1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License Version 1.0 (CDDL-1.0). 6 * You can obtain a copy of the license from the top-level file 7 * "OPENSOLARIS.LICENSE" or at <http://opensource.org/licenses/CDDL-1.0>. 8 * You may not use this file except in compliance with the license. 9 * 10 * CDDL HEADER END 11 */ 12 13 /* 14 * Copyright (c) 2016, Intel Corporation. 15 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com> 16 */ 17 18 #include <libnvpair.h> 19 #include <libzfs.h> 20 #include <stddef.h> 21 #include <stdlib.h> 22 #include <string.h> 23 #include <sys/list.h> 24 #include <sys/time.h> 25 #include <sys/sysevent/eventdefs.h> 26 #include <sys/sysevent/dev.h> 27 #include <sys/fm/protocol.h> 28 #include <sys/fm/fs/zfs.h> 29 #include <pthread.h> 30 #include <unistd.h> 31 32 #include "zfs_agents.h" 33 #include "fmd_api.h" 34 #include "../zed_log.h" 35 36 /* 37 * agent dispatch code 38 */ 39 40 static pthread_mutex_t agent_lock = PTHREAD_MUTEX_INITIALIZER; 41 static pthread_cond_t agent_cond = PTHREAD_COND_INITIALIZER; 42 static list_t agent_events; /* list of pending events */ 43 static int agent_exiting; 44 45 typedef struct agent_event { 46 char ae_class[64]; 47 char ae_subclass[32]; 48 nvlist_t *ae_nvl; 49 list_node_t ae_node; 50 } agent_event_t; 51 52 pthread_t g_agents_tid; 53 54 libzfs_handle_t *g_zfs_hdl; 55 56 /* guid search data */ 57 typedef enum device_type { 58 DEVICE_TYPE_L2ARC, /* l2arc device */ 59 DEVICE_TYPE_SPARE, /* spare device */ 60 DEVICE_TYPE_PRIMARY /* any primary pool storage device */ 61 } device_type_t; 62 63 typedef struct guid_search { 64 uint64_t gs_pool_guid; 65 uint64_t gs_vdev_guid; 66 char *gs_devid; 67 device_type_t gs_vdev_type; 68 uint64_t gs_vdev_expandtime; /* vdev expansion time */ 69 } guid_search_t; 70 71 /* 72 * Walks the vdev tree recursively looking for a matching devid. 73 * Returns B_TRUE as soon as a matching device is found, B_FALSE otherwise. 74 */ 75 static boolean_t 76 zfs_agent_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *arg) 77 { 78 guid_search_t *gsp = arg; 79 char *path = NULL; 80 uint_t c, children; 81 nvlist_t **child; 82 83 /* 84 * First iterate over any children. 85 */ 86 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, 87 &child, &children) == 0) { 88 for (c = 0; c < children; c++) { 89 if (zfs_agent_iter_vdev(zhp, child[c], gsp)) { 90 gsp->gs_vdev_type = DEVICE_TYPE_PRIMARY; 91 return (B_TRUE); 92 } 93 } 94 } 95 /* 96 * Iterate over any spares and cache devices 97 */ 98 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_SPARES, 99 &child, &children) == 0) { 100 for (c = 0; c < children; c++) { 101 if (zfs_agent_iter_vdev(zhp, child[c], gsp)) { 102 gsp->gs_vdev_type = DEVICE_TYPE_L2ARC; 103 return (B_TRUE); 104 } 105 } 106 } 107 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_L2CACHE, 108 &child, &children) == 0) { 109 for (c = 0; c < children; c++) { 110 if (zfs_agent_iter_vdev(zhp, child[c], gsp)) { 111 gsp->gs_vdev_type = DEVICE_TYPE_SPARE; 112 return (B_TRUE); 113 } 114 } 115 } 116 /* 117 * On a devid match, grab the vdev guid and expansion time, if any. 118 */ 119 if (gsp->gs_devid != NULL && 120 (nvlist_lookup_string(nvl, ZPOOL_CONFIG_DEVID, &path) == 0) && 121 (strcmp(gsp->gs_devid, path) == 0)) { 122 (void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID, 123 &gsp->gs_vdev_guid); 124 (void) nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_EXPANSION_TIME, 125 &gsp->gs_vdev_expandtime); 126 return (B_TRUE); 127 } 128 129 return (B_FALSE); 130 } 131 132 static int 133 zfs_agent_iter_pool(zpool_handle_t *zhp, void *arg) 134 { 135 guid_search_t *gsp = arg; 136 nvlist_t *config, *nvl; 137 138 /* 139 * For each vdev in this pool, look for a match by devid 140 */ 141 if ((config = zpool_get_config(zhp, NULL)) != NULL) { 142 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 143 &nvl) == 0) { 144 (void) zfs_agent_iter_vdev(zhp, nvl, gsp); 145 } 146 } 147 /* 148 * if a match was found then grab the pool guid 149 */ 150 if (gsp->gs_vdev_guid) { 151 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 152 &gsp->gs_pool_guid); 153 } 154 155 zpool_close(zhp); 156 return (gsp->gs_vdev_guid != 0); 157 } 158 159 void 160 zfs_agent_post_event(const char *class, const char *subclass, nvlist_t *nvl) 161 { 162 agent_event_t *event; 163 164 if (subclass == NULL) 165 subclass = ""; 166 167 event = malloc(sizeof (agent_event_t)); 168 if (event == NULL || nvlist_dup(nvl, &event->ae_nvl, 0) != 0) { 169 if (event) 170 free(event); 171 return; 172 } 173 174 if (strcmp(class, "sysevent.fs.zfs.vdev_check") == 0) { 175 class = EC_ZFS; 176 subclass = ESC_ZFS_VDEV_CHECK; 177 } 178 179 /* 180 * On ZFS on Linux, we don't get the expected FM_RESOURCE_REMOVED 181 * ereport from vdev_disk layer after a hot unplug. Fortunately we 182 * get a EC_DEV_REMOVE from our disk monitor and it is a suitable 183 * proxy so we remap it here for the benefit of the diagnosis engine. 184 */ 185 if ((strcmp(class, EC_DEV_REMOVE) == 0) && 186 (strcmp(subclass, ESC_DISK) == 0) && 187 (nvlist_exists(nvl, ZFS_EV_VDEV_GUID) || 188 nvlist_exists(nvl, DEV_IDENTIFIER))) { 189 nvlist_t *payload = event->ae_nvl; 190 struct timeval tv; 191 int64_t tod[2]; 192 uint64_t pool_guid = 0, vdev_guid = 0; 193 guid_search_t search = { 0 }; 194 device_type_t devtype = DEVICE_TYPE_PRIMARY; 195 196 class = "resource.fs.zfs.removed"; 197 subclass = ""; 198 199 (void) nvlist_add_string(payload, FM_CLASS, class); 200 (void) nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID, &pool_guid); 201 (void) nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &vdev_guid); 202 203 (void) gettimeofday(&tv, NULL); 204 tod[0] = tv.tv_sec; 205 tod[1] = tv.tv_usec; 206 (void) nvlist_add_int64_array(payload, FM_EREPORT_TIME, tod, 2); 207 208 /* 209 * For multipath, spare and l2arc devices ZFS_EV_VDEV_GUID or 210 * ZFS_EV_POOL_GUID may be missing so find them. 211 */ 212 (void) nvlist_lookup_string(nvl, DEV_IDENTIFIER, 213 &search.gs_devid); 214 (void) zpool_iter(g_zfs_hdl, zfs_agent_iter_pool, &search); 215 pool_guid = search.gs_pool_guid; 216 vdev_guid = search.gs_vdev_guid; 217 devtype = search.gs_vdev_type; 218 219 /* 220 * We want to avoid reporting "remove" events coming from 221 * libudev for VDEVs which were expanded recently (10s) and 222 * avoid activating spares in response to partitions being 223 * deleted and created in rapid succession. 224 */ 225 if (search.gs_vdev_expandtime != 0 && 226 search.gs_vdev_expandtime + 10 > tv.tv_sec) { 227 zed_log_msg(LOG_INFO, "agent post event: ignoring '%s' " 228 "for recently expanded device '%s'", EC_DEV_REMOVE, 229 search.gs_devid); 230 goto out; 231 } 232 233 (void) nvlist_add_uint64(payload, 234 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, pool_guid); 235 (void) nvlist_add_uint64(payload, 236 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vdev_guid); 237 switch (devtype) { 238 case DEVICE_TYPE_L2ARC: 239 (void) nvlist_add_string(payload, 240 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 241 VDEV_TYPE_L2CACHE); 242 break; 243 case DEVICE_TYPE_SPARE: 244 (void) nvlist_add_string(payload, 245 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, VDEV_TYPE_SPARE); 246 break; 247 case DEVICE_TYPE_PRIMARY: 248 (void) nvlist_add_string(payload, 249 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, VDEV_TYPE_DISK); 250 break; 251 } 252 253 zed_log_msg(LOG_INFO, "agent post event: mapping '%s' to '%s'", 254 EC_DEV_REMOVE, class); 255 } 256 257 (void) strlcpy(event->ae_class, class, sizeof (event->ae_class)); 258 (void) strlcpy(event->ae_subclass, subclass, 259 sizeof (event->ae_subclass)); 260 261 (void) pthread_mutex_lock(&agent_lock); 262 list_insert_tail(&agent_events, event); 263 (void) pthread_mutex_unlock(&agent_lock); 264 265 out: 266 (void) pthread_cond_signal(&agent_cond); 267 } 268 269 static void 270 zfs_agent_dispatch(const char *class, const char *subclass, nvlist_t *nvl) 271 { 272 /* 273 * The diagnosis engine subscribes to the following events. 274 * On illumos these subscriptions reside in: 275 * /usr/lib/fm/fmd/plugins/zfs-diagnosis.conf 276 */ 277 if (strstr(class, "ereport.fs.zfs.") != NULL || 278 strstr(class, "resource.fs.zfs.") != NULL || 279 strcmp(class, "sysevent.fs.zfs.vdev_remove") == 0 || 280 strcmp(class, "sysevent.fs.zfs.vdev_remove_dev") == 0 || 281 strcmp(class, "sysevent.fs.zfs.pool_destroy") == 0) { 282 fmd_module_recv(fmd_module_hdl("zfs-diagnosis"), nvl, class); 283 } 284 285 /* 286 * The retire agent subscribes to the following events. 287 * On illumos these subscriptions reside in: 288 * /usr/lib/fm/fmd/plugins/zfs-retire.conf 289 * 290 * NOTE: faults events come directly from our diagnosis engine 291 * and will not pass through the zfs kernel module. 292 */ 293 if (strcmp(class, FM_LIST_SUSPECT_CLASS) == 0 || 294 strcmp(class, "resource.fs.zfs.removed") == 0 || 295 strcmp(class, "resource.fs.zfs.statechange") == 0 || 296 strcmp(class, "sysevent.fs.zfs.vdev_remove") == 0) { 297 fmd_module_recv(fmd_module_hdl("zfs-retire"), nvl, class); 298 } 299 300 /* 301 * The SLM module only consumes disk events and vdev check events 302 * 303 * NOTE: disk events come directly from disk monitor and will 304 * not pass through the zfs kernel module. 305 */ 306 if (strstr(class, "EC_dev_") != NULL || 307 strcmp(class, EC_ZFS) == 0) { 308 (void) zfs_slm_event(class, subclass, nvl); 309 } 310 } 311 312 /* 313 * Events are consumed and dispatched from this thread 314 * An agent can also post an event so event list lock 315 * is not held when calling an agent. 316 * One event is consumed at a time. 317 */ 318 static void * 319 zfs_agent_consumer_thread(void *arg) 320 { 321 for (;;) { 322 agent_event_t *event; 323 324 (void) pthread_mutex_lock(&agent_lock); 325 326 /* wait for an event to show up */ 327 while (!agent_exiting && list_is_empty(&agent_events)) 328 (void) pthread_cond_wait(&agent_cond, &agent_lock); 329 330 if (agent_exiting) { 331 (void) pthread_mutex_unlock(&agent_lock); 332 zed_log_msg(LOG_INFO, "zfs_agent_consumer_thread: " 333 "exiting"); 334 return (NULL); 335 } 336 337 if ((event = (list_head(&agent_events))) != NULL) { 338 list_remove(&agent_events, event); 339 340 (void) pthread_mutex_unlock(&agent_lock); 341 342 /* dispatch to all event subscribers */ 343 zfs_agent_dispatch(event->ae_class, event->ae_subclass, 344 event->ae_nvl); 345 346 nvlist_free(event->ae_nvl); 347 free(event); 348 continue; 349 } 350 351 (void) pthread_mutex_unlock(&agent_lock); 352 } 353 354 return (NULL); 355 } 356 357 void 358 zfs_agent_init(libzfs_handle_t *zfs_hdl) 359 { 360 fmd_hdl_t *hdl; 361 362 g_zfs_hdl = zfs_hdl; 363 364 if (zfs_slm_init() != 0) 365 zed_log_die("Failed to initialize zfs slm"); 366 zed_log_msg(LOG_INFO, "Add Agent: init"); 367 368 hdl = fmd_module_hdl("zfs-diagnosis"); 369 _zfs_diagnosis_init(hdl); 370 if (!fmd_module_initialized(hdl)) 371 zed_log_die("Failed to initialize zfs diagnosis"); 372 373 hdl = fmd_module_hdl("zfs-retire"); 374 _zfs_retire_init(hdl); 375 if (!fmd_module_initialized(hdl)) 376 zed_log_die("Failed to initialize zfs retire"); 377 378 list_create(&agent_events, sizeof (agent_event_t), 379 offsetof(struct agent_event, ae_node)); 380 381 if (pthread_create(&g_agents_tid, NULL, zfs_agent_consumer_thread, 382 NULL) != 0) { 383 list_destroy(&agent_events); 384 zed_log_die("Failed to initialize agents"); 385 } 386 } 387 388 void 389 zfs_agent_fini(void) 390 { 391 fmd_hdl_t *hdl; 392 agent_event_t *event; 393 394 agent_exiting = 1; 395 (void) pthread_cond_signal(&agent_cond); 396 397 /* wait for zfs_enum_pools thread to complete */ 398 (void) pthread_join(g_agents_tid, NULL); 399 400 /* drain any pending events */ 401 while ((event = (list_head(&agent_events))) != NULL) { 402 list_remove(&agent_events, event); 403 nvlist_free(event->ae_nvl); 404 free(event); 405 } 406 407 list_destroy(&agent_events); 408 409 if ((hdl = fmd_module_hdl("zfs-retire")) != NULL) { 410 _zfs_retire_fini(hdl); 411 fmd_hdl_unregister(hdl); 412 } 413 if ((hdl = fmd_module_hdl("zfs-diagnosis")) != NULL) { 414 _zfs_diagnosis_fini(hdl); 415 fmd_hdl_unregister(hdl); 416 } 417 418 zed_log_msg(LOG_INFO, "Add Agent: fini"); 419 zfs_slm_fini(); 420 421 g_zfs_hdl = NULL; 422 } 423