events from 947 * open transports and tell them to proceed with fmd_xprt_recv(). 948 */ 949 fmd_xprt_resume_all(); 950 fmd_gc(dp, 0, 0); 951 fmd_clear_aged_rsrcs(dp, 0, 0); 952 953 dp->d_booted = 1; 954 } 955 956 void 957 fmd_help(fmd_t *dp) 958 { 959 const fmd_conf_mode_t *cmp; 960 961 (void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname); 962 963 for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++) 964 (void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc); 965 } 966