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_uint32, "0" }, /* 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 { "resolvecode", &fmd_conf_string, "-" }, /* diagcode for list.resolved */ 276 { "updatecode", &fmd_conf_string, "-" }, /* diagcode for list.updated */ 277 { "osrelease", &fmd_conf_string, _fmd_uts.release }, /* release (uname -r) */ 278 { "osversion", &fmd_conf_string, _fmd_uts.version }, /* version (uname -v) */ 279 { "platform", &fmd_conf_string, _fmd_plat }, /* platform string (uname -i) */ 280 { "plugin.close", &fmd_conf_bool, "true" }, /* dlclose plugins on fini */ 281 { "plugin.path", &fmd_conf_path, _fmd_plugin_path }, /* path for plugin mods */ 282 { "product", &fmd_conf_string, _fmd_prod }, /* product name string */ 283 { "rootdir", &fmd_conf_string, "" }, /* root directory for paths */ 284 { "rpc.adm.path", &fmd_conf_string, NULL }, /* FMD_ADM rendezvous file */ 285 { "rpc.adm.prog", &fmd_conf_uint32, "100169" }, /* FMD_ADM rpc program num */ 286 { "rpc.api.path", &fmd_conf_string, NULL }, /* FMD_API rendezvous file */ 287 { "rpc.api.prog", &fmd_conf_uint32, "100170" }, /* FMD_API rpc program num */ 288 { "rpc.rcvsize", &fmd_conf_size, "128k" }, /* rpc receive buffer size */ 289 { "rpc.sndsize", &fmd_conf_size, "128k" }, /* rpc send buffer size */ 290 { "rsrc.age", &fmd_conf_time, "30d" }, /* max age of old rsrc log */ 291 { "rsrc.zero", &fmd_conf_bool, "false" }, /* zero rsrc cache on start? */ 292 { "schemedir", &fmd_conf_string, _fmd_scheme_path }, /* path for scheme mods */ 293 { "self.name", &fmd_conf_string, "fmd-self-diagnosis" }, /* self-diag module */ 294 { "self.dict", &fmd_conf_list, "FMD.dict" }, /* self-diag dictionary list */ 295 { "server", &fmd_conf_string, _fmd_uts.nodename }, /* server id for de auth */ 296 { "strbuckets", &fmd_conf_uint32, "211" }, /* size of string hashes */ 297 #ifdef DEBUG 298 { "trace.mode", &fmd_trmode_ops, "full" }, /* trace mode: none/lite/full */ 299 #else 300 { "trace.mode", &fmd_trmode_ops, "lite" }, /* trace mode: none/lite/full */ 301 #endif 302 { "trace.recs", &fmd_conf_uint32, "128" }, /* trace records per thread */ 303 { "trace.frames", &fmd_conf_uint32, "16" }, /* max trace rec stack frames */ 304 { "uuidlen", &fmd_conf_uint32, "36" }, /* UUID ASCII string length */ 305 { "xprt.ttl", &fmd_conf_uint8, "1" }, /* default event time-to-live */ 306 }; 307 308 /* 309 * Statistics maintained by fmd itself on behalf of various global subsystems. 310 * NOTE: FMD_TYPE_STRING statistics should not be used here. If they are 311 * required in the future, the FMD_ADM_MODGSTAT service routine must change. 312 */ 313 static fmd_statistics_t _fmd_stats = { 314 { "errlog.replayed", FMD_TYPE_UINT64, "total events replayed from errlog" }, 315 { "errlog.partials", FMD_TYPE_UINT64, "events partially committed in errlog" }, 316 { "errlog.enospc", FMD_TYPE_UINT64, "events not appended to errlog (ENOSPC)" }, 317 { "fltlog.enospc", FMD_TYPE_UINT64, "events not appended to fltlog (ENOSPC)" }, 318 { "log.enospc", FMD_TYPE_UINT64, "events not appended to other logs (ENOSPC)" }, 319 { "dr.gen", FMD_TYPE_UINT64, "dynamic reconfiguration generation" }, 320 { "topo.gen", FMD_TYPE_UINT64, "topology snapshot generation" }, 321 { "topo.drgen", FMD_TYPE_UINT64, "current topology DR generation number" }, 322 }; 323 324 void 325 fmd_create(fmd_t *dp, const char *arg0, const char *root, const char *conf) 326 { 327 fmd_conf_path_t *pap; 328 char file[PATH_MAX]; 329 const char *name; 330 fmd_stat_t *sp; 331 int i; 332 333 smbios_hdl_t *shp; 334 smbios_system_t s1; 335 smbios_info_t s2; 336 id_t id; 337 338 di_prom_handle_t promh = DI_PROM_HANDLE_NIL; 339 di_node_t rooth = DI_NODE_NIL; 340 char *bufp; 341 342 (void) sysinfo(SI_PLATFORM, _fmd_plat, sizeof (_fmd_plat)); 343 (void) sysinfo(SI_ARCHITECTURE, _fmd_isa, sizeof (_fmd_isa)); 344 (void) uname(&_fmd_uts); 345 346 if ((shp = smbios_open(NULL, SMB_VERSION, 0, NULL)) != NULL) { 347 if ((id = smbios_info_system(shp, &s1)) != SMB_ERR && 348 smbios_info_common(shp, id, &s2) != SMB_ERR) { 349 (void) strlcpy(_fmd_prod, s2.smbi_product, MAXNAMELEN); 350 (void) strlcpy(_fmd_csn, s2.smbi_serial, MAXNAMELEN); 351 } 352 smbios_close(shp); 353 } else if ((rooth = di_init("/", DINFOPROP)) != DI_NODE_NIL && 354 (promh = di_prom_init()) != DI_PROM_HANDLE_NIL) { 355 if (di_prom_prop_lookup_bytes(promh, rooth, "chassis-sn", 356 (unsigned char **)&bufp) != -1) { 357 (void) strlcpy(_fmd_csn, bufp, MAXNAMELEN); 358 } 359 } 360 361 if (promh != DI_PROM_HANDLE_NIL) 362 di_prom_fini(promh); 363 if (rooth != DI_NODE_NIL) 364 di_fini(rooth); 365 366 bzero(dp, sizeof (fmd_t)); 367 368 dp->d_version = _fmd_version; 369 dp->d_pname = fmd_strbasename(arg0); 370 dp->d_pid = getpid(); 371 372 if (pthread_key_create(&dp->d_key, NULL) != 0) 373 fmd_error(EFMD_EXIT, "failed to create pthread key"); 374 375 (void) pthread_mutex_init(&dp->d_xprt_lock, NULL); 376 (void) pthread_mutex_init(&dp->d_err_lock, NULL); 377 (void) pthread_mutex_init(&dp->d_thr_lock, NULL); 378 (void) pthread_mutex_init(&dp->d_mod_lock, NULL); 379 (void) pthread_mutex_init(&dp->d_stats_lock, NULL); 380 (void) pthread_mutex_init(&dp->d_topo_lock, NULL); 381 (void) pthread_rwlock_init(&dp->d_log_lock, NULL); 382 (void) pthread_mutex_init(&dp->d_fmd_lock, NULL); 383 (void) pthread_cond_init(&dp->d_fmd_cv, NULL); 384 385 /* 386 * A small number of properties must be set manually before we open 387 * the root configuration file. These include any settings for our 388 * memory allocator and path expansion token values, because these 389 * values are needed by the routines in fmd_conf.c itself. After 390 * the root configuration file is processed, we reset these properties 391 * based upon the latest values from the configuration file. 392 */ 393 dp->d_alloc_msecs = 10; 394 dp->d_alloc_tries = 3; 395 dp->d_str_buckets = 211; 396 397 dp->d_rootdir = root ? root : ""; 398 dp->d_platform = _fmd_plat; 399 dp->d_machine = _fmd_uts.machine; 400 dp->d_isaname = _fmd_isa; 401 402 dp->d_conf = fmd_conf_open(conf, sizeof (_fmd_conf) / 403 sizeof (_fmd_conf[0]), _fmd_conf, FMD_CONF_DEFER); 404 405 if (dp->d_conf == NULL) { 406 fmd_error(EFMD_EXIT, 407 "failed to load required configuration properties\n"); 408 } 409 410 (void) fmd_conf_getprop(dp->d_conf, "alloc.msecs", &dp->d_alloc_msecs); 411 (void) fmd_conf_getprop(dp->d_conf, "alloc.tries", &dp->d_alloc_tries); 412 (void) fmd_conf_getprop(dp->d_conf, "strbuckets", &dp->d_str_buckets); 413 414 (void) fmd_conf_getprop(dp->d_conf, "platform", &dp->d_platform); 415 (void) fmd_conf_getprop(dp->d_conf, "machine", &dp->d_machine); 416 (void) fmd_conf_getprop(dp->d_conf, "isaname", &dp->d_isaname); 417 418 /* 419 * Manually specified rootdirs override config files, so only update 420 * d_rootdir based on the config files we parsed if no 'root' was set. 421 */ 422 if (root == NULL) 423 (void) fmd_conf_getprop(dp->d_conf, "rootdir", &dp->d_rootdir); 424 else 425 (void) fmd_conf_setprop(dp->d_conf, "rootdir", dp->d_rootdir); 426 427 /* 428 * Once the base conf file properties are loaded, lookup the values 429 * of $conf_path and $conf_file and merge in any other conf files. 430 */ 431 (void) fmd_conf_getprop(dp->d_conf, "conf_path", &pap); 432 (void) fmd_conf_getprop(dp->d_conf, "conf_file", &name); 433 434 for (i = 0; i < pap->cpa_argc; i++) { 435 (void) snprintf(file, sizeof (file), 436 "%s/%s", pap->cpa_argv[i], name); 437 if (access(file, F_OK) == 0) 438 fmd_conf_merge(dp->d_conf, file); 439 } 440 441 /* 442 * Update the value of fmd.d_fg based on "fg". We cache this property 443 * because it must be accessed deep within fmd at fmd_verror() time. 444 * Update any other properties that must be cached for performance. 445 */ 446 (void) fmd_conf_getprop(fmd.d_conf, "fg", &fmd.d_fg); 447 (void) fmd_conf_getprop(fmd.d_conf, "xprt.ttl", &fmd.d_xprt_ttl); 448 449 /* 450 * Initialize our custom libnvpair allocator and create an nvlist for 451 * authority elements corresponding to this instance of the daemon. 452 */ 453 (void) nv_alloc_init(&dp->d_nva, &fmd_nv_alloc_ops); 454 dp->d_auth = fmd_protocol_authority(); 455 456 /* 457 * The fmd_module_t for the root module must be created manually. Most 458 * of it remains unused and zero, except for the few things we fill in. 459 */ 460 dp->d_rmod = fmd_zalloc(sizeof (fmd_module_t), FMD_SLEEP); 461 dp->d_rmod->mod_name = fmd_strdup(dp->d_pname, FMD_SLEEP); 462 dp->d_rmod->mod_fmri = fmd_protocol_fmri_module(dp->d_rmod); 463 464 fmd_list_append(&dp->d_mod_list, dp->d_rmod); 465 fmd_module_hold(dp->d_rmod); 466 467 (void) pthread_mutex_init(&dp->d_rmod->mod_lock, NULL); 468 (void) pthread_cond_init(&dp->d_rmod->mod_cv, NULL); 469 (void) pthread_mutex_init(&dp->d_rmod->mod_stats_lock, NULL); 470 471 dp->d_rmod->mod_thread = fmd_thread_xcreate(dp->d_rmod, pthread_self()); 472 dp->d_rmod->mod_stats = fmd_zalloc(sizeof (fmd_modstat_t), FMD_SLEEP); 473 dp->d_rmod->mod_ustat = fmd_ustat_create(); 474 475 if (pthread_setspecific(dp->d_key, dp->d_rmod->mod_thread) != 0) 476 fmd_error(EFMD_EXIT, "failed to attach main thread key"); 477 478 if ((dp->d_stats = (fmd_statistics_t *)fmd_ustat_insert( 479 dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, sizeof (_fmd_stats) / 480 sizeof (fmd_stat_t), (fmd_stat_t *)&_fmd_stats, NULL)) == NULL) 481 fmd_error(EFMD_EXIT, "failed to initialize statistics"); 482 483 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 484 dp->d_rmod->mod_flags |= FMD_MOD_INIT; 485 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 486 487 /* 488 * In addition to inserting the _fmd_stats collection of program-wide 489 * statistics, we also insert a statistic named after each of our 490 * errors and update these counts in fmd_verror() (see fmd_subr.c). 491 */ 492 dp->d_errstats = sp = fmd_zalloc(sizeof (fmd_stat_t) * 493 (EFMD_END - EFMD_UNKNOWN), FMD_SLEEP); 494 495 for (i = 0; i < EFMD_END - EFMD_UNKNOWN; i++, sp++) { 496 (void) snprintf(sp->fmds_name, sizeof (sp->fmds_name), "err.%s", 497 strrchr(fmd_errclass(EFMD_UNKNOWN + i), '.') + 1); 498 sp->fmds_type = FMD_TYPE_UINT64; 499 } 500 501 (void) fmd_ustat_insert(dp->d_rmod->mod_ustat, FMD_USTAT_NOALLOC, 502 EFMD_END - EFMD_UNKNOWN, dp->d_errstats, NULL); 503 } 504 505 void 506 fmd_destroy(fmd_t *dp) 507 { 508 fmd_module_t *mp; 509 fmd_case_t *cp; 510 int core; 511 512 (void) fmd_conf_getprop(fmd.d_conf, "core", &core); 513 514 fmd_rpc_fini(); 515 516 if (dp->d_xprt_ids != NULL) 517 fmd_xprt_suspend_all(); 518 519 /* 520 * Unload the self-diagnosis module first. This ensures that it does 521 * not get confused as we start unloading other modules, etc. We must 522 * hold the dispq lock as a writer while doing so since it uses d_self. 523 */ 524 if (dp->d_self != NULL) { 525 fmd_module_t *self; 526 527 (void) pthread_rwlock_wrlock(&dp->d_disp->dq_lock); 528 self = dp->d_self; 529 dp->d_self = NULL; 530 (void) pthread_rwlock_unlock(&dp->d_disp->dq_lock); 531 532 fmd_module_unload(self); 533 fmd_module_rele(self); 534 } 535 536 /* 537 * Unload modules in reverse order *except* for the root module, which 538 * is first in the list. This allows it to keep its thread and trace. 539 */ 540 for (mp = fmd_list_prev(&dp->d_mod_list); mp != dp->d_rmod; ) { 541 fmd_module_unload(mp); 542 mp = fmd_list_prev(mp); 543 } 544 545 if (dp->d_mod_hash != NULL) { 546 fmd_modhash_destroy(dp->d_mod_hash); 547 dp->d_mod_hash = NULL; 548 } 549 550 /* 551 * Close both log files now that modules are no longer active. We must 552 * set these pointers to NULL in case any subsequent errors occur. 553 */ 554 if (dp->d_errlog != NULL) { 555 fmd_log_rele(dp->d_errlog); 556 dp->d_errlog = NULL; 557 } 558 559 if (dp->d_fltlog != NULL) { 560 fmd_log_rele(dp->d_fltlog); 561 dp->d_fltlog = NULL; 562 } 563 564 /* 565 * Now destroy the resource cache: each ASRU contains a case reference, 566 * which may in turn contain a pointer to a referenced owning module. 567 */ 568 if (dp->d_asrus != NULL) { 569 fmd_asru_hash_destroy(dp->d_asrus); 570 dp->d_asrus = NULL; 571 } 572 573 /* 574 * Now that all data structures that refer to modules are torn down, 575 * no modules should be remaining on the module list except for d_rmod. 576 * If we trip one of these assertions, we're missing a rele somewhere. 577 */ 578 ASSERT(fmd_list_prev(&dp->d_mod_list) == dp->d_rmod); 579 ASSERT(fmd_list_next(&dp->d_mod_list) == dp->d_rmod); 580 581 /* 582 * Now destroy the root module. We clear its thread key first so any 583 * calls to fmd_trace() inside of the module code will be ignored. 584 */ 585 (void) pthread_setspecific(dp->d_key, NULL); 586 fmd_module_lock(dp->d_rmod); 587 588 while ((cp = fmd_list_next(&dp->d_rmod->mod_cases)) != NULL) 589 fmd_case_discard(cp); 590 591 fmd_module_unlock(dp->d_rmod); 592 fmd_free(dp->d_rmod->mod_stats, sizeof (fmd_modstat_t)); 593 dp->d_rmod->mod_stats = NULL; 594 595 (void) pthread_mutex_lock(&dp->d_rmod->mod_lock); 596 dp->d_rmod->mod_flags |= FMD_MOD_FINI; 597 (void) pthread_mutex_unlock(&dp->d_rmod->mod_lock); 598 599 fmd_module_rele(dp->d_rmod); 600 ASSERT(fmd_list_next(&dp->d_mod_list) == NULL); 601 602 /* 603 * Now destroy the remaining global data structures. If 'core' was 604 * set to true, force a core dump so we can check for memory leaks. 605 */ 606 if (dp->d_cases != NULL) 607 fmd_case_hash_destroy(dp->d_cases); 608 if (dp->d_disp != NULL) 609 fmd_dispq_destroy(dp->d_disp); 610 if (dp->d_timers != NULL) 611 fmd_timerq_destroy(dp->d_timers); 612 if (dp->d_schemes != NULL) 613 fmd_scheme_hash_destroy(dp->d_schemes); 614 if (dp->d_xprt_ids != NULL) 615 fmd_idspace_destroy(dp->d_xprt_ids); 616 617 if (dp->d_errstats != NULL) { 618 fmd_free(dp->d_errstats, 619 sizeof (fmd_stat_t) * (EFMD_END - EFMD_UNKNOWN)); 620 } 621 622 if (dp->d_conf != NULL) 623 fmd_conf_close(dp->d_conf); 624 625 fmd_topo_fini(); 626 627 nvlist_free(dp->d_auth); 628 (void) nv_alloc_fini(&dp->d_nva); 629 dp->d_clockops->fto_fini(dp->d_clockptr); 630 631 (void) pthread_key_delete(dp->d_key); 632 bzero(dp, sizeof (fmd_t)); 633 634 if (core) 635 fmd_panic("forcing core dump at user request\n"); 636 } 637 638 /*ARGSUSED*/ 639 static void 640 fmd_gc(fmd_t *dp, id_t id, hrtime_t hrt) 641 { 642 hrtime_t delta; 643 644 if (id != 0) { 645 TRACE((FMD_DBG_MOD, "garbage collect start")); 646 fmd_modhash_apply(dp->d_mod_hash, fmd_module_gc); 647 TRACE((FMD_DBG_MOD, "garbage collect end")); 648 649 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 650 fmd_log_update(dp->d_errlog); 651 (void) pthread_rwlock_unlock(&dp->d_log_lock); 652 } 653 654 (void) fmd_conf_getprop(dp->d_conf, "gc_interval", &delta); 655 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 656 (fmd_timer_f *)fmd_gc, dp, NULL, delta); 657 } 658 659 /*ARGSUSED*/ 660 static void 661 fmd_clear_aged_rsrcs(fmd_t *dp, id_t id, hrtime_t hrt) 662 { 663 hrtime_t delta; 664 665 fmd_asru_clear_aged_rsrcs(); 666 (void) fmd_conf_getprop(dp->d_conf, "rsrc.age", &delta); 667 (void) fmd_timerq_install(dp->d_timers, dp->d_rmod->mod_timerids, 668 (fmd_timer_f *)fmd_clear_aged_rsrcs, dp, NULL, delta/10); 669 } 670 671 /* 672 * Events are committed to the errlog after cases are checkpointed. If fmd 673 * crashes before an event is ever associated with a module, this function will 674 * be called to replay it to all subscribers. If fmd crashes in between the 675 * subscriber checkpointing and committing the event in the error log, the 676 * module will have seen the event and we don't want to replay it. So we look 677 * for the event in all modules and transition it to the proper state. If 678 * it is found, we commit it to the error log and do not replay it. The in- 679 * memory case search used by fmd_module_contains() et al isn't particularly 680 * efficient, but it is faster than doing read i/o's on every case event to 681 * check their status or write i/o's on every event to replay to update states. 682 * We can improve the efficiency of this lookup algorithm later if necessary. 683 */ 684 /*ARGSUSED*/ 685 static void 686 fmd_err_replay(fmd_log_t *lp, fmd_event_t *ep, fmd_t *dp) 687 { 688 fmd_module_t *mp; 689 fmd_stat_t *sp; 690 691 (void) pthread_mutex_lock(&dp->d_mod_lock); 692 693 for (mp = fmd_list_next(&dp->d_mod_list); 694 mp != NULL; mp = fmd_list_next(mp)) { 695 if (fmd_module_contains(mp, ep)) { 696 fmd_module_hold(mp); 697 break; 698 } 699 } 700 701 (void) pthread_mutex_unlock(&dp->d_mod_lock); 702 703 if (mp != NULL) { 704 fmd_event_commit(ep); 705 fmd_module_rele(mp); 706 sp = &dp->d_stats->ds_log_partials; 707 } else { 708 fmd_dispq_dispatch(dp->d_disp, ep, FMD_EVENT_DATA(ep)); 709 sp = &dp->d_stats->ds_log_replayed; 710 } 711 712 (void) pthread_mutex_lock(&dp->d_stats_lock); 713 sp->fmds_value.ui64++; 714 (void) pthread_mutex_unlock(&dp->d_stats_lock); 715 } 716 717 void 718 fmd_door_server(void *dip) 719 { 720 fmd_dprintf(FMD_DBG_XPRT, "door server starting for %p\n", dip); 721 (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 722 (void) door_return(NULL, 0, NULL, 0); 723 } 724 725 /* 726 * Custom door server create callback. Any fmd services that use doors will 727 * require those threads to have their fmd-specific TSD initialized, etc. 728 */ 729 static void 730 fmd_door(door_info_t *dip) 731 { 732 if (fmd_thread_create(fmd.d_rmod, fmd_door_server, dip) == NULL) 733 fmd_panic("failed to create server for door %p", (void *)dip); 734 } 735 736 /* 737 * This signal handler is installed for the client.thrsig signal to be used to 738 * force an auxiliary thread to wake up from a system call and return EINTR in 739 * response to a module's use of fmd_thr_signal(). We also trace the event. 740 */ 741 static void 742 fmd_signal(int sig) 743 { 744 TRACE((FMD_DBG_MOD, "module thread received sig #%d", sig)); 745 } 746 747 void 748 fmd_run(fmd_t *dp, int pfd) 749 { 750 char *nodc_key[] = { FMD_FLT_NODC, NULL }; 751 char *repair_key[] = { FM_LIST_REPAIRED_CLASS, NULL }; 752 char *resolve_key[] = { FM_LIST_RESOLVED_CLASS, NULL }; 753 char *update_key[] = { FM_LIST_UPDATED_CLASS, NULL }; 754 char code_str[128]; 755 struct sigaction act; 756 757 int status = FMD_EXIT_SUCCESS; 758 const char *name; 759 fmd_conf_path_t *pap; 760 fmd_event_t *e; 761 int dbout; 762 763 /* 764 * Cache all the current debug property settings in d_fmd_debug, 765 * d_fmd_dbout, d_hdl_debug, and d_hdl_dbout. If a given debug mask 766 * is non-zero and the corresponding dbout mask is zero, set dbout 767 * to a sensible default value based on whether we have daemonized. 768 */ 769 (void) fmd_conf_getprop(dp->d_conf, "dbout", &dbout); 770 771 if (dp->d_fmd_debug != 0 && dbout == 0) 772 dp->d_fmd_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 773 else 774 dp->d_fmd_dbout = dbout; 775 776 (void) fmd_conf_getprop(dp->d_conf, "client.debug", &dp->d_hdl_debug); 777 (void) fmd_conf_getprop(dp->d_conf, "client.dbout", &dbout); 778 779 if (dp->d_hdl_debug != 0 && dbout == 0) 780 dp->d_hdl_dbout = dp->d_fg? FMD_DBOUT_STDERR : FMD_DBOUT_SYSLOG; 781 else 782 dp->d_hdl_dbout = dbout; 783 784 /* 785 * Initialize remaining major program data structures such as the 786 * clock, dispatch queues, log files, module hash collections, etc. 787 * This work is done here rather than in fmd_create() to permit the -o 788 * command-line option to modify properties after fmd_create() is done. 789 */ 790 name = dp->d_rootdir != NULL && 791 *dp->d_rootdir != '\0' ? dp->d_rootdir : NULL; 792 793 /* 794 * The clock must be initialized before fmd_topo_init() because 795 * fmd_topo_update() calls fmd_time_gethrtime(). 796 */ 797 dp->d_clockptr = dp->d_clockops->fto_init(); 798 799 fmd_topo_init(); 800 801 dp->d_xprt_ids = fmd_idspace_create("xprt_ids", 1, INT_MAX); 802 fmd_xprt_suspend_all(); 803 804 (void) door_server_create(fmd_door); 805 806 dp->d_rmod->mod_timerids = fmd_idspace_create(dp->d_pname, 1, 16); 807 dp->d_timers = fmd_timerq_create(); 808 dp->d_disp = fmd_dispq_create(); 809 dp->d_cases = fmd_case_hash_create(); 810 811 /* 812 * The root module's mod_queue is created with limit zero, making it 813 * act like /dev/null; anything inserted here is simply ignored. 814 */ 815 dp->d_rmod->mod_queue = fmd_eventq_create(dp->d_rmod, 816 &dp->d_rmod->mod_stats->ms_evqstat, &dp->d_rmod->mod_stats_lock, 0); 817 818 /* 819 * Once our subsystems that use signals have been set up, install the 820 * signal handler for the fmd_thr_signal() API. Verify that the signal 821 * being used for this purpose doesn't conflict with something else. 822 */ 823 (void) fmd_conf_getprop(dp->d_conf, "client.thrsig", &dp->d_thr_sig); 824 825 if (sigaction(dp->d_thr_sig, NULL, &act) != 0) { 826 fmd_error(EFMD_EXIT, "invalid signal selected for " 827 "client.thrsig property: %d\n", dp->d_thr_sig); 828 } 829 830 if (act.sa_handler != SIG_IGN && act.sa_handler != SIG_DFL) { 831 fmd_error(EFMD_EXIT, "signal selected for client.thrsig " 832 "property is already in use: %d\n", dp->d_thr_sig); 833 } 834 835 act.sa_handler = fmd_signal; 836 act.sa_flags = 0; 837 838 (void) sigemptyset(&act.sa_mask); 839 (void) sigaction(dp->d_thr_sig, &act, NULL); 840 841 (void) fmd_conf_getprop(dp->d_conf, "schemedir", &name); 842 dp->d_schemes = fmd_scheme_hash_create(dp->d_rootdir, name); 843 844 (void) fmd_conf_getprop(dp->d_conf, "log.rsrc", &name); 845 dp->d_asrus = fmd_asru_hash_create(dp->d_rootdir, name); 846 847 (void) fmd_conf_getprop(dp->d_conf, "log.error", &name); 848 dp->d_errlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_ERROR); 849 850 (void) fmd_conf_getprop(dp->d_conf, "log.fault", &name); 851 dp->d_fltlog = fmd_log_open(dp->d_rootdir, name, FMD_LOG_FAULT); 852 853 if (dp->d_asrus == NULL || dp->d_errlog == NULL || dp->d_fltlog == NULL) 854 fmd_error(EFMD_EXIT, "failed to initialize log files\n"); 855 856 /* 857 * Before loading modules, create an empty control event which will act 858 * as a global barrier for module event processing. Each module we 859 * load successfully will insert it at their head of their event queue, 860 * and then pause inside of fmd_ctl_rele() after dequeuing the event. 861 * This module barrier is required for two reasons: 862 * 863 * (a) During module loading, the restoration of case checkpoints may 864 * result in a list.* event being recreated for which the intended 865 * subscriber has not yet loaded depending on the load order. Such 866 * events could then result in spurious "no subscriber" errors. 867 * 868 * (b) During errlog replay, a sequence of errors from a long time ago 869 * may be replayed, and the module may attempt to install relative 870 * timers associated with one or more of these events. If errlog 871 * replay were "racing" with active module threads, an event E1 872 * that resulted in a relative timer T at time E1 + N nsec could 873 * fire prior to an event E2 being enqueued, even if the relative 874 * time ordering was E1 < E2 < E1 + N, causing mis-diagnosis. 875 */ 876 dp->d_mod_event = e = fmd_event_create(FMD_EVT_CTL, 877 FMD_HRT_NOW, NULL, fmd_ctl_init(NULL)); 878 879 fmd_event_hold(e); 880 881 /* 882 * Once all data structures are initialized, we load all of our modules 883 * in order according to class in order to load up any subscriptions. 884 * Once built-in modules are loaded, we detach from our waiting parent. 885 */ 886 dp->d_mod_hash = fmd_modhash_create(); 887 888 if (fmd_builtin_loadall(dp->d_mod_hash) != 0 && !dp->d_fg) 889 fmd_error(EFMD_EXIT, "failed to initialize fault manager\n"); 890 891 (void) fmd_conf_getprop(dp->d_conf, "self.name", &name); 892 dp->d_self = fmd_modhash_lookup(dp->d_mod_hash, name); 893 894 if (dp->d_self != NULL) { 895 if (fmd_module_dc_key2code(dp->d_self, nodc_key, code_str, 896 sizeof (code_str)) == 0) 897 (void) fmd_conf_setprop(dp->d_conf, "nodiagcode", 898 code_str); 899 if (fmd_module_dc_key2code(dp->d_self, repair_key, code_str, 900 sizeof (code_str)) == 0) 901 (void) fmd_conf_setprop(dp->d_conf, "repaircode", 902 code_str); 903 if (fmd_module_dc_key2code(dp->d_self, resolve_key, code_str, 904 sizeof (code_str)) == 0) 905 (void) fmd_conf_setprop(dp->d_conf, "resolvecode", 906 code_str); 907 if (fmd_module_dc_key2code(dp->d_self, update_key, code_str, 908 sizeof (code_str)) == 0) 909 (void) fmd_conf_setprop(dp->d_conf, "updatecode", 910 code_str); 911 } 912 913 fmd_rpc_init(); 914 dp->d_running = 1; /* we are now officially an active fmd */ 915 916 /* 917 * Now that we're running, if a pipe fd was specified, write an exit 918 * status to it to indicate that our parent process can safely detach. 919 * Then proceed to loading the remaining non-built-in modules. 920 */ 921 if (pfd >= 0) 922 (void) write(pfd, &status, sizeof (status)); 923 924 /* 925 * Before loading all modules, repopulate the ASRU cache from its 926 * persistent repository on disk. Then during module loading, the 927 * restoration of checkpoint files will reparent any active cases. 928 */ 929 fmd_asru_hash_refresh(dp->d_asrus); 930 931 (void) fmd_conf_getprop(dp->d_conf, "plugin.path", &pap); 932 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_rtld_ops, ".so"); 933 934 (void) fmd_conf_getprop(dp->d_conf, "agent.path", &pap); 935 fmd_modhash_loadall(dp->d_mod_hash, pap, &fmd_proc_ops, NULL); 936 937 /* 938 * With all modules loaded, replay fault events from the ASRU cache for 939 * any ASRUs that must be retired, replay error events from the errlog 940 * that did not finish processing the last time ran, and then release 941 * the global module barrier by executing a final rele on d_mod_event. 942 */ 943 fmd_asru_hash_replay(dp->d_asrus); 944 945 (void) pthread_rwlock_rdlock(&dp->d_log_lock); 946 fmd_log_replay(dp->d_errlog, (fmd_log_f *)fmd_err_replay, dp); 947 fmd_log_update(dp->d_errlog); 948 (void) pthread_rwlock_unlock(&dp->d_log_lock); 949 950 dp->d_mod_event = NULL; 951 fmd_event_rele(e); 952 953 /* 954 * Now replay list.updated and list.repaired events 955 */ 956 fmd_case_repair_replay(); 957 958 /* 959 * Finally, awaken any threads associated with receiving events from 960 * open transports and tell them to proceed with fmd_xprt_recv(). 961 */ 962 fmd_xprt_resume_all(); 963 fmd_gc(dp, 0, 0); 964 fmd_clear_aged_rsrcs(dp, 0, 0); 965 966 (void) pthread_mutex_lock(&dp->d_fmd_lock); 967 dp->d_booted = 1; 968 (void) pthread_cond_broadcast(&dp->d_fmd_cv); 969 (void) pthread_mutex_unlock(&dp->d_fmd_lock); 970 } 971 972 void 973 fmd_help(fmd_t *dp) 974 { 975 const fmd_conf_mode_t *cmp; 976 977 (void) printf("Usage: %s -o debug=mode[,mode]\n", dp->d_pname); 978 979 for (cmp = _fmd_debug_modes; cmp->cm_name != NULL; cmp++) 980 (void) printf("\t%s\t%s\n", cmp->cm_name, cmp->cm_desc); 981 } 982