spa_config.c revision 209962
130302Sjoerg/* 288551Sjoerg * CDDL HEADER START 330302Sjoerg * 430302Sjoerg * The contents of this file are subject to the terms of the 530302Sjoerg * Common Development and Distribution License (the "License"). 630302Sjoerg * You may not use this file except in compliance with the License. 730302Sjoerg * 830302Sjoerg * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 930302Sjoerg * or http://www.opensolaris.org/os/licensing. 1030302Sjoerg * See the License for the specific language governing permissions 1130302Sjoerg * and limitations under the License. 1230302Sjoerg * 1330302Sjoerg * When distributing Covered Code, include this CDDL HEADER in each 1430302Sjoerg * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 1530302Sjoerg * If applicable, add the following below this CDDL HEADER, with the 1630302Sjoerg * fields enclosed by brackets "[]" replaced with your own identifying 1730302Sjoerg * information: Portions Copyright [yyyy] [name of copyright owner] 1830302Sjoerg * 1930302Sjoerg * CDDL HEADER END 2030302Sjoerg */ 2130302Sjoerg 2230302Sjoerg/* 2330302Sjoerg * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 2430302Sjoerg * Use is subject to license terms. 2530302Sjoerg */ 2630302Sjoerg 27114589Sobrien#include <sys/zfs_context.h> 28114589Sobrien#include <sys/spa.h> 2932274Scharnier#include <sys/spa_impl.h> 3088559Sjoerg#include <sys/nvpair.h> 3130302Sjoerg#include <sys/uio.h> 3230302Sjoerg#include <sys/fs/zfs.h> 3330302Sjoerg#include <sys/vdev_impl.h> 3430302Sjoerg#include <sys/zfs_ioctl.h> 3530302Sjoerg#include <sys/utsname.h> 3630302Sjoerg#include <sys/sunddi.h> 3730302Sjoerg#ifdef _KERNEL 3830302Sjoerg#include <sys/kobj.h> 3978732Sdd#endif 4030302Sjoerg 4130302Sjoerg/* 4230302Sjoerg * Pool configuration repository. 4330302Sjoerg * 4432274Scharnier * Pool configuration is stored as a packed nvlist on the filesystem. By 4530302Sjoerg * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot 4630302Sjoerg * (when the ZFS module is loaded). Pools can also have the 'cachefile' 4730302Sjoerg * property set that allows them to be stored in an alternate location until 4830302Sjoerg * the control of external software. 4930302Sjoerg * 5030302Sjoerg * For each cache file, we have a single nvlist which holds all the 5130302Sjoerg * configuration information. When the module loads, we read this information 5230302Sjoerg * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is 5330302Sjoerg * maintained independently in spa.c. Whenever the namespace is modified, or 5430302Sjoerg * the configuration of a pool is changed, we call spa_config_sync(), which 5530302Sjoerg * walks through all the active pools and writes the configuration to disk. 5630302Sjoerg */ 5730302Sjoerg 5830302Sjoergstatic uint64_t spa_config_generation = 1; 5988551Sjoerg 6088551Sjoerg/* 6130302Sjoerg * This can be overridden in userland to preserve an alternate namespace for 6230302Sjoerg * userland pools when doing testing. 6330302Sjoerg */ 6430302Sjoergconst char *spa_config_path = ZPOOL_CACHE; 6530302Sjoerg 6630302Sjoerg/* 6730302Sjoerg * Called when the module is first loaded, this routine loads the configuration 6830302Sjoerg * file into the SPA namespace. It does not actually open or load the pools; it 6930302Sjoerg * only populates the namespace. 7030302Sjoerg */ 7130302Sjoergvoid 7230302Sjoergspa_config_load(void) 7330302Sjoerg{ 7430302Sjoerg void *buf = NULL; 7530302Sjoerg nvlist_t *nvlist, *child; 7630302Sjoerg nvpair_t *nvpair; 7730302Sjoerg spa_t *spa; 7830302Sjoerg char *pathname; 7930302Sjoerg struct _buf *file; 8030302Sjoerg uint64_t fsize; 8130302Sjoerg 8230302Sjoerg /* 8330302Sjoerg * Open the configuration file. 8430302Sjoerg */ 8530302Sjoerg pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 8630302Sjoerg 8730302Sjoerg (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path); 8830302Sjoerg 8930302Sjoerg file = kobj_open_file(pathname); 9030302Sjoerg 91173717Sjb if (file == (struct _buf *)-1) { 9230302Sjoerg ZFS_LOG(1, "Cannot open %s.", pathname); 9330302Sjoerg goto out; 9430302Sjoerg } 9530302Sjoerg 9630302Sjoerg if (kobj_get_filesize(file, &fsize) != 0) { 9730302Sjoerg ZFS_LOG(1, "Cannot get size of %s.", pathname); 9830302Sjoerg goto out; 9930302Sjoerg } 10030302Sjoerg 10130302Sjoerg buf = kmem_alloc(fsize, KM_SLEEP); 10230302Sjoerg 10330302Sjoerg /* 10430302Sjoerg * Read the nvlist from the file. 10530302Sjoerg */ 10630302Sjoerg if (kobj_read_file(file, buf, fsize, 0) < 0) { 10730302Sjoerg ZFS_LOG(1, "Cannot read %s.", pathname); 10830302Sjoerg goto out; 10930302Sjoerg } 11030302Sjoerg 11130302Sjoerg /* 11230302Sjoerg * Unpack the nvlist. 11330302Sjoerg */ 11430302Sjoerg if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0) 11530302Sjoerg goto out; 11630302Sjoerg 11730302Sjoerg ZFS_LOG(1, "File %s loaded.", pathname); 11830302Sjoerg 11930302Sjoerg /* 12030302Sjoerg * Iterate over all elements in the nvlist, creating a new spa_t for 12130302Sjoerg * each one with the specified configuration. 12230302Sjoerg */ 12330302Sjoerg mutex_enter(&spa_namespace_lock); 12430302Sjoerg nvpair = NULL; 12530302Sjoerg while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) { 12630302Sjoerg 12730302Sjoerg if (nvpair_type(nvpair) != DATA_TYPE_NVLIST) 12830302Sjoerg continue; 12930302Sjoerg 13030302Sjoerg VERIFY(nvpair_value_nvlist(nvpair, &child) == 0); 13130302Sjoerg 13230302Sjoerg if (spa_lookup(nvpair_name(nvpair)) != NULL) 13330302Sjoerg continue; 13430302Sjoerg spa = spa_add(nvpair_name(nvpair), NULL); 13530302Sjoerg 13630302Sjoerg /* 13730302Sjoerg * We blindly duplicate the configuration here. If it's 13830302Sjoerg * invalid, we will catch it when the pool is first opened. 13930302Sjoerg */ 14030302Sjoerg VERIFY(nvlist_dup(child, &spa->spa_config, 0) == 0); 14130302Sjoerg } 14230302Sjoerg mutex_exit(&spa_namespace_lock); 14330302Sjoerg 14430302Sjoerg nvlist_free(nvlist); 14530302Sjoerg 14630302Sjoergout: 14730302Sjoerg kmem_free(pathname, MAXPATHLEN); 14830302Sjoerg if (buf != NULL) 14930302Sjoerg kmem_free(buf, fsize); 15030302Sjoerg if (file != (struct _buf *)-1) 15130302Sjoerg kobj_close_file(file); 15230302Sjoerg} 15330302Sjoerg 15430302Sjoergstatic void 15530302Sjoergspa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl) 15630302Sjoerg{ 15730302Sjoerg int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX; 15830302Sjoerg char *buf, *temp; 15930302Sjoerg size_t buflen; 16030302Sjoerg vnode_t *vp; 16188551Sjoerg 16288551Sjoerg /* 16388551Sjoerg * If the nvlist is empty (NULL), then remove the old cachefile. 16488551Sjoerg */ 16588551Sjoerg if (nvl == NULL) { 16688551Sjoerg (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE); 16788551Sjoerg return; 16888551Sjoerg } 16988551Sjoerg 17088551Sjoerg /* 17188551Sjoerg * Pack the configuration into a buffer. 17288551Sjoerg */ 17388551Sjoerg VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0); 17488551Sjoerg 17588536Sjoerg buf = kmem_alloc(buflen, KM_SLEEP); 17688536Sjoerg temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 17788536Sjoerg 17888724Sjoerg VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR, 17988724Sjoerg KM_SLEEP) == 0); 18088724Sjoerg 18188724Sjoerg /* 18230302Sjoerg * Write the configuration to disk. We need to do the traditional 18330302Sjoerg * 'write to temporary file, sync, move over original' to make sure we 18430302Sjoerg * always have a consistent view of the data. 18530302Sjoerg */ 18630302Sjoerg (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path); 18730302Sjoerg 18830302Sjoerg if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) { 189173717Sjb if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE, 19030302Sjoerg 0, RLIM64_INFINITY, kcred, NULL) == 0 && 19130302Sjoerg VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) { 19230302Sjoerg (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE); 19330302Sjoerg } 19430302Sjoerg (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL); 19530302Sjoerg } 19630302Sjoerg 19730302Sjoerg (void) vn_remove(temp, UIO_SYSSPACE, RMFILE); 19830302Sjoerg 19930302Sjoerg kmem_free(buf, buflen); 20032274Scharnier kmem_free(temp, MAXPATHLEN); 20130302Sjoerg} 20230302Sjoerg 20332274Scharnier/* 20432274Scharnier * Synchronize pool configuration to disk. This must be called with the 20532274Scharnier * namespace lock held. 20632274Scharnier */ 20730302Sjoergvoid 20830302Sjoergspa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent) 20930302Sjoerg{ 21030302Sjoerg spa_config_dirent_t *dp, *tdp; 21130302Sjoerg nvlist_t *nvl; 21230302Sjoerg 21330302Sjoerg ASSERT(MUTEX_HELD(&spa_namespace_lock)); 21430302Sjoerg 21530302Sjoerg if (rootdir == NULL || !(spa_mode_global & FWRITE)) 21630302Sjoerg return; 21730302Sjoerg 21830302Sjoerg /* 21930302Sjoerg * Iterate over all cachefiles for the pool, past or present. When the 22030302Sjoerg * cachefile is changed, the new one is pushed onto this list, allowing 22130302Sjoerg * us to update previous cachefiles that no longer contain this pool. 22230302Sjoerg */ 22330302Sjoerg for (dp = list_head(&target->spa_config_list); dp != NULL; 22488551Sjoerg dp = list_next(&target->spa_config_list, dp)) { 22588536Sjoerg spa_t *spa = NULL; 22688724Sjoerg if (dp->scd_path == NULL) 22730302Sjoerg continue; 22830302Sjoerg 22930302Sjoerg /* 23030302Sjoerg * Iterate over all pools, adding any matching pools to 'nvl'. 23130302Sjoerg */ 23230302Sjoerg nvl = NULL; 23330302Sjoerg while ((spa = spa_next(spa)) != NULL) { 23430302Sjoerg if (spa == target && removing) 23530302Sjoerg continue; 23630302Sjoerg 23730302Sjoerg mutex_enter(&spa->spa_props_lock); 23830302Sjoerg tdp = list_head(&spa->spa_config_list); 23930302Sjoerg if (spa->spa_config == NULL || 24030302Sjoerg tdp->scd_path == NULL || 24130302Sjoerg strcmp(tdp->scd_path, dp->scd_path) != 0) { 24230302Sjoerg mutex_exit(&spa->spa_props_lock); 24330302Sjoerg continue; 24430302Sjoerg } 24530302Sjoerg 24630302Sjoerg if (nvl == NULL) 24730302Sjoerg VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, 24830302Sjoerg KM_SLEEP) == 0); 24930302Sjoerg 25030302Sjoerg VERIFY(nvlist_add_nvlist(nvl, spa->spa_name, 25130302Sjoerg spa->spa_config) == 0); 25230302Sjoerg mutex_exit(&spa->spa_props_lock); 25330302Sjoerg } 25430302Sjoerg 25530302Sjoerg spa_config_write(dp, nvl); 25630302Sjoerg nvlist_free(nvl); 25744345Sgj } 25830302Sjoerg 25930302Sjoerg /* 26030302Sjoerg * Remove any config entries older than the current one. 26130302Sjoerg */ 26230302Sjoerg dp = list_head(&target->spa_config_list); 26330302Sjoerg while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) { 26430302Sjoerg list_remove(&target->spa_config_list, tdp); 265 if (tdp->scd_path != NULL) 266 spa_strfree(tdp->scd_path); 267 kmem_free(tdp, sizeof (spa_config_dirent_t)); 268 } 269 270 spa_config_generation++; 271 272 if (postsysevent) 273 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC); 274} 275 276/* 277 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache, 278 * and we don't want to allow the local zone to see all the pools anyway. 279 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration 280 * information for all pool visible within the zone. 281 */ 282nvlist_t * 283spa_all_configs(uint64_t *generation) 284{ 285 nvlist_t *pools; 286 spa_t *spa = NULL; 287 288 if (*generation == spa_config_generation) 289 return (NULL); 290 291 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0); 292 293 mutex_enter(&spa_namespace_lock); 294 while ((spa = spa_next(spa)) != NULL) { 295 if (INGLOBALZONE(curthread) || 296 zone_dataset_visible(spa_name(spa), NULL)) { 297 mutex_enter(&spa->spa_props_lock); 298 VERIFY(nvlist_add_nvlist(pools, spa_name(spa), 299 spa->spa_config) == 0); 300 mutex_exit(&spa->spa_props_lock); 301 } 302 } 303 *generation = spa_config_generation; 304 mutex_exit(&spa_namespace_lock); 305 306 return (pools); 307} 308 309void 310spa_config_set(spa_t *spa, nvlist_t *config) 311{ 312 mutex_enter(&spa->spa_props_lock); 313 if (spa->spa_config != NULL) 314 nvlist_free(spa->spa_config); 315 spa->spa_config = config; 316 mutex_exit(&spa->spa_props_lock); 317} 318 319/* 320 * Generate the pool's configuration based on the current in-core state. 321 * We infer whether to generate a complete config or just one top-level config 322 * based on whether vd is the root vdev. 323 */ 324nvlist_t * 325spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) 326{ 327 nvlist_t *config, *nvroot; 328 vdev_t *rvd = spa->spa_root_vdev; 329 unsigned long hostid = 0; 330 boolean_t locked = B_FALSE; 331 332 if (vd == NULL) { 333 vd = rvd; 334 locked = B_TRUE; 335 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 336 } 337 338 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) == 339 (SCL_CONFIG | SCL_STATE)); 340 341 /* 342 * If txg is -1, report the current value of spa->spa_config_txg. 343 */ 344 if (txg == -1ULL) 345 txg = spa->spa_config_txg; 346 347 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0); 348 349 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 350 spa_version(spa)) == 0); 351 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 352 spa_name(spa)) == 0); 353 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 354 spa_state(spa)) == 0); 355 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 356 txg) == 0); 357 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 358 spa_guid(spa)) == 0); 359 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid); 360 if (hostid != 0) { 361 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, 362 hostid) == 0); 363 } 364 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, 365 utsname.nodename) == 0); 366 367 if (vd != rvd) { 368 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID, 369 vd->vdev_top->vdev_guid) == 0); 370 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID, 371 vd->vdev_guid) == 0); 372 if (vd->vdev_isspare) 373 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE, 374 1ULL) == 0); 375 if (vd->vdev_islog) 376 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG, 377 1ULL) == 0); 378 vd = vd->vdev_top; /* label contains top config */ 379 } 380 381 nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE); 382 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 383 nvlist_free(nvroot); 384 385 if (locked) 386 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 387 388 return (config); 389} 390 391/* 392 * Update all disk labels, generate a fresh config based on the current 393 * in-core state, and sync the global config cache (do not sync the config 394 * cache if this is a booting rootpool). 395 */ 396void 397spa_config_update(spa_t *spa, int what) 398{ 399 vdev_t *rvd = spa->spa_root_vdev; 400 uint64_t txg; 401 int c; 402 403 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 404 405 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 406 txg = spa_last_synced_txg(spa) + 1; 407 if (what == SPA_CONFIG_UPDATE_POOL) { 408 vdev_config_dirty(rvd); 409 } else { 410 /* 411 * If we have top-level vdevs that were added but have 412 * not yet been prepared for allocation, do that now. 413 * (It's safe now because the config cache is up to date, 414 * so it will be able to translate the new DVAs.) 415 * See comments in spa_vdev_add() for full details. 416 */ 417 for (c = 0; c < rvd->vdev_children; c++) { 418 vdev_t *tvd = rvd->vdev_child[c]; 419 if (tvd->vdev_ms_array == 0) { 420 vdev_init(tvd, txg); 421 vdev_config_dirty(tvd); 422 } 423 } 424 } 425 spa_config_exit(spa, SCL_ALL, FTAG); 426 427 /* 428 * Wait for the mosconfig to be regenerated and synced. 429 */ 430 txg_wait_synced(spa->spa_dsl_pool, txg); 431 432 /* 433 * Update the global config cache to reflect the new mosconfig. 434 */ 435 if (!spa->spa_is_root) 436 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL); 437 438 if (what == SPA_CONFIG_UPDATE_POOL) 439 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); 440} 441