1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 241b497ab8SAdam H. Leventhal * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25*14372834SHans Rosenfeld * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 275aeb9474SGarrett D'Amore */ 28fa9e4066Sahrens 29fa9e4066Sahrens /* 303e30c24aSWill Andrews * SPA: Storage Pool Allocator 313e30c24aSWill Andrews * 32fa9e4066Sahrens * This file contains all the routines used when modifying on-disk SPA state. 33fa9e4066Sahrens * This includes opening, importing, destroying, exporting a pool, and syncing a 34fa9e4066Sahrens * pool. 35fa9e4066Sahrens */ 36fa9e4066Sahrens 37fa9e4066Sahrens #include <sys/zfs_context.h> 38ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 39fa9e4066Sahrens #include <sys/spa_impl.h> 40fa9e4066Sahrens #include <sys/zio.h> 41fa9e4066Sahrens #include <sys/zio_checksum.h> 42fa9e4066Sahrens #include <sys/dmu.h> 43fa9e4066Sahrens #include <sys/dmu_tx.h> 44fa9e4066Sahrens #include <sys/zap.h> 45fa9e4066Sahrens #include <sys/zil.h> 46b24ab676SJeff Bonwick #include <sys/ddt.h> 47fa9e4066Sahrens #include <sys/vdev_impl.h> 48fa9e4066Sahrens #include <sys/metaslab.h> 4988ecc943SGeorge Wilson #include <sys/metaslab_impl.h> 50fa9e4066Sahrens #include <sys/uberblock_impl.h> 51fa9e4066Sahrens #include <sys/txg.h> 52fa9e4066Sahrens #include <sys/avl.h> 53fa9e4066Sahrens #include <sys/dmu_traverse.h> 54b1b8ab34Slling #include <sys/dmu_objset.h> 55fa9e4066Sahrens #include <sys/unique.h> 56fa9e4066Sahrens #include <sys/dsl_pool.h> 57b1b8ab34Slling #include <sys/dsl_dataset.h> 58fa9e4066Sahrens #include <sys/dsl_dir.h> 59fa9e4066Sahrens #include <sys/dsl_prop.h> 60b1b8ab34Slling #include <sys/dsl_synctask.h> 61fa9e4066Sahrens #include <sys/fs/zfs.h> 62fa94a07fSbrendan #include <sys/arc.h> 63fa9e4066Sahrens #include <sys/callb.h> 6495173954Sek #include <sys/systeminfo.h> 65e7cbe64fSgw #include <sys/spa_boot.h> 66573ca77eSGeorge Wilson #include <sys/zfs_ioctl.h> 673f9d6ad7SLin Ling #include <sys/dsl_scan.h> 68ad135b5dSChristopher Siden #include <sys/zfeature.h> 693b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 70fa9e4066Sahrens 715679c89fSjv #ifdef _KERNEL 72dedec472SJack Meng #include <sys/bootprops.h> 7335a5a358SJonathan Adams #include <sys/callb.h> 7435a5a358SJonathan Adams #include <sys/cpupart.h> 7535a5a358SJonathan Adams #include <sys/pool.h> 7635a5a358SJonathan Adams #include <sys/sysdc.h> 7735a5a358SJonathan Adams #include <sys/zone.h> 785679c89fSjv #endif /* _KERNEL */ 795679c89fSjv 80990b4856Slling #include "zfs_prop.h" 81b7b97454Sperrin #include "zfs_comutil.h" 82990b4856Slling 833cb69f73SWill Andrews /* 843cb69f73SWill Andrews * The interval, in seconds, at which failed configuration cache file writes 853cb69f73SWill Andrews * should be retried. 863cb69f73SWill Andrews */ 873cb69f73SWill Andrews static int zfs_ccw_retry_interval = 300; 883cb69f73SWill Andrews 8935a5a358SJonathan Adams typedef enum zti_modes { 90ec94d322SAdam Leventhal ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 91ec94d322SAdam Leventhal ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 92ec94d322SAdam Leventhal ZTI_MODE_NULL, /* don't create a taskq */ 93ec94d322SAdam Leventhal ZTI_NMODES 9435a5a358SJonathan Adams } zti_modes_t; 95416e0cd8Sek 96ec94d322SAdam Leventhal #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 97ec94d322SAdam Leventhal #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 98ec94d322SAdam Leventhal #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 992e0c549eSJonathan Adams 100ec94d322SAdam Leventhal #define ZTI_N(n) ZTI_P(n, 1) 101ec94d322SAdam Leventhal #define ZTI_ONE ZTI_N(1) 1022e0c549eSJonathan Adams 1032e0c549eSJonathan Adams typedef struct zio_taskq_info { 104ec94d322SAdam Leventhal zti_modes_t zti_mode; 10580eb36f2SGeorge Wilson uint_t zti_value; 106ec94d322SAdam Leventhal uint_t zti_count; 1072e0c549eSJonathan Adams } zio_taskq_info_t; 1082e0c549eSJonathan Adams 1092e0c549eSJonathan Adams static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 11035a5a358SJonathan Adams "issue", "issue_high", "intr", "intr_high" 1112e0c549eSJonathan Adams }; 1122e0c549eSJonathan Adams 11380eb36f2SGeorge Wilson /* 114ec94d322SAdam Leventhal * This table defines the taskq settings for each ZFS I/O type. When 115ec94d322SAdam Leventhal * initializing a pool, we use this table to create an appropriately sized 116ec94d322SAdam Leventhal * taskq. Some operations are low volume and therefore have a small, static 117ec94d322SAdam Leventhal * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 118ec94d322SAdam Leventhal * macros. Other operations process a large amount of data; the ZTI_BATCH 119ec94d322SAdam Leventhal * macro causes us to create a taskq oriented for throughput. Some operations 120ec94d322SAdam Leventhal * are so high frequency and short-lived that the taskq itself can become a a 121ec94d322SAdam Leventhal * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 122ec94d322SAdam Leventhal * additional degree of parallelism specified by the number of threads per- 123ec94d322SAdam Leventhal * taskq and the number of taskqs; when dispatching an event in this case, the 124ec94d322SAdam Leventhal * particular taskq is chosen at random. 125ec94d322SAdam Leventhal * 126ec94d322SAdam Leventhal * The different taskq priorities are to handle the different contexts (issue 127ec94d322SAdam Leventhal * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 128ec94d322SAdam Leventhal * need to be handled with minimum delay. 12980eb36f2SGeorge Wilson */ 13080eb36f2SGeorge Wilson const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 13180eb36f2SGeorge Wilson /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 132ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 1331b497ab8SAdam H. Leventhal { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 134ec94d322SAdam Leventhal { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 135ec94d322SAdam Leventhal { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 136ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 137ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 1382e0c549eSJonathan Adams }; 1392e0c549eSJonathan Adams 1403b2aab18SMatthew Ahrens static void spa_sync_version(void *arg, dmu_tx_t *tx); 1413b2aab18SMatthew Ahrens static void spa_sync_props(void *arg, dmu_tx_t *tx); 14289a89ebfSlling static boolean_t spa_has_active_shared_spare(spa_t *spa); 1431195e687SMark J Musante static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 1441195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 1451195e687SMark J Musante char **ereport); 146cb04b873SMark J Musante static void spa_vdev_resilver_done(spa_t *spa); 147990b4856Slling 14869962b56SMatthew Ahrens uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 14935a5a358SJonathan Adams id_t zio_taskq_psrset_bind = PS_NONE; 15035a5a358SJonathan Adams boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 15135a5a358SJonathan Adams uint_t zio_taskq_basedc = 80; /* base duty cycle */ 15235a5a358SJonathan Adams 15335a5a358SJonathan Adams boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 15401f55e48SGeorge Wilson extern int zfs_sync_pass_deferred_free; 15535a5a358SJonathan Adams 15635a5a358SJonathan Adams /* 15735a5a358SJonathan Adams * This (illegal) pool name is used when temporarily importing a spa_t in order 15835a5a358SJonathan Adams * to get the vdev stats associated with the imported devices. 15935a5a358SJonathan Adams */ 16035a5a358SJonathan Adams #define TRYIMPORT_NAME "$import" 16135a5a358SJonathan Adams 162990b4856Slling /* 163990b4856Slling * ========================================================================== 164990b4856Slling * SPA properties routines 165990b4856Slling * ========================================================================== 166990b4856Slling */ 167990b4856Slling 168990b4856Slling /* 169990b4856Slling * Add a (source=src, propname=propval) list to an nvlist. 170990b4856Slling */ 1719d82f4f6Slling static void 172990b4856Slling spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 173990b4856Slling uint64_t intval, zprop_source_t src) 174990b4856Slling { 175990b4856Slling const char *propname = zpool_prop_to_name(prop); 176990b4856Slling nvlist_t *propval; 177990b4856Slling 1789d82f4f6Slling VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1799d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 180990b4856Slling 1819d82f4f6Slling if (strval != NULL) 1829d82f4f6Slling VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 1839d82f4f6Slling else 1849d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 185990b4856Slling 1869d82f4f6Slling VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 187990b4856Slling nvlist_free(propval); 188990b4856Slling } 189990b4856Slling 190990b4856Slling /* 191990b4856Slling * Get property values from the spa configuration. 192990b4856Slling */ 1939d82f4f6Slling static void 194990b4856Slling spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 195990b4856Slling { 1964263d13fSGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 197ad135b5dSChristopher Siden dsl_pool_t *pool = spa->spa_dsl_pool; 1982e4c9986SGeorge Wilson uint64_t size, alloc, cap, version; 199990b4856Slling zprop_source_t src = ZPROP_SRC_NONE; 200c5904d13Seschrock spa_config_dirent_t *dp; 2012e4c9986SGeorge Wilson metaslab_class_t *mc = spa_normal_class(spa); 202990b4856Slling 203e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 204e14bb325SJeff Bonwick 2054263d13fSGeorge Wilson if (rvd != NULL) { 206485bbbf5SGeorge Wilson alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 207b24ab676SJeff Bonwick size = metaslab_class_get_space(spa_normal_class(spa)); 208379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 209379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 210485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 211485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 212485bbbf5SGeorge Wilson size - alloc, src); 2134263d13fSGeorge Wilson 2142e4c9986SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 2152e4c9986SGeorge Wilson metaslab_class_fragmentation(mc), src); 2162e4c9986SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 2172e4c9986SGeorge Wilson metaslab_class_expandable_space(mc), src); 218f9af39baSGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 219f9af39baSGeorge Wilson (spa_mode(spa) == FREAD), src); 220379c004dSEric Schrock 221485bbbf5SGeorge Wilson cap = (size == 0) ? 0 : (alloc * 100 / size); 222379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 223379c004dSEric Schrock 224b24ab676SJeff Bonwick spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 225b24ab676SJeff Bonwick ddt_get_pool_dedup_ratio(spa), src); 226b24ab676SJeff Bonwick 227379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 2284263d13fSGeorge Wilson rvd->vdev_state, src); 229379c004dSEric Schrock 230379c004dSEric Schrock version = spa_version(spa); 231379c004dSEric Schrock if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 232379c004dSEric Schrock src = ZPROP_SRC_DEFAULT; 233379c004dSEric Schrock else 234379c004dSEric Schrock src = ZPROP_SRC_LOCAL; 235379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 236379c004dSEric Schrock } 237990b4856Slling 238ad135b5dSChristopher Siden if (pool != NULL) { 239ad135b5dSChristopher Siden /* 240ad135b5dSChristopher Siden * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 241ad135b5dSChristopher Siden * when opening pools before this version freedir will be NULL. 242ad135b5dSChristopher Siden */ 2437fd05ac4SMatthew Ahrens if (pool->dp_free_dir != NULL) { 244ad135b5dSChristopher Siden spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 245c1379625SJustin T. Gibbs dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 246c1379625SJustin T. Gibbs src); 247ad135b5dSChristopher Siden } else { 248ad135b5dSChristopher Siden spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 249ad135b5dSChristopher Siden NULL, 0, src); 250ad135b5dSChristopher Siden } 2517fd05ac4SMatthew Ahrens 2527fd05ac4SMatthew Ahrens if (pool->dp_leak_dir != NULL) { 2537fd05ac4SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 254c1379625SJustin T. Gibbs dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 255c1379625SJustin T. Gibbs src); 2567fd05ac4SMatthew Ahrens } else { 2577fd05ac4SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 2587fd05ac4SMatthew Ahrens NULL, 0, src); 2597fd05ac4SMatthew Ahrens } 260ad135b5dSChristopher Siden } 261ad135b5dSChristopher Siden 2629d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 263990b4856Slling 2648704186eSDan McDonald if (spa->spa_comment != NULL) { 2658704186eSDan McDonald spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 2668704186eSDan McDonald 0, ZPROP_SRC_LOCAL); 2678704186eSDan McDonald } 2688704186eSDan McDonald 2699d82f4f6Slling if (spa->spa_root != NULL) 2709d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 2719d82f4f6Slling 0, ZPROP_SRC_LOCAL); 272990b4856Slling 273b5152584SMatthew Ahrens if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 274b5152584SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 275b5152584SMatthew Ahrens MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 276b5152584SMatthew Ahrens } else { 277b5152584SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 278b5152584SMatthew Ahrens SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 279b5152584SMatthew Ahrens } 280b5152584SMatthew Ahrens 281c5904d13Seschrock if ((dp = list_head(&spa->spa_config_list)) != NULL) { 282c5904d13Seschrock if (dp->scd_path == NULL) { 2839d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 284c5904d13Seschrock "none", 0, ZPROP_SRC_LOCAL); 285c5904d13Seschrock } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 2869d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 287c5904d13Seschrock dp->scd_path, 0, ZPROP_SRC_LOCAL); 2882f8aaab3Seschrock } 2892f8aaab3Seschrock } 290990b4856Slling } 291990b4856Slling 292990b4856Slling /* 293990b4856Slling * Get zpool property values. 294990b4856Slling */ 295990b4856Slling int 296990b4856Slling spa_prop_get(spa_t *spa, nvlist_t **nvp) 297990b4856Slling { 298b24ab676SJeff Bonwick objset_t *mos = spa->spa_meta_objset; 299990b4856Slling zap_cursor_t zc; 300990b4856Slling zap_attribute_t za; 301990b4856Slling int err; 302990b4856Slling 3039d82f4f6Slling VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 304990b4856Slling 305e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 306e14bb325SJeff Bonwick 307990b4856Slling /* 308990b4856Slling * Get properties from the spa config. 309990b4856Slling */ 3109d82f4f6Slling spa_prop_get_config(spa, nvp); 311990b4856Slling 312990b4856Slling /* If no pool property object, no more prop to get. */ 313afee20e4SGeorge Wilson if (mos == NULL || spa->spa_pool_props_object == 0) { 314990b4856Slling mutex_exit(&spa->spa_props_lock); 315990b4856Slling return (0); 316990b4856Slling } 317990b4856Slling 318990b4856Slling /* 319990b4856Slling * Get properties from the MOS pool property object. 320990b4856Slling */ 321990b4856Slling for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 322990b4856Slling (err = zap_cursor_retrieve(&zc, &za)) == 0; 323990b4856Slling zap_cursor_advance(&zc)) { 324990b4856Slling uint64_t intval = 0; 325990b4856Slling char *strval = NULL; 326990b4856Slling zprop_source_t src = ZPROP_SRC_DEFAULT; 327990b4856Slling zpool_prop_t prop; 328990b4856Slling 329990b4856Slling if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 330990b4856Slling continue; 331990b4856Slling 332990b4856Slling switch (za.za_integer_length) { 333990b4856Slling case 8: 334990b4856Slling /* integer property */ 335990b4856Slling if (za.za_first_integer != 336990b4856Slling zpool_prop_default_numeric(prop)) 337990b4856Slling src = ZPROP_SRC_LOCAL; 338990b4856Slling 339990b4856Slling if (prop == ZPOOL_PROP_BOOTFS) { 340990b4856Slling dsl_pool_t *dp; 341990b4856Slling dsl_dataset_t *ds = NULL; 342990b4856Slling 343990b4856Slling dp = spa_get_dsl(spa); 3443b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 345745cd3c5Smaybee if (err = dsl_dataset_hold_obj(dp, 346745cd3c5Smaybee za.za_first_integer, FTAG, &ds)) { 3473b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 348990b4856Slling break; 349990b4856Slling } 350990b4856Slling 351990b4856Slling strval = kmem_alloc( 352990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 353990b4856Slling KM_SLEEP); 354990b4856Slling dsl_dataset_name(ds, strval); 355745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 3563b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 357990b4856Slling } else { 358990b4856Slling strval = NULL; 359990b4856Slling intval = za.za_first_integer; 360990b4856Slling } 361990b4856Slling 3629d82f4f6Slling spa_prop_add_list(*nvp, prop, strval, intval, src); 363990b4856Slling 364990b4856Slling if (strval != NULL) 365990b4856Slling kmem_free(strval, 366990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 367990b4856Slling 368990b4856Slling break; 369990b4856Slling 370990b4856Slling case 1: 371990b4856Slling /* string property */ 372990b4856Slling strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 373990b4856Slling err = zap_lookup(mos, spa->spa_pool_props_object, 374990b4856Slling za.za_name, 1, za.za_num_integers, strval); 375990b4856Slling if (err) { 376990b4856Slling kmem_free(strval, za.za_num_integers); 377990b4856Slling break; 378990b4856Slling } 3799d82f4f6Slling spa_prop_add_list(*nvp, prop, strval, 0, src); 380990b4856Slling kmem_free(strval, za.za_num_integers); 381990b4856Slling break; 382990b4856Slling 383990b4856Slling default: 384990b4856Slling break; 385990b4856Slling } 386990b4856Slling } 387990b4856Slling zap_cursor_fini(&zc); 388990b4856Slling mutex_exit(&spa->spa_props_lock); 389990b4856Slling out: 390990b4856Slling if (err && err != ENOENT) { 391990b4856Slling nvlist_free(*nvp); 3929d82f4f6Slling *nvp = NULL; 393990b4856Slling return (err); 394990b4856Slling } 395990b4856Slling 396990b4856Slling return (0); 397990b4856Slling } 398990b4856Slling 399990b4856Slling /* 400990b4856Slling * Validate the given pool properties nvlist and modify the list 401990b4856Slling * for the property values to be set. 402990b4856Slling */ 403990b4856Slling static int 404990b4856Slling spa_prop_validate(spa_t *spa, nvlist_t *props) 405990b4856Slling { 406990b4856Slling nvpair_t *elem; 407990b4856Slling int error = 0, reset_bootfs = 0; 408d5285caeSGeorge Wilson uint64_t objnum = 0; 409ad135b5dSChristopher Siden boolean_t has_feature = B_FALSE; 410990b4856Slling 411990b4856Slling elem = NULL; 412990b4856Slling while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 413990b4856Slling uint64_t intval; 414ad135b5dSChristopher Siden char *strval, *slash, *check, *fname; 415ad135b5dSChristopher Siden const char *propname = nvpair_name(elem); 416ad135b5dSChristopher Siden zpool_prop_t prop = zpool_name_to_prop(propname); 417ad135b5dSChristopher Siden 418ad135b5dSChristopher Siden switch (prop) { 419ad135b5dSChristopher Siden case ZPROP_INVAL: 420ad135b5dSChristopher Siden if (!zpool_prop_feature(propname)) { 421be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 422ad135b5dSChristopher Siden break; 423ad135b5dSChristopher Siden } 424990b4856Slling 425ad135b5dSChristopher Siden /* 426ad135b5dSChristopher Siden * Sanitize the input. 427ad135b5dSChristopher Siden */ 428ad135b5dSChristopher Siden if (nvpair_type(elem) != DATA_TYPE_UINT64) { 429be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 430ad135b5dSChristopher Siden break; 431ad135b5dSChristopher Siden } 432990b4856Slling 433ad135b5dSChristopher Siden if (nvpair_value_uint64(elem, &intval) != 0) { 434be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 435ad135b5dSChristopher Siden break; 436ad135b5dSChristopher Siden } 437ad135b5dSChristopher Siden 438ad135b5dSChristopher Siden if (intval != 0) { 439be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 440ad135b5dSChristopher Siden break; 441ad135b5dSChristopher Siden } 442ad135b5dSChristopher Siden 443ad135b5dSChristopher Siden fname = strchr(propname, '@') + 1; 444ad135b5dSChristopher Siden if (zfeature_lookup_name(fname, NULL) != 0) { 445be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 446ad135b5dSChristopher Siden break; 447ad135b5dSChristopher Siden } 448ad135b5dSChristopher Siden 449ad135b5dSChristopher Siden has_feature = B_TRUE; 450ad135b5dSChristopher Siden break; 451990b4856Slling 452990b4856Slling case ZPOOL_PROP_VERSION: 453990b4856Slling error = nvpair_value_uint64(elem, &intval); 454990b4856Slling if (!error && 455ad135b5dSChristopher Siden (intval < spa_version(spa) || 456ad135b5dSChristopher Siden intval > SPA_VERSION_BEFORE_FEATURES || 457ad135b5dSChristopher Siden has_feature)) 458be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 459990b4856Slling break; 460990b4856Slling 461990b4856Slling case ZPOOL_PROP_DELEGATION: 462990b4856Slling case ZPOOL_PROP_AUTOREPLACE: 463d5b5bb25SRich Morris case ZPOOL_PROP_LISTSNAPS: 464573ca77eSGeorge Wilson case ZPOOL_PROP_AUTOEXPAND: 465990b4856Slling error = nvpair_value_uint64(elem, &intval); 466990b4856Slling if (!error && intval > 1) 467be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 468990b4856Slling break; 469990b4856Slling 470990b4856Slling case ZPOOL_PROP_BOOTFS: 47125f89ee2SJeff Bonwick /* 47225f89ee2SJeff Bonwick * If the pool version is less than SPA_VERSION_BOOTFS, 47325f89ee2SJeff Bonwick * or the pool is still being created (version == 0), 47425f89ee2SJeff Bonwick * the bootfs property cannot be set. 47525f89ee2SJeff Bonwick */ 476990b4856Slling if (spa_version(spa) < SPA_VERSION_BOOTFS) { 477be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 478990b4856Slling break; 479990b4856Slling } 480990b4856Slling 481990b4856Slling /* 48215e6edf1Sgw * Make sure the vdev config is bootable 483990b4856Slling */ 48415e6edf1Sgw if (!vdev_is_bootable(spa->spa_root_vdev)) { 485be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 486990b4856Slling break; 487990b4856Slling } 488990b4856Slling 489990b4856Slling reset_bootfs = 1; 490990b4856Slling 491990b4856Slling error = nvpair_value_string(elem, &strval); 492990b4856Slling 493990b4856Slling if (!error) { 494ad135b5dSChristopher Siden objset_t *os; 495b5152584SMatthew Ahrens uint64_t propval; 49615e6edf1Sgw 497990b4856Slling if (strval == NULL || strval[0] == '\0') { 498990b4856Slling objnum = zpool_prop_default_numeric( 499990b4856Slling ZPOOL_PROP_BOOTFS); 500990b4856Slling break; 501990b4856Slling } 502990b4856Slling 503503ad85cSMatthew Ahrens if (error = dmu_objset_hold(strval, FTAG, &os)) 504990b4856Slling break; 50515e6edf1Sgw 506b5152584SMatthew Ahrens /* 507b5152584SMatthew Ahrens * Must be ZPL, and its property settings 508b5152584SMatthew Ahrens * must be supported by GRUB (compression 509b5152584SMatthew Ahrens * is not gzip, and large blocks are not used). 510b5152584SMatthew Ahrens */ 511503ad85cSMatthew Ahrens 512503ad85cSMatthew Ahrens if (dmu_objset_type(os) != DMU_OST_ZFS) { 513be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 5143b2aab18SMatthew Ahrens } else if ((error = 5153b2aab18SMatthew Ahrens dsl_prop_get_int_ds(dmu_objset_ds(os), 51615e6edf1Sgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 517b5152584SMatthew Ahrens &propval)) == 0 && 518b5152584SMatthew Ahrens !BOOTFS_COMPRESS_VALID(propval)) { 519b5152584SMatthew Ahrens error = SET_ERROR(ENOTSUP); 520b5152584SMatthew Ahrens } else if ((error = 521b5152584SMatthew Ahrens dsl_prop_get_int_ds(dmu_objset_ds(os), 522b5152584SMatthew Ahrens zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 523b5152584SMatthew Ahrens &propval)) == 0 && 524b5152584SMatthew Ahrens propval > SPA_OLD_MAXBLOCKSIZE) { 525be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 52615e6edf1Sgw } else { 52715e6edf1Sgw objnum = dmu_objset_id(os); 52815e6edf1Sgw } 529503ad85cSMatthew Ahrens dmu_objset_rele(os, FTAG); 530990b4856Slling } 531990b4856Slling break; 532e14bb325SJeff Bonwick 5330a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 5340a4e9518Sgw error = nvpair_value_uint64(elem, &intval); 5350a4e9518Sgw if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 5360a4e9518Sgw intval > ZIO_FAILURE_MODE_PANIC)) 537be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 5380a4e9518Sgw 5390a4e9518Sgw /* 5400a4e9518Sgw * This is a special case which only occurs when 5410a4e9518Sgw * the pool has completely failed. This allows 5420a4e9518Sgw * the user to change the in-core failmode property 5430a4e9518Sgw * without syncing it out to disk (I/Os might 5440a4e9518Sgw * currently be blocked). We do this by returning 5450a4e9518Sgw * EIO to the caller (spa_prop_set) to trick it 5460a4e9518Sgw * into thinking we encountered a property validation 5470a4e9518Sgw * error. 5480a4e9518Sgw */ 549e14bb325SJeff Bonwick if (!error && spa_suspended(spa)) { 5500a4e9518Sgw spa->spa_failmode = intval; 551be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 5520a4e9518Sgw } 5530a4e9518Sgw break; 5542f8aaab3Seschrock 5552f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 5562f8aaab3Seschrock if ((error = nvpair_value_string(elem, &strval)) != 0) 5572f8aaab3Seschrock break; 5582f8aaab3Seschrock 5592f8aaab3Seschrock if (strval[0] == '\0') 5602f8aaab3Seschrock break; 5612f8aaab3Seschrock 5622f8aaab3Seschrock if (strcmp(strval, "none") == 0) 5632f8aaab3Seschrock break; 5642f8aaab3Seschrock 5652f8aaab3Seschrock if (strval[0] != '/') { 566be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 5672f8aaab3Seschrock break; 5682f8aaab3Seschrock } 5692f8aaab3Seschrock 5702f8aaab3Seschrock slash = strrchr(strval, '/'); 5712f8aaab3Seschrock ASSERT(slash != NULL); 5722f8aaab3Seschrock 5732f8aaab3Seschrock if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 5742f8aaab3Seschrock strcmp(slash, "/..") == 0) 575be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 5762f8aaab3Seschrock break; 577b24ab676SJeff Bonwick 5788704186eSDan McDonald case ZPOOL_PROP_COMMENT: 5798704186eSDan McDonald if ((error = nvpair_value_string(elem, &strval)) != 0) 5808704186eSDan McDonald break; 5818704186eSDan McDonald for (check = strval; *check != '\0'; check++) { 5828704186eSDan McDonald /* 5838704186eSDan McDonald * The kernel doesn't have an easy isprint() 5848704186eSDan McDonald * check. For this kernel check, we merely 5858704186eSDan McDonald * check ASCII apart from DEL. Fix this if 5868704186eSDan McDonald * there is an easy-to-use kernel isprint(). 5878704186eSDan McDonald */ 5888704186eSDan McDonald if (*check >= 0x7f) { 589be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 5908704186eSDan McDonald break; 5918704186eSDan McDonald } 5928704186eSDan McDonald check++; 5938704186eSDan McDonald } 5948704186eSDan McDonald if (strlen(strval) > ZPROP_MAX_COMMENT) 5958704186eSDan McDonald error = E2BIG; 5968704186eSDan McDonald break; 5978704186eSDan McDonald 598b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 599b24ab676SJeff Bonwick if (spa_version(spa) < SPA_VERSION_DEDUP) 600be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 601b24ab676SJeff Bonwick else 602b24ab676SJeff Bonwick error = nvpair_value_uint64(elem, &intval); 603b24ab676SJeff Bonwick if (error == 0 && 604b24ab676SJeff Bonwick intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 605be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 606b24ab676SJeff Bonwick break; 607990b4856Slling } 608990b4856Slling 609990b4856Slling if (error) 610990b4856Slling break; 611990b4856Slling } 612990b4856Slling 613990b4856Slling if (!error && reset_bootfs) { 614990b4856Slling error = nvlist_remove(props, 615990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 616990b4856Slling 617990b4856Slling if (!error) { 618990b4856Slling error = nvlist_add_uint64(props, 619990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 620990b4856Slling } 621990b4856Slling } 622990b4856Slling 623990b4856Slling return (error); 624990b4856Slling } 625990b4856Slling 626379c004dSEric Schrock void 627379c004dSEric Schrock spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 628379c004dSEric Schrock { 629379c004dSEric Schrock char *cachefile; 630379c004dSEric Schrock spa_config_dirent_t *dp; 631379c004dSEric Schrock 632379c004dSEric Schrock if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 633379c004dSEric Schrock &cachefile) != 0) 634379c004dSEric Schrock return; 635379c004dSEric Schrock 636379c004dSEric Schrock dp = kmem_alloc(sizeof (spa_config_dirent_t), 637379c004dSEric Schrock KM_SLEEP); 638379c004dSEric Schrock 639379c004dSEric Schrock if (cachefile[0] == '\0') 640379c004dSEric Schrock dp->scd_path = spa_strdup(spa_config_path); 641379c004dSEric Schrock else if (strcmp(cachefile, "none") == 0) 642379c004dSEric Schrock dp->scd_path = NULL; 643379c004dSEric Schrock else 644379c004dSEric Schrock dp->scd_path = spa_strdup(cachefile); 645379c004dSEric Schrock 646379c004dSEric Schrock list_insert_head(&spa->spa_config_list, dp); 647379c004dSEric Schrock if (need_sync) 648379c004dSEric Schrock spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 649379c004dSEric Schrock } 650379c004dSEric Schrock 651990b4856Slling int 652990b4856Slling spa_prop_set(spa_t *spa, nvlist_t *nvp) 653990b4856Slling { 654990b4856Slling int error; 655ad135b5dSChristopher Siden nvpair_t *elem = NULL; 656379c004dSEric Schrock boolean_t need_sync = B_FALSE; 657990b4856Slling 658990b4856Slling if ((error = spa_prop_validate(spa, nvp)) != 0) 659990b4856Slling return (error); 660990b4856Slling 661379c004dSEric Schrock while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 662ad135b5dSChristopher Siden zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 663379c004dSEric Schrock 664f9af39baSGeorge Wilson if (prop == ZPOOL_PROP_CACHEFILE || 665f9af39baSGeorge Wilson prop == ZPOOL_PROP_ALTROOT || 666f9af39baSGeorge Wilson prop == ZPOOL_PROP_READONLY) 667379c004dSEric Schrock continue; 668379c004dSEric Schrock 669ad135b5dSChristopher Siden if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 670ad135b5dSChristopher Siden uint64_t ver; 671ad135b5dSChristopher Siden 672ad135b5dSChristopher Siden if (prop == ZPOOL_PROP_VERSION) { 673ad135b5dSChristopher Siden VERIFY(nvpair_value_uint64(elem, &ver) == 0); 674ad135b5dSChristopher Siden } else { 675ad135b5dSChristopher Siden ASSERT(zpool_prop_feature(nvpair_name(elem))); 676ad135b5dSChristopher Siden ver = SPA_VERSION_FEATURES; 677ad135b5dSChristopher Siden need_sync = B_TRUE; 678ad135b5dSChristopher Siden } 679ad135b5dSChristopher Siden 680ad135b5dSChristopher Siden /* Save time if the version is already set. */ 681ad135b5dSChristopher Siden if (ver == spa_version(spa)) 682ad135b5dSChristopher Siden continue; 683ad135b5dSChristopher Siden 684ad135b5dSChristopher Siden /* 685ad135b5dSChristopher Siden * In addition to the pool directory object, we might 686ad135b5dSChristopher Siden * create the pool properties object, the features for 687ad135b5dSChristopher Siden * read object, the features for write object, or the 688ad135b5dSChristopher Siden * feature descriptions object. 689ad135b5dSChristopher Siden */ 6903b2aab18SMatthew Ahrens error = dsl_sync_task(spa->spa_name, NULL, 6917d46dc6cSMatthew Ahrens spa_sync_version, &ver, 6927d46dc6cSMatthew Ahrens 6, ZFS_SPACE_CHECK_RESERVED); 693ad135b5dSChristopher Siden if (error) 694ad135b5dSChristopher Siden return (error); 695ad135b5dSChristopher Siden continue; 696ad135b5dSChristopher Siden } 697ad135b5dSChristopher Siden 698379c004dSEric Schrock need_sync = B_TRUE; 699379c004dSEric Schrock break; 700379c004dSEric Schrock } 701379c004dSEric Schrock 702ad135b5dSChristopher Siden if (need_sync) { 7033b2aab18SMatthew Ahrens return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 7047d46dc6cSMatthew Ahrens nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 705ad135b5dSChristopher Siden } 706ad135b5dSChristopher Siden 707ad135b5dSChristopher Siden return (0); 708990b4856Slling } 709990b4856Slling 710990b4856Slling /* 711990b4856Slling * If the bootfs property value is dsobj, clear it. 712990b4856Slling */ 713990b4856Slling void 714990b4856Slling spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 715990b4856Slling { 716990b4856Slling if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 717990b4856Slling VERIFY(zap_remove(spa->spa_meta_objset, 718990b4856Slling spa->spa_pool_props_object, 719990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 720990b4856Slling spa->spa_bootfs = 0; 721990b4856Slling } 722990b4856Slling } 723990b4856Slling 724dfbb9432SGeorge Wilson /*ARGSUSED*/ 725dfbb9432SGeorge Wilson static int 7263b2aab18SMatthew Ahrens spa_change_guid_check(void *arg, dmu_tx_t *tx) 727dfbb9432SGeorge Wilson { 7283b2aab18SMatthew Ahrens uint64_t *newguid = arg; 7293b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 730dfbb9432SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 731dfbb9432SGeorge Wilson uint64_t vdev_state; 732dfbb9432SGeorge Wilson 733dfbb9432SGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 734dfbb9432SGeorge Wilson vdev_state = rvd->vdev_state; 735dfbb9432SGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 736dfbb9432SGeorge Wilson 737dfbb9432SGeorge Wilson if (vdev_state != VDEV_STATE_HEALTHY) 738be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 739dfbb9432SGeorge Wilson 740dfbb9432SGeorge Wilson ASSERT3U(spa_guid(spa), !=, *newguid); 741dfbb9432SGeorge Wilson 742dfbb9432SGeorge Wilson return (0); 743dfbb9432SGeorge Wilson } 744dfbb9432SGeorge Wilson 745dfbb9432SGeorge Wilson static void 7463b2aab18SMatthew Ahrens spa_change_guid_sync(void *arg, dmu_tx_t *tx) 747dfbb9432SGeorge Wilson { 7483b2aab18SMatthew Ahrens uint64_t *newguid = arg; 7493b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 750dfbb9432SGeorge Wilson uint64_t oldguid; 751dfbb9432SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 752dfbb9432SGeorge Wilson 753dfbb9432SGeorge Wilson oldguid = spa_guid(spa); 754dfbb9432SGeorge Wilson 755dfbb9432SGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 756dfbb9432SGeorge Wilson rvd->vdev_guid = *newguid; 757dfbb9432SGeorge Wilson rvd->vdev_guid_sum += (*newguid - oldguid); 758dfbb9432SGeorge Wilson vdev_config_dirty(rvd); 759dfbb9432SGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 760dfbb9432SGeorge Wilson 76120128a08SGeorge Wilson spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 762dfbb9432SGeorge Wilson oldguid, *newguid); 763dfbb9432SGeorge Wilson } 764dfbb9432SGeorge Wilson 765e9103aaeSGarrett D'Amore /* 766e9103aaeSGarrett D'Amore * Change the GUID for the pool. This is done so that we can later 767e9103aaeSGarrett D'Amore * re-import a pool built from a clone of our own vdevs. We will modify 768e9103aaeSGarrett D'Amore * the root vdev's guid, our own pool guid, and then mark all of our 769e9103aaeSGarrett D'Amore * vdevs dirty. Note that we must make sure that all our vdevs are 770e9103aaeSGarrett D'Amore * online when we do this, or else any vdevs that weren't present 771e9103aaeSGarrett D'Amore * would be orphaned from our pool. We are also going to issue a 772e9103aaeSGarrett D'Amore * sysevent to update any watchers. 773e9103aaeSGarrett D'Amore */ 774e9103aaeSGarrett D'Amore int 775e9103aaeSGarrett D'Amore spa_change_guid(spa_t *spa) 776e9103aaeSGarrett D'Amore { 777dfbb9432SGeorge Wilson int error; 778dfbb9432SGeorge Wilson uint64_t guid; 779e9103aaeSGarrett D'Amore 7802c1e2b44SGeorge Wilson mutex_enter(&spa->spa_vdev_top_lock); 781dfbb9432SGeorge Wilson mutex_enter(&spa_namespace_lock); 782dfbb9432SGeorge Wilson guid = spa_generate_guid(NULL); 783e9103aaeSGarrett D'Amore 7843b2aab18SMatthew Ahrens error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 7857d46dc6cSMatthew Ahrens spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 786e9103aaeSGarrett D'Amore 787dfbb9432SGeorge Wilson if (error == 0) { 788dfbb9432SGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 789dfbb9432SGeorge Wilson spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 790dfbb9432SGeorge Wilson } 791e9103aaeSGarrett D'Amore 792dfbb9432SGeorge Wilson mutex_exit(&spa_namespace_lock); 7932c1e2b44SGeorge Wilson mutex_exit(&spa->spa_vdev_top_lock); 794e9103aaeSGarrett D'Amore 795dfbb9432SGeorge Wilson return (error); 796e9103aaeSGarrett D'Amore } 797e9103aaeSGarrett D'Amore 798fa9e4066Sahrens /* 799fa9e4066Sahrens * ========================================================================== 800fa9e4066Sahrens * SPA state manipulation (open/create/destroy/import/export) 801fa9e4066Sahrens * ========================================================================== 802fa9e4066Sahrens */ 803fa9e4066Sahrens 804ea8dc4b6Seschrock static int 805ea8dc4b6Seschrock spa_error_entry_compare(const void *a, const void *b) 806ea8dc4b6Seschrock { 807ea8dc4b6Seschrock spa_error_entry_t *sa = (spa_error_entry_t *)a; 808ea8dc4b6Seschrock spa_error_entry_t *sb = (spa_error_entry_t *)b; 809ea8dc4b6Seschrock int ret; 810ea8dc4b6Seschrock 811ea8dc4b6Seschrock ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 8127802d7bfSMatthew Ahrens sizeof (zbookmark_phys_t)); 813ea8dc4b6Seschrock 814ea8dc4b6Seschrock if (ret < 0) 815ea8dc4b6Seschrock return (-1); 816ea8dc4b6Seschrock else if (ret > 0) 817ea8dc4b6Seschrock return (1); 818ea8dc4b6Seschrock else 819ea8dc4b6Seschrock return (0); 820ea8dc4b6Seschrock } 821ea8dc4b6Seschrock 822ea8dc4b6Seschrock /* 823ea8dc4b6Seschrock * Utility function which retrieves copies of the current logs and 824ea8dc4b6Seschrock * re-initializes them in the process. 825ea8dc4b6Seschrock */ 826ea8dc4b6Seschrock void 827ea8dc4b6Seschrock spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 828ea8dc4b6Seschrock { 829ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 830ea8dc4b6Seschrock 831ea8dc4b6Seschrock bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 832ea8dc4b6Seschrock bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 833ea8dc4b6Seschrock 834ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 835ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 836ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 837ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 838ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 839ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 840ea8dc4b6Seschrock } 841ea8dc4b6Seschrock 842ec94d322SAdam Leventhal static void 843ec94d322SAdam Leventhal spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 844fa9e4066Sahrens { 845ec94d322SAdam Leventhal const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 846ec94d322SAdam Leventhal enum zti_modes mode = ztip->zti_mode; 847ec94d322SAdam Leventhal uint_t value = ztip->zti_value; 848ec94d322SAdam Leventhal uint_t count = ztip->zti_count; 849ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 850ec94d322SAdam Leventhal char name[32]; 8515aeb9474SGarrett D'Amore uint_t flags = 0; 85235a5a358SJonathan Adams boolean_t batch = B_FALSE; 853fa9e4066Sahrens 854ec94d322SAdam Leventhal if (mode == ZTI_MODE_NULL) { 855ec94d322SAdam Leventhal tqs->stqs_count = 0; 856ec94d322SAdam Leventhal tqs->stqs_taskq = NULL; 857ec94d322SAdam Leventhal return; 858ec94d322SAdam Leventhal } 859fa9e4066Sahrens 860ec94d322SAdam Leventhal ASSERT3U(count, >, 0); 861fa9e4066Sahrens 862ec94d322SAdam Leventhal tqs->stqs_count = count; 863ec94d322SAdam Leventhal tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 86435a5a358SJonathan Adams 86569962b56SMatthew Ahrens switch (mode) { 86669962b56SMatthew Ahrens case ZTI_MODE_FIXED: 86769962b56SMatthew Ahrens ASSERT3U(value, >=, 1); 86869962b56SMatthew Ahrens value = MAX(value, 1); 86969962b56SMatthew Ahrens break; 870ec94d322SAdam Leventhal 87169962b56SMatthew Ahrens case ZTI_MODE_BATCH: 87269962b56SMatthew Ahrens batch = B_TRUE; 87369962b56SMatthew Ahrens flags |= TASKQ_THREADS_CPU_PCT; 87469962b56SMatthew Ahrens value = zio_taskq_batch_pct; 87569962b56SMatthew Ahrens break; 876ec94d322SAdam Leventhal 87769962b56SMatthew Ahrens default: 87869962b56SMatthew Ahrens panic("unrecognized mode for %s_%s taskq (%u:%u) in " 87969962b56SMatthew Ahrens "spa_activate()", 88069962b56SMatthew Ahrens zio_type_name[t], zio_taskq_types[q], mode, value); 88169962b56SMatthew Ahrens break; 88269962b56SMatthew Ahrens } 883ec94d322SAdam Leventhal 88469962b56SMatthew Ahrens for (uint_t i = 0; i < count; i++) { 88569962b56SMatthew Ahrens taskq_t *tq; 886ec94d322SAdam Leventhal 887ec94d322SAdam Leventhal if (count > 1) { 888ec94d322SAdam Leventhal (void) snprintf(name, sizeof (name), "%s_%s_%u", 889ec94d322SAdam Leventhal zio_type_name[t], zio_taskq_types[q], i); 890ec94d322SAdam Leventhal } else { 891ec94d322SAdam Leventhal (void) snprintf(name, sizeof (name), "%s_%s", 892ec94d322SAdam Leventhal zio_type_name[t], zio_taskq_types[q]); 893ec94d322SAdam Leventhal } 894ec94d322SAdam Leventhal 895ec94d322SAdam Leventhal if (zio_taskq_sysdc && spa->spa_proc != &p0) { 896ec94d322SAdam Leventhal if (batch) 897ec94d322SAdam Leventhal flags |= TASKQ_DC_BATCH; 898ec94d322SAdam Leventhal 899ec94d322SAdam Leventhal tq = taskq_create_sysdc(name, value, 50, INT_MAX, 900ec94d322SAdam Leventhal spa->spa_proc, zio_taskq_basedc, flags); 901ec94d322SAdam Leventhal } else { 90269962b56SMatthew Ahrens pri_t pri = maxclsyspri; 90369962b56SMatthew Ahrens /* 90469962b56SMatthew Ahrens * The write issue taskq can be extremely CPU 90569962b56SMatthew Ahrens * intensive. Run it at slightly lower priority 90669962b56SMatthew Ahrens * than the other taskqs. 90769962b56SMatthew Ahrens */ 90869962b56SMatthew Ahrens if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 90969962b56SMatthew Ahrens pri--; 91069962b56SMatthew Ahrens 91169962b56SMatthew Ahrens tq = taskq_create_proc(name, value, pri, 50, 912ec94d322SAdam Leventhal INT_MAX, spa->spa_proc, flags); 913ec94d322SAdam Leventhal } 914ec94d322SAdam Leventhal 915ec94d322SAdam Leventhal tqs->stqs_taskq[i] = tq; 916ec94d322SAdam Leventhal } 917ec94d322SAdam Leventhal } 918ec94d322SAdam Leventhal 919ec94d322SAdam Leventhal static void 920ec94d322SAdam Leventhal spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 921ec94d322SAdam Leventhal { 922ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 923ec94d322SAdam Leventhal 924ec94d322SAdam Leventhal if (tqs->stqs_taskq == NULL) { 925ec94d322SAdam Leventhal ASSERT0(tqs->stqs_count); 926ec94d322SAdam Leventhal return; 927ec94d322SAdam Leventhal } 928ec94d322SAdam Leventhal 929ec94d322SAdam Leventhal for (uint_t i = 0; i < tqs->stqs_count; i++) { 930ec94d322SAdam Leventhal ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 931ec94d322SAdam Leventhal taskq_destroy(tqs->stqs_taskq[i]); 93235a5a358SJonathan Adams } 93335a5a358SJonathan Adams 934ec94d322SAdam Leventhal kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 935ec94d322SAdam Leventhal tqs->stqs_taskq = NULL; 936ec94d322SAdam Leventhal } 93735a5a358SJonathan Adams 938ec94d322SAdam Leventhal /* 939ec94d322SAdam Leventhal * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 940ec94d322SAdam Leventhal * Note that a type may have multiple discrete taskqs to avoid lock contention 941ec94d322SAdam Leventhal * on the taskq itself. In that case we choose which taskq at random by using 942ec94d322SAdam Leventhal * the low bits of gethrtime(). 943ec94d322SAdam Leventhal */ 944ec94d322SAdam Leventhal void 945ec94d322SAdam Leventhal spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 946ec94d322SAdam Leventhal task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 947ec94d322SAdam Leventhal { 948ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 949ec94d322SAdam Leventhal taskq_t *tq; 950ec94d322SAdam Leventhal 951ec94d322SAdam Leventhal ASSERT3P(tqs->stqs_taskq, !=, NULL); 952ec94d322SAdam Leventhal ASSERT3U(tqs->stqs_count, !=, 0); 953ec94d322SAdam Leventhal 954ec94d322SAdam Leventhal if (tqs->stqs_count == 1) { 955ec94d322SAdam Leventhal tq = tqs->stqs_taskq[0]; 956ec94d322SAdam Leventhal } else { 957ec94d322SAdam Leventhal tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 95835a5a358SJonathan Adams } 959ec94d322SAdam Leventhal 960ec94d322SAdam Leventhal taskq_dispatch_ent(tq, func, arg, flags, ent); 96135a5a358SJonathan Adams } 96235a5a358SJonathan Adams 96335a5a358SJonathan Adams static void 96435a5a358SJonathan Adams spa_create_zio_taskqs(spa_t *spa) 96535a5a358SJonathan Adams { 966e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 967e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 968ec94d322SAdam Leventhal spa_taskqs_init(spa, t, q); 96935a5a358SJonathan Adams } 97035a5a358SJonathan Adams } 97135a5a358SJonathan Adams } 97235a5a358SJonathan Adams 97335a5a358SJonathan Adams #ifdef _KERNEL 97435a5a358SJonathan Adams static void 97535a5a358SJonathan Adams spa_thread(void *arg) 97635a5a358SJonathan Adams { 97735a5a358SJonathan Adams callb_cpr_t cprinfo; 9782e0c549eSJonathan Adams 97935a5a358SJonathan Adams spa_t *spa = arg; 98035a5a358SJonathan Adams user_t *pu = PTOU(curproc); 9812e0c549eSJonathan Adams 98235a5a358SJonathan Adams CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 98335a5a358SJonathan Adams spa->spa_name); 9842e0c549eSJonathan Adams 98535a5a358SJonathan Adams ASSERT(curproc != &p0); 98635a5a358SJonathan Adams (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 98735a5a358SJonathan Adams "zpool-%s", spa->spa_name); 98835a5a358SJonathan Adams (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 9892e0c549eSJonathan Adams 99035a5a358SJonathan Adams /* bind this thread to the requested psrset */ 99135a5a358SJonathan Adams if (zio_taskq_psrset_bind != PS_NONE) { 99235a5a358SJonathan Adams pool_lock(); 99335a5a358SJonathan Adams mutex_enter(&cpu_lock); 99435a5a358SJonathan Adams mutex_enter(&pidlock); 99535a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 99680eb36f2SGeorge Wilson 99735a5a358SJonathan Adams if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 99835a5a358SJonathan Adams 0, NULL, NULL) == 0) { 99935a5a358SJonathan Adams curthread->t_bind_pset = zio_taskq_psrset_bind; 100035a5a358SJonathan Adams } else { 100135a5a358SJonathan Adams cmn_err(CE_WARN, 100235a5a358SJonathan Adams "Couldn't bind process for zfs pool \"%s\" to " 100335a5a358SJonathan Adams "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 100435a5a358SJonathan Adams } 100535a5a358SJonathan Adams 100635a5a358SJonathan Adams mutex_exit(&curproc->p_lock); 100735a5a358SJonathan Adams mutex_exit(&pidlock); 100835a5a358SJonathan Adams mutex_exit(&cpu_lock); 100935a5a358SJonathan Adams pool_unlock(); 101035a5a358SJonathan Adams } 101135a5a358SJonathan Adams 101235a5a358SJonathan Adams if (zio_taskq_sysdc) { 101335a5a358SJonathan Adams sysdc_thread_enter(curthread, 100, 0); 101435a5a358SJonathan Adams } 101535a5a358SJonathan Adams 101635a5a358SJonathan Adams spa->spa_proc = curproc; 101735a5a358SJonathan Adams spa->spa_did = curthread->t_did; 101835a5a358SJonathan Adams 101935a5a358SJonathan Adams spa_create_zio_taskqs(spa); 102035a5a358SJonathan Adams 102135a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 102235a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 102335a5a358SJonathan Adams 102435a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_ACTIVE; 102535a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 102635a5a358SJonathan Adams 102735a5a358SJonathan Adams CALLB_CPR_SAFE_BEGIN(&cprinfo); 102835a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_ACTIVE) 102935a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 103035a5a358SJonathan Adams CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 103135a5a358SJonathan Adams 103235a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 103335a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_GONE; 103435a5a358SJonathan Adams spa->spa_proc = &p0; 103535a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 103635a5a358SJonathan Adams CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 103735a5a358SJonathan Adams 103835a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 103935a5a358SJonathan Adams lwp_exit(); 104035a5a358SJonathan Adams } 104135a5a358SJonathan Adams #endif 104235a5a358SJonathan Adams 104335a5a358SJonathan Adams /* 104435a5a358SJonathan Adams * Activate an uninitialized pool. 104535a5a358SJonathan Adams */ 104635a5a358SJonathan Adams static void 104735a5a358SJonathan Adams spa_activate(spa_t *spa, int mode) 104835a5a358SJonathan Adams { 104935a5a358SJonathan Adams ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 105035a5a358SJonathan Adams 105135a5a358SJonathan Adams spa->spa_state = POOL_STATE_ACTIVE; 105235a5a358SJonathan Adams spa->spa_mode = mode; 105335a5a358SJonathan Adams 105435a5a358SJonathan Adams spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 105535a5a358SJonathan Adams spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 105635a5a358SJonathan Adams 105735a5a358SJonathan Adams /* Try to create a covering process */ 105835a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 105935a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 106035a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 106135a5a358SJonathan Adams spa->spa_did = 0; 106235a5a358SJonathan Adams 106335a5a358SJonathan Adams /* Only create a process if we're going to be around a while. */ 106435a5a358SJonathan Adams if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 106535a5a358SJonathan Adams if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 106635a5a358SJonathan Adams NULL, 0) == 0) { 106735a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_CREATED; 106835a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_CREATED) { 106935a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, 107035a5a358SJonathan Adams &spa->spa_proc_lock); 10712e0c549eSJonathan Adams } 107235a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 107335a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 107435a5a358SJonathan Adams ASSERT(spa->spa_did != 0); 107535a5a358SJonathan Adams } else { 107635a5a358SJonathan Adams #ifdef _KERNEL 107735a5a358SJonathan Adams cmn_err(CE_WARN, 107835a5a358SJonathan Adams "Couldn't create process for zfs pool \"%s\"\n", 107935a5a358SJonathan Adams spa->spa_name); 108035a5a358SJonathan Adams #endif 1081e14bb325SJeff Bonwick } 1082fa9e4066Sahrens } 108335a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 108435a5a358SJonathan Adams 108535a5a358SJonathan Adams /* If we didn't create a process, we need to create our taskqs. */ 108635a5a358SJonathan Adams if (spa->spa_proc == &p0) { 108735a5a358SJonathan Adams spa_create_zio_taskqs(spa); 108835a5a358SJonathan Adams } 1089fa9e4066Sahrens 1090e14bb325SJeff Bonwick list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1091e14bb325SJeff Bonwick offsetof(vdev_t, vdev_config_dirty_node)); 1092bc9014e6SJustin Gibbs list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1093bc9014e6SJustin Gibbs offsetof(objset_t, os_evicting_node)); 1094e14bb325SJeff Bonwick list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1095e14bb325SJeff Bonwick offsetof(vdev_t, vdev_state_dirty_node)); 1096fa9e4066Sahrens 1097fa9e4066Sahrens txg_list_create(&spa->spa_vdev_txg_list, 1098fa9e4066Sahrens offsetof(struct vdev, vdev_txg_node)); 1099ea8dc4b6Seschrock 1100ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 1101ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 1102ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 1103ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 1104ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 1105ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 1106fa9e4066Sahrens } 1107fa9e4066Sahrens 1108fa9e4066Sahrens /* 1109fa9e4066Sahrens * Opposite of spa_activate(). 1110fa9e4066Sahrens */ 1111fa9e4066Sahrens static void 1112fa9e4066Sahrens spa_deactivate(spa_t *spa) 1113fa9e4066Sahrens { 1114fa9e4066Sahrens ASSERT(spa->spa_sync_on == B_FALSE); 1115fa9e4066Sahrens ASSERT(spa->spa_dsl_pool == NULL); 1116fa9e4066Sahrens ASSERT(spa->spa_root_vdev == NULL); 111725f89ee2SJeff Bonwick ASSERT(spa->spa_async_zio_root == NULL); 1118fa9e4066Sahrens ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1119fa9e4066Sahrens 1120bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 1121bc9014e6SJustin Gibbs 1122fa9e4066Sahrens txg_list_destroy(&spa->spa_vdev_txg_list); 1123fa9e4066Sahrens 1124e14bb325SJeff Bonwick list_destroy(&spa->spa_config_dirty_list); 1125bc9014e6SJustin Gibbs list_destroy(&spa->spa_evicting_os_list); 1126e14bb325SJeff Bonwick list_destroy(&spa->spa_state_dirty_list); 1127fa9e4066Sahrens 1128e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 1129e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1130ec94d322SAdam Leventhal spa_taskqs_fini(spa, t, q); 1131e14bb325SJeff Bonwick } 1132fa9e4066Sahrens } 1133fa9e4066Sahrens 1134fa9e4066Sahrens metaslab_class_destroy(spa->spa_normal_class); 1135fa9e4066Sahrens spa->spa_normal_class = NULL; 1136fa9e4066Sahrens 11378654d025Sperrin metaslab_class_destroy(spa->spa_log_class); 11388654d025Sperrin spa->spa_log_class = NULL; 11398654d025Sperrin 1140ea8dc4b6Seschrock /* 1141ea8dc4b6Seschrock * If this was part of an import or the open otherwise failed, we may 1142ea8dc4b6Seschrock * still have errors left in the queues. Empty them just in case. 1143ea8dc4b6Seschrock */ 1144ea8dc4b6Seschrock spa_errlog_drain(spa); 1145ea8dc4b6Seschrock 1146ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_scrub); 1147ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_last); 1148ea8dc4b6Seschrock 1149fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 115035a5a358SJonathan Adams 115135a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 115235a5a358SJonathan Adams if (spa->spa_proc_state != SPA_PROC_NONE) { 115335a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 115435a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_DEACTIVATE; 115535a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 115635a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 115735a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 115835a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 115935a5a358SJonathan Adams } 116035a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 116135a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 116235a5a358SJonathan Adams } 116335a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 116435a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 116535a5a358SJonathan Adams 116635a5a358SJonathan Adams /* 116735a5a358SJonathan Adams * We want to make sure spa_thread() has actually exited the ZFS 116835a5a358SJonathan Adams * module, so that the module can't be unloaded out from underneath 116935a5a358SJonathan Adams * it. 117035a5a358SJonathan Adams */ 117135a5a358SJonathan Adams if (spa->spa_did != 0) { 117235a5a358SJonathan Adams thread_join(spa->spa_did); 117335a5a358SJonathan Adams spa->spa_did = 0; 117435a5a358SJonathan Adams } 1175fa9e4066Sahrens } 1176fa9e4066Sahrens 1177fa9e4066Sahrens /* 1178fa9e4066Sahrens * Verify a pool configuration, and construct the vdev tree appropriately. This 1179fa9e4066Sahrens * will create all the necessary vdevs in the appropriate layout, with each vdev 1180fa9e4066Sahrens * in the CLOSED state. This will prep the pool before open/creation/import. 1181fa9e4066Sahrens * All vdev validation is done by the vdev_alloc() routine. 1182fa9e4066Sahrens */ 118399653d4eSeschrock static int 118499653d4eSeschrock spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 118599653d4eSeschrock uint_t id, int atype) 1186fa9e4066Sahrens { 1187fa9e4066Sahrens nvlist_t **child; 1188573ca77eSGeorge Wilson uint_t children; 118999653d4eSeschrock int error; 1190fa9e4066Sahrens 119199653d4eSeschrock if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 119299653d4eSeschrock return (error); 1193fa9e4066Sahrens 119499653d4eSeschrock if ((*vdp)->vdev_ops->vdev_op_leaf) 119599653d4eSeschrock return (0); 1196fa9e4066Sahrens 1197e14bb325SJeff Bonwick error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1198e14bb325SJeff Bonwick &child, &children); 1199e14bb325SJeff Bonwick 1200e14bb325SJeff Bonwick if (error == ENOENT) 1201e14bb325SJeff Bonwick return (0); 1202e14bb325SJeff Bonwick 1203e14bb325SJeff Bonwick if (error) { 120499653d4eSeschrock vdev_free(*vdp); 120599653d4eSeschrock *vdp = NULL; 1206be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1207fa9e4066Sahrens } 1208fa9e4066Sahrens 1209573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 121099653d4eSeschrock vdev_t *vd; 121199653d4eSeschrock if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 121299653d4eSeschrock atype)) != 0) { 121399653d4eSeschrock vdev_free(*vdp); 121499653d4eSeschrock *vdp = NULL; 121599653d4eSeschrock return (error); 1216fa9e4066Sahrens } 1217fa9e4066Sahrens } 1218fa9e4066Sahrens 121999653d4eSeschrock ASSERT(*vdp != NULL); 122099653d4eSeschrock 122199653d4eSeschrock return (0); 1222fa9e4066Sahrens } 1223fa9e4066Sahrens 1224fa9e4066Sahrens /* 1225fa9e4066Sahrens * Opposite of spa_load(). 1226fa9e4066Sahrens */ 1227fa9e4066Sahrens static void 1228fa9e4066Sahrens spa_unload(spa_t *spa) 1229fa9e4066Sahrens { 123099653d4eSeschrock int i; 123199653d4eSeschrock 1232e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1233e14bb325SJeff Bonwick 1234ea8dc4b6Seschrock /* 1235ea8dc4b6Seschrock * Stop async tasks. 1236ea8dc4b6Seschrock */ 1237ea8dc4b6Seschrock spa_async_suspend(spa); 1238ea8dc4b6Seschrock 1239fa9e4066Sahrens /* 1240fa9e4066Sahrens * Stop syncing. 1241fa9e4066Sahrens */ 1242fa9e4066Sahrens if (spa->spa_sync_on) { 1243fa9e4066Sahrens txg_sync_stop(spa->spa_dsl_pool); 1244fa9e4066Sahrens spa->spa_sync_on = B_FALSE; 1245fa9e4066Sahrens } 1246fa9e4066Sahrens 1247fa9e4066Sahrens /* 1248e14bb325SJeff Bonwick * Wait for any outstanding async I/O to complete. 1249fa9e4066Sahrens */ 125054d692b7SGeorge Wilson if (spa->spa_async_zio_root != NULL) { 12516f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) 12526f834bc1SMatthew Ahrens (void) zio_wait(spa->spa_async_zio_root[i]); 12536f834bc1SMatthew Ahrens kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 125454d692b7SGeorge Wilson spa->spa_async_zio_root = NULL; 125554d692b7SGeorge Wilson } 1256fa9e4066Sahrens 1257cde58dbcSMatthew Ahrens bpobj_close(&spa->spa_deferred_bpobj); 1258cde58dbcSMatthew Ahrens 12590713e232SGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 12600713e232SGeorge Wilson 12610713e232SGeorge Wilson /* 12620713e232SGeorge Wilson * Close all vdevs. 12630713e232SGeorge Wilson */ 12640713e232SGeorge Wilson if (spa->spa_root_vdev) 12650713e232SGeorge Wilson vdev_free(spa->spa_root_vdev); 12660713e232SGeorge Wilson ASSERT(spa->spa_root_vdev == NULL); 12670713e232SGeorge Wilson 1268fa9e4066Sahrens /* 1269fa9e4066Sahrens * Close the dsl pool. 1270fa9e4066Sahrens */ 1271fa9e4066Sahrens if (spa->spa_dsl_pool) { 1272fa9e4066Sahrens dsl_pool_close(spa->spa_dsl_pool); 1273fa9e4066Sahrens spa->spa_dsl_pool = NULL; 1274afee20e4SGeorge Wilson spa->spa_meta_objset = NULL; 1275fa9e4066Sahrens } 1276fa9e4066Sahrens 1277b24ab676SJeff Bonwick ddt_unload(spa); 1278b24ab676SJeff Bonwick 12798ad4d6ddSJeff Bonwick 12808ad4d6ddSJeff Bonwick /* 12818ad4d6ddSJeff Bonwick * Drop and purge level 2 cache 12828ad4d6ddSJeff Bonwick */ 12838ad4d6ddSJeff Bonwick spa_l2cache_drop(spa); 12848ad4d6ddSJeff Bonwick 1285fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 1286fa94a07fSbrendan vdev_free(spa->spa_spares.sav_vdevs[i]); 1287fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) { 1288fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 1289fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 1290fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 129199653d4eSeschrock } 1292fa94a07fSbrendan if (spa->spa_spares.sav_config) { 1293fa94a07fSbrendan nvlist_free(spa->spa_spares.sav_config); 1294fa94a07fSbrendan spa->spa_spares.sav_config = NULL; 1295fa94a07fSbrendan } 12962ce8af81SEric Schrock spa->spa_spares.sav_count = 0; 1297fa94a07fSbrendan 1298cd0837ccSGeorge Wilson for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1299cd0837ccSGeorge Wilson vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1300fa94a07fSbrendan vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1301cd0837ccSGeorge Wilson } 1302fa94a07fSbrendan if (spa->spa_l2cache.sav_vdevs) { 1303fa94a07fSbrendan kmem_free(spa->spa_l2cache.sav_vdevs, 1304fa94a07fSbrendan spa->spa_l2cache.sav_count * sizeof (void *)); 1305fa94a07fSbrendan spa->spa_l2cache.sav_vdevs = NULL; 1306fa94a07fSbrendan } 1307fa94a07fSbrendan if (spa->spa_l2cache.sav_config) { 1308fa94a07fSbrendan nvlist_free(spa->spa_l2cache.sav_config); 1309fa94a07fSbrendan spa->spa_l2cache.sav_config = NULL; 131099653d4eSeschrock } 13112ce8af81SEric Schrock spa->spa_l2cache.sav_count = 0; 131299653d4eSeschrock 1313ea8dc4b6Seschrock spa->spa_async_suspended = 0; 13148ad4d6ddSJeff Bonwick 13158704186eSDan McDonald if (spa->spa_comment != NULL) { 13168704186eSDan McDonald spa_strfree(spa->spa_comment); 13178704186eSDan McDonald spa->spa_comment = NULL; 13188704186eSDan McDonald } 13198704186eSDan McDonald 13208ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1321fa9e4066Sahrens } 1322fa9e4066Sahrens 132399653d4eSeschrock /* 132499653d4eSeschrock * Load (or re-load) the current list of vdevs describing the active spares for 132599653d4eSeschrock * this pool. When this is called, we have some form of basic information in 1326fa94a07fSbrendan * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1327fa94a07fSbrendan * then re-generate a more complete list including status information. 132899653d4eSeschrock */ 132999653d4eSeschrock static void 133099653d4eSeschrock spa_load_spares(spa_t *spa) 133199653d4eSeschrock { 133299653d4eSeschrock nvlist_t **spares; 133399653d4eSeschrock uint_t nspares; 133499653d4eSeschrock int i; 133539c23413Seschrock vdev_t *vd, *tvd; 133699653d4eSeschrock 1337e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1338e14bb325SJeff Bonwick 133999653d4eSeschrock /* 134099653d4eSeschrock * First, close and free any existing spare vdevs. 134199653d4eSeschrock */ 1342fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 1343fa94a07fSbrendan vd = spa->spa_spares.sav_vdevs[i]; 134439c23413Seschrock 134539c23413Seschrock /* Undo the call to spa_activate() below */ 1346c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1347c5904d13Seschrock B_FALSE)) != NULL && tvd->vdev_isspare) 134839c23413Seschrock spa_spare_remove(tvd); 134939c23413Seschrock vdev_close(vd); 135039c23413Seschrock vdev_free(vd); 135199653d4eSeschrock } 135239c23413Seschrock 1353fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) 1354fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 1355fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 135699653d4eSeschrock 1357fa94a07fSbrendan if (spa->spa_spares.sav_config == NULL) 135899653d4eSeschrock nspares = 0; 135999653d4eSeschrock else 1360fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 136199653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 136299653d4eSeschrock 1363fa94a07fSbrendan spa->spa_spares.sav_count = (int)nspares; 1364fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 136599653d4eSeschrock 136699653d4eSeschrock if (nspares == 0) 136799653d4eSeschrock return; 136899653d4eSeschrock 136999653d4eSeschrock /* 137099653d4eSeschrock * Construct the array of vdevs, opening them to get status in the 137139c23413Seschrock * process. For each spare, there is potentially two different vdev_t 137239c23413Seschrock * structures associated with it: one in the list of spares (used only 137339c23413Seschrock * for basic validation purposes) and one in the active vdev 137439c23413Seschrock * configuration (if it's spared in). During this phase we open and 137539c23413Seschrock * validate each vdev on the spare list. If the vdev also exists in the 137639c23413Seschrock * active configuration, then we also mark this vdev as an active spare. 137799653d4eSeschrock */ 1378fa94a07fSbrendan spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1379fa94a07fSbrendan KM_SLEEP); 1380fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 138199653d4eSeschrock VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 138299653d4eSeschrock VDEV_ALLOC_SPARE) == 0); 138399653d4eSeschrock ASSERT(vd != NULL); 138499653d4eSeschrock 1385fa94a07fSbrendan spa->spa_spares.sav_vdevs[i] = vd; 138699653d4eSeschrock 1387c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1388c5904d13Seschrock B_FALSE)) != NULL) { 138939c23413Seschrock if (!tvd->vdev_isspare) 139039c23413Seschrock spa_spare_add(tvd); 139139c23413Seschrock 139239c23413Seschrock /* 139339c23413Seschrock * We only mark the spare active if we were successfully 139439c23413Seschrock * able to load the vdev. Otherwise, importing a pool 139539c23413Seschrock * with a bad active spare would result in strange 139639c23413Seschrock * behavior, because multiple pool would think the spare 139739c23413Seschrock * is actively in use. 139839c23413Seschrock * 139939c23413Seschrock * There is a vulnerability here to an equally bizarre 140039c23413Seschrock * circumstance, where a dead active spare is later 140139c23413Seschrock * brought back to life (onlined or otherwise). Given 140239c23413Seschrock * the rarity of this scenario, and the extra complexity 140339c23413Seschrock * it adds, we ignore the possibility. 140439c23413Seschrock */ 140539c23413Seschrock if (!vdev_is_dead(tvd)) 140639c23413Seschrock spa_spare_activate(tvd); 140739c23413Seschrock } 140839c23413Seschrock 1409e14bb325SJeff Bonwick vd->vdev_top = vd; 14106809eb4eSEric Schrock vd->vdev_aux = &spa->spa_spares; 1411e14bb325SJeff Bonwick 141299653d4eSeschrock if (vdev_open(vd) != 0) 141399653d4eSeschrock continue; 141499653d4eSeschrock 1415fa94a07fSbrendan if (vdev_validate_aux(vd) == 0) 1416fa94a07fSbrendan spa_spare_add(vd); 141799653d4eSeschrock } 141899653d4eSeschrock 141999653d4eSeschrock /* 142099653d4eSeschrock * Recompute the stashed list of spares, with status information 142199653d4eSeschrock * this time. 142299653d4eSeschrock */ 1423fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 142499653d4eSeschrock DATA_TYPE_NVLIST_ARRAY) == 0); 142599653d4eSeschrock 1426fa94a07fSbrendan spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1427fa94a07fSbrendan KM_SLEEP); 1428fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 1429fa94a07fSbrendan spares[i] = vdev_config_generate(spa, 14303f9d6ad7SLin Ling spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1431fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1432fa94a07fSbrendan ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1433fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 143499653d4eSeschrock nvlist_free(spares[i]); 1435fa94a07fSbrendan kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1436fa94a07fSbrendan } 1437fa94a07fSbrendan 1438fa94a07fSbrendan /* 1439fa94a07fSbrendan * Load (or re-load) the current list of vdevs describing the active l2cache for 1440fa94a07fSbrendan * this pool. When this is called, we have some form of basic information in 1441fa94a07fSbrendan * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1442fa94a07fSbrendan * then re-generate a more complete list including status information. 1443fa94a07fSbrendan * Devices which are already active have their details maintained, and are 1444fa94a07fSbrendan * not re-opened. 1445fa94a07fSbrendan */ 1446fa94a07fSbrendan static void 1447fa94a07fSbrendan spa_load_l2cache(spa_t *spa) 1448fa94a07fSbrendan { 1449fa94a07fSbrendan nvlist_t **l2cache; 1450fa94a07fSbrendan uint_t nl2cache; 1451fa94a07fSbrendan int i, j, oldnvdevs; 1452573ca77eSGeorge Wilson uint64_t guid; 1453fa94a07fSbrendan vdev_t *vd, **oldvdevs, **newvdevs; 1454fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 1455fa94a07fSbrendan 1456e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1457e14bb325SJeff Bonwick 1458fa94a07fSbrendan if (sav->sav_config != NULL) { 1459fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1460fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1461fa94a07fSbrendan newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1462fa94a07fSbrendan } else { 1463fa94a07fSbrendan nl2cache = 0; 1464d5285caeSGeorge Wilson newvdevs = NULL; 1465fa94a07fSbrendan } 1466fa94a07fSbrendan 1467fa94a07fSbrendan oldvdevs = sav->sav_vdevs; 1468fa94a07fSbrendan oldnvdevs = sav->sav_count; 1469fa94a07fSbrendan sav->sav_vdevs = NULL; 1470fa94a07fSbrendan sav->sav_count = 0; 1471fa94a07fSbrendan 1472fa94a07fSbrendan /* 1473fa94a07fSbrendan * Process new nvlist of vdevs. 1474fa94a07fSbrendan */ 1475fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 1476fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1477fa94a07fSbrendan &guid) == 0); 1478fa94a07fSbrendan 1479fa94a07fSbrendan newvdevs[i] = NULL; 1480fa94a07fSbrendan for (j = 0; j < oldnvdevs; j++) { 1481fa94a07fSbrendan vd = oldvdevs[j]; 1482fa94a07fSbrendan if (vd != NULL && guid == vd->vdev_guid) { 1483fa94a07fSbrendan /* 1484fa94a07fSbrendan * Retain previous vdev for add/remove ops. 1485fa94a07fSbrendan */ 1486fa94a07fSbrendan newvdevs[i] = vd; 1487fa94a07fSbrendan oldvdevs[j] = NULL; 1488fa94a07fSbrendan break; 1489fa94a07fSbrendan } 1490fa94a07fSbrendan } 1491fa94a07fSbrendan 1492fa94a07fSbrendan if (newvdevs[i] == NULL) { 1493fa94a07fSbrendan /* 1494fa94a07fSbrendan * Create new vdev 1495fa94a07fSbrendan */ 1496fa94a07fSbrendan VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1497fa94a07fSbrendan VDEV_ALLOC_L2CACHE) == 0); 1498fa94a07fSbrendan ASSERT(vd != NULL); 1499fa94a07fSbrendan newvdevs[i] = vd; 1500fa94a07fSbrendan 1501fa94a07fSbrendan /* 1502fa94a07fSbrendan * Commit this vdev as an l2cache device, 1503fa94a07fSbrendan * even if it fails to open. 1504fa94a07fSbrendan */ 1505fa94a07fSbrendan spa_l2cache_add(vd); 1506fa94a07fSbrendan 1507c5904d13Seschrock vd->vdev_top = vd; 1508c5904d13Seschrock vd->vdev_aux = sav; 1509c5904d13Seschrock 1510c5904d13Seschrock spa_l2cache_activate(vd); 1511c5904d13Seschrock 1512fa94a07fSbrendan if (vdev_open(vd) != 0) 1513fa94a07fSbrendan continue; 1514fa94a07fSbrendan 1515fa94a07fSbrendan (void) vdev_validate_aux(vd); 1516fa94a07fSbrendan 1517573ca77eSGeorge Wilson if (!vdev_is_dead(vd)) 1518573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1519fa94a07fSbrendan } 1520fa94a07fSbrendan } 1521fa94a07fSbrendan 1522fa94a07fSbrendan /* 1523fa94a07fSbrendan * Purge vdevs that were dropped 1524fa94a07fSbrendan */ 1525fa94a07fSbrendan for (i = 0; i < oldnvdevs; i++) { 1526fa94a07fSbrendan uint64_t pool; 1527fa94a07fSbrendan 1528fa94a07fSbrendan vd = oldvdevs[i]; 1529fa94a07fSbrendan if (vd != NULL) { 1530cd0837ccSGeorge Wilson ASSERT(vd->vdev_isl2cache); 1531cd0837ccSGeorge Wilson 15328ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 15338ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 1534fa94a07fSbrendan l2arc_remove_vdev(vd); 1535cd0837ccSGeorge Wilson vdev_clear_stats(vd); 1536cd0837ccSGeorge Wilson vdev_free(vd); 1537fa94a07fSbrendan } 1538fa94a07fSbrendan } 1539fa94a07fSbrendan 1540fa94a07fSbrendan if (oldvdevs) 1541fa94a07fSbrendan kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1542fa94a07fSbrendan 1543fa94a07fSbrendan if (sav->sav_config == NULL) 1544fa94a07fSbrendan goto out; 1545fa94a07fSbrendan 1546fa94a07fSbrendan sav->sav_vdevs = newvdevs; 1547fa94a07fSbrendan sav->sav_count = (int)nl2cache; 1548fa94a07fSbrendan 1549fa94a07fSbrendan /* 1550fa94a07fSbrendan * Recompute the stashed list of l2cache devices, with status 1551fa94a07fSbrendan * information this time. 1552fa94a07fSbrendan */ 1553fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1554fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 1555fa94a07fSbrendan 1556fa94a07fSbrendan l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1557fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1558fa94a07fSbrendan l2cache[i] = vdev_config_generate(spa, 15593f9d6ad7SLin Ling sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1560fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1561fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1562fa94a07fSbrendan out: 1563fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1564fa94a07fSbrendan nvlist_free(l2cache[i]); 1565fa94a07fSbrendan if (sav->sav_count) 1566fa94a07fSbrendan kmem_free(l2cache, sav->sav_count * sizeof (void *)); 156799653d4eSeschrock } 156899653d4eSeschrock 156999653d4eSeschrock static int 157099653d4eSeschrock load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 157199653d4eSeschrock { 157299653d4eSeschrock dmu_buf_t *db; 157399653d4eSeschrock char *packed = NULL; 157499653d4eSeschrock size_t nvsize = 0; 157599653d4eSeschrock int error; 157699653d4eSeschrock *value = NULL; 157799653d4eSeschrock 1578a45f1c3cSBrian Behlendorf error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1579a45f1c3cSBrian Behlendorf if (error != 0) 1580a45f1c3cSBrian Behlendorf return (error); 1581a45f1c3cSBrian Behlendorf 158299653d4eSeschrock nvsize = *(uint64_t *)db->db_data; 158399653d4eSeschrock dmu_buf_rele(db, FTAG); 158499653d4eSeschrock 158599653d4eSeschrock packed = kmem_alloc(nvsize, KM_SLEEP); 15867bfdf011SNeil Perrin error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 15877bfdf011SNeil Perrin DMU_READ_PREFETCH); 158899653d4eSeschrock if (error == 0) 158999653d4eSeschrock error = nvlist_unpack(packed, nvsize, value, 0); 159099653d4eSeschrock kmem_free(packed, nvsize); 159199653d4eSeschrock 159299653d4eSeschrock return (error); 159399653d4eSeschrock } 159499653d4eSeschrock 15953d7072f8Seschrock /* 15963d7072f8Seschrock * Checks to see if the given vdev could not be opened, in which case we post a 15973d7072f8Seschrock * sysevent to notify the autoreplace code that the device has been removed. 15983d7072f8Seschrock */ 15993d7072f8Seschrock static void 16003d7072f8Seschrock spa_check_removed(vdev_t *vd) 16013d7072f8Seschrock { 1602573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16033d7072f8Seschrock spa_check_removed(vd->vdev_child[c]); 16043d7072f8Seschrock 1605efb4a871SYuri Pankov if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1606efb4a871SYuri Pankov !vd->vdev_ishole) { 16073d7072f8Seschrock zfs_post_autoreplace(vd->vdev_spa, vd); 16083d7072f8Seschrock spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 16093d7072f8Seschrock } 16103d7072f8Seschrock } 16113d7072f8Seschrock 1612e6ca193dSGeorge Wilson /* 16134b964adaSGeorge Wilson * Validate the current config against the MOS config 1614e6ca193dSGeorge Wilson */ 16154b964adaSGeorge Wilson static boolean_t 16164b964adaSGeorge Wilson spa_config_valid(spa_t *spa, nvlist_t *config) 1617e6ca193dSGeorge Wilson { 16184b964adaSGeorge Wilson vdev_t *mrvd, *rvd = spa->spa_root_vdev; 16194b964adaSGeorge Wilson nvlist_t *nv; 16204b964adaSGeorge Wilson 16214b964adaSGeorge Wilson VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 16224b964adaSGeorge Wilson 16234b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 16244b964adaSGeorge Wilson VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 16254b964adaSGeorge Wilson 16264b964adaSGeorge Wilson ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1627e6ca193dSGeorge Wilson 162888ecc943SGeorge Wilson /* 16294b964adaSGeorge Wilson * If we're doing a normal import, then build up any additional 16304b964adaSGeorge Wilson * diagnostic information about missing devices in this config. 16314b964adaSGeorge Wilson * We'll pass this up to the user for further processing. 163288ecc943SGeorge Wilson */ 16334b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 16344b964adaSGeorge Wilson nvlist_t **child, *nv; 16354b964adaSGeorge Wilson uint64_t idx = 0; 16364b964adaSGeorge Wilson 16374b964adaSGeorge Wilson child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 16384b964adaSGeorge Wilson KM_SLEEP); 16394b964adaSGeorge Wilson VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1640e6ca193dSGeorge Wilson 16414b964adaSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 16424b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 16434b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 16444b964adaSGeorge Wilson 16454b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 16464b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops && 16474b964adaSGeorge Wilson mtvd->vdev_islog) 16484b964adaSGeorge Wilson child[idx++] = vdev_config_generate(spa, mtvd, 16494b964adaSGeorge Wilson B_FALSE, 0); 16504b964adaSGeorge Wilson } 16514b964adaSGeorge Wilson 16524b964adaSGeorge Wilson if (idx) { 16534b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist_array(nv, 16544b964adaSGeorge Wilson ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 16554b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(spa->spa_load_info, 16564b964adaSGeorge Wilson ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 16574b964adaSGeorge Wilson 16584b964adaSGeorge Wilson for (int i = 0; i < idx; i++) 16594b964adaSGeorge Wilson nvlist_free(child[i]); 16604b964adaSGeorge Wilson } 16614b964adaSGeorge Wilson nvlist_free(nv); 16624b964adaSGeorge Wilson kmem_free(child, rvd->vdev_children * sizeof (char **)); 16634b964adaSGeorge Wilson } 16644b964adaSGeorge Wilson 16654b964adaSGeorge Wilson /* 16664b964adaSGeorge Wilson * Compare the root vdev tree with the information we have 16674b964adaSGeorge Wilson * from the MOS config (mrvd). Check each top-level vdev 16684b964adaSGeorge Wilson * with the corresponding MOS config top-level (mtvd). 16694b964adaSGeorge Wilson */ 167088ecc943SGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 16714b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 16724b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 16734b964adaSGeorge Wilson 16744b964adaSGeorge Wilson /* 16754b964adaSGeorge Wilson * Resolve any "missing" vdevs in the current configuration. 16764b964adaSGeorge Wilson * If we find that the MOS config has more accurate information 16774b964adaSGeorge Wilson * about the top-level vdev then use that vdev instead. 16784b964adaSGeorge Wilson */ 16794b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 16804b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops) { 16814b964adaSGeorge Wilson 16824b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 16834b964adaSGeorge Wilson continue; 16844b964adaSGeorge Wilson 16854b964adaSGeorge Wilson /* 16864b964adaSGeorge Wilson * Device specific actions. 16874b964adaSGeorge Wilson */ 16884b964adaSGeorge Wilson if (mtvd->vdev_islog) { 16894b964adaSGeorge Wilson spa_set_log_state(spa, SPA_LOG_CLEAR); 16904b964adaSGeorge Wilson } else { 16914b964adaSGeorge Wilson /* 16924b964adaSGeorge Wilson * XXX - once we have 'readonly' pool 16934b964adaSGeorge Wilson * support we should be able to handle 16944b964adaSGeorge Wilson * missing data devices by transitioning 16954b964adaSGeorge Wilson * the pool to readonly. 16964b964adaSGeorge Wilson */ 16974b964adaSGeorge Wilson continue; 16984b964adaSGeorge Wilson } 16994b964adaSGeorge Wilson 17004b964adaSGeorge Wilson /* 17014b964adaSGeorge Wilson * Swap the missing vdev with the data we were 17024b964adaSGeorge Wilson * able to obtain from the MOS config. 17034b964adaSGeorge Wilson */ 17044b964adaSGeorge Wilson vdev_remove_child(rvd, tvd); 17054b964adaSGeorge Wilson vdev_remove_child(mrvd, mtvd); 17064b964adaSGeorge Wilson 17074b964adaSGeorge Wilson vdev_add_child(rvd, mtvd); 17084b964adaSGeorge Wilson vdev_add_child(mrvd, tvd); 17094b964adaSGeorge Wilson 17104b964adaSGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 17114b964adaSGeorge Wilson vdev_load(mtvd); 17124b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 17134b964adaSGeorge Wilson 17144b964adaSGeorge Wilson vdev_reopen(rvd); 17154b964adaSGeorge Wilson } else if (mtvd->vdev_islog) { 17164b964adaSGeorge Wilson /* 17174b964adaSGeorge Wilson * Load the slog device's state from the MOS config 17184b964adaSGeorge Wilson * since it's possible that the label does not 17194b964adaSGeorge Wilson * contain the most up-to-date information. 17204b964adaSGeorge Wilson */ 17214b964adaSGeorge Wilson vdev_load_log_state(tvd, mtvd); 17224b964adaSGeorge Wilson vdev_reopen(tvd); 17234b964adaSGeorge Wilson } 1724e6ca193dSGeorge Wilson } 17254b964adaSGeorge Wilson vdev_free(mrvd); 172688ecc943SGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 17274b964adaSGeorge Wilson 17284b964adaSGeorge Wilson /* 17294b964adaSGeorge Wilson * Ensure we were able to validate the config. 17304b964adaSGeorge Wilson */ 17314b964adaSGeorge Wilson return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1732e6ca193dSGeorge Wilson } 1733e6ca193dSGeorge Wilson 1734b87f3af3Sperrin /* 1735b87f3af3Sperrin * Check for missing log devices 1736b87f3af3Sperrin */ 17373b2aab18SMatthew Ahrens static boolean_t 1738b87f3af3Sperrin spa_check_logs(spa_t *spa) 1739b87f3af3Sperrin { 17403b2aab18SMatthew Ahrens boolean_t rv = B_FALSE; 174112380e1eSArne Jansen dsl_pool_t *dp = spa_get_dsl(spa); 17423b2aab18SMatthew Ahrens 1743b87f3af3Sperrin switch (spa->spa_log_state) { 1744b87f3af3Sperrin case SPA_LOG_MISSING: 1745b87f3af3Sperrin /* need to recheck in case slog has been restored */ 1746b87f3af3Sperrin case SPA_LOG_UNKNOWN: 174712380e1eSArne Jansen rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 174812380e1eSArne Jansen zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 17493b2aab18SMatthew Ahrens if (rv) 17501195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_MISSING); 1751b87f3af3Sperrin break; 1752b87f3af3Sperrin } 17533b2aab18SMatthew Ahrens return (rv); 1754b87f3af3Sperrin } 1755b87f3af3Sperrin 17561195e687SMark J Musante static boolean_t 17571195e687SMark J Musante spa_passivate_log(spa_t *spa) 17581195e687SMark J Musante { 17591195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 17601195e687SMark J Musante boolean_t slog_found = B_FALSE; 17611195e687SMark J Musante 17621195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 17631195e687SMark J Musante 17641195e687SMark J Musante if (!spa_has_slogs(spa)) 17651195e687SMark J Musante return (B_FALSE); 17661195e687SMark J Musante 17671195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 17681195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 17691195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 17701195e687SMark J Musante 17711195e687SMark J Musante if (tvd->vdev_islog) { 17721195e687SMark J Musante metaslab_group_passivate(mg); 17731195e687SMark J Musante slog_found = B_TRUE; 17741195e687SMark J Musante } 17751195e687SMark J Musante } 17761195e687SMark J Musante 17771195e687SMark J Musante return (slog_found); 17781195e687SMark J Musante } 17791195e687SMark J Musante 17801195e687SMark J Musante static void 17811195e687SMark J Musante spa_activate_log(spa_t *spa) 17821195e687SMark J Musante { 17831195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 17841195e687SMark J Musante 17851195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 17861195e687SMark J Musante 17871195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 17881195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 17891195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 17901195e687SMark J Musante 17911195e687SMark J Musante if (tvd->vdev_islog) 17921195e687SMark J Musante metaslab_group_activate(mg); 17931195e687SMark J Musante } 17941195e687SMark J Musante } 17951195e687SMark J Musante 17961195e687SMark J Musante int 17971195e687SMark J Musante spa_offline_log(spa_t *spa) 17981195e687SMark J Musante { 17993b2aab18SMatthew Ahrens int error; 18001195e687SMark J Musante 18013b2aab18SMatthew Ahrens error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 18023b2aab18SMatthew Ahrens NULL, DS_FIND_CHILDREN); 18033b2aab18SMatthew Ahrens if (error == 0) { 18041195e687SMark J Musante /* 18051195e687SMark J Musante * We successfully offlined the log device, sync out the 18061195e687SMark J Musante * current txg so that the "stubby" block can be removed 18071195e687SMark J Musante * by zil_sync(). 18081195e687SMark J Musante */ 18091195e687SMark J Musante txg_wait_synced(spa->spa_dsl_pool, 0); 18101195e687SMark J Musante } 18111195e687SMark J Musante return (error); 18121195e687SMark J Musante } 18131195e687SMark J Musante 1814b693757aSEric Schrock static void 1815b693757aSEric Schrock spa_aux_check_removed(spa_aux_vdev_t *sav) 1816b693757aSEric Schrock { 1817b24ab676SJeff Bonwick for (int i = 0; i < sav->sav_count; i++) 1818b693757aSEric Schrock spa_check_removed(sav->sav_vdevs[i]); 1819b693757aSEric Schrock } 1820b693757aSEric Schrock 1821b24ab676SJeff Bonwick void 1822b24ab676SJeff Bonwick spa_claim_notify(zio_t *zio) 1823b24ab676SJeff Bonwick { 1824b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 1825b24ab676SJeff Bonwick 1826b24ab676SJeff Bonwick if (zio->io_error) 1827b24ab676SJeff Bonwick return; 1828b24ab676SJeff Bonwick 1829b24ab676SJeff Bonwick mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1830b24ab676SJeff Bonwick if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1831b24ab676SJeff Bonwick spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1832b24ab676SJeff Bonwick mutex_exit(&spa->spa_props_lock); 1833b24ab676SJeff Bonwick } 1834b24ab676SJeff Bonwick 1835468c413aSTim Haley typedef struct spa_load_error { 1836c8ee1847SVictor Latushkin uint64_t sle_meta_count; 1837468c413aSTim Haley uint64_t sle_data_count; 1838468c413aSTim Haley } spa_load_error_t; 1839468c413aSTim Haley 1840468c413aSTim Haley static void 1841468c413aSTim Haley spa_load_verify_done(zio_t *zio) 1842468c413aSTim Haley { 1843468c413aSTim Haley blkptr_t *bp = zio->io_bp; 1844468c413aSTim Haley spa_load_error_t *sle = zio->io_private; 1845468c413aSTim Haley dmu_object_type_t type = BP_GET_TYPE(bp); 1846468c413aSTim Haley int error = zio->io_error; 1847e42d2059SMatthew Ahrens spa_t *spa = zio->io_spa; 1848468c413aSTim Haley 1849468c413aSTim Haley if (error) { 1850ad135b5dSChristopher Siden if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1851468c413aSTim Haley type != DMU_OT_INTENT_LOG) 18521a5e258fSJosef 'Jeff' Sipek atomic_inc_64(&sle->sle_meta_count); 1853468c413aSTim Haley else 18541a5e258fSJosef 'Jeff' Sipek atomic_inc_64(&sle->sle_data_count); 1855468c413aSTim Haley } 1856468c413aSTim Haley zio_data_buf_free(zio->io_data, zio->io_size); 1857e42d2059SMatthew Ahrens 1858e42d2059SMatthew Ahrens mutex_enter(&spa->spa_scrub_lock); 1859e42d2059SMatthew Ahrens spa->spa_scrub_inflight--; 1860e42d2059SMatthew Ahrens cv_broadcast(&spa->spa_scrub_io_cv); 1861e42d2059SMatthew Ahrens mutex_exit(&spa->spa_scrub_lock); 1862468c413aSTim Haley } 1863468c413aSTim Haley 1864e42d2059SMatthew Ahrens /* 1865e42d2059SMatthew Ahrens * Maximum number of concurrent scrub i/os to create while verifying 1866e42d2059SMatthew Ahrens * a pool while importing it. 1867e42d2059SMatthew Ahrens */ 1868e42d2059SMatthew Ahrens int spa_load_verify_maxinflight = 10000; 1869e42d2059SMatthew Ahrens boolean_t spa_load_verify_metadata = B_TRUE; 1870e42d2059SMatthew Ahrens boolean_t spa_load_verify_data = B_TRUE; 1871e42d2059SMatthew Ahrens 1872468c413aSTim Haley /*ARGSUSED*/ 1873468c413aSTim Haley static int 1874b24ab676SJeff Bonwick spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 18757802d7bfSMatthew Ahrens const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1876468c413aSTim Haley { 1877a2cdcdd2SPaul Dagnelie if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1878e42d2059SMatthew Ahrens return (0); 1879e42d2059SMatthew Ahrens /* 1880e42d2059SMatthew Ahrens * Note: normally this routine will not be called if 1881e42d2059SMatthew Ahrens * spa_load_verify_metadata is not set. However, it may be useful 1882e42d2059SMatthew Ahrens * to manually set the flag after the traversal has begun. 1883e42d2059SMatthew Ahrens */ 1884e42d2059SMatthew Ahrens if (!spa_load_verify_metadata) 1885e42d2059SMatthew Ahrens return (0); 1886e42d2059SMatthew Ahrens if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1887e42d2059SMatthew Ahrens return (0); 1888468c413aSTim Haley 1889e42d2059SMatthew Ahrens zio_t *rio = arg; 1890e42d2059SMatthew Ahrens size_t size = BP_GET_PSIZE(bp); 1891e42d2059SMatthew Ahrens void *data = zio_data_buf_alloc(size); 1892e42d2059SMatthew Ahrens 1893e42d2059SMatthew Ahrens mutex_enter(&spa->spa_scrub_lock); 1894e42d2059SMatthew Ahrens while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1895e42d2059SMatthew Ahrens cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1896e42d2059SMatthew Ahrens spa->spa_scrub_inflight++; 1897e42d2059SMatthew Ahrens mutex_exit(&spa->spa_scrub_lock); 1898e42d2059SMatthew Ahrens 1899e42d2059SMatthew Ahrens zio_nowait(zio_read(rio, spa, bp, data, size, 1900e42d2059SMatthew Ahrens spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1901e42d2059SMatthew Ahrens ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1902e42d2059SMatthew Ahrens ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1903468c413aSTim Haley return (0); 1904468c413aSTim Haley } 1905468c413aSTim Haley 1906468c413aSTim Haley static int 1907468c413aSTim Haley spa_load_verify(spa_t *spa) 1908468c413aSTim Haley { 1909468c413aSTim Haley zio_t *rio; 1910468c413aSTim Haley spa_load_error_t sle = { 0 }; 1911468c413aSTim Haley zpool_rewind_policy_t policy; 1912468c413aSTim Haley boolean_t verify_ok = B_FALSE; 1913e42d2059SMatthew Ahrens int error = 0; 1914468c413aSTim Haley 1915c8ee1847SVictor Latushkin zpool_get_rewind_policy(spa->spa_config, &policy); 1916c8ee1847SVictor Latushkin 1917c8ee1847SVictor Latushkin if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1918c8ee1847SVictor Latushkin return (0); 1919c8ee1847SVictor Latushkin 1920468c413aSTim Haley rio = zio_root(spa, NULL, &sle, 1921468c413aSTim Haley ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1922468c413aSTim Haley 1923e42d2059SMatthew Ahrens if (spa_load_verify_metadata) { 1924e42d2059SMatthew Ahrens error = traverse_pool(spa, spa->spa_verify_min_txg, 1925e42d2059SMatthew Ahrens TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1926e42d2059SMatthew Ahrens spa_load_verify_cb, rio); 1927e42d2059SMatthew Ahrens } 1928468c413aSTim Haley 1929468c413aSTim Haley (void) zio_wait(rio); 1930468c413aSTim Haley 1931c8ee1847SVictor Latushkin spa->spa_load_meta_errors = sle.sle_meta_count; 1932468c413aSTim Haley spa->spa_load_data_errors = sle.sle_data_count; 1933468c413aSTim Haley 1934c8ee1847SVictor Latushkin if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1935468c413aSTim Haley sle.sle_data_count <= policy.zrp_maxdata) { 19364b964adaSGeorge Wilson int64_t loss = 0; 19374b964adaSGeorge Wilson 1938468c413aSTim Haley verify_ok = B_TRUE; 1939468c413aSTim Haley spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1940468c413aSTim Haley spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 19414b964adaSGeorge Wilson 19424b964adaSGeorge Wilson loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 19434b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 19444b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 19454b964adaSGeorge Wilson VERIFY(nvlist_add_int64(spa->spa_load_info, 19464b964adaSGeorge Wilson ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 19474b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 19484b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1949a33cae98STim Haley } else { 1950a33cae98STim Haley spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1951468c413aSTim Haley } 1952468c413aSTim Haley 1953468c413aSTim Haley if (error) { 1954468c413aSTim Haley if (error != ENXIO && error != EIO) 1955be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 1956468c413aSTim Haley return (error); 1957468c413aSTim Haley } 1958468c413aSTim Haley 1959468c413aSTim Haley return (verify_ok ? 0 : EIO); 1960468c413aSTim Haley } 1961468c413aSTim Haley 19621195e687SMark J Musante /* 19631195e687SMark J Musante * Find a value in the pool props object. 19641195e687SMark J Musante */ 19651195e687SMark J Musante static void 19661195e687SMark J Musante spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 19671195e687SMark J Musante { 19681195e687SMark J Musante (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 19691195e687SMark J Musante zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 19701195e687SMark J Musante } 19711195e687SMark J Musante 19721195e687SMark J Musante /* 19731195e687SMark J Musante * Find a value in the pool directory object. 19741195e687SMark J Musante */ 19751195e687SMark J Musante static int 19761195e687SMark J Musante spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 19771195e687SMark J Musante { 19781195e687SMark J Musante return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 19791195e687SMark J Musante name, sizeof (uint64_t), 1, val)); 19801195e687SMark J Musante } 19811195e687SMark J Musante 19821195e687SMark J Musante static int 19831195e687SMark J Musante spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 19841195e687SMark J Musante { 19851195e687SMark J Musante vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 19861195e687SMark J Musante return (err); 19871195e687SMark J Musante } 19881195e687SMark J Musante 19891195e687SMark J Musante /* 19901195e687SMark J Musante * Fix up config after a partly-completed split. This is done with the 19911195e687SMark J Musante * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 19921195e687SMark J Musante * pool have that entry in their config, but only the splitting one contains 19931195e687SMark J Musante * a list of all the guids of the vdevs that are being split off. 19941195e687SMark J Musante * 19951195e687SMark J Musante * This function determines what to do with that list: either rejoin 19961195e687SMark J Musante * all the disks to the pool, or complete the splitting process. To attempt 19971195e687SMark J Musante * the rejoin, each disk that is offlined is marked online again, and 19981195e687SMark J Musante * we do a reopen() call. If the vdev label for every disk that was 19991195e687SMark J Musante * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 20001195e687SMark J Musante * then we call vdev_split() on each disk, and complete the split. 20011195e687SMark J Musante * 2002d41c4376SMark J Musante * Otherwise we leave the config alone, with all the vdevs in place in 2003d41c4376SMark J Musante * the original pool. 20041195e687SMark J Musante */ 20051195e687SMark J Musante static void 20061195e687SMark J Musante spa_try_repair(spa_t *spa, nvlist_t *config) 20071195e687SMark J Musante { 20081195e687SMark J Musante uint_t extracted; 20091195e687SMark J Musante uint64_t *glist; 20101195e687SMark J Musante uint_t i, gcount; 20111195e687SMark J Musante nvlist_t *nvl; 20121195e687SMark J Musante vdev_t **vd; 20131195e687SMark J Musante boolean_t attempt_reopen; 20141195e687SMark J Musante 20151195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 20161195e687SMark J Musante return; 20171195e687SMark J Musante 20181195e687SMark J Musante /* check that the config is complete */ 20191195e687SMark J Musante if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 20201195e687SMark J Musante &glist, &gcount) != 0) 20211195e687SMark J Musante return; 20221195e687SMark J Musante 20231195e687SMark J Musante vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 20241195e687SMark J Musante 20251195e687SMark J Musante /* attempt to online all the vdevs & validate */ 20261195e687SMark J Musante attempt_reopen = B_TRUE; 20271195e687SMark J Musante for (i = 0; i < gcount; i++) { 20281195e687SMark J Musante if (glist[i] == 0) /* vdev is hole */ 20291195e687SMark J Musante continue; 20301195e687SMark J Musante 20311195e687SMark J Musante vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 20321195e687SMark J Musante if (vd[i] == NULL) { 20331195e687SMark J Musante /* 20341195e687SMark J Musante * Don't bother attempting to reopen the disks; 20351195e687SMark J Musante * just do the split. 20361195e687SMark J Musante */ 20371195e687SMark J Musante attempt_reopen = B_FALSE; 20381195e687SMark J Musante } else { 20391195e687SMark J Musante /* attempt to re-online it */ 20401195e687SMark J Musante vd[i]->vdev_offline = B_FALSE; 20411195e687SMark J Musante } 20421195e687SMark J Musante } 20431195e687SMark J Musante 20441195e687SMark J Musante if (attempt_reopen) { 20451195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 20461195e687SMark J Musante 20471195e687SMark J Musante /* check each device to see what state it's in */ 20481195e687SMark J Musante for (extracted = 0, i = 0; i < gcount; i++) { 20491195e687SMark J Musante if (vd[i] != NULL && 20501195e687SMark J Musante vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 20511195e687SMark J Musante break; 20521195e687SMark J Musante ++extracted; 20531195e687SMark J Musante } 20541195e687SMark J Musante } 20551195e687SMark J Musante 20561195e687SMark J Musante /* 20571195e687SMark J Musante * If every disk has been moved to the new pool, or if we never 20581195e687SMark J Musante * even attempted to look at them, then we split them off for 20591195e687SMark J Musante * good. 20601195e687SMark J Musante */ 20611195e687SMark J Musante if (!attempt_reopen || gcount == extracted) { 20621195e687SMark J Musante for (i = 0; i < gcount; i++) 20631195e687SMark J Musante if (vd[i] != NULL) 20641195e687SMark J Musante vdev_split(vd[i]); 20651195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 20661195e687SMark J Musante } 20671195e687SMark J Musante 20681195e687SMark J Musante kmem_free(vd, gcount * sizeof (vdev_t *)); 20691195e687SMark J Musante } 20701195e687SMark J Musante 20711195e687SMark J Musante static int 20721195e687SMark J Musante spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 20731195e687SMark J Musante boolean_t mosconfig) 20741195e687SMark J Musante { 20751195e687SMark J Musante nvlist_t *config = spa->spa_config; 20761195e687SMark J Musante char *ereport = FM_EREPORT_ZFS_POOL; 20778704186eSDan McDonald char *comment; 20781195e687SMark J Musante int error; 20791195e687SMark J Musante uint64_t pool_guid; 20801195e687SMark J Musante nvlist_t *nvl; 20811195e687SMark J Musante 20821195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2083be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 20841195e687SMark J Musante 20858704186eSDan McDonald ASSERT(spa->spa_comment == NULL); 20868704186eSDan McDonald if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 20878704186eSDan McDonald spa->spa_comment = spa_strdup(comment); 20888704186eSDan McDonald 20891195e687SMark J Musante /* 20901195e687SMark J Musante * Versioning wasn't explicitly added to the label until later, so if 20911195e687SMark J Musante * it's not present treat it as the initial version. 20921195e687SMark J Musante */ 20931195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 20941195e687SMark J Musante &spa->spa_ubsync.ub_version) != 0) 20951195e687SMark J Musante spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 20961195e687SMark J Musante 20971195e687SMark J Musante (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 20981195e687SMark J Musante &spa->spa_config_txg); 20991195e687SMark J Musante 21001195e687SMark J Musante if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 21011195e687SMark J Musante spa_guid_exists(pool_guid, 0)) { 2102be6fd75aSMatthew Ahrens error = SET_ERROR(EEXIST); 21031195e687SMark J Musante } else { 2104e9103aaeSGarrett D'Amore spa->spa_config_guid = pool_guid; 21051195e687SMark J Musante 21061195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 21071195e687SMark J Musante &nvl) == 0) { 21081195e687SMark J Musante VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 21091195e687SMark J Musante KM_SLEEP) == 0); 21101195e687SMark J Musante } 21111195e687SMark J Musante 2112ad135b5dSChristopher Siden nvlist_free(spa->spa_load_info); 2113ad135b5dSChristopher Siden spa->spa_load_info = fnvlist_alloc(); 2114ad135b5dSChristopher Siden 211511027bc7STim Haley gethrestime(&spa->spa_loaded_ts); 21161195e687SMark J Musante error = spa_load_impl(spa, pool_guid, config, state, type, 21171195e687SMark J Musante mosconfig, &ereport); 21181195e687SMark J Musante } 21191195e687SMark J Musante 2120bc9014e6SJustin Gibbs /* 2121bc9014e6SJustin Gibbs * Don't count references from objsets that are already closed 2122bc9014e6SJustin Gibbs * and are making their way through the eviction process. 2123bc9014e6SJustin Gibbs */ 2124bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 21251195e687SMark J Musante spa->spa_minref = refcount_count(&spa->spa_refcount); 212611027bc7STim Haley if (error) { 212711027bc7STim Haley if (error != EEXIST) { 212811027bc7STim Haley spa->spa_loaded_ts.tv_sec = 0; 212911027bc7STim Haley spa->spa_loaded_ts.tv_nsec = 0; 213011027bc7STim Haley } 213111027bc7STim Haley if (error != EBADF) { 213211027bc7STim Haley zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 213311027bc7STim Haley } 213411027bc7STim Haley } 21351195e687SMark J Musante spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 21361195e687SMark J Musante spa->spa_ena = 0; 21371195e687SMark J Musante 21381195e687SMark J Musante return (error); 21391195e687SMark J Musante } 21401195e687SMark J Musante 2141fa9e4066Sahrens /* 2142fa9e4066Sahrens * Load an existing storage pool, using the pool's builtin spa_config as a 2143ea8dc4b6Seschrock * source of configuration information. 2144fa9e4066Sahrens */ 2145fa9e4066Sahrens static int 21461195e687SMark J Musante spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 21471195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 21481195e687SMark J Musante char **ereport) 2149fa9e4066Sahrens { 2150fa9e4066Sahrens int error = 0; 2151871a9500SMark J Musante nvlist_t *nvroot = NULL; 2152ad135b5dSChristopher Siden nvlist_t *label; 2153fa9e4066Sahrens vdev_t *rvd; 2154fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 21554b964adaSGeorge Wilson uint64_t children, config_cache_txg = spa->spa_config_txg; 21568ad4d6ddSJeff Bonwick int orig_mode = spa->spa_mode; 21571195e687SMark J Musante int parse; 2158cde58dbcSMatthew Ahrens uint64_t obj; 2159ad135b5dSChristopher Siden boolean_t missing_feat_write = B_FALSE; 2160fa9e4066Sahrens 21618ad4d6ddSJeff Bonwick /* 21628ad4d6ddSJeff Bonwick * If this is an untrusted config, access the pool in read-only mode. 21638ad4d6ddSJeff Bonwick * This prevents things like resilvering recently removed devices. 21648ad4d6ddSJeff Bonwick */ 21658ad4d6ddSJeff Bonwick if (!mosconfig) 21668ad4d6ddSJeff Bonwick spa->spa_mode = FREAD; 21678ad4d6ddSJeff Bonwick 2168e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2169e14bb325SJeff Bonwick 2170ea8dc4b6Seschrock spa->spa_load_state = state; 21710373e76bSbonwick 21721195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2173be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2174fa9e4066Sahrens 21751195e687SMark J Musante parse = (type == SPA_IMPORT_EXISTING ? 21761195e687SMark J Musante VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2177b5989ec7Seschrock 217854d692b7SGeorge Wilson /* 217954d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 218054d692b7SGeorge Wilson */ 21816f834bc1SMatthew Ahrens spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 21826f834bc1SMatthew Ahrens KM_SLEEP); 21836f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) { 21846f834bc1SMatthew Ahrens spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 21856f834bc1SMatthew Ahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 21866f834bc1SMatthew Ahrens ZIO_FLAG_GODFATHER); 21876f834bc1SMatthew Ahrens } 218854d692b7SGeorge Wilson 2189fa9e4066Sahrens /* 219099653d4eSeschrock * Parse the configuration into a vdev tree. We explicitly set the 219199653d4eSeschrock * value that will be returned by spa_version() since parsing the 219299653d4eSeschrock * configuration requires knowing the version number. 2193fa9e4066Sahrens */ 2194e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 21951195e687SMark J Musante error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2196e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2197fa9e4066Sahrens 219899653d4eSeschrock if (error != 0) 21991195e687SMark J Musante return (error); 2200fa9e4066Sahrens 22010e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == rvd); 220281cd5c55SMatthew Ahrens ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 220381cd5c55SMatthew Ahrens ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 22041195e687SMark J Musante 22051195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 22061195e687SMark J Musante ASSERT(spa_guid(spa) == pool_guid); 22071195e687SMark J Musante } 2208fa9e4066Sahrens 2209fa9e4066Sahrens /* 2210fa9e4066Sahrens * Try to open all vdevs, loading each label in the process. 2211fa9e4066Sahrens */ 2212e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 22130bf246f5Smc error = vdev_open(rvd); 2214e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 22150bf246f5Smc if (error != 0) 22161195e687SMark J Musante return (error); 2217fa9e4066Sahrens 2218560e6e96Seschrock /* 221977e3a39cSMark J Musante * We need to validate the vdev labels against the configuration that 222077e3a39cSMark J Musante * we have in hand, which is dependent on the setting of mosconfig. If 222177e3a39cSMark J Musante * mosconfig is true then we're validating the vdev labels based on 22221195e687SMark J Musante * that config. Otherwise, we're validating against the cached config 222377e3a39cSMark J Musante * (zpool.cache) that was read when we loaded the zfs module, and then 222477e3a39cSMark J Musante * later we will recursively call spa_load() and validate against 222577e3a39cSMark J Musante * the vdev config. 22261195e687SMark J Musante * 22271195e687SMark J Musante * If we're assembling a new pool that's been split off from an 22281195e687SMark J Musante * existing pool, the labels haven't yet been updated so we skip 22291195e687SMark J Musante * validation for now. 2230560e6e96Seschrock */ 22311195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 22321195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2233d7f601efSGeorge Wilson error = vdev_validate(rvd, mosconfig); 22341195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 2235560e6e96Seschrock 22361195e687SMark J Musante if (error != 0) 22371195e687SMark J Musante return (error); 22381195e687SMark J Musante 22391195e687SMark J Musante if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2240be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 2241560e6e96Seschrock } 2242560e6e96Seschrock 2243fa9e4066Sahrens /* 2244fa9e4066Sahrens * Find the best uberblock. 2245fa9e4066Sahrens */ 2246ad135b5dSChristopher Siden vdev_uberblock_load(rvd, ub, &label); 2247fa9e4066Sahrens 2248fa9e4066Sahrens /* 2249fa9e4066Sahrens * If we weren't able to find a single valid uberblock, return failure. 2250fa9e4066Sahrens */ 2251ad135b5dSChristopher Siden if (ub->ub_txg == 0) { 2252ad135b5dSChristopher Siden nvlist_free(label); 22531195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2254ad135b5dSChristopher Siden } 2255ea8dc4b6Seschrock 2256ea8dc4b6Seschrock /* 2257ad135b5dSChristopher Siden * If the pool has an unsupported version we can't open it. 2258ea8dc4b6Seschrock */ 2259ad135b5dSChristopher Siden if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2260ad135b5dSChristopher Siden nvlist_free(label); 22611195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2262ad135b5dSChristopher Siden } 2263ad135b5dSChristopher Siden 2264ad135b5dSChristopher Siden if (ub->ub_version >= SPA_VERSION_FEATURES) { 2265ad135b5dSChristopher Siden nvlist_t *features; 2266ad135b5dSChristopher Siden 2267ad135b5dSChristopher Siden /* 2268ad135b5dSChristopher Siden * If we weren't able to find what's necessary for reading the 2269ad135b5dSChristopher Siden * MOS in the label, return failure. 2270ad135b5dSChristopher Siden */ 2271ad135b5dSChristopher Siden if (label == NULL || nvlist_lookup_nvlist(label, 2272ad135b5dSChristopher Siden ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2273ad135b5dSChristopher Siden nvlist_free(label); 2274ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2275ad135b5dSChristopher Siden ENXIO)); 2276ad135b5dSChristopher Siden } 2277ad135b5dSChristopher Siden 2278ad135b5dSChristopher Siden /* 2279ad135b5dSChristopher Siden * Update our in-core representation with the definitive values 2280ad135b5dSChristopher Siden * from the label. 2281ad135b5dSChristopher Siden */ 2282ad135b5dSChristopher Siden nvlist_free(spa->spa_label_features); 2283ad135b5dSChristopher Siden VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2284ad135b5dSChristopher Siden } 2285ad135b5dSChristopher Siden 2286ad135b5dSChristopher Siden nvlist_free(label); 2287ad135b5dSChristopher Siden 2288ad135b5dSChristopher Siden /* 2289ad135b5dSChristopher Siden * Look through entries in the label nvlist's features_for_read. If 2290ad135b5dSChristopher Siden * there is a feature listed there which we don't understand then we 2291ad135b5dSChristopher Siden * cannot open a pool. 2292ad135b5dSChristopher Siden */ 2293ad135b5dSChristopher Siden if (ub->ub_version >= SPA_VERSION_FEATURES) { 2294ad135b5dSChristopher Siden nvlist_t *unsup_feat; 2295ad135b5dSChristopher Siden 2296ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2297ad135b5dSChristopher Siden 0); 2298ad135b5dSChristopher Siden 2299ad135b5dSChristopher Siden for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2300ad135b5dSChristopher Siden NULL); nvp != NULL; 2301ad135b5dSChristopher Siden nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2302ad135b5dSChristopher Siden if (!zfeature_is_supported(nvpair_name(nvp))) { 2303ad135b5dSChristopher Siden VERIFY(nvlist_add_string(unsup_feat, 2304ad135b5dSChristopher Siden nvpair_name(nvp), "") == 0); 2305ad135b5dSChristopher Siden } 2306ad135b5dSChristopher Siden } 2307ad135b5dSChristopher Siden 2308ad135b5dSChristopher Siden if (!nvlist_empty(unsup_feat)) { 2309ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2310ad135b5dSChristopher Siden ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2311ad135b5dSChristopher Siden nvlist_free(unsup_feat); 2312ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2313ad135b5dSChristopher Siden ENOTSUP)); 2314ad135b5dSChristopher Siden } 2315ad135b5dSChristopher Siden 2316ad135b5dSChristopher Siden nvlist_free(unsup_feat); 2317ad135b5dSChristopher Siden } 2318fa9e4066Sahrens 2319fa9e4066Sahrens /* 2320fa9e4066Sahrens * If the vdev guid sum doesn't match the uberblock, we have an 23214b964adaSGeorge Wilson * incomplete configuration. We first check to see if the pool 23224b964adaSGeorge Wilson * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 23234b964adaSGeorge Wilson * If it is, defer the vdev_guid_sum check till later so we 23244b964adaSGeorge Wilson * can handle missing vdevs. 2325fa9e4066Sahrens */ 23264b964adaSGeorge Wilson if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 23274b964adaSGeorge Wilson &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 23281195e687SMark J Musante rvd->vdev_guid_sum != ub->ub_guid_sum) 23291195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 23301195e687SMark J Musante 23311195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 23321195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 23331195e687SMark J Musante spa_try_repair(spa, config); 23341195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 23351195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 23361195e687SMark J Musante spa->spa_config_splitting = NULL; 2337fa9e4066Sahrens } 2338fa9e4066Sahrens 2339fa9e4066Sahrens /* 2340fa9e4066Sahrens * Initialize internal SPA structures. 2341fa9e4066Sahrens */ 2342fa9e4066Sahrens spa->spa_state = POOL_STATE_ACTIVE; 2343fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 2344468c413aSTim Haley spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2345c8ee1847SVictor Latushkin TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2346468c413aSTim Haley spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2347468c413aSTim Haley spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2348b24ab676SJeff Bonwick spa->spa_claim_max_txg = spa->spa_first_txg; 23493f9d6ad7SLin Ling spa->spa_prev_software_version = ub->ub_software_version; 2350b24ab676SJeff Bonwick 2351ad135b5dSChristopher Siden error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 23521195e687SMark J Musante if (error) 23531195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2354fa9e4066Sahrens spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2355fa9e4066Sahrens 23561195e687SMark J Musante if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 23571195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2358fa9e4066Sahrens 2359ad135b5dSChristopher Siden if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2360ad135b5dSChristopher Siden boolean_t missing_feat_read = B_FALSE; 236157221772SChristopher Siden nvlist_t *unsup_feat, *enabled_feat; 2362ad135b5dSChristopher Siden 2363ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2364ad135b5dSChristopher Siden &spa->spa_feat_for_read_obj) != 0) { 2365ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2366ad135b5dSChristopher Siden } 2367ad135b5dSChristopher Siden 2368ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2369ad135b5dSChristopher Siden &spa->spa_feat_for_write_obj) != 0) { 2370ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2371ad135b5dSChristopher Siden } 2372ad135b5dSChristopher Siden 2373ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2374ad135b5dSChristopher Siden &spa->spa_feat_desc_obj) != 0) { 2375ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2376ad135b5dSChristopher Siden } 2377ad135b5dSChristopher Siden 237857221772SChristopher Siden enabled_feat = fnvlist_alloc(); 237957221772SChristopher Siden unsup_feat = fnvlist_alloc(); 2380ad135b5dSChristopher Siden 23812acef22dSMatthew Ahrens if (!spa_features_check(spa, B_FALSE, 238257221772SChristopher Siden unsup_feat, enabled_feat)) 2383ad135b5dSChristopher Siden missing_feat_read = B_TRUE; 2384ad135b5dSChristopher Siden 2385ad135b5dSChristopher Siden if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 23862acef22dSMatthew Ahrens if (!spa_features_check(spa, B_TRUE, 238757221772SChristopher Siden unsup_feat, enabled_feat)) { 2388ad135b5dSChristopher Siden missing_feat_write = B_TRUE; 238957221772SChristopher Siden } 2390ad135b5dSChristopher Siden } 2391ad135b5dSChristopher Siden 239257221772SChristopher Siden fnvlist_add_nvlist(spa->spa_load_info, 239357221772SChristopher Siden ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 239457221772SChristopher Siden 2395ad135b5dSChristopher Siden if (!nvlist_empty(unsup_feat)) { 239657221772SChristopher Siden fnvlist_add_nvlist(spa->spa_load_info, 239757221772SChristopher Siden ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2398ad135b5dSChristopher Siden } 2399ad135b5dSChristopher Siden 240057221772SChristopher Siden fnvlist_free(enabled_feat); 240157221772SChristopher Siden fnvlist_free(unsup_feat); 2402ad135b5dSChristopher Siden 2403ad135b5dSChristopher Siden if (!missing_feat_read) { 2404ad135b5dSChristopher Siden fnvlist_add_boolean(spa->spa_load_info, 2405ad135b5dSChristopher Siden ZPOOL_CONFIG_CAN_RDONLY); 2406ad135b5dSChristopher Siden } 2407ad135b5dSChristopher Siden 2408ad135b5dSChristopher Siden /* 2409ad135b5dSChristopher Siden * If the state is SPA_LOAD_TRYIMPORT, our objective is 2410ad135b5dSChristopher Siden * twofold: to determine whether the pool is available for 2411ad135b5dSChristopher Siden * import in read-write mode and (if it is not) whether the 2412ad135b5dSChristopher Siden * pool is available for import in read-only mode. If the pool 2413ad135b5dSChristopher Siden * is available for import in read-write mode, it is displayed 2414ad135b5dSChristopher Siden * as available in userland; if it is not available for import 2415ad135b5dSChristopher Siden * in read-only mode, it is displayed as unavailable in 2416ad135b5dSChristopher Siden * userland. If the pool is available for import in read-only 2417ad135b5dSChristopher Siden * mode but not read-write mode, it is displayed as unavailable 2418ad135b5dSChristopher Siden * in userland with a special note that the pool is actually 2419ad135b5dSChristopher Siden * available for open in read-only mode. 2420ad135b5dSChristopher Siden * 2421ad135b5dSChristopher Siden * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2422ad135b5dSChristopher Siden * missing a feature for write, we must first determine whether 2423ad135b5dSChristopher Siden * the pool can be opened read-only before returning to 2424ad135b5dSChristopher Siden * userland in order to know whether to display the 2425ad135b5dSChristopher Siden * abovementioned note. 2426ad135b5dSChristopher Siden */ 2427ad135b5dSChristopher Siden if (missing_feat_read || (missing_feat_write && 2428ad135b5dSChristopher Siden spa_writeable(spa))) { 2429ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2430ad135b5dSChristopher Siden ENOTSUP)); 2431ad135b5dSChristopher Siden } 243243466aaeSMax Grossman 243343466aaeSMax Grossman /* 243443466aaeSMax Grossman * Load refcounts for ZFS features from disk into an in-memory 243543466aaeSMax Grossman * cache during SPA initialization. 243643466aaeSMax Grossman */ 243743466aaeSMax Grossman for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 243843466aaeSMax Grossman uint64_t refcount; 243943466aaeSMax Grossman 244043466aaeSMax Grossman error = feature_get_refcount_from_disk(spa, 244143466aaeSMax Grossman &spa_feature_table[i], &refcount); 244243466aaeSMax Grossman if (error == 0) { 244343466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = refcount; 244443466aaeSMax Grossman } else if (error == ENOTSUP) { 244543466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = 244643466aaeSMax Grossman SPA_FEATURE_DISABLED; 244743466aaeSMax Grossman } else { 244843466aaeSMax Grossman return (spa_vdev_err(rvd, 244943466aaeSMax Grossman VDEV_AUX_CORRUPT_DATA, EIO)); 245043466aaeSMax Grossman } 245143466aaeSMax Grossman } 245243466aaeSMax Grossman } 245343466aaeSMax Grossman 245443466aaeSMax Grossman if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 245543466aaeSMax Grossman if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 24565d7b4d43SMatthew Ahrens &spa->spa_feat_enabled_txg_obj) != 0) 245743466aaeSMax Grossman return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2458ad135b5dSChristopher Siden } 2459ad135b5dSChristopher Siden 2460ad135b5dSChristopher Siden spa->spa_is_initializing = B_TRUE; 2461ad135b5dSChristopher Siden error = dsl_pool_open(spa->spa_dsl_pool); 2462ad135b5dSChristopher Siden spa->spa_is_initializing = B_FALSE; 2463ad135b5dSChristopher Siden if (error != 0) 2464ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2465ad135b5dSChristopher Siden 2466fa9e4066Sahrens if (!mosconfig) { 246795173954Sek uint64_t hostid; 2468871a9500SMark J Musante nvlist_t *policy = NULL, *nvconfig; 2469871a9500SMark J Musante 2470871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2471871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2472fa9e4066Sahrens 247388ecc943SGeorge Wilson if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 247477650510SLin Ling ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 247595173954Sek char *hostname; 247695173954Sek unsigned long myhostid = 0; 247795173954Sek 247888ecc943SGeorge Wilson VERIFY(nvlist_lookup_string(nvconfig, 247995173954Sek ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 248095173954Sek 24815679c89fSjv #ifdef _KERNEL 24825679c89fSjv myhostid = zone_get_hostid(NULL); 24835679c89fSjv #else /* _KERNEL */ 24845679c89fSjv /* 24855679c89fSjv * We're emulating the system's hostid in userland, so 24865679c89fSjv * we can't use zone_get_hostid(). 24875679c89fSjv */ 248895173954Sek (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 24895679c89fSjv #endif /* _KERNEL */ 249017194a52Slling if (hostid != 0 && myhostid != 0 && 24915679c89fSjv hostid != myhostid) { 2492871a9500SMark J Musante nvlist_free(nvconfig); 249395173954Sek cmn_err(CE_WARN, "pool '%s' could not be " 249495173954Sek "loaded as it was last accessed by " 249577650510SLin Ling "another system (host: %s hostid: 0x%lx). " 2496654b400cSJoshua M. Clulow "See: http://illumos.org/msg/ZFS-8000-EY", 2497e14bb325SJeff Bonwick spa_name(spa), hostname, 249895173954Sek (unsigned long)hostid); 2499be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 250095173954Sek } 250195173954Sek } 2502c8ee1847SVictor Latushkin if (nvlist_lookup_nvlist(spa->spa_config, 2503c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, &policy) == 0) 2504c8ee1847SVictor Latushkin VERIFY(nvlist_add_nvlist(nvconfig, 2505c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, policy) == 0); 250695173954Sek 250788ecc943SGeorge Wilson spa_config_set(spa, nvconfig); 2508fa9e4066Sahrens spa_unload(spa); 2509fa9e4066Sahrens spa_deactivate(spa); 25108ad4d6ddSJeff Bonwick spa_activate(spa, orig_mode); 2511fa9e4066Sahrens 25121195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2513fa9e4066Sahrens } 2514fa9e4066Sahrens 2515cde58dbcSMatthew Ahrens if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2516cde58dbcSMatthew Ahrens return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2517cde58dbcSMatthew Ahrens error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2518cde58dbcSMatthew Ahrens if (error != 0) 25191195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2520fa9e4066Sahrens 252199653d4eSeschrock /* 252299653d4eSeschrock * Load the bit that tells us to use the new accounting function 252399653d4eSeschrock * (raid-z deflation). If we have an older pool, this will not 252499653d4eSeschrock * be present. 252599653d4eSeschrock */ 25261195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 25271195e687SMark J Musante if (error != 0 && error != ENOENT) 25281195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 252999653d4eSeschrock 25303f9d6ad7SLin Ling error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 25313f9d6ad7SLin Ling &spa->spa_creation_version); 25323f9d6ad7SLin Ling if (error != 0 && error != ENOENT) 25333f9d6ad7SLin Ling return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 25343f9d6ad7SLin Ling 2535fa9e4066Sahrens /* 2536ea8dc4b6Seschrock * Load the persistent error log. If we have an older pool, this will 2537ea8dc4b6Seschrock * not be present. 2538fa9e4066Sahrens */ 25391195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 25401195e687SMark J Musante if (error != 0 && error != ENOENT) 25411195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2542ea8dc4b6Seschrock 25431195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 25441195e687SMark J Musante &spa->spa_errlog_scrub); 25451195e687SMark J Musante if (error != 0 && error != ENOENT) 25461195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2547ea8dc4b6Seschrock 254806eeb2adSek /* 254906eeb2adSek * Load the history object. If we have an older pool, this 255006eeb2adSek * will not be present. 255106eeb2adSek */ 25521195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 25531195e687SMark J Musante if (error != 0 && error != ENOENT) 25541195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 25551195e687SMark J Musante 25561195e687SMark J Musante /* 25571195e687SMark J Musante * If we're assembling the pool from the split-off vdevs of 25581195e687SMark J Musante * an existing pool, we don't want to attach the spares & cache 25591195e687SMark J Musante * devices. 25601195e687SMark J Musante */ 256106eeb2adSek 256299653d4eSeschrock /* 256399653d4eSeschrock * Load any hot spares for this pool. 256499653d4eSeschrock */ 25651195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 25661195e687SMark J Musante if (error != 0 && error != ENOENT) 25671195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 25681195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2569e7437265Sahrens ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2570fa94a07fSbrendan if (load_nvlist(spa, spa->spa_spares.sav_object, 25711195e687SMark J Musante &spa->spa_spares.sav_config) != 0) 25721195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 257399653d4eSeschrock 2574e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 257599653d4eSeschrock spa_load_spares(spa); 2576e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 25771195e687SMark J Musante } else if (error == 0) { 25781195e687SMark J Musante spa->spa_spares.sav_sync = B_TRUE; 257999653d4eSeschrock } 258099653d4eSeschrock 2581fa94a07fSbrendan /* 2582fa94a07fSbrendan * Load any level 2 ARC devices for this pool. 2583fa94a07fSbrendan */ 25841195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2585fa94a07fSbrendan &spa->spa_l2cache.sav_object); 25861195e687SMark J Musante if (error != 0 && error != ENOENT) 25871195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 25881195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2589fa94a07fSbrendan ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2590fa94a07fSbrendan if (load_nvlist(spa, spa->spa_l2cache.sav_object, 25911195e687SMark J Musante &spa->spa_l2cache.sav_config) != 0) 25921195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2593fa94a07fSbrendan 2594e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2595fa94a07fSbrendan spa_load_l2cache(spa); 2596e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 25971195e687SMark J Musante } else if (error == 0) { 25981195e687SMark J Musante spa->spa_l2cache.sav_sync = B_TRUE; 2599fa94a07fSbrendan } 2600fa94a07fSbrendan 2601990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2602ecd6cf80Smarks 26031195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 26041195e687SMark J Musante if (error && error != ENOENT) 26051195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2606b1b8ab34Slling 2607b1b8ab34Slling if (error == 0) { 26081195e687SMark J Musante uint64_t autoreplace; 26091195e687SMark J Musante 26101195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 26111195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 26121195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 26131195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 26141195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 26151195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 26161195e687SMark J Musante &spa->spa_dedup_ditto); 26171195e687SMark J Musante 2618b693757aSEric Schrock spa->spa_autoreplace = (autoreplace != 0); 2619b1b8ab34Slling } 2620b1b8ab34Slling 26213d7072f8Seschrock /* 26223d7072f8Seschrock * If the 'autoreplace' property is set, then post a resource notifying 26233d7072f8Seschrock * the ZFS DE that it should not issue any faults for unopenable 26243d7072f8Seschrock * devices. We also iterate over the vdevs, and post a sysevent for any 26253d7072f8Seschrock * unopenable vdevs so that the normal autoreplace handler can take 26263d7072f8Seschrock * over. 26273d7072f8Seschrock */ 2628b693757aSEric Schrock if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 26293d7072f8Seschrock spa_check_removed(spa->spa_root_vdev); 2630b693757aSEric Schrock /* 2631b693757aSEric Schrock * For the import case, this is done in spa_import(), because 2632b693757aSEric Schrock * at this point we're using the spare definitions from 2633b693757aSEric Schrock * the MOS config, not necessarily from the userland config. 2634b693757aSEric Schrock */ 2635b693757aSEric Schrock if (state != SPA_LOAD_IMPORT) { 2636b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 2637b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 2638b693757aSEric Schrock } 2639b693757aSEric Schrock } 26403d7072f8Seschrock 2641ea8dc4b6Seschrock /* 2642560e6e96Seschrock * Load the vdev state for all toplevel vdevs. 2643ea8dc4b6Seschrock */ 2644560e6e96Seschrock vdev_load(rvd); 26450373e76bSbonwick 2646fa9e4066Sahrens /* 2647fa9e4066Sahrens * Propagate the leaf DTLs we just loaded all the way up the tree. 2648fa9e4066Sahrens */ 2649e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2650fa9e4066Sahrens vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2651e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2652fa9e4066Sahrens 2653b24ab676SJeff Bonwick /* 2654b24ab676SJeff Bonwick * Load the DDTs (dedup tables). 2655b24ab676SJeff Bonwick */ 2656b24ab676SJeff Bonwick error = ddt_load(spa); 26571195e687SMark J Musante if (error != 0) 26581195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2659b24ab676SJeff Bonwick 2660485bbbf5SGeorge Wilson spa_update_dspace(spa); 2661485bbbf5SGeorge Wilson 2662b24ab676SJeff Bonwick /* 26634b964adaSGeorge Wilson * Validate the config, using the MOS config to fill in any 26644b964adaSGeorge Wilson * information which might be missing. If we fail to validate 26654b964adaSGeorge Wilson * the config then declare the pool unfit for use. If we're 26664b964adaSGeorge Wilson * assembling a pool from a split, the log is not transferred 26674b964adaSGeorge Wilson * over. 2668b24ab676SJeff Bonwick */ 26691195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 2670871a9500SMark J Musante nvlist_t *nvconfig; 2671871a9500SMark J Musante 2672871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2673871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2674871a9500SMark J Musante 26754b964adaSGeorge Wilson if (!spa_config_valid(spa, nvconfig)) { 26764b964adaSGeorge Wilson nvlist_free(nvconfig); 26774b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 26784b964adaSGeorge Wilson ENXIO)); 26794b964adaSGeorge Wilson } 26801195e687SMark J Musante nvlist_free(nvconfig); 26811195e687SMark J Musante 26824b964adaSGeorge Wilson /* 2683ad135b5dSChristopher Siden * Now that we've validated the config, check the state of the 26844b964adaSGeorge Wilson * root vdev. If it can't be opened, it indicates one or 26854b964adaSGeorge Wilson * more toplevel vdevs are faulted. 26864b964adaSGeorge Wilson */ 26874b964adaSGeorge Wilson if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2688be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 26894b964adaSGeorge Wilson 269023367a2fSMatthew Ahrens if (spa_writeable(spa) && spa_check_logs(spa)) { 26911195e687SMark J Musante *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 26921195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 26931195e687SMark J Musante } 2694b24ab676SJeff Bonwick } 2695b24ab676SJeff Bonwick 2696ad135b5dSChristopher Siden if (missing_feat_write) { 2697ad135b5dSChristopher Siden ASSERT(state == SPA_LOAD_TRYIMPORT); 2698ad135b5dSChristopher Siden 2699ad135b5dSChristopher Siden /* 2700ad135b5dSChristopher Siden * At this point, we know that we can open the pool in 2701ad135b5dSChristopher Siden * read-only mode but not read-write mode. We now have enough 2702ad135b5dSChristopher Siden * information and can return to userland. 2703ad135b5dSChristopher Siden */ 2704ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2705ad135b5dSChristopher Siden } 2706ad135b5dSChristopher Siden 27074b964adaSGeorge Wilson /* 27084b964adaSGeorge Wilson * We've successfully opened the pool, verify that we're ready 27094b964adaSGeorge Wilson * to start pushing transactions. 27104b964adaSGeorge Wilson */ 27114b964adaSGeorge Wilson if (state != SPA_LOAD_TRYIMPORT) { 27124b964adaSGeorge Wilson if (error = spa_load_verify(spa)) 27134b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 27144b964adaSGeorge Wilson error)); 27154b964adaSGeorge Wilson } 27164b964adaSGeorge Wilson 2717468c413aSTim Haley if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2718468c413aSTim Haley spa->spa_load_max_txg == UINT64_MAX)) { 27195dabedeeSbonwick dmu_tx_t *tx; 27200373e76bSbonwick int need_update = B_FALSE; 272112380e1eSArne Jansen dsl_pool_t *dp = spa_get_dsl(spa); 27228ad4d6ddSJeff Bonwick 27238ad4d6ddSJeff Bonwick ASSERT(state != SPA_LOAD_TRYIMPORT); 27245dabedeeSbonwick 27250373e76bSbonwick /* 27260373e76bSbonwick * Claim log blocks that haven't been committed yet. 27270373e76bSbonwick * This must all happen in a single txg. 2728b24ab676SJeff Bonwick * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2729b24ab676SJeff Bonwick * invoked from zil_claim_log_block()'s i/o done callback. 2730468c413aSTim Haley * Price of rollback is that we abandon the log. 27310373e76bSbonwick */ 2732b24ab676SJeff Bonwick spa->spa_claiming = B_TRUE; 2733b24ab676SJeff Bonwick 273412380e1eSArne Jansen tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 273512380e1eSArne Jansen (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 27360b69c2f0Sahrens zil_claim, tx, DS_FIND_CHILDREN); 2737fa9e4066Sahrens dmu_tx_commit(tx); 2738fa9e4066Sahrens 2739b24ab676SJeff Bonwick spa->spa_claiming = B_FALSE; 2740b24ab676SJeff Bonwick 27411195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_GOOD); 2742fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 2743fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 2744fa9e4066Sahrens 2745fa9e4066Sahrens /* 2746b24ab676SJeff Bonwick * Wait for all claims to sync. We sync up to the highest 2747b24ab676SJeff Bonwick * claimed log block birth time so that claimed log blocks 2748b24ab676SJeff Bonwick * don't appear to be from the future. spa_claim_max_txg 2749b24ab676SJeff Bonwick * will have been set for us by either zil_check_log_chain() 2750b24ab676SJeff Bonwick * (invoked from spa_check_logs()) or zil_claim() above. 2751fa9e4066Sahrens */ 2752b24ab676SJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 27530e34b6a7Sbonwick 27540e34b6a7Sbonwick /* 27550373e76bSbonwick * If the config cache is stale, or we have uninitialized 27560373e76bSbonwick * metaslabs (see spa_vdev_add()), then update the config. 2757bc758434SLin Ling * 27584b964adaSGeorge Wilson * If this is a verbatim import, trust the current 2759bc758434SLin Ling * in-core spa_config and update the disk labels. 27600e34b6a7Sbonwick */ 27610373e76bSbonwick if (config_cache_txg != spa->spa_config_txg || 27624b964adaSGeorge Wilson state == SPA_LOAD_IMPORT || 27634b964adaSGeorge Wilson state == SPA_LOAD_RECOVER || 27644b964adaSGeorge Wilson (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 27650373e76bSbonwick need_update = B_TRUE; 27660373e76bSbonwick 27678ad4d6ddSJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) 27680373e76bSbonwick if (rvd->vdev_child[c]->vdev_ms_array == 0) 27690373e76bSbonwick need_update = B_TRUE; 27700e34b6a7Sbonwick 27710e34b6a7Sbonwick /* 27720373e76bSbonwick * Update the config cache asychronously in case we're the 27730373e76bSbonwick * root pool, in which case the config cache isn't writable yet. 27740e34b6a7Sbonwick */ 27750373e76bSbonwick if (need_update) 27760373e76bSbonwick spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 27778ad4d6ddSJeff Bonwick 27788ad4d6ddSJeff Bonwick /* 27798ad4d6ddSJeff Bonwick * Check all DTLs to see if anything needs resilvering. 27808ad4d6ddSJeff Bonwick */ 27813f9d6ad7SLin Ling if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 27823f9d6ad7SLin Ling vdev_resilver_needed(rvd, NULL, NULL)) 27838ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 2784503ad85cSMatthew Ahrens 27854445fffbSMatthew Ahrens /* 27864445fffbSMatthew Ahrens * Log the fact that we booted up (so that we can detect if 27874445fffbSMatthew Ahrens * we rebooted in the middle of an operation). 27884445fffbSMatthew Ahrens */ 27894445fffbSMatthew Ahrens spa_history_log_version(spa, "open"); 27904445fffbSMatthew Ahrens 2791503ad85cSMatthew Ahrens /* 2792503ad85cSMatthew Ahrens * Delete any inconsistent datasets. 2793503ad85cSMatthew Ahrens */ 2794503ad85cSMatthew Ahrens (void) dmu_objset_find(spa_name(spa), 2795503ad85cSMatthew Ahrens dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2796ca45db41SChris Kirby 2797ca45db41SChris Kirby /* 2798ca45db41SChris Kirby * Clean up any stale temporary dataset userrefs. 2799ca45db41SChris Kirby */ 2800ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2801fa9e4066Sahrens } 2802fa9e4066Sahrens 28031195e687SMark J Musante return (0); 2804fa9e4066Sahrens } 2805fa9e4066Sahrens 2806468c413aSTim Haley static int 2807468c413aSTim Haley spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2808468c413aSTim Haley { 2809f9af39baSGeorge Wilson int mode = spa->spa_mode; 2810f9af39baSGeorge Wilson 2811468c413aSTim Haley spa_unload(spa); 2812468c413aSTim Haley spa_deactivate(spa); 2813468c413aSTim Haley 2814e42d2059SMatthew Ahrens spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2815468c413aSTim Haley 2816f9af39baSGeorge Wilson spa_activate(spa, mode); 2817468c413aSTim Haley spa_async_suspend(spa); 2818468c413aSTim Haley 28191195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2820468c413aSTim Haley } 2821468c413aSTim Haley 2822ad135b5dSChristopher Siden /* 2823ad135b5dSChristopher Siden * If spa_load() fails this function will try loading prior txg's. If 2824ad135b5dSChristopher Siden * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2825ad135b5dSChristopher Siden * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2826ad135b5dSChristopher Siden * function will not rewind the pool and will return the same error as 2827ad135b5dSChristopher Siden * spa_load(). 2828ad135b5dSChristopher Siden */ 2829468c413aSTim Haley static int 2830468c413aSTim Haley spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2831c8ee1847SVictor Latushkin uint64_t max_request, int rewind_flags) 2832468c413aSTim Haley { 2833ad135b5dSChristopher Siden nvlist_t *loadinfo = NULL; 2834468c413aSTim Haley nvlist_t *config = NULL; 2835468c413aSTim Haley int load_error, rewind_error; 2836c8ee1847SVictor Latushkin uint64_t safe_rewind_txg; 2837468c413aSTim Haley uint64_t min_txg; 2838468c413aSTim Haley 2839a33cae98STim Haley if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2840468c413aSTim Haley spa->spa_load_max_txg = spa->spa_load_txg; 28411195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2842a33cae98STim Haley } else { 2843468c413aSTim Haley spa->spa_load_max_txg = max_request; 2844e42d2059SMatthew Ahrens if (max_request != UINT64_MAX) 2845e42d2059SMatthew Ahrens spa->spa_extreme_rewind = B_TRUE; 2846a33cae98STim Haley } 2847468c413aSTim Haley 28481195e687SMark J Musante load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 28491195e687SMark J Musante mosconfig); 2850468c413aSTim Haley if (load_error == 0) 2851468c413aSTim Haley return (0); 2852468c413aSTim Haley 2853468c413aSTim Haley if (spa->spa_root_vdev != NULL) 2854468c413aSTim Haley config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2855468c413aSTim Haley 2856468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2857468c413aSTim Haley spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2858468c413aSTim Haley 2859c8ee1847SVictor Latushkin if (rewind_flags & ZPOOL_NEVER_REWIND) { 2860468c413aSTim Haley nvlist_free(config); 2861468c413aSTim Haley return (load_error); 2862468c413aSTim Haley } 2863468c413aSTim Haley 2864ad135b5dSChristopher Siden if (state == SPA_LOAD_RECOVER) { 2865ad135b5dSChristopher Siden /* Price of rolling back is discarding txgs, including log */ 28661195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2867ad135b5dSChristopher Siden } else { 2868ad135b5dSChristopher Siden /* 2869ad135b5dSChristopher Siden * If we aren't rolling back save the load info from our first 2870ad135b5dSChristopher Siden * import attempt so that we can restore it after attempting 2871ad135b5dSChristopher Siden * to rewind. 2872ad135b5dSChristopher Siden */ 2873ad135b5dSChristopher Siden loadinfo = spa->spa_load_info; 2874ad135b5dSChristopher Siden spa->spa_load_info = fnvlist_alloc(); 2875ad135b5dSChristopher Siden } 2876468c413aSTim Haley 2877c8ee1847SVictor Latushkin spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2878c8ee1847SVictor Latushkin safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2879c8ee1847SVictor Latushkin min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2880c8ee1847SVictor Latushkin TXG_INITIAL : safe_rewind_txg; 2881468c413aSTim Haley 2882c8ee1847SVictor Latushkin /* 2883c8ee1847SVictor Latushkin * Continue as long as we're finding errors, we're still within 2884c8ee1847SVictor Latushkin * the acceptable rewind range, and we're still finding uberblocks 2885c8ee1847SVictor Latushkin */ 2886c8ee1847SVictor Latushkin while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2887c8ee1847SVictor Latushkin spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2888c8ee1847SVictor Latushkin if (spa->spa_load_max_txg < safe_rewind_txg) 2889468c413aSTim Haley spa->spa_extreme_rewind = B_TRUE; 2890468c413aSTim Haley rewind_error = spa_load_retry(spa, state, mosconfig); 2891468c413aSTim Haley } 2892468c413aSTim Haley 2893468c413aSTim Haley spa->spa_extreme_rewind = B_FALSE; 2894468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 2895468c413aSTim Haley 2896468c413aSTim Haley if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2897468c413aSTim Haley spa_config_set(spa, config); 2898468c413aSTim Haley 2899ad135b5dSChristopher Siden if (state == SPA_LOAD_RECOVER) { 2900ad135b5dSChristopher Siden ASSERT3P(loadinfo, ==, NULL); 2901ad135b5dSChristopher Siden return (rewind_error); 2902ad135b5dSChristopher Siden } else { 2903ad135b5dSChristopher Siden /* Store the rewind info as part of the initial load info */ 2904ad135b5dSChristopher Siden fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2905ad135b5dSChristopher Siden spa->spa_load_info); 2906ad135b5dSChristopher Siden 2907ad135b5dSChristopher Siden /* Restore the initial load info */ 2908ad135b5dSChristopher Siden fnvlist_free(spa->spa_load_info); 2909ad135b5dSChristopher Siden spa->spa_load_info = loadinfo; 2910ad135b5dSChristopher Siden 2911ad135b5dSChristopher Siden return (load_error); 2912ad135b5dSChristopher Siden } 2913468c413aSTim Haley } 2914468c413aSTim Haley 2915fa9e4066Sahrens /* 2916fa9e4066Sahrens * Pool Open/Import 2917fa9e4066Sahrens * 2918fa9e4066Sahrens * The import case is identical to an open except that the configuration is sent 2919fa9e4066Sahrens * down from userland, instead of grabbed from the configuration cache. For the 2920fa9e4066Sahrens * case of an open, the pool configuration will exist in the 29213d7072f8Seschrock * POOL_STATE_UNINITIALIZED state. 2922fa9e4066Sahrens * 2923fa9e4066Sahrens * The stats information (gen/count/ustats) is used to gather vdev statistics at 2924fa9e4066Sahrens * the same time open the pool, without having to keep around the spa_t in some 2925fa9e4066Sahrens * ambiguous state. 2926fa9e4066Sahrens */ 2927fa9e4066Sahrens static int 2928468c413aSTim Haley spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2929468c413aSTim Haley nvlist_t **config) 2930fa9e4066Sahrens { 2931fa9e4066Sahrens spa_t *spa; 29324b964adaSGeorge Wilson spa_load_state_t state = SPA_LOAD_OPEN; 2933fa9e4066Sahrens int error; 2934fa9e4066Sahrens int locked = B_FALSE; 2935fa9e4066Sahrens 2936fa9e4066Sahrens *spapp = NULL; 2937fa9e4066Sahrens 2938fa9e4066Sahrens /* 2939fa9e4066Sahrens * As disgusting as this is, we need to support recursive calls to this 2940fa9e4066Sahrens * function because dsl_dir_open() is called during spa_load(), and ends 2941fa9e4066Sahrens * up calling spa_open() again. The real fix is to figure out how to 2942fa9e4066Sahrens * avoid dsl_dir_open() calling this in the first place. 2943fa9e4066Sahrens */ 2944fa9e4066Sahrens if (mutex_owner(&spa_namespace_lock) != curthread) { 2945fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 2946fa9e4066Sahrens locked = B_TRUE; 2947fa9e4066Sahrens } 2948fa9e4066Sahrens 2949fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 2950fa9e4066Sahrens if (locked) 2951fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2952be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 2953fa9e4066Sahrens } 2954468c413aSTim Haley 2955fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 29564b44c88cSTim Haley zpool_rewind_policy_t policy; 29574b44c88cSTim Haley 29584b44c88cSTim Haley zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 29594b44c88cSTim Haley &policy); 29604b44c88cSTim Haley if (policy.zrp_request & ZPOOL_DO_REWIND) 29614b44c88cSTim Haley state = SPA_LOAD_RECOVER; 2962fa9e4066Sahrens 29638ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 2964fa9e4066Sahrens 2965468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 2966468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2967468c413aSTim Haley 2968468c413aSTim Haley error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2969c8ee1847SVictor Latushkin policy.zrp_request); 2970fa9e4066Sahrens 2971fa9e4066Sahrens if (error == EBADF) { 2972fa9e4066Sahrens /* 2973560e6e96Seschrock * If vdev_validate() returns failure (indicated by 2974560e6e96Seschrock * EBADF), it indicates that one of the vdevs indicates 2975560e6e96Seschrock * that the pool has been exported or destroyed. If 2976560e6e96Seschrock * this is the case, the config cache is out of sync and 2977560e6e96Seschrock * we should remove the pool from the namespace. 2978fa9e4066Sahrens */ 2979fa9e4066Sahrens spa_unload(spa); 2980fa9e4066Sahrens spa_deactivate(spa); 2981c5904d13Seschrock spa_config_sync(spa, B_TRUE, B_TRUE); 2982fa9e4066Sahrens spa_remove(spa); 2983fa9e4066Sahrens if (locked) 2984fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2985be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 2986ea8dc4b6Seschrock } 2987ea8dc4b6Seschrock 2988ea8dc4b6Seschrock if (error) { 2989fa9e4066Sahrens /* 2990fa9e4066Sahrens * We can't open the pool, but we still have useful 2991fa9e4066Sahrens * information: the state of each vdev after the 2992fa9e4066Sahrens * attempted vdev_open(). Return this to the user. 2993fa9e4066Sahrens */ 29944b964adaSGeorge Wilson if (config != NULL && spa->spa_config) { 2995468c413aSTim Haley VERIFY(nvlist_dup(spa->spa_config, config, 2996468c413aSTim Haley KM_SLEEP) == 0); 29974b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, 29984b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_INFO, 29994b964adaSGeorge Wilson spa->spa_load_info) == 0); 30004b964adaSGeorge Wilson } 3001fa9e4066Sahrens spa_unload(spa); 3002fa9e4066Sahrens spa_deactivate(spa); 3003468c413aSTim Haley spa->spa_last_open_failed = error; 3004fa9e4066Sahrens if (locked) 3005fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3006fa9e4066Sahrens *spapp = NULL; 3007fa9e4066Sahrens return (error); 3008fa9e4066Sahrens } 3009fa9e4066Sahrens } 3010fa9e4066Sahrens 3011fa9e4066Sahrens spa_open_ref(spa, tag); 30123d7072f8Seschrock 3013468c413aSTim Haley if (config != NULL) 3014468c413aSTim Haley *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3015468c413aSTim Haley 30164b964adaSGeorge Wilson /* 30174b964adaSGeorge Wilson * If we've recovered the pool, pass back any information we 30184b964adaSGeorge Wilson * gathered while doing the load. 30194b964adaSGeorge Wilson */ 30204b964adaSGeorge Wilson if (state == SPA_LOAD_RECOVER) { 30214b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 30224b964adaSGeorge Wilson spa->spa_load_info) == 0); 30234b964adaSGeorge Wilson } 30244b964adaSGeorge Wilson 3025a33cae98STim Haley if (locked) { 3026a33cae98STim Haley spa->spa_last_open_failed = 0; 3027a33cae98STim Haley spa->spa_last_ubsync_txg = 0; 3028a33cae98STim Haley spa->spa_load_txg = 0; 3029fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3030a33cae98STim Haley } 3031fa9e4066Sahrens 3032fa9e4066Sahrens *spapp = spa; 3033fa9e4066Sahrens 3034fa9e4066Sahrens return (0); 3035fa9e4066Sahrens } 3036fa9e4066Sahrens 3037468c413aSTim Haley int 3038468c413aSTim Haley spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3039468c413aSTim Haley nvlist_t **config) 3040468c413aSTim Haley { 3041468c413aSTim Haley return (spa_open_common(name, spapp, tag, policy, config)); 3042468c413aSTim Haley } 3043468c413aSTim Haley 3044fa9e4066Sahrens int 3045fa9e4066Sahrens spa_open(const char *name, spa_t **spapp, void *tag) 3046fa9e4066Sahrens { 3047468c413aSTim Haley return (spa_open_common(name, spapp, tag, NULL, NULL)); 3048fa9e4066Sahrens } 3049fa9e4066Sahrens 3050ea8dc4b6Seschrock /* 3051ea8dc4b6Seschrock * Lookup the given spa_t, incrementing the inject count in the process, 3052ea8dc4b6Seschrock * preventing it from being exported or destroyed. 3053ea8dc4b6Seschrock */ 3054ea8dc4b6Seschrock spa_t * 3055ea8dc4b6Seschrock spa_inject_addref(char *name) 3056ea8dc4b6Seschrock { 3057ea8dc4b6Seschrock spa_t *spa; 3058ea8dc4b6Seschrock 3059ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3060ea8dc4b6Seschrock if ((spa = spa_lookup(name)) == NULL) { 3061ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3062ea8dc4b6Seschrock return (NULL); 3063ea8dc4b6Seschrock } 3064ea8dc4b6Seschrock spa->spa_inject_ref++; 3065ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3066ea8dc4b6Seschrock 3067ea8dc4b6Seschrock return (spa); 3068ea8dc4b6Seschrock } 3069ea8dc4b6Seschrock 3070ea8dc4b6Seschrock void 3071ea8dc4b6Seschrock spa_inject_delref(spa_t *spa) 3072ea8dc4b6Seschrock { 3073ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3074ea8dc4b6Seschrock spa->spa_inject_ref--; 3075ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3076ea8dc4b6Seschrock } 3077ea8dc4b6Seschrock 3078fa94a07fSbrendan /* 3079fa94a07fSbrendan * Add spares device information to the nvlist. 3080fa94a07fSbrendan */ 308199653d4eSeschrock static void 308299653d4eSeschrock spa_add_spares(spa_t *spa, nvlist_t *config) 308399653d4eSeschrock { 308499653d4eSeschrock nvlist_t **spares; 308599653d4eSeschrock uint_t i, nspares; 308699653d4eSeschrock nvlist_t *nvroot; 308799653d4eSeschrock uint64_t guid; 308899653d4eSeschrock vdev_stat_t *vs; 308999653d4eSeschrock uint_t vsc; 309039c23413Seschrock uint64_t pool; 309199653d4eSeschrock 30926809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 30936809eb4eSEric Schrock 3094fa94a07fSbrendan if (spa->spa_spares.sav_count == 0) 309599653d4eSeschrock return; 309699653d4eSeschrock 309799653d4eSeschrock VERIFY(nvlist_lookup_nvlist(config, 309899653d4eSeschrock ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3099fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 310099653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 310199653d4eSeschrock if (nspares != 0) { 310299653d4eSeschrock VERIFY(nvlist_add_nvlist_array(nvroot, 310399653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 310499653d4eSeschrock VERIFY(nvlist_lookup_nvlist_array(nvroot, 310599653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 310699653d4eSeschrock 310799653d4eSeschrock /* 310899653d4eSeschrock * Go through and find any spares which have since been 310999653d4eSeschrock * repurposed as an active spare. If this is the case, update 311099653d4eSeschrock * their status appropriately. 311199653d4eSeschrock */ 311299653d4eSeschrock for (i = 0; i < nspares; i++) { 311399653d4eSeschrock VERIFY(nvlist_lookup_uint64(spares[i], 311499653d4eSeschrock ZPOOL_CONFIG_GUID, &guid) == 0); 311589a89ebfSlling if (spa_spare_exists(guid, &pool, NULL) && 311689a89ebfSlling pool != 0ULL) { 311799653d4eSeschrock VERIFY(nvlist_lookup_uint64_array( 31183f9d6ad7SLin Ling spares[i], ZPOOL_CONFIG_VDEV_STATS, 311999653d4eSeschrock (uint64_t **)&vs, &vsc) == 0); 312099653d4eSeschrock vs->vs_state = VDEV_STATE_CANT_OPEN; 312199653d4eSeschrock vs->vs_aux = VDEV_AUX_SPARED; 312299653d4eSeschrock } 312399653d4eSeschrock } 312499653d4eSeschrock } 312599653d4eSeschrock } 312699653d4eSeschrock 3127fa94a07fSbrendan /* 3128fa94a07fSbrendan * Add l2cache device information to the nvlist, including vdev stats. 3129fa94a07fSbrendan */ 3130fa94a07fSbrendan static void 3131fa94a07fSbrendan spa_add_l2cache(spa_t *spa, nvlist_t *config) 3132fa94a07fSbrendan { 3133fa94a07fSbrendan nvlist_t **l2cache; 3134fa94a07fSbrendan uint_t i, j, nl2cache; 3135fa94a07fSbrendan nvlist_t *nvroot; 3136fa94a07fSbrendan uint64_t guid; 3137fa94a07fSbrendan vdev_t *vd; 3138fa94a07fSbrendan vdev_stat_t *vs; 3139fa94a07fSbrendan uint_t vsc; 3140fa94a07fSbrendan 31416809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 31426809eb4eSEric Schrock 3143fa94a07fSbrendan if (spa->spa_l2cache.sav_count == 0) 3144fa94a07fSbrendan return; 3145fa94a07fSbrendan 3146fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist(config, 3147fa94a07fSbrendan ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3148fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3149fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3150fa94a07fSbrendan if (nl2cache != 0) { 3151fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, 3152fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3153fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(nvroot, 3154fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3155fa94a07fSbrendan 3156fa94a07fSbrendan /* 3157fa94a07fSbrendan * Update level 2 cache device stats. 3158fa94a07fSbrendan */ 3159fa94a07fSbrendan 3160fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 3161fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 3162fa94a07fSbrendan ZPOOL_CONFIG_GUID, &guid) == 0); 3163fa94a07fSbrendan 3164fa94a07fSbrendan vd = NULL; 3165fa94a07fSbrendan for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3166fa94a07fSbrendan if (guid == 3167fa94a07fSbrendan spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3168fa94a07fSbrendan vd = spa->spa_l2cache.sav_vdevs[j]; 3169fa94a07fSbrendan break; 3170fa94a07fSbrendan } 3171fa94a07fSbrendan } 3172fa94a07fSbrendan ASSERT(vd != NULL); 3173fa94a07fSbrendan 3174fa94a07fSbrendan VERIFY(nvlist_lookup_uint64_array(l2cache[i], 31753f9d6ad7SLin Ling ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 31763f9d6ad7SLin Ling == 0); 3177fa94a07fSbrendan vdev_get_stats(vd, vs); 3178fa94a07fSbrendan } 3179fa94a07fSbrendan } 3180fa94a07fSbrendan } 3181fa94a07fSbrendan 3182ad135b5dSChristopher Siden static void 3183ad135b5dSChristopher Siden spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3184ad135b5dSChristopher Siden { 3185ad135b5dSChristopher Siden nvlist_t *features; 3186ad135b5dSChristopher Siden zap_cursor_t zc; 3187ad135b5dSChristopher Siden zap_attribute_t za; 3188ad135b5dSChristopher Siden 3189ad135b5dSChristopher Siden ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3190ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3191ad135b5dSChristopher Siden 3192ad135b5dSChristopher Siden if (spa->spa_feat_for_read_obj != 0) { 3193ad135b5dSChristopher Siden for (zap_cursor_init(&zc, spa->spa_meta_objset, 3194ad135b5dSChristopher Siden spa->spa_feat_for_read_obj); 3195ad135b5dSChristopher Siden zap_cursor_retrieve(&zc, &za) == 0; 3196ad135b5dSChristopher Siden zap_cursor_advance(&zc)) { 3197ad135b5dSChristopher Siden ASSERT(za.za_integer_length == sizeof (uint64_t) && 3198ad135b5dSChristopher Siden za.za_num_integers == 1); 3199b420f3adSRichard Lowe VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3200ad135b5dSChristopher Siden za.za_first_integer)); 3201ad135b5dSChristopher Siden } 3202ad135b5dSChristopher Siden zap_cursor_fini(&zc); 3203ad135b5dSChristopher Siden } 3204ad135b5dSChristopher Siden 3205ad135b5dSChristopher Siden if (spa->spa_feat_for_write_obj != 0) { 3206ad135b5dSChristopher Siden for (zap_cursor_init(&zc, spa->spa_meta_objset, 3207ad135b5dSChristopher Siden spa->spa_feat_for_write_obj); 3208ad135b5dSChristopher Siden zap_cursor_retrieve(&zc, &za) == 0; 3209ad135b5dSChristopher Siden zap_cursor_advance(&zc)) { 3210ad135b5dSChristopher Siden ASSERT(za.za_integer_length == sizeof (uint64_t) && 3211ad135b5dSChristopher Siden za.za_num_integers == 1); 3212b420f3adSRichard Lowe VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3213ad135b5dSChristopher Siden za.za_first_integer)); 3214ad135b5dSChristopher Siden } 3215ad135b5dSChristopher Siden zap_cursor_fini(&zc); 3216ad135b5dSChristopher Siden } 3217ad135b5dSChristopher Siden 3218ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3219ad135b5dSChristopher Siden features) == 0); 3220ad135b5dSChristopher Siden nvlist_free(features); 3221ad135b5dSChristopher Siden } 3222ad135b5dSChristopher Siden 3223fa9e4066Sahrens int 3224ad135b5dSChristopher Siden spa_get_stats(const char *name, nvlist_t **config, 3225ad135b5dSChristopher Siden char *altroot, size_t buflen) 3226fa9e4066Sahrens { 3227fa9e4066Sahrens int error; 3228fa9e4066Sahrens spa_t *spa; 3229fa9e4066Sahrens 3230fa9e4066Sahrens *config = NULL; 3231468c413aSTim Haley error = spa_open_common(name, &spa, FTAG, NULL, config); 3232fa9e4066Sahrens 32336809eb4eSEric Schrock if (spa != NULL) { 32346809eb4eSEric Schrock /* 32356809eb4eSEric Schrock * This still leaves a window of inconsistency where the spares 32366809eb4eSEric Schrock * or l2cache devices could change and the config would be 32376809eb4eSEric Schrock * self-inconsistent. 32386809eb4eSEric Schrock */ 32396809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3240ea8dc4b6Seschrock 32416809eb4eSEric Schrock if (*config != NULL) { 324211027bc7STim Haley uint64_t loadtimes[2]; 324311027bc7STim Haley 324411027bc7STim Haley loadtimes[0] = spa->spa_loaded_ts.tv_sec; 324511027bc7STim Haley loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 324611027bc7STim Haley VERIFY(nvlist_add_uint64_array(*config, 324711027bc7STim Haley ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 324811027bc7STim Haley 3249e14bb325SJeff Bonwick VERIFY(nvlist_add_uint64(*config, 32506809eb4eSEric Schrock ZPOOL_CONFIG_ERRCOUNT, 32516809eb4eSEric Schrock spa_get_errlog_size(spa)) == 0); 3252e14bb325SJeff Bonwick 32536809eb4eSEric Schrock if (spa_suspended(spa)) 32546809eb4eSEric Schrock VERIFY(nvlist_add_uint64(*config, 32556809eb4eSEric Schrock ZPOOL_CONFIG_SUSPENDED, 32566809eb4eSEric Schrock spa->spa_failmode) == 0); 32576809eb4eSEric Schrock 32586809eb4eSEric Schrock spa_add_spares(spa, *config); 32596809eb4eSEric Schrock spa_add_l2cache(spa, *config); 3260ad135b5dSChristopher Siden spa_add_feature_stats(spa, *config); 32616809eb4eSEric Schrock } 326299653d4eSeschrock } 326399653d4eSeschrock 3264ea8dc4b6Seschrock /* 3265ea8dc4b6Seschrock * We want to get the alternate root even for faulted pools, so we cheat 3266ea8dc4b6Seschrock * and call spa_lookup() directly. 3267ea8dc4b6Seschrock */ 3268ea8dc4b6Seschrock if (altroot) { 3269ea8dc4b6Seschrock if (spa == NULL) { 3270ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3271ea8dc4b6Seschrock spa = spa_lookup(name); 3272ea8dc4b6Seschrock if (spa) 3273ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 3274ea8dc4b6Seschrock else 3275ea8dc4b6Seschrock altroot[0] = '\0'; 3276ea8dc4b6Seschrock spa = NULL; 3277ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3278ea8dc4b6Seschrock } else { 3279ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 3280ea8dc4b6Seschrock } 3281ea8dc4b6Seschrock } 3282ea8dc4b6Seschrock 32836809eb4eSEric Schrock if (spa != NULL) { 32846809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 3285fa9e4066Sahrens spa_close(spa, FTAG); 32866809eb4eSEric Schrock } 3287fa9e4066Sahrens 3288fa9e4066Sahrens return (error); 3289fa9e4066Sahrens } 3290fa9e4066Sahrens 329199653d4eSeschrock /* 3292fa94a07fSbrendan * Validate that the auxiliary device array is well formed. We must have an 3293fa94a07fSbrendan * array of nvlists, each which describes a valid leaf vdev. If this is an 3294fa94a07fSbrendan * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3295fa94a07fSbrendan * specified, as long as they are well-formed. 329699653d4eSeschrock */ 329799653d4eSeschrock static int 3298fa94a07fSbrendan spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3299fa94a07fSbrendan spa_aux_vdev_t *sav, const char *config, uint64_t version, 3300fa94a07fSbrendan vdev_labeltype_t label) 330199653d4eSeschrock { 3302fa94a07fSbrendan nvlist_t **dev; 3303fa94a07fSbrendan uint_t i, ndev; 330499653d4eSeschrock vdev_t *vd; 330599653d4eSeschrock int error; 330699653d4eSeschrock 3307e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3308e14bb325SJeff Bonwick 330999653d4eSeschrock /* 3310fa94a07fSbrendan * It's acceptable to have no devs specified. 331199653d4eSeschrock */ 3312fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 331399653d4eSeschrock return (0); 331499653d4eSeschrock 3315fa94a07fSbrendan if (ndev == 0) 3316be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 331799653d4eSeschrock 331899653d4eSeschrock /* 3319fa94a07fSbrendan * Make sure the pool is formatted with a version that supports this 3320fa94a07fSbrendan * device type. 332199653d4eSeschrock */ 3322fa94a07fSbrendan if (spa_version(spa) < version) 3323be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 332499653d4eSeschrock 332539c23413Seschrock /* 3326fa94a07fSbrendan * Set the pending device list so we correctly handle device in-use 332739c23413Seschrock * checking. 332839c23413Seschrock */ 3329fa94a07fSbrendan sav->sav_pending = dev; 3330fa94a07fSbrendan sav->sav_npending = ndev; 333139c23413Seschrock 3332fa94a07fSbrendan for (i = 0; i < ndev; i++) { 3333fa94a07fSbrendan if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 333499653d4eSeschrock mode)) != 0) 333539c23413Seschrock goto out; 333699653d4eSeschrock 333799653d4eSeschrock if (!vd->vdev_ops->vdev_op_leaf) { 333899653d4eSeschrock vdev_free(vd); 3339be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 334039c23413Seschrock goto out; 334199653d4eSeschrock } 334299653d4eSeschrock 3343fa94a07fSbrendan /* 3344e14bb325SJeff Bonwick * The L2ARC currently only supports disk devices in 3345e14bb325SJeff Bonwick * kernel context. For user-level testing, we allow it. 3346fa94a07fSbrendan */ 3347e14bb325SJeff Bonwick #ifdef _KERNEL 3348fa94a07fSbrendan if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3349fa94a07fSbrendan strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3350be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTBLK); 3351cd0837ccSGeorge Wilson vdev_free(vd); 3352fa94a07fSbrendan goto out; 3353fa94a07fSbrendan } 3354e14bb325SJeff Bonwick #endif 335599653d4eSeschrock vd->vdev_top = vd; 335699653d4eSeschrock 335739c23413Seschrock if ((error = vdev_open(vd)) == 0 && 3358fa94a07fSbrendan (error = vdev_label_init(vd, crtxg, label)) == 0) { 3359fa94a07fSbrendan VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 336039c23413Seschrock vd->vdev_guid) == 0); 336139c23413Seschrock } 336299653d4eSeschrock 336399653d4eSeschrock vdev_free(vd); 336439c23413Seschrock 3365fa94a07fSbrendan if (error && 3366fa94a07fSbrendan (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 336739c23413Seschrock goto out; 336839c23413Seschrock else 336939c23413Seschrock error = 0; 337099653d4eSeschrock } 337199653d4eSeschrock 337239c23413Seschrock out: 3373fa94a07fSbrendan sav->sav_pending = NULL; 3374fa94a07fSbrendan sav->sav_npending = 0; 337539c23413Seschrock return (error); 337699653d4eSeschrock } 337799653d4eSeschrock 3378fa94a07fSbrendan static int 3379fa94a07fSbrendan spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3380fa94a07fSbrendan { 3381fa94a07fSbrendan int error; 3382fa94a07fSbrendan 3383e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3384e14bb325SJeff Bonwick 3385fa94a07fSbrendan if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3386fa94a07fSbrendan &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3387fa94a07fSbrendan VDEV_LABEL_SPARE)) != 0) { 3388fa94a07fSbrendan return (error); 3389fa94a07fSbrendan } 3390fa94a07fSbrendan 3391fa94a07fSbrendan return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3392fa94a07fSbrendan &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3393fa94a07fSbrendan VDEV_LABEL_L2CACHE)); 3394fa94a07fSbrendan } 3395fa94a07fSbrendan 3396fa94a07fSbrendan static void 3397fa94a07fSbrendan spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3398fa94a07fSbrendan const char *config) 3399fa94a07fSbrendan { 3400fa94a07fSbrendan int i; 3401fa94a07fSbrendan 3402fa94a07fSbrendan if (sav->sav_config != NULL) { 3403fa94a07fSbrendan nvlist_t **olddevs; 3404fa94a07fSbrendan uint_t oldndevs; 3405fa94a07fSbrendan nvlist_t **newdevs; 3406fa94a07fSbrendan 3407fa94a07fSbrendan /* 3408fa94a07fSbrendan * Generate new dev list by concatentating with the 3409fa94a07fSbrendan * current dev list. 3410fa94a07fSbrendan */ 3411fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3412fa94a07fSbrendan &olddevs, &oldndevs) == 0); 3413fa94a07fSbrendan 3414fa94a07fSbrendan newdevs = kmem_alloc(sizeof (void *) * 3415fa94a07fSbrendan (ndevs + oldndevs), KM_SLEEP); 3416fa94a07fSbrendan for (i = 0; i < oldndevs; i++) 3417fa94a07fSbrendan VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3418fa94a07fSbrendan KM_SLEEP) == 0); 3419fa94a07fSbrendan for (i = 0; i < ndevs; i++) 3420fa94a07fSbrendan VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3421fa94a07fSbrendan KM_SLEEP) == 0); 3422fa94a07fSbrendan 3423fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, config, 3424fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 3425fa94a07fSbrendan 3426fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3427fa94a07fSbrendan config, newdevs, ndevs + oldndevs) == 0); 3428fa94a07fSbrendan for (i = 0; i < oldndevs + ndevs; i++) 3429fa94a07fSbrendan nvlist_free(newdevs[i]); 3430fa94a07fSbrendan kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3431fa94a07fSbrendan } else { 3432fa94a07fSbrendan /* 3433fa94a07fSbrendan * Generate a new dev list. 3434fa94a07fSbrendan */ 3435fa94a07fSbrendan VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3436fa94a07fSbrendan KM_SLEEP) == 0); 3437fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3438fa94a07fSbrendan devs, ndevs) == 0); 3439fa94a07fSbrendan } 3440fa94a07fSbrendan } 3441fa94a07fSbrendan 3442fa94a07fSbrendan /* 3443fa94a07fSbrendan * Stop and drop level 2 ARC devices 3444fa94a07fSbrendan */ 3445fa94a07fSbrendan void 3446fa94a07fSbrendan spa_l2cache_drop(spa_t *spa) 3447fa94a07fSbrendan { 3448fa94a07fSbrendan vdev_t *vd; 3449fa94a07fSbrendan int i; 3450fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 3451fa94a07fSbrendan 3452fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) { 3453fa94a07fSbrendan uint64_t pool; 3454fa94a07fSbrendan 3455fa94a07fSbrendan vd = sav->sav_vdevs[i]; 3456fa94a07fSbrendan ASSERT(vd != NULL); 3457fa94a07fSbrendan 34588ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 34598ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 3460fa94a07fSbrendan l2arc_remove_vdev(vd); 3461fa94a07fSbrendan } 3462fa94a07fSbrendan } 3463fa94a07fSbrendan 3464fa9e4066Sahrens /* 3465fa9e4066Sahrens * Pool Creation 3466fa9e4066Sahrens */ 3467fa9e4066Sahrens int 3468990b4856Slling spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 34694445fffbSMatthew Ahrens nvlist_t *zplprops) 3470fa9e4066Sahrens { 3471fa9e4066Sahrens spa_t *spa; 3472990b4856Slling char *altroot = NULL; 34730373e76bSbonwick vdev_t *rvd; 3474fa9e4066Sahrens dsl_pool_t *dp; 3475fa9e4066Sahrens dmu_tx_t *tx; 3476573ca77eSGeorge Wilson int error = 0; 3477fa9e4066Sahrens uint64_t txg = TXG_INITIAL; 3478fa94a07fSbrendan nvlist_t **spares, **l2cache; 3479fa94a07fSbrendan uint_t nspares, nl2cache; 3480cde58dbcSMatthew Ahrens uint64_t version, obj; 3481ad135b5dSChristopher Siden boolean_t has_features; 3482fa9e4066Sahrens 3483fa9e4066Sahrens /* 3484fa9e4066Sahrens * If this pool already exists, return failure. 3485fa9e4066Sahrens */ 3486fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 3487fa9e4066Sahrens if (spa_lookup(pool) != NULL) { 3488fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3489be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 3490fa9e4066Sahrens } 3491fa9e4066Sahrens 3492fa9e4066Sahrens /* 3493fa9e4066Sahrens * Allocate a new spa_t structure. 3494fa9e4066Sahrens */ 3495990b4856Slling (void) nvlist_lookup_string(props, 3496990b4856Slling zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3497468c413aSTim Haley spa = spa_add(pool, NULL, altroot); 34988ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 3499fa9e4066Sahrens 3500990b4856Slling if (props && (error = spa_prop_validate(spa, props))) { 3501990b4856Slling spa_deactivate(spa); 3502990b4856Slling spa_remove(spa); 3503c5904d13Seschrock mutex_exit(&spa_namespace_lock); 3504990b4856Slling return (error); 3505990b4856Slling } 3506990b4856Slling 3507ad135b5dSChristopher Siden has_features = B_FALSE; 3508ad135b5dSChristopher Siden for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3509ad135b5dSChristopher Siden elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3510ad135b5dSChristopher Siden if (zpool_prop_feature(nvpair_name(elem))) 3511ad135b5dSChristopher Siden has_features = B_TRUE; 3512ad135b5dSChristopher Siden } 3513ad135b5dSChristopher Siden 3514ad135b5dSChristopher Siden if (has_features || nvlist_lookup_uint64(props, 3515ad135b5dSChristopher Siden zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3516990b4856Slling version = SPA_VERSION; 3517ad135b5dSChristopher Siden } 3518ad135b5dSChristopher Siden ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3519b24ab676SJeff Bonwick 3520b24ab676SJeff Bonwick spa->spa_first_txg = txg; 3521b24ab676SJeff Bonwick spa->spa_uberblock.ub_txg = txg - 1; 3522990b4856Slling spa->spa_uberblock.ub_version = version; 3523fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 3524fa9e4066Sahrens 352554d692b7SGeorge Wilson /* 352654d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 352754d692b7SGeorge Wilson */ 35286f834bc1SMatthew Ahrens spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 35296f834bc1SMatthew Ahrens KM_SLEEP); 35306f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) { 35316f834bc1SMatthew Ahrens spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 35326f834bc1SMatthew Ahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 35336f834bc1SMatthew Ahrens ZIO_FLAG_GODFATHER); 35346f834bc1SMatthew Ahrens } 353554d692b7SGeorge Wilson 35360373e76bSbonwick /* 35370373e76bSbonwick * Create the root vdev. 35380373e76bSbonwick */ 3539e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 35400373e76bSbonwick 354199653d4eSeschrock error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 35420373e76bSbonwick 354399653d4eSeschrock ASSERT(error != 0 || rvd != NULL); 354499653d4eSeschrock ASSERT(error != 0 || spa->spa_root_vdev == rvd); 35450373e76bSbonwick 3546b7b97454Sperrin if (error == 0 && !zfs_allocatable_devs(nvroot)) 3547be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 354899653d4eSeschrock 354999653d4eSeschrock if (error == 0 && 355099653d4eSeschrock (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3551fa94a07fSbrendan (error = spa_validate_aux(spa, nvroot, txg, 355299653d4eSeschrock VDEV_ALLOC_ADD)) == 0) { 3553573ca77eSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 3554573ca77eSGeorge Wilson vdev_metaslab_set_size(rvd->vdev_child[c]); 3555573ca77eSGeorge Wilson vdev_expand(rvd->vdev_child[c], txg); 3556573ca77eSGeorge Wilson } 35570373e76bSbonwick } 35580373e76bSbonwick 3559e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3560fa9e4066Sahrens 356199653d4eSeschrock if (error != 0) { 3562fa9e4066Sahrens spa_unload(spa); 3563fa9e4066Sahrens spa_deactivate(spa); 3564fa9e4066Sahrens spa_remove(spa); 3565fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3566fa9e4066Sahrens return (error); 3567fa9e4066Sahrens } 3568fa9e4066Sahrens 356999653d4eSeschrock /* 357099653d4eSeschrock * Get the list of spares, if specified. 357199653d4eSeschrock */ 357299653d4eSeschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 357399653d4eSeschrock &spares, &nspares) == 0) { 3574fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 357599653d4eSeschrock KM_SLEEP) == 0); 3576fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 357799653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3578e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 357999653d4eSeschrock spa_load_spares(spa); 3580e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3581fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 3582fa94a07fSbrendan } 3583fa94a07fSbrendan 3584fa94a07fSbrendan /* 3585fa94a07fSbrendan * Get the list of level 2 cache devices, if specified. 3586fa94a07fSbrendan */ 3587fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3588fa94a07fSbrendan &l2cache, &nl2cache) == 0) { 3589fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3590fa94a07fSbrendan NV_UNIQUE_NAME, KM_SLEEP) == 0); 3591fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3592fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3593e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3594fa94a07fSbrendan spa_load_l2cache(spa); 3595e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3596fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 359799653d4eSeschrock } 359899653d4eSeschrock 3599ad135b5dSChristopher Siden spa->spa_is_initializing = B_TRUE; 36000a48a24eStimh spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3601fa9e4066Sahrens spa->spa_meta_objset = dp->dp_meta_objset; 3602ad135b5dSChristopher Siden spa->spa_is_initializing = B_FALSE; 3603fa9e4066Sahrens 3604485bbbf5SGeorge Wilson /* 3605485bbbf5SGeorge Wilson * Create DDTs (dedup tables). 3606485bbbf5SGeorge Wilson */ 3607485bbbf5SGeorge Wilson ddt_create(spa); 3608485bbbf5SGeorge Wilson 3609485bbbf5SGeorge Wilson spa_update_dspace(spa); 3610485bbbf5SGeorge Wilson 3611fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 3612fa9e4066Sahrens 3613fa9e4066Sahrens /* 3614fa9e4066Sahrens * Create the pool config object. 3615fa9e4066Sahrens */ 3616fa9e4066Sahrens spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3617f7991ba4STim Haley DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3618fa9e4066Sahrens DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3619fa9e4066Sahrens 3620ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 3621fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3622ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3623ea8dc4b6Seschrock cmn_err(CE_PANIC, "failed to add pool config"); 3624ea8dc4b6Seschrock } 3625fa9e4066Sahrens 3626ad135b5dSChristopher Siden if (spa_version(spa) >= SPA_VERSION_FEATURES) 3627ad135b5dSChristopher Siden spa_feature_create_zap_objects(spa, tx); 3628ad135b5dSChristopher Siden 36293f9d6ad7SLin Ling if (zap_add(spa->spa_meta_objset, 36303f9d6ad7SLin Ling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 36313f9d6ad7SLin Ling sizeof (uint64_t), 1, &version, tx) != 0) { 36323f9d6ad7SLin Ling cmn_err(CE_PANIC, "failed to add pool version"); 36333f9d6ad7SLin Ling } 36343f9d6ad7SLin Ling 3635990b4856Slling /* Newly created pools with the right version are always deflated. */ 3636990b4856Slling if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3637990b4856Slling spa->spa_deflate = TRUE; 3638990b4856Slling if (zap_add(spa->spa_meta_objset, 3639990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3640990b4856Slling sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3641990b4856Slling cmn_err(CE_PANIC, "failed to add deflate"); 3642990b4856Slling } 364399653d4eSeschrock } 364499653d4eSeschrock 3645fa9e4066Sahrens /* 3646cde58dbcSMatthew Ahrens * Create the deferred-free bpobj. Turn off compression 3647fa9e4066Sahrens * because sync-to-convergence takes longer if the blocksize 3648fa9e4066Sahrens * keeps changing. 3649fa9e4066Sahrens */ 3650cde58dbcSMatthew Ahrens obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3651cde58dbcSMatthew Ahrens dmu_object_set_compress(spa->spa_meta_objset, obj, 3652cde58dbcSMatthew Ahrens ZIO_COMPRESS_OFF, tx); 3653ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 3654cde58dbcSMatthew Ahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3655cde58dbcSMatthew Ahrens sizeof (uint64_t), 1, &obj, tx) != 0) { 3656cde58dbcSMatthew Ahrens cmn_err(CE_PANIC, "failed to add bpobj"); 3657ea8dc4b6Seschrock } 3658b420f3adSRichard Lowe VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3659cde58dbcSMatthew Ahrens spa->spa_meta_objset, obj)); 3660fa9e4066Sahrens 366106eeb2adSek /* 366206eeb2adSek * Create the pool's history object. 366306eeb2adSek */ 3664990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY) 3665990b4856Slling spa_history_create_obj(spa, tx); 3666990b4856Slling 3667990b4856Slling /* 3668990b4856Slling * Set pool properties. 3669990b4856Slling */ 3670990b4856Slling spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3671990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 36720a4e9518Sgw spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3673573ca77eSGeorge Wilson spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3674b24ab676SJeff Bonwick 3675379c004dSEric Schrock if (props != NULL) { 3676379c004dSEric Schrock spa_configfile_set(spa, props, B_FALSE); 36773b2aab18SMatthew Ahrens spa_sync_props(props, tx); 3678379c004dSEric Schrock } 367906eeb2adSek 3680fa9e4066Sahrens dmu_tx_commit(tx); 3681fa9e4066Sahrens 3682fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 3683fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 3684fa9e4066Sahrens 3685fa9e4066Sahrens /* 3686fa9e4066Sahrens * We explicitly wait for the first transaction to complete so that our 3687fa9e4066Sahrens * bean counters are appropriately updated. 3688fa9e4066Sahrens */ 3689fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 3690fa9e4066Sahrens 3691c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 3692*14372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3693fa9e4066Sahrens 36944445fffbSMatthew Ahrens spa_history_log_version(spa, "create"); 3695228975ccSek 3696bc9014e6SJustin Gibbs /* 3697bc9014e6SJustin Gibbs * Don't count references from objsets that are already closed 3698bc9014e6SJustin Gibbs * and are making their way through the eviction process. 3699bc9014e6SJustin Gibbs */ 3700bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 3701088f3894Sahrens spa->spa_minref = refcount_count(&spa->spa_refcount); 3702088f3894Sahrens 3703daaa36a7SGeorge Wilson mutex_exit(&spa_namespace_lock); 3704daaa36a7SGeorge Wilson 3705fa9e4066Sahrens return (0); 3706fa9e4066Sahrens } 3707fa9e4066Sahrens 3708e7cbe64fSgw #ifdef _KERNEL 3709e7cbe64fSgw /* 371021ecdf64SLin Ling * Get the root pool information from the root disk, then import the root pool 371121ecdf64SLin Ling * during the system boot up time. 3712e7cbe64fSgw */ 371321ecdf64SLin Ling extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 371421ecdf64SLin Ling 371521ecdf64SLin Ling static nvlist_t * 371621ecdf64SLin Ling spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3717e7cbe64fSgw { 371821ecdf64SLin Ling nvlist_t *config; 3719e7cbe64fSgw nvlist_t *nvtop, *nvroot; 3720e7cbe64fSgw uint64_t pgid; 3721e7cbe64fSgw 372221ecdf64SLin Ling if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 372321ecdf64SLin Ling return (NULL); 372421ecdf64SLin Ling 3725e7cbe64fSgw /* 3726e7cbe64fSgw * Add this top-level vdev to the child array. 3727e7cbe64fSgw */ 372821ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 372921ecdf64SLin Ling &nvtop) == 0); 373021ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 373121ecdf64SLin Ling &pgid) == 0); 373221ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3733e7cbe64fSgw 3734e7cbe64fSgw /* 3735e7cbe64fSgw * Put this pool's top-level vdevs into a root vdev. 3736e7cbe64fSgw */ 3737e7cbe64fSgw VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 373821ecdf64SLin Ling VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 373921ecdf64SLin Ling VDEV_TYPE_ROOT) == 0); 3740e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3741e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3742e7cbe64fSgw VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3743e7cbe64fSgw &nvtop, 1) == 0); 3744e7cbe64fSgw 3745e7cbe64fSgw /* 3746e7cbe64fSgw * Replace the existing vdev_tree with the new root vdev in 3747e7cbe64fSgw * this pool's configuration (remove the old, add the new). 3748e7cbe64fSgw */ 3749e7cbe64fSgw VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3750e7cbe64fSgw nvlist_free(nvroot); 375121ecdf64SLin Ling return (config); 3752e7cbe64fSgw } 3753e7cbe64fSgw 3754e7cbe64fSgw /* 375521ecdf64SLin Ling * Walk the vdev tree and see if we can find a device with "better" 375621ecdf64SLin Ling * configuration. A configuration is "better" if the label on that 375721ecdf64SLin Ling * device has a more recent txg. 3758051aabe6Staylor */ 375921ecdf64SLin Ling static void 376021ecdf64SLin Ling spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3761051aabe6Staylor { 3762573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 376321ecdf64SLin Ling spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3764051aabe6Staylor 376521ecdf64SLin Ling if (vd->vdev_ops->vdev_op_leaf) { 376621ecdf64SLin Ling nvlist_t *label; 376721ecdf64SLin Ling uint64_t label_txg; 3768051aabe6Staylor 376921ecdf64SLin Ling if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 377021ecdf64SLin Ling &label) != 0) 377121ecdf64SLin Ling return; 3772051aabe6Staylor 377321ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 377421ecdf64SLin Ling &label_txg) == 0); 3775051aabe6Staylor 377621ecdf64SLin Ling /* 377721ecdf64SLin Ling * Do we have a better boot device? 377821ecdf64SLin Ling */ 377921ecdf64SLin Ling if (label_txg > *txg) { 378021ecdf64SLin Ling *txg = label_txg; 378121ecdf64SLin Ling *avd = vd; 3782051aabe6Staylor } 378321ecdf64SLin Ling nvlist_free(label); 3784051aabe6Staylor } 3785051aabe6Staylor } 3786051aabe6Staylor 3787e7cbe64fSgw /* 3788e7cbe64fSgw * Import a root pool. 3789e7cbe64fSgw * 3790051aabe6Staylor * For x86. devpath_list will consist of devid and/or physpath name of 3791051aabe6Staylor * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3792051aabe6Staylor * The GRUB "findroot" command will return the vdev we should boot. 3793e7cbe64fSgw * 3794e7cbe64fSgw * For Sparc, devpath_list consists the physpath name of the booting device 3795e7cbe64fSgw * no matter the rootpool is a single device pool or a mirrored pool. 3796e7cbe64fSgw * e.g. 3797e7cbe64fSgw * "/pci@1f,0/ide@d/disk@0,0:a" 3798e7cbe64fSgw */ 3799e7cbe64fSgw int 3800051aabe6Staylor spa_import_rootpool(char *devpath, char *devid) 3801e7cbe64fSgw { 380221ecdf64SLin Ling spa_t *spa; 380321ecdf64SLin Ling vdev_t *rvd, *bvd, *avd = NULL; 380421ecdf64SLin Ling nvlist_t *config, *nvtop; 380521ecdf64SLin Ling uint64_t guid, txg; 3806e7cbe64fSgw char *pname; 3807e7cbe64fSgw int error; 3808e7cbe64fSgw 3809e7cbe64fSgw /* 381021ecdf64SLin Ling * Read the label from the boot device and generate a configuration. 3811e7cbe64fSgw */ 3812dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3813dedec472SJack Meng #if defined(_OBP) && defined(_KERNEL) 3814dedec472SJack Meng if (config == NULL) { 3815dedec472SJack Meng if (strstr(devpath, "/iscsi/ssd") != NULL) { 3816dedec472SJack Meng /* iscsi boot */ 3817dedec472SJack Meng get_iscsi_bootpath_phy(devpath); 3818dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3819dedec472SJack Meng } 3820dedec472SJack Meng } 3821dedec472SJack Meng #endif 3822dedec472SJack Meng if (config == NULL) { 3823ad135b5dSChristopher Siden cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 382421ecdf64SLin Ling devpath); 3825be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 382621ecdf64SLin Ling } 3827e7cbe64fSgw 382821ecdf64SLin Ling VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 382921ecdf64SLin Ling &pname) == 0); 383021ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3831e7cbe64fSgw 38326809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 38336809eb4eSEric Schrock if ((spa = spa_lookup(pname)) != NULL) { 38346809eb4eSEric Schrock /* 38356809eb4eSEric Schrock * Remove the existing root pool from the namespace so that we 38366809eb4eSEric Schrock * can replace it with the correct config we just read in. 38376809eb4eSEric Schrock */ 38386809eb4eSEric Schrock spa_remove(spa); 38396809eb4eSEric Schrock } 38406809eb4eSEric Schrock 3841468c413aSTim Haley spa = spa_add(pname, config, NULL); 38426809eb4eSEric Schrock spa->spa_is_root = B_TRUE; 38434b964adaSGeorge Wilson spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3844e7cbe64fSgw 384521ecdf64SLin Ling /* 384621ecdf64SLin Ling * Build up a vdev tree based on the boot device's label config. 384721ecdf64SLin Ling */ 384821ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 384921ecdf64SLin Ling &nvtop) == 0); 385021ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 385121ecdf64SLin Ling error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 385221ecdf64SLin Ling VDEV_ALLOC_ROOTPOOL); 385321ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 385421ecdf64SLin Ling if (error) { 385521ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 385621ecdf64SLin Ling nvlist_free(config); 385721ecdf64SLin Ling cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 385821ecdf64SLin Ling pname); 385921ecdf64SLin Ling return (error); 386021ecdf64SLin Ling } 386121ecdf64SLin Ling 386221ecdf64SLin Ling /* 386321ecdf64SLin Ling * Get the boot vdev. 386421ecdf64SLin Ling */ 386521ecdf64SLin Ling if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 386621ecdf64SLin Ling cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 386721ecdf64SLin Ling (u_longlong_t)guid); 3868be6fd75aSMatthew Ahrens error = SET_ERROR(ENOENT); 386921ecdf64SLin Ling goto out; 387021ecdf64SLin Ling } 3871e7cbe64fSgw 387221ecdf64SLin Ling /* 387321ecdf64SLin Ling * Determine if there is a better boot device. 387421ecdf64SLin Ling */ 387521ecdf64SLin Ling avd = bvd; 387621ecdf64SLin Ling spa_alt_rootvdev(rvd, &avd, &txg); 387721ecdf64SLin Ling if (avd != bvd) { 387821ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 387921ecdf64SLin Ling "try booting from '%s'", avd->vdev_path); 3880be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 388121ecdf64SLin Ling goto out; 388221ecdf64SLin Ling } 3883e7cbe64fSgw 388421ecdf64SLin Ling /* 388521ecdf64SLin Ling * If the boot device is part of a spare vdev then ensure that 388621ecdf64SLin Ling * we're booting off the active spare. 388721ecdf64SLin Ling */ 388821ecdf64SLin Ling if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 388921ecdf64SLin Ling !bvd->vdev_isspare) { 389021ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is currently spared. Please " 389121ecdf64SLin Ling "try booting from '%s'", 3892cb04b873SMark J Musante bvd->vdev_parent-> 3893cb04b873SMark J Musante vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3894be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 389521ecdf64SLin Ling goto out; 389621ecdf64SLin Ling } 389721ecdf64SLin Ling 389821ecdf64SLin Ling error = 0; 389921ecdf64SLin Ling out: 390021ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 390121ecdf64SLin Ling vdev_free(rvd); 390221ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 390321ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 390421ecdf64SLin Ling 390521ecdf64SLin Ling nvlist_free(config); 3906e7cbe64fSgw return (error); 3907e7cbe64fSgw } 390821ecdf64SLin Ling 3909e7cbe64fSgw #endif 3910e7cbe64fSgw 39116809eb4eSEric Schrock /* 39126809eb4eSEric Schrock * Import a non-root pool into the system. 39136809eb4eSEric Schrock */ 3914c5904d13Seschrock int 39154b964adaSGeorge Wilson spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3916c5904d13Seschrock { 39176809eb4eSEric Schrock spa_t *spa; 39186809eb4eSEric Schrock char *altroot = NULL; 3919468c413aSTim Haley spa_load_state_t state = SPA_LOAD_IMPORT; 3920468c413aSTim Haley zpool_rewind_policy_t policy; 3921f9af39baSGeorge Wilson uint64_t mode = spa_mode_global; 3922f9af39baSGeorge Wilson uint64_t readonly = B_FALSE; 39236809eb4eSEric Schrock int error; 39246809eb4eSEric Schrock nvlist_t *nvroot; 39256809eb4eSEric Schrock nvlist_t **spares, **l2cache; 39266809eb4eSEric Schrock uint_t nspares, nl2cache; 39276809eb4eSEric Schrock 39286809eb4eSEric Schrock /* 39296809eb4eSEric Schrock * If a pool with this name exists, return failure. 39306809eb4eSEric Schrock */ 39316809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 39321195e687SMark J Musante if (spa_lookup(pool) != NULL) { 39336809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 3934be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 39356809eb4eSEric Schrock } 39366809eb4eSEric Schrock 39376809eb4eSEric Schrock /* 39386809eb4eSEric Schrock * Create and initialize the spa structure. 39396809eb4eSEric Schrock */ 39406809eb4eSEric Schrock (void) nvlist_lookup_string(props, 39416809eb4eSEric Schrock zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3942f9af39baSGeorge Wilson (void) nvlist_lookup_uint64(props, 3943f9af39baSGeorge Wilson zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 3944f9af39baSGeorge Wilson if (readonly) 3945f9af39baSGeorge Wilson mode = FREAD; 3946468c413aSTim Haley spa = spa_add(pool, config, altroot); 39474b964adaSGeorge Wilson spa->spa_import_flags = flags; 39484b964adaSGeorge Wilson 39494b964adaSGeorge Wilson /* 39504b964adaSGeorge Wilson * Verbatim import - Take a pool and insert it into the namespace 39514b964adaSGeorge Wilson * as if it had been loaded at boot. 39524b964adaSGeorge Wilson */ 39534b964adaSGeorge Wilson if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 39544b964adaSGeorge Wilson if (props != NULL) 39554b964adaSGeorge Wilson spa_configfile_set(spa, props, B_FALSE); 39564b964adaSGeorge Wilson 39574b964adaSGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 3958*14372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 39594b964adaSGeorge Wilson 39604b964adaSGeorge Wilson mutex_exit(&spa_namespace_lock); 39614b964adaSGeorge Wilson return (0); 39624b964adaSGeorge Wilson } 39634b964adaSGeorge Wilson 3964f9af39baSGeorge Wilson spa_activate(spa, mode); 39656809eb4eSEric Schrock 396625f89ee2SJeff Bonwick /* 396725f89ee2SJeff Bonwick * Don't start async tasks until we know everything is healthy. 396825f89ee2SJeff Bonwick */ 396925f89ee2SJeff Bonwick spa_async_suspend(spa); 397025f89ee2SJeff Bonwick 39714b964adaSGeorge Wilson zpool_get_rewind_policy(config, &policy); 39724b964adaSGeorge Wilson if (policy.zrp_request & ZPOOL_DO_REWIND) 39734b964adaSGeorge Wilson state = SPA_LOAD_RECOVER; 39744b964adaSGeorge Wilson 39756809eb4eSEric Schrock /* 39766809eb4eSEric Schrock * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 39776809eb4eSEric Schrock * because the user-supplied config is actually the one to trust when 39786809eb4eSEric Schrock * doing an import. 39796809eb4eSEric Schrock */ 3980468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 3981468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 39824b964adaSGeorge Wilson 3983468c413aSTim Haley error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 3984c8ee1847SVictor Latushkin policy.zrp_request); 3985468c413aSTim Haley 3986468c413aSTim Haley /* 39874b964adaSGeorge Wilson * Propagate anything learned while loading the pool and pass it 39884b964adaSGeorge Wilson * back to caller (i.e. rewind info, missing devices, etc). 3989468c413aSTim Haley */ 39904b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 39914b964adaSGeorge Wilson spa->spa_load_info) == 0); 39926809eb4eSEric Schrock 39936809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 39946809eb4eSEric Schrock /* 39956809eb4eSEric Schrock * Toss any existing sparelist, as it doesn't have any validity 39966809eb4eSEric Schrock * anymore, and conflicts with spa_has_spare(). 39976809eb4eSEric Schrock */ 39986809eb4eSEric Schrock if (spa->spa_spares.sav_config) { 39996809eb4eSEric Schrock nvlist_free(spa->spa_spares.sav_config); 40006809eb4eSEric Schrock spa->spa_spares.sav_config = NULL; 40016809eb4eSEric Schrock spa_load_spares(spa); 40026809eb4eSEric Schrock } 40036809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) { 40046809eb4eSEric Schrock nvlist_free(spa->spa_l2cache.sav_config); 40056809eb4eSEric Schrock spa->spa_l2cache.sav_config = NULL; 40066809eb4eSEric Schrock spa_load_l2cache(spa); 40076809eb4eSEric Schrock } 40086809eb4eSEric Schrock 40096809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 40106809eb4eSEric Schrock &nvroot) == 0); 40116809eb4eSEric Schrock if (error == 0) 40126809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 40136809eb4eSEric Schrock VDEV_ALLOC_SPARE); 40146809eb4eSEric Schrock if (error == 0) 40156809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 40166809eb4eSEric Schrock VDEV_ALLOC_L2CACHE); 40176809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 40186809eb4eSEric Schrock 40196809eb4eSEric Schrock if (props != NULL) 40206809eb4eSEric Schrock spa_configfile_set(spa, props, B_FALSE); 40216809eb4eSEric Schrock 40226809eb4eSEric Schrock if (error != 0 || (props && spa_writeable(spa) && 40236809eb4eSEric Schrock (error = spa_prop_set(spa, props)))) { 40246809eb4eSEric Schrock spa_unload(spa); 40256809eb4eSEric Schrock spa_deactivate(spa); 40266809eb4eSEric Schrock spa_remove(spa); 40276809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 40286809eb4eSEric Schrock return (error); 40296809eb4eSEric Schrock } 40306809eb4eSEric Schrock 4031955ef359SLin Ling spa_async_resume(spa); 4032955ef359SLin Ling 40336809eb4eSEric Schrock /* 40346809eb4eSEric Schrock * Override any spares and level 2 cache devices as specified by 40356809eb4eSEric Schrock * the user, as these may have correct device names/devids, etc. 40366809eb4eSEric Schrock */ 40376809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 40386809eb4eSEric Schrock &spares, &nspares) == 0) { 40396809eb4eSEric Schrock if (spa->spa_spares.sav_config) 40406809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_spares.sav_config, 40416809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 40426809eb4eSEric Schrock else 40436809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 40446809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 40456809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 40466809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 40476809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 40486809eb4eSEric Schrock spa_load_spares(spa); 40496809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 40506809eb4eSEric Schrock spa->spa_spares.sav_sync = B_TRUE; 40516809eb4eSEric Schrock } 40526809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 40536809eb4eSEric Schrock &l2cache, &nl2cache) == 0) { 40546809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) 40556809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 40566809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 40576809eb4eSEric Schrock else 40586809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 40596809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 40606809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 40616809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 40626809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 40636809eb4eSEric Schrock spa_load_l2cache(spa); 40646809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 40656809eb4eSEric Schrock spa->spa_l2cache.sav_sync = B_TRUE; 40666809eb4eSEric Schrock } 40676809eb4eSEric Schrock 4068b693757aSEric Schrock /* 4069b693757aSEric Schrock * Check for any removed devices. 4070b693757aSEric Schrock */ 4071b693757aSEric Schrock if (spa->spa_autoreplace) { 4072b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 4073b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 4074b693757aSEric Schrock } 4075b693757aSEric Schrock 40766809eb4eSEric Schrock if (spa_writeable(spa)) { 40776809eb4eSEric Schrock /* 40786809eb4eSEric Schrock * Update the config cache to include the newly-imported pool. 40796809eb4eSEric Schrock */ 4080bc758434SLin Ling spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 40816809eb4eSEric Schrock } 40826809eb4eSEric Schrock 4083573ca77eSGeorge Wilson /* 4084573ca77eSGeorge Wilson * It's possible that the pool was expanded while it was exported. 4085573ca77eSGeorge Wilson * We kick off an async task to handle this for us. 4086573ca77eSGeorge Wilson */ 4087573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4088573ca77eSGeorge Wilson 40894445fffbSMatthew Ahrens spa_history_log_version(spa, "import"); 40906809eb4eSEric Schrock 4091*14372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4092*14372834SHans Rosenfeld 4093*14372834SHans Rosenfeld mutex_exit(&spa_namespace_lock); 4094*14372834SHans Rosenfeld 40956809eb4eSEric Schrock return (0); 4096c5904d13Seschrock } 4097c5904d13Seschrock 4098fa9e4066Sahrens nvlist_t * 4099fa9e4066Sahrens spa_tryimport(nvlist_t *tryconfig) 4100fa9e4066Sahrens { 4101fa9e4066Sahrens nvlist_t *config = NULL; 4102fa9e4066Sahrens char *poolname; 4103fa9e4066Sahrens spa_t *spa; 4104fa9e4066Sahrens uint64_t state; 41057b7154beSLin Ling int error; 4106fa9e4066Sahrens 4107fa9e4066Sahrens if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4108fa9e4066Sahrens return (NULL); 4109fa9e4066Sahrens 4110fa9e4066Sahrens if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4111fa9e4066Sahrens return (NULL); 4112fa9e4066Sahrens 4113fa9e4066Sahrens /* 41140373e76bSbonwick * Create and initialize the spa structure. 4115fa9e4066Sahrens */ 41160373e76bSbonwick mutex_enter(&spa_namespace_lock); 4117468c413aSTim Haley spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 41188ad4d6ddSJeff Bonwick spa_activate(spa, FREAD); 4119fa9e4066Sahrens 4120fa9e4066Sahrens /* 41210373e76bSbonwick * Pass off the heavy lifting to spa_load(). 4122ecc2d604Sbonwick * Pass TRUE for mosconfig because the user-supplied config 4123ecc2d604Sbonwick * is actually the one to trust when doing an import. 4124fa9e4066Sahrens */ 41251195e687SMark J Musante error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4126fa9e4066Sahrens 4127fa9e4066Sahrens /* 4128fa9e4066Sahrens * If 'tryconfig' was at least parsable, return the current config. 4129fa9e4066Sahrens */ 4130fa9e4066Sahrens if (spa->spa_root_vdev != NULL) { 4131fa9e4066Sahrens config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4132fa9e4066Sahrens VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4133fa9e4066Sahrens poolname) == 0); 4134fa9e4066Sahrens VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4135fa9e4066Sahrens state) == 0); 413695173954Sek VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 413795173954Sek spa->spa_uberblock.ub_timestamp) == 0); 4138ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4139ad135b5dSChristopher Siden spa->spa_load_info) == 0); 414099653d4eSeschrock 4141e7cbe64fSgw /* 4142e7cbe64fSgw * If the bootfs property exists on this pool then we 4143e7cbe64fSgw * copy it out so that external consumers can tell which 4144e7cbe64fSgw * pools are bootable. 4145e7cbe64fSgw */ 41467b7154beSLin Ling if ((!error || error == EEXIST) && spa->spa_bootfs) { 4147e7cbe64fSgw char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4148e7cbe64fSgw 4149e7cbe64fSgw /* 4150e7cbe64fSgw * We have to play games with the name since the 4151e7cbe64fSgw * pool was opened as TRYIMPORT_NAME. 4152e7cbe64fSgw */ 4153e14bb325SJeff Bonwick if (dsl_dsobj_to_dsname(spa_name(spa), 4154e7cbe64fSgw spa->spa_bootfs, tmpname) == 0) { 4155e7cbe64fSgw char *cp; 4156e7cbe64fSgw char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4157e7cbe64fSgw 4158e7cbe64fSgw cp = strchr(tmpname, '/'); 4159e7cbe64fSgw if (cp == NULL) { 4160e7cbe64fSgw (void) strlcpy(dsname, tmpname, 4161e7cbe64fSgw MAXPATHLEN); 4162e7cbe64fSgw } else { 4163e7cbe64fSgw (void) snprintf(dsname, MAXPATHLEN, 4164e7cbe64fSgw "%s/%s", poolname, ++cp); 4165e7cbe64fSgw } 4166e7cbe64fSgw VERIFY(nvlist_add_string(config, 4167e7cbe64fSgw ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4168e7cbe64fSgw kmem_free(dsname, MAXPATHLEN); 4169e7cbe64fSgw } 4170e7cbe64fSgw kmem_free(tmpname, MAXPATHLEN); 4171e7cbe64fSgw } 4172e7cbe64fSgw 417399653d4eSeschrock /* 4174fa94a07fSbrendan * Add the list of hot spares and level 2 cache devices. 417599653d4eSeschrock */ 41766809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 417799653d4eSeschrock spa_add_spares(spa, config); 4178fa94a07fSbrendan spa_add_l2cache(spa, config); 41796809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 4180fa9e4066Sahrens } 4181fa9e4066Sahrens 4182fa9e4066Sahrens spa_unload(spa); 4183fa9e4066Sahrens spa_deactivate(spa); 4184fa9e4066Sahrens spa_remove(spa); 4185fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4186fa9e4066Sahrens 4187fa9e4066Sahrens return (config); 4188fa9e4066Sahrens } 4189fa9e4066Sahrens 4190fa9e4066Sahrens /* 4191fa9e4066Sahrens * Pool export/destroy 4192fa9e4066Sahrens * 4193fa9e4066Sahrens * The act of destroying or exporting a pool is very simple. We make sure there 4194fa9e4066Sahrens * is no more pending I/O and any references to the pool are gone. Then, we 4195fa9e4066Sahrens * update the pool state and sync all the labels to disk, removing the 4196394ab0cbSGeorge Wilson * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4197394ab0cbSGeorge Wilson * we don't sync the labels or remove the configuration cache. 4198fa9e4066Sahrens */ 4199fa9e4066Sahrens static int 420089a89ebfSlling spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4201394ab0cbSGeorge Wilson boolean_t force, boolean_t hardforce) 4202fa9e4066Sahrens { 4203fa9e4066Sahrens spa_t *spa; 4204fa9e4066Sahrens 420544cd46caSbillm if (oldconfig) 420644cd46caSbillm *oldconfig = NULL; 420744cd46caSbillm 42088ad4d6ddSJeff Bonwick if (!(spa_mode_global & FWRITE)) 4209be6fd75aSMatthew Ahrens return (SET_ERROR(EROFS)); 4210fa9e4066Sahrens 4211fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4212fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 4213fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4214be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 4215fa9e4066Sahrens } 4216fa9e4066Sahrens 4217ea8dc4b6Seschrock /* 4218ea8dc4b6Seschrock * Put a hold on the pool, drop the namespace lock, stop async tasks, 4219ea8dc4b6Seschrock * reacquire the namespace lock, and see if we can export. 4220ea8dc4b6Seschrock */ 4221ea8dc4b6Seschrock spa_open_ref(spa, FTAG); 4222ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 4223ea8dc4b6Seschrock spa_async_suspend(spa); 4224ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 4225ea8dc4b6Seschrock spa_close(spa, FTAG); 4226ea8dc4b6Seschrock 4227fa9e4066Sahrens /* 4228fa9e4066Sahrens * The pool will be in core if it's openable, 4229fa9e4066Sahrens * in which case we can modify its state. 4230fa9e4066Sahrens */ 4231fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4232fa9e4066Sahrens /* 4233fa9e4066Sahrens * Objsets may be open only because they're dirty, so we 4234fa9e4066Sahrens * have to force it to sync before checking spa_refcnt. 4235fa9e4066Sahrens */ 4236fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, 0); 4237bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 4238fa9e4066Sahrens 4239ea8dc4b6Seschrock /* 4240ea8dc4b6Seschrock * A pool cannot be exported or destroyed if there are active 4241ea8dc4b6Seschrock * references. If we are resetting a pool, allow references by 4242ea8dc4b6Seschrock * fault injection handlers. 4243ea8dc4b6Seschrock */ 4244ea8dc4b6Seschrock if (!spa_refcount_zero(spa) || 4245ea8dc4b6Seschrock (spa->spa_inject_ref != 0 && 4246ea8dc4b6Seschrock new_state != POOL_STATE_UNINITIALIZED)) { 4247ea8dc4b6Seschrock spa_async_resume(spa); 4248fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4249be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 4250fa9e4066Sahrens } 4251fa9e4066Sahrens 425289a89ebfSlling /* 425389a89ebfSlling * A pool cannot be exported if it has an active shared spare. 425489a89ebfSlling * This is to prevent other pools stealing the active spare 425589a89ebfSlling * from an exported pool. At user's own will, such pool can 425689a89ebfSlling * be forcedly exported. 425789a89ebfSlling */ 425889a89ebfSlling if (!force && new_state == POOL_STATE_EXPORTED && 425989a89ebfSlling spa_has_active_shared_spare(spa)) { 426089a89ebfSlling spa_async_resume(spa); 426189a89ebfSlling mutex_exit(&spa_namespace_lock); 4262be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 426389a89ebfSlling } 426489a89ebfSlling 4265fa9e4066Sahrens /* 4266fa9e4066Sahrens * We want this to be reflected on every label, 4267fa9e4066Sahrens * so mark them all dirty. spa_unload() will do the 4268fa9e4066Sahrens * final sync that pushes these changes out. 4269fa9e4066Sahrens */ 4270394ab0cbSGeorge Wilson if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4271e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4272ea8dc4b6Seschrock spa->spa_state = new_state; 42733f9d6ad7SLin Ling spa->spa_final_txg = spa_last_synced_txg(spa) + 42743f9d6ad7SLin Ling TXG_DEFER_SIZE + 1; 4275ea8dc4b6Seschrock vdev_config_dirty(spa->spa_root_vdev); 4276e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 4277ea8dc4b6Seschrock } 4278fa9e4066Sahrens } 4279fa9e4066Sahrens 42803d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 42813d7072f8Seschrock 4282fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4283fa9e4066Sahrens spa_unload(spa); 4284fa9e4066Sahrens spa_deactivate(spa); 4285fa9e4066Sahrens } 4286fa9e4066Sahrens 428744cd46caSbillm if (oldconfig && spa->spa_config) 428844cd46caSbillm VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 428944cd46caSbillm 4290ea8dc4b6Seschrock if (new_state != POOL_STATE_UNINITIALIZED) { 4291394ab0cbSGeorge Wilson if (!hardforce) 4292394ab0cbSGeorge Wilson spa_config_sync(spa, B_TRUE, B_TRUE); 4293ea8dc4b6Seschrock spa_remove(spa); 4294ea8dc4b6Seschrock } 4295fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4296fa9e4066Sahrens 4297fa9e4066Sahrens return (0); 4298fa9e4066Sahrens } 4299fa9e4066Sahrens 4300fa9e4066Sahrens /* 4301fa9e4066Sahrens * Destroy a storage pool. 4302fa9e4066Sahrens */ 4303fa9e4066Sahrens int 4304fa9e4066Sahrens spa_destroy(char *pool) 4305fa9e4066Sahrens { 4306394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4307394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 4308fa9e4066Sahrens } 4309fa9e4066Sahrens 4310fa9e4066Sahrens /* 4311fa9e4066Sahrens * Export a storage pool. 4312fa9e4066Sahrens */ 4313fa9e4066Sahrens int 4314394ab0cbSGeorge Wilson spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4315394ab0cbSGeorge Wilson boolean_t hardforce) 4316fa9e4066Sahrens { 4317394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4318394ab0cbSGeorge Wilson force, hardforce)); 4319fa9e4066Sahrens } 4320fa9e4066Sahrens 4321ea8dc4b6Seschrock /* 4322ea8dc4b6Seschrock * Similar to spa_export(), this unloads the spa_t without actually removing it 4323ea8dc4b6Seschrock * from the namespace in any way. 4324ea8dc4b6Seschrock */ 4325ea8dc4b6Seschrock int 4326ea8dc4b6Seschrock spa_reset(char *pool) 4327ea8dc4b6Seschrock { 432889a89ebfSlling return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4329394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 4330ea8dc4b6Seschrock } 4331ea8dc4b6Seschrock 4332fa9e4066Sahrens /* 4333fa9e4066Sahrens * ========================================================================== 4334fa9e4066Sahrens * Device manipulation 4335fa9e4066Sahrens * ========================================================================== 4336fa9e4066Sahrens */ 4337fa9e4066Sahrens 4338fa9e4066Sahrens /* 43398654d025Sperrin * Add a device to a storage pool. 4340fa9e4066Sahrens */ 4341fa9e4066Sahrens int 4342fa9e4066Sahrens spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4343fa9e4066Sahrens { 434488ecc943SGeorge Wilson uint64_t txg, id; 43458ad4d6ddSJeff Bonwick int error; 4346fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 43470e34b6a7Sbonwick vdev_t *vd, *tvd; 4348fa94a07fSbrendan nvlist_t **spares, **l2cache; 4349fa94a07fSbrendan uint_t nspares, nl2cache; 4350fa9e4066Sahrens 4351f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4352f9af39baSGeorge Wilson 4353fa9e4066Sahrens txg = spa_vdev_enter(spa); 4354fa9e4066Sahrens 435599653d4eSeschrock if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 435699653d4eSeschrock VDEV_ALLOC_ADD)) != 0) 435799653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, error)); 4358fa9e4066Sahrens 4359e14bb325SJeff Bonwick spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 436099653d4eSeschrock 4361fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4362fa94a07fSbrendan &nspares) != 0) 436399653d4eSeschrock nspares = 0; 436499653d4eSeschrock 4365fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4366fa94a07fSbrendan &nl2cache) != 0) 4367fa94a07fSbrendan nl2cache = 0; 4368fa94a07fSbrendan 4369e14bb325SJeff Bonwick if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4370fa9e4066Sahrens return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4371fa9e4066Sahrens 4372e14bb325SJeff Bonwick if (vd->vdev_children != 0 && 4373e14bb325SJeff Bonwick (error = vdev_create(vd, txg, B_FALSE)) != 0) 4374e14bb325SJeff Bonwick return (spa_vdev_exit(spa, vd, txg, error)); 437599653d4eSeschrock 437639c23413Seschrock /* 4377fa94a07fSbrendan * We must validate the spares and l2cache devices after checking the 4378fa94a07fSbrendan * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 437939c23413Seschrock */ 4380e14bb325SJeff Bonwick if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 438139c23413Seschrock return (spa_vdev_exit(spa, vd, txg, error)); 438239c23413Seschrock 438339c23413Seschrock /* 438439c23413Seschrock * Transfer each new top-level vdev from vd to rvd. 438539c23413Seschrock */ 43868ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 438788ecc943SGeorge Wilson 438888ecc943SGeorge Wilson /* 438988ecc943SGeorge Wilson * Set the vdev id to the first hole, if one exists. 439088ecc943SGeorge Wilson */ 439188ecc943SGeorge Wilson for (id = 0; id < rvd->vdev_children; id++) { 439288ecc943SGeorge Wilson if (rvd->vdev_child[id]->vdev_ishole) { 439388ecc943SGeorge Wilson vdev_free(rvd->vdev_child[id]); 439488ecc943SGeorge Wilson break; 439588ecc943SGeorge Wilson } 439688ecc943SGeorge Wilson } 439739c23413Seschrock tvd = vd->vdev_child[c]; 439839c23413Seschrock vdev_remove_child(vd, tvd); 439988ecc943SGeorge Wilson tvd->vdev_id = id; 440039c23413Seschrock vdev_add_child(rvd, tvd); 440139c23413Seschrock vdev_config_dirty(tvd); 440239c23413Seschrock } 440339c23413Seschrock 440499653d4eSeschrock if (nspares != 0) { 4405fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4406fa94a07fSbrendan ZPOOL_CONFIG_SPARES); 440799653d4eSeschrock spa_load_spares(spa); 4408fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 4409fa94a07fSbrendan } 4410fa94a07fSbrendan 4411fa94a07fSbrendan if (nl2cache != 0) { 4412fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4413fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE); 4414fa94a07fSbrendan spa_load_l2cache(spa); 4415fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 4416fa9e4066Sahrens } 4417fa9e4066Sahrens 4418fa9e4066Sahrens /* 44190e34b6a7Sbonwick * We have to be careful when adding new vdevs to an existing pool. 44200e34b6a7Sbonwick * If other threads start allocating from these vdevs before we 44210e34b6a7Sbonwick * sync the config cache, and we lose power, then upon reboot we may 44220e34b6a7Sbonwick * fail to open the pool because there are DVAs that the config cache 44230e34b6a7Sbonwick * can't translate. Therefore, we first add the vdevs without 44240e34b6a7Sbonwick * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 44250373e76bSbonwick * and then let spa_config_update() initialize the new metaslabs. 44260e34b6a7Sbonwick * 44270e34b6a7Sbonwick * spa_load() checks for added-but-not-initialized vdevs, so that 44280e34b6a7Sbonwick * if we lose power at any point in this sequence, the remaining 44290e34b6a7Sbonwick * steps will be completed the next time we load the pool. 44300e34b6a7Sbonwick */ 44310373e76bSbonwick (void) spa_vdev_exit(spa, vd, txg, 0); 44320e34b6a7Sbonwick 44330373e76bSbonwick mutex_enter(&spa_namespace_lock); 44340373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4435*14372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 44360373e76bSbonwick mutex_exit(&spa_namespace_lock); 4437fa9e4066Sahrens 44380373e76bSbonwick return (0); 4439fa9e4066Sahrens } 4440fa9e4066Sahrens 4441fa9e4066Sahrens /* 4442fa9e4066Sahrens * Attach a device to a mirror. The arguments are the path to any device 4443fa9e4066Sahrens * in the mirror, and the nvroot for the new device. If the path specifies 4444fa9e4066Sahrens * a device that is not mirrored, we automatically insert the mirror vdev. 4445fa9e4066Sahrens * 4446fa9e4066Sahrens * If 'replacing' is specified, the new device is intended to replace the 4447fa9e4066Sahrens * existing device; in this case the two devices are made into their own 44483d7072f8Seschrock * mirror using the 'replacing' vdev, which is functionally identical to 4449fa9e4066Sahrens * the mirror vdev (it actually reuses all the same ops) but has a few 4450fa9e4066Sahrens * extra rules: you can't attach to it after it's been created, and upon 4451fa9e4066Sahrens * completion of resilvering, the first disk (the one being replaced) 4452fa9e4066Sahrens * is automatically detached. 4453fa9e4066Sahrens */ 4454fa9e4066Sahrens int 4455ea8dc4b6Seschrock spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4456fa9e4066Sahrens { 44573f9d6ad7SLin Ling uint64_t txg, dtl_max_txg; 4458fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4459fa9e4066Sahrens vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 446099653d4eSeschrock vdev_ops_t *pvops; 44619b3f6b42SEric Kustarz char *oldvdpath, *newvdpath; 44629b3f6b42SEric Kustarz int newvd_isspare; 44639b3f6b42SEric Kustarz int error; 4464fa9e4066Sahrens 4465f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4466f9af39baSGeorge Wilson 4467fa9e4066Sahrens txg = spa_vdev_enter(spa); 4468fa9e4066Sahrens 4469c5904d13Seschrock oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4470fa9e4066Sahrens 4471fa9e4066Sahrens if (oldvd == NULL) 4472fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4473fa9e4066Sahrens 44740e34b6a7Sbonwick if (!oldvd->vdev_ops->vdev_op_leaf) 44750e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 44760e34b6a7Sbonwick 4477fa9e4066Sahrens pvd = oldvd->vdev_parent; 4478fa9e4066Sahrens 447999653d4eSeschrock if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4480cd0837ccSGeorge Wilson VDEV_ALLOC_ATTACH)) != 0) 44813d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 44823d7072f8Seschrock 44833d7072f8Seschrock if (newrootvd->vdev_children != 1) 4484fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4485fa9e4066Sahrens 4486fa9e4066Sahrens newvd = newrootvd->vdev_child[0]; 4487fa9e4066Sahrens 4488fa9e4066Sahrens if (!newvd->vdev_ops->vdev_op_leaf) 4489fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4490fa9e4066Sahrens 449199653d4eSeschrock if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4492fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, error)); 4493fa9e4066Sahrens 44948654d025Sperrin /* 44958654d025Sperrin * Spares can't replace logs 44968654d025Sperrin */ 4497ee0eb9f2SEric Schrock if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 44988654d025Sperrin return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 44998654d025Sperrin 450099653d4eSeschrock if (!replacing) { 450199653d4eSeschrock /* 450299653d4eSeschrock * For attach, the only allowable parent is a mirror or the root 450399653d4eSeschrock * vdev. 450499653d4eSeschrock */ 450599653d4eSeschrock if (pvd->vdev_ops != &vdev_mirror_ops && 450699653d4eSeschrock pvd->vdev_ops != &vdev_root_ops) 450799653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 450899653d4eSeschrock 450999653d4eSeschrock pvops = &vdev_mirror_ops; 451099653d4eSeschrock } else { 451199653d4eSeschrock /* 451299653d4eSeschrock * Active hot spares can only be replaced by inactive hot 451399653d4eSeschrock * spares. 451499653d4eSeschrock */ 451599653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 4516cb04b873SMark J Musante oldvd->vdev_isspare && 451799653d4eSeschrock !spa_has_spare(spa, newvd->vdev_guid)) 451899653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 451999653d4eSeschrock 452099653d4eSeschrock /* 452199653d4eSeschrock * If the source is a hot spare, and the parent isn't already a 452299653d4eSeschrock * spare, then we want to create a new hot spare. Otherwise, we 452339c23413Seschrock * want to create a replacing vdev. The user is not allowed to 452439c23413Seschrock * attach to a spared vdev child unless the 'isspare' state is 452539c23413Seschrock * the same (spare replaces spare, non-spare replaces 452639c23413Seschrock * non-spare). 452799653d4eSeschrock */ 4528cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && 4529cb04b873SMark J Musante spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 453099653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4531cb04b873SMark J Musante } else if (pvd->vdev_ops == &vdev_spare_ops && 4532cb04b873SMark J Musante newvd->vdev_isspare != oldvd->vdev_isspare) { 453339c23413Seschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4534cb04b873SMark J Musante } 4535cb04b873SMark J Musante 4536cb04b873SMark J Musante if (newvd->vdev_isspare) 453799653d4eSeschrock pvops = &vdev_spare_ops; 453899653d4eSeschrock else 453999653d4eSeschrock pvops = &vdev_replacing_ops; 454099653d4eSeschrock } 454199653d4eSeschrock 45422a79c5feSlling /* 4543573ca77eSGeorge Wilson * Make sure the new device is big enough. 45442a79c5feSlling */ 4545573ca77eSGeorge Wilson if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4546fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4547fa9e4066Sahrens 4548ecc2d604Sbonwick /* 4549ecc2d604Sbonwick * The new device cannot have a higher alignment requirement 4550ecc2d604Sbonwick * than the top-level vdev. 4551ecc2d604Sbonwick */ 4552ecc2d604Sbonwick if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4553fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4554fa9e4066Sahrens 4555fa9e4066Sahrens /* 4556fa9e4066Sahrens * If this is an in-place replacement, update oldvd's path and devid 4557fa9e4066Sahrens * to make it distinguishable from newvd, and unopenable from now on. 4558fa9e4066Sahrens */ 4559fa9e4066Sahrens if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4560fa9e4066Sahrens spa_strfree(oldvd->vdev_path); 4561fa9e4066Sahrens oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4562fa9e4066Sahrens KM_SLEEP); 4563fa9e4066Sahrens (void) sprintf(oldvd->vdev_path, "%s/%s", 4564fa9e4066Sahrens newvd->vdev_path, "old"); 4565fa9e4066Sahrens if (oldvd->vdev_devid != NULL) { 4566fa9e4066Sahrens spa_strfree(oldvd->vdev_devid); 4567fa9e4066Sahrens oldvd->vdev_devid = NULL; 4568fa9e4066Sahrens } 4569fa9e4066Sahrens } 4570fa9e4066Sahrens 4571cb04b873SMark J Musante /* mark the device being resilvered */ 4572b4952e17SGeorge Wilson newvd->vdev_resilver_txg = txg; 4573cb04b873SMark J Musante 4574fa9e4066Sahrens /* 457599653d4eSeschrock * If the parent is not a mirror, or if we're replacing, insert the new 457699653d4eSeschrock * mirror/replacing/spare vdev above oldvd. 4577fa9e4066Sahrens */ 4578fa9e4066Sahrens if (pvd->vdev_ops != pvops) 4579fa9e4066Sahrens pvd = vdev_add_parent(oldvd, pvops); 4580fa9e4066Sahrens 4581fa9e4066Sahrens ASSERT(pvd->vdev_top->vdev_parent == rvd); 4582fa9e4066Sahrens ASSERT(pvd->vdev_ops == pvops); 4583fa9e4066Sahrens ASSERT(oldvd->vdev_parent == pvd); 4584fa9e4066Sahrens 4585fa9e4066Sahrens /* 4586fa9e4066Sahrens * Extract the new device from its root and add it to pvd. 4587fa9e4066Sahrens */ 4588fa9e4066Sahrens vdev_remove_child(newrootvd, newvd); 4589fa9e4066Sahrens newvd->vdev_id = pvd->vdev_children; 459088ecc943SGeorge Wilson newvd->vdev_crtxg = oldvd->vdev_crtxg; 4591fa9e4066Sahrens vdev_add_child(pvd, newvd); 4592fa9e4066Sahrens 4593fa9e4066Sahrens tvd = newvd->vdev_top; 4594fa9e4066Sahrens ASSERT(pvd->vdev_top == tvd); 4595fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 4596fa9e4066Sahrens 4597fa9e4066Sahrens vdev_config_dirty(tvd); 4598fa9e4066Sahrens 4599fa9e4066Sahrens /* 46003f9d6ad7SLin Ling * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 46013f9d6ad7SLin Ling * for any dmu_sync-ed blocks. It will propagate upward when 46023f9d6ad7SLin Ling * spa_vdev_exit() calls vdev_dtl_reassess(). 4603fa9e4066Sahrens */ 46043f9d6ad7SLin Ling dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4605fa9e4066Sahrens 46063f9d6ad7SLin Ling vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 46073f9d6ad7SLin Ling dtl_max_txg - TXG_INITIAL); 4608fa9e4066Sahrens 46096809eb4eSEric Schrock if (newvd->vdev_isspare) { 461039c23413Seschrock spa_spare_activate(newvd); 46116809eb4eSEric Schrock spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 46126809eb4eSEric Schrock } 46136809eb4eSEric Schrock 4614e14bb325SJeff Bonwick oldvdpath = spa_strdup(oldvd->vdev_path); 4615e14bb325SJeff Bonwick newvdpath = spa_strdup(newvd->vdev_path); 46169b3f6b42SEric Kustarz newvd_isspare = newvd->vdev_isspare; 4617ea8dc4b6Seschrock 4618fa9e4066Sahrens /* 4619fa9e4066Sahrens * Mark newvd's DTL dirty in this txg. 4620fa9e4066Sahrens */ 4621ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, newvd, txg); 4622fa9e4066Sahrens 46233f9d6ad7SLin Ling /* 46240713e232SGeorge Wilson * Schedule the resilver to restart in the future. We do this to 46250713e232SGeorge Wilson * ensure that dmu_sync-ed blocks have been stitched into the 46260713e232SGeorge Wilson * respective datasets. 46273f9d6ad7SLin Ling */ 46283f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 46293f9d6ad7SLin Ling 4630*14372834SHans Rosenfeld if (spa->spa_bootfs) 4631*14372834SHans Rosenfeld spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4632*14372834SHans Rosenfeld 4633*14372834SHans Rosenfeld spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 4634*14372834SHans Rosenfeld 46353f9d6ad7SLin Ling /* 46363f9d6ad7SLin Ling * Commit the config 46373f9d6ad7SLin Ling */ 46383f9d6ad7SLin Ling (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4639fa9e4066Sahrens 46404445fffbSMatthew Ahrens spa_history_log_internal(spa, "vdev attach", NULL, 46413f9d6ad7SLin Ling "%s vdev=%s %s vdev=%s", 4642c8e1f6d2SMark J Musante replacing && newvd_isspare ? "spare in" : 4643c8e1f6d2SMark J Musante replacing ? "replace" : "attach", newvdpath, 4644c8e1f6d2SMark J Musante replacing ? "for" : "to", oldvdpath); 46459b3f6b42SEric Kustarz 46469b3f6b42SEric Kustarz spa_strfree(oldvdpath); 46479b3f6b42SEric Kustarz spa_strfree(newvdpath); 46489b3f6b42SEric Kustarz 4649fa9e4066Sahrens return (0); 4650fa9e4066Sahrens } 4651fa9e4066Sahrens 4652fa9e4066Sahrens /* 4653fa9e4066Sahrens * Detach a device from a mirror or replacing vdev. 4654f7170741SWill Andrews * 4655fa9e4066Sahrens * If 'replace_done' is specified, only detach if the parent 4656fa9e4066Sahrens * is a replacing vdev. 4657fa9e4066Sahrens */ 4658fa9e4066Sahrens int 46598ad4d6ddSJeff Bonwick spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4660fa9e4066Sahrens { 4661fa9e4066Sahrens uint64_t txg; 46628ad4d6ddSJeff Bonwick int error; 4663fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4664fa9e4066Sahrens vdev_t *vd, *pvd, *cvd, *tvd; 466599653d4eSeschrock boolean_t unspare = B_FALSE; 4666d5285caeSGeorge Wilson uint64_t unspare_guid = 0; 46671195e687SMark J Musante char *vdpath; 4668fa9e4066Sahrens 4669f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4670f9af39baSGeorge Wilson 4671fa9e4066Sahrens txg = spa_vdev_enter(spa); 4672fa9e4066Sahrens 4673c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4674fa9e4066Sahrens 4675fa9e4066Sahrens if (vd == NULL) 4676fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4677fa9e4066Sahrens 46780e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 46790e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 46800e34b6a7Sbonwick 4681fa9e4066Sahrens pvd = vd->vdev_parent; 4682fa9e4066Sahrens 46838ad4d6ddSJeff Bonwick /* 46848ad4d6ddSJeff Bonwick * If the parent/child relationship is not as expected, don't do it. 46858ad4d6ddSJeff Bonwick * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 46868ad4d6ddSJeff Bonwick * vdev that's replacing B with C. The user's intent in replacing 46878ad4d6ddSJeff Bonwick * is to go from M(A,B) to M(A,C). If the user decides to cancel 46888ad4d6ddSJeff Bonwick * the replace by detaching C, the expected behavior is to end up 46898ad4d6ddSJeff Bonwick * M(A,B). But suppose that right after deciding to detach C, 46908ad4d6ddSJeff Bonwick * the replacement of B completes. We would have M(A,C), and then 46918ad4d6ddSJeff Bonwick * ask to detach C, which would leave us with just A -- not what 46928ad4d6ddSJeff Bonwick * the user wanted. To prevent this, we make sure that the 46938ad4d6ddSJeff Bonwick * parent/child relationship hasn't changed -- in this example, 46948ad4d6ddSJeff Bonwick * that C's parent is still the replacing vdev R. 46958ad4d6ddSJeff Bonwick */ 46968ad4d6ddSJeff Bonwick if (pvd->vdev_guid != pguid && pguid != 0) 46978ad4d6ddSJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 46988ad4d6ddSJeff Bonwick 4699fa9e4066Sahrens /* 4700cb04b873SMark J Musante * Only 'replacing' or 'spare' vdevs can be replaced. 470199653d4eSeschrock */ 4702cb04b873SMark J Musante if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4703cb04b873SMark J Musante pvd->vdev_ops != &vdev_spare_ops) 4704cb04b873SMark J Musante return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 470599653d4eSeschrock 470699653d4eSeschrock ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4707e7437265Sahrens spa_version(spa) >= SPA_VERSION_SPARES); 4708fa9e4066Sahrens 4709fa9e4066Sahrens /* 471099653d4eSeschrock * Only mirror, replacing, and spare vdevs support detach. 4711fa9e4066Sahrens */ 4712fa9e4066Sahrens if (pvd->vdev_ops != &vdev_replacing_ops && 471399653d4eSeschrock pvd->vdev_ops != &vdev_mirror_ops && 471499653d4eSeschrock pvd->vdev_ops != &vdev_spare_ops) 4715fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4716fa9e4066Sahrens 4717fa9e4066Sahrens /* 47188ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 47198ad4d6ddSJeff Bonwick * we cannot safely detach it. 4720fa9e4066Sahrens */ 47218ad4d6ddSJeff Bonwick if (vdev_dtl_required(vd)) 4722fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4723fa9e4066Sahrens 47248ad4d6ddSJeff Bonwick ASSERT(pvd->vdev_children >= 2); 4725fa9e4066Sahrens 4726bf82a41bSeschrock /* 4727bf82a41bSeschrock * If we are detaching the second disk from a replacing vdev, then 4728bf82a41bSeschrock * check to see if we changed the original vdev's path to have "/old" 4729bf82a41bSeschrock * at the end in spa_vdev_attach(). If so, undo that change now. 4730bf82a41bSeschrock */ 4731cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4732cb04b873SMark J Musante vd->vdev_path != NULL) { 4733cb04b873SMark J Musante size_t len = strlen(vd->vdev_path); 4734cb04b873SMark J Musante 4735cb04b873SMark J Musante for (int c = 0; c < pvd->vdev_children; c++) { 4736cb04b873SMark J Musante cvd = pvd->vdev_child[c]; 4737cb04b873SMark J Musante 4738cb04b873SMark J Musante if (cvd == vd || cvd->vdev_path == NULL) 4739cb04b873SMark J Musante continue; 4740cb04b873SMark J Musante 4741cb04b873SMark J Musante if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4742cb04b873SMark J Musante strcmp(cvd->vdev_path + len, "/old") == 0) { 4743cb04b873SMark J Musante spa_strfree(cvd->vdev_path); 4744cb04b873SMark J Musante cvd->vdev_path = spa_strdup(vd->vdev_path); 4745cb04b873SMark J Musante break; 4746cb04b873SMark J Musante } 4747bf82a41bSeschrock } 4748bf82a41bSeschrock } 4749bf82a41bSeschrock 475099653d4eSeschrock /* 475199653d4eSeschrock * If we are detaching the original disk from a spare, then it implies 475299653d4eSeschrock * that the spare should become a real disk, and be removed from the 475399653d4eSeschrock * active spare list for the pool. 475499653d4eSeschrock */ 475599653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 4756cb04b873SMark J Musante vd->vdev_id == 0 && 4757cb04b873SMark J Musante pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 475899653d4eSeschrock unspare = B_TRUE; 475999653d4eSeschrock 4760fa9e4066Sahrens /* 4761fa9e4066Sahrens * Erase the disk labels so the disk can be used for other things. 4762fa9e4066Sahrens * This must be done after all other error cases are handled, 4763fa9e4066Sahrens * but before we disembowel vd (so we can still do I/O to it). 4764fa9e4066Sahrens * But if we can't do it, don't treat the error as fatal -- 4765fa9e4066Sahrens * it may be that the unwritability of the disk is the reason 4766fa9e4066Sahrens * it's being detached! 4767fa9e4066Sahrens */ 476839c23413Seschrock error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4769fa9e4066Sahrens 4770fa9e4066Sahrens /* 4771fa9e4066Sahrens * Remove vd from its parent and compact the parent's children. 4772fa9e4066Sahrens */ 4773fa9e4066Sahrens vdev_remove_child(pvd, vd); 4774fa9e4066Sahrens vdev_compact_children(pvd); 4775fa9e4066Sahrens 4776fa9e4066Sahrens /* 4777fa9e4066Sahrens * Remember one of the remaining children so we can get tvd below. 4778fa9e4066Sahrens */ 4779cb04b873SMark J Musante cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4780fa9e4066Sahrens 478199653d4eSeschrock /* 478299653d4eSeschrock * If we need to remove the remaining child from the list of hot spares, 47838ad4d6ddSJeff Bonwick * do it now, marking the vdev as no longer a spare in the process. 47848ad4d6ddSJeff Bonwick * We must do this before vdev_remove_parent(), because that can 47858ad4d6ddSJeff Bonwick * change the GUID if it creates a new toplevel GUID. For a similar 47868ad4d6ddSJeff Bonwick * reason, we must remove the spare now, in the same txg as the detach; 47878ad4d6ddSJeff Bonwick * otherwise someone could attach a new sibling, change the GUID, and 47888ad4d6ddSJeff Bonwick * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 478999653d4eSeschrock */ 479099653d4eSeschrock if (unspare) { 479199653d4eSeschrock ASSERT(cvd->vdev_isspare); 479239c23413Seschrock spa_spare_remove(cvd); 479399653d4eSeschrock unspare_guid = cvd->vdev_guid; 47948ad4d6ddSJeff Bonwick (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4795cb04b873SMark J Musante cvd->vdev_unspare = B_TRUE; 479699653d4eSeschrock } 479799653d4eSeschrock 4798fa9e4066Sahrens /* 4799fa9e4066Sahrens * If the parent mirror/replacing vdev only has one child, 4800fa9e4066Sahrens * the parent is no longer needed. Remove it from the tree. 4801fa9e4066Sahrens */ 4802cb04b873SMark J Musante if (pvd->vdev_children == 1) { 4803cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_spare_ops) 4804cb04b873SMark J Musante cvd->vdev_unspare = B_FALSE; 4805fa9e4066Sahrens vdev_remove_parent(cvd); 4806cb04b873SMark J Musante } 4807cb04b873SMark J Musante 4808fa9e4066Sahrens 4809fa9e4066Sahrens /* 4810fa9e4066Sahrens * We don't set tvd until now because the parent we just removed 4811fa9e4066Sahrens * may have been the previous top-level vdev. 4812fa9e4066Sahrens */ 4813fa9e4066Sahrens tvd = cvd->vdev_top; 4814fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 4815fa9e4066Sahrens 4816fa9e4066Sahrens /* 481739c23413Seschrock * Reevaluate the parent vdev state. 4818fa9e4066Sahrens */ 48193d7072f8Seschrock vdev_propagate_state(cvd); 4820fa9e4066Sahrens 4821fa9e4066Sahrens /* 4822573ca77eSGeorge Wilson * If the 'autoexpand' property is set on the pool then automatically 4823573ca77eSGeorge Wilson * try to expand the size of the pool. For example if the device we 4824573ca77eSGeorge Wilson * just detached was smaller than the others, it may be possible to 4825573ca77eSGeorge Wilson * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4826573ca77eSGeorge Wilson * first so that we can obtain the updated sizes of the leaf vdevs. 4827fa9e4066Sahrens */ 4828573ca77eSGeorge Wilson if (spa->spa_autoexpand) { 4829573ca77eSGeorge Wilson vdev_reopen(tvd); 4830573ca77eSGeorge Wilson vdev_expand(tvd, txg); 4831573ca77eSGeorge Wilson } 4832fa9e4066Sahrens 4833fa9e4066Sahrens vdev_config_dirty(tvd); 4834fa9e4066Sahrens 4835fa9e4066Sahrens /* 483639c23413Seschrock * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 483739c23413Seschrock * vd->vdev_detached is set and free vd's DTL object in syncing context. 483839c23413Seschrock * But first make sure we're not on any *other* txg's DTL list, to 483939c23413Seschrock * prevent vd from being accessed after it's freed. 4840fa9e4066Sahrens */ 48411195e687SMark J Musante vdpath = spa_strdup(vd->vdev_path); 48428ad4d6ddSJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 4843fa9e4066Sahrens (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4844ecc2d604Sbonwick vd->vdev_detached = B_TRUE; 4845ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, vd, txg); 4846fa9e4066Sahrens 48473d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 48483d7072f8Seschrock 4849cb04b873SMark J Musante /* hang on to the spa before we release the lock */ 4850cb04b873SMark J Musante spa_open_ref(spa, FTAG); 4851cb04b873SMark J Musante 485299653d4eSeschrock error = spa_vdev_exit(spa, vd, txg, 0); 485399653d4eSeschrock 48544445fffbSMatthew Ahrens spa_history_log_internal(spa, "detach", NULL, 48551195e687SMark J Musante "vdev=%s", vdpath); 48561195e687SMark J Musante spa_strfree(vdpath); 48571195e687SMark J Musante 485899653d4eSeschrock /* 485939c23413Seschrock * If this was the removal of the original device in a hot spare vdev, 486039c23413Seschrock * then we want to go through and remove the device from the hot spare 486139c23413Seschrock * list of every other pool. 486299653d4eSeschrock */ 486399653d4eSeschrock if (unspare) { 4864cb04b873SMark J Musante spa_t *altspa = NULL; 4865cb04b873SMark J Musante 486699653d4eSeschrock mutex_enter(&spa_namespace_lock); 4867cb04b873SMark J Musante while ((altspa = spa_next(altspa)) != NULL) { 4868cb04b873SMark J Musante if (altspa->spa_state != POOL_STATE_ACTIVE || 4869cb04b873SMark J Musante altspa == spa) 487099653d4eSeschrock continue; 4871cb04b873SMark J Musante 4872cb04b873SMark J Musante spa_open_ref(altspa, FTAG); 48739af0a4dfSJeff Bonwick mutex_exit(&spa_namespace_lock); 4874cb04b873SMark J Musante (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 48759af0a4dfSJeff Bonwick mutex_enter(&spa_namespace_lock); 4876cb04b873SMark J Musante spa_close(altspa, FTAG); 487799653d4eSeschrock } 487899653d4eSeschrock mutex_exit(&spa_namespace_lock); 4879cb04b873SMark J Musante 4880cb04b873SMark J Musante /* search the rest of the vdevs for spares to remove */ 4881cb04b873SMark J Musante spa_vdev_resilver_done(spa); 488299653d4eSeschrock } 488399653d4eSeschrock 4884cb04b873SMark J Musante /* all done with the spa; OK to release */ 4885cb04b873SMark J Musante mutex_enter(&spa_namespace_lock); 4886cb04b873SMark J Musante spa_close(spa, FTAG); 4887cb04b873SMark J Musante mutex_exit(&spa_namespace_lock); 4888cb04b873SMark J Musante 488999653d4eSeschrock return (error); 489099653d4eSeschrock } 489199653d4eSeschrock 48921195e687SMark J Musante /* 48931195e687SMark J Musante * Split a set of devices from their mirrors, and create a new pool from them. 48941195e687SMark J Musante */ 48951195e687SMark J Musante int 48961195e687SMark J Musante spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 48971195e687SMark J Musante nvlist_t *props, boolean_t exp) 48981195e687SMark J Musante { 48991195e687SMark J Musante int error = 0; 49001195e687SMark J Musante uint64_t txg, *glist; 49011195e687SMark J Musante spa_t *newspa; 49021195e687SMark J Musante uint_t c, children, lastlog; 49031195e687SMark J Musante nvlist_t **child, *nvl, *tmp; 49041195e687SMark J Musante dmu_tx_t *tx; 49051195e687SMark J Musante char *altroot = NULL; 49061195e687SMark J Musante vdev_t *rvd, **vml = NULL; /* vdev modify list */ 49071195e687SMark J Musante boolean_t activate_slog; 49081195e687SMark J Musante 4909f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 49101195e687SMark J Musante 49111195e687SMark J Musante txg = spa_vdev_enter(spa); 49121195e687SMark J Musante 49131195e687SMark J Musante /* clear the log and flush everything up to now */ 49141195e687SMark J Musante activate_slog = spa_passivate_log(spa); 49151195e687SMark J Musante (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 49161195e687SMark J Musante error = spa_offline_log(spa); 49171195e687SMark J Musante txg = spa_vdev_config_enter(spa); 49181195e687SMark J Musante 49191195e687SMark J Musante if (activate_slog) 49201195e687SMark J Musante spa_activate_log(spa); 49211195e687SMark J Musante 49221195e687SMark J Musante if (error != 0) 49231195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 49241195e687SMark J Musante 49251195e687SMark J Musante /* check new spa name before going any further */ 49261195e687SMark J Musante if (spa_lookup(newname) != NULL) 49271195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 49281195e687SMark J Musante 49291195e687SMark J Musante /* 49301195e687SMark J Musante * scan through all the children to ensure they're all mirrors 49311195e687SMark J Musante */ 49321195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 49331195e687SMark J Musante nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 49341195e687SMark J Musante &children) != 0) 49351195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 49361195e687SMark J Musante 49371195e687SMark J Musante /* first, check to ensure we've got the right child count */ 49381195e687SMark J Musante rvd = spa->spa_root_vdev; 49391195e687SMark J Musante lastlog = 0; 49401195e687SMark J Musante for (c = 0; c < rvd->vdev_children; c++) { 49411195e687SMark J Musante vdev_t *vd = rvd->vdev_child[c]; 49421195e687SMark J Musante 49431195e687SMark J Musante /* don't count the holes & logs as children */ 49441195e687SMark J Musante if (vd->vdev_islog || vd->vdev_ishole) { 49451195e687SMark J Musante if (lastlog == 0) 49461195e687SMark J Musante lastlog = c; 49471195e687SMark J Musante continue; 49481195e687SMark J Musante } 49491195e687SMark J Musante 49501195e687SMark J Musante lastlog = 0; 49511195e687SMark J Musante } 49521195e687SMark J Musante if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 49531195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 49541195e687SMark J Musante 49551195e687SMark J Musante /* next, ensure no spare or cache devices are part of the split */ 49561195e687SMark J Musante if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 49571195e687SMark J Musante nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 49581195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 49591195e687SMark J Musante 49601195e687SMark J Musante vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 49611195e687SMark J Musante glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 49621195e687SMark J Musante 49631195e687SMark J Musante /* then, loop over each vdev and validate it */ 49641195e687SMark J Musante for (c = 0; c < children; c++) { 49651195e687SMark J Musante uint64_t is_hole = 0; 49661195e687SMark J Musante 49671195e687SMark J Musante (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 49681195e687SMark J Musante &is_hole); 49691195e687SMark J Musante 49701195e687SMark J Musante if (is_hole != 0) { 49711195e687SMark J Musante if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 49721195e687SMark J Musante spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 49731195e687SMark J Musante continue; 49741195e687SMark J Musante } else { 4975be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 49761195e687SMark J Musante break; 49771195e687SMark J Musante } 49781195e687SMark J Musante } 49791195e687SMark J Musante 49801195e687SMark J Musante /* which disk is going to be split? */ 49811195e687SMark J Musante if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 49821195e687SMark J Musante &glist[c]) != 0) { 4983be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 49841195e687SMark J Musante break; 49851195e687SMark J Musante } 49861195e687SMark J Musante 49871195e687SMark J Musante /* look it up in the spa */ 49881195e687SMark J Musante vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 49891195e687SMark J Musante if (vml[c] == NULL) { 4990be6fd75aSMatthew Ahrens error = SET_ERROR(ENODEV); 49911195e687SMark J Musante break; 49921195e687SMark J Musante } 49931195e687SMark J Musante 49941195e687SMark J Musante /* make sure there's nothing stopping the split */ 49951195e687SMark J Musante if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 49961195e687SMark J Musante vml[c]->vdev_islog || 49971195e687SMark J Musante vml[c]->vdev_ishole || 49981195e687SMark J Musante vml[c]->vdev_isspare || 49991195e687SMark J Musante vml[c]->vdev_isl2cache || 50001195e687SMark J Musante !vdev_writeable(vml[c]) || 5001d41c4376SMark J Musante vml[c]->vdev_children != 0 || 50021195e687SMark J Musante vml[c]->vdev_state != VDEV_STATE_HEALTHY || 50031195e687SMark J Musante c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5004be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 50051195e687SMark J Musante break; 50061195e687SMark J Musante } 50071195e687SMark J Musante 50081195e687SMark J Musante if (vdev_dtl_required(vml[c])) { 5009be6fd75aSMatthew Ahrens error = SET_ERROR(EBUSY); 50101195e687SMark J Musante break; 50111195e687SMark J Musante } 50121195e687SMark J Musante 50131195e687SMark J Musante /* we need certain info from the top level */ 50141195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 50151195e687SMark J Musante vml[c]->vdev_top->vdev_ms_array) == 0); 50161195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 50171195e687SMark J Musante vml[c]->vdev_top->vdev_ms_shift) == 0); 50181195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 50191195e687SMark J Musante vml[c]->vdev_top->vdev_asize) == 0); 50201195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 50211195e687SMark J Musante vml[c]->vdev_top->vdev_ashift) == 0); 50221195e687SMark J Musante } 50231195e687SMark J Musante 50241195e687SMark J Musante if (error != 0) { 50251195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 50261195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 50271195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 50281195e687SMark J Musante } 50291195e687SMark J Musante 50301195e687SMark J Musante /* stop writers from using the disks */ 50311195e687SMark J Musante for (c = 0; c < children; c++) { 50321195e687SMark J Musante if (vml[c] != NULL) 50331195e687SMark J Musante vml[c]->vdev_offline = B_TRUE; 50341195e687SMark J Musante } 50351195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 50361195e687SMark J Musante 50371195e687SMark J Musante /* 50381195e687SMark J Musante * Temporarily record the splitting vdevs in the spa config. This 50391195e687SMark J Musante * will disappear once the config is regenerated. 50401195e687SMark J Musante */ 50411195e687SMark J Musante VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 50421195e687SMark J Musante VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 50431195e687SMark J Musante glist, children) == 0); 50441195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 50451195e687SMark J Musante 504698295d61SMark J Musante mutex_enter(&spa->spa_props_lock); 50471195e687SMark J Musante VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 50481195e687SMark J Musante nvl) == 0); 504998295d61SMark J Musante mutex_exit(&spa->spa_props_lock); 50501195e687SMark J Musante spa->spa_config_splitting = nvl; 50511195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 50521195e687SMark J Musante 50531195e687SMark J Musante /* configure and create the new pool */ 50541195e687SMark J Musante VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 50551195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 50561195e687SMark J Musante exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 50571195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 50581195e687SMark J Musante spa_version(spa)) == 0); 50591195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 50601195e687SMark J Musante spa->spa_config_txg) == 0); 50611195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 50621195e687SMark J Musante spa_generate_guid(NULL)) == 0); 50631195e687SMark J Musante (void) nvlist_lookup_string(props, 50641195e687SMark J Musante zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 50651195e687SMark J Musante 5066d41c4376SMark J Musante /* add the new pool to the namespace */ 50671195e687SMark J Musante newspa = spa_add(newname, config, altroot); 50681195e687SMark J Musante newspa->spa_config_txg = spa->spa_config_txg; 50691195e687SMark J Musante spa_set_log_state(newspa, SPA_LOG_CLEAR); 50701195e687SMark J Musante 50711195e687SMark J Musante /* release the spa config lock, retaining the namespace lock */ 50721195e687SMark J Musante spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 50731195e687SMark J Musante 50741195e687SMark J Musante if (zio_injection_enabled) 50751195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 1); 50761195e687SMark J Musante 50771195e687SMark J Musante spa_activate(newspa, spa_mode_global); 50781195e687SMark J Musante spa_async_suspend(newspa); 50791195e687SMark J Musante 50801195e687SMark J Musante /* create the new pool from the disks of the original pool */ 50811195e687SMark J Musante error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 50821195e687SMark J Musante if (error) 50831195e687SMark J Musante goto out; 50841195e687SMark J Musante 50851195e687SMark J Musante /* if that worked, generate a real config for the new pool */ 50861195e687SMark J Musante if (newspa->spa_root_vdev != NULL) { 50871195e687SMark J Musante VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 50881195e687SMark J Musante NV_UNIQUE_NAME, KM_SLEEP) == 0); 50891195e687SMark J Musante VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 50901195e687SMark J Musante ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 50911195e687SMark J Musante spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 50921195e687SMark J Musante B_TRUE)); 50931195e687SMark J Musante } 50941195e687SMark J Musante 50951195e687SMark J Musante /* set the props */ 50961195e687SMark J Musante if (props != NULL) { 50971195e687SMark J Musante spa_configfile_set(newspa, props, B_FALSE); 50981195e687SMark J Musante error = spa_prop_set(newspa, props); 50991195e687SMark J Musante if (error) 51001195e687SMark J Musante goto out; 51011195e687SMark J Musante } 51021195e687SMark J Musante 51031195e687SMark J Musante /* flush everything */ 51041195e687SMark J Musante txg = spa_vdev_config_enter(newspa); 51051195e687SMark J Musante vdev_config_dirty(newspa->spa_root_vdev); 51061195e687SMark J Musante (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 51071195e687SMark J Musante 51081195e687SMark J Musante if (zio_injection_enabled) 51091195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 2); 51101195e687SMark J Musante 51111195e687SMark J Musante spa_async_resume(newspa); 51121195e687SMark J Musante 51131195e687SMark J Musante /* finally, update the original pool's config */ 51141195e687SMark J Musante txg = spa_vdev_config_enter(spa); 51151195e687SMark J Musante tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 51161195e687SMark J Musante error = dmu_tx_assign(tx, TXG_WAIT); 51171195e687SMark J Musante if (error != 0) 51181195e687SMark J Musante dmu_tx_abort(tx); 51191195e687SMark J Musante for (c = 0; c < children; c++) { 51201195e687SMark J Musante if (vml[c] != NULL) { 51211195e687SMark J Musante vdev_split(vml[c]); 51221195e687SMark J Musante if (error == 0) 51234445fffbSMatthew Ahrens spa_history_log_internal(spa, "detach", tx, 51244445fffbSMatthew Ahrens "vdev=%s", vml[c]->vdev_path); 51251195e687SMark J Musante vdev_free(vml[c]); 51261195e687SMark J Musante } 51271195e687SMark J Musante } 51281195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 51291195e687SMark J Musante spa->spa_config_splitting = NULL; 51301195e687SMark J Musante nvlist_free(nvl); 51311195e687SMark J Musante if (error == 0) 51321195e687SMark J Musante dmu_tx_commit(tx); 51331195e687SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, 0); 51341195e687SMark J Musante 51351195e687SMark J Musante if (zio_injection_enabled) 51361195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 3); 51371195e687SMark J Musante 51381195e687SMark J Musante /* split is complete; log a history record */ 51394445fffbSMatthew Ahrens spa_history_log_internal(newspa, "split", NULL, 51404445fffbSMatthew Ahrens "from pool %s", spa_name(spa)); 51411195e687SMark J Musante 51421195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 51431195e687SMark J Musante 51441195e687SMark J Musante /* if we're not going to mount the filesystems in userland, export */ 51451195e687SMark J Musante if (exp) 51461195e687SMark J Musante error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 51471195e687SMark J Musante B_FALSE, B_FALSE); 51481195e687SMark J Musante 51491195e687SMark J Musante return (error); 51501195e687SMark J Musante 51511195e687SMark J Musante out: 51521195e687SMark J Musante spa_unload(newspa); 51531195e687SMark J Musante spa_deactivate(newspa); 51541195e687SMark J Musante spa_remove(newspa); 51551195e687SMark J Musante 51561195e687SMark J Musante txg = spa_vdev_config_enter(spa); 515798295d61SMark J Musante 515898295d61SMark J Musante /* re-online all offlined disks */ 515998295d61SMark J Musante for (c = 0; c < children; c++) { 516098295d61SMark J Musante if (vml[c] != NULL) 516198295d61SMark J Musante vml[c]->vdev_offline = B_FALSE; 516298295d61SMark J Musante } 516398295d61SMark J Musante vdev_reopen(spa->spa_root_vdev); 516498295d61SMark J Musante 51651195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 51661195e687SMark J Musante spa->spa_config_splitting = NULL; 5167d41c4376SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, error); 51681195e687SMark J Musante 51691195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 51701195e687SMark J Musante return (error); 51711195e687SMark J Musante } 51721195e687SMark J Musante 5173e14bb325SJeff Bonwick static nvlist_t * 5174e14bb325SJeff Bonwick spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 517599653d4eSeschrock { 5176e14bb325SJeff Bonwick for (int i = 0; i < count; i++) { 5177e14bb325SJeff Bonwick uint64_t guid; 517899653d4eSeschrock 5179e14bb325SJeff Bonwick VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5180e14bb325SJeff Bonwick &guid) == 0); 518199653d4eSeschrock 5182e14bb325SJeff Bonwick if (guid == target_guid) 5183e14bb325SJeff Bonwick return (nvpp[i]); 518499653d4eSeschrock } 518599653d4eSeschrock 5186e14bb325SJeff Bonwick return (NULL); 5187fa94a07fSbrendan } 5188fa94a07fSbrendan 5189e14bb325SJeff Bonwick static void 5190e14bb325SJeff Bonwick spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5191e14bb325SJeff Bonwick nvlist_t *dev_to_remove) 5192fa94a07fSbrendan { 5193e14bb325SJeff Bonwick nvlist_t **newdev = NULL; 5194fa94a07fSbrendan 5195e14bb325SJeff Bonwick if (count > 1) 5196e14bb325SJeff Bonwick newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5197fa94a07fSbrendan 5198e14bb325SJeff Bonwick for (int i = 0, j = 0; i < count; i++) { 5199e14bb325SJeff Bonwick if (dev[i] == dev_to_remove) 5200e14bb325SJeff Bonwick continue; 5201e14bb325SJeff Bonwick VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5202fa94a07fSbrendan } 5203fa94a07fSbrendan 5204e14bb325SJeff Bonwick VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5205e14bb325SJeff Bonwick VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5206fa94a07fSbrendan 5207e14bb325SJeff Bonwick for (int i = 0; i < count - 1; i++) 5208e14bb325SJeff Bonwick nvlist_free(newdev[i]); 5209fa94a07fSbrendan 5210e14bb325SJeff Bonwick if (count > 1) 5211e14bb325SJeff Bonwick kmem_free(newdev, (count - 1) * sizeof (void *)); 5212fa94a07fSbrendan } 5213fa94a07fSbrendan 521488ecc943SGeorge Wilson /* 521588ecc943SGeorge Wilson * Evacuate the device. 521688ecc943SGeorge Wilson */ 52173f9d6ad7SLin Ling static int 521888ecc943SGeorge Wilson spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 521988ecc943SGeorge Wilson { 522088ecc943SGeorge Wilson uint64_t txg; 52213f9d6ad7SLin Ling int error = 0; 522288ecc943SGeorge Wilson 522388ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 522488ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5225b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 522688ecc943SGeorge Wilson 522788ecc943SGeorge Wilson /* 522888ecc943SGeorge Wilson * Evacuate the device. We don't hold the config lock as writer 522988ecc943SGeorge Wilson * since we need to do I/O but we do keep the 523088ecc943SGeorge Wilson * spa_namespace_lock held. Once this completes the device 523188ecc943SGeorge Wilson * should no longer have any blocks allocated on it. 523288ecc943SGeorge Wilson */ 523388ecc943SGeorge Wilson if (vd->vdev_islog) { 52343f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 52353f9d6ad7SLin Ling error = spa_offline_log(spa); 5236a1521560SJeff Bonwick } else { 5237be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 523888ecc943SGeorge Wilson } 523988ecc943SGeorge Wilson 5240a1521560SJeff Bonwick if (error) 5241a1521560SJeff Bonwick return (error); 5242a1521560SJeff Bonwick 524388ecc943SGeorge Wilson /* 5244a1521560SJeff Bonwick * The evacuation succeeded. Remove any remaining MOS metadata 5245a1521560SJeff Bonwick * associated with this vdev, and wait for these changes to sync. 524688ecc943SGeorge Wilson */ 5247fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 524888ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 524988ecc943SGeorge Wilson vd->vdev_removing = B_TRUE; 52500713e232SGeorge Wilson vdev_dirty_leaves(vd, VDD_DTL, txg); 525188ecc943SGeorge Wilson vdev_config_dirty(vd); 525288ecc943SGeorge Wilson spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 525388ecc943SGeorge Wilson 525488ecc943SGeorge Wilson return (0); 525588ecc943SGeorge Wilson } 525688ecc943SGeorge Wilson 525788ecc943SGeorge Wilson /* 525888ecc943SGeorge Wilson * Complete the removal by cleaning up the namespace. 525988ecc943SGeorge Wilson */ 52603f9d6ad7SLin Ling static void 5261a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 526288ecc943SGeorge Wilson { 526388ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 526488ecc943SGeorge Wilson uint64_t id = vd->vdev_id; 526588ecc943SGeorge Wilson boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 526688ecc943SGeorge Wilson 526788ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 526888ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5269b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 527088ecc943SGeorge Wilson 52713f9d6ad7SLin Ling /* 52723f9d6ad7SLin Ling * Only remove any devices which are empty. 52733f9d6ad7SLin Ling */ 52743f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 52753f9d6ad7SLin Ling return; 52763f9d6ad7SLin Ling 527788ecc943SGeorge Wilson (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5278b24ab676SJeff Bonwick 5279b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_state_dirty_node)) 5280b24ab676SJeff Bonwick vdev_state_clean(vd); 5281b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_config_dirty_node)) 5282b24ab676SJeff Bonwick vdev_config_clean(vd); 5283b24ab676SJeff Bonwick 528488ecc943SGeorge Wilson vdev_free(vd); 528588ecc943SGeorge Wilson 528688ecc943SGeorge Wilson if (last_vdev) { 528788ecc943SGeorge Wilson vdev_compact_children(rvd); 528888ecc943SGeorge Wilson } else { 528988ecc943SGeorge Wilson vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 529088ecc943SGeorge Wilson vdev_add_child(rvd, vd); 529188ecc943SGeorge Wilson } 5292fcbfa62bSLin Ling vdev_config_dirty(rvd); 5293fcbfa62bSLin Ling 5294fcbfa62bSLin Ling /* 5295fcbfa62bSLin Ling * Reassess the health of our root vdev. 5296fcbfa62bSLin Ling */ 5297fcbfa62bSLin Ling vdev_reopen(rvd); 529888ecc943SGeorge Wilson } 529988ecc943SGeorge Wilson 53003f9d6ad7SLin Ling /* 53013f9d6ad7SLin Ling * Remove a device from the pool - 53023f9d6ad7SLin Ling * 53033f9d6ad7SLin Ling * Removing a device from the vdev namespace requires several steps 53043f9d6ad7SLin Ling * and can take a significant amount of time. As a result we use 53053f9d6ad7SLin Ling * the spa_vdev_config_[enter/exit] functions which allow us to 53063f9d6ad7SLin Ling * grab and release the spa_config_lock while still holding the namespace 53073f9d6ad7SLin Ling * lock. During each step the configuration is synced out. 5308f7170741SWill Andrews * 5309f7170741SWill Andrews * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5310f7170741SWill Andrews * devices. 5311fa94a07fSbrendan */ 5312fa94a07fSbrendan int 5313fa94a07fSbrendan spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5314fa94a07fSbrendan { 5315fa94a07fSbrendan vdev_t *vd; 5316a1521560SJeff Bonwick metaslab_group_t *mg; 5317e14bb325SJeff Bonwick nvlist_t **spares, **l2cache, *nv; 53188ad4d6ddSJeff Bonwick uint64_t txg = 0; 531988ecc943SGeorge Wilson uint_t nspares, nl2cache; 5320fa94a07fSbrendan int error = 0; 53218ad4d6ddSJeff Bonwick boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5322fa94a07fSbrendan 5323f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 5324f9af39baSGeorge Wilson 53258ad4d6ddSJeff Bonwick if (!locked) 53268ad4d6ddSJeff Bonwick txg = spa_vdev_enter(spa); 5327fa94a07fSbrendan 5328c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5329fa94a07fSbrendan 5330fa94a07fSbrendan if (spa->spa_spares.sav_vdevs != NULL && 5331fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5332e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5333e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5334e14bb325SJeff Bonwick /* 5335e14bb325SJeff Bonwick * Only remove the hot spare if it's not currently in use 5336e14bb325SJeff Bonwick * in this pool. 5337e14bb325SJeff Bonwick */ 5338e14bb325SJeff Bonwick if (vd == NULL || unspare) { 5339e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_spares.sav_config, 5340e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5341e14bb325SJeff Bonwick spa_load_spares(spa); 5342e14bb325SJeff Bonwick spa->spa_spares.sav_sync = B_TRUE; 5343e14bb325SJeff Bonwick } else { 5344be6fd75aSMatthew Ahrens error = SET_ERROR(EBUSY); 5345e14bb325SJeff Bonwick } 5346e14bb325SJeff Bonwick } else if (spa->spa_l2cache.sav_vdevs != NULL && 5347fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5348e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5349e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5350e14bb325SJeff Bonwick /* 5351e14bb325SJeff Bonwick * Cache devices can always be removed. 5352e14bb325SJeff Bonwick */ 5353e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5354e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5355fa94a07fSbrendan spa_load_l2cache(spa); 5356fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 535788ecc943SGeorge Wilson } else if (vd != NULL && vd->vdev_islog) { 535888ecc943SGeorge Wilson ASSERT(!locked); 5359b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 536088ecc943SGeorge Wilson 5361a1521560SJeff Bonwick mg = vd->vdev_mg; 5362a1521560SJeff Bonwick 536388ecc943SGeorge Wilson /* 5364a1521560SJeff Bonwick * Stop allocating from this vdev. 536588ecc943SGeorge Wilson */ 5366a1521560SJeff Bonwick metaslab_group_passivate(mg); 536788ecc943SGeorge Wilson 5368b24ab676SJeff Bonwick /* 5369b24ab676SJeff Bonwick * Wait for the youngest allocations and frees to sync, 5370b24ab676SJeff Bonwick * and then wait for the deferral of those frees to finish. 5371b24ab676SJeff Bonwick */ 5372b24ab676SJeff Bonwick spa_vdev_config_exit(spa, NULL, 5373b24ab676SJeff Bonwick txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5374b24ab676SJeff Bonwick 5375a1521560SJeff Bonwick /* 5376a1521560SJeff Bonwick * Attempt to evacuate the vdev. 5377a1521560SJeff Bonwick */ 5378a1521560SJeff Bonwick error = spa_vdev_remove_evacuate(spa, vd); 5379a1521560SJeff Bonwick 538088ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 538188ecc943SGeorge Wilson 5382a1521560SJeff Bonwick /* 5383a1521560SJeff Bonwick * If we couldn't evacuate the vdev, unwind. 5384a1521560SJeff Bonwick */ 5385a1521560SJeff Bonwick if (error) { 5386a1521560SJeff Bonwick metaslab_group_activate(mg); 5387a1521560SJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, error)); 5388a1521560SJeff Bonwick } 5389a1521560SJeff Bonwick 5390a1521560SJeff Bonwick /* 5391a1521560SJeff Bonwick * Clean up the vdev namespace. 5392a1521560SJeff Bonwick */ 5393a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa, vd); 539488ecc943SGeorge Wilson 5395e14bb325SJeff Bonwick } else if (vd != NULL) { 5396e14bb325SJeff Bonwick /* 5397e14bb325SJeff Bonwick * Normal vdevs cannot be removed (yet). 5398e14bb325SJeff Bonwick */ 5399be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 5400e14bb325SJeff Bonwick } else { 5401e14bb325SJeff Bonwick /* 5402e14bb325SJeff Bonwick * There is no vdev of any kind with the specified guid. 5403e14bb325SJeff Bonwick */ 5404be6fd75aSMatthew Ahrens error = SET_ERROR(ENOENT); 5405fa94a07fSbrendan } 540699653d4eSeschrock 54078ad4d6ddSJeff Bonwick if (!locked) 54088ad4d6ddSJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, error)); 54098ad4d6ddSJeff Bonwick 54108ad4d6ddSJeff Bonwick return (error); 5411fa9e4066Sahrens } 5412fa9e4066Sahrens 5413fa9e4066Sahrens /* 54143d7072f8Seschrock * Find any device that's done replacing, or a vdev marked 'unspare' that's 5415f7170741SWill Andrews * currently spared, so we can detach it. 5416fa9e4066Sahrens */ 5417ea8dc4b6Seschrock static vdev_t * 54183d7072f8Seschrock spa_vdev_resilver_done_hunt(vdev_t *vd) 5419fa9e4066Sahrens { 5420ea8dc4b6Seschrock vdev_t *newvd, *oldvd; 5421fa9e4066Sahrens 5422573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 54233d7072f8Seschrock oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5424ea8dc4b6Seschrock if (oldvd != NULL) 5425ea8dc4b6Seschrock return (oldvd); 5426ea8dc4b6Seschrock } 5427fa9e4066Sahrens 54283d7072f8Seschrock /* 5429cb04b873SMark J Musante * Check for a completed replacement. We always consider the first 5430cb04b873SMark J Musante * vdev in the list to be the oldest vdev, and the last one to be 5431cb04b873SMark J Musante * the newest (see spa_vdev_attach() for how that works). In 5432cb04b873SMark J Musante * the case where the newest vdev is faulted, we will not automatically 5433cb04b873SMark J Musante * remove it after a resilver completes. This is OK as it will require 5434cb04b873SMark J Musante * user intervention to determine which disk the admin wishes to keep. 54353d7072f8Seschrock */ 5436cb04b873SMark J Musante if (vd->vdev_ops == &vdev_replacing_ops) { 5437cb04b873SMark J Musante ASSERT(vd->vdev_children > 1); 5438cb04b873SMark J Musante 5439cb04b873SMark J Musante newvd = vd->vdev_child[vd->vdev_children - 1]; 5440ea8dc4b6Seschrock oldvd = vd->vdev_child[0]; 5441ea8dc4b6Seschrock 54428ad4d6ddSJeff Bonwick if (vdev_dtl_empty(newvd, DTL_MISSING) && 5443e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 54448ad4d6ddSJeff Bonwick !vdev_dtl_required(oldvd)) 5445ea8dc4b6Seschrock return (oldvd); 5446fa9e4066Sahrens } 5447ea8dc4b6Seschrock 54483d7072f8Seschrock /* 54493d7072f8Seschrock * Check for a completed resilver with the 'unspare' flag set. 54503d7072f8Seschrock */ 5451cb04b873SMark J Musante if (vd->vdev_ops == &vdev_spare_ops) { 5452cb04b873SMark J Musante vdev_t *first = vd->vdev_child[0]; 5453cb04b873SMark J Musante vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5454cb04b873SMark J Musante 5455cb04b873SMark J Musante if (last->vdev_unspare) { 5456cb04b873SMark J Musante oldvd = first; 5457cb04b873SMark J Musante newvd = last; 5458cb04b873SMark J Musante } else if (first->vdev_unspare) { 5459cb04b873SMark J Musante oldvd = last; 5460cb04b873SMark J Musante newvd = first; 5461cb04b873SMark J Musante } else { 5462cb04b873SMark J Musante oldvd = NULL; 5463cb04b873SMark J Musante } 54643d7072f8Seschrock 5465cb04b873SMark J Musante if (oldvd != NULL && 54668ad4d6ddSJeff Bonwick vdev_dtl_empty(newvd, DTL_MISSING) && 5467e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 5468cb04b873SMark J Musante !vdev_dtl_required(oldvd)) 54693d7072f8Seschrock return (oldvd); 5470cb04b873SMark J Musante 5471cb04b873SMark J Musante /* 5472cb04b873SMark J Musante * If there are more than two spares attached to a disk, 5473cb04b873SMark J Musante * and those spares are not required, then we want to 5474cb04b873SMark J Musante * attempt to free them up now so that they can be used 5475cb04b873SMark J Musante * by other pools. Once we're back down to a single 5476cb04b873SMark J Musante * disk+spare, we stop removing them. 5477cb04b873SMark J Musante */ 5478cb04b873SMark J Musante if (vd->vdev_children > 2) { 5479cb04b873SMark J Musante newvd = vd->vdev_child[1]; 5480cb04b873SMark J Musante 5481cb04b873SMark J Musante if (newvd->vdev_isspare && last->vdev_isspare && 5482cb04b873SMark J Musante vdev_dtl_empty(last, DTL_MISSING) && 5483cb04b873SMark J Musante vdev_dtl_empty(last, DTL_OUTAGE) && 5484cb04b873SMark J Musante !vdev_dtl_required(newvd)) 5485cb04b873SMark J Musante return (newvd); 54863d7072f8Seschrock } 54873d7072f8Seschrock } 54883d7072f8Seschrock 5489ea8dc4b6Seschrock return (NULL); 5490fa9e4066Sahrens } 5491fa9e4066Sahrens 5492ea8dc4b6Seschrock static void 54933d7072f8Seschrock spa_vdev_resilver_done(spa_t *spa) 5494fa9e4066Sahrens { 54958ad4d6ddSJeff Bonwick vdev_t *vd, *pvd, *ppvd; 54968ad4d6ddSJeff Bonwick uint64_t guid, sguid, pguid, ppguid; 5497ea8dc4b6Seschrock 54988ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5499ea8dc4b6Seschrock 55003d7072f8Seschrock while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 55018ad4d6ddSJeff Bonwick pvd = vd->vdev_parent; 55028ad4d6ddSJeff Bonwick ppvd = pvd->vdev_parent; 5503ea8dc4b6Seschrock guid = vd->vdev_guid; 55048ad4d6ddSJeff Bonwick pguid = pvd->vdev_guid; 55058ad4d6ddSJeff Bonwick ppguid = ppvd->vdev_guid; 55068ad4d6ddSJeff Bonwick sguid = 0; 550799653d4eSeschrock /* 550899653d4eSeschrock * If we have just finished replacing a hot spared device, then 550999653d4eSeschrock * we need to detach the parent's first child (the original hot 551099653d4eSeschrock * spare) as well. 551199653d4eSeschrock */ 5512cb04b873SMark J Musante if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5513cb04b873SMark J Musante ppvd->vdev_children == 2) { 551499653d4eSeschrock ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 55158ad4d6ddSJeff Bonwick sguid = ppvd->vdev_child[1]->vdev_guid; 551699653d4eSeschrock } 5517b4952e17SGeorge Wilson ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5518b4952e17SGeorge Wilson 55198ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 55208ad4d6ddSJeff Bonwick if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5521ea8dc4b6Seschrock return; 55228ad4d6ddSJeff Bonwick if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 552399653d4eSeschrock return; 55248ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5525fa9e4066Sahrens } 5526fa9e4066Sahrens 55278ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 5528fa9e4066Sahrens } 5529fa9e4066Sahrens 5530c67d9675Seschrock /* 5531b3388e4fSEric Taylor * Update the stored path or FRU for this vdev. 5532c67d9675Seschrock */ 5533c67d9675Seschrock int 55346809eb4eSEric Schrock spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 55356809eb4eSEric Schrock boolean_t ispath) 5536c67d9675Seschrock { 5537c5904d13Seschrock vdev_t *vd; 5538208044b8SGeorge Wilson boolean_t sync = B_FALSE; 5539c67d9675Seschrock 5540f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 5541f9af39baSGeorge Wilson 5542b3388e4fSEric Taylor spa_vdev_state_enter(spa, SCL_ALL); 5543c67d9675Seschrock 55446809eb4eSEric Schrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5545b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5546c67d9675Seschrock 55470e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 5548b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 55490e34b6a7Sbonwick 55506809eb4eSEric Schrock if (ispath) { 5551208044b8SGeorge Wilson if (strcmp(value, vd->vdev_path) != 0) { 5552208044b8SGeorge Wilson spa_strfree(vd->vdev_path); 5553208044b8SGeorge Wilson vd->vdev_path = spa_strdup(value); 5554208044b8SGeorge Wilson sync = B_TRUE; 5555208044b8SGeorge Wilson } 55566809eb4eSEric Schrock } else { 5557208044b8SGeorge Wilson if (vd->vdev_fru == NULL) { 5558208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 5559208044b8SGeorge Wilson sync = B_TRUE; 5560208044b8SGeorge Wilson } else if (strcmp(value, vd->vdev_fru) != 0) { 55616809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 5562208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 5563208044b8SGeorge Wilson sync = B_TRUE; 5564208044b8SGeorge Wilson } 55656809eb4eSEric Schrock } 5566c67d9675Seschrock 5567208044b8SGeorge Wilson return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5568c67d9675Seschrock } 5569c67d9675Seschrock 55706809eb4eSEric Schrock int 55716809eb4eSEric Schrock spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 55726809eb4eSEric Schrock { 55736809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 55746809eb4eSEric Schrock } 55756809eb4eSEric Schrock 55766809eb4eSEric Schrock int 55776809eb4eSEric Schrock spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 55786809eb4eSEric Schrock { 55796809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 55806809eb4eSEric Schrock } 55816809eb4eSEric Schrock 5582fa9e4066Sahrens /* 5583fa9e4066Sahrens * ========================================================================== 55843f9d6ad7SLin Ling * SPA Scanning 5585fa9e4066Sahrens * ========================================================================== 5586fa9e4066Sahrens */ 5587fa9e4066Sahrens 5588ea8dc4b6Seschrock int 55893f9d6ad7SLin Ling spa_scan_stop(spa_t *spa) 5590fa9e4066Sahrens { 5591e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 55923f9d6ad7SLin Ling if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5593be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 55943f9d6ad7SLin Ling return (dsl_scan_cancel(spa->spa_dsl_pool)); 55953f9d6ad7SLin Ling } 5596bb8b5132Sek 55973f9d6ad7SLin Ling int 55983f9d6ad7SLin Ling spa_scan(spa_t *spa, pool_scan_func_t func) 55993f9d6ad7SLin Ling { 56003f9d6ad7SLin Ling ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 56013f9d6ad7SLin Ling 56023f9d6ad7SLin Ling if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5603be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 5604fa9e4066Sahrens 5605fa9e4066Sahrens /* 5606088f3894Sahrens * If a resilver was requested, but there is no DTL on a 5607088f3894Sahrens * writeable leaf device, we have nothing to do. 5608fa9e4066Sahrens */ 56093f9d6ad7SLin Ling if (func == POOL_SCAN_RESILVER && 5610088f3894Sahrens !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5611088f3894Sahrens spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5612ea8dc4b6Seschrock return (0); 5613ea8dc4b6Seschrock } 5614fa9e4066Sahrens 56153f9d6ad7SLin Ling return (dsl_scan(spa->spa_dsl_pool, func)); 5616fa9e4066Sahrens } 5617fa9e4066Sahrens 5618ea8dc4b6Seschrock /* 5619ea8dc4b6Seschrock * ========================================================================== 5620ea8dc4b6Seschrock * SPA async task processing 5621ea8dc4b6Seschrock * ========================================================================== 5622ea8dc4b6Seschrock */ 5623ea8dc4b6Seschrock 5624ea8dc4b6Seschrock static void 56253d7072f8Seschrock spa_async_remove(spa_t *spa, vdev_t *vd) 5626fa9e4066Sahrens { 562749cf58c0SBrendan Gregg - Sun Microsystems if (vd->vdev_remove_wanted) { 562898d1cbfeSGeorge Wilson vd->vdev_remove_wanted = B_FALSE; 562998d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 563049cf58c0SBrendan Gregg - Sun Microsystems vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 56311d713200SEric Schrock 56321d713200SEric Schrock /* 56331d713200SEric Schrock * We want to clear the stats, but we don't want to do a full 56341d713200SEric Schrock * vdev_clear() as that will cause us to throw away 56351d713200SEric Schrock * degraded/faulted state as well as attempt to reopen the 56361d713200SEric Schrock * device, all of which is a waste. 56371d713200SEric Schrock */ 56381d713200SEric Schrock vd->vdev_stat.vs_read_errors = 0; 56391d713200SEric Schrock vd->vdev_stat.vs_write_errors = 0; 56401d713200SEric Schrock vd->vdev_stat.vs_checksum_errors = 0; 56411d713200SEric Schrock 5642e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 5643ea8dc4b6Seschrock } 564449cf58c0SBrendan Gregg - Sun Microsystems 5645e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 564649cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, vd->vdev_child[c]); 5647ea8dc4b6Seschrock } 5648fa9e4066Sahrens 5649e14bb325SJeff Bonwick static void 5650e14bb325SJeff Bonwick spa_async_probe(spa_t *spa, vdev_t *vd) 5651e14bb325SJeff Bonwick { 5652e14bb325SJeff Bonwick if (vd->vdev_probe_wanted) { 565398d1cbfeSGeorge Wilson vd->vdev_probe_wanted = B_FALSE; 5654e14bb325SJeff Bonwick vdev_reopen(vd); /* vdev_open() does the actual probe */ 5655e14bb325SJeff Bonwick } 5656e14bb325SJeff Bonwick 5657e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 5658e14bb325SJeff Bonwick spa_async_probe(spa, vd->vdev_child[c]); 5659e14bb325SJeff Bonwick } 5660e14bb325SJeff Bonwick 5661573ca77eSGeorge Wilson static void 5662573ca77eSGeorge Wilson spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5663573ca77eSGeorge Wilson { 5664573ca77eSGeorge Wilson sysevent_id_t eid; 5665573ca77eSGeorge Wilson nvlist_t *attr; 5666573ca77eSGeorge Wilson char *physpath; 5667573ca77eSGeorge Wilson 5668573ca77eSGeorge Wilson if (!spa->spa_autoexpand) 5669573ca77eSGeorge Wilson return; 5670573ca77eSGeorge Wilson 5671573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 5672573ca77eSGeorge Wilson vdev_t *cvd = vd->vdev_child[c]; 5673573ca77eSGeorge Wilson spa_async_autoexpand(spa, cvd); 5674573ca77eSGeorge Wilson } 5675573ca77eSGeorge Wilson 5676573ca77eSGeorge Wilson if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5677573ca77eSGeorge Wilson return; 5678573ca77eSGeorge Wilson 5679573ca77eSGeorge Wilson physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5680573ca77eSGeorge Wilson (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5681573ca77eSGeorge Wilson 5682573ca77eSGeorge Wilson VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5683573ca77eSGeorge Wilson VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5684573ca77eSGeorge Wilson 5685573ca77eSGeorge Wilson (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5686573ca77eSGeorge Wilson ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 5687573ca77eSGeorge Wilson 5688573ca77eSGeorge Wilson nvlist_free(attr); 5689573ca77eSGeorge Wilson kmem_free(physpath, MAXPATHLEN); 5690573ca77eSGeorge Wilson } 5691573ca77eSGeorge Wilson 5692ea8dc4b6Seschrock static void 5693ea8dc4b6Seschrock spa_async_thread(spa_t *spa) 5694ea8dc4b6Seschrock { 5695e14bb325SJeff Bonwick int tasks; 5696ea8dc4b6Seschrock 5697ea8dc4b6Seschrock ASSERT(spa->spa_sync_on); 5698ea8dc4b6Seschrock 5699ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5700ea8dc4b6Seschrock tasks = spa->spa_async_tasks; 5701ea8dc4b6Seschrock spa->spa_async_tasks = 0; 5702ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5703ea8dc4b6Seschrock 57040373e76bSbonwick /* 57050373e76bSbonwick * See if the config needs to be updated. 57060373e76bSbonwick */ 57070373e76bSbonwick if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5708b24ab676SJeff Bonwick uint64_t old_space, new_space; 5709573ca77eSGeorge Wilson 57100373e76bSbonwick mutex_enter(&spa_namespace_lock); 5711b24ab676SJeff Bonwick old_space = metaslab_class_get_space(spa_normal_class(spa)); 57120373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5713b24ab676SJeff Bonwick new_space = metaslab_class_get_space(spa_normal_class(spa)); 57140373e76bSbonwick mutex_exit(&spa_namespace_lock); 5715573ca77eSGeorge Wilson 5716573ca77eSGeorge Wilson /* 5717573ca77eSGeorge Wilson * If the pool grew as a result of the config update, 5718573ca77eSGeorge Wilson * then log an internal history event. 5719573ca77eSGeorge Wilson */ 5720b24ab676SJeff Bonwick if (new_space != old_space) { 57214445fffbSMatthew Ahrens spa_history_log_internal(spa, "vdev online", NULL, 5722c8e1f6d2SMark J Musante "pool '%s' size: %llu(+%llu)", 5723b24ab676SJeff Bonwick spa_name(spa), new_space, new_space - old_space); 5724573ca77eSGeorge Wilson } 57250373e76bSbonwick } 57260373e76bSbonwick 5727ea8dc4b6Seschrock /* 57283d7072f8Seschrock * See if any devices need to be marked REMOVED. 5729ea8dc4b6Seschrock */ 5730e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_REMOVE) { 57318f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 57323d7072f8Seschrock spa_async_remove(spa, spa->spa_root_vdev); 5733e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 573449cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5735e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_spares.sav_count; i++) 573649cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5737e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 5738e14bb325SJeff Bonwick } 5739e14bb325SJeff Bonwick 5740573ca77eSGeorge Wilson if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5741573ca77eSGeorge Wilson spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5742573ca77eSGeorge Wilson spa_async_autoexpand(spa, spa->spa_root_vdev); 5743573ca77eSGeorge Wilson spa_config_exit(spa, SCL_CONFIG, FTAG); 5744573ca77eSGeorge Wilson } 5745573ca77eSGeorge Wilson 5746e14bb325SJeff Bonwick /* 5747e14bb325SJeff Bonwick * See if any devices need to be probed. 5748e14bb325SJeff Bonwick */ 5749e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_PROBE) { 57508f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 5751e14bb325SJeff Bonwick spa_async_probe(spa, spa->spa_root_vdev); 5752e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 57533d7072f8Seschrock } 5754ea8dc4b6Seschrock 5755ea8dc4b6Seschrock /* 5756ea8dc4b6Seschrock * If any devices are done replacing, detach them. 5757ea8dc4b6Seschrock */ 57583d7072f8Seschrock if (tasks & SPA_ASYNC_RESILVER_DONE) 57593d7072f8Seschrock spa_vdev_resilver_done(spa); 5760fa9e4066Sahrens 5761ea8dc4b6Seschrock /* 5762ea8dc4b6Seschrock * Kick off a resilver. 5763ea8dc4b6Seschrock */ 5764088f3894Sahrens if (tasks & SPA_ASYNC_RESILVER) 57653f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, 0); 5766ea8dc4b6Seschrock 5767ea8dc4b6Seschrock /* 5768ea8dc4b6Seschrock * Let the world know that we're done. 5769ea8dc4b6Seschrock */ 5770ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5771ea8dc4b6Seschrock spa->spa_async_thread = NULL; 5772ea8dc4b6Seschrock cv_broadcast(&spa->spa_async_cv); 5773ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5774ea8dc4b6Seschrock thread_exit(); 5775ea8dc4b6Seschrock } 5776ea8dc4b6Seschrock 5777ea8dc4b6Seschrock void 5778ea8dc4b6Seschrock spa_async_suspend(spa_t *spa) 5779ea8dc4b6Seschrock { 5780ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5781ea8dc4b6Seschrock spa->spa_async_suspended++; 5782ea8dc4b6Seschrock while (spa->spa_async_thread != NULL) 5783ea8dc4b6Seschrock cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5784ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5785ea8dc4b6Seschrock } 5786ea8dc4b6Seschrock 5787ea8dc4b6Seschrock void 5788ea8dc4b6Seschrock spa_async_resume(spa_t *spa) 5789ea8dc4b6Seschrock { 5790ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5791ea8dc4b6Seschrock ASSERT(spa->spa_async_suspended != 0); 5792ea8dc4b6Seschrock spa->spa_async_suspended--; 5793ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5794ea8dc4b6Seschrock } 5795ea8dc4b6Seschrock 57963cb69f73SWill Andrews static boolean_t 57973cb69f73SWill Andrews spa_async_tasks_pending(spa_t *spa) 57983cb69f73SWill Andrews { 57993cb69f73SWill Andrews uint_t non_config_tasks; 58003cb69f73SWill Andrews uint_t config_task; 58013cb69f73SWill Andrews boolean_t config_task_suspended; 58023cb69f73SWill Andrews 58033cb69f73SWill Andrews non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; 58043cb69f73SWill Andrews config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 58053cb69f73SWill Andrews if (spa->spa_ccw_fail_time == 0) { 58063cb69f73SWill Andrews config_task_suspended = B_FALSE; 58073cb69f73SWill Andrews } else { 58083cb69f73SWill Andrews config_task_suspended = 58093cb69f73SWill Andrews (gethrtime() - spa->spa_ccw_fail_time) < 58103cb69f73SWill Andrews (zfs_ccw_retry_interval * NANOSEC); 58113cb69f73SWill Andrews } 58123cb69f73SWill Andrews 58133cb69f73SWill Andrews return (non_config_tasks || (config_task && !config_task_suspended)); 58143cb69f73SWill Andrews } 58153cb69f73SWill Andrews 5816ea8dc4b6Seschrock static void 5817ea8dc4b6Seschrock spa_async_dispatch(spa_t *spa) 5818ea8dc4b6Seschrock { 5819ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 58203cb69f73SWill Andrews if (spa_async_tasks_pending(spa) && 58213cb69f73SWill Andrews !spa->spa_async_suspended && 58220373e76bSbonwick spa->spa_async_thread == NULL && 58233cb69f73SWill Andrews rootdir != NULL) 5824ea8dc4b6Seschrock spa->spa_async_thread = thread_create(NULL, 0, 5825ea8dc4b6Seschrock spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5826ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5827ea8dc4b6Seschrock } 5828ea8dc4b6Seschrock 5829ea8dc4b6Seschrock void 5830ea8dc4b6Seschrock spa_async_request(spa_t *spa, int task) 5831ea8dc4b6Seschrock { 58323f9d6ad7SLin Ling zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5833ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5834ea8dc4b6Seschrock spa->spa_async_tasks |= task; 5835ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5836fa9e4066Sahrens } 5837fa9e4066Sahrens 5838fa9e4066Sahrens /* 5839fa9e4066Sahrens * ========================================================================== 5840fa9e4066Sahrens * SPA syncing routines 5841fa9e4066Sahrens * ========================================================================== 5842fa9e4066Sahrens */ 5843fa9e4066Sahrens 5844cde58dbcSMatthew Ahrens static int 5845cde58dbcSMatthew Ahrens bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5846cde58dbcSMatthew Ahrens { 5847cde58dbcSMatthew Ahrens bpobj_t *bpo = arg; 5848cde58dbcSMatthew Ahrens bpobj_enqueue(bpo, bp, tx); 5849cde58dbcSMatthew Ahrens return (0); 5850b24ab676SJeff Bonwick } 5851b24ab676SJeff Bonwick 5852cde58dbcSMatthew Ahrens static int 5853cde58dbcSMatthew Ahrens spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5854b24ab676SJeff Bonwick { 5855b24ab676SJeff Bonwick zio_t *zio = arg; 5856b24ab676SJeff Bonwick 5857b24ab676SJeff Bonwick zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5858b24ab676SJeff Bonwick zio->io_flags)); 5859cde58dbcSMatthew Ahrens return (0); 5860fa9e4066Sahrens } 5861fa9e4066Sahrens 586269962b56SMatthew Ahrens /* 586369962b56SMatthew Ahrens * Note: this simple function is not inlined to make it easier to dtrace the 586469962b56SMatthew Ahrens * amount of time spent syncing frees. 586569962b56SMatthew Ahrens */ 586669962b56SMatthew Ahrens static void 586769962b56SMatthew Ahrens spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 586869962b56SMatthew Ahrens { 586969962b56SMatthew Ahrens zio_t *zio = zio_root(spa, NULL, NULL, 0); 587069962b56SMatthew Ahrens bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 587169962b56SMatthew Ahrens VERIFY(zio_wait(zio) == 0); 587269962b56SMatthew Ahrens } 587369962b56SMatthew Ahrens 587469962b56SMatthew Ahrens /* 587569962b56SMatthew Ahrens * Note: this simple function is not inlined to make it easier to dtrace the 587669962b56SMatthew Ahrens * amount of time spent syncing deferred frees. 587769962b56SMatthew Ahrens */ 587869962b56SMatthew Ahrens static void 587969962b56SMatthew Ahrens spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 588069962b56SMatthew Ahrens { 588169962b56SMatthew Ahrens zio_t *zio = zio_root(spa, NULL, NULL, 0); 588269962b56SMatthew Ahrens VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 588369962b56SMatthew Ahrens spa_free_sync_cb, zio, tx), ==, 0); 588469962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 588569962b56SMatthew Ahrens } 588669962b56SMatthew Ahrens 588769962b56SMatthew Ahrens 5888fa9e4066Sahrens static void 588999653d4eSeschrock spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5890fa9e4066Sahrens { 5891fa9e4066Sahrens char *packed = NULL; 5892f7991ba4STim Haley size_t bufsize; 5893fa9e4066Sahrens size_t nvsize = 0; 5894fa9e4066Sahrens dmu_buf_t *db; 5895fa9e4066Sahrens 589699653d4eSeschrock VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5897fa9e4066Sahrens 5898f7991ba4STim Haley /* 5899f7991ba4STim Haley * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 590043466aaeSMax Grossman * information. This avoids the dmu_buf_will_dirty() path and 5901f7991ba4STim Haley * saves us a pre-read to get data we don't actually care about. 5902f7991ba4STim Haley */ 5903ad135b5dSChristopher Siden bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5904f7991ba4STim Haley packed = kmem_alloc(bufsize, KM_SLEEP); 5905fa9e4066Sahrens 590699653d4eSeschrock VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5907ea8dc4b6Seschrock KM_SLEEP) == 0); 5908f7991ba4STim Haley bzero(packed + nvsize, bufsize - nvsize); 5909fa9e4066Sahrens 5910f7991ba4STim Haley dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5911fa9e4066Sahrens 5912f7991ba4STim Haley kmem_free(packed, bufsize); 5913fa9e4066Sahrens 591499653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5915fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 5916fa9e4066Sahrens *(uint64_t *)db->db_data = nvsize; 5917ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 5918fa9e4066Sahrens } 5919fa9e4066Sahrens 592099653d4eSeschrock static void 5921fa94a07fSbrendan spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5922fa94a07fSbrendan const char *config, const char *entry) 592399653d4eSeschrock { 592499653d4eSeschrock nvlist_t *nvroot; 5925fa94a07fSbrendan nvlist_t **list; 592699653d4eSeschrock int i; 592799653d4eSeschrock 5928fa94a07fSbrendan if (!sav->sav_sync) 592999653d4eSeschrock return; 593099653d4eSeschrock 593199653d4eSeschrock /* 5932fa94a07fSbrendan * Update the MOS nvlist describing the list of available devices. 5933fa94a07fSbrendan * spa_validate_aux() will have already made sure this nvlist is 59343d7072f8Seschrock * valid and the vdevs are labeled appropriately. 593599653d4eSeschrock */ 5936fa94a07fSbrendan if (sav->sav_object == 0) { 5937fa94a07fSbrendan sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5938fa94a07fSbrendan DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5939fa94a07fSbrendan sizeof (uint64_t), tx); 594099653d4eSeschrock VERIFY(zap_update(spa->spa_meta_objset, 5941fa94a07fSbrendan DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5942fa94a07fSbrendan &sav->sav_object, tx) == 0); 594399653d4eSeschrock } 594499653d4eSeschrock 594599653d4eSeschrock VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5946fa94a07fSbrendan if (sav->sav_count == 0) { 5947fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 594899653d4eSeschrock } else { 5949fa94a07fSbrendan list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5950fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 5951fa94a07fSbrendan list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 59523f9d6ad7SLin Ling B_FALSE, VDEV_CONFIG_L2CACHE); 5953fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5954fa94a07fSbrendan sav->sav_count) == 0); 5955fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 5956fa94a07fSbrendan nvlist_free(list[i]); 5957fa94a07fSbrendan kmem_free(list, sav->sav_count * sizeof (void *)); 595899653d4eSeschrock } 595999653d4eSeschrock 5960fa94a07fSbrendan spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 596106eeb2adSek nvlist_free(nvroot); 596299653d4eSeschrock 5963fa94a07fSbrendan sav->sav_sync = B_FALSE; 596499653d4eSeschrock } 596599653d4eSeschrock 596699653d4eSeschrock static void 596799653d4eSeschrock spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 596899653d4eSeschrock { 596999653d4eSeschrock nvlist_t *config; 597099653d4eSeschrock 5971e14bb325SJeff Bonwick if (list_is_empty(&spa->spa_config_dirty_list)) 597299653d4eSeschrock return; 597399653d4eSeschrock 5974e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5975e14bb325SJeff Bonwick 5976e14bb325SJeff Bonwick config = spa_config_generate(spa, spa->spa_root_vdev, 5977e14bb325SJeff Bonwick dmu_tx_get_txg(tx), B_FALSE); 5978e14bb325SJeff Bonwick 597925345e46SGeorge Wilson /* 598025345e46SGeorge Wilson * If we're upgrading the spa version then make sure that 598125345e46SGeorge Wilson * the config object gets updated with the correct version. 598225345e46SGeorge Wilson */ 598325345e46SGeorge Wilson if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 598425345e46SGeorge Wilson fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 598525345e46SGeorge Wilson spa->spa_uberblock.ub_version); 598625345e46SGeorge Wilson 5987e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 598899653d4eSeschrock 598999653d4eSeschrock if (spa->spa_config_syncing) 599099653d4eSeschrock nvlist_free(spa->spa_config_syncing); 599199653d4eSeschrock spa->spa_config_syncing = config; 599299653d4eSeschrock 599399653d4eSeschrock spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 599499653d4eSeschrock } 599599653d4eSeschrock 5996ad135b5dSChristopher Siden static void 59973b2aab18SMatthew Ahrens spa_sync_version(void *arg, dmu_tx_t *tx) 5998ad135b5dSChristopher Siden { 59993b2aab18SMatthew Ahrens uint64_t *versionp = arg; 60003b2aab18SMatthew Ahrens uint64_t version = *versionp; 60013b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6002ad135b5dSChristopher Siden 6003ad135b5dSChristopher Siden /* 6004ad135b5dSChristopher Siden * Setting the version is special cased when first creating the pool. 6005ad135b5dSChristopher Siden */ 6006ad135b5dSChristopher Siden ASSERT(tx->tx_txg != TXG_INITIAL); 6007ad135b5dSChristopher Siden 600862eae887SRichard Yao ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6009ad135b5dSChristopher Siden ASSERT(version >= spa_version(spa)); 6010ad135b5dSChristopher Siden 6011ad135b5dSChristopher Siden spa->spa_uberblock.ub_version = version; 6012ad135b5dSChristopher Siden vdev_config_dirty(spa->spa_root_vdev); 60134445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6014ad135b5dSChristopher Siden } 6015ad135b5dSChristopher Siden 6016990b4856Slling /* 6017990b4856Slling * Set zpool properties. 6018990b4856Slling */ 6019b1b8ab34Slling static void 60203b2aab18SMatthew Ahrens spa_sync_props(void *arg, dmu_tx_t *tx) 6021b1b8ab34Slling { 60223b2aab18SMatthew Ahrens nvlist_t *nvp = arg; 60233b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6024b1b8ab34Slling objset_t *mos = spa->spa_meta_objset; 6025ad135b5dSChristopher Siden nvpair_t *elem = NULL; 6026b1b8ab34Slling 6027e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 6028e14bb325SJeff Bonwick 6029990b4856Slling while ((elem = nvlist_next_nvpair(nvp, elem))) { 6030ad135b5dSChristopher Siden uint64_t intval; 6031ad135b5dSChristopher Siden char *strval, *fname; 6032ad135b5dSChristopher Siden zpool_prop_t prop; 6033ad135b5dSChristopher Siden const char *propname; 6034ad135b5dSChristopher Siden zprop_type_t proptype; 60352acef22dSMatthew Ahrens spa_feature_t fid; 6036ad135b5dSChristopher Siden 6037990b4856Slling switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6038ad135b5dSChristopher Siden case ZPROP_INVAL: 6039ad135b5dSChristopher Siden /* 6040ad135b5dSChristopher Siden * We checked this earlier in spa_prop_validate(). 6041ad135b5dSChristopher Siden */ 6042ad135b5dSChristopher Siden ASSERT(zpool_prop_feature(nvpair_name(elem))); 6043ad135b5dSChristopher Siden 6044ad135b5dSChristopher Siden fname = strchr(nvpair_name(elem), '@') + 1; 60452acef22dSMatthew Ahrens VERIFY0(zfeature_lookup_name(fname, &fid)); 6046ad135b5dSChristopher Siden 60472acef22dSMatthew Ahrens spa_feature_enable(spa, fid, tx); 60484445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 60494445fffbSMatthew Ahrens "%s=enabled", nvpair_name(elem)); 6050ad135b5dSChristopher Siden break; 6051ad135b5dSChristopher Siden 6052990b4856Slling case ZPOOL_PROP_VERSION: 60530713e232SGeorge Wilson intval = fnvpair_value_uint64(elem); 6054990b4856Slling /* 6055ad135b5dSChristopher Siden * The version is synced seperatly before other 6056ad135b5dSChristopher Siden * properties and should be correct by now. 6057990b4856Slling */ 6058ad135b5dSChristopher Siden ASSERT3U(spa_version(spa), >=, intval); 6059ecd6cf80Smarks break; 6060990b4856Slling 6061990b4856Slling case ZPOOL_PROP_ALTROOT: 6062990b4856Slling /* 6063990b4856Slling * 'altroot' is a non-persistent property. It should 6064990b4856Slling * have been set temporarily at creation or import time. 6065990b4856Slling */ 6066990b4856Slling ASSERT(spa->spa_root != NULL); 6067b1b8ab34Slling break; 60683d7072f8Seschrock 6069f9af39baSGeorge Wilson case ZPOOL_PROP_READONLY: 60702f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 6071990b4856Slling /* 6072f9af39baSGeorge Wilson * 'readonly' and 'cachefile' are also non-persisitent 6073f9af39baSGeorge Wilson * properties. 6074990b4856Slling */ 60753d7072f8Seschrock break; 60768704186eSDan McDonald case ZPOOL_PROP_COMMENT: 60770713e232SGeorge Wilson strval = fnvpair_value_string(elem); 60788704186eSDan McDonald if (spa->spa_comment != NULL) 60798704186eSDan McDonald spa_strfree(spa->spa_comment); 60808704186eSDan McDonald spa->spa_comment = spa_strdup(strval); 60818704186eSDan McDonald /* 60828704186eSDan McDonald * We need to dirty the configuration on all the vdevs 60838704186eSDan McDonald * so that their labels get updated. It's unnecessary 60848704186eSDan McDonald * to do this for pool creation since the vdev's 60858704186eSDan McDonald * configuratoin has already been dirtied. 60868704186eSDan McDonald */ 60878704186eSDan McDonald if (tx->tx_txg != TXG_INITIAL) 60888704186eSDan McDonald vdev_config_dirty(spa->spa_root_vdev); 60894445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 60904445fffbSMatthew Ahrens "%s=%s", nvpair_name(elem), strval); 60918704186eSDan McDonald break; 6092990b4856Slling default: 6093990b4856Slling /* 6094990b4856Slling * Set pool property values in the poolprops mos object. 6095990b4856Slling */ 6096990b4856Slling if (spa->spa_pool_props_object == 0) { 6097ad135b5dSChristopher Siden spa->spa_pool_props_object = 6098ad135b5dSChristopher Siden zap_create_link(mos, DMU_OT_POOL_PROPS, 6099990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6100ad135b5dSChristopher Siden tx); 6101990b4856Slling } 6102990b4856Slling 6103990b4856Slling /* normalize the property name */ 6104990b4856Slling propname = zpool_prop_to_name(prop); 6105990b4856Slling proptype = zpool_prop_get_type(prop); 6106990b4856Slling 6107990b4856Slling if (nvpair_type(elem) == DATA_TYPE_STRING) { 6108990b4856Slling ASSERT(proptype == PROP_TYPE_STRING); 61090713e232SGeorge Wilson strval = fnvpair_value_string(elem); 61100713e232SGeorge Wilson VERIFY0(zap_update(mos, 6111990b4856Slling spa->spa_pool_props_object, propname, 61120713e232SGeorge Wilson 1, strlen(strval) + 1, strval, tx)); 61134445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 61144445fffbSMatthew Ahrens "%s=%s", nvpair_name(elem), strval); 6115990b4856Slling } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 61160713e232SGeorge Wilson intval = fnvpair_value_uint64(elem); 6117990b4856Slling 6118990b4856Slling if (proptype == PROP_TYPE_INDEX) { 6119990b4856Slling const char *unused; 61200713e232SGeorge Wilson VERIFY0(zpool_prop_index_to_string( 61210713e232SGeorge Wilson prop, intval, &unused)); 6122990b4856Slling } 61230713e232SGeorge Wilson VERIFY0(zap_update(mos, 6124990b4856Slling spa->spa_pool_props_object, propname, 61250713e232SGeorge Wilson 8, 1, &intval, tx)); 61264445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 61274445fffbSMatthew Ahrens "%s=%lld", nvpair_name(elem), intval); 6128990b4856Slling } else { 6129990b4856Slling ASSERT(0); /* not allowed */ 6130990b4856Slling } 6131990b4856Slling 61320a4e9518Sgw switch (prop) { 61330a4e9518Sgw case ZPOOL_PROP_DELEGATION: 6134990b4856Slling spa->spa_delegation = intval; 61350a4e9518Sgw break; 61360a4e9518Sgw case ZPOOL_PROP_BOOTFS: 6137990b4856Slling spa->spa_bootfs = intval; 61380a4e9518Sgw break; 61390a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 61400a4e9518Sgw spa->spa_failmode = intval; 61410a4e9518Sgw break; 6142573ca77eSGeorge Wilson case ZPOOL_PROP_AUTOEXPAND: 6143573ca77eSGeorge Wilson spa->spa_autoexpand = intval; 6144b98131cfSEric Taylor if (tx->tx_txg != TXG_INITIAL) 6145b98131cfSEric Taylor spa_async_request(spa, 6146b98131cfSEric Taylor SPA_ASYNC_AUTOEXPAND); 6147573ca77eSGeorge Wilson break; 6148b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 6149b24ab676SJeff Bonwick spa->spa_dedup_ditto = intval; 6150b24ab676SJeff Bonwick break; 61510a4e9518Sgw default: 61520a4e9518Sgw break; 61530a4e9518Sgw } 6154990b4856Slling } 6155990b4856Slling 6156b1b8ab34Slling } 6157e14bb325SJeff Bonwick 6158e14bb325SJeff Bonwick mutex_exit(&spa->spa_props_lock); 6159b1b8ab34Slling } 6160b1b8ab34Slling 6161cde58dbcSMatthew Ahrens /* 6162cde58dbcSMatthew Ahrens * Perform one-time upgrade on-disk changes. spa_version() does not 6163cde58dbcSMatthew Ahrens * reflect the new version this txg, so there must be no changes this 6164cde58dbcSMatthew Ahrens * txg to anything that the upgrade code depends on after it executes. 6165cde58dbcSMatthew Ahrens * Therefore this must be called after dsl_pool_sync() does the sync 6166cde58dbcSMatthew Ahrens * tasks. 6167cde58dbcSMatthew Ahrens */ 6168cde58dbcSMatthew Ahrens static void 6169cde58dbcSMatthew Ahrens spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6170cde58dbcSMatthew Ahrens { 6171cde58dbcSMatthew Ahrens dsl_pool_t *dp = spa->spa_dsl_pool; 6172cde58dbcSMatthew Ahrens 6173cde58dbcSMatthew Ahrens ASSERT(spa->spa_sync_pass == 1); 6174cde58dbcSMatthew Ahrens 61753b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 61763b2aab18SMatthew Ahrens 6177cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6178cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6179cde58dbcSMatthew Ahrens dsl_pool_create_origin(dp, tx); 6180cde58dbcSMatthew Ahrens 6181cde58dbcSMatthew Ahrens /* Keeping the origin open increases spa_minref */ 6182cde58dbcSMatthew Ahrens spa->spa_minref += 3; 6183cde58dbcSMatthew Ahrens } 6184cde58dbcSMatthew Ahrens 6185cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6186cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6187cde58dbcSMatthew Ahrens dsl_pool_upgrade_clones(dp, tx); 6188cde58dbcSMatthew Ahrens } 6189cde58dbcSMatthew Ahrens 6190cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6191cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6192cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dp, tx); 6193cde58dbcSMatthew Ahrens 6194cde58dbcSMatthew Ahrens /* Keeping the freedir open increases spa_minref */ 6195cde58dbcSMatthew Ahrens spa->spa_minref += 3; 6196cde58dbcSMatthew Ahrens } 6197ad135b5dSChristopher Siden 6198ad135b5dSChristopher Siden if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6199ad135b5dSChristopher Siden spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6200ad135b5dSChristopher Siden spa_feature_create_zap_objects(spa, tx); 6201ad135b5dSChristopher Siden } 6202b8289d24SDaniil Lunev 6203b8289d24SDaniil Lunev /* 6204b8289d24SDaniil Lunev * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6205b8289d24SDaniil Lunev * when possibility to use lz4 compression for metadata was added 6206b8289d24SDaniil Lunev * Old pools that have this feature enabled must be upgraded to have 6207b8289d24SDaniil Lunev * this feature active 6208b8289d24SDaniil Lunev */ 6209b8289d24SDaniil Lunev if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6210b8289d24SDaniil Lunev boolean_t lz4_en = spa_feature_is_enabled(spa, 6211b8289d24SDaniil Lunev SPA_FEATURE_LZ4_COMPRESS); 6212b8289d24SDaniil Lunev boolean_t lz4_ac = spa_feature_is_active(spa, 6213b8289d24SDaniil Lunev SPA_FEATURE_LZ4_COMPRESS); 6214b8289d24SDaniil Lunev 6215b8289d24SDaniil Lunev if (lz4_en && !lz4_ac) 6216b8289d24SDaniil Lunev spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6217b8289d24SDaniil Lunev } 62183b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 6219cde58dbcSMatthew Ahrens } 6220cde58dbcSMatthew Ahrens 6221fa9e4066Sahrens /* 6222fa9e4066Sahrens * Sync the specified transaction group. New blocks may be dirtied as 6223fa9e4066Sahrens * part of the process, so we iterate until it converges. 6224fa9e4066Sahrens */ 6225fa9e4066Sahrens void 6226fa9e4066Sahrens spa_sync(spa_t *spa, uint64_t txg) 6227fa9e4066Sahrens { 6228fa9e4066Sahrens dsl_pool_t *dp = spa->spa_dsl_pool; 6229fa9e4066Sahrens objset_t *mos = spa->spa_meta_objset; 6230b24ab676SJeff Bonwick bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 62310373e76bSbonwick vdev_t *rvd = spa->spa_root_vdev; 6232fa9e4066Sahrens vdev_t *vd; 6233fa9e4066Sahrens dmu_tx_t *tx; 6234e14bb325SJeff Bonwick int error; 6235fa9e4066Sahrens 6236f9af39baSGeorge Wilson VERIFY(spa_writeable(spa)); 6237f9af39baSGeorge Wilson 6238fa9e4066Sahrens /* 6239fa9e4066Sahrens * Lock out configuration changes. 6240fa9e4066Sahrens */ 6241e14bb325SJeff Bonwick spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6242fa9e4066Sahrens 6243fa9e4066Sahrens spa->spa_syncing_txg = txg; 6244fa9e4066Sahrens spa->spa_sync_pass = 0; 6245fa9e4066Sahrens 6246e14bb325SJeff Bonwick /* 6247e14bb325SJeff Bonwick * If there are any pending vdev state changes, convert them 6248e14bb325SJeff Bonwick * into config changes that go out with this transaction group. 6249e14bb325SJeff Bonwick */ 6250e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 62518ad4d6ddSJeff Bonwick while (list_head(&spa->spa_state_dirty_list) != NULL) { 62528ad4d6ddSJeff Bonwick /* 62538ad4d6ddSJeff Bonwick * We need the write lock here because, for aux vdevs, 62548ad4d6ddSJeff Bonwick * calling vdev_config_dirty() modifies sav_config. 62558ad4d6ddSJeff Bonwick * This is ugly and will become unnecessary when we 62568ad4d6ddSJeff Bonwick * eliminate the aux vdev wart by integrating all vdevs 62578ad4d6ddSJeff Bonwick * into the root vdev tree. 62588ad4d6ddSJeff Bonwick */ 62598ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 62608ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 62618ad4d6ddSJeff Bonwick while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 62628ad4d6ddSJeff Bonwick vdev_state_clean(vd); 62638ad4d6ddSJeff Bonwick vdev_config_dirty(vd); 62648ad4d6ddSJeff Bonwick } 62658ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 62668ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6267e14bb325SJeff Bonwick } 6268e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 6269e14bb325SJeff Bonwick 627099653d4eSeschrock tx = dmu_tx_create_assigned(dp, txg); 627199653d4eSeschrock 6272283b8460SGeorge.Wilson spa->spa_sync_starttime = gethrtime(); 6273283b8460SGeorge.Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6274283b8460SGeorge.Wilson spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6275283b8460SGeorge.Wilson 627699653d4eSeschrock /* 6277e7437265Sahrens * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 627899653d4eSeschrock * set spa_deflate if we have no raid-z vdevs. 627999653d4eSeschrock */ 6280e7437265Sahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6281e7437265Sahrens spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 628299653d4eSeschrock int i; 628399653d4eSeschrock 628499653d4eSeschrock for (i = 0; i < rvd->vdev_children; i++) { 628599653d4eSeschrock vd = rvd->vdev_child[i]; 628699653d4eSeschrock if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 628799653d4eSeschrock break; 628899653d4eSeschrock } 628999653d4eSeschrock if (i == rvd->vdev_children) { 629099653d4eSeschrock spa->spa_deflate = TRUE; 629199653d4eSeschrock VERIFY(0 == zap_add(spa->spa_meta_objset, 629299653d4eSeschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 629399653d4eSeschrock sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 629499653d4eSeschrock } 629599653d4eSeschrock } 629699653d4eSeschrock 6297fa9e4066Sahrens /* 6298fa9e4066Sahrens * Iterate to convergence. 6299fa9e4066Sahrens */ 6300fa9e4066Sahrens do { 6301b24ab676SJeff Bonwick int pass = ++spa->spa_sync_pass; 6302fa9e4066Sahrens 6303fa9e4066Sahrens spa_sync_config_object(spa, tx); 6304fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6305fa94a07fSbrendan ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6306fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6307fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6308ea8dc4b6Seschrock spa_errlog_sync(spa, txg); 6309fa9e4066Sahrens dsl_pool_sync(dp, txg); 6310fa9e4066Sahrens 631101f55e48SGeorge Wilson if (pass < zfs_sync_pass_deferred_free) { 631269962b56SMatthew Ahrens spa_sync_frees(spa, free_bpl, tx); 6313b24ab676SJeff Bonwick } else { 6314231aab85SMatthew Ahrens /* 6315231aab85SMatthew Ahrens * We can not defer frees in pass 1, because 6316231aab85SMatthew Ahrens * we sync the deferred frees later in pass 1. 6317231aab85SMatthew Ahrens */ 6318231aab85SMatthew Ahrens ASSERT3U(pass, >, 1); 6319cde58dbcSMatthew Ahrens bplist_iterate(free_bpl, bpobj_enqueue_cb, 632069962b56SMatthew Ahrens &spa->spa_deferred_bpobj, tx); 6321fa9e4066Sahrens } 6322fa9e4066Sahrens 6323b24ab676SJeff Bonwick ddt_sync(spa, txg); 63243f9d6ad7SLin Ling dsl_scan_sync(dp, tx); 6325afee20e4SGeorge Wilson 6326b24ab676SJeff Bonwick while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6327b24ab676SJeff Bonwick vdev_sync(vd, txg); 6328b24ab676SJeff Bonwick 6329231aab85SMatthew Ahrens if (pass == 1) { 6330cde58dbcSMatthew Ahrens spa_sync_upgrades(spa, tx); 6331231aab85SMatthew Ahrens ASSERT3U(txg, >=, 6332231aab85SMatthew Ahrens spa->spa_uberblock.ub_rootbp.blk_birth); 6333231aab85SMatthew Ahrens /* 6334231aab85SMatthew Ahrens * Note: We need to check if the MOS is dirty 6335231aab85SMatthew Ahrens * because we could have marked the MOS dirty 6336231aab85SMatthew Ahrens * without updating the uberblock (e.g. if we 6337231aab85SMatthew Ahrens * have sync tasks but no dirty user data). We 6338231aab85SMatthew Ahrens * need to check the uberblock's rootbp because 6339231aab85SMatthew Ahrens * it is updated if we have synced out dirty 6340231aab85SMatthew Ahrens * data (though in this case the MOS will most 6341231aab85SMatthew Ahrens * likely also be dirty due to second order 6342231aab85SMatthew Ahrens * effects, we don't want to rely on that here). 6343231aab85SMatthew Ahrens */ 6344231aab85SMatthew Ahrens if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6345231aab85SMatthew Ahrens !dmu_objset_is_dirty(mos, txg)) { 6346231aab85SMatthew Ahrens /* 6347231aab85SMatthew Ahrens * Nothing changed on the first pass, 6348231aab85SMatthew Ahrens * therefore this TXG is a no-op. Avoid 6349231aab85SMatthew Ahrens * syncing deferred frees, so that we 6350231aab85SMatthew Ahrens * can keep this TXG as a no-op. 6351231aab85SMatthew Ahrens */ 6352231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6353231aab85SMatthew Ahrens txg)); 6354231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6355231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6356231aab85SMatthew Ahrens break; 6357231aab85SMatthew Ahrens } 6358231aab85SMatthew Ahrens spa_sync_deferred_frees(spa, tx); 6359231aab85SMatthew Ahrens } 6360fa9e4066Sahrens 6361cde58dbcSMatthew Ahrens } while (dmu_objset_is_dirty(mos, txg)); 6362fa9e4066Sahrens 6363fa9e4066Sahrens /* 6364fa9e4066Sahrens * Rewrite the vdev configuration (which includes the uberblock) 6365fa9e4066Sahrens * to commit the transaction group. 63660373e76bSbonwick * 636717f17c2dSbonwick * If there are no dirty vdevs, we sync the uberblock to a few 636817f17c2dSbonwick * random top-level vdevs that are known to be visible in the 6369e14bb325SJeff Bonwick * config cache (see spa_vdev_add() for a complete description). 6370e14bb325SJeff Bonwick * If there *are* dirty vdevs, sync the uberblock to all vdevs. 63710373e76bSbonwick */ 6372e14bb325SJeff Bonwick for (;;) { 6373e14bb325SJeff Bonwick /* 6374e14bb325SJeff Bonwick * We hold SCL_STATE to prevent vdev open/close/etc. 6375e14bb325SJeff Bonwick * while we're attempting to write the vdev labels. 6376e14bb325SJeff Bonwick */ 6377e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6378e14bb325SJeff Bonwick 6379e14bb325SJeff Bonwick if (list_is_empty(&spa->spa_config_dirty_list)) { 6380e14bb325SJeff Bonwick vdev_t *svd[SPA_DVAS_PER_BP]; 6381e14bb325SJeff Bonwick int svdcount = 0; 6382e14bb325SJeff Bonwick int children = rvd->vdev_children; 6383e14bb325SJeff Bonwick int c0 = spa_get_random(children); 6384e14bb325SJeff Bonwick 6385573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 6386e14bb325SJeff Bonwick vd = rvd->vdev_child[(c0 + c) % children]; 6387e14bb325SJeff Bonwick if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6388e14bb325SJeff Bonwick continue; 6389e14bb325SJeff Bonwick svd[svdcount++] = vd; 6390e14bb325SJeff Bonwick if (svdcount == SPA_DVAS_PER_BP) 6391e14bb325SJeff Bonwick break; 6392e14bb325SJeff Bonwick } 63938956713aSEric Schrock error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 63948956713aSEric Schrock if (error != 0) 63958956713aSEric Schrock error = vdev_config_sync(svd, svdcount, txg, 63968956713aSEric Schrock B_TRUE); 6397e14bb325SJeff Bonwick } else { 6398e14bb325SJeff Bonwick error = vdev_config_sync(rvd->vdev_child, 63998956713aSEric Schrock rvd->vdev_children, txg, B_FALSE); 64008956713aSEric Schrock if (error != 0) 64018956713aSEric Schrock error = vdev_config_sync(rvd->vdev_child, 64028956713aSEric Schrock rvd->vdev_children, txg, B_TRUE); 64030373e76bSbonwick } 6404e14bb325SJeff Bonwick 6405dfbb9432SGeorge Wilson if (error == 0) 6406dfbb9432SGeorge Wilson spa->spa_last_synced_guid = rvd->vdev_guid; 6407dfbb9432SGeorge Wilson 6408e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 6409e14bb325SJeff Bonwick 6410e14bb325SJeff Bonwick if (error == 0) 6411e14bb325SJeff Bonwick break; 6412e14bb325SJeff Bonwick zio_suspend(spa, NULL); 6413e14bb325SJeff Bonwick zio_resume_wait(spa); 64140373e76bSbonwick } 641599653d4eSeschrock dmu_tx_commit(tx); 641699653d4eSeschrock 6417283b8460SGeorge.Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6418283b8460SGeorge.Wilson 64190373e76bSbonwick /* 64200373e76bSbonwick * Clear the dirty config list. 6421fa9e4066Sahrens */ 6422e14bb325SJeff Bonwick while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 64230373e76bSbonwick vdev_config_clean(vd); 64240373e76bSbonwick 64250373e76bSbonwick /* 64260373e76bSbonwick * Now that the new config has synced transactionally, 64270373e76bSbonwick * let it become visible to the config cache. 64280373e76bSbonwick */ 64290373e76bSbonwick if (spa->spa_config_syncing != NULL) { 64300373e76bSbonwick spa_config_set(spa, spa->spa_config_syncing); 64310373e76bSbonwick spa->spa_config_txg = txg; 64320373e76bSbonwick spa->spa_config_syncing = NULL; 64330373e76bSbonwick } 6434fa9e4066Sahrens 6435fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 6436fa9e4066Sahrens 6437b24ab676SJeff Bonwick dsl_pool_sync_done(dp, txg); 6438fa9e4066Sahrens 6439fa9e4066Sahrens /* 6440fa9e4066Sahrens * Update usable space statistics. 6441fa9e4066Sahrens */ 6442fa9e4066Sahrens while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6443fa9e4066Sahrens vdev_sync_done(vd, txg); 6444fa9e4066Sahrens 6445485bbbf5SGeorge Wilson spa_update_dspace(spa); 6446485bbbf5SGeorge Wilson 6447fa9e4066Sahrens /* 6448fa9e4066Sahrens * It had better be the case that we didn't dirty anything 644999653d4eSeschrock * since vdev_config_sync(). 6450fa9e4066Sahrens */ 6451fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6452fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6453fa9e4066Sahrens ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6454b24ab676SJeff Bonwick 6455b24ab676SJeff Bonwick spa->spa_sync_pass = 0; 6456fa9e4066Sahrens 6457e14bb325SJeff Bonwick spa_config_exit(spa, SCL_CONFIG, FTAG); 6458ea8dc4b6Seschrock 6459468c413aSTim Haley spa_handle_ignored_writes(spa); 6460468c413aSTim Haley 6461ea8dc4b6Seschrock /* 6462ea8dc4b6Seschrock * If any async tasks have been requested, kick them off. 6463ea8dc4b6Seschrock */ 6464ea8dc4b6Seschrock spa_async_dispatch(spa); 6465fa9e4066Sahrens } 6466fa9e4066Sahrens 6467fa9e4066Sahrens /* 6468fa9e4066Sahrens * Sync all pools. We don't want to hold the namespace lock across these 6469fa9e4066Sahrens * operations, so we take a reference on the spa_t and drop the lock during the 6470fa9e4066Sahrens * sync. 6471fa9e4066Sahrens */ 6472fa9e4066Sahrens void 6473fa9e4066Sahrens spa_sync_allpools(void) 6474fa9e4066Sahrens { 6475fa9e4066Sahrens spa_t *spa = NULL; 6476fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6477fa9e4066Sahrens while ((spa = spa_next(spa)) != NULL) { 6478f9af39baSGeorge Wilson if (spa_state(spa) != POOL_STATE_ACTIVE || 6479f9af39baSGeorge Wilson !spa_writeable(spa) || spa_suspended(spa)) 6480fa9e4066Sahrens continue; 6481fa9e4066Sahrens spa_open_ref(spa, FTAG); 6482fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6483fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), 0); 6484fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6485fa9e4066Sahrens spa_close(spa, FTAG); 6486fa9e4066Sahrens } 6487fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6488fa9e4066Sahrens } 6489fa9e4066Sahrens 6490fa9e4066Sahrens /* 6491fa9e4066Sahrens * ========================================================================== 6492fa9e4066Sahrens * Miscellaneous routines 6493fa9e4066Sahrens * ========================================================================== 6494fa9e4066Sahrens */ 6495fa9e4066Sahrens 6496fa9e4066Sahrens /* 6497fa9e4066Sahrens * Remove all pools in the system. 6498fa9e4066Sahrens */ 6499fa9e4066Sahrens void 6500fa9e4066Sahrens spa_evict_all(void) 6501fa9e4066Sahrens { 6502fa9e4066Sahrens spa_t *spa; 6503fa9e4066Sahrens 6504fa9e4066Sahrens /* 6505fa9e4066Sahrens * Remove all cached state. All pools should be closed now, 6506fa9e4066Sahrens * so every spa in the AVL tree should be unreferenced. 6507fa9e4066Sahrens */ 6508fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6509fa9e4066Sahrens while ((spa = spa_next(NULL)) != NULL) { 6510fa9e4066Sahrens /* 6511ea8dc4b6Seschrock * Stop async tasks. The async thread may need to detach 6512ea8dc4b6Seschrock * a device that's been replaced, which requires grabbing 6513ea8dc4b6Seschrock * spa_namespace_lock, so we must drop it here. 6514fa9e4066Sahrens */ 6515fa9e4066Sahrens spa_open_ref(spa, FTAG); 6516fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6517ea8dc4b6Seschrock spa_async_suspend(spa); 6518fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6519fa9e4066Sahrens spa_close(spa, FTAG); 6520fa9e4066Sahrens 6521fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6522fa9e4066Sahrens spa_unload(spa); 6523fa9e4066Sahrens spa_deactivate(spa); 6524fa9e4066Sahrens } 6525fa9e4066Sahrens spa_remove(spa); 6526fa9e4066Sahrens } 6527fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6528fa9e4066Sahrens } 6529ea8dc4b6Seschrock 6530ea8dc4b6Seschrock vdev_t * 65316809eb4eSEric Schrock spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6532ea8dc4b6Seschrock { 6533c5904d13Seschrock vdev_t *vd; 6534c5904d13Seschrock int i; 6535c5904d13Seschrock 6536c5904d13Seschrock if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6537c5904d13Seschrock return (vd); 6538c5904d13Seschrock 65396809eb4eSEric Schrock if (aux) { 6540c5904d13Seschrock for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6541c5904d13Seschrock vd = spa->spa_l2cache.sav_vdevs[i]; 65426809eb4eSEric Schrock if (vd->vdev_guid == guid) 65436809eb4eSEric Schrock return (vd); 65446809eb4eSEric Schrock } 65456809eb4eSEric Schrock 65466809eb4eSEric Schrock for (i = 0; i < spa->spa_spares.sav_count; i++) { 65476809eb4eSEric Schrock vd = spa->spa_spares.sav_vdevs[i]; 6548c5904d13Seschrock if (vd->vdev_guid == guid) 6549c5904d13Seschrock return (vd); 6550c5904d13Seschrock } 6551c5904d13Seschrock } 6552c5904d13Seschrock 6553c5904d13Seschrock return (NULL); 6554ea8dc4b6Seschrock } 6555eaca9bbdSeschrock 6556eaca9bbdSeschrock void 6557990b4856Slling spa_upgrade(spa_t *spa, uint64_t version) 6558eaca9bbdSeschrock { 6559f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 6560f9af39baSGeorge Wilson 6561e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6562eaca9bbdSeschrock 6563eaca9bbdSeschrock /* 6564eaca9bbdSeschrock * This should only be called for a non-faulted pool, and since a 6565eaca9bbdSeschrock * future version would result in an unopenable pool, this shouldn't be 6566eaca9bbdSeschrock * possible. 6567eaca9bbdSeschrock */ 656862eae887SRichard Yao ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 65695d7b4d43SMatthew Ahrens ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6570eaca9bbdSeschrock 6571990b4856Slling spa->spa_uberblock.ub_version = version; 6572eaca9bbdSeschrock vdev_config_dirty(spa->spa_root_vdev); 6573eaca9bbdSeschrock 6574e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 657599653d4eSeschrock 657699653d4eSeschrock txg_wait_synced(spa_get_dsl(spa), 0); 657799653d4eSeschrock } 657899653d4eSeschrock 657999653d4eSeschrock boolean_t 658099653d4eSeschrock spa_has_spare(spa_t *spa, uint64_t guid) 658199653d4eSeschrock { 658299653d4eSeschrock int i; 658339c23413Seschrock uint64_t spareguid; 6584fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_spares; 658599653d4eSeschrock 6586fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 6587fa94a07fSbrendan if (sav->sav_vdevs[i]->vdev_guid == guid) 658899653d4eSeschrock return (B_TRUE); 658999653d4eSeschrock 6590fa94a07fSbrendan for (i = 0; i < sav->sav_npending; i++) { 6591fa94a07fSbrendan if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6592fa94a07fSbrendan &spareguid) == 0 && spareguid == guid) 659339c23413Seschrock return (B_TRUE); 659439c23413Seschrock } 659539c23413Seschrock 659699653d4eSeschrock return (B_FALSE); 6597eaca9bbdSeschrock } 6598b1b8ab34Slling 659989a89ebfSlling /* 660089a89ebfSlling * Check if a pool has an active shared spare device. 660189a89ebfSlling * Note: reference count of an active spare is 2, as a spare and as a replace 660289a89ebfSlling */ 660389a89ebfSlling static boolean_t 660489a89ebfSlling spa_has_active_shared_spare(spa_t *spa) 660589a89ebfSlling { 660689a89ebfSlling int i, refcnt; 660789a89ebfSlling uint64_t pool; 660889a89ebfSlling spa_aux_vdev_t *sav = &spa->spa_spares; 660989a89ebfSlling 661089a89ebfSlling for (i = 0; i < sav->sav_count; i++) { 661189a89ebfSlling if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 661289a89ebfSlling &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 661389a89ebfSlling refcnt > 2) 661489a89ebfSlling return (B_TRUE); 661589a89ebfSlling } 661689a89ebfSlling 661789a89ebfSlling return (B_FALSE); 661889a89ebfSlling } 661989a89ebfSlling 66203d7072f8Seschrock /* 66213d7072f8Seschrock * Post a sysevent corresponding to the given event. The 'name' must be one of 66223d7072f8Seschrock * the event definitions in sys/sysevent/eventdefs.h. The payload will be 66233d7072f8Seschrock * filled in from the spa and (optionally) the vdev. This doesn't do anything 66243d7072f8Seschrock * in the userland libzpool, as we don't want consumers to misinterpret ztest 66253d7072f8Seschrock * or zdb as real changes. 66263d7072f8Seschrock */ 66273d7072f8Seschrock void 66283d7072f8Seschrock spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 66293d7072f8Seschrock { 66303d7072f8Seschrock #ifdef _KERNEL 66313d7072f8Seschrock sysevent_t *ev; 66323d7072f8Seschrock sysevent_attr_list_t *attr = NULL; 66333d7072f8Seschrock sysevent_value_t value; 66343d7072f8Seschrock sysevent_id_t eid; 66353d7072f8Seschrock 66363d7072f8Seschrock ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 66373d7072f8Seschrock SE_SLEEP); 66383d7072f8Seschrock 66393d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 66403d7072f8Seschrock value.value.sv_string = spa_name(spa); 66413d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 66423d7072f8Seschrock goto done; 66433d7072f8Seschrock 66443d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 66453d7072f8Seschrock value.value.sv_uint64 = spa_guid(spa); 66463d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 66473d7072f8Seschrock goto done; 66483d7072f8Seschrock 66493d7072f8Seschrock if (vd) { 66503d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 66513d7072f8Seschrock value.value.sv_uint64 = vd->vdev_guid; 66523d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 66533d7072f8Seschrock SE_SLEEP) != 0) 66543d7072f8Seschrock goto done; 66553d7072f8Seschrock 66563d7072f8Seschrock if (vd->vdev_path) { 66573d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 66583d7072f8Seschrock value.value.sv_string = vd->vdev_path; 66593d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 66603d7072f8Seschrock &value, SE_SLEEP) != 0) 66613d7072f8Seschrock goto done; 66623d7072f8Seschrock } 66633d7072f8Seschrock } 66643d7072f8Seschrock 6665b01c3b58Seschrock if (sysevent_attach_attributes(ev, attr) != 0) 6666b01c3b58Seschrock goto done; 6667b01c3b58Seschrock attr = NULL; 6668b01c3b58Seschrock 66693d7072f8Seschrock (void) log_sysevent(ev, SE_SLEEP, &eid); 66703d7072f8Seschrock 66713d7072f8Seschrock done: 66723d7072f8Seschrock if (attr) 66733d7072f8Seschrock sysevent_free_attr(attr); 66743d7072f8Seschrock sysevent_free(ev); 66753d7072f8Seschrock #endif 66763d7072f8Seschrock } 6677