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. 2514372834SHans Rosenfeld * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 2745818ee1SMatthew Ahrens * Copyright 2013 Saso Kiselkov. All rights reserved. 28c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 295aeb9474SGarrett D'Amore */ 30fa9e4066Sahrens 31fa9e4066Sahrens /* 323e30c24aSWill Andrews * SPA: Storage Pool Allocator 333e30c24aSWill Andrews * 34fa9e4066Sahrens * This file contains all the routines used when modifying on-disk SPA state. 35fa9e4066Sahrens * This includes opening, importing, destroying, exporting a pool, and syncing a 36fa9e4066Sahrens * pool. 37fa9e4066Sahrens */ 38fa9e4066Sahrens 39fa9e4066Sahrens #include <sys/zfs_context.h> 40ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 41fa9e4066Sahrens #include <sys/spa_impl.h> 42fa9e4066Sahrens #include <sys/zio.h> 43fa9e4066Sahrens #include <sys/zio_checksum.h> 44fa9e4066Sahrens #include <sys/dmu.h> 45fa9e4066Sahrens #include <sys/dmu_tx.h> 46fa9e4066Sahrens #include <sys/zap.h> 47fa9e4066Sahrens #include <sys/zil.h> 48b24ab676SJeff Bonwick #include <sys/ddt.h> 49fa9e4066Sahrens #include <sys/vdev_impl.h> 50fa9e4066Sahrens #include <sys/metaslab.h> 5188ecc943SGeorge Wilson #include <sys/metaslab_impl.h> 52fa9e4066Sahrens #include <sys/uberblock_impl.h> 53fa9e4066Sahrens #include <sys/txg.h> 54fa9e4066Sahrens #include <sys/avl.h> 55fa9e4066Sahrens #include <sys/dmu_traverse.h> 56b1b8ab34Slling #include <sys/dmu_objset.h> 57fa9e4066Sahrens #include <sys/unique.h> 58fa9e4066Sahrens #include <sys/dsl_pool.h> 59b1b8ab34Slling #include <sys/dsl_dataset.h> 60fa9e4066Sahrens #include <sys/dsl_dir.h> 61fa9e4066Sahrens #include <sys/dsl_prop.h> 62b1b8ab34Slling #include <sys/dsl_synctask.h> 63fa9e4066Sahrens #include <sys/fs/zfs.h> 64fa94a07fSbrendan #include <sys/arc.h> 65fa9e4066Sahrens #include <sys/callb.h> 6695173954Sek #include <sys/systeminfo.h> 67e7cbe64fSgw #include <sys/spa_boot.h> 68573ca77eSGeorge Wilson #include <sys/zfs_ioctl.h> 693f9d6ad7SLin Ling #include <sys/dsl_scan.h> 70ad135b5dSChristopher Siden #include <sys/zfeature.h> 713b2aab18SMatthew Ahrens #include <sys/dsl_destroy.h> 72fa9e4066Sahrens 735679c89fSjv #ifdef _KERNEL 74dedec472SJack Meng #include <sys/bootprops.h> 7535a5a358SJonathan Adams #include <sys/callb.h> 7635a5a358SJonathan Adams #include <sys/cpupart.h> 7735a5a358SJonathan Adams #include <sys/pool.h> 7835a5a358SJonathan Adams #include <sys/sysdc.h> 7935a5a358SJonathan Adams #include <sys/zone.h> 805679c89fSjv #endif /* _KERNEL */ 815679c89fSjv 82990b4856Slling #include "zfs_prop.h" 83b7b97454Sperrin #include "zfs_comutil.h" 84990b4856Slling 853cb69f73SWill Andrews /* 863cb69f73SWill Andrews * The interval, in seconds, at which failed configuration cache file writes 873cb69f73SWill Andrews * should be retried. 883cb69f73SWill Andrews */ 893cb69f73SWill Andrews static int zfs_ccw_retry_interval = 300; 903cb69f73SWill Andrews 9135a5a358SJonathan Adams typedef enum zti_modes { 92ec94d322SAdam Leventhal ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 93ec94d322SAdam Leventhal ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 94ec94d322SAdam Leventhal ZTI_MODE_NULL, /* don't create a taskq */ 95ec94d322SAdam Leventhal ZTI_NMODES 9635a5a358SJonathan Adams } zti_modes_t; 97416e0cd8Sek 98ec94d322SAdam Leventhal #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 99ec94d322SAdam Leventhal #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 100ec94d322SAdam Leventhal #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 1012e0c549eSJonathan Adams 102ec94d322SAdam Leventhal #define ZTI_N(n) ZTI_P(n, 1) 103ec94d322SAdam Leventhal #define ZTI_ONE ZTI_N(1) 1042e0c549eSJonathan Adams 1052e0c549eSJonathan Adams typedef struct zio_taskq_info { 106ec94d322SAdam Leventhal zti_modes_t zti_mode; 10780eb36f2SGeorge Wilson uint_t zti_value; 108ec94d322SAdam Leventhal uint_t zti_count; 1092e0c549eSJonathan Adams } zio_taskq_info_t; 1102e0c549eSJonathan Adams 1112e0c549eSJonathan Adams static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 11235a5a358SJonathan Adams "issue", "issue_high", "intr", "intr_high" 1132e0c549eSJonathan Adams }; 1142e0c549eSJonathan Adams 11580eb36f2SGeorge Wilson /* 116ec94d322SAdam Leventhal * This table defines the taskq settings for each ZFS I/O type. When 117ec94d322SAdam Leventhal * initializing a pool, we use this table to create an appropriately sized 118ec94d322SAdam Leventhal * taskq. Some operations are low volume and therefore have a small, static 119ec94d322SAdam Leventhal * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 120ec94d322SAdam Leventhal * macros. Other operations process a large amount of data; the ZTI_BATCH 121ec94d322SAdam Leventhal * macro causes us to create a taskq oriented for throughput. Some operations 122ec94d322SAdam Leventhal * are so high frequency and short-lived that the taskq itself can become a a 123ec94d322SAdam Leventhal * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 124ec94d322SAdam Leventhal * additional degree of parallelism specified by the number of threads per- 125ec94d322SAdam Leventhal * taskq and the number of taskqs; when dispatching an event in this case, the 126ec94d322SAdam Leventhal * particular taskq is chosen at random. 127ec94d322SAdam Leventhal * 128ec94d322SAdam Leventhal * The different taskq priorities are to handle the different contexts (issue 129ec94d322SAdam Leventhal * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 130ec94d322SAdam Leventhal * need to be handled with minimum delay. 13180eb36f2SGeorge Wilson */ 13280eb36f2SGeorge Wilson const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 13380eb36f2SGeorge Wilson /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 134ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 1351b497ab8SAdam H. Leventhal { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 136ec94d322SAdam Leventhal { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 137ec94d322SAdam Leventhal { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 138ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 139ec94d322SAdam Leventhal { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 1402e0c549eSJonathan Adams }; 1412e0c549eSJonathan Adams 1423b2aab18SMatthew Ahrens static void spa_sync_version(void *arg, dmu_tx_t *tx); 1433b2aab18SMatthew Ahrens static void spa_sync_props(void *arg, dmu_tx_t *tx); 14489a89ebfSlling static boolean_t spa_has_active_shared_spare(spa_t *spa); 1451195e687SMark J Musante static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 1461195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 1471195e687SMark J Musante char **ereport); 148cb04b873SMark J Musante static void spa_vdev_resilver_done(spa_t *spa); 149990b4856Slling 15069962b56SMatthew Ahrens uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 15135a5a358SJonathan Adams id_t zio_taskq_psrset_bind = PS_NONE; 15235a5a358SJonathan Adams boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 15335a5a358SJonathan Adams uint_t zio_taskq_basedc = 80; /* base duty cycle */ 15435a5a358SJonathan Adams 15535a5a358SJonathan Adams boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 15601f55e48SGeorge Wilson extern int zfs_sync_pass_deferred_free; 15735a5a358SJonathan Adams 15835a5a358SJonathan Adams /* 15935a5a358SJonathan Adams * This (illegal) pool name is used when temporarily importing a spa_t in order 16035a5a358SJonathan Adams * to get the vdev stats associated with the imported devices. 16135a5a358SJonathan Adams */ 16235a5a358SJonathan Adams #define TRYIMPORT_NAME "$import" 16335a5a358SJonathan Adams 164990b4856Slling /* 165990b4856Slling * ========================================================================== 166990b4856Slling * SPA properties routines 167990b4856Slling * ========================================================================== 168990b4856Slling */ 169990b4856Slling 170990b4856Slling /* 171990b4856Slling * Add a (source=src, propname=propval) list to an nvlist. 172990b4856Slling */ 1739d82f4f6Slling static void 174990b4856Slling spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 175990b4856Slling uint64_t intval, zprop_source_t src) 176990b4856Slling { 177990b4856Slling const char *propname = zpool_prop_to_name(prop); 178990b4856Slling nvlist_t *propval; 179990b4856Slling 1809d82f4f6Slling VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1819d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 182990b4856Slling 1839d82f4f6Slling if (strval != NULL) 1849d82f4f6Slling VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 1859d82f4f6Slling else 1869d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 187990b4856Slling 1889d82f4f6Slling VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 189990b4856Slling nvlist_free(propval); 190990b4856Slling } 191990b4856Slling 192990b4856Slling /* 193990b4856Slling * Get property values from the spa configuration. 194990b4856Slling */ 1959d82f4f6Slling static void 196990b4856Slling spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 197990b4856Slling { 1984263d13fSGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 199ad135b5dSChristopher Siden dsl_pool_t *pool = spa->spa_dsl_pool; 2002e4c9986SGeorge Wilson uint64_t size, alloc, cap, version; 201990b4856Slling zprop_source_t src = ZPROP_SRC_NONE; 202c5904d13Seschrock spa_config_dirent_t *dp; 2032e4c9986SGeorge Wilson metaslab_class_t *mc = spa_normal_class(spa); 204990b4856Slling 205e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 206e14bb325SJeff Bonwick 2074263d13fSGeorge Wilson if (rvd != NULL) { 208485bbbf5SGeorge Wilson alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 209b24ab676SJeff Bonwick size = metaslab_class_get_space(spa_normal_class(spa)); 210379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 211379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 212485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 213485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 214485bbbf5SGeorge Wilson size - alloc, src); 2154263d13fSGeorge Wilson 2162e4c9986SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 2172e4c9986SGeorge Wilson metaslab_class_fragmentation(mc), src); 2182e4c9986SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 2192e4c9986SGeorge Wilson metaslab_class_expandable_space(mc), src); 220f9af39baSGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 221f9af39baSGeorge Wilson (spa_mode(spa) == FREAD), src); 222379c004dSEric Schrock 223485bbbf5SGeorge Wilson cap = (size == 0) ? 0 : (alloc * 100 / size); 224379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 225379c004dSEric Schrock 226b24ab676SJeff Bonwick spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 227b24ab676SJeff Bonwick ddt_get_pool_dedup_ratio(spa), src); 228b24ab676SJeff Bonwick 229379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 2304263d13fSGeorge Wilson rvd->vdev_state, src); 231379c004dSEric Schrock 232379c004dSEric Schrock version = spa_version(spa); 233379c004dSEric Schrock if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 234379c004dSEric Schrock src = ZPROP_SRC_DEFAULT; 235379c004dSEric Schrock else 236379c004dSEric Schrock src = ZPROP_SRC_LOCAL; 237379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 238379c004dSEric Schrock } 239990b4856Slling 240ad135b5dSChristopher Siden if (pool != NULL) { 241ad135b5dSChristopher Siden /* 242ad135b5dSChristopher Siden * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 243ad135b5dSChristopher Siden * when opening pools before this version freedir will be NULL. 244ad135b5dSChristopher Siden */ 2457fd05ac4SMatthew Ahrens if (pool->dp_free_dir != NULL) { 246ad135b5dSChristopher Siden spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 247c1379625SJustin T. Gibbs dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 248c1379625SJustin T. Gibbs src); 249ad135b5dSChristopher Siden } else { 250ad135b5dSChristopher Siden spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 251ad135b5dSChristopher Siden NULL, 0, src); 252ad135b5dSChristopher Siden } 2537fd05ac4SMatthew Ahrens 2547fd05ac4SMatthew Ahrens if (pool->dp_leak_dir != NULL) { 2557fd05ac4SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 256c1379625SJustin T. Gibbs dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 257c1379625SJustin T. Gibbs src); 2587fd05ac4SMatthew Ahrens } else { 2597fd05ac4SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 2607fd05ac4SMatthew Ahrens NULL, 0, src); 2617fd05ac4SMatthew Ahrens } 262ad135b5dSChristopher Siden } 263ad135b5dSChristopher Siden 2649d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 265990b4856Slling 2668704186eSDan McDonald if (spa->spa_comment != NULL) { 2678704186eSDan McDonald spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 2688704186eSDan McDonald 0, ZPROP_SRC_LOCAL); 2698704186eSDan McDonald } 2708704186eSDan McDonald 2719d82f4f6Slling if (spa->spa_root != NULL) 2729d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 2739d82f4f6Slling 0, ZPROP_SRC_LOCAL); 274990b4856Slling 275b5152584SMatthew Ahrens if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 276b5152584SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 277b5152584SMatthew Ahrens MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 278b5152584SMatthew Ahrens } else { 279b5152584SMatthew Ahrens spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 280b5152584SMatthew Ahrens SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 281b5152584SMatthew Ahrens } 282b5152584SMatthew Ahrens 283c5904d13Seschrock if ((dp = list_head(&spa->spa_config_list)) != NULL) { 284c5904d13Seschrock if (dp->scd_path == NULL) { 2859d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 286c5904d13Seschrock "none", 0, ZPROP_SRC_LOCAL); 287c5904d13Seschrock } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 2889d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 289c5904d13Seschrock dp->scd_path, 0, ZPROP_SRC_LOCAL); 2902f8aaab3Seschrock } 2912f8aaab3Seschrock } 292990b4856Slling } 293990b4856Slling 294990b4856Slling /* 295990b4856Slling * Get zpool property values. 296990b4856Slling */ 297990b4856Slling int 298990b4856Slling spa_prop_get(spa_t *spa, nvlist_t **nvp) 299990b4856Slling { 300b24ab676SJeff Bonwick objset_t *mos = spa->spa_meta_objset; 301990b4856Slling zap_cursor_t zc; 302990b4856Slling zap_attribute_t za; 303990b4856Slling int err; 304990b4856Slling 3059d82f4f6Slling VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 306990b4856Slling 307e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 308e14bb325SJeff Bonwick 309990b4856Slling /* 310990b4856Slling * Get properties from the spa config. 311990b4856Slling */ 3129d82f4f6Slling spa_prop_get_config(spa, nvp); 313990b4856Slling 314990b4856Slling /* If no pool property object, no more prop to get. */ 315afee20e4SGeorge Wilson if (mos == NULL || spa->spa_pool_props_object == 0) { 316990b4856Slling mutex_exit(&spa->spa_props_lock); 317990b4856Slling return (0); 318990b4856Slling } 319990b4856Slling 320990b4856Slling /* 321990b4856Slling * Get properties from the MOS pool property object. 322990b4856Slling */ 323990b4856Slling for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 324990b4856Slling (err = zap_cursor_retrieve(&zc, &za)) == 0; 325990b4856Slling zap_cursor_advance(&zc)) { 326990b4856Slling uint64_t intval = 0; 327990b4856Slling char *strval = NULL; 328990b4856Slling zprop_source_t src = ZPROP_SRC_DEFAULT; 329990b4856Slling zpool_prop_t prop; 330990b4856Slling 331990b4856Slling if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 332990b4856Slling continue; 333990b4856Slling 334990b4856Slling switch (za.za_integer_length) { 335990b4856Slling case 8: 336990b4856Slling /* integer property */ 337990b4856Slling if (za.za_first_integer != 338990b4856Slling zpool_prop_default_numeric(prop)) 339990b4856Slling src = ZPROP_SRC_LOCAL; 340990b4856Slling 341990b4856Slling if (prop == ZPOOL_PROP_BOOTFS) { 342990b4856Slling dsl_pool_t *dp; 343990b4856Slling dsl_dataset_t *ds = NULL; 344990b4856Slling 345990b4856Slling dp = spa_get_dsl(spa); 3463b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 347745cd3c5Smaybee if (err = dsl_dataset_hold_obj(dp, 348745cd3c5Smaybee za.za_first_integer, FTAG, &ds)) { 3493b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 350990b4856Slling break; 351990b4856Slling } 352990b4856Slling 353*9adfa60dSMatthew Ahrens strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 354990b4856Slling KM_SLEEP); 355990b4856Slling dsl_dataset_name(ds, strval); 356745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 3573b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 358990b4856Slling } else { 359990b4856Slling strval = NULL; 360990b4856Slling intval = za.za_first_integer; 361990b4856Slling } 362990b4856Slling 3639d82f4f6Slling spa_prop_add_list(*nvp, prop, strval, intval, src); 364990b4856Slling 365990b4856Slling if (strval != NULL) 366*9adfa60dSMatthew Ahrens kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 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 } 5938704186eSDan McDonald if (strlen(strval) > ZPROP_MAX_COMMENT) 5948704186eSDan McDonald error = E2BIG; 5958704186eSDan McDonald break; 5968704186eSDan McDonald 597b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 598b24ab676SJeff Bonwick if (spa_version(spa) < SPA_VERSION_DEDUP) 599be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 600b24ab676SJeff Bonwick else 601b24ab676SJeff Bonwick error = nvpair_value_uint64(elem, &intval); 602b24ab676SJeff Bonwick if (error == 0 && 603b24ab676SJeff Bonwick intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 604be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 605b24ab676SJeff Bonwick break; 606990b4856Slling } 607990b4856Slling 608990b4856Slling if (error) 609990b4856Slling break; 610990b4856Slling } 611990b4856Slling 612990b4856Slling if (!error && reset_bootfs) { 613990b4856Slling error = nvlist_remove(props, 614990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 615990b4856Slling 616990b4856Slling if (!error) { 617990b4856Slling error = nvlist_add_uint64(props, 618990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 619990b4856Slling } 620990b4856Slling } 621990b4856Slling 622990b4856Slling return (error); 623990b4856Slling } 624990b4856Slling 625379c004dSEric Schrock void 626379c004dSEric Schrock spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 627379c004dSEric Schrock { 628379c004dSEric Schrock char *cachefile; 629379c004dSEric Schrock spa_config_dirent_t *dp; 630379c004dSEric Schrock 631379c004dSEric Schrock if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 632379c004dSEric Schrock &cachefile) != 0) 633379c004dSEric Schrock return; 634379c004dSEric Schrock 635379c004dSEric Schrock dp = kmem_alloc(sizeof (spa_config_dirent_t), 636379c004dSEric Schrock KM_SLEEP); 637379c004dSEric Schrock 638379c004dSEric Schrock if (cachefile[0] == '\0') 639379c004dSEric Schrock dp->scd_path = spa_strdup(spa_config_path); 640379c004dSEric Schrock else if (strcmp(cachefile, "none") == 0) 641379c004dSEric Schrock dp->scd_path = NULL; 642379c004dSEric Schrock else 643379c004dSEric Schrock dp->scd_path = spa_strdup(cachefile); 644379c004dSEric Schrock 645379c004dSEric Schrock list_insert_head(&spa->spa_config_list, dp); 646379c004dSEric Schrock if (need_sync) 647379c004dSEric Schrock spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 648379c004dSEric Schrock } 649379c004dSEric Schrock 650990b4856Slling int 651990b4856Slling spa_prop_set(spa_t *spa, nvlist_t *nvp) 652990b4856Slling { 653990b4856Slling int error; 654ad135b5dSChristopher Siden nvpair_t *elem = NULL; 655379c004dSEric Schrock boolean_t need_sync = B_FALSE; 656990b4856Slling 657990b4856Slling if ((error = spa_prop_validate(spa, nvp)) != 0) 658990b4856Slling return (error); 659990b4856Slling 660379c004dSEric Schrock while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 661ad135b5dSChristopher Siden zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 662379c004dSEric Schrock 663f9af39baSGeorge Wilson if (prop == ZPOOL_PROP_CACHEFILE || 664f9af39baSGeorge Wilson prop == ZPOOL_PROP_ALTROOT || 665f9af39baSGeorge Wilson prop == ZPOOL_PROP_READONLY) 666379c004dSEric Schrock continue; 667379c004dSEric Schrock 668ad135b5dSChristopher Siden if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 669ad135b5dSChristopher Siden uint64_t ver; 670ad135b5dSChristopher Siden 671ad135b5dSChristopher Siden if (prop == ZPOOL_PROP_VERSION) { 672ad135b5dSChristopher Siden VERIFY(nvpair_value_uint64(elem, &ver) == 0); 673ad135b5dSChristopher Siden } else { 674ad135b5dSChristopher Siden ASSERT(zpool_prop_feature(nvpair_name(elem))); 675ad135b5dSChristopher Siden ver = SPA_VERSION_FEATURES; 676ad135b5dSChristopher Siden need_sync = B_TRUE; 677ad135b5dSChristopher Siden } 678ad135b5dSChristopher Siden 679ad135b5dSChristopher Siden /* Save time if the version is already set. */ 680ad135b5dSChristopher Siden if (ver == spa_version(spa)) 681ad135b5dSChristopher Siden continue; 682ad135b5dSChristopher Siden 683ad135b5dSChristopher Siden /* 684ad135b5dSChristopher Siden * In addition to the pool directory object, we might 685ad135b5dSChristopher Siden * create the pool properties object, the features for 686ad135b5dSChristopher Siden * read object, the features for write object, or the 687ad135b5dSChristopher Siden * feature descriptions object. 688ad135b5dSChristopher Siden */ 6893b2aab18SMatthew Ahrens error = dsl_sync_task(spa->spa_name, NULL, 6907d46dc6cSMatthew Ahrens spa_sync_version, &ver, 6917d46dc6cSMatthew Ahrens 6, ZFS_SPACE_CHECK_RESERVED); 692ad135b5dSChristopher Siden if (error) 693ad135b5dSChristopher Siden return (error); 694ad135b5dSChristopher Siden continue; 695ad135b5dSChristopher Siden } 696ad135b5dSChristopher Siden 697379c004dSEric Schrock need_sync = B_TRUE; 698379c004dSEric Schrock break; 699379c004dSEric Schrock } 700379c004dSEric Schrock 701ad135b5dSChristopher Siden if (need_sync) { 7023b2aab18SMatthew Ahrens return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 7037d46dc6cSMatthew Ahrens nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 704ad135b5dSChristopher Siden } 705ad135b5dSChristopher Siden 706ad135b5dSChristopher Siden return (0); 707990b4856Slling } 708990b4856Slling 709990b4856Slling /* 710990b4856Slling * If the bootfs property value is dsobj, clear it. 711990b4856Slling */ 712990b4856Slling void 713990b4856Slling spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 714990b4856Slling { 715990b4856Slling if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 716990b4856Slling VERIFY(zap_remove(spa->spa_meta_objset, 717990b4856Slling spa->spa_pool_props_object, 718990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 719990b4856Slling spa->spa_bootfs = 0; 720990b4856Slling } 721990b4856Slling } 722990b4856Slling 723dfbb9432SGeorge Wilson /*ARGSUSED*/ 724dfbb9432SGeorge Wilson static int 7253b2aab18SMatthew Ahrens spa_change_guid_check(void *arg, dmu_tx_t *tx) 726dfbb9432SGeorge Wilson { 7273b2aab18SMatthew Ahrens uint64_t *newguid = arg; 7283b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 729dfbb9432SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 730dfbb9432SGeorge Wilson uint64_t vdev_state; 731dfbb9432SGeorge Wilson 732dfbb9432SGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 733dfbb9432SGeorge Wilson vdev_state = rvd->vdev_state; 734dfbb9432SGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 735dfbb9432SGeorge Wilson 736dfbb9432SGeorge Wilson if (vdev_state != VDEV_STATE_HEALTHY) 737be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 738dfbb9432SGeorge Wilson 739dfbb9432SGeorge Wilson ASSERT3U(spa_guid(spa), !=, *newguid); 740dfbb9432SGeorge Wilson 741dfbb9432SGeorge Wilson return (0); 742dfbb9432SGeorge Wilson } 743dfbb9432SGeorge Wilson 744dfbb9432SGeorge Wilson static void 7453b2aab18SMatthew Ahrens spa_change_guid_sync(void *arg, dmu_tx_t *tx) 746dfbb9432SGeorge Wilson { 7473b2aab18SMatthew Ahrens uint64_t *newguid = arg; 7483b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 749dfbb9432SGeorge Wilson uint64_t oldguid; 750dfbb9432SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 751dfbb9432SGeorge Wilson 752dfbb9432SGeorge Wilson oldguid = spa_guid(spa); 753dfbb9432SGeorge Wilson 754dfbb9432SGeorge Wilson spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 755dfbb9432SGeorge Wilson rvd->vdev_guid = *newguid; 756dfbb9432SGeorge Wilson rvd->vdev_guid_sum += (*newguid - oldguid); 757dfbb9432SGeorge Wilson vdev_config_dirty(rvd); 758dfbb9432SGeorge Wilson spa_config_exit(spa, SCL_STATE, FTAG); 759dfbb9432SGeorge Wilson 76020128a08SGeorge Wilson spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 761dfbb9432SGeorge Wilson oldguid, *newguid); 762dfbb9432SGeorge Wilson } 763dfbb9432SGeorge Wilson 764e9103aaeSGarrett D'Amore /* 765e9103aaeSGarrett D'Amore * Change the GUID for the pool. This is done so that we can later 766e9103aaeSGarrett D'Amore * re-import a pool built from a clone of our own vdevs. We will modify 767e9103aaeSGarrett D'Amore * the root vdev's guid, our own pool guid, and then mark all of our 768e9103aaeSGarrett D'Amore * vdevs dirty. Note that we must make sure that all our vdevs are 769e9103aaeSGarrett D'Amore * online when we do this, or else any vdevs that weren't present 770e9103aaeSGarrett D'Amore * would be orphaned from our pool. We are also going to issue a 771e9103aaeSGarrett D'Amore * sysevent to update any watchers. 772e9103aaeSGarrett D'Amore */ 773e9103aaeSGarrett D'Amore int 774e9103aaeSGarrett D'Amore spa_change_guid(spa_t *spa) 775e9103aaeSGarrett D'Amore { 776dfbb9432SGeorge Wilson int error; 777dfbb9432SGeorge Wilson uint64_t guid; 778e9103aaeSGarrett D'Amore 7792c1e2b44SGeorge Wilson mutex_enter(&spa->spa_vdev_top_lock); 780dfbb9432SGeorge Wilson mutex_enter(&spa_namespace_lock); 781dfbb9432SGeorge Wilson guid = spa_generate_guid(NULL); 782e9103aaeSGarrett D'Amore 7833b2aab18SMatthew Ahrens error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 7847d46dc6cSMatthew Ahrens spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 785e9103aaeSGarrett D'Amore 786dfbb9432SGeorge Wilson if (error == 0) { 787dfbb9432SGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 788dfbb9432SGeorge Wilson spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 789dfbb9432SGeorge Wilson } 790e9103aaeSGarrett D'Amore 791dfbb9432SGeorge Wilson mutex_exit(&spa_namespace_lock); 7922c1e2b44SGeorge Wilson mutex_exit(&spa->spa_vdev_top_lock); 793e9103aaeSGarrett D'Amore 794dfbb9432SGeorge Wilson return (error); 795e9103aaeSGarrett D'Amore } 796e9103aaeSGarrett D'Amore 797fa9e4066Sahrens /* 798fa9e4066Sahrens * ========================================================================== 799fa9e4066Sahrens * SPA state manipulation (open/create/destroy/import/export) 800fa9e4066Sahrens * ========================================================================== 801fa9e4066Sahrens */ 802fa9e4066Sahrens 803ea8dc4b6Seschrock static int 804ea8dc4b6Seschrock spa_error_entry_compare(const void *a, const void *b) 805ea8dc4b6Seschrock { 806ea8dc4b6Seschrock spa_error_entry_t *sa = (spa_error_entry_t *)a; 807ea8dc4b6Seschrock spa_error_entry_t *sb = (spa_error_entry_t *)b; 808ea8dc4b6Seschrock int ret; 809ea8dc4b6Seschrock 810ea8dc4b6Seschrock ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 8117802d7bfSMatthew Ahrens sizeof (zbookmark_phys_t)); 812ea8dc4b6Seschrock 813ea8dc4b6Seschrock if (ret < 0) 814ea8dc4b6Seschrock return (-1); 815ea8dc4b6Seschrock else if (ret > 0) 816ea8dc4b6Seschrock return (1); 817ea8dc4b6Seschrock else 818ea8dc4b6Seschrock return (0); 819ea8dc4b6Seschrock } 820ea8dc4b6Seschrock 821ea8dc4b6Seschrock /* 822ea8dc4b6Seschrock * Utility function which retrieves copies of the current logs and 823ea8dc4b6Seschrock * re-initializes them in the process. 824ea8dc4b6Seschrock */ 825ea8dc4b6Seschrock void 826ea8dc4b6Seschrock spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 827ea8dc4b6Seschrock { 828ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 829ea8dc4b6Seschrock 830ea8dc4b6Seschrock bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 831ea8dc4b6Seschrock bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 832ea8dc4b6Seschrock 833ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 834ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 835ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 836ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 837ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 838ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 839ea8dc4b6Seschrock } 840ea8dc4b6Seschrock 841ec94d322SAdam Leventhal static void 842ec94d322SAdam Leventhal spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 843fa9e4066Sahrens { 844ec94d322SAdam Leventhal const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 845ec94d322SAdam Leventhal enum zti_modes mode = ztip->zti_mode; 846ec94d322SAdam Leventhal uint_t value = ztip->zti_value; 847ec94d322SAdam Leventhal uint_t count = ztip->zti_count; 848ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 849ec94d322SAdam Leventhal char name[32]; 8505aeb9474SGarrett D'Amore uint_t flags = 0; 85135a5a358SJonathan Adams boolean_t batch = B_FALSE; 852fa9e4066Sahrens 853ec94d322SAdam Leventhal if (mode == ZTI_MODE_NULL) { 854ec94d322SAdam Leventhal tqs->stqs_count = 0; 855ec94d322SAdam Leventhal tqs->stqs_taskq = NULL; 856ec94d322SAdam Leventhal return; 857ec94d322SAdam Leventhal } 858fa9e4066Sahrens 859ec94d322SAdam Leventhal ASSERT3U(count, >, 0); 860fa9e4066Sahrens 861ec94d322SAdam Leventhal tqs->stqs_count = count; 862ec94d322SAdam Leventhal tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 86335a5a358SJonathan Adams 86469962b56SMatthew Ahrens switch (mode) { 86569962b56SMatthew Ahrens case ZTI_MODE_FIXED: 86669962b56SMatthew Ahrens ASSERT3U(value, >=, 1); 86769962b56SMatthew Ahrens value = MAX(value, 1); 86869962b56SMatthew Ahrens break; 869ec94d322SAdam Leventhal 87069962b56SMatthew Ahrens case ZTI_MODE_BATCH: 87169962b56SMatthew Ahrens batch = B_TRUE; 87269962b56SMatthew Ahrens flags |= TASKQ_THREADS_CPU_PCT; 87369962b56SMatthew Ahrens value = zio_taskq_batch_pct; 87469962b56SMatthew Ahrens break; 875ec94d322SAdam Leventhal 87669962b56SMatthew Ahrens default: 87769962b56SMatthew Ahrens panic("unrecognized mode for %s_%s taskq (%u:%u) in " 87869962b56SMatthew Ahrens "spa_activate()", 87969962b56SMatthew Ahrens zio_type_name[t], zio_taskq_types[q], mode, value); 88069962b56SMatthew Ahrens break; 88169962b56SMatthew Ahrens } 882ec94d322SAdam Leventhal 88369962b56SMatthew Ahrens for (uint_t i = 0; i < count; i++) { 88469962b56SMatthew Ahrens taskq_t *tq; 885ec94d322SAdam Leventhal 886ec94d322SAdam Leventhal if (count > 1) { 887ec94d322SAdam Leventhal (void) snprintf(name, sizeof (name), "%s_%s_%u", 888ec94d322SAdam Leventhal zio_type_name[t], zio_taskq_types[q], i); 889ec94d322SAdam Leventhal } else { 890ec94d322SAdam Leventhal (void) snprintf(name, sizeof (name), "%s_%s", 891ec94d322SAdam Leventhal zio_type_name[t], zio_taskq_types[q]); 892ec94d322SAdam Leventhal } 893ec94d322SAdam Leventhal 894ec94d322SAdam Leventhal if (zio_taskq_sysdc && spa->spa_proc != &p0) { 895ec94d322SAdam Leventhal if (batch) 896ec94d322SAdam Leventhal flags |= TASKQ_DC_BATCH; 897ec94d322SAdam Leventhal 898ec94d322SAdam Leventhal tq = taskq_create_sysdc(name, value, 50, INT_MAX, 899ec94d322SAdam Leventhal spa->spa_proc, zio_taskq_basedc, flags); 900ec94d322SAdam Leventhal } else { 90169962b56SMatthew Ahrens pri_t pri = maxclsyspri; 90269962b56SMatthew Ahrens /* 90369962b56SMatthew Ahrens * The write issue taskq can be extremely CPU 90469962b56SMatthew Ahrens * intensive. Run it at slightly lower priority 90569962b56SMatthew Ahrens * than the other taskqs. 90669962b56SMatthew Ahrens */ 90769962b56SMatthew Ahrens if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 90869962b56SMatthew Ahrens pri--; 90969962b56SMatthew Ahrens 91069962b56SMatthew Ahrens tq = taskq_create_proc(name, value, pri, 50, 911ec94d322SAdam Leventhal INT_MAX, spa->spa_proc, flags); 912ec94d322SAdam Leventhal } 913ec94d322SAdam Leventhal 914ec94d322SAdam Leventhal tqs->stqs_taskq[i] = tq; 915ec94d322SAdam Leventhal } 916ec94d322SAdam Leventhal } 917ec94d322SAdam Leventhal 918ec94d322SAdam Leventhal static void 919ec94d322SAdam Leventhal spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 920ec94d322SAdam Leventhal { 921ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 922ec94d322SAdam Leventhal 923ec94d322SAdam Leventhal if (tqs->stqs_taskq == NULL) { 924ec94d322SAdam Leventhal ASSERT0(tqs->stqs_count); 925ec94d322SAdam Leventhal return; 926ec94d322SAdam Leventhal } 927ec94d322SAdam Leventhal 928ec94d322SAdam Leventhal for (uint_t i = 0; i < tqs->stqs_count; i++) { 929ec94d322SAdam Leventhal ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 930ec94d322SAdam Leventhal taskq_destroy(tqs->stqs_taskq[i]); 93135a5a358SJonathan Adams } 93235a5a358SJonathan Adams 933ec94d322SAdam Leventhal kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 934ec94d322SAdam Leventhal tqs->stqs_taskq = NULL; 935ec94d322SAdam Leventhal } 93635a5a358SJonathan Adams 937ec94d322SAdam Leventhal /* 938ec94d322SAdam Leventhal * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 939ec94d322SAdam Leventhal * Note that a type may have multiple discrete taskqs to avoid lock contention 940ec94d322SAdam Leventhal * on the taskq itself. In that case we choose which taskq at random by using 941ec94d322SAdam Leventhal * the low bits of gethrtime(). 942ec94d322SAdam Leventhal */ 943ec94d322SAdam Leventhal void 944ec94d322SAdam Leventhal spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 945ec94d322SAdam Leventhal task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 946ec94d322SAdam Leventhal { 947ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 948ec94d322SAdam Leventhal taskq_t *tq; 949ec94d322SAdam Leventhal 950ec94d322SAdam Leventhal ASSERT3P(tqs->stqs_taskq, !=, NULL); 951ec94d322SAdam Leventhal ASSERT3U(tqs->stqs_count, !=, 0); 952ec94d322SAdam Leventhal 953ec94d322SAdam Leventhal if (tqs->stqs_count == 1) { 954ec94d322SAdam Leventhal tq = tqs->stqs_taskq[0]; 955ec94d322SAdam Leventhal } else { 956ec94d322SAdam Leventhal tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 95735a5a358SJonathan Adams } 958ec94d322SAdam Leventhal 959ec94d322SAdam Leventhal taskq_dispatch_ent(tq, func, arg, flags, ent); 96035a5a358SJonathan Adams } 96135a5a358SJonathan Adams 96235a5a358SJonathan Adams static void 96335a5a358SJonathan Adams spa_create_zio_taskqs(spa_t *spa) 96435a5a358SJonathan Adams { 965e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 966e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 967ec94d322SAdam Leventhal spa_taskqs_init(spa, t, q); 96835a5a358SJonathan Adams } 96935a5a358SJonathan Adams } 97035a5a358SJonathan Adams } 97135a5a358SJonathan Adams 97235a5a358SJonathan Adams #ifdef _KERNEL 97335a5a358SJonathan Adams static void 97435a5a358SJonathan Adams spa_thread(void *arg) 97535a5a358SJonathan Adams { 97635a5a358SJonathan Adams callb_cpr_t cprinfo; 9772e0c549eSJonathan Adams 97835a5a358SJonathan Adams spa_t *spa = arg; 97935a5a358SJonathan Adams user_t *pu = PTOU(curproc); 9802e0c549eSJonathan Adams 98135a5a358SJonathan Adams CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 98235a5a358SJonathan Adams spa->spa_name); 9832e0c549eSJonathan Adams 98435a5a358SJonathan Adams ASSERT(curproc != &p0); 98535a5a358SJonathan Adams (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 98635a5a358SJonathan Adams "zpool-%s", spa->spa_name); 98735a5a358SJonathan Adams (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 9882e0c549eSJonathan Adams 98935a5a358SJonathan Adams /* bind this thread to the requested psrset */ 99035a5a358SJonathan Adams if (zio_taskq_psrset_bind != PS_NONE) { 99135a5a358SJonathan Adams pool_lock(); 99235a5a358SJonathan Adams mutex_enter(&cpu_lock); 99335a5a358SJonathan Adams mutex_enter(&pidlock); 99435a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 99580eb36f2SGeorge Wilson 99635a5a358SJonathan Adams if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 99735a5a358SJonathan Adams 0, NULL, NULL) == 0) { 99835a5a358SJonathan Adams curthread->t_bind_pset = zio_taskq_psrset_bind; 99935a5a358SJonathan Adams } else { 100035a5a358SJonathan Adams cmn_err(CE_WARN, 100135a5a358SJonathan Adams "Couldn't bind process for zfs pool \"%s\" to " 100235a5a358SJonathan Adams "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 100335a5a358SJonathan Adams } 100435a5a358SJonathan Adams 100535a5a358SJonathan Adams mutex_exit(&curproc->p_lock); 100635a5a358SJonathan Adams mutex_exit(&pidlock); 100735a5a358SJonathan Adams mutex_exit(&cpu_lock); 100835a5a358SJonathan Adams pool_unlock(); 100935a5a358SJonathan Adams } 101035a5a358SJonathan Adams 101135a5a358SJonathan Adams if (zio_taskq_sysdc) { 101235a5a358SJonathan Adams sysdc_thread_enter(curthread, 100, 0); 101335a5a358SJonathan Adams } 101435a5a358SJonathan Adams 101535a5a358SJonathan Adams spa->spa_proc = curproc; 101635a5a358SJonathan Adams spa->spa_did = curthread->t_did; 101735a5a358SJonathan Adams 101835a5a358SJonathan Adams spa_create_zio_taskqs(spa); 101935a5a358SJonathan Adams 102035a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 102135a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 102235a5a358SJonathan Adams 102335a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_ACTIVE; 102435a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 102535a5a358SJonathan Adams 102635a5a358SJonathan Adams CALLB_CPR_SAFE_BEGIN(&cprinfo); 102735a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_ACTIVE) 102835a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 102935a5a358SJonathan Adams CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 103035a5a358SJonathan Adams 103135a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 103235a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_GONE; 103335a5a358SJonathan Adams spa->spa_proc = &p0; 103435a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 103535a5a358SJonathan Adams CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 103635a5a358SJonathan Adams 103735a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 103835a5a358SJonathan Adams lwp_exit(); 103935a5a358SJonathan Adams } 104035a5a358SJonathan Adams #endif 104135a5a358SJonathan Adams 104235a5a358SJonathan Adams /* 104335a5a358SJonathan Adams * Activate an uninitialized pool. 104435a5a358SJonathan Adams */ 104535a5a358SJonathan Adams static void 104635a5a358SJonathan Adams spa_activate(spa_t *spa, int mode) 104735a5a358SJonathan Adams { 104835a5a358SJonathan Adams ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 104935a5a358SJonathan Adams 105035a5a358SJonathan Adams spa->spa_state = POOL_STATE_ACTIVE; 105135a5a358SJonathan Adams spa->spa_mode = mode; 105235a5a358SJonathan Adams 105335a5a358SJonathan Adams spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 105435a5a358SJonathan Adams spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 105535a5a358SJonathan Adams 105635a5a358SJonathan Adams /* Try to create a covering process */ 105735a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 105835a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 105935a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 106035a5a358SJonathan Adams spa->spa_did = 0; 106135a5a358SJonathan Adams 106235a5a358SJonathan Adams /* Only create a process if we're going to be around a while. */ 106335a5a358SJonathan Adams if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 106435a5a358SJonathan Adams if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 106535a5a358SJonathan Adams NULL, 0) == 0) { 106635a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_CREATED; 106735a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_CREATED) { 106835a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, 106935a5a358SJonathan Adams &spa->spa_proc_lock); 10702e0c549eSJonathan Adams } 107135a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 107235a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 107335a5a358SJonathan Adams ASSERT(spa->spa_did != 0); 107435a5a358SJonathan Adams } else { 107535a5a358SJonathan Adams #ifdef _KERNEL 107635a5a358SJonathan Adams cmn_err(CE_WARN, 107735a5a358SJonathan Adams "Couldn't create process for zfs pool \"%s\"\n", 107835a5a358SJonathan Adams spa->spa_name); 107935a5a358SJonathan Adams #endif 1080e14bb325SJeff Bonwick } 1081fa9e4066Sahrens } 108235a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 108335a5a358SJonathan Adams 108435a5a358SJonathan Adams /* If we didn't create a process, we need to create our taskqs. */ 108535a5a358SJonathan Adams if (spa->spa_proc == &p0) { 108635a5a358SJonathan Adams spa_create_zio_taskqs(spa); 108735a5a358SJonathan Adams } 1088fa9e4066Sahrens 1089e14bb325SJeff Bonwick list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1090e14bb325SJeff Bonwick offsetof(vdev_t, vdev_config_dirty_node)); 1091bc9014e6SJustin Gibbs list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1092bc9014e6SJustin Gibbs offsetof(objset_t, os_evicting_node)); 1093e14bb325SJeff Bonwick list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1094e14bb325SJeff Bonwick offsetof(vdev_t, vdev_state_dirty_node)); 1095fa9e4066Sahrens 1096fa9e4066Sahrens txg_list_create(&spa->spa_vdev_txg_list, 1097fa9e4066Sahrens offsetof(struct vdev, vdev_txg_node)); 1098ea8dc4b6Seschrock 1099ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 1100ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 1101ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 1102ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 1103ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 1104ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 1105fa9e4066Sahrens } 1106fa9e4066Sahrens 1107fa9e4066Sahrens /* 1108fa9e4066Sahrens * Opposite of spa_activate(). 1109fa9e4066Sahrens */ 1110fa9e4066Sahrens static void 1111fa9e4066Sahrens spa_deactivate(spa_t *spa) 1112fa9e4066Sahrens { 1113fa9e4066Sahrens ASSERT(spa->spa_sync_on == B_FALSE); 1114fa9e4066Sahrens ASSERT(spa->spa_dsl_pool == NULL); 1115fa9e4066Sahrens ASSERT(spa->spa_root_vdev == NULL); 111625f89ee2SJeff Bonwick ASSERT(spa->spa_async_zio_root == NULL); 1117fa9e4066Sahrens ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1118fa9e4066Sahrens 1119bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 1120bc9014e6SJustin Gibbs 1121fa9e4066Sahrens txg_list_destroy(&spa->spa_vdev_txg_list); 1122fa9e4066Sahrens 1123e14bb325SJeff Bonwick list_destroy(&spa->spa_config_dirty_list); 1124bc9014e6SJustin Gibbs list_destroy(&spa->spa_evicting_os_list); 1125e14bb325SJeff Bonwick list_destroy(&spa->spa_state_dirty_list); 1126fa9e4066Sahrens 1127e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 1128e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1129ec94d322SAdam Leventhal spa_taskqs_fini(spa, t, q); 1130e14bb325SJeff Bonwick } 1131fa9e4066Sahrens } 1132fa9e4066Sahrens 1133fa9e4066Sahrens metaslab_class_destroy(spa->spa_normal_class); 1134fa9e4066Sahrens spa->spa_normal_class = NULL; 1135fa9e4066Sahrens 11368654d025Sperrin metaslab_class_destroy(spa->spa_log_class); 11378654d025Sperrin spa->spa_log_class = NULL; 11388654d025Sperrin 1139ea8dc4b6Seschrock /* 1140ea8dc4b6Seschrock * If this was part of an import or the open otherwise failed, we may 1141ea8dc4b6Seschrock * still have errors left in the queues. Empty them just in case. 1142ea8dc4b6Seschrock */ 1143ea8dc4b6Seschrock spa_errlog_drain(spa); 1144ea8dc4b6Seschrock 1145ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_scrub); 1146ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_last); 1147ea8dc4b6Seschrock 1148fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 114935a5a358SJonathan Adams 115035a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 115135a5a358SJonathan Adams if (spa->spa_proc_state != SPA_PROC_NONE) { 115235a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 115335a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_DEACTIVATE; 115435a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 115535a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 115635a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 115735a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 115835a5a358SJonathan Adams } 115935a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 116035a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 116135a5a358SJonathan Adams } 116235a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 116335a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 116435a5a358SJonathan Adams 116535a5a358SJonathan Adams /* 116635a5a358SJonathan Adams * We want to make sure spa_thread() has actually exited the ZFS 116735a5a358SJonathan Adams * module, so that the module can't be unloaded out from underneath 116835a5a358SJonathan Adams * it. 116935a5a358SJonathan Adams */ 117035a5a358SJonathan Adams if (spa->spa_did != 0) { 117135a5a358SJonathan Adams thread_join(spa->spa_did); 117235a5a358SJonathan Adams spa->spa_did = 0; 117335a5a358SJonathan Adams } 1174fa9e4066Sahrens } 1175fa9e4066Sahrens 1176fa9e4066Sahrens /* 1177fa9e4066Sahrens * Verify a pool configuration, and construct the vdev tree appropriately. This 1178fa9e4066Sahrens * will create all the necessary vdevs in the appropriate layout, with each vdev 1179fa9e4066Sahrens * in the CLOSED state. This will prep the pool before open/creation/import. 1180fa9e4066Sahrens * All vdev validation is done by the vdev_alloc() routine. 1181fa9e4066Sahrens */ 118299653d4eSeschrock static int 118399653d4eSeschrock spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 118499653d4eSeschrock uint_t id, int atype) 1185fa9e4066Sahrens { 1186fa9e4066Sahrens nvlist_t **child; 1187573ca77eSGeorge Wilson uint_t children; 118899653d4eSeschrock int error; 1189fa9e4066Sahrens 119099653d4eSeschrock if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 119199653d4eSeschrock return (error); 1192fa9e4066Sahrens 119399653d4eSeschrock if ((*vdp)->vdev_ops->vdev_op_leaf) 119499653d4eSeschrock return (0); 1195fa9e4066Sahrens 1196e14bb325SJeff Bonwick error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1197e14bb325SJeff Bonwick &child, &children); 1198e14bb325SJeff Bonwick 1199e14bb325SJeff Bonwick if (error == ENOENT) 1200e14bb325SJeff Bonwick return (0); 1201e14bb325SJeff Bonwick 1202e14bb325SJeff Bonwick if (error) { 120399653d4eSeschrock vdev_free(*vdp); 120499653d4eSeschrock *vdp = NULL; 1205be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1206fa9e4066Sahrens } 1207fa9e4066Sahrens 1208573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 120999653d4eSeschrock vdev_t *vd; 121099653d4eSeschrock if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 121199653d4eSeschrock atype)) != 0) { 121299653d4eSeschrock vdev_free(*vdp); 121399653d4eSeschrock *vdp = NULL; 121499653d4eSeschrock return (error); 1215fa9e4066Sahrens } 1216fa9e4066Sahrens } 1217fa9e4066Sahrens 121899653d4eSeschrock ASSERT(*vdp != NULL); 121999653d4eSeschrock 122099653d4eSeschrock return (0); 1221fa9e4066Sahrens } 1222fa9e4066Sahrens 1223fa9e4066Sahrens /* 1224fa9e4066Sahrens * Opposite of spa_load(). 1225fa9e4066Sahrens */ 1226fa9e4066Sahrens static void 1227fa9e4066Sahrens spa_unload(spa_t *spa) 1228fa9e4066Sahrens { 122999653d4eSeschrock int i; 123099653d4eSeschrock 1231e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1232e14bb325SJeff Bonwick 1233ea8dc4b6Seschrock /* 1234ea8dc4b6Seschrock * Stop async tasks. 1235ea8dc4b6Seschrock */ 1236ea8dc4b6Seschrock spa_async_suspend(spa); 1237ea8dc4b6Seschrock 1238fa9e4066Sahrens /* 1239fa9e4066Sahrens * Stop syncing. 1240fa9e4066Sahrens */ 1241fa9e4066Sahrens if (spa->spa_sync_on) { 1242fa9e4066Sahrens txg_sync_stop(spa->spa_dsl_pool); 1243fa9e4066Sahrens spa->spa_sync_on = B_FALSE; 1244fa9e4066Sahrens } 1245fa9e4066Sahrens 1246fa9e4066Sahrens /* 1247e14bb325SJeff Bonwick * Wait for any outstanding async I/O to complete. 1248fa9e4066Sahrens */ 124954d692b7SGeorge Wilson if (spa->spa_async_zio_root != NULL) { 12506f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) 12516f834bc1SMatthew Ahrens (void) zio_wait(spa->spa_async_zio_root[i]); 12526f834bc1SMatthew Ahrens kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 125354d692b7SGeorge Wilson spa->spa_async_zio_root = NULL; 125454d692b7SGeorge Wilson } 1255fa9e4066Sahrens 1256cde58dbcSMatthew Ahrens bpobj_close(&spa->spa_deferred_bpobj); 1257cde58dbcSMatthew Ahrens 12580713e232SGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 12590713e232SGeorge Wilson 12600713e232SGeorge Wilson /* 12610713e232SGeorge Wilson * Close all vdevs. 12620713e232SGeorge Wilson */ 12630713e232SGeorge Wilson if (spa->spa_root_vdev) 12640713e232SGeorge Wilson vdev_free(spa->spa_root_vdev); 12650713e232SGeorge Wilson ASSERT(spa->spa_root_vdev == NULL); 12660713e232SGeorge Wilson 1267fa9e4066Sahrens /* 1268fa9e4066Sahrens * Close the dsl pool. 1269fa9e4066Sahrens */ 1270fa9e4066Sahrens if (spa->spa_dsl_pool) { 1271fa9e4066Sahrens dsl_pool_close(spa->spa_dsl_pool); 1272fa9e4066Sahrens spa->spa_dsl_pool = NULL; 1273afee20e4SGeorge Wilson spa->spa_meta_objset = NULL; 1274fa9e4066Sahrens } 1275fa9e4066Sahrens 1276b24ab676SJeff Bonwick ddt_unload(spa); 1277b24ab676SJeff Bonwick 12788ad4d6ddSJeff Bonwick 12798ad4d6ddSJeff Bonwick /* 12808ad4d6ddSJeff Bonwick * Drop and purge level 2 cache 12818ad4d6ddSJeff Bonwick */ 12828ad4d6ddSJeff Bonwick spa_l2cache_drop(spa); 12838ad4d6ddSJeff Bonwick 1284fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 1285fa94a07fSbrendan vdev_free(spa->spa_spares.sav_vdevs[i]); 1286fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) { 1287fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 1288fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 1289fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 129099653d4eSeschrock } 1291fa94a07fSbrendan if (spa->spa_spares.sav_config) { 1292fa94a07fSbrendan nvlist_free(spa->spa_spares.sav_config); 1293fa94a07fSbrendan spa->spa_spares.sav_config = NULL; 1294fa94a07fSbrendan } 12952ce8af81SEric Schrock spa->spa_spares.sav_count = 0; 1296fa94a07fSbrendan 1297cd0837ccSGeorge Wilson for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1298cd0837ccSGeorge Wilson vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1299fa94a07fSbrendan vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1300cd0837ccSGeorge Wilson } 1301fa94a07fSbrendan if (spa->spa_l2cache.sav_vdevs) { 1302fa94a07fSbrendan kmem_free(spa->spa_l2cache.sav_vdevs, 1303fa94a07fSbrendan spa->spa_l2cache.sav_count * sizeof (void *)); 1304fa94a07fSbrendan spa->spa_l2cache.sav_vdevs = NULL; 1305fa94a07fSbrendan } 1306fa94a07fSbrendan if (spa->spa_l2cache.sav_config) { 1307fa94a07fSbrendan nvlist_free(spa->spa_l2cache.sav_config); 1308fa94a07fSbrendan spa->spa_l2cache.sav_config = NULL; 130999653d4eSeschrock } 13102ce8af81SEric Schrock spa->spa_l2cache.sav_count = 0; 131199653d4eSeschrock 1312ea8dc4b6Seschrock spa->spa_async_suspended = 0; 13138ad4d6ddSJeff Bonwick 13148704186eSDan McDonald if (spa->spa_comment != NULL) { 13158704186eSDan McDonald spa_strfree(spa->spa_comment); 13168704186eSDan McDonald spa->spa_comment = NULL; 13178704186eSDan McDonald } 13188704186eSDan McDonald 13198ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1320fa9e4066Sahrens } 1321fa9e4066Sahrens 132299653d4eSeschrock /* 132399653d4eSeschrock * Load (or re-load) the current list of vdevs describing the active spares for 132499653d4eSeschrock * this pool. When this is called, we have some form of basic information in 1325fa94a07fSbrendan * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1326fa94a07fSbrendan * then re-generate a more complete list including status information. 132799653d4eSeschrock */ 132899653d4eSeschrock static void 132999653d4eSeschrock spa_load_spares(spa_t *spa) 133099653d4eSeschrock { 133199653d4eSeschrock nvlist_t **spares; 133299653d4eSeschrock uint_t nspares; 133399653d4eSeschrock int i; 133439c23413Seschrock vdev_t *vd, *tvd; 133599653d4eSeschrock 1336e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1337e14bb325SJeff Bonwick 133899653d4eSeschrock /* 133999653d4eSeschrock * First, close and free any existing spare vdevs. 134099653d4eSeschrock */ 1341fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 1342fa94a07fSbrendan vd = spa->spa_spares.sav_vdevs[i]; 134339c23413Seschrock 134439c23413Seschrock /* Undo the call to spa_activate() below */ 1345c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1346c5904d13Seschrock B_FALSE)) != NULL && tvd->vdev_isspare) 134739c23413Seschrock spa_spare_remove(tvd); 134839c23413Seschrock vdev_close(vd); 134939c23413Seschrock vdev_free(vd); 135099653d4eSeschrock } 135139c23413Seschrock 1352fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) 1353fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 1354fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 135599653d4eSeschrock 1356fa94a07fSbrendan if (spa->spa_spares.sav_config == NULL) 135799653d4eSeschrock nspares = 0; 135899653d4eSeschrock else 1359fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 136099653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 136199653d4eSeschrock 1362fa94a07fSbrendan spa->spa_spares.sav_count = (int)nspares; 1363fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 136499653d4eSeschrock 136599653d4eSeschrock if (nspares == 0) 136699653d4eSeschrock return; 136799653d4eSeschrock 136899653d4eSeschrock /* 136999653d4eSeschrock * Construct the array of vdevs, opening them to get status in the 137039c23413Seschrock * process. For each spare, there is potentially two different vdev_t 137139c23413Seschrock * structures associated with it: one in the list of spares (used only 137239c23413Seschrock * for basic validation purposes) and one in the active vdev 137339c23413Seschrock * configuration (if it's spared in). During this phase we open and 137439c23413Seschrock * validate each vdev on the spare list. If the vdev also exists in the 137539c23413Seschrock * active configuration, then we also mark this vdev as an active spare. 137699653d4eSeschrock */ 1377fa94a07fSbrendan spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1378fa94a07fSbrendan KM_SLEEP); 1379fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 138099653d4eSeschrock VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 138199653d4eSeschrock VDEV_ALLOC_SPARE) == 0); 138299653d4eSeschrock ASSERT(vd != NULL); 138399653d4eSeschrock 1384fa94a07fSbrendan spa->spa_spares.sav_vdevs[i] = vd; 138599653d4eSeschrock 1386c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1387c5904d13Seschrock B_FALSE)) != NULL) { 138839c23413Seschrock if (!tvd->vdev_isspare) 138939c23413Seschrock spa_spare_add(tvd); 139039c23413Seschrock 139139c23413Seschrock /* 139239c23413Seschrock * We only mark the spare active if we were successfully 139339c23413Seschrock * able to load the vdev. Otherwise, importing a pool 139439c23413Seschrock * with a bad active spare would result in strange 139539c23413Seschrock * behavior, because multiple pool would think the spare 139639c23413Seschrock * is actively in use. 139739c23413Seschrock * 139839c23413Seschrock * There is a vulnerability here to an equally bizarre 139939c23413Seschrock * circumstance, where a dead active spare is later 140039c23413Seschrock * brought back to life (onlined or otherwise). Given 140139c23413Seschrock * the rarity of this scenario, and the extra complexity 140239c23413Seschrock * it adds, we ignore the possibility. 140339c23413Seschrock */ 140439c23413Seschrock if (!vdev_is_dead(tvd)) 140539c23413Seschrock spa_spare_activate(tvd); 140639c23413Seschrock } 140739c23413Seschrock 1408e14bb325SJeff Bonwick vd->vdev_top = vd; 14096809eb4eSEric Schrock vd->vdev_aux = &spa->spa_spares; 1410e14bb325SJeff Bonwick 141199653d4eSeschrock if (vdev_open(vd) != 0) 141299653d4eSeschrock continue; 141399653d4eSeschrock 1414fa94a07fSbrendan if (vdev_validate_aux(vd) == 0) 1415fa94a07fSbrendan spa_spare_add(vd); 141699653d4eSeschrock } 141799653d4eSeschrock 141899653d4eSeschrock /* 141999653d4eSeschrock * Recompute the stashed list of spares, with status information 142099653d4eSeschrock * this time. 142199653d4eSeschrock */ 1422fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 142399653d4eSeschrock DATA_TYPE_NVLIST_ARRAY) == 0); 142499653d4eSeschrock 1425fa94a07fSbrendan spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1426fa94a07fSbrendan KM_SLEEP); 1427fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 1428fa94a07fSbrendan spares[i] = vdev_config_generate(spa, 14293f9d6ad7SLin Ling spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1430fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1431fa94a07fSbrendan ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1432fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 143399653d4eSeschrock nvlist_free(spares[i]); 1434fa94a07fSbrendan kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1435fa94a07fSbrendan } 1436fa94a07fSbrendan 1437fa94a07fSbrendan /* 1438fa94a07fSbrendan * Load (or re-load) the current list of vdevs describing the active l2cache for 1439fa94a07fSbrendan * this pool. When this is called, we have some form of basic information in 1440fa94a07fSbrendan * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1441fa94a07fSbrendan * then re-generate a more complete list including status information. 1442fa94a07fSbrendan * Devices which are already active have their details maintained, and are 1443fa94a07fSbrendan * not re-opened. 1444fa94a07fSbrendan */ 1445fa94a07fSbrendan static void 1446fa94a07fSbrendan spa_load_l2cache(spa_t *spa) 1447fa94a07fSbrendan { 1448fa94a07fSbrendan nvlist_t **l2cache; 1449fa94a07fSbrendan uint_t nl2cache; 1450fa94a07fSbrendan int i, j, oldnvdevs; 1451573ca77eSGeorge Wilson uint64_t guid; 1452fa94a07fSbrendan vdev_t *vd, **oldvdevs, **newvdevs; 1453fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 1454fa94a07fSbrendan 1455e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1456e14bb325SJeff Bonwick 1457fa94a07fSbrendan if (sav->sav_config != NULL) { 1458fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1459fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1460fa94a07fSbrendan newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1461fa94a07fSbrendan } else { 1462fa94a07fSbrendan nl2cache = 0; 1463d5285caeSGeorge Wilson newvdevs = NULL; 1464fa94a07fSbrendan } 1465fa94a07fSbrendan 1466fa94a07fSbrendan oldvdevs = sav->sav_vdevs; 1467fa94a07fSbrendan oldnvdevs = sav->sav_count; 1468fa94a07fSbrendan sav->sav_vdevs = NULL; 1469fa94a07fSbrendan sav->sav_count = 0; 1470fa94a07fSbrendan 1471fa94a07fSbrendan /* 1472fa94a07fSbrendan * Process new nvlist of vdevs. 1473fa94a07fSbrendan */ 1474fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 1475fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1476fa94a07fSbrendan &guid) == 0); 1477fa94a07fSbrendan 1478fa94a07fSbrendan newvdevs[i] = NULL; 1479fa94a07fSbrendan for (j = 0; j < oldnvdevs; j++) { 1480fa94a07fSbrendan vd = oldvdevs[j]; 1481fa94a07fSbrendan if (vd != NULL && guid == vd->vdev_guid) { 1482fa94a07fSbrendan /* 1483fa94a07fSbrendan * Retain previous vdev for add/remove ops. 1484fa94a07fSbrendan */ 1485fa94a07fSbrendan newvdevs[i] = vd; 1486fa94a07fSbrendan oldvdevs[j] = NULL; 1487fa94a07fSbrendan break; 1488fa94a07fSbrendan } 1489fa94a07fSbrendan } 1490fa94a07fSbrendan 1491fa94a07fSbrendan if (newvdevs[i] == NULL) { 1492fa94a07fSbrendan /* 1493fa94a07fSbrendan * Create new vdev 1494fa94a07fSbrendan */ 1495fa94a07fSbrendan VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1496fa94a07fSbrendan VDEV_ALLOC_L2CACHE) == 0); 1497fa94a07fSbrendan ASSERT(vd != NULL); 1498fa94a07fSbrendan newvdevs[i] = vd; 1499fa94a07fSbrendan 1500fa94a07fSbrendan /* 1501fa94a07fSbrendan * Commit this vdev as an l2cache device, 1502fa94a07fSbrendan * even if it fails to open. 1503fa94a07fSbrendan */ 1504fa94a07fSbrendan spa_l2cache_add(vd); 1505fa94a07fSbrendan 1506c5904d13Seschrock vd->vdev_top = vd; 1507c5904d13Seschrock vd->vdev_aux = sav; 1508c5904d13Seschrock 1509c5904d13Seschrock spa_l2cache_activate(vd); 1510c5904d13Seschrock 1511fa94a07fSbrendan if (vdev_open(vd) != 0) 1512fa94a07fSbrendan continue; 1513fa94a07fSbrendan 1514fa94a07fSbrendan (void) vdev_validate_aux(vd); 1515fa94a07fSbrendan 1516573ca77eSGeorge Wilson if (!vdev_is_dead(vd)) 1517573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1518fa94a07fSbrendan } 1519fa94a07fSbrendan } 1520fa94a07fSbrendan 1521fa94a07fSbrendan /* 1522fa94a07fSbrendan * Purge vdevs that were dropped 1523fa94a07fSbrendan */ 1524fa94a07fSbrendan for (i = 0; i < oldnvdevs; i++) { 1525fa94a07fSbrendan uint64_t pool; 1526fa94a07fSbrendan 1527fa94a07fSbrendan vd = oldvdevs[i]; 1528fa94a07fSbrendan if (vd != NULL) { 1529cd0837ccSGeorge Wilson ASSERT(vd->vdev_isl2cache); 1530cd0837ccSGeorge Wilson 15318ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 15328ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 1533fa94a07fSbrendan l2arc_remove_vdev(vd); 1534cd0837ccSGeorge Wilson vdev_clear_stats(vd); 1535cd0837ccSGeorge Wilson vdev_free(vd); 1536fa94a07fSbrendan } 1537fa94a07fSbrendan } 1538fa94a07fSbrendan 1539fa94a07fSbrendan if (oldvdevs) 1540fa94a07fSbrendan kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1541fa94a07fSbrendan 1542fa94a07fSbrendan if (sav->sav_config == NULL) 1543fa94a07fSbrendan goto out; 1544fa94a07fSbrendan 1545fa94a07fSbrendan sav->sav_vdevs = newvdevs; 1546fa94a07fSbrendan sav->sav_count = (int)nl2cache; 1547fa94a07fSbrendan 1548fa94a07fSbrendan /* 1549fa94a07fSbrendan * Recompute the stashed list of l2cache devices, with status 1550fa94a07fSbrendan * information this time. 1551fa94a07fSbrendan */ 1552fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1553fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 1554fa94a07fSbrendan 1555fa94a07fSbrendan l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1556fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1557fa94a07fSbrendan l2cache[i] = vdev_config_generate(spa, 15583f9d6ad7SLin Ling sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1559fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1560fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1561fa94a07fSbrendan out: 1562fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1563fa94a07fSbrendan nvlist_free(l2cache[i]); 1564fa94a07fSbrendan if (sav->sav_count) 1565fa94a07fSbrendan kmem_free(l2cache, sav->sav_count * sizeof (void *)); 156699653d4eSeschrock } 156799653d4eSeschrock 156899653d4eSeschrock static int 156999653d4eSeschrock load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 157099653d4eSeschrock { 157199653d4eSeschrock dmu_buf_t *db; 157299653d4eSeschrock char *packed = NULL; 157399653d4eSeschrock size_t nvsize = 0; 157499653d4eSeschrock int error; 157599653d4eSeschrock *value = NULL; 157699653d4eSeschrock 1577a45f1c3cSBrian Behlendorf error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1578a45f1c3cSBrian Behlendorf if (error != 0) 1579a45f1c3cSBrian Behlendorf return (error); 1580a45f1c3cSBrian Behlendorf 158199653d4eSeschrock nvsize = *(uint64_t *)db->db_data; 158299653d4eSeschrock dmu_buf_rele(db, FTAG); 158399653d4eSeschrock 158499653d4eSeschrock packed = kmem_alloc(nvsize, KM_SLEEP); 15857bfdf011SNeil Perrin error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 15867bfdf011SNeil Perrin DMU_READ_PREFETCH); 158799653d4eSeschrock if (error == 0) 158899653d4eSeschrock error = nvlist_unpack(packed, nvsize, value, 0); 158999653d4eSeschrock kmem_free(packed, nvsize); 159099653d4eSeschrock 159199653d4eSeschrock return (error); 159299653d4eSeschrock } 159399653d4eSeschrock 15943d7072f8Seschrock /* 15953d7072f8Seschrock * Checks to see if the given vdev could not be opened, in which case we post a 15963d7072f8Seschrock * sysevent to notify the autoreplace code that the device has been removed. 15973d7072f8Seschrock */ 15983d7072f8Seschrock static void 15993d7072f8Seschrock spa_check_removed(vdev_t *vd) 16003d7072f8Seschrock { 1601573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16023d7072f8Seschrock spa_check_removed(vd->vdev_child[c]); 16033d7072f8Seschrock 1604efb4a871SYuri Pankov if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1605efb4a871SYuri Pankov !vd->vdev_ishole) { 16063d7072f8Seschrock zfs_post_autoreplace(vd->vdev_spa, vd); 16073d7072f8Seschrock spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 16083d7072f8Seschrock } 16093d7072f8Seschrock } 16103d7072f8Seschrock 1611215198a6SJoe Stein static void 1612215198a6SJoe Stein spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) 1613215198a6SJoe Stein { 1614215198a6SJoe Stein ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); 1615215198a6SJoe Stein 1616215198a6SJoe Stein vd->vdev_top_zap = mvd->vdev_top_zap; 1617215198a6SJoe Stein vd->vdev_leaf_zap = mvd->vdev_leaf_zap; 1618215198a6SJoe Stein 1619215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 1620215198a6SJoe Stein spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); 1621215198a6SJoe Stein } 1622215198a6SJoe Stein } 1623215198a6SJoe Stein 1624e6ca193dSGeorge Wilson /* 16254b964adaSGeorge Wilson * Validate the current config against the MOS config 1626e6ca193dSGeorge Wilson */ 16274b964adaSGeorge Wilson static boolean_t 16284b964adaSGeorge Wilson spa_config_valid(spa_t *spa, nvlist_t *config) 1629e6ca193dSGeorge Wilson { 16304b964adaSGeorge Wilson vdev_t *mrvd, *rvd = spa->spa_root_vdev; 16314b964adaSGeorge Wilson nvlist_t *nv; 16324b964adaSGeorge Wilson 16334b964adaSGeorge Wilson VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 16344b964adaSGeorge Wilson 16354b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 16364b964adaSGeorge Wilson VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 16374b964adaSGeorge Wilson 16384b964adaSGeorge Wilson ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1639e6ca193dSGeorge Wilson 164088ecc943SGeorge Wilson /* 16414b964adaSGeorge Wilson * If we're doing a normal import, then build up any additional 16424b964adaSGeorge Wilson * diagnostic information about missing devices in this config. 16434b964adaSGeorge Wilson * We'll pass this up to the user for further processing. 164488ecc943SGeorge Wilson */ 16454b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 16464b964adaSGeorge Wilson nvlist_t **child, *nv; 16474b964adaSGeorge Wilson uint64_t idx = 0; 16484b964adaSGeorge Wilson 16494b964adaSGeorge Wilson child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 16504b964adaSGeorge Wilson KM_SLEEP); 16514b964adaSGeorge Wilson VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1652e6ca193dSGeorge Wilson 16534b964adaSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 16544b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 16554b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 16564b964adaSGeorge Wilson 16574b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 16584b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops && 16594b964adaSGeorge Wilson mtvd->vdev_islog) 16604b964adaSGeorge Wilson child[idx++] = vdev_config_generate(spa, mtvd, 16614b964adaSGeorge Wilson B_FALSE, 0); 16624b964adaSGeorge Wilson } 16634b964adaSGeorge Wilson 16644b964adaSGeorge Wilson if (idx) { 16654b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist_array(nv, 16664b964adaSGeorge Wilson ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 16674b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(spa->spa_load_info, 16684b964adaSGeorge Wilson ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 16694b964adaSGeorge Wilson 16704b964adaSGeorge Wilson for (int i = 0; i < idx; i++) 16714b964adaSGeorge Wilson nvlist_free(child[i]); 16724b964adaSGeorge Wilson } 16734b964adaSGeorge Wilson nvlist_free(nv); 16744b964adaSGeorge Wilson kmem_free(child, rvd->vdev_children * sizeof (char **)); 16754b964adaSGeorge Wilson } 16764b964adaSGeorge Wilson 16774b964adaSGeorge Wilson /* 16784b964adaSGeorge Wilson * Compare the root vdev tree with the information we have 16794b964adaSGeorge Wilson * from the MOS config (mrvd). Check each top-level vdev 16804b964adaSGeorge Wilson * with the corresponding MOS config top-level (mtvd). 16814b964adaSGeorge Wilson */ 168288ecc943SGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 16834b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 16844b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 16854b964adaSGeorge Wilson 16864b964adaSGeorge Wilson /* 16874b964adaSGeorge Wilson * Resolve any "missing" vdevs in the current configuration. 16884b964adaSGeorge Wilson * If we find that the MOS config has more accurate information 16894b964adaSGeorge Wilson * about the top-level vdev then use that vdev instead. 16904b964adaSGeorge Wilson */ 16914b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 16924b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops) { 16934b964adaSGeorge Wilson 16944b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 16954b964adaSGeorge Wilson continue; 16964b964adaSGeorge Wilson 16974b964adaSGeorge Wilson /* 16984b964adaSGeorge Wilson * Device specific actions. 16994b964adaSGeorge Wilson */ 17004b964adaSGeorge Wilson if (mtvd->vdev_islog) { 17014b964adaSGeorge Wilson spa_set_log_state(spa, SPA_LOG_CLEAR); 17024b964adaSGeorge Wilson } else { 17034b964adaSGeorge Wilson /* 17044b964adaSGeorge Wilson * XXX - once we have 'readonly' pool 17054b964adaSGeorge Wilson * support we should be able to handle 17064b964adaSGeorge Wilson * missing data devices by transitioning 17074b964adaSGeorge Wilson * the pool to readonly. 17084b964adaSGeorge Wilson */ 17094b964adaSGeorge Wilson continue; 17104b964adaSGeorge Wilson } 17114b964adaSGeorge Wilson 17124b964adaSGeorge Wilson /* 17134b964adaSGeorge Wilson * Swap the missing vdev with the data we were 17144b964adaSGeorge Wilson * able to obtain from the MOS config. 17154b964adaSGeorge Wilson */ 17164b964adaSGeorge Wilson vdev_remove_child(rvd, tvd); 17174b964adaSGeorge Wilson vdev_remove_child(mrvd, mtvd); 17184b964adaSGeorge Wilson 17194b964adaSGeorge Wilson vdev_add_child(rvd, mtvd); 17204b964adaSGeorge Wilson vdev_add_child(mrvd, tvd); 17214b964adaSGeorge Wilson 17224b964adaSGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 17234b964adaSGeorge Wilson vdev_load(mtvd); 17244b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 17254b964adaSGeorge Wilson 17264b964adaSGeorge Wilson vdev_reopen(rvd); 1727215198a6SJoe Stein } else { 1728215198a6SJoe Stein if (mtvd->vdev_islog) { 1729215198a6SJoe Stein /* 1730215198a6SJoe Stein * Load the slog device's state from the MOS 1731215198a6SJoe Stein * config since it's possible that the label 1732215198a6SJoe Stein * does not contain the most up-to-date 1733215198a6SJoe Stein * information. 1734215198a6SJoe Stein */ 1735215198a6SJoe Stein vdev_load_log_state(tvd, mtvd); 1736215198a6SJoe Stein vdev_reopen(tvd); 1737215198a6SJoe Stein } 1738215198a6SJoe Stein 17394b964adaSGeorge Wilson /* 1740215198a6SJoe Stein * Per-vdev ZAP info is stored exclusively in the MOS. 17414b964adaSGeorge Wilson */ 1742215198a6SJoe Stein spa_config_valid_zaps(tvd, mtvd); 17434b964adaSGeorge Wilson } 1744e6ca193dSGeorge Wilson } 1745215198a6SJoe Stein 17464b964adaSGeorge Wilson vdev_free(mrvd); 174788ecc943SGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 17484b964adaSGeorge Wilson 17494b964adaSGeorge Wilson /* 17504b964adaSGeorge Wilson * Ensure we were able to validate the config. 17514b964adaSGeorge Wilson */ 17524b964adaSGeorge Wilson return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1753e6ca193dSGeorge Wilson } 1754e6ca193dSGeorge Wilson 1755b87f3af3Sperrin /* 1756b87f3af3Sperrin * Check for missing log devices 1757b87f3af3Sperrin */ 17583b2aab18SMatthew Ahrens static boolean_t 1759b87f3af3Sperrin spa_check_logs(spa_t *spa) 1760b87f3af3Sperrin { 17613b2aab18SMatthew Ahrens boolean_t rv = B_FALSE; 176212380e1eSArne Jansen dsl_pool_t *dp = spa_get_dsl(spa); 17633b2aab18SMatthew Ahrens 1764b87f3af3Sperrin switch (spa->spa_log_state) { 1765b87f3af3Sperrin case SPA_LOG_MISSING: 1766b87f3af3Sperrin /* need to recheck in case slog has been restored */ 1767b87f3af3Sperrin case SPA_LOG_UNKNOWN: 176812380e1eSArne Jansen rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 176912380e1eSArne Jansen zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 17703b2aab18SMatthew Ahrens if (rv) 17711195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_MISSING); 1772b87f3af3Sperrin break; 1773b87f3af3Sperrin } 17743b2aab18SMatthew Ahrens return (rv); 1775b87f3af3Sperrin } 1776b87f3af3Sperrin 17771195e687SMark J Musante static boolean_t 17781195e687SMark J Musante spa_passivate_log(spa_t *spa) 17791195e687SMark J Musante { 17801195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 17811195e687SMark J Musante boolean_t slog_found = B_FALSE; 17821195e687SMark J Musante 17831195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 17841195e687SMark J Musante 17851195e687SMark J Musante if (!spa_has_slogs(spa)) 17861195e687SMark J Musante return (B_FALSE); 17871195e687SMark J Musante 17881195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 17891195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 17901195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 17911195e687SMark J Musante 17921195e687SMark J Musante if (tvd->vdev_islog) { 17931195e687SMark J Musante metaslab_group_passivate(mg); 17941195e687SMark J Musante slog_found = B_TRUE; 17951195e687SMark J Musante } 17961195e687SMark J Musante } 17971195e687SMark J Musante 17981195e687SMark J Musante return (slog_found); 17991195e687SMark J Musante } 18001195e687SMark J Musante 18011195e687SMark J Musante static void 18021195e687SMark J Musante spa_activate_log(spa_t *spa) 18031195e687SMark J Musante { 18041195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 18051195e687SMark J Musante 18061195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 18071195e687SMark J Musante 18081195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 18091195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 18101195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 18111195e687SMark J Musante 18121195e687SMark J Musante if (tvd->vdev_islog) 18131195e687SMark J Musante metaslab_group_activate(mg); 18141195e687SMark J Musante } 18151195e687SMark J Musante } 18161195e687SMark J Musante 18171195e687SMark J Musante int 18181195e687SMark J Musante spa_offline_log(spa_t *spa) 18191195e687SMark J Musante { 18203b2aab18SMatthew Ahrens int error; 18211195e687SMark J Musante 18223b2aab18SMatthew Ahrens error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 18233b2aab18SMatthew Ahrens NULL, DS_FIND_CHILDREN); 18243b2aab18SMatthew Ahrens if (error == 0) { 18251195e687SMark J Musante /* 18261195e687SMark J Musante * We successfully offlined the log device, sync out the 18271195e687SMark J Musante * current txg so that the "stubby" block can be removed 18281195e687SMark J Musante * by zil_sync(). 18291195e687SMark J Musante */ 18301195e687SMark J Musante txg_wait_synced(spa->spa_dsl_pool, 0); 18311195e687SMark J Musante } 18321195e687SMark J Musante return (error); 18331195e687SMark J Musante } 18341195e687SMark J Musante 1835b693757aSEric Schrock static void 1836b693757aSEric Schrock spa_aux_check_removed(spa_aux_vdev_t *sav) 1837b693757aSEric Schrock { 1838b24ab676SJeff Bonwick for (int i = 0; i < sav->sav_count; i++) 1839b693757aSEric Schrock spa_check_removed(sav->sav_vdevs[i]); 1840b693757aSEric Schrock } 1841b693757aSEric Schrock 1842b24ab676SJeff Bonwick void 1843b24ab676SJeff Bonwick spa_claim_notify(zio_t *zio) 1844b24ab676SJeff Bonwick { 1845b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 1846b24ab676SJeff Bonwick 1847b24ab676SJeff Bonwick if (zio->io_error) 1848b24ab676SJeff Bonwick return; 1849b24ab676SJeff Bonwick 1850b24ab676SJeff Bonwick mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1851b24ab676SJeff Bonwick if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1852b24ab676SJeff Bonwick spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1853b24ab676SJeff Bonwick mutex_exit(&spa->spa_props_lock); 1854b24ab676SJeff Bonwick } 1855b24ab676SJeff Bonwick 1856468c413aSTim Haley typedef struct spa_load_error { 1857c8ee1847SVictor Latushkin uint64_t sle_meta_count; 1858468c413aSTim Haley uint64_t sle_data_count; 1859468c413aSTim Haley } spa_load_error_t; 1860468c413aSTim Haley 1861468c413aSTim Haley static void 1862468c413aSTim Haley spa_load_verify_done(zio_t *zio) 1863468c413aSTim Haley { 1864468c413aSTim Haley blkptr_t *bp = zio->io_bp; 1865468c413aSTim Haley spa_load_error_t *sle = zio->io_private; 1866468c413aSTim Haley dmu_object_type_t type = BP_GET_TYPE(bp); 1867468c413aSTim Haley int error = zio->io_error; 1868e42d2059SMatthew Ahrens spa_t *spa = zio->io_spa; 1869468c413aSTim Haley 1870468c413aSTim Haley if (error) { 1871ad135b5dSChristopher Siden if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1872468c413aSTim Haley type != DMU_OT_INTENT_LOG) 18731a5e258fSJosef 'Jeff' Sipek atomic_inc_64(&sle->sle_meta_count); 1874468c413aSTim Haley else 18751a5e258fSJosef 'Jeff' Sipek atomic_inc_64(&sle->sle_data_count); 1876468c413aSTim Haley } 1877468c413aSTim Haley zio_data_buf_free(zio->io_data, zio->io_size); 1878e42d2059SMatthew Ahrens 1879e42d2059SMatthew Ahrens mutex_enter(&spa->spa_scrub_lock); 1880e42d2059SMatthew Ahrens spa->spa_scrub_inflight--; 1881e42d2059SMatthew Ahrens cv_broadcast(&spa->spa_scrub_io_cv); 1882e42d2059SMatthew Ahrens mutex_exit(&spa->spa_scrub_lock); 1883468c413aSTim Haley } 1884468c413aSTim Haley 1885e42d2059SMatthew Ahrens /* 1886e42d2059SMatthew Ahrens * Maximum number of concurrent scrub i/os to create while verifying 1887e42d2059SMatthew Ahrens * a pool while importing it. 1888e42d2059SMatthew Ahrens */ 1889e42d2059SMatthew Ahrens int spa_load_verify_maxinflight = 10000; 1890e42d2059SMatthew Ahrens boolean_t spa_load_verify_metadata = B_TRUE; 1891e42d2059SMatthew Ahrens boolean_t spa_load_verify_data = B_TRUE; 1892e42d2059SMatthew Ahrens 1893468c413aSTim Haley /*ARGSUSED*/ 1894468c413aSTim Haley static int 1895b24ab676SJeff Bonwick spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 18967802d7bfSMatthew Ahrens const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1897468c413aSTim Haley { 1898a2cdcdd2SPaul Dagnelie if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1899e42d2059SMatthew Ahrens return (0); 1900e42d2059SMatthew Ahrens /* 1901e42d2059SMatthew Ahrens * Note: normally this routine will not be called if 1902e42d2059SMatthew Ahrens * spa_load_verify_metadata is not set. However, it may be useful 1903e42d2059SMatthew Ahrens * to manually set the flag after the traversal has begun. 1904e42d2059SMatthew Ahrens */ 1905e42d2059SMatthew Ahrens if (!spa_load_verify_metadata) 1906e42d2059SMatthew Ahrens return (0); 1907e42d2059SMatthew Ahrens if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1908e42d2059SMatthew Ahrens return (0); 1909468c413aSTim Haley 1910e42d2059SMatthew Ahrens zio_t *rio = arg; 1911e42d2059SMatthew Ahrens size_t size = BP_GET_PSIZE(bp); 1912e42d2059SMatthew Ahrens void *data = zio_data_buf_alloc(size); 1913e42d2059SMatthew Ahrens 1914e42d2059SMatthew Ahrens mutex_enter(&spa->spa_scrub_lock); 1915e42d2059SMatthew Ahrens while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1916e42d2059SMatthew Ahrens cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1917e42d2059SMatthew Ahrens spa->spa_scrub_inflight++; 1918e42d2059SMatthew Ahrens mutex_exit(&spa->spa_scrub_lock); 1919e42d2059SMatthew Ahrens 1920e42d2059SMatthew Ahrens zio_nowait(zio_read(rio, spa, bp, data, size, 1921e42d2059SMatthew Ahrens spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1922e42d2059SMatthew Ahrens ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1923e42d2059SMatthew Ahrens ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1924468c413aSTim Haley return (0); 1925468c413aSTim Haley } 1926468c413aSTim Haley 1927c971037bSPaul Dagnelie /* ARGSUSED */ 1928c971037bSPaul Dagnelie int 1929c971037bSPaul Dagnelie verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1930c971037bSPaul Dagnelie { 1931*9adfa60dSMatthew Ahrens if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 1932c971037bSPaul Dagnelie return (SET_ERROR(ENAMETOOLONG)); 1933c971037bSPaul Dagnelie 1934c971037bSPaul Dagnelie return (0); 1935c971037bSPaul Dagnelie } 1936c971037bSPaul Dagnelie 1937468c413aSTim Haley static int 1938468c413aSTim Haley spa_load_verify(spa_t *spa) 1939468c413aSTim Haley { 1940468c413aSTim Haley zio_t *rio; 1941468c413aSTim Haley spa_load_error_t sle = { 0 }; 1942468c413aSTim Haley zpool_rewind_policy_t policy; 1943468c413aSTim Haley boolean_t verify_ok = B_FALSE; 1944e42d2059SMatthew Ahrens int error = 0; 1945468c413aSTim Haley 1946c8ee1847SVictor Latushkin zpool_get_rewind_policy(spa->spa_config, &policy); 1947c8ee1847SVictor Latushkin 1948c8ee1847SVictor Latushkin if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1949c8ee1847SVictor Latushkin return (0); 1950c8ee1847SVictor Latushkin 1951c971037bSPaul Dagnelie dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 1952c971037bSPaul Dagnelie error = dmu_objset_find_dp(spa->spa_dsl_pool, 1953c971037bSPaul Dagnelie spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 1954c971037bSPaul Dagnelie DS_FIND_CHILDREN); 1955c971037bSPaul Dagnelie dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 1956c971037bSPaul Dagnelie if (error != 0) 1957c971037bSPaul Dagnelie return (error); 1958c971037bSPaul Dagnelie 1959468c413aSTim Haley rio = zio_root(spa, NULL, &sle, 1960468c413aSTim Haley ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1961468c413aSTim Haley 1962e42d2059SMatthew Ahrens if (spa_load_verify_metadata) { 1963e42d2059SMatthew Ahrens error = traverse_pool(spa, spa->spa_verify_min_txg, 1964e42d2059SMatthew Ahrens TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1965e42d2059SMatthew Ahrens spa_load_verify_cb, rio); 1966e42d2059SMatthew Ahrens } 1967468c413aSTim Haley 1968468c413aSTim Haley (void) zio_wait(rio); 1969468c413aSTim Haley 1970c8ee1847SVictor Latushkin spa->spa_load_meta_errors = sle.sle_meta_count; 1971468c413aSTim Haley spa->spa_load_data_errors = sle.sle_data_count; 1972468c413aSTim Haley 1973c8ee1847SVictor Latushkin if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1974468c413aSTim Haley sle.sle_data_count <= policy.zrp_maxdata) { 19754b964adaSGeorge Wilson int64_t loss = 0; 19764b964adaSGeorge Wilson 1977468c413aSTim Haley verify_ok = B_TRUE; 1978468c413aSTim Haley spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1979468c413aSTim Haley spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 19804b964adaSGeorge Wilson 19814b964adaSGeorge Wilson loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 19824b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 19834b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 19844b964adaSGeorge Wilson VERIFY(nvlist_add_int64(spa->spa_load_info, 19854b964adaSGeorge Wilson ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 19864b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 19874b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1988a33cae98STim Haley } else { 1989a33cae98STim Haley spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1990468c413aSTim Haley } 1991468c413aSTim Haley 1992468c413aSTim Haley if (error) { 1993468c413aSTim Haley if (error != ENXIO && error != EIO) 1994be6fd75aSMatthew Ahrens error = SET_ERROR(EIO); 1995468c413aSTim Haley return (error); 1996468c413aSTim Haley } 1997468c413aSTim Haley 1998468c413aSTim Haley return (verify_ok ? 0 : EIO); 1999468c413aSTim Haley } 2000468c413aSTim Haley 20011195e687SMark J Musante /* 20021195e687SMark J Musante * Find a value in the pool props object. 20031195e687SMark J Musante */ 20041195e687SMark J Musante static void 20051195e687SMark J Musante spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 20061195e687SMark J Musante { 20071195e687SMark J Musante (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 20081195e687SMark J Musante zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 20091195e687SMark J Musante } 20101195e687SMark J Musante 20111195e687SMark J Musante /* 20121195e687SMark J Musante * Find a value in the pool directory object. 20131195e687SMark J Musante */ 20141195e687SMark J Musante static int 20151195e687SMark J Musante spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 20161195e687SMark J Musante { 20171195e687SMark J Musante return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 20181195e687SMark J Musante name, sizeof (uint64_t), 1, val)); 20191195e687SMark J Musante } 20201195e687SMark J Musante 20211195e687SMark J Musante static int 20221195e687SMark J Musante spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 20231195e687SMark J Musante { 20241195e687SMark J Musante vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 20251195e687SMark J Musante return (err); 20261195e687SMark J Musante } 20271195e687SMark J Musante 20281195e687SMark J Musante /* 20291195e687SMark J Musante * Fix up config after a partly-completed split. This is done with the 20301195e687SMark J Musante * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 20311195e687SMark J Musante * pool have that entry in their config, but only the splitting one contains 20321195e687SMark J Musante * a list of all the guids of the vdevs that are being split off. 20331195e687SMark J Musante * 20341195e687SMark J Musante * This function determines what to do with that list: either rejoin 20351195e687SMark J Musante * all the disks to the pool, or complete the splitting process. To attempt 20361195e687SMark J Musante * the rejoin, each disk that is offlined is marked online again, and 20371195e687SMark J Musante * we do a reopen() call. If the vdev label for every disk that was 20381195e687SMark J Musante * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 20391195e687SMark J Musante * then we call vdev_split() on each disk, and complete the split. 20401195e687SMark J Musante * 2041d41c4376SMark J Musante * Otherwise we leave the config alone, with all the vdevs in place in 2042d41c4376SMark J Musante * the original pool. 20431195e687SMark J Musante */ 20441195e687SMark J Musante static void 20451195e687SMark J Musante spa_try_repair(spa_t *spa, nvlist_t *config) 20461195e687SMark J Musante { 20471195e687SMark J Musante uint_t extracted; 20481195e687SMark J Musante uint64_t *glist; 20491195e687SMark J Musante uint_t i, gcount; 20501195e687SMark J Musante nvlist_t *nvl; 20511195e687SMark J Musante vdev_t **vd; 20521195e687SMark J Musante boolean_t attempt_reopen; 20531195e687SMark J Musante 20541195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 20551195e687SMark J Musante return; 20561195e687SMark J Musante 20571195e687SMark J Musante /* check that the config is complete */ 20581195e687SMark J Musante if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 20591195e687SMark J Musante &glist, &gcount) != 0) 20601195e687SMark J Musante return; 20611195e687SMark J Musante 20621195e687SMark J Musante vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 20631195e687SMark J Musante 20641195e687SMark J Musante /* attempt to online all the vdevs & validate */ 20651195e687SMark J Musante attempt_reopen = B_TRUE; 20661195e687SMark J Musante for (i = 0; i < gcount; i++) { 20671195e687SMark J Musante if (glist[i] == 0) /* vdev is hole */ 20681195e687SMark J Musante continue; 20691195e687SMark J Musante 20701195e687SMark J Musante vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 20711195e687SMark J Musante if (vd[i] == NULL) { 20721195e687SMark J Musante /* 20731195e687SMark J Musante * Don't bother attempting to reopen the disks; 20741195e687SMark J Musante * just do the split. 20751195e687SMark J Musante */ 20761195e687SMark J Musante attempt_reopen = B_FALSE; 20771195e687SMark J Musante } else { 20781195e687SMark J Musante /* attempt to re-online it */ 20791195e687SMark J Musante vd[i]->vdev_offline = B_FALSE; 20801195e687SMark J Musante } 20811195e687SMark J Musante } 20821195e687SMark J Musante 20831195e687SMark J Musante if (attempt_reopen) { 20841195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 20851195e687SMark J Musante 20861195e687SMark J Musante /* check each device to see what state it's in */ 20871195e687SMark J Musante for (extracted = 0, i = 0; i < gcount; i++) { 20881195e687SMark J Musante if (vd[i] != NULL && 20891195e687SMark J Musante vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 20901195e687SMark J Musante break; 20911195e687SMark J Musante ++extracted; 20921195e687SMark J Musante } 20931195e687SMark J Musante } 20941195e687SMark J Musante 20951195e687SMark J Musante /* 20961195e687SMark J Musante * If every disk has been moved to the new pool, or if we never 20971195e687SMark J Musante * even attempted to look at them, then we split them off for 20981195e687SMark J Musante * good. 20991195e687SMark J Musante */ 21001195e687SMark J Musante if (!attempt_reopen || gcount == extracted) { 21011195e687SMark J Musante for (i = 0; i < gcount; i++) 21021195e687SMark J Musante if (vd[i] != NULL) 21031195e687SMark J Musante vdev_split(vd[i]); 21041195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 21051195e687SMark J Musante } 21061195e687SMark J Musante 21071195e687SMark J Musante kmem_free(vd, gcount * sizeof (vdev_t *)); 21081195e687SMark J Musante } 21091195e687SMark J Musante 21101195e687SMark J Musante static int 21111195e687SMark J Musante spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 21121195e687SMark J Musante boolean_t mosconfig) 21131195e687SMark J Musante { 21141195e687SMark J Musante nvlist_t *config = spa->spa_config; 21151195e687SMark J Musante char *ereport = FM_EREPORT_ZFS_POOL; 21168704186eSDan McDonald char *comment; 21171195e687SMark J Musante int error; 21181195e687SMark J Musante uint64_t pool_guid; 21191195e687SMark J Musante nvlist_t *nvl; 21201195e687SMark J Musante 21211195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2122be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 21231195e687SMark J Musante 21248704186eSDan McDonald ASSERT(spa->spa_comment == NULL); 21258704186eSDan McDonald if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 21268704186eSDan McDonald spa->spa_comment = spa_strdup(comment); 21278704186eSDan McDonald 21281195e687SMark J Musante /* 21291195e687SMark J Musante * Versioning wasn't explicitly added to the label until later, so if 21301195e687SMark J Musante * it's not present treat it as the initial version. 21311195e687SMark J Musante */ 21321195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 21331195e687SMark J Musante &spa->spa_ubsync.ub_version) != 0) 21341195e687SMark J Musante spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 21351195e687SMark J Musante 21361195e687SMark J Musante (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 21371195e687SMark J Musante &spa->spa_config_txg); 21381195e687SMark J Musante 21391195e687SMark J Musante if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 21401195e687SMark J Musante spa_guid_exists(pool_guid, 0)) { 2141be6fd75aSMatthew Ahrens error = SET_ERROR(EEXIST); 21421195e687SMark J Musante } else { 2143e9103aaeSGarrett D'Amore spa->spa_config_guid = pool_guid; 21441195e687SMark J Musante 21451195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 21461195e687SMark J Musante &nvl) == 0) { 21471195e687SMark J Musante VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 21481195e687SMark J Musante KM_SLEEP) == 0); 21491195e687SMark J Musante } 21501195e687SMark J Musante 2151ad135b5dSChristopher Siden nvlist_free(spa->spa_load_info); 2152ad135b5dSChristopher Siden spa->spa_load_info = fnvlist_alloc(); 2153ad135b5dSChristopher Siden 215411027bc7STim Haley gethrestime(&spa->spa_loaded_ts); 21551195e687SMark J Musante error = spa_load_impl(spa, pool_guid, config, state, type, 21561195e687SMark J Musante mosconfig, &ereport); 21571195e687SMark J Musante } 21581195e687SMark J Musante 2159bc9014e6SJustin Gibbs /* 2160bc9014e6SJustin Gibbs * Don't count references from objsets that are already closed 2161bc9014e6SJustin Gibbs * and are making their way through the eviction process. 2162bc9014e6SJustin Gibbs */ 2163bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 21641195e687SMark J Musante spa->spa_minref = refcount_count(&spa->spa_refcount); 216511027bc7STim Haley if (error) { 216611027bc7STim Haley if (error != EEXIST) { 216711027bc7STim Haley spa->spa_loaded_ts.tv_sec = 0; 216811027bc7STim Haley spa->spa_loaded_ts.tv_nsec = 0; 216911027bc7STim Haley } 217011027bc7STim Haley if (error != EBADF) { 217111027bc7STim Haley zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 217211027bc7STim Haley } 217311027bc7STim Haley } 21741195e687SMark J Musante spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 21751195e687SMark J Musante spa->spa_ena = 0; 21761195e687SMark J Musante 21771195e687SMark J Musante return (error); 21781195e687SMark J Musante } 21791195e687SMark J Musante 2180215198a6SJoe Stein /* 2181215198a6SJoe Stein * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2182215198a6SJoe Stein * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2183215198a6SJoe Stein * spa's per-vdev ZAP list. 2184215198a6SJoe Stein */ 2185215198a6SJoe Stein static uint64_t 2186215198a6SJoe Stein vdev_count_verify_zaps(vdev_t *vd) 2187215198a6SJoe Stein { 2188215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2189215198a6SJoe Stein uint64_t total = 0; 2190215198a6SJoe Stein if (vd->vdev_top_zap != 0) { 2191215198a6SJoe Stein total++; 2192215198a6SJoe Stein ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2193215198a6SJoe Stein spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2194215198a6SJoe Stein } 2195215198a6SJoe Stein if (vd->vdev_leaf_zap != 0) { 2196215198a6SJoe Stein total++; 2197215198a6SJoe Stein ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2198215198a6SJoe Stein spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2199215198a6SJoe Stein } 2200215198a6SJoe Stein 2201215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2202215198a6SJoe Stein total += vdev_count_verify_zaps(vd->vdev_child[i]); 2203215198a6SJoe Stein } 2204215198a6SJoe Stein 2205215198a6SJoe Stein return (total); 2206215198a6SJoe Stein } 2207215198a6SJoe Stein 2208fa9e4066Sahrens /* 2209fa9e4066Sahrens * Load an existing storage pool, using the pool's builtin spa_config as a 2210ea8dc4b6Seschrock * source of configuration information. 2211fa9e4066Sahrens */ 2212fa9e4066Sahrens static int 22131195e687SMark J Musante spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 22141195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 22151195e687SMark J Musante char **ereport) 2216fa9e4066Sahrens { 2217fa9e4066Sahrens int error = 0; 2218871a9500SMark J Musante nvlist_t *nvroot = NULL; 2219ad135b5dSChristopher Siden nvlist_t *label; 2220fa9e4066Sahrens vdev_t *rvd; 2221fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 22224b964adaSGeorge Wilson uint64_t children, config_cache_txg = spa->spa_config_txg; 22238ad4d6ddSJeff Bonwick int orig_mode = spa->spa_mode; 22241195e687SMark J Musante int parse; 2225cde58dbcSMatthew Ahrens uint64_t obj; 2226ad135b5dSChristopher Siden boolean_t missing_feat_write = B_FALSE; 2227fa9e4066Sahrens 22288ad4d6ddSJeff Bonwick /* 22298ad4d6ddSJeff Bonwick * If this is an untrusted config, access the pool in read-only mode. 22308ad4d6ddSJeff Bonwick * This prevents things like resilvering recently removed devices. 22318ad4d6ddSJeff Bonwick */ 22328ad4d6ddSJeff Bonwick if (!mosconfig) 22338ad4d6ddSJeff Bonwick spa->spa_mode = FREAD; 22348ad4d6ddSJeff Bonwick 2235e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2236e14bb325SJeff Bonwick 2237ea8dc4b6Seschrock spa->spa_load_state = state; 22380373e76bSbonwick 22391195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2240be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 2241fa9e4066Sahrens 22421195e687SMark J Musante parse = (type == SPA_IMPORT_EXISTING ? 22431195e687SMark J Musante VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2244b5989ec7Seschrock 224554d692b7SGeorge Wilson /* 224654d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 224754d692b7SGeorge Wilson */ 22486f834bc1SMatthew Ahrens spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 22496f834bc1SMatthew Ahrens KM_SLEEP); 22506f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) { 22516f834bc1SMatthew Ahrens spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 22526f834bc1SMatthew Ahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 22536f834bc1SMatthew Ahrens ZIO_FLAG_GODFATHER); 22546f834bc1SMatthew Ahrens } 225554d692b7SGeorge Wilson 2256fa9e4066Sahrens /* 225799653d4eSeschrock * Parse the configuration into a vdev tree. We explicitly set the 225899653d4eSeschrock * value that will be returned by spa_version() since parsing the 225999653d4eSeschrock * configuration requires knowing the version number. 2260fa9e4066Sahrens */ 2261e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 22621195e687SMark J Musante error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2263e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2264fa9e4066Sahrens 226599653d4eSeschrock if (error != 0) 22661195e687SMark J Musante return (error); 2267fa9e4066Sahrens 22680e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == rvd); 226981cd5c55SMatthew Ahrens ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 227081cd5c55SMatthew Ahrens ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 22711195e687SMark J Musante 22721195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 22731195e687SMark J Musante ASSERT(spa_guid(spa) == pool_guid); 22741195e687SMark J Musante } 2275fa9e4066Sahrens 2276fa9e4066Sahrens /* 2277fa9e4066Sahrens * Try to open all vdevs, loading each label in the process. 2278fa9e4066Sahrens */ 2279e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 22800bf246f5Smc error = vdev_open(rvd); 2281e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 22820bf246f5Smc if (error != 0) 22831195e687SMark J Musante return (error); 2284fa9e4066Sahrens 2285560e6e96Seschrock /* 228677e3a39cSMark J Musante * We need to validate the vdev labels against the configuration that 228777e3a39cSMark J Musante * we have in hand, which is dependent on the setting of mosconfig. If 228877e3a39cSMark J Musante * mosconfig is true then we're validating the vdev labels based on 22891195e687SMark J Musante * that config. Otherwise, we're validating against the cached config 229077e3a39cSMark J Musante * (zpool.cache) that was read when we loaded the zfs module, and then 229177e3a39cSMark J Musante * later we will recursively call spa_load() and validate against 229277e3a39cSMark J Musante * the vdev config. 22931195e687SMark J Musante * 22941195e687SMark J Musante * If we're assembling a new pool that's been split off from an 22951195e687SMark J Musante * existing pool, the labels haven't yet been updated so we skip 22961195e687SMark J Musante * validation for now. 2297560e6e96Seschrock */ 22981195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 22991195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2300d7f601efSGeorge Wilson error = vdev_validate(rvd, mosconfig); 23011195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 2302560e6e96Seschrock 23031195e687SMark J Musante if (error != 0) 23041195e687SMark J Musante return (error); 23051195e687SMark J Musante 23061195e687SMark J Musante if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2307be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 2308560e6e96Seschrock } 2309560e6e96Seschrock 2310fa9e4066Sahrens /* 2311fa9e4066Sahrens * Find the best uberblock. 2312fa9e4066Sahrens */ 2313ad135b5dSChristopher Siden vdev_uberblock_load(rvd, ub, &label); 2314fa9e4066Sahrens 2315fa9e4066Sahrens /* 2316fa9e4066Sahrens * If we weren't able to find a single valid uberblock, return failure. 2317fa9e4066Sahrens */ 2318ad135b5dSChristopher Siden if (ub->ub_txg == 0) { 2319ad135b5dSChristopher Siden nvlist_free(label); 23201195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2321ad135b5dSChristopher Siden } 2322ea8dc4b6Seschrock 2323ea8dc4b6Seschrock /* 2324ad135b5dSChristopher Siden * If the pool has an unsupported version we can't open it. 2325ea8dc4b6Seschrock */ 2326ad135b5dSChristopher Siden if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2327ad135b5dSChristopher Siden nvlist_free(label); 23281195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2329ad135b5dSChristopher Siden } 2330ad135b5dSChristopher Siden 2331ad135b5dSChristopher Siden if (ub->ub_version >= SPA_VERSION_FEATURES) { 2332ad135b5dSChristopher Siden nvlist_t *features; 2333ad135b5dSChristopher Siden 2334ad135b5dSChristopher Siden /* 2335ad135b5dSChristopher Siden * If we weren't able to find what's necessary for reading the 2336ad135b5dSChristopher Siden * MOS in the label, return failure. 2337ad135b5dSChristopher Siden */ 2338ad135b5dSChristopher Siden if (label == NULL || nvlist_lookup_nvlist(label, 2339ad135b5dSChristopher Siden ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2340ad135b5dSChristopher Siden nvlist_free(label); 2341ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2342ad135b5dSChristopher Siden ENXIO)); 2343ad135b5dSChristopher Siden } 2344ad135b5dSChristopher Siden 2345ad135b5dSChristopher Siden /* 2346ad135b5dSChristopher Siden * Update our in-core representation with the definitive values 2347ad135b5dSChristopher Siden * from the label. 2348ad135b5dSChristopher Siden */ 2349ad135b5dSChristopher Siden nvlist_free(spa->spa_label_features); 2350ad135b5dSChristopher Siden VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2351ad135b5dSChristopher Siden } 2352ad135b5dSChristopher Siden 2353ad135b5dSChristopher Siden nvlist_free(label); 2354ad135b5dSChristopher Siden 2355ad135b5dSChristopher Siden /* 2356ad135b5dSChristopher Siden * Look through entries in the label nvlist's features_for_read. If 2357ad135b5dSChristopher Siden * there is a feature listed there which we don't understand then we 2358ad135b5dSChristopher Siden * cannot open a pool. 2359ad135b5dSChristopher Siden */ 2360ad135b5dSChristopher Siden if (ub->ub_version >= SPA_VERSION_FEATURES) { 2361ad135b5dSChristopher Siden nvlist_t *unsup_feat; 2362ad135b5dSChristopher Siden 2363ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2364ad135b5dSChristopher Siden 0); 2365ad135b5dSChristopher Siden 2366ad135b5dSChristopher Siden for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2367ad135b5dSChristopher Siden NULL); nvp != NULL; 2368ad135b5dSChristopher Siden nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2369ad135b5dSChristopher Siden if (!zfeature_is_supported(nvpair_name(nvp))) { 2370ad135b5dSChristopher Siden VERIFY(nvlist_add_string(unsup_feat, 2371ad135b5dSChristopher Siden nvpair_name(nvp), "") == 0); 2372ad135b5dSChristopher Siden } 2373ad135b5dSChristopher Siden } 2374ad135b5dSChristopher Siden 2375ad135b5dSChristopher Siden if (!nvlist_empty(unsup_feat)) { 2376ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2377ad135b5dSChristopher Siden ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2378ad135b5dSChristopher Siden nvlist_free(unsup_feat); 2379ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2380ad135b5dSChristopher Siden ENOTSUP)); 2381ad135b5dSChristopher Siden } 2382ad135b5dSChristopher Siden 2383ad135b5dSChristopher Siden nvlist_free(unsup_feat); 2384ad135b5dSChristopher Siden } 2385fa9e4066Sahrens 2386fa9e4066Sahrens /* 2387fa9e4066Sahrens * If the vdev guid sum doesn't match the uberblock, we have an 23884b964adaSGeorge Wilson * incomplete configuration. We first check to see if the pool 23894b964adaSGeorge Wilson * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 23904b964adaSGeorge Wilson * If it is, defer the vdev_guid_sum check till later so we 23914b964adaSGeorge Wilson * can handle missing vdevs. 2392fa9e4066Sahrens */ 23934b964adaSGeorge Wilson if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 23944b964adaSGeorge Wilson &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 23951195e687SMark J Musante rvd->vdev_guid_sum != ub->ub_guid_sum) 23961195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 23971195e687SMark J Musante 23981195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 23991195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 24001195e687SMark J Musante spa_try_repair(spa, config); 24011195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 24021195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 24031195e687SMark J Musante spa->spa_config_splitting = NULL; 2404fa9e4066Sahrens } 2405fa9e4066Sahrens 2406fa9e4066Sahrens /* 2407fa9e4066Sahrens * Initialize internal SPA structures. 2408fa9e4066Sahrens */ 2409fa9e4066Sahrens spa->spa_state = POOL_STATE_ACTIVE; 2410fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 2411468c413aSTim Haley spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2412c8ee1847SVictor Latushkin TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2413468c413aSTim Haley spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2414468c413aSTim Haley spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2415b24ab676SJeff Bonwick spa->spa_claim_max_txg = spa->spa_first_txg; 24163f9d6ad7SLin Ling spa->spa_prev_software_version = ub->ub_software_version; 2417b24ab676SJeff Bonwick 2418ad135b5dSChristopher Siden error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 24191195e687SMark J Musante if (error) 24201195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421fa9e4066Sahrens spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2422fa9e4066Sahrens 24231195e687SMark J Musante if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 24241195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2425fa9e4066Sahrens 2426ad135b5dSChristopher Siden if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2427ad135b5dSChristopher Siden boolean_t missing_feat_read = B_FALSE; 242857221772SChristopher Siden nvlist_t *unsup_feat, *enabled_feat; 2429ad135b5dSChristopher Siden 2430ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2431ad135b5dSChristopher Siden &spa->spa_feat_for_read_obj) != 0) { 2432ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2433ad135b5dSChristopher Siden } 2434ad135b5dSChristopher Siden 2435ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2436ad135b5dSChristopher Siden &spa->spa_feat_for_write_obj) != 0) { 2437ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2438ad135b5dSChristopher Siden } 2439ad135b5dSChristopher Siden 2440ad135b5dSChristopher Siden if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2441ad135b5dSChristopher Siden &spa->spa_feat_desc_obj) != 0) { 2442ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2443ad135b5dSChristopher Siden } 2444ad135b5dSChristopher Siden 244557221772SChristopher Siden enabled_feat = fnvlist_alloc(); 244657221772SChristopher Siden unsup_feat = fnvlist_alloc(); 2447ad135b5dSChristopher Siden 24482acef22dSMatthew Ahrens if (!spa_features_check(spa, B_FALSE, 244957221772SChristopher Siden unsup_feat, enabled_feat)) 2450ad135b5dSChristopher Siden missing_feat_read = B_TRUE; 2451ad135b5dSChristopher Siden 2452ad135b5dSChristopher Siden if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 24532acef22dSMatthew Ahrens if (!spa_features_check(spa, B_TRUE, 245457221772SChristopher Siden unsup_feat, enabled_feat)) { 2455ad135b5dSChristopher Siden missing_feat_write = B_TRUE; 245657221772SChristopher Siden } 2457ad135b5dSChristopher Siden } 2458ad135b5dSChristopher Siden 245957221772SChristopher Siden fnvlist_add_nvlist(spa->spa_load_info, 246057221772SChristopher Siden ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 246157221772SChristopher Siden 2462ad135b5dSChristopher Siden if (!nvlist_empty(unsup_feat)) { 246357221772SChristopher Siden fnvlist_add_nvlist(spa->spa_load_info, 246457221772SChristopher Siden ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2465ad135b5dSChristopher Siden } 2466ad135b5dSChristopher Siden 246757221772SChristopher Siden fnvlist_free(enabled_feat); 246857221772SChristopher Siden fnvlist_free(unsup_feat); 2469ad135b5dSChristopher Siden 2470ad135b5dSChristopher Siden if (!missing_feat_read) { 2471ad135b5dSChristopher Siden fnvlist_add_boolean(spa->spa_load_info, 2472ad135b5dSChristopher Siden ZPOOL_CONFIG_CAN_RDONLY); 2473ad135b5dSChristopher Siden } 2474ad135b5dSChristopher Siden 2475ad135b5dSChristopher Siden /* 2476ad135b5dSChristopher Siden * If the state is SPA_LOAD_TRYIMPORT, our objective is 2477ad135b5dSChristopher Siden * twofold: to determine whether the pool is available for 2478ad135b5dSChristopher Siden * import in read-write mode and (if it is not) whether the 2479ad135b5dSChristopher Siden * pool is available for import in read-only mode. If the pool 2480ad135b5dSChristopher Siden * is available for import in read-write mode, it is displayed 2481ad135b5dSChristopher Siden * as available in userland; if it is not available for import 2482ad135b5dSChristopher Siden * in read-only mode, it is displayed as unavailable in 2483ad135b5dSChristopher Siden * userland. If the pool is available for import in read-only 2484ad135b5dSChristopher Siden * mode but not read-write mode, it is displayed as unavailable 2485ad135b5dSChristopher Siden * in userland with a special note that the pool is actually 2486ad135b5dSChristopher Siden * available for open in read-only mode. 2487ad135b5dSChristopher Siden * 2488ad135b5dSChristopher Siden * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2489ad135b5dSChristopher Siden * missing a feature for write, we must first determine whether 2490ad135b5dSChristopher Siden * the pool can be opened read-only before returning to 2491ad135b5dSChristopher Siden * userland in order to know whether to display the 2492ad135b5dSChristopher Siden * abovementioned note. 2493ad135b5dSChristopher Siden */ 2494ad135b5dSChristopher Siden if (missing_feat_read || (missing_feat_write && 2495ad135b5dSChristopher Siden spa_writeable(spa))) { 2496ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2497ad135b5dSChristopher Siden ENOTSUP)); 2498ad135b5dSChristopher Siden } 249943466aaeSMax Grossman 250043466aaeSMax Grossman /* 250143466aaeSMax Grossman * Load refcounts for ZFS features from disk into an in-memory 250243466aaeSMax Grossman * cache during SPA initialization. 250343466aaeSMax Grossman */ 250443466aaeSMax Grossman for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 250543466aaeSMax Grossman uint64_t refcount; 250643466aaeSMax Grossman 250743466aaeSMax Grossman error = feature_get_refcount_from_disk(spa, 250843466aaeSMax Grossman &spa_feature_table[i], &refcount); 250943466aaeSMax Grossman if (error == 0) { 251043466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = refcount; 251143466aaeSMax Grossman } else if (error == ENOTSUP) { 251243466aaeSMax Grossman spa->spa_feat_refcount_cache[i] = 251343466aaeSMax Grossman SPA_FEATURE_DISABLED; 251443466aaeSMax Grossman } else { 251543466aaeSMax Grossman return (spa_vdev_err(rvd, 251643466aaeSMax Grossman VDEV_AUX_CORRUPT_DATA, EIO)); 251743466aaeSMax Grossman } 251843466aaeSMax Grossman } 251943466aaeSMax Grossman } 252043466aaeSMax Grossman 252143466aaeSMax Grossman if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 252243466aaeSMax Grossman if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 25235d7b4d43SMatthew Ahrens &spa->spa_feat_enabled_txg_obj) != 0) 252443466aaeSMax Grossman return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2525ad135b5dSChristopher Siden } 2526ad135b5dSChristopher Siden 2527ad135b5dSChristopher Siden spa->spa_is_initializing = B_TRUE; 2528ad135b5dSChristopher Siden error = dsl_pool_open(spa->spa_dsl_pool); 2529ad135b5dSChristopher Siden spa->spa_is_initializing = B_FALSE; 2530ad135b5dSChristopher Siden if (error != 0) 2531ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2532ad135b5dSChristopher Siden 2533fa9e4066Sahrens if (!mosconfig) { 253495173954Sek uint64_t hostid; 2535871a9500SMark J Musante nvlist_t *policy = NULL, *nvconfig; 2536871a9500SMark J Musante 2537871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2538871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2539fa9e4066Sahrens 254088ecc943SGeorge Wilson if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 254177650510SLin Ling ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 254295173954Sek char *hostname; 254395173954Sek unsigned long myhostid = 0; 254495173954Sek 254588ecc943SGeorge Wilson VERIFY(nvlist_lookup_string(nvconfig, 254695173954Sek ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 254795173954Sek 25485679c89fSjv #ifdef _KERNEL 25495679c89fSjv myhostid = zone_get_hostid(NULL); 25505679c89fSjv #else /* _KERNEL */ 25515679c89fSjv /* 25525679c89fSjv * We're emulating the system's hostid in userland, so 25535679c89fSjv * we can't use zone_get_hostid(). 25545679c89fSjv */ 255595173954Sek (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 25565679c89fSjv #endif /* _KERNEL */ 255717194a52Slling if (hostid != 0 && myhostid != 0 && 25585679c89fSjv hostid != myhostid) { 2559871a9500SMark J Musante nvlist_free(nvconfig); 256095173954Sek cmn_err(CE_WARN, "pool '%s' could not be " 256195173954Sek "loaded as it was last accessed by " 256277650510SLin Ling "another system (host: %s hostid: 0x%lx). " 2563654b400cSJoshua M. Clulow "See: http://illumos.org/msg/ZFS-8000-EY", 2564e14bb325SJeff Bonwick spa_name(spa), hostname, 256595173954Sek (unsigned long)hostid); 2566be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 256795173954Sek } 256895173954Sek } 2569c8ee1847SVictor Latushkin if (nvlist_lookup_nvlist(spa->spa_config, 2570c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, &policy) == 0) 2571c8ee1847SVictor Latushkin VERIFY(nvlist_add_nvlist(nvconfig, 2572c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, policy) == 0); 257395173954Sek 257488ecc943SGeorge Wilson spa_config_set(spa, nvconfig); 2575fa9e4066Sahrens spa_unload(spa); 2576fa9e4066Sahrens spa_deactivate(spa); 25778ad4d6ddSJeff Bonwick spa_activate(spa, orig_mode); 2578fa9e4066Sahrens 25791195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2580fa9e4066Sahrens } 2581fa9e4066Sahrens 258245818ee1SMatthew Ahrens /* Grab the secret checksum salt from the MOS. */ 258345818ee1SMatthew Ahrens error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 258445818ee1SMatthew Ahrens DMU_POOL_CHECKSUM_SALT, 1, 258545818ee1SMatthew Ahrens sizeof (spa->spa_cksum_salt.zcs_bytes), 258645818ee1SMatthew Ahrens spa->spa_cksum_salt.zcs_bytes); 258745818ee1SMatthew Ahrens if (error == ENOENT) { 258845818ee1SMatthew Ahrens /* Generate a new salt for subsequent use */ 258945818ee1SMatthew Ahrens (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 259045818ee1SMatthew Ahrens sizeof (spa->spa_cksum_salt.zcs_bytes)); 259145818ee1SMatthew Ahrens } else if (error != 0) { 259245818ee1SMatthew Ahrens return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 259345818ee1SMatthew Ahrens } 259445818ee1SMatthew Ahrens 2595cde58dbcSMatthew Ahrens if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2596cde58dbcSMatthew Ahrens return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2597cde58dbcSMatthew Ahrens error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2598cde58dbcSMatthew Ahrens if (error != 0) 25991195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2600fa9e4066Sahrens 260199653d4eSeschrock /* 260299653d4eSeschrock * Load the bit that tells us to use the new accounting function 260399653d4eSeschrock * (raid-z deflation). If we have an older pool, this will not 260499653d4eSeschrock * be present. 260599653d4eSeschrock */ 26061195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 26071195e687SMark J Musante if (error != 0 && error != ENOENT) 26081195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 260999653d4eSeschrock 26103f9d6ad7SLin Ling error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 26113f9d6ad7SLin Ling &spa->spa_creation_version); 26123f9d6ad7SLin Ling if (error != 0 && error != ENOENT) 26133f9d6ad7SLin Ling return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 26143f9d6ad7SLin Ling 2615fa9e4066Sahrens /* 2616ea8dc4b6Seschrock * Load the persistent error log. If we have an older pool, this will 2617ea8dc4b6Seschrock * not be present. 2618fa9e4066Sahrens */ 26191195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 26201195e687SMark J Musante if (error != 0 && error != ENOENT) 26211195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2622ea8dc4b6Seschrock 26231195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 26241195e687SMark J Musante &spa->spa_errlog_scrub); 26251195e687SMark J Musante if (error != 0 && error != ENOENT) 26261195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2627ea8dc4b6Seschrock 262806eeb2adSek /* 262906eeb2adSek * Load the history object. If we have an older pool, this 263006eeb2adSek * will not be present. 263106eeb2adSek */ 26321195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 26331195e687SMark J Musante if (error != 0 && error != ENOENT) 26341195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 26351195e687SMark J Musante 2636215198a6SJoe Stein /* 2637215198a6SJoe Stein * Load the per-vdev ZAP map. If we have an older pool, this will not 2638215198a6SJoe Stein * be present; in this case, defer its creation to a later time to 2639215198a6SJoe Stein * avoid dirtying the MOS this early / out of sync context. See 2640215198a6SJoe Stein * spa_sync_config_object. 2641215198a6SJoe Stein */ 2642215198a6SJoe Stein 2643215198a6SJoe Stein /* The sentinel is only available in the MOS config. */ 2644215198a6SJoe Stein nvlist_t *mos_config; 2645215198a6SJoe Stein if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) 2646215198a6SJoe Stein return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2647215198a6SJoe Stein 2648215198a6SJoe Stein error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 2649215198a6SJoe Stein &spa->spa_all_vdev_zaps); 2650215198a6SJoe Stein 2651215198a6SJoe Stein if (error != ENOENT && error != 0) { 2652215198a6SJoe Stein return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2653215198a6SJoe Stein } else if (error == 0 && !nvlist_exists(mos_config, 2654215198a6SJoe Stein ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 2655215198a6SJoe Stein /* 2656215198a6SJoe Stein * An older version of ZFS overwrote the sentinel value, so 2657215198a6SJoe Stein * we have orphaned per-vdev ZAPs in the MOS. Defer their 2658215198a6SJoe Stein * destruction to later; see spa_sync_config_object. 2659215198a6SJoe Stein */ 2660215198a6SJoe Stein spa->spa_avz_action = AVZ_ACTION_DESTROY; 2661215198a6SJoe Stein /* 2662215198a6SJoe Stein * We're assuming that no vdevs have had their ZAPs created 2663215198a6SJoe Stein * before this. Better be sure of it. 2664215198a6SJoe Stein */ 2665215198a6SJoe Stein ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 2666215198a6SJoe Stein } 2667215198a6SJoe Stein nvlist_free(mos_config); 2668215198a6SJoe Stein 26691195e687SMark J Musante /* 26701195e687SMark J Musante * If we're assembling the pool from the split-off vdevs of 26711195e687SMark J Musante * an existing pool, we don't want to attach the spares & cache 26721195e687SMark J Musante * devices. 26731195e687SMark J Musante */ 267406eeb2adSek 267599653d4eSeschrock /* 267699653d4eSeschrock * Load any hot spares for this pool. 267799653d4eSeschrock */ 26781195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 26791195e687SMark J Musante if (error != 0 && error != ENOENT) 26801195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 26811195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2682e7437265Sahrens ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2683fa94a07fSbrendan if (load_nvlist(spa, spa->spa_spares.sav_object, 26841195e687SMark J Musante &spa->spa_spares.sav_config) != 0) 26851195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 268699653d4eSeschrock 2687e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 268899653d4eSeschrock spa_load_spares(spa); 2689e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 26901195e687SMark J Musante } else if (error == 0) { 26911195e687SMark J Musante spa->spa_spares.sav_sync = B_TRUE; 269299653d4eSeschrock } 269399653d4eSeschrock 2694fa94a07fSbrendan /* 2695fa94a07fSbrendan * Load any level 2 ARC devices for this pool. 2696fa94a07fSbrendan */ 26971195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2698fa94a07fSbrendan &spa->spa_l2cache.sav_object); 26991195e687SMark J Musante if (error != 0 && error != ENOENT) 27001195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 27011195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2702fa94a07fSbrendan ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2703fa94a07fSbrendan if (load_nvlist(spa, spa->spa_l2cache.sav_object, 27041195e687SMark J Musante &spa->spa_l2cache.sav_config) != 0) 27051195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2706fa94a07fSbrendan 2707e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2708fa94a07fSbrendan spa_load_l2cache(spa); 2709e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 27101195e687SMark J Musante } else if (error == 0) { 27111195e687SMark J Musante spa->spa_l2cache.sav_sync = B_TRUE; 2712fa94a07fSbrendan } 2713fa94a07fSbrendan 2714990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2715ecd6cf80Smarks 27161195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 27171195e687SMark J Musante if (error && error != ENOENT) 27181195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2719b1b8ab34Slling 2720b1b8ab34Slling if (error == 0) { 27211195e687SMark J Musante uint64_t autoreplace; 27221195e687SMark J Musante 27231195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 27241195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 27251195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 27261195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 27271195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 27281195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 27291195e687SMark J Musante &spa->spa_dedup_ditto); 27301195e687SMark J Musante 2731b693757aSEric Schrock spa->spa_autoreplace = (autoreplace != 0); 2732b1b8ab34Slling } 2733b1b8ab34Slling 27343d7072f8Seschrock /* 27353d7072f8Seschrock * If the 'autoreplace' property is set, then post a resource notifying 27363d7072f8Seschrock * the ZFS DE that it should not issue any faults for unopenable 27373d7072f8Seschrock * devices. We also iterate over the vdevs, and post a sysevent for any 27383d7072f8Seschrock * unopenable vdevs so that the normal autoreplace handler can take 27393d7072f8Seschrock * over. 27403d7072f8Seschrock */ 2741b693757aSEric Schrock if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 27423d7072f8Seschrock spa_check_removed(spa->spa_root_vdev); 2743b693757aSEric Schrock /* 2744b693757aSEric Schrock * For the import case, this is done in spa_import(), because 2745b693757aSEric Schrock * at this point we're using the spare definitions from 2746b693757aSEric Schrock * the MOS config, not necessarily from the userland config. 2747b693757aSEric Schrock */ 2748b693757aSEric Schrock if (state != SPA_LOAD_IMPORT) { 2749b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 2750b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 2751b693757aSEric Schrock } 2752b693757aSEric Schrock } 27533d7072f8Seschrock 2754ea8dc4b6Seschrock /* 2755560e6e96Seschrock * Load the vdev state for all toplevel vdevs. 2756ea8dc4b6Seschrock */ 2757560e6e96Seschrock vdev_load(rvd); 27580373e76bSbonwick 2759fa9e4066Sahrens /* 2760fa9e4066Sahrens * Propagate the leaf DTLs we just loaded all the way up the tree. 2761fa9e4066Sahrens */ 2762e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2763fa9e4066Sahrens vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2764e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2765fa9e4066Sahrens 2766b24ab676SJeff Bonwick /* 2767b24ab676SJeff Bonwick * Load the DDTs (dedup tables). 2768b24ab676SJeff Bonwick */ 2769b24ab676SJeff Bonwick error = ddt_load(spa); 27701195e687SMark J Musante if (error != 0) 27711195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2772b24ab676SJeff Bonwick 2773485bbbf5SGeorge Wilson spa_update_dspace(spa); 2774485bbbf5SGeorge Wilson 2775b24ab676SJeff Bonwick /* 27764b964adaSGeorge Wilson * Validate the config, using the MOS config to fill in any 27774b964adaSGeorge Wilson * information which might be missing. If we fail to validate 27784b964adaSGeorge Wilson * the config then declare the pool unfit for use. If we're 27794b964adaSGeorge Wilson * assembling a pool from a split, the log is not transferred 27804b964adaSGeorge Wilson * over. 2781b24ab676SJeff Bonwick */ 27821195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 2783871a9500SMark J Musante nvlist_t *nvconfig; 2784871a9500SMark J Musante 2785871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2786871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2787871a9500SMark J Musante 27884b964adaSGeorge Wilson if (!spa_config_valid(spa, nvconfig)) { 27894b964adaSGeorge Wilson nvlist_free(nvconfig); 27904b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 27914b964adaSGeorge Wilson ENXIO)); 27924b964adaSGeorge Wilson } 27931195e687SMark J Musante nvlist_free(nvconfig); 27941195e687SMark J Musante 27954b964adaSGeorge Wilson /* 2796ad135b5dSChristopher Siden * Now that we've validated the config, check the state of the 27974b964adaSGeorge Wilson * root vdev. If it can't be opened, it indicates one or 27984b964adaSGeorge Wilson * more toplevel vdevs are faulted. 27994b964adaSGeorge Wilson */ 28004b964adaSGeorge Wilson if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2801be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 28024b964adaSGeorge Wilson 280323367a2fSMatthew Ahrens if (spa_writeable(spa) && spa_check_logs(spa)) { 28041195e687SMark J Musante *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 28051195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 28061195e687SMark J Musante } 2807b24ab676SJeff Bonwick } 2808b24ab676SJeff Bonwick 2809ad135b5dSChristopher Siden if (missing_feat_write) { 2810ad135b5dSChristopher Siden ASSERT(state == SPA_LOAD_TRYIMPORT); 2811ad135b5dSChristopher Siden 2812ad135b5dSChristopher Siden /* 2813ad135b5dSChristopher Siden * At this point, we know that we can open the pool in 2814ad135b5dSChristopher Siden * read-only mode but not read-write mode. We now have enough 2815ad135b5dSChristopher Siden * information and can return to userland. 2816ad135b5dSChristopher Siden */ 2817ad135b5dSChristopher Siden return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2818ad135b5dSChristopher Siden } 2819ad135b5dSChristopher Siden 28204b964adaSGeorge Wilson /* 28214b964adaSGeorge Wilson * We've successfully opened the pool, verify that we're ready 28224b964adaSGeorge Wilson * to start pushing transactions. 28234b964adaSGeorge Wilson */ 28244b964adaSGeorge Wilson if (state != SPA_LOAD_TRYIMPORT) { 28254b964adaSGeorge Wilson if (error = spa_load_verify(spa)) 28264b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 28274b964adaSGeorge Wilson error)); 28284b964adaSGeorge Wilson } 28294b964adaSGeorge Wilson 2830468c413aSTim Haley if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2831468c413aSTim Haley spa->spa_load_max_txg == UINT64_MAX)) { 28325dabedeeSbonwick dmu_tx_t *tx; 28330373e76bSbonwick int need_update = B_FALSE; 283412380e1eSArne Jansen dsl_pool_t *dp = spa_get_dsl(spa); 28358ad4d6ddSJeff Bonwick 28368ad4d6ddSJeff Bonwick ASSERT(state != SPA_LOAD_TRYIMPORT); 28375dabedeeSbonwick 28380373e76bSbonwick /* 28390373e76bSbonwick * Claim log blocks that haven't been committed yet. 28400373e76bSbonwick * This must all happen in a single txg. 2841b24ab676SJeff Bonwick * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2842b24ab676SJeff Bonwick * invoked from zil_claim_log_block()'s i/o done callback. 2843468c413aSTim Haley * Price of rollback is that we abandon the log. 28440373e76bSbonwick */ 2845b24ab676SJeff Bonwick spa->spa_claiming = B_TRUE; 2846b24ab676SJeff Bonwick 284712380e1eSArne Jansen tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 284812380e1eSArne Jansen (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 28490b69c2f0Sahrens zil_claim, tx, DS_FIND_CHILDREN); 2850fa9e4066Sahrens dmu_tx_commit(tx); 2851fa9e4066Sahrens 2852b24ab676SJeff Bonwick spa->spa_claiming = B_FALSE; 2853b24ab676SJeff Bonwick 28541195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_GOOD); 2855fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 2856fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 2857fa9e4066Sahrens 2858fa9e4066Sahrens /* 2859b24ab676SJeff Bonwick * Wait for all claims to sync. We sync up to the highest 2860b24ab676SJeff Bonwick * claimed log block birth time so that claimed log blocks 2861b24ab676SJeff Bonwick * don't appear to be from the future. spa_claim_max_txg 2862b24ab676SJeff Bonwick * will have been set for us by either zil_check_log_chain() 2863b24ab676SJeff Bonwick * (invoked from spa_check_logs()) or zil_claim() above. 2864fa9e4066Sahrens */ 2865b24ab676SJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 28660e34b6a7Sbonwick 28670e34b6a7Sbonwick /* 28680373e76bSbonwick * If the config cache is stale, or we have uninitialized 28690373e76bSbonwick * metaslabs (see spa_vdev_add()), then update the config. 2870bc758434SLin Ling * 28714b964adaSGeorge Wilson * If this is a verbatim import, trust the current 2872bc758434SLin Ling * in-core spa_config and update the disk labels. 28730e34b6a7Sbonwick */ 28740373e76bSbonwick if (config_cache_txg != spa->spa_config_txg || 28754b964adaSGeorge Wilson state == SPA_LOAD_IMPORT || 28764b964adaSGeorge Wilson state == SPA_LOAD_RECOVER || 28774b964adaSGeorge Wilson (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 28780373e76bSbonwick need_update = B_TRUE; 28790373e76bSbonwick 28808ad4d6ddSJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) 28810373e76bSbonwick if (rvd->vdev_child[c]->vdev_ms_array == 0) 28820373e76bSbonwick need_update = B_TRUE; 28830e34b6a7Sbonwick 28840e34b6a7Sbonwick /* 28850373e76bSbonwick * Update the config cache asychronously in case we're the 28860373e76bSbonwick * root pool, in which case the config cache isn't writable yet. 28870e34b6a7Sbonwick */ 28880373e76bSbonwick if (need_update) 28890373e76bSbonwick spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 28908ad4d6ddSJeff Bonwick 28918ad4d6ddSJeff Bonwick /* 28928ad4d6ddSJeff Bonwick * Check all DTLs to see if anything needs resilvering. 28938ad4d6ddSJeff Bonwick */ 28943f9d6ad7SLin Ling if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 28953f9d6ad7SLin Ling vdev_resilver_needed(rvd, NULL, NULL)) 28968ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 2897503ad85cSMatthew Ahrens 28984445fffbSMatthew Ahrens /* 28994445fffbSMatthew Ahrens * Log the fact that we booted up (so that we can detect if 29004445fffbSMatthew Ahrens * we rebooted in the middle of an operation). 29014445fffbSMatthew Ahrens */ 29024445fffbSMatthew Ahrens spa_history_log_version(spa, "open"); 29034445fffbSMatthew Ahrens 2904503ad85cSMatthew Ahrens /* 2905503ad85cSMatthew Ahrens * Delete any inconsistent datasets. 2906503ad85cSMatthew Ahrens */ 2907503ad85cSMatthew Ahrens (void) dmu_objset_find(spa_name(spa), 2908503ad85cSMatthew Ahrens dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2909ca45db41SChris Kirby 2910ca45db41SChris Kirby /* 2911ca45db41SChris Kirby * Clean up any stale temporary dataset userrefs. 2912ca45db41SChris Kirby */ 2913ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2914fa9e4066Sahrens } 2915fa9e4066Sahrens 29161195e687SMark J Musante return (0); 2917fa9e4066Sahrens } 2918fa9e4066Sahrens 2919468c413aSTim Haley static int 2920468c413aSTim Haley spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2921468c413aSTim Haley { 2922f9af39baSGeorge Wilson int mode = spa->spa_mode; 2923f9af39baSGeorge Wilson 2924468c413aSTim Haley spa_unload(spa); 2925468c413aSTim Haley spa_deactivate(spa); 2926468c413aSTim Haley 2927e42d2059SMatthew Ahrens spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2928468c413aSTim Haley 2929f9af39baSGeorge Wilson spa_activate(spa, mode); 2930468c413aSTim Haley spa_async_suspend(spa); 2931468c413aSTim Haley 29321195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2933468c413aSTim Haley } 2934468c413aSTim Haley 2935ad135b5dSChristopher Siden /* 2936ad135b5dSChristopher Siden * If spa_load() fails this function will try loading prior txg's. If 2937ad135b5dSChristopher Siden * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2938ad135b5dSChristopher Siden * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2939ad135b5dSChristopher Siden * function will not rewind the pool and will return the same error as 2940ad135b5dSChristopher Siden * spa_load(). 2941ad135b5dSChristopher Siden */ 2942468c413aSTim Haley static int 2943468c413aSTim Haley spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2944c8ee1847SVictor Latushkin uint64_t max_request, int rewind_flags) 2945468c413aSTim Haley { 2946ad135b5dSChristopher Siden nvlist_t *loadinfo = NULL; 2947468c413aSTim Haley nvlist_t *config = NULL; 2948468c413aSTim Haley int load_error, rewind_error; 2949c8ee1847SVictor Latushkin uint64_t safe_rewind_txg; 2950468c413aSTim Haley uint64_t min_txg; 2951468c413aSTim Haley 2952a33cae98STim Haley if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2953468c413aSTim Haley spa->spa_load_max_txg = spa->spa_load_txg; 29541195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2955a33cae98STim Haley } else { 2956468c413aSTim Haley spa->spa_load_max_txg = max_request; 2957e42d2059SMatthew Ahrens if (max_request != UINT64_MAX) 2958e42d2059SMatthew Ahrens spa->spa_extreme_rewind = B_TRUE; 2959a33cae98STim Haley } 2960468c413aSTim Haley 29611195e687SMark J Musante load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 29621195e687SMark J Musante mosconfig); 2963468c413aSTim Haley if (load_error == 0) 2964468c413aSTim Haley return (0); 2965468c413aSTim Haley 2966468c413aSTim Haley if (spa->spa_root_vdev != NULL) 2967468c413aSTim Haley config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2968468c413aSTim Haley 2969468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2970468c413aSTim Haley spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2971468c413aSTim Haley 2972c8ee1847SVictor Latushkin if (rewind_flags & ZPOOL_NEVER_REWIND) { 2973468c413aSTim Haley nvlist_free(config); 2974468c413aSTim Haley return (load_error); 2975468c413aSTim Haley } 2976468c413aSTim Haley 2977ad135b5dSChristopher Siden if (state == SPA_LOAD_RECOVER) { 2978ad135b5dSChristopher Siden /* Price of rolling back is discarding txgs, including log */ 29791195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2980ad135b5dSChristopher Siden } else { 2981ad135b5dSChristopher Siden /* 2982ad135b5dSChristopher Siden * If we aren't rolling back save the load info from our first 2983ad135b5dSChristopher Siden * import attempt so that we can restore it after attempting 2984ad135b5dSChristopher Siden * to rewind. 2985ad135b5dSChristopher Siden */ 2986ad135b5dSChristopher Siden loadinfo = spa->spa_load_info; 2987ad135b5dSChristopher Siden spa->spa_load_info = fnvlist_alloc(); 2988ad135b5dSChristopher Siden } 2989468c413aSTim Haley 2990c8ee1847SVictor Latushkin spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2991c8ee1847SVictor Latushkin safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2992c8ee1847SVictor Latushkin min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2993c8ee1847SVictor Latushkin TXG_INITIAL : safe_rewind_txg; 2994468c413aSTim Haley 2995c8ee1847SVictor Latushkin /* 2996c8ee1847SVictor Latushkin * Continue as long as we're finding errors, we're still within 2997c8ee1847SVictor Latushkin * the acceptable rewind range, and we're still finding uberblocks 2998c8ee1847SVictor Latushkin */ 2999c8ee1847SVictor Latushkin while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 3000c8ee1847SVictor Latushkin spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 3001c8ee1847SVictor Latushkin if (spa->spa_load_max_txg < safe_rewind_txg) 3002468c413aSTim Haley spa->spa_extreme_rewind = B_TRUE; 3003468c413aSTim Haley rewind_error = spa_load_retry(spa, state, mosconfig); 3004468c413aSTim Haley } 3005468c413aSTim Haley 3006468c413aSTim Haley spa->spa_extreme_rewind = B_FALSE; 3007468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 3008468c413aSTim Haley 3009468c413aSTim Haley if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3010468c413aSTim Haley spa_config_set(spa, config); 3011468c413aSTim Haley 3012ad135b5dSChristopher Siden if (state == SPA_LOAD_RECOVER) { 3013ad135b5dSChristopher Siden ASSERT3P(loadinfo, ==, NULL); 3014ad135b5dSChristopher Siden return (rewind_error); 3015ad135b5dSChristopher Siden } else { 3016ad135b5dSChristopher Siden /* Store the rewind info as part of the initial load info */ 3017ad135b5dSChristopher Siden fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3018ad135b5dSChristopher Siden spa->spa_load_info); 3019ad135b5dSChristopher Siden 3020ad135b5dSChristopher Siden /* Restore the initial load info */ 3021ad135b5dSChristopher Siden fnvlist_free(spa->spa_load_info); 3022ad135b5dSChristopher Siden spa->spa_load_info = loadinfo; 3023ad135b5dSChristopher Siden 3024ad135b5dSChristopher Siden return (load_error); 3025ad135b5dSChristopher Siden } 3026468c413aSTim Haley } 3027468c413aSTim Haley 3028fa9e4066Sahrens /* 3029fa9e4066Sahrens * Pool Open/Import 3030fa9e4066Sahrens * 3031fa9e4066Sahrens * The import case is identical to an open except that the configuration is sent 3032fa9e4066Sahrens * down from userland, instead of grabbed from the configuration cache. For the 3033fa9e4066Sahrens * case of an open, the pool configuration will exist in the 30343d7072f8Seschrock * POOL_STATE_UNINITIALIZED state. 3035fa9e4066Sahrens * 3036fa9e4066Sahrens * The stats information (gen/count/ustats) is used to gather vdev statistics at 3037fa9e4066Sahrens * the same time open the pool, without having to keep around the spa_t in some 3038fa9e4066Sahrens * ambiguous state. 3039fa9e4066Sahrens */ 3040fa9e4066Sahrens static int 3041468c413aSTim Haley spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3042468c413aSTim Haley nvlist_t **config) 3043fa9e4066Sahrens { 3044fa9e4066Sahrens spa_t *spa; 30454b964adaSGeorge Wilson spa_load_state_t state = SPA_LOAD_OPEN; 3046fa9e4066Sahrens int error; 3047fa9e4066Sahrens int locked = B_FALSE; 3048fa9e4066Sahrens 3049fa9e4066Sahrens *spapp = NULL; 3050fa9e4066Sahrens 3051fa9e4066Sahrens /* 3052fa9e4066Sahrens * As disgusting as this is, we need to support recursive calls to this 3053fa9e4066Sahrens * function because dsl_dir_open() is called during spa_load(), and ends 3054fa9e4066Sahrens * up calling spa_open() again. The real fix is to figure out how to 3055fa9e4066Sahrens * avoid dsl_dir_open() calling this in the first place. 3056fa9e4066Sahrens */ 3057fa9e4066Sahrens if (mutex_owner(&spa_namespace_lock) != curthread) { 3058fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 3059fa9e4066Sahrens locked = B_TRUE; 3060fa9e4066Sahrens } 3061fa9e4066Sahrens 3062fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 3063fa9e4066Sahrens if (locked) 3064fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3065be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 3066fa9e4066Sahrens } 3067468c413aSTim Haley 3068fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 30694b44c88cSTim Haley zpool_rewind_policy_t policy; 30704b44c88cSTim Haley 30714b44c88cSTim Haley zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 30724b44c88cSTim Haley &policy); 30734b44c88cSTim Haley if (policy.zrp_request & ZPOOL_DO_REWIND) 30744b44c88cSTim Haley state = SPA_LOAD_RECOVER; 3075fa9e4066Sahrens 30768ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 3077fa9e4066Sahrens 3078468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 3079468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3080468c413aSTim Haley 3081468c413aSTim Haley error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3082c8ee1847SVictor Latushkin policy.zrp_request); 3083fa9e4066Sahrens 3084fa9e4066Sahrens if (error == EBADF) { 3085fa9e4066Sahrens /* 3086560e6e96Seschrock * If vdev_validate() returns failure (indicated by 3087560e6e96Seschrock * EBADF), it indicates that one of the vdevs indicates 3088560e6e96Seschrock * that the pool has been exported or destroyed. If 3089560e6e96Seschrock * this is the case, the config cache is out of sync and 3090560e6e96Seschrock * we should remove the pool from the namespace. 3091fa9e4066Sahrens */ 3092fa9e4066Sahrens spa_unload(spa); 3093fa9e4066Sahrens spa_deactivate(spa); 3094c5904d13Seschrock spa_config_sync(spa, B_TRUE, B_TRUE); 3095fa9e4066Sahrens spa_remove(spa); 3096fa9e4066Sahrens if (locked) 3097fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3098be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 3099ea8dc4b6Seschrock } 3100ea8dc4b6Seschrock 3101ea8dc4b6Seschrock if (error) { 3102fa9e4066Sahrens /* 3103fa9e4066Sahrens * We can't open the pool, but we still have useful 3104fa9e4066Sahrens * information: the state of each vdev after the 3105fa9e4066Sahrens * attempted vdev_open(). Return this to the user. 3106fa9e4066Sahrens */ 31074b964adaSGeorge Wilson if (config != NULL && spa->spa_config) { 3108468c413aSTim Haley VERIFY(nvlist_dup(spa->spa_config, config, 3109468c413aSTim Haley KM_SLEEP) == 0); 31104b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, 31114b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_INFO, 31124b964adaSGeorge Wilson spa->spa_load_info) == 0); 31134b964adaSGeorge Wilson } 3114fa9e4066Sahrens spa_unload(spa); 3115fa9e4066Sahrens spa_deactivate(spa); 3116468c413aSTim Haley spa->spa_last_open_failed = error; 3117fa9e4066Sahrens if (locked) 3118fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3119fa9e4066Sahrens *spapp = NULL; 3120fa9e4066Sahrens return (error); 3121fa9e4066Sahrens } 3122fa9e4066Sahrens } 3123fa9e4066Sahrens 3124fa9e4066Sahrens spa_open_ref(spa, tag); 31253d7072f8Seschrock 3126468c413aSTim Haley if (config != NULL) 3127468c413aSTim Haley *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3128468c413aSTim Haley 31294b964adaSGeorge Wilson /* 31304b964adaSGeorge Wilson * If we've recovered the pool, pass back any information we 31314b964adaSGeorge Wilson * gathered while doing the load. 31324b964adaSGeorge Wilson */ 31334b964adaSGeorge Wilson if (state == SPA_LOAD_RECOVER) { 31344b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 31354b964adaSGeorge Wilson spa->spa_load_info) == 0); 31364b964adaSGeorge Wilson } 31374b964adaSGeorge Wilson 3138a33cae98STim Haley if (locked) { 3139a33cae98STim Haley spa->spa_last_open_failed = 0; 3140a33cae98STim Haley spa->spa_last_ubsync_txg = 0; 3141a33cae98STim Haley spa->spa_load_txg = 0; 3142fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3143a33cae98STim Haley } 3144fa9e4066Sahrens 3145fa9e4066Sahrens *spapp = spa; 3146fa9e4066Sahrens 3147fa9e4066Sahrens return (0); 3148fa9e4066Sahrens } 3149fa9e4066Sahrens 3150468c413aSTim Haley int 3151468c413aSTim Haley spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3152468c413aSTim Haley nvlist_t **config) 3153468c413aSTim Haley { 3154468c413aSTim Haley return (spa_open_common(name, spapp, tag, policy, config)); 3155468c413aSTim Haley } 3156468c413aSTim Haley 3157fa9e4066Sahrens int 3158fa9e4066Sahrens spa_open(const char *name, spa_t **spapp, void *tag) 3159fa9e4066Sahrens { 3160468c413aSTim Haley return (spa_open_common(name, spapp, tag, NULL, NULL)); 3161fa9e4066Sahrens } 3162fa9e4066Sahrens 3163ea8dc4b6Seschrock /* 3164ea8dc4b6Seschrock * Lookup the given spa_t, incrementing the inject count in the process, 3165ea8dc4b6Seschrock * preventing it from being exported or destroyed. 3166ea8dc4b6Seschrock */ 3167ea8dc4b6Seschrock spa_t * 3168ea8dc4b6Seschrock spa_inject_addref(char *name) 3169ea8dc4b6Seschrock { 3170ea8dc4b6Seschrock spa_t *spa; 3171ea8dc4b6Seschrock 3172ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3173ea8dc4b6Seschrock if ((spa = spa_lookup(name)) == NULL) { 3174ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3175ea8dc4b6Seschrock return (NULL); 3176ea8dc4b6Seschrock } 3177ea8dc4b6Seschrock spa->spa_inject_ref++; 3178ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3179ea8dc4b6Seschrock 3180ea8dc4b6Seschrock return (spa); 3181ea8dc4b6Seschrock } 3182ea8dc4b6Seschrock 3183ea8dc4b6Seschrock void 3184ea8dc4b6Seschrock spa_inject_delref(spa_t *spa) 3185ea8dc4b6Seschrock { 3186ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3187ea8dc4b6Seschrock spa->spa_inject_ref--; 3188ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3189ea8dc4b6Seschrock } 3190ea8dc4b6Seschrock 3191fa94a07fSbrendan /* 3192fa94a07fSbrendan * Add spares device information to the nvlist. 3193fa94a07fSbrendan */ 319499653d4eSeschrock static void 319599653d4eSeschrock spa_add_spares(spa_t *spa, nvlist_t *config) 319699653d4eSeschrock { 319799653d4eSeschrock nvlist_t **spares; 319899653d4eSeschrock uint_t i, nspares; 319999653d4eSeschrock nvlist_t *nvroot; 320099653d4eSeschrock uint64_t guid; 320199653d4eSeschrock vdev_stat_t *vs; 320299653d4eSeschrock uint_t vsc; 320339c23413Seschrock uint64_t pool; 320499653d4eSeschrock 32056809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 32066809eb4eSEric Schrock 3207fa94a07fSbrendan if (spa->spa_spares.sav_count == 0) 320899653d4eSeschrock return; 320999653d4eSeschrock 321099653d4eSeschrock VERIFY(nvlist_lookup_nvlist(config, 321199653d4eSeschrock ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3212fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 321399653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 321499653d4eSeschrock if (nspares != 0) { 321599653d4eSeschrock VERIFY(nvlist_add_nvlist_array(nvroot, 321699653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 321799653d4eSeschrock VERIFY(nvlist_lookup_nvlist_array(nvroot, 321899653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 321999653d4eSeschrock 322099653d4eSeschrock /* 322199653d4eSeschrock * Go through and find any spares which have since been 322299653d4eSeschrock * repurposed as an active spare. If this is the case, update 322399653d4eSeschrock * their status appropriately. 322499653d4eSeschrock */ 322599653d4eSeschrock for (i = 0; i < nspares; i++) { 322699653d4eSeschrock VERIFY(nvlist_lookup_uint64(spares[i], 322799653d4eSeschrock ZPOOL_CONFIG_GUID, &guid) == 0); 322889a89ebfSlling if (spa_spare_exists(guid, &pool, NULL) && 322989a89ebfSlling pool != 0ULL) { 323099653d4eSeschrock VERIFY(nvlist_lookup_uint64_array( 32313f9d6ad7SLin Ling spares[i], ZPOOL_CONFIG_VDEV_STATS, 323299653d4eSeschrock (uint64_t **)&vs, &vsc) == 0); 323399653d4eSeschrock vs->vs_state = VDEV_STATE_CANT_OPEN; 323499653d4eSeschrock vs->vs_aux = VDEV_AUX_SPARED; 323599653d4eSeschrock } 323699653d4eSeschrock } 323799653d4eSeschrock } 323899653d4eSeschrock } 323999653d4eSeschrock 3240fa94a07fSbrendan /* 3241fa94a07fSbrendan * Add l2cache device information to the nvlist, including vdev stats. 3242fa94a07fSbrendan */ 3243fa94a07fSbrendan static void 3244fa94a07fSbrendan spa_add_l2cache(spa_t *spa, nvlist_t *config) 3245fa94a07fSbrendan { 3246fa94a07fSbrendan nvlist_t **l2cache; 3247fa94a07fSbrendan uint_t i, j, nl2cache; 3248fa94a07fSbrendan nvlist_t *nvroot; 3249fa94a07fSbrendan uint64_t guid; 3250fa94a07fSbrendan vdev_t *vd; 3251fa94a07fSbrendan vdev_stat_t *vs; 3252fa94a07fSbrendan uint_t vsc; 3253fa94a07fSbrendan 32546809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 32556809eb4eSEric Schrock 3256fa94a07fSbrendan if (spa->spa_l2cache.sav_count == 0) 3257fa94a07fSbrendan return; 3258fa94a07fSbrendan 3259fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist(config, 3260fa94a07fSbrendan ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3261fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3262fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3263fa94a07fSbrendan if (nl2cache != 0) { 3264fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, 3265fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3266fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(nvroot, 3267fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3268fa94a07fSbrendan 3269fa94a07fSbrendan /* 3270fa94a07fSbrendan * Update level 2 cache device stats. 3271fa94a07fSbrendan */ 3272fa94a07fSbrendan 3273fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 3274fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 3275fa94a07fSbrendan ZPOOL_CONFIG_GUID, &guid) == 0); 3276fa94a07fSbrendan 3277fa94a07fSbrendan vd = NULL; 3278fa94a07fSbrendan for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3279fa94a07fSbrendan if (guid == 3280fa94a07fSbrendan spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3281fa94a07fSbrendan vd = spa->spa_l2cache.sav_vdevs[j]; 3282fa94a07fSbrendan break; 3283fa94a07fSbrendan } 3284fa94a07fSbrendan } 3285fa94a07fSbrendan ASSERT(vd != NULL); 3286fa94a07fSbrendan 3287fa94a07fSbrendan VERIFY(nvlist_lookup_uint64_array(l2cache[i], 32883f9d6ad7SLin Ling ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 32893f9d6ad7SLin Ling == 0); 3290fa94a07fSbrendan vdev_get_stats(vd, vs); 3291fa94a07fSbrendan } 3292fa94a07fSbrendan } 3293fa94a07fSbrendan } 3294fa94a07fSbrendan 3295ad135b5dSChristopher Siden static void 3296ad135b5dSChristopher Siden spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3297ad135b5dSChristopher Siden { 3298ad135b5dSChristopher Siden nvlist_t *features; 3299ad135b5dSChristopher Siden zap_cursor_t zc; 3300ad135b5dSChristopher Siden zap_attribute_t za; 3301ad135b5dSChristopher Siden 3302ad135b5dSChristopher Siden ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3303ad135b5dSChristopher Siden VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3304ad135b5dSChristopher Siden 3305ad135b5dSChristopher Siden if (spa->spa_feat_for_read_obj != 0) { 3306ad135b5dSChristopher Siden for (zap_cursor_init(&zc, spa->spa_meta_objset, 3307ad135b5dSChristopher Siden spa->spa_feat_for_read_obj); 3308ad135b5dSChristopher Siden zap_cursor_retrieve(&zc, &za) == 0; 3309ad135b5dSChristopher Siden zap_cursor_advance(&zc)) { 3310ad135b5dSChristopher Siden ASSERT(za.za_integer_length == sizeof (uint64_t) && 3311ad135b5dSChristopher Siden za.za_num_integers == 1); 3312b420f3adSRichard Lowe VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3313ad135b5dSChristopher Siden za.za_first_integer)); 3314ad135b5dSChristopher Siden } 3315ad135b5dSChristopher Siden zap_cursor_fini(&zc); 3316ad135b5dSChristopher Siden } 3317ad135b5dSChristopher Siden 3318ad135b5dSChristopher Siden if (spa->spa_feat_for_write_obj != 0) { 3319ad135b5dSChristopher Siden for (zap_cursor_init(&zc, spa->spa_meta_objset, 3320ad135b5dSChristopher Siden spa->spa_feat_for_write_obj); 3321ad135b5dSChristopher Siden zap_cursor_retrieve(&zc, &za) == 0; 3322ad135b5dSChristopher Siden zap_cursor_advance(&zc)) { 3323ad135b5dSChristopher Siden ASSERT(za.za_integer_length == sizeof (uint64_t) && 3324ad135b5dSChristopher Siden za.za_num_integers == 1); 3325b420f3adSRichard Lowe VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3326ad135b5dSChristopher Siden za.za_first_integer)); 3327ad135b5dSChristopher Siden } 3328ad135b5dSChristopher Siden zap_cursor_fini(&zc); 3329ad135b5dSChristopher Siden } 3330ad135b5dSChristopher Siden 3331ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3332ad135b5dSChristopher Siden features) == 0); 3333ad135b5dSChristopher Siden nvlist_free(features); 3334ad135b5dSChristopher Siden } 3335ad135b5dSChristopher Siden 3336fa9e4066Sahrens int 3337ad135b5dSChristopher Siden spa_get_stats(const char *name, nvlist_t **config, 3338ad135b5dSChristopher Siden char *altroot, size_t buflen) 3339fa9e4066Sahrens { 3340fa9e4066Sahrens int error; 3341fa9e4066Sahrens spa_t *spa; 3342fa9e4066Sahrens 3343fa9e4066Sahrens *config = NULL; 3344468c413aSTim Haley error = spa_open_common(name, &spa, FTAG, NULL, config); 3345fa9e4066Sahrens 33466809eb4eSEric Schrock if (spa != NULL) { 33476809eb4eSEric Schrock /* 33486809eb4eSEric Schrock * This still leaves a window of inconsistency where the spares 33496809eb4eSEric Schrock * or l2cache devices could change and the config would be 33506809eb4eSEric Schrock * self-inconsistent. 33516809eb4eSEric Schrock */ 33526809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3353ea8dc4b6Seschrock 33546809eb4eSEric Schrock if (*config != NULL) { 335511027bc7STim Haley uint64_t loadtimes[2]; 335611027bc7STim Haley 335711027bc7STim Haley loadtimes[0] = spa->spa_loaded_ts.tv_sec; 335811027bc7STim Haley loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 335911027bc7STim Haley VERIFY(nvlist_add_uint64_array(*config, 336011027bc7STim Haley ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 336111027bc7STim Haley 3362e14bb325SJeff Bonwick VERIFY(nvlist_add_uint64(*config, 33636809eb4eSEric Schrock ZPOOL_CONFIG_ERRCOUNT, 33646809eb4eSEric Schrock spa_get_errlog_size(spa)) == 0); 3365e14bb325SJeff Bonwick 33666809eb4eSEric Schrock if (spa_suspended(spa)) 33676809eb4eSEric Schrock VERIFY(nvlist_add_uint64(*config, 33686809eb4eSEric Schrock ZPOOL_CONFIG_SUSPENDED, 33696809eb4eSEric Schrock spa->spa_failmode) == 0); 33706809eb4eSEric Schrock 33716809eb4eSEric Schrock spa_add_spares(spa, *config); 33726809eb4eSEric Schrock spa_add_l2cache(spa, *config); 3373ad135b5dSChristopher Siden spa_add_feature_stats(spa, *config); 33746809eb4eSEric Schrock } 337599653d4eSeschrock } 337699653d4eSeschrock 3377ea8dc4b6Seschrock /* 3378ea8dc4b6Seschrock * We want to get the alternate root even for faulted pools, so we cheat 3379ea8dc4b6Seschrock * and call spa_lookup() directly. 3380ea8dc4b6Seschrock */ 3381ea8dc4b6Seschrock if (altroot) { 3382ea8dc4b6Seschrock if (spa == NULL) { 3383ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3384ea8dc4b6Seschrock spa = spa_lookup(name); 3385ea8dc4b6Seschrock if (spa) 3386ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 3387ea8dc4b6Seschrock else 3388ea8dc4b6Seschrock altroot[0] = '\0'; 3389ea8dc4b6Seschrock spa = NULL; 3390ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3391ea8dc4b6Seschrock } else { 3392ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 3393ea8dc4b6Seschrock } 3394ea8dc4b6Seschrock } 3395ea8dc4b6Seschrock 33966809eb4eSEric Schrock if (spa != NULL) { 33976809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 3398fa9e4066Sahrens spa_close(spa, FTAG); 33996809eb4eSEric Schrock } 3400fa9e4066Sahrens 3401fa9e4066Sahrens return (error); 3402fa9e4066Sahrens } 3403fa9e4066Sahrens 340499653d4eSeschrock /* 3405fa94a07fSbrendan * Validate that the auxiliary device array is well formed. We must have an 3406fa94a07fSbrendan * array of nvlists, each which describes a valid leaf vdev. If this is an 3407fa94a07fSbrendan * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3408fa94a07fSbrendan * specified, as long as they are well-formed. 340999653d4eSeschrock */ 341099653d4eSeschrock static int 3411fa94a07fSbrendan spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3412fa94a07fSbrendan spa_aux_vdev_t *sav, const char *config, uint64_t version, 3413fa94a07fSbrendan vdev_labeltype_t label) 341499653d4eSeschrock { 3415fa94a07fSbrendan nvlist_t **dev; 3416fa94a07fSbrendan uint_t i, ndev; 341799653d4eSeschrock vdev_t *vd; 341899653d4eSeschrock int error; 341999653d4eSeschrock 3420e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3421e14bb325SJeff Bonwick 342299653d4eSeschrock /* 3423fa94a07fSbrendan * It's acceptable to have no devs specified. 342499653d4eSeschrock */ 3425fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 342699653d4eSeschrock return (0); 342799653d4eSeschrock 3428fa94a07fSbrendan if (ndev == 0) 3429be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 343099653d4eSeschrock 343199653d4eSeschrock /* 3432fa94a07fSbrendan * Make sure the pool is formatted with a version that supports this 3433fa94a07fSbrendan * device type. 343499653d4eSeschrock */ 3435fa94a07fSbrendan if (spa_version(spa) < version) 3436be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 343799653d4eSeschrock 343839c23413Seschrock /* 3439fa94a07fSbrendan * Set the pending device list so we correctly handle device in-use 344039c23413Seschrock * checking. 344139c23413Seschrock */ 3442fa94a07fSbrendan sav->sav_pending = dev; 3443fa94a07fSbrendan sav->sav_npending = ndev; 344439c23413Seschrock 3445fa94a07fSbrendan for (i = 0; i < ndev; i++) { 3446fa94a07fSbrendan if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 344799653d4eSeschrock mode)) != 0) 344839c23413Seschrock goto out; 344999653d4eSeschrock 345099653d4eSeschrock if (!vd->vdev_ops->vdev_op_leaf) { 345199653d4eSeschrock vdev_free(vd); 3452be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 345339c23413Seschrock goto out; 345499653d4eSeschrock } 345599653d4eSeschrock 3456fa94a07fSbrendan /* 3457e14bb325SJeff Bonwick * The L2ARC currently only supports disk devices in 3458e14bb325SJeff Bonwick * kernel context. For user-level testing, we allow it. 3459fa94a07fSbrendan */ 3460e14bb325SJeff Bonwick #ifdef _KERNEL 3461fa94a07fSbrendan if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3462fa94a07fSbrendan strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3463be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTBLK); 3464cd0837ccSGeorge Wilson vdev_free(vd); 3465fa94a07fSbrendan goto out; 3466fa94a07fSbrendan } 3467e14bb325SJeff Bonwick #endif 346899653d4eSeschrock vd->vdev_top = vd; 346999653d4eSeschrock 347039c23413Seschrock if ((error = vdev_open(vd)) == 0 && 3471fa94a07fSbrendan (error = vdev_label_init(vd, crtxg, label)) == 0) { 3472fa94a07fSbrendan VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 347339c23413Seschrock vd->vdev_guid) == 0); 347439c23413Seschrock } 347599653d4eSeschrock 347699653d4eSeschrock vdev_free(vd); 347739c23413Seschrock 3478fa94a07fSbrendan if (error && 3479fa94a07fSbrendan (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 348039c23413Seschrock goto out; 348139c23413Seschrock else 348239c23413Seschrock error = 0; 348399653d4eSeschrock } 348499653d4eSeschrock 348539c23413Seschrock out: 3486fa94a07fSbrendan sav->sav_pending = NULL; 3487fa94a07fSbrendan sav->sav_npending = 0; 348839c23413Seschrock return (error); 348999653d4eSeschrock } 349099653d4eSeschrock 3491fa94a07fSbrendan static int 3492fa94a07fSbrendan spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3493fa94a07fSbrendan { 3494fa94a07fSbrendan int error; 3495fa94a07fSbrendan 3496e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3497e14bb325SJeff Bonwick 3498fa94a07fSbrendan if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3499fa94a07fSbrendan &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3500fa94a07fSbrendan VDEV_LABEL_SPARE)) != 0) { 3501fa94a07fSbrendan return (error); 3502fa94a07fSbrendan } 3503fa94a07fSbrendan 3504fa94a07fSbrendan return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3505fa94a07fSbrendan &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3506fa94a07fSbrendan VDEV_LABEL_L2CACHE)); 3507fa94a07fSbrendan } 3508fa94a07fSbrendan 3509fa94a07fSbrendan static void 3510fa94a07fSbrendan spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3511fa94a07fSbrendan const char *config) 3512fa94a07fSbrendan { 3513fa94a07fSbrendan int i; 3514fa94a07fSbrendan 3515fa94a07fSbrendan if (sav->sav_config != NULL) { 3516fa94a07fSbrendan nvlist_t **olddevs; 3517fa94a07fSbrendan uint_t oldndevs; 3518fa94a07fSbrendan nvlist_t **newdevs; 3519fa94a07fSbrendan 3520fa94a07fSbrendan /* 3521fa94a07fSbrendan * Generate new dev list by concatentating with the 3522fa94a07fSbrendan * current dev list. 3523fa94a07fSbrendan */ 3524fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3525fa94a07fSbrendan &olddevs, &oldndevs) == 0); 3526fa94a07fSbrendan 3527fa94a07fSbrendan newdevs = kmem_alloc(sizeof (void *) * 3528fa94a07fSbrendan (ndevs + oldndevs), KM_SLEEP); 3529fa94a07fSbrendan for (i = 0; i < oldndevs; i++) 3530fa94a07fSbrendan VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3531fa94a07fSbrendan KM_SLEEP) == 0); 3532fa94a07fSbrendan for (i = 0; i < ndevs; i++) 3533fa94a07fSbrendan VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3534fa94a07fSbrendan KM_SLEEP) == 0); 3535fa94a07fSbrendan 3536fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, config, 3537fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 3538fa94a07fSbrendan 3539fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3540fa94a07fSbrendan config, newdevs, ndevs + oldndevs) == 0); 3541fa94a07fSbrendan for (i = 0; i < oldndevs + ndevs; i++) 3542fa94a07fSbrendan nvlist_free(newdevs[i]); 3543fa94a07fSbrendan kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3544fa94a07fSbrendan } else { 3545fa94a07fSbrendan /* 3546fa94a07fSbrendan * Generate a new dev list. 3547fa94a07fSbrendan */ 3548fa94a07fSbrendan VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3549fa94a07fSbrendan KM_SLEEP) == 0); 3550fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3551fa94a07fSbrendan devs, ndevs) == 0); 3552fa94a07fSbrendan } 3553fa94a07fSbrendan } 3554fa94a07fSbrendan 3555fa94a07fSbrendan /* 3556fa94a07fSbrendan * Stop and drop level 2 ARC devices 3557fa94a07fSbrendan */ 3558fa94a07fSbrendan void 3559fa94a07fSbrendan spa_l2cache_drop(spa_t *spa) 3560fa94a07fSbrendan { 3561fa94a07fSbrendan vdev_t *vd; 3562fa94a07fSbrendan int i; 3563fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 3564fa94a07fSbrendan 3565fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) { 3566fa94a07fSbrendan uint64_t pool; 3567fa94a07fSbrendan 3568fa94a07fSbrendan vd = sav->sav_vdevs[i]; 3569fa94a07fSbrendan ASSERT(vd != NULL); 3570fa94a07fSbrendan 35718ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 35728ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 3573fa94a07fSbrendan l2arc_remove_vdev(vd); 3574fa94a07fSbrendan } 3575fa94a07fSbrendan } 3576fa94a07fSbrendan 3577fa9e4066Sahrens /* 3578fa9e4066Sahrens * Pool Creation 3579fa9e4066Sahrens */ 3580fa9e4066Sahrens int 3581990b4856Slling spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 35824445fffbSMatthew Ahrens nvlist_t *zplprops) 3583fa9e4066Sahrens { 3584fa9e4066Sahrens spa_t *spa; 3585990b4856Slling char *altroot = NULL; 35860373e76bSbonwick vdev_t *rvd; 3587fa9e4066Sahrens dsl_pool_t *dp; 3588fa9e4066Sahrens dmu_tx_t *tx; 3589573ca77eSGeorge Wilson int error = 0; 3590fa9e4066Sahrens uint64_t txg = TXG_INITIAL; 3591fa94a07fSbrendan nvlist_t **spares, **l2cache; 3592fa94a07fSbrendan uint_t nspares, nl2cache; 3593cde58dbcSMatthew Ahrens uint64_t version, obj; 3594ad135b5dSChristopher Siden boolean_t has_features; 3595fa9e4066Sahrens 3596fa9e4066Sahrens /* 3597fa9e4066Sahrens * If this pool already exists, return failure. 3598fa9e4066Sahrens */ 3599fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 3600fa9e4066Sahrens if (spa_lookup(pool) != NULL) { 3601fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3602be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 3603fa9e4066Sahrens } 3604fa9e4066Sahrens 3605fa9e4066Sahrens /* 3606fa9e4066Sahrens * Allocate a new spa_t structure. 3607fa9e4066Sahrens */ 3608990b4856Slling (void) nvlist_lookup_string(props, 3609990b4856Slling zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3610468c413aSTim Haley spa = spa_add(pool, NULL, altroot); 36118ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 3612fa9e4066Sahrens 3613990b4856Slling if (props && (error = spa_prop_validate(spa, props))) { 3614990b4856Slling spa_deactivate(spa); 3615990b4856Slling spa_remove(spa); 3616c5904d13Seschrock mutex_exit(&spa_namespace_lock); 3617990b4856Slling return (error); 3618990b4856Slling } 3619990b4856Slling 3620ad135b5dSChristopher Siden has_features = B_FALSE; 3621ad135b5dSChristopher Siden for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3622ad135b5dSChristopher Siden elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3623ad135b5dSChristopher Siden if (zpool_prop_feature(nvpair_name(elem))) 3624ad135b5dSChristopher Siden has_features = B_TRUE; 3625ad135b5dSChristopher Siden } 3626ad135b5dSChristopher Siden 3627ad135b5dSChristopher Siden if (has_features || nvlist_lookup_uint64(props, 3628ad135b5dSChristopher Siden zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3629990b4856Slling version = SPA_VERSION; 3630ad135b5dSChristopher Siden } 3631ad135b5dSChristopher Siden ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3632b24ab676SJeff Bonwick 3633b24ab676SJeff Bonwick spa->spa_first_txg = txg; 3634b24ab676SJeff Bonwick spa->spa_uberblock.ub_txg = txg - 1; 3635990b4856Slling spa->spa_uberblock.ub_version = version; 3636fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 3637fa9e4066Sahrens 363854d692b7SGeorge Wilson /* 363954d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 364054d692b7SGeorge Wilson */ 36416f834bc1SMatthew Ahrens spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 36426f834bc1SMatthew Ahrens KM_SLEEP); 36436f834bc1SMatthew Ahrens for (int i = 0; i < max_ncpus; i++) { 36446f834bc1SMatthew Ahrens spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 36456f834bc1SMatthew Ahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 36466f834bc1SMatthew Ahrens ZIO_FLAG_GODFATHER); 36476f834bc1SMatthew Ahrens } 364854d692b7SGeorge Wilson 36490373e76bSbonwick /* 36500373e76bSbonwick * Create the root vdev. 36510373e76bSbonwick */ 3652e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 36530373e76bSbonwick 365499653d4eSeschrock error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 36550373e76bSbonwick 365699653d4eSeschrock ASSERT(error != 0 || rvd != NULL); 365799653d4eSeschrock ASSERT(error != 0 || spa->spa_root_vdev == rvd); 36580373e76bSbonwick 3659b7b97454Sperrin if (error == 0 && !zfs_allocatable_devs(nvroot)) 3660be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 366199653d4eSeschrock 366299653d4eSeschrock if (error == 0 && 366399653d4eSeschrock (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3664fa94a07fSbrendan (error = spa_validate_aux(spa, nvroot, txg, 366599653d4eSeschrock VDEV_ALLOC_ADD)) == 0) { 3666573ca77eSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 3667573ca77eSGeorge Wilson vdev_metaslab_set_size(rvd->vdev_child[c]); 3668573ca77eSGeorge Wilson vdev_expand(rvd->vdev_child[c], txg); 3669573ca77eSGeorge Wilson } 36700373e76bSbonwick } 36710373e76bSbonwick 3672e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3673fa9e4066Sahrens 367499653d4eSeschrock if (error != 0) { 3675fa9e4066Sahrens spa_unload(spa); 3676fa9e4066Sahrens spa_deactivate(spa); 3677fa9e4066Sahrens spa_remove(spa); 3678fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3679fa9e4066Sahrens return (error); 3680fa9e4066Sahrens } 3681fa9e4066Sahrens 368299653d4eSeschrock /* 368399653d4eSeschrock * Get the list of spares, if specified. 368499653d4eSeschrock */ 368599653d4eSeschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 368699653d4eSeschrock &spares, &nspares) == 0) { 3687fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 368899653d4eSeschrock KM_SLEEP) == 0); 3689fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 369099653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3691e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 369299653d4eSeschrock spa_load_spares(spa); 3693e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3694fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 3695fa94a07fSbrendan } 3696fa94a07fSbrendan 3697fa94a07fSbrendan /* 3698fa94a07fSbrendan * Get the list of level 2 cache devices, if specified. 3699fa94a07fSbrendan */ 3700fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3701fa94a07fSbrendan &l2cache, &nl2cache) == 0) { 3702fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3703fa94a07fSbrendan NV_UNIQUE_NAME, KM_SLEEP) == 0); 3704fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3705fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3706e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3707fa94a07fSbrendan spa_load_l2cache(spa); 3708e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3709fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 371099653d4eSeschrock } 371199653d4eSeschrock 3712ad135b5dSChristopher Siden spa->spa_is_initializing = B_TRUE; 37130a48a24eStimh spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3714fa9e4066Sahrens spa->spa_meta_objset = dp->dp_meta_objset; 3715ad135b5dSChristopher Siden spa->spa_is_initializing = B_FALSE; 3716fa9e4066Sahrens 3717485bbbf5SGeorge Wilson /* 3718485bbbf5SGeorge Wilson * Create DDTs (dedup tables). 3719485bbbf5SGeorge Wilson */ 3720485bbbf5SGeorge Wilson ddt_create(spa); 3721485bbbf5SGeorge Wilson 3722485bbbf5SGeorge Wilson spa_update_dspace(spa); 3723485bbbf5SGeorge Wilson 3724fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 3725fa9e4066Sahrens 3726fa9e4066Sahrens /* 3727fa9e4066Sahrens * Create the pool config object. 3728fa9e4066Sahrens */ 3729fa9e4066Sahrens spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3730f7991ba4STim Haley DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3731fa9e4066Sahrens DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3732fa9e4066Sahrens 3733ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 3734fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3735ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3736ea8dc4b6Seschrock cmn_err(CE_PANIC, "failed to add pool config"); 3737ea8dc4b6Seschrock } 3738fa9e4066Sahrens 3739ad135b5dSChristopher Siden if (spa_version(spa) >= SPA_VERSION_FEATURES) 3740ad135b5dSChristopher Siden spa_feature_create_zap_objects(spa, tx); 3741ad135b5dSChristopher Siden 37423f9d6ad7SLin Ling if (zap_add(spa->spa_meta_objset, 37433f9d6ad7SLin Ling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 37443f9d6ad7SLin Ling sizeof (uint64_t), 1, &version, tx) != 0) { 37453f9d6ad7SLin Ling cmn_err(CE_PANIC, "failed to add pool version"); 37463f9d6ad7SLin Ling } 37473f9d6ad7SLin Ling 3748990b4856Slling /* Newly created pools with the right version are always deflated. */ 3749990b4856Slling if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3750990b4856Slling spa->spa_deflate = TRUE; 3751990b4856Slling if (zap_add(spa->spa_meta_objset, 3752990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3753990b4856Slling sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3754990b4856Slling cmn_err(CE_PANIC, "failed to add deflate"); 3755990b4856Slling } 375699653d4eSeschrock } 375799653d4eSeschrock 3758fa9e4066Sahrens /* 3759cde58dbcSMatthew Ahrens * Create the deferred-free bpobj. Turn off compression 3760fa9e4066Sahrens * because sync-to-convergence takes longer if the blocksize 3761fa9e4066Sahrens * keeps changing. 3762fa9e4066Sahrens */ 3763cde58dbcSMatthew Ahrens obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3764cde58dbcSMatthew Ahrens dmu_object_set_compress(spa->spa_meta_objset, obj, 3765cde58dbcSMatthew Ahrens ZIO_COMPRESS_OFF, tx); 3766ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 3767cde58dbcSMatthew Ahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3768cde58dbcSMatthew Ahrens sizeof (uint64_t), 1, &obj, tx) != 0) { 3769cde58dbcSMatthew Ahrens cmn_err(CE_PANIC, "failed to add bpobj"); 3770ea8dc4b6Seschrock } 3771b420f3adSRichard Lowe VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3772cde58dbcSMatthew Ahrens spa->spa_meta_objset, obj)); 3773fa9e4066Sahrens 377406eeb2adSek /* 377506eeb2adSek * Create the pool's history object. 377606eeb2adSek */ 3777990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY) 3778990b4856Slling spa_history_create_obj(spa, tx); 3779990b4856Slling 378045818ee1SMatthew Ahrens /* 378145818ee1SMatthew Ahrens * Generate some random noise for salted checksums to operate on. 378245818ee1SMatthew Ahrens */ 378345818ee1SMatthew Ahrens (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 378445818ee1SMatthew Ahrens sizeof (spa->spa_cksum_salt.zcs_bytes)); 378545818ee1SMatthew Ahrens 3786990b4856Slling /* 3787990b4856Slling * Set pool properties. 3788990b4856Slling */ 3789990b4856Slling spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3790990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 37910a4e9518Sgw spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3792573ca77eSGeorge Wilson spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3793b24ab676SJeff Bonwick 3794379c004dSEric Schrock if (props != NULL) { 3795379c004dSEric Schrock spa_configfile_set(spa, props, B_FALSE); 37963b2aab18SMatthew Ahrens spa_sync_props(props, tx); 3797379c004dSEric Schrock } 379806eeb2adSek 3799fa9e4066Sahrens dmu_tx_commit(tx); 3800fa9e4066Sahrens 3801fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 3802fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 3803fa9e4066Sahrens 3804fa9e4066Sahrens /* 3805fa9e4066Sahrens * We explicitly wait for the first transaction to complete so that our 3806fa9e4066Sahrens * bean counters are appropriately updated. 3807fa9e4066Sahrens */ 3808fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 3809fa9e4066Sahrens 3810c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 381114372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3812fa9e4066Sahrens 38134445fffbSMatthew Ahrens spa_history_log_version(spa, "create"); 3814228975ccSek 3815bc9014e6SJustin Gibbs /* 3816bc9014e6SJustin Gibbs * Don't count references from objsets that are already closed 3817bc9014e6SJustin Gibbs * and are making their way through the eviction process. 3818bc9014e6SJustin Gibbs */ 3819bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 3820088f3894Sahrens spa->spa_minref = refcount_count(&spa->spa_refcount); 3821088f3894Sahrens 3822daaa36a7SGeorge Wilson mutex_exit(&spa_namespace_lock); 3823daaa36a7SGeorge Wilson 3824fa9e4066Sahrens return (0); 3825fa9e4066Sahrens } 3826fa9e4066Sahrens 3827e7cbe64fSgw #ifdef _KERNEL 3828e7cbe64fSgw /* 382921ecdf64SLin Ling * Get the root pool information from the root disk, then import the root pool 383021ecdf64SLin Ling * during the system boot up time. 3831e7cbe64fSgw */ 383221ecdf64SLin Ling extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 383321ecdf64SLin Ling 383421ecdf64SLin Ling static nvlist_t * 383521ecdf64SLin Ling spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3836e7cbe64fSgw { 383721ecdf64SLin Ling nvlist_t *config; 3838e7cbe64fSgw nvlist_t *nvtop, *nvroot; 3839e7cbe64fSgw uint64_t pgid; 3840e7cbe64fSgw 384121ecdf64SLin Ling if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 384221ecdf64SLin Ling return (NULL); 384321ecdf64SLin Ling 3844e7cbe64fSgw /* 3845e7cbe64fSgw * Add this top-level vdev to the child array. 3846e7cbe64fSgw */ 384721ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 384821ecdf64SLin Ling &nvtop) == 0); 384921ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 385021ecdf64SLin Ling &pgid) == 0); 385121ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3852e7cbe64fSgw 3853e7cbe64fSgw /* 3854e7cbe64fSgw * Put this pool's top-level vdevs into a root vdev. 3855e7cbe64fSgw */ 3856e7cbe64fSgw VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 385721ecdf64SLin Ling VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 385821ecdf64SLin Ling VDEV_TYPE_ROOT) == 0); 3859e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3860e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3861e7cbe64fSgw VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3862e7cbe64fSgw &nvtop, 1) == 0); 3863e7cbe64fSgw 3864e7cbe64fSgw /* 3865e7cbe64fSgw * Replace the existing vdev_tree with the new root vdev in 3866e7cbe64fSgw * this pool's configuration (remove the old, add the new). 3867e7cbe64fSgw */ 3868e7cbe64fSgw VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3869e7cbe64fSgw nvlist_free(nvroot); 387021ecdf64SLin Ling return (config); 3871e7cbe64fSgw } 3872e7cbe64fSgw 3873e7cbe64fSgw /* 387421ecdf64SLin Ling * Walk the vdev tree and see if we can find a device with "better" 387521ecdf64SLin Ling * configuration. A configuration is "better" if the label on that 387621ecdf64SLin Ling * device has a more recent txg. 3877051aabe6Staylor */ 387821ecdf64SLin Ling static void 387921ecdf64SLin Ling spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3880051aabe6Staylor { 3881573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 388221ecdf64SLin Ling spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3883051aabe6Staylor 388421ecdf64SLin Ling if (vd->vdev_ops->vdev_op_leaf) { 388521ecdf64SLin Ling nvlist_t *label; 388621ecdf64SLin Ling uint64_t label_txg; 3887051aabe6Staylor 388821ecdf64SLin Ling if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 388921ecdf64SLin Ling &label) != 0) 389021ecdf64SLin Ling return; 3891051aabe6Staylor 389221ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 389321ecdf64SLin Ling &label_txg) == 0); 3894051aabe6Staylor 389521ecdf64SLin Ling /* 389621ecdf64SLin Ling * Do we have a better boot device? 389721ecdf64SLin Ling */ 389821ecdf64SLin Ling if (label_txg > *txg) { 389921ecdf64SLin Ling *txg = label_txg; 390021ecdf64SLin Ling *avd = vd; 3901051aabe6Staylor } 390221ecdf64SLin Ling nvlist_free(label); 3903051aabe6Staylor } 3904051aabe6Staylor } 3905051aabe6Staylor 3906e7cbe64fSgw /* 3907e7cbe64fSgw * Import a root pool. 3908e7cbe64fSgw * 3909051aabe6Staylor * For x86. devpath_list will consist of devid and/or physpath name of 3910051aabe6Staylor * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3911051aabe6Staylor * The GRUB "findroot" command will return the vdev we should boot. 3912e7cbe64fSgw * 3913e7cbe64fSgw * For Sparc, devpath_list consists the physpath name of the booting device 3914e7cbe64fSgw * no matter the rootpool is a single device pool or a mirrored pool. 3915e7cbe64fSgw * e.g. 3916e7cbe64fSgw * "/pci@1f,0/ide@d/disk@0,0:a" 3917e7cbe64fSgw */ 3918e7cbe64fSgw int 3919051aabe6Staylor spa_import_rootpool(char *devpath, char *devid) 3920e7cbe64fSgw { 392121ecdf64SLin Ling spa_t *spa; 392221ecdf64SLin Ling vdev_t *rvd, *bvd, *avd = NULL; 392321ecdf64SLin Ling nvlist_t *config, *nvtop; 392421ecdf64SLin Ling uint64_t guid, txg; 3925e7cbe64fSgw char *pname; 3926e7cbe64fSgw int error; 3927e7cbe64fSgw 3928e7cbe64fSgw /* 392921ecdf64SLin Ling * Read the label from the boot device and generate a configuration. 3930e7cbe64fSgw */ 3931dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3932dedec472SJack Meng #if defined(_OBP) && defined(_KERNEL) 3933dedec472SJack Meng if (config == NULL) { 3934dedec472SJack Meng if (strstr(devpath, "/iscsi/ssd") != NULL) { 3935dedec472SJack Meng /* iscsi boot */ 3936dedec472SJack Meng get_iscsi_bootpath_phy(devpath); 3937dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3938dedec472SJack Meng } 3939dedec472SJack Meng } 3940dedec472SJack Meng #endif 3941dedec472SJack Meng if (config == NULL) { 3942ad135b5dSChristopher Siden cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 394321ecdf64SLin Ling devpath); 3944be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 394521ecdf64SLin Ling } 3946e7cbe64fSgw 394721ecdf64SLin Ling VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 394821ecdf64SLin Ling &pname) == 0); 394921ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3950e7cbe64fSgw 39516809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 39526809eb4eSEric Schrock if ((spa = spa_lookup(pname)) != NULL) { 39536809eb4eSEric Schrock /* 39546809eb4eSEric Schrock * Remove the existing root pool from the namespace so that we 39556809eb4eSEric Schrock * can replace it with the correct config we just read in. 39566809eb4eSEric Schrock */ 39576809eb4eSEric Schrock spa_remove(spa); 39586809eb4eSEric Schrock } 39596809eb4eSEric Schrock 3960468c413aSTim Haley spa = spa_add(pname, config, NULL); 39616809eb4eSEric Schrock spa->spa_is_root = B_TRUE; 39624b964adaSGeorge Wilson spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3963e7cbe64fSgw 396421ecdf64SLin Ling /* 396521ecdf64SLin Ling * Build up a vdev tree based on the boot device's label config. 396621ecdf64SLin Ling */ 396721ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 396821ecdf64SLin Ling &nvtop) == 0); 396921ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 397021ecdf64SLin Ling error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 397121ecdf64SLin Ling VDEV_ALLOC_ROOTPOOL); 397221ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 397321ecdf64SLin Ling if (error) { 397421ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 397521ecdf64SLin Ling nvlist_free(config); 397621ecdf64SLin Ling cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 397721ecdf64SLin Ling pname); 397821ecdf64SLin Ling return (error); 397921ecdf64SLin Ling } 398021ecdf64SLin Ling 398121ecdf64SLin Ling /* 398221ecdf64SLin Ling * Get the boot vdev. 398321ecdf64SLin Ling */ 398421ecdf64SLin Ling if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 398521ecdf64SLin Ling cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 398621ecdf64SLin Ling (u_longlong_t)guid); 3987be6fd75aSMatthew Ahrens error = SET_ERROR(ENOENT); 398821ecdf64SLin Ling goto out; 398921ecdf64SLin Ling } 3990e7cbe64fSgw 399121ecdf64SLin Ling /* 399221ecdf64SLin Ling * Determine if there is a better boot device. 399321ecdf64SLin Ling */ 399421ecdf64SLin Ling avd = bvd; 399521ecdf64SLin Ling spa_alt_rootvdev(rvd, &avd, &txg); 399621ecdf64SLin Ling if (avd != bvd) { 399721ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 399821ecdf64SLin Ling "try booting from '%s'", avd->vdev_path); 3999be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 400021ecdf64SLin Ling goto out; 400121ecdf64SLin Ling } 4002e7cbe64fSgw 400321ecdf64SLin Ling /* 400421ecdf64SLin Ling * If the boot device is part of a spare vdev then ensure that 400521ecdf64SLin Ling * we're booting off the active spare. 400621ecdf64SLin Ling */ 400721ecdf64SLin Ling if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 400821ecdf64SLin Ling !bvd->vdev_isspare) { 400921ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is currently spared. Please " 401021ecdf64SLin Ling "try booting from '%s'", 4011cb04b873SMark J Musante bvd->vdev_parent-> 4012cb04b873SMark J Musante vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4013be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 401421ecdf64SLin Ling goto out; 401521ecdf64SLin Ling } 401621ecdf64SLin Ling 401721ecdf64SLin Ling error = 0; 401821ecdf64SLin Ling out: 401921ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 402021ecdf64SLin Ling vdev_free(rvd); 402121ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 402221ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 402321ecdf64SLin Ling 402421ecdf64SLin Ling nvlist_free(config); 4025e7cbe64fSgw return (error); 4026e7cbe64fSgw } 402721ecdf64SLin Ling 4028e7cbe64fSgw #endif 4029e7cbe64fSgw 40306809eb4eSEric Schrock /* 40316809eb4eSEric Schrock * Import a non-root pool into the system. 40326809eb4eSEric Schrock */ 4033c5904d13Seschrock int 40344b964adaSGeorge Wilson spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4035c5904d13Seschrock { 40366809eb4eSEric Schrock spa_t *spa; 40376809eb4eSEric Schrock char *altroot = NULL; 4038468c413aSTim Haley spa_load_state_t state = SPA_LOAD_IMPORT; 4039468c413aSTim Haley zpool_rewind_policy_t policy; 4040f9af39baSGeorge Wilson uint64_t mode = spa_mode_global; 4041f9af39baSGeorge Wilson uint64_t readonly = B_FALSE; 40426809eb4eSEric Schrock int error; 40436809eb4eSEric Schrock nvlist_t *nvroot; 40446809eb4eSEric Schrock nvlist_t **spares, **l2cache; 40456809eb4eSEric Schrock uint_t nspares, nl2cache; 40466809eb4eSEric Schrock 40476809eb4eSEric Schrock /* 40486809eb4eSEric Schrock * If a pool with this name exists, return failure. 40496809eb4eSEric Schrock */ 40506809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 40511195e687SMark J Musante if (spa_lookup(pool) != NULL) { 40526809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 4053be6fd75aSMatthew Ahrens return (SET_ERROR(EEXIST)); 40546809eb4eSEric Schrock } 40556809eb4eSEric Schrock 40566809eb4eSEric Schrock /* 40576809eb4eSEric Schrock * Create and initialize the spa structure. 40586809eb4eSEric Schrock */ 40596809eb4eSEric Schrock (void) nvlist_lookup_string(props, 40606809eb4eSEric Schrock zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4061f9af39baSGeorge Wilson (void) nvlist_lookup_uint64(props, 4062f9af39baSGeorge Wilson zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4063f9af39baSGeorge Wilson if (readonly) 4064f9af39baSGeorge Wilson mode = FREAD; 4065468c413aSTim Haley spa = spa_add(pool, config, altroot); 40664b964adaSGeorge Wilson spa->spa_import_flags = flags; 40674b964adaSGeorge Wilson 40684b964adaSGeorge Wilson /* 40694b964adaSGeorge Wilson * Verbatim import - Take a pool and insert it into the namespace 40704b964adaSGeorge Wilson * as if it had been loaded at boot. 40714b964adaSGeorge Wilson */ 40724b964adaSGeorge Wilson if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 40734b964adaSGeorge Wilson if (props != NULL) 40744b964adaSGeorge Wilson spa_configfile_set(spa, props, B_FALSE); 40754b964adaSGeorge Wilson 40764b964adaSGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 407714372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 40784b964adaSGeorge Wilson 40794b964adaSGeorge Wilson mutex_exit(&spa_namespace_lock); 40804b964adaSGeorge Wilson return (0); 40814b964adaSGeorge Wilson } 40824b964adaSGeorge Wilson 4083f9af39baSGeorge Wilson spa_activate(spa, mode); 40846809eb4eSEric Schrock 408525f89ee2SJeff Bonwick /* 408625f89ee2SJeff Bonwick * Don't start async tasks until we know everything is healthy. 408725f89ee2SJeff Bonwick */ 408825f89ee2SJeff Bonwick spa_async_suspend(spa); 408925f89ee2SJeff Bonwick 40904b964adaSGeorge Wilson zpool_get_rewind_policy(config, &policy); 40914b964adaSGeorge Wilson if (policy.zrp_request & ZPOOL_DO_REWIND) 40924b964adaSGeorge Wilson state = SPA_LOAD_RECOVER; 40934b964adaSGeorge Wilson 40946809eb4eSEric Schrock /* 40956809eb4eSEric Schrock * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 40966809eb4eSEric Schrock * because the user-supplied config is actually the one to trust when 40976809eb4eSEric Schrock * doing an import. 40986809eb4eSEric Schrock */ 4099468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 4100468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 41014b964adaSGeorge Wilson 4102468c413aSTim Haley error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4103c8ee1847SVictor Latushkin policy.zrp_request); 4104468c413aSTim Haley 4105468c413aSTim Haley /* 41064b964adaSGeorge Wilson * Propagate anything learned while loading the pool and pass it 41074b964adaSGeorge Wilson * back to caller (i.e. rewind info, missing devices, etc). 4108468c413aSTim Haley */ 41094b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 41104b964adaSGeorge Wilson spa->spa_load_info) == 0); 41116809eb4eSEric Schrock 41126809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 41136809eb4eSEric Schrock /* 41146809eb4eSEric Schrock * Toss any existing sparelist, as it doesn't have any validity 41156809eb4eSEric Schrock * anymore, and conflicts with spa_has_spare(). 41166809eb4eSEric Schrock */ 41176809eb4eSEric Schrock if (spa->spa_spares.sav_config) { 41186809eb4eSEric Schrock nvlist_free(spa->spa_spares.sav_config); 41196809eb4eSEric Schrock spa->spa_spares.sav_config = NULL; 41206809eb4eSEric Schrock spa_load_spares(spa); 41216809eb4eSEric Schrock } 41226809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) { 41236809eb4eSEric Schrock nvlist_free(spa->spa_l2cache.sav_config); 41246809eb4eSEric Schrock spa->spa_l2cache.sav_config = NULL; 41256809eb4eSEric Schrock spa_load_l2cache(spa); 41266809eb4eSEric Schrock } 41276809eb4eSEric Schrock 41286809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 41296809eb4eSEric Schrock &nvroot) == 0); 41306809eb4eSEric Schrock if (error == 0) 41316809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 41326809eb4eSEric Schrock VDEV_ALLOC_SPARE); 41336809eb4eSEric Schrock if (error == 0) 41346809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 41356809eb4eSEric Schrock VDEV_ALLOC_L2CACHE); 41366809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 41376809eb4eSEric Schrock 41386809eb4eSEric Schrock if (props != NULL) 41396809eb4eSEric Schrock spa_configfile_set(spa, props, B_FALSE); 41406809eb4eSEric Schrock 41416809eb4eSEric Schrock if (error != 0 || (props && spa_writeable(spa) && 41426809eb4eSEric Schrock (error = spa_prop_set(spa, props)))) { 41436809eb4eSEric Schrock spa_unload(spa); 41446809eb4eSEric Schrock spa_deactivate(spa); 41456809eb4eSEric Schrock spa_remove(spa); 41466809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 41476809eb4eSEric Schrock return (error); 41486809eb4eSEric Schrock } 41496809eb4eSEric Schrock 4150955ef359SLin Ling spa_async_resume(spa); 4151955ef359SLin Ling 41526809eb4eSEric Schrock /* 41536809eb4eSEric Schrock * Override any spares and level 2 cache devices as specified by 41546809eb4eSEric Schrock * the user, as these may have correct device names/devids, etc. 41556809eb4eSEric Schrock */ 41566809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 41576809eb4eSEric Schrock &spares, &nspares) == 0) { 41586809eb4eSEric Schrock if (spa->spa_spares.sav_config) 41596809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_spares.sav_config, 41606809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 41616809eb4eSEric Schrock else 41626809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 41636809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 41646809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 41656809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 41666809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 41676809eb4eSEric Schrock spa_load_spares(spa); 41686809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 41696809eb4eSEric Schrock spa->spa_spares.sav_sync = B_TRUE; 41706809eb4eSEric Schrock } 41716809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 41726809eb4eSEric Schrock &l2cache, &nl2cache) == 0) { 41736809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) 41746809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 41756809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 41766809eb4eSEric Schrock else 41776809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 41786809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 41796809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 41806809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 41816809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 41826809eb4eSEric Schrock spa_load_l2cache(spa); 41836809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 41846809eb4eSEric Schrock spa->spa_l2cache.sav_sync = B_TRUE; 41856809eb4eSEric Schrock } 41866809eb4eSEric Schrock 4187b693757aSEric Schrock /* 4188b693757aSEric Schrock * Check for any removed devices. 4189b693757aSEric Schrock */ 4190b693757aSEric Schrock if (spa->spa_autoreplace) { 4191b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 4192b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 4193b693757aSEric Schrock } 4194b693757aSEric Schrock 41956809eb4eSEric Schrock if (spa_writeable(spa)) { 41966809eb4eSEric Schrock /* 41976809eb4eSEric Schrock * Update the config cache to include the newly-imported pool. 41986809eb4eSEric Schrock */ 4199bc758434SLin Ling spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 42006809eb4eSEric Schrock } 42016809eb4eSEric Schrock 4202573ca77eSGeorge Wilson /* 4203573ca77eSGeorge Wilson * It's possible that the pool was expanded while it was exported. 4204573ca77eSGeorge Wilson * We kick off an async task to handle this for us. 4205573ca77eSGeorge Wilson */ 4206573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4207573ca77eSGeorge Wilson 42084445fffbSMatthew Ahrens spa_history_log_version(spa, "import"); 42096809eb4eSEric Schrock 421014372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 421114372834SHans Rosenfeld 421214372834SHans Rosenfeld mutex_exit(&spa_namespace_lock); 421314372834SHans Rosenfeld 42146809eb4eSEric Schrock return (0); 4215c5904d13Seschrock } 4216c5904d13Seschrock 4217fa9e4066Sahrens nvlist_t * 4218fa9e4066Sahrens spa_tryimport(nvlist_t *tryconfig) 4219fa9e4066Sahrens { 4220fa9e4066Sahrens nvlist_t *config = NULL; 4221fa9e4066Sahrens char *poolname; 4222fa9e4066Sahrens spa_t *spa; 4223fa9e4066Sahrens uint64_t state; 42247b7154beSLin Ling int error; 4225fa9e4066Sahrens 4226fa9e4066Sahrens if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4227fa9e4066Sahrens return (NULL); 4228fa9e4066Sahrens 4229fa9e4066Sahrens if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4230fa9e4066Sahrens return (NULL); 4231fa9e4066Sahrens 4232fa9e4066Sahrens /* 42330373e76bSbonwick * Create and initialize the spa structure. 4234fa9e4066Sahrens */ 42350373e76bSbonwick mutex_enter(&spa_namespace_lock); 4236468c413aSTim Haley spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 42378ad4d6ddSJeff Bonwick spa_activate(spa, FREAD); 4238fa9e4066Sahrens 4239fa9e4066Sahrens /* 42400373e76bSbonwick * Pass off the heavy lifting to spa_load(). 4241ecc2d604Sbonwick * Pass TRUE for mosconfig because the user-supplied config 4242ecc2d604Sbonwick * is actually the one to trust when doing an import. 4243fa9e4066Sahrens */ 42441195e687SMark J Musante error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4245fa9e4066Sahrens 4246fa9e4066Sahrens /* 4247fa9e4066Sahrens * If 'tryconfig' was at least parsable, return the current config. 4248fa9e4066Sahrens */ 4249fa9e4066Sahrens if (spa->spa_root_vdev != NULL) { 4250fa9e4066Sahrens config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4251fa9e4066Sahrens VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4252fa9e4066Sahrens poolname) == 0); 4253fa9e4066Sahrens VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4254fa9e4066Sahrens state) == 0); 425595173954Sek VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 425695173954Sek spa->spa_uberblock.ub_timestamp) == 0); 4257ad135b5dSChristopher Siden VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4258ad135b5dSChristopher Siden spa->spa_load_info) == 0); 425999653d4eSeschrock 4260e7cbe64fSgw /* 4261e7cbe64fSgw * If the bootfs property exists on this pool then we 4262e7cbe64fSgw * copy it out so that external consumers can tell which 4263e7cbe64fSgw * pools are bootable. 4264e7cbe64fSgw */ 42657b7154beSLin Ling if ((!error || error == EEXIST) && spa->spa_bootfs) { 4266e7cbe64fSgw char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4267e7cbe64fSgw 4268e7cbe64fSgw /* 4269e7cbe64fSgw * We have to play games with the name since the 4270e7cbe64fSgw * pool was opened as TRYIMPORT_NAME. 4271e7cbe64fSgw */ 4272e14bb325SJeff Bonwick if (dsl_dsobj_to_dsname(spa_name(spa), 4273e7cbe64fSgw spa->spa_bootfs, tmpname) == 0) { 4274e7cbe64fSgw char *cp; 4275e7cbe64fSgw char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4276e7cbe64fSgw 4277e7cbe64fSgw cp = strchr(tmpname, '/'); 4278e7cbe64fSgw if (cp == NULL) { 4279e7cbe64fSgw (void) strlcpy(dsname, tmpname, 4280e7cbe64fSgw MAXPATHLEN); 4281e7cbe64fSgw } else { 4282e7cbe64fSgw (void) snprintf(dsname, MAXPATHLEN, 4283e7cbe64fSgw "%s/%s", poolname, ++cp); 4284e7cbe64fSgw } 4285e7cbe64fSgw VERIFY(nvlist_add_string(config, 4286e7cbe64fSgw ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4287e7cbe64fSgw kmem_free(dsname, MAXPATHLEN); 4288e7cbe64fSgw } 4289e7cbe64fSgw kmem_free(tmpname, MAXPATHLEN); 4290e7cbe64fSgw } 4291e7cbe64fSgw 429299653d4eSeschrock /* 4293fa94a07fSbrendan * Add the list of hot spares and level 2 cache devices. 429499653d4eSeschrock */ 42956809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 429699653d4eSeschrock spa_add_spares(spa, config); 4297fa94a07fSbrendan spa_add_l2cache(spa, config); 42986809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 4299fa9e4066Sahrens } 4300fa9e4066Sahrens 4301fa9e4066Sahrens spa_unload(spa); 4302fa9e4066Sahrens spa_deactivate(spa); 4303fa9e4066Sahrens spa_remove(spa); 4304fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4305fa9e4066Sahrens 4306fa9e4066Sahrens return (config); 4307fa9e4066Sahrens } 4308fa9e4066Sahrens 4309fa9e4066Sahrens /* 4310fa9e4066Sahrens * Pool export/destroy 4311fa9e4066Sahrens * 4312fa9e4066Sahrens * The act of destroying or exporting a pool is very simple. We make sure there 4313fa9e4066Sahrens * is no more pending I/O and any references to the pool are gone. Then, we 4314fa9e4066Sahrens * update the pool state and sync all the labels to disk, removing the 4315394ab0cbSGeorge Wilson * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4316394ab0cbSGeorge Wilson * we don't sync the labels or remove the configuration cache. 4317fa9e4066Sahrens */ 4318fa9e4066Sahrens static int 431989a89ebfSlling spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4320394ab0cbSGeorge Wilson boolean_t force, boolean_t hardforce) 4321fa9e4066Sahrens { 4322fa9e4066Sahrens spa_t *spa; 4323fa9e4066Sahrens 432444cd46caSbillm if (oldconfig) 432544cd46caSbillm *oldconfig = NULL; 432644cd46caSbillm 43278ad4d6ddSJeff Bonwick if (!(spa_mode_global & FWRITE)) 4328be6fd75aSMatthew Ahrens return (SET_ERROR(EROFS)); 4329fa9e4066Sahrens 4330fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4331fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 4332fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4333be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 4334fa9e4066Sahrens } 4335fa9e4066Sahrens 4336ea8dc4b6Seschrock /* 4337ea8dc4b6Seschrock * Put a hold on the pool, drop the namespace lock, stop async tasks, 4338ea8dc4b6Seschrock * reacquire the namespace lock, and see if we can export. 4339ea8dc4b6Seschrock */ 4340ea8dc4b6Seschrock spa_open_ref(spa, FTAG); 4341ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 4342ea8dc4b6Seschrock spa_async_suspend(spa); 4343ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 4344ea8dc4b6Seschrock spa_close(spa, FTAG); 4345ea8dc4b6Seschrock 4346fa9e4066Sahrens /* 4347fa9e4066Sahrens * The pool will be in core if it's openable, 4348fa9e4066Sahrens * in which case we can modify its state. 4349fa9e4066Sahrens */ 4350fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4351fa9e4066Sahrens /* 4352fa9e4066Sahrens * Objsets may be open only because they're dirty, so we 4353fa9e4066Sahrens * have to force it to sync before checking spa_refcnt. 4354fa9e4066Sahrens */ 4355fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, 0); 4356bc9014e6SJustin Gibbs spa_evicting_os_wait(spa); 4357fa9e4066Sahrens 4358ea8dc4b6Seschrock /* 4359ea8dc4b6Seschrock * A pool cannot be exported or destroyed if there are active 4360ea8dc4b6Seschrock * references. If we are resetting a pool, allow references by 4361ea8dc4b6Seschrock * fault injection handlers. 4362ea8dc4b6Seschrock */ 4363ea8dc4b6Seschrock if (!spa_refcount_zero(spa) || 4364ea8dc4b6Seschrock (spa->spa_inject_ref != 0 && 4365ea8dc4b6Seschrock new_state != POOL_STATE_UNINITIALIZED)) { 4366ea8dc4b6Seschrock spa_async_resume(spa); 4367fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4368be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 4369fa9e4066Sahrens } 4370fa9e4066Sahrens 437189a89ebfSlling /* 437289a89ebfSlling * A pool cannot be exported if it has an active shared spare. 437389a89ebfSlling * This is to prevent other pools stealing the active spare 437489a89ebfSlling * from an exported pool. At user's own will, such pool can 437589a89ebfSlling * be forcedly exported. 437689a89ebfSlling */ 437789a89ebfSlling if (!force && new_state == POOL_STATE_EXPORTED && 437889a89ebfSlling spa_has_active_shared_spare(spa)) { 437989a89ebfSlling spa_async_resume(spa); 438089a89ebfSlling mutex_exit(&spa_namespace_lock); 4381be6fd75aSMatthew Ahrens return (SET_ERROR(EXDEV)); 438289a89ebfSlling } 438389a89ebfSlling 4384fa9e4066Sahrens /* 4385fa9e4066Sahrens * We want this to be reflected on every label, 4386fa9e4066Sahrens * so mark them all dirty. spa_unload() will do the 4387fa9e4066Sahrens * final sync that pushes these changes out. 4388fa9e4066Sahrens */ 4389394ab0cbSGeorge Wilson if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4390e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4391ea8dc4b6Seschrock spa->spa_state = new_state; 43923f9d6ad7SLin Ling spa->spa_final_txg = spa_last_synced_txg(spa) + 43933f9d6ad7SLin Ling TXG_DEFER_SIZE + 1; 4394ea8dc4b6Seschrock vdev_config_dirty(spa->spa_root_vdev); 4395e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 4396ea8dc4b6Seschrock } 4397fa9e4066Sahrens } 4398fa9e4066Sahrens 43993d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 44003d7072f8Seschrock 4401fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4402fa9e4066Sahrens spa_unload(spa); 4403fa9e4066Sahrens spa_deactivate(spa); 4404fa9e4066Sahrens } 4405fa9e4066Sahrens 440644cd46caSbillm if (oldconfig && spa->spa_config) 440744cd46caSbillm VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 440844cd46caSbillm 4409ea8dc4b6Seschrock if (new_state != POOL_STATE_UNINITIALIZED) { 4410394ab0cbSGeorge Wilson if (!hardforce) 4411394ab0cbSGeorge Wilson spa_config_sync(spa, B_TRUE, B_TRUE); 4412ea8dc4b6Seschrock spa_remove(spa); 4413ea8dc4b6Seschrock } 4414fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4415fa9e4066Sahrens 4416fa9e4066Sahrens return (0); 4417fa9e4066Sahrens } 4418fa9e4066Sahrens 4419fa9e4066Sahrens /* 4420fa9e4066Sahrens * Destroy a storage pool. 4421fa9e4066Sahrens */ 4422fa9e4066Sahrens int 4423fa9e4066Sahrens spa_destroy(char *pool) 4424fa9e4066Sahrens { 4425394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4426394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 4427fa9e4066Sahrens } 4428fa9e4066Sahrens 4429fa9e4066Sahrens /* 4430fa9e4066Sahrens * Export a storage pool. 4431fa9e4066Sahrens */ 4432fa9e4066Sahrens int 4433394ab0cbSGeorge Wilson spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4434394ab0cbSGeorge Wilson boolean_t hardforce) 4435fa9e4066Sahrens { 4436394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4437394ab0cbSGeorge Wilson force, hardforce)); 4438fa9e4066Sahrens } 4439fa9e4066Sahrens 4440ea8dc4b6Seschrock /* 4441ea8dc4b6Seschrock * Similar to spa_export(), this unloads the spa_t without actually removing it 4442ea8dc4b6Seschrock * from the namespace in any way. 4443ea8dc4b6Seschrock */ 4444ea8dc4b6Seschrock int 4445ea8dc4b6Seschrock spa_reset(char *pool) 4446ea8dc4b6Seschrock { 444789a89ebfSlling return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4448394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 4449ea8dc4b6Seschrock } 4450ea8dc4b6Seschrock 4451fa9e4066Sahrens /* 4452fa9e4066Sahrens * ========================================================================== 4453fa9e4066Sahrens * Device manipulation 4454fa9e4066Sahrens * ========================================================================== 4455fa9e4066Sahrens */ 4456fa9e4066Sahrens 4457fa9e4066Sahrens /* 44588654d025Sperrin * Add a device to a storage pool. 4459fa9e4066Sahrens */ 4460fa9e4066Sahrens int 4461fa9e4066Sahrens spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4462fa9e4066Sahrens { 446388ecc943SGeorge Wilson uint64_t txg, id; 44648ad4d6ddSJeff Bonwick int error; 4465fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 44660e34b6a7Sbonwick vdev_t *vd, *tvd; 4467fa94a07fSbrendan nvlist_t **spares, **l2cache; 4468fa94a07fSbrendan uint_t nspares, nl2cache; 4469fa9e4066Sahrens 4470f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4471f9af39baSGeorge Wilson 4472fa9e4066Sahrens txg = spa_vdev_enter(spa); 4473fa9e4066Sahrens 447499653d4eSeschrock if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 447599653d4eSeschrock VDEV_ALLOC_ADD)) != 0) 447699653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, error)); 4477fa9e4066Sahrens 4478e14bb325SJeff Bonwick spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 447999653d4eSeschrock 4480fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4481fa94a07fSbrendan &nspares) != 0) 448299653d4eSeschrock nspares = 0; 448399653d4eSeschrock 4484fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4485fa94a07fSbrendan &nl2cache) != 0) 4486fa94a07fSbrendan nl2cache = 0; 4487fa94a07fSbrendan 4488e14bb325SJeff Bonwick if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4489fa9e4066Sahrens return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4490fa9e4066Sahrens 4491e14bb325SJeff Bonwick if (vd->vdev_children != 0 && 4492e14bb325SJeff Bonwick (error = vdev_create(vd, txg, B_FALSE)) != 0) 4493e14bb325SJeff Bonwick return (spa_vdev_exit(spa, vd, txg, error)); 449499653d4eSeschrock 449539c23413Seschrock /* 4496fa94a07fSbrendan * We must validate the spares and l2cache devices after checking the 4497fa94a07fSbrendan * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 449839c23413Seschrock */ 4499e14bb325SJeff Bonwick if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 450039c23413Seschrock return (spa_vdev_exit(spa, vd, txg, error)); 450139c23413Seschrock 450239c23413Seschrock /* 450339c23413Seschrock * Transfer each new top-level vdev from vd to rvd. 450439c23413Seschrock */ 45058ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 450688ecc943SGeorge Wilson 450788ecc943SGeorge Wilson /* 450888ecc943SGeorge Wilson * Set the vdev id to the first hole, if one exists. 450988ecc943SGeorge Wilson */ 451088ecc943SGeorge Wilson for (id = 0; id < rvd->vdev_children; id++) { 451188ecc943SGeorge Wilson if (rvd->vdev_child[id]->vdev_ishole) { 451288ecc943SGeorge Wilson vdev_free(rvd->vdev_child[id]); 451388ecc943SGeorge Wilson break; 451488ecc943SGeorge Wilson } 451588ecc943SGeorge Wilson } 451639c23413Seschrock tvd = vd->vdev_child[c]; 451739c23413Seschrock vdev_remove_child(vd, tvd); 451888ecc943SGeorge Wilson tvd->vdev_id = id; 451939c23413Seschrock vdev_add_child(rvd, tvd); 452039c23413Seschrock vdev_config_dirty(tvd); 452139c23413Seschrock } 452239c23413Seschrock 452399653d4eSeschrock if (nspares != 0) { 4524fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4525fa94a07fSbrendan ZPOOL_CONFIG_SPARES); 452699653d4eSeschrock spa_load_spares(spa); 4527fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 4528fa94a07fSbrendan } 4529fa94a07fSbrendan 4530fa94a07fSbrendan if (nl2cache != 0) { 4531fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4532fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE); 4533fa94a07fSbrendan spa_load_l2cache(spa); 4534fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 4535fa9e4066Sahrens } 4536fa9e4066Sahrens 4537fa9e4066Sahrens /* 45380e34b6a7Sbonwick * We have to be careful when adding new vdevs to an existing pool. 45390e34b6a7Sbonwick * If other threads start allocating from these vdevs before we 45400e34b6a7Sbonwick * sync the config cache, and we lose power, then upon reboot we may 45410e34b6a7Sbonwick * fail to open the pool because there are DVAs that the config cache 45420e34b6a7Sbonwick * can't translate. Therefore, we first add the vdevs without 45430e34b6a7Sbonwick * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 45440373e76bSbonwick * and then let spa_config_update() initialize the new metaslabs. 45450e34b6a7Sbonwick * 45460e34b6a7Sbonwick * spa_load() checks for added-but-not-initialized vdevs, so that 45470e34b6a7Sbonwick * if we lose power at any point in this sequence, the remaining 45480e34b6a7Sbonwick * steps will be completed the next time we load the pool. 45490e34b6a7Sbonwick */ 45500373e76bSbonwick (void) spa_vdev_exit(spa, vd, txg, 0); 45510e34b6a7Sbonwick 45520373e76bSbonwick mutex_enter(&spa_namespace_lock); 45530373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 455414372834SHans Rosenfeld spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 45550373e76bSbonwick mutex_exit(&spa_namespace_lock); 4556fa9e4066Sahrens 45570373e76bSbonwick return (0); 4558fa9e4066Sahrens } 4559fa9e4066Sahrens 4560fa9e4066Sahrens /* 4561fa9e4066Sahrens * Attach a device to a mirror. The arguments are the path to any device 4562fa9e4066Sahrens * in the mirror, and the nvroot for the new device. If the path specifies 4563fa9e4066Sahrens * a device that is not mirrored, we automatically insert the mirror vdev. 4564fa9e4066Sahrens * 4565fa9e4066Sahrens * If 'replacing' is specified, the new device is intended to replace the 4566fa9e4066Sahrens * existing device; in this case the two devices are made into their own 45673d7072f8Seschrock * mirror using the 'replacing' vdev, which is functionally identical to 4568fa9e4066Sahrens * the mirror vdev (it actually reuses all the same ops) but has a few 4569fa9e4066Sahrens * extra rules: you can't attach to it after it's been created, and upon 4570fa9e4066Sahrens * completion of resilvering, the first disk (the one being replaced) 4571fa9e4066Sahrens * is automatically detached. 4572fa9e4066Sahrens */ 4573fa9e4066Sahrens int 4574ea8dc4b6Seschrock spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4575fa9e4066Sahrens { 45763f9d6ad7SLin Ling uint64_t txg, dtl_max_txg; 4577fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4578fa9e4066Sahrens vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 457999653d4eSeschrock vdev_ops_t *pvops; 45809b3f6b42SEric Kustarz char *oldvdpath, *newvdpath; 45819b3f6b42SEric Kustarz int newvd_isspare; 45829b3f6b42SEric Kustarz int error; 4583fa9e4066Sahrens 4584f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4585f9af39baSGeorge Wilson 4586fa9e4066Sahrens txg = spa_vdev_enter(spa); 4587fa9e4066Sahrens 4588c5904d13Seschrock oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4589fa9e4066Sahrens 4590fa9e4066Sahrens if (oldvd == NULL) 4591fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4592fa9e4066Sahrens 45930e34b6a7Sbonwick if (!oldvd->vdev_ops->vdev_op_leaf) 45940e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 45950e34b6a7Sbonwick 4596fa9e4066Sahrens pvd = oldvd->vdev_parent; 4597fa9e4066Sahrens 459899653d4eSeschrock if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4599cd0837ccSGeorge Wilson VDEV_ALLOC_ATTACH)) != 0) 46003d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 46013d7072f8Seschrock 46023d7072f8Seschrock if (newrootvd->vdev_children != 1) 4603fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4604fa9e4066Sahrens 4605fa9e4066Sahrens newvd = newrootvd->vdev_child[0]; 4606fa9e4066Sahrens 4607fa9e4066Sahrens if (!newvd->vdev_ops->vdev_op_leaf) 4608fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4609fa9e4066Sahrens 461099653d4eSeschrock if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4611fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, error)); 4612fa9e4066Sahrens 46138654d025Sperrin /* 46148654d025Sperrin * Spares can't replace logs 46158654d025Sperrin */ 4616ee0eb9f2SEric Schrock if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 46178654d025Sperrin return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 46188654d025Sperrin 461999653d4eSeschrock if (!replacing) { 462099653d4eSeschrock /* 462199653d4eSeschrock * For attach, the only allowable parent is a mirror or the root 462299653d4eSeschrock * vdev. 462399653d4eSeschrock */ 462499653d4eSeschrock if (pvd->vdev_ops != &vdev_mirror_ops && 462599653d4eSeschrock pvd->vdev_ops != &vdev_root_ops) 462699653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 462799653d4eSeschrock 462899653d4eSeschrock pvops = &vdev_mirror_ops; 462999653d4eSeschrock } else { 463099653d4eSeschrock /* 463199653d4eSeschrock * Active hot spares can only be replaced by inactive hot 463299653d4eSeschrock * spares. 463399653d4eSeschrock */ 463499653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 4635cb04b873SMark J Musante oldvd->vdev_isspare && 463699653d4eSeschrock !spa_has_spare(spa, newvd->vdev_guid)) 463799653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 463899653d4eSeschrock 463999653d4eSeschrock /* 464099653d4eSeschrock * If the source is a hot spare, and the parent isn't already a 464199653d4eSeschrock * spare, then we want to create a new hot spare. Otherwise, we 464239c23413Seschrock * want to create a replacing vdev. The user is not allowed to 464339c23413Seschrock * attach to a spared vdev child unless the 'isspare' state is 464439c23413Seschrock * the same (spare replaces spare, non-spare replaces 464539c23413Seschrock * non-spare). 464699653d4eSeschrock */ 4647cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && 4648cb04b873SMark J Musante spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 464999653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4650cb04b873SMark J Musante } else if (pvd->vdev_ops == &vdev_spare_ops && 4651cb04b873SMark J Musante newvd->vdev_isspare != oldvd->vdev_isspare) { 465239c23413Seschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4653cb04b873SMark J Musante } 4654cb04b873SMark J Musante 4655cb04b873SMark J Musante if (newvd->vdev_isspare) 465699653d4eSeschrock pvops = &vdev_spare_ops; 465799653d4eSeschrock else 465899653d4eSeschrock pvops = &vdev_replacing_ops; 465999653d4eSeschrock } 466099653d4eSeschrock 46612a79c5feSlling /* 4662573ca77eSGeorge Wilson * Make sure the new device is big enough. 46632a79c5feSlling */ 4664573ca77eSGeorge Wilson if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4665fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4666fa9e4066Sahrens 4667ecc2d604Sbonwick /* 4668ecc2d604Sbonwick * The new device cannot have a higher alignment requirement 4669ecc2d604Sbonwick * than the top-level vdev. 4670ecc2d604Sbonwick */ 4671ecc2d604Sbonwick if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4672fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4673fa9e4066Sahrens 4674fa9e4066Sahrens /* 4675fa9e4066Sahrens * If this is an in-place replacement, update oldvd's path and devid 4676fa9e4066Sahrens * to make it distinguishable from newvd, and unopenable from now on. 4677fa9e4066Sahrens */ 4678fa9e4066Sahrens if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4679fa9e4066Sahrens spa_strfree(oldvd->vdev_path); 4680fa9e4066Sahrens oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4681fa9e4066Sahrens KM_SLEEP); 4682fa9e4066Sahrens (void) sprintf(oldvd->vdev_path, "%s/%s", 4683fa9e4066Sahrens newvd->vdev_path, "old"); 4684fa9e4066Sahrens if (oldvd->vdev_devid != NULL) { 4685fa9e4066Sahrens spa_strfree(oldvd->vdev_devid); 4686fa9e4066Sahrens oldvd->vdev_devid = NULL; 4687fa9e4066Sahrens } 4688fa9e4066Sahrens } 4689fa9e4066Sahrens 4690cb04b873SMark J Musante /* mark the device being resilvered */ 4691b4952e17SGeorge Wilson newvd->vdev_resilver_txg = txg; 4692cb04b873SMark J Musante 4693fa9e4066Sahrens /* 469499653d4eSeschrock * If the parent is not a mirror, or if we're replacing, insert the new 469599653d4eSeschrock * mirror/replacing/spare vdev above oldvd. 4696fa9e4066Sahrens */ 4697fa9e4066Sahrens if (pvd->vdev_ops != pvops) 4698fa9e4066Sahrens pvd = vdev_add_parent(oldvd, pvops); 4699fa9e4066Sahrens 4700fa9e4066Sahrens ASSERT(pvd->vdev_top->vdev_parent == rvd); 4701fa9e4066Sahrens ASSERT(pvd->vdev_ops == pvops); 4702fa9e4066Sahrens ASSERT(oldvd->vdev_parent == pvd); 4703fa9e4066Sahrens 4704fa9e4066Sahrens /* 4705fa9e4066Sahrens * Extract the new device from its root and add it to pvd. 4706fa9e4066Sahrens */ 4707fa9e4066Sahrens vdev_remove_child(newrootvd, newvd); 4708fa9e4066Sahrens newvd->vdev_id = pvd->vdev_children; 470988ecc943SGeorge Wilson newvd->vdev_crtxg = oldvd->vdev_crtxg; 4710fa9e4066Sahrens vdev_add_child(pvd, newvd); 4711fa9e4066Sahrens 4712fa9e4066Sahrens tvd = newvd->vdev_top; 4713fa9e4066Sahrens ASSERT(pvd->vdev_top == tvd); 4714fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 4715fa9e4066Sahrens 4716fa9e4066Sahrens vdev_config_dirty(tvd); 4717fa9e4066Sahrens 4718fa9e4066Sahrens /* 47193f9d6ad7SLin Ling * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 47203f9d6ad7SLin Ling * for any dmu_sync-ed blocks. It will propagate upward when 47213f9d6ad7SLin Ling * spa_vdev_exit() calls vdev_dtl_reassess(). 4722fa9e4066Sahrens */ 47233f9d6ad7SLin Ling dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4724fa9e4066Sahrens 47253f9d6ad7SLin Ling vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 47263f9d6ad7SLin Ling dtl_max_txg - TXG_INITIAL); 4727fa9e4066Sahrens 47286809eb4eSEric Schrock if (newvd->vdev_isspare) { 472939c23413Seschrock spa_spare_activate(newvd); 47306809eb4eSEric Schrock spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 47316809eb4eSEric Schrock } 47326809eb4eSEric Schrock 4733e14bb325SJeff Bonwick oldvdpath = spa_strdup(oldvd->vdev_path); 4734e14bb325SJeff Bonwick newvdpath = spa_strdup(newvd->vdev_path); 47359b3f6b42SEric Kustarz newvd_isspare = newvd->vdev_isspare; 4736ea8dc4b6Seschrock 4737fa9e4066Sahrens /* 4738fa9e4066Sahrens * Mark newvd's DTL dirty in this txg. 4739fa9e4066Sahrens */ 4740ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, newvd, txg); 4741fa9e4066Sahrens 47423f9d6ad7SLin Ling /* 47430713e232SGeorge Wilson * Schedule the resilver to restart in the future. We do this to 47440713e232SGeorge Wilson * ensure that dmu_sync-ed blocks have been stitched into the 47450713e232SGeorge Wilson * respective datasets. 47463f9d6ad7SLin Ling */ 47473f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 47483f9d6ad7SLin Ling 474914372834SHans Rosenfeld if (spa->spa_bootfs) 475014372834SHans Rosenfeld spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 475114372834SHans Rosenfeld 475214372834SHans Rosenfeld spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 475314372834SHans Rosenfeld 47543f9d6ad7SLin Ling /* 47553f9d6ad7SLin Ling * Commit the config 47563f9d6ad7SLin Ling */ 47573f9d6ad7SLin Ling (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4758fa9e4066Sahrens 47594445fffbSMatthew Ahrens spa_history_log_internal(spa, "vdev attach", NULL, 47603f9d6ad7SLin Ling "%s vdev=%s %s vdev=%s", 4761c8e1f6d2SMark J Musante replacing && newvd_isspare ? "spare in" : 4762c8e1f6d2SMark J Musante replacing ? "replace" : "attach", newvdpath, 4763c8e1f6d2SMark J Musante replacing ? "for" : "to", oldvdpath); 47649b3f6b42SEric Kustarz 47659b3f6b42SEric Kustarz spa_strfree(oldvdpath); 47669b3f6b42SEric Kustarz spa_strfree(newvdpath); 47679b3f6b42SEric Kustarz 4768fa9e4066Sahrens return (0); 4769fa9e4066Sahrens } 4770fa9e4066Sahrens 4771fa9e4066Sahrens /* 4772fa9e4066Sahrens * Detach a device from a mirror or replacing vdev. 4773f7170741SWill Andrews * 4774fa9e4066Sahrens * If 'replace_done' is specified, only detach if the parent 4775fa9e4066Sahrens * is a replacing vdev. 4776fa9e4066Sahrens */ 4777fa9e4066Sahrens int 47788ad4d6ddSJeff Bonwick spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4779fa9e4066Sahrens { 4780fa9e4066Sahrens uint64_t txg; 47818ad4d6ddSJeff Bonwick int error; 4782fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4783fa9e4066Sahrens vdev_t *vd, *pvd, *cvd, *tvd; 478499653d4eSeschrock boolean_t unspare = B_FALSE; 4785d5285caeSGeorge Wilson uint64_t unspare_guid = 0; 47861195e687SMark J Musante char *vdpath; 4787fa9e4066Sahrens 4788f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4789f9af39baSGeorge Wilson 4790fa9e4066Sahrens txg = spa_vdev_enter(spa); 4791fa9e4066Sahrens 4792c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4793fa9e4066Sahrens 4794fa9e4066Sahrens if (vd == NULL) 4795fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4796fa9e4066Sahrens 47970e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 47980e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 47990e34b6a7Sbonwick 4800fa9e4066Sahrens pvd = vd->vdev_parent; 4801fa9e4066Sahrens 48028ad4d6ddSJeff Bonwick /* 48038ad4d6ddSJeff Bonwick * If the parent/child relationship is not as expected, don't do it. 48048ad4d6ddSJeff Bonwick * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 48058ad4d6ddSJeff Bonwick * vdev that's replacing B with C. The user's intent in replacing 48068ad4d6ddSJeff Bonwick * is to go from M(A,B) to M(A,C). If the user decides to cancel 48078ad4d6ddSJeff Bonwick * the replace by detaching C, the expected behavior is to end up 48088ad4d6ddSJeff Bonwick * M(A,B). But suppose that right after deciding to detach C, 48098ad4d6ddSJeff Bonwick * the replacement of B completes. We would have M(A,C), and then 48108ad4d6ddSJeff Bonwick * ask to detach C, which would leave us with just A -- not what 48118ad4d6ddSJeff Bonwick * the user wanted. To prevent this, we make sure that the 48128ad4d6ddSJeff Bonwick * parent/child relationship hasn't changed -- in this example, 48138ad4d6ddSJeff Bonwick * that C's parent is still the replacing vdev R. 48148ad4d6ddSJeff Bonwick */ 48158ad4d6ddSJeff Bonwick if (pvd->vdev_guid != pguid && pguid != 0) 48168ad4d6ddSJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 48178ad4d6ddSJeff Bonwick 4818fa9e4066Sahrens /* 4819cb04b873SMark J Musante * Only 'replacing' or 'spare' vdevs can be replaced. 482099653d4eSeschrock */ 4821cb04b873SMark J Musante if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4822cb04b873SMark J Musante pvd->vdev_ops != &vdev_spare_ops) 4823cb04b873SMark J Musante return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 482499653d4eSeschrock 482599653d4eSeschrock ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4826e7437265Sahrens spa_version(spa) >= SPA_VERSION_SPARES); 4827fa9e4066Sahrens 4828fa9e4066Sahrens /* 482999653d4eSeschrock * Only mirror, replacing, and spare vdevs support detach. 4830fa9e4066Sahrens */ 4831fa9e4066Sahrens if (pvd->vdev_ops != &vdev_replacing_ops && 483299653d4eSeschrock pvd->vdev_ops != &vdev_mirror_ops && 483399653d4eSeschrock pvd->vdev_ops != &vdev_spare_ops) 4834fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4835fa9e4066Sahrens 4836fa9e4066Sahrens /* 48378ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 48388ad4d6ddSJeff Bonwick * we cannot safely detach it. 4839fa9e4066Sahrens */ 48408ad4d6ddSJeff Bonwick if (vdev_dtl_required(vd)) 4841fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4842fa9e4066Sahrens 48438ad4d6ddSJeff Bonwick ASSERT(pvd->vdev_children >= 2); 4844fa9e4066Sahrens 4845bf82a41bSeschrock /* 4846bf82a41bSeschrock * If we are detaching the second disk from a replacing vdev, then 4847bf82a41bSeschrock * check to see if we changed the original vdev's path to have "/old" 4848bf82a41bSeschrock * at the end in spa_vdev_attach(). If so, undo that change now. 4849bf82a41bSeschrock */ 4850cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4851cb04b873SMark J Musante vd->vdev_path != NULL) { 4852cb04b873SMark J Musante size_t len = strlen(vd->vdev_path); 4853cb04b873SMark J Musante 4854cb04b873SMark J Musante for (int c = 0; c < pvd->vdev_children; c++) { 4855cb04b873SMark J Musante cvd = pvd->vdev_child[c]; 4856cb04b873SMark J Musante 4857cb04b873SMark J Musante if (cvd == vd || cvd->vdev_path == NULL) 4858cb04b873SMark J Musante continue; 4859cb04b873SMark J Musante 4860cb04b873SMark J Musante if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4861cb04b873SMark J Musante strcmp(cvd->vdev_path + len, "/old") == 0) { 4862cb04b873SMark J Musante spa_strfree(cvd->vdev_path); 4863cb04b873SMark J Musante cvd->vdev_path = spa_strdup(vd->vdev_path); 4864cb04b873SMark J Musante break; 4865cb04b873SMark J Musante } 4866bf82a41bSeschrock } 4867bf82a41bSeschrock } 4868bf82a41bSeschrock 486999653d4eSeschrock /* 487099653d4eSeschrock * If we are detaching the original disk from a spare, then it implies 487199653d4eSeschrock * that the spare should become a real disk, and be removed from the 487299653d4eSeschrock * active spare list for the pool. 487399653d4eSeschrock */ 487499653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 4875cb04b873SMark J Musante vd->vdev_id == 0 && 4876cb04b873SMark J Musante pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 487799653d4eSeschrock unspare = B_TRUE; 487899653d4eSeschrock 4879fa9e4066Sahrens /* 4880fa9e4066Sahrens * Erase the disk labels so the disk can be used for other things. 4881fa9e4066Sahrens * This must be done after all other error cases are handled, 4882fa9e4066Sahrens * but before we disembowel vd (so we can still do I/O to it). 4883fa9e4066Sahrens * But if we can't do it, don't treat the error as fatal -- 4884fa9e4066Sahrens * it may be that the unwritability of the disk is the reason 4885fa9e4066Sahrens * it's being detached! 4886fa9e4066Sahrens */ 488739c23413Seschrock error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4888fa9e4066Sahrens 4889fa9e4066Sahrens /* 4890fa9e4066Sahrens * Remove vd from its parent and compact the parent's children. 4891fa9e4066Sahrens */ 4892fa9e4066Sahrens vdev_remove_child(pvd, vd); 4893fa9e4066Sahrens vdev_compact_children(pvd); 4894fa9e4066Sahrens 4895fa9e4066Sahrens /* 4896fa9e4066Sahrens * Remember one of the remaining children so we can get tvd below. 4897fa9e4066Sahrens */ 4898cb04b873SMark J Musante cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4899fa9e4066Sahrens 490099653d4eSeschrock /* 490199653d4eSeschrock * If we need to remove the remaining child from the list of hot spares, 49028ad4d6ddSJeff Bonwick * do it now, marking the vdev as no longer a spare in the process. 49038ad4d6ddSJeff Bonwick * We must do this before vdev_remove_parent(), because that can 49048ad4d6ddSJeff Bonwick * change the GUID if it creates a new toplevel GUID. For a similar 49058ad4d6ddSJeff Bonwick * reason, we must remove the spare now, in the same txg as the detach; 49068ad4d6ddSJeff Bonwick * otherwise someone could attach a new sibling, change the GUID, and 49078ad4d6ddSJeff Bonwick * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 490899653d4eSeschrock */ 490999653d4eSeschrock if (unspare) { 491099653d4eSeschrock ASSERT(cvd->vdev_isspare); 491139c23413Seschrock spa_spare_remove(cvd); 491299653d4eSeschrock unspare_guid = cvd->vdev_guid; 49138ad4d6ddSJeff Bonwick (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4914cb04b873SMark J Musante cvd->vdev_unspare = B_TRUE; 491599653d4eSeschrock } 491699653d4eSeschrock 4917fa9e4066Sahrens /* 4918fa9e4066Sahrens * If the parent mirror/replacing vdev only has one child, 4919fa9e4066Sahrens * the parent is no longer needed. Remove it from the tree. 4920fa9e4066Sahrens */ 4921cb04b873SMark J Musante if (pvd->vdev_children == 1) { 4922cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_spare_ops) 4923cb04b873SMark J Musante cvd->vdev_unspare = B_FALSE; 4924fa9e4066Sahrens vdev_remove_parent(cvd); 4925cb04b873SMark J Musante } 4926cb04b873SMark J Musante 4927fa9e4066Sahrens 4928fa9e4066Sahrens /* 4929fa9e4066Sahrens * We don't set tvd until now because the parent we just removed 4930fa9e4066Sahrens * may have been the previous top-level vdev. 4931fa9e4066Sahrens */ 4932fa9e4066Sahrens tvd = cvd->vdev_top; 4933fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 4934fa9e4066Sahrens 4935fa9e4066Sahrens /* 493639c23413Seschrock * Reevaluate the parent vdev state. 4937fa9e4066Sahrens */ 49383d7072f8Seschrock vdev_propagate_state(cvd); 4939fa9e4066Sahrens 4940fa9e4066Sahrens /* 4941573ca77eSGeorge Wilson * If the 'autoexpand' property is set on the pool then automatically 4942573ca77eSGeorge Wilson * try to expand the size of the pool. For example if the device we 4943573ca77eSGeorge Wilson * just detached was smaller than the others, it may be possible to 4944573ca77eSGeorge Wilson * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4945573ca77eSGeorge Wilson * first so that we can obtain the updated sizes of the leaf vdevs. 4946fa9e4066Sahrens */ 4947573ca77eSGeorge Wilson if (spa->spa_autoexpand) { 4948573ca77eSGeorge Wilson vdev_reopen(tvd); 4949573ca77eSGeorge Wilson vdev_expand(tvd, txg); 4950573ca77eSGeorge Wilson } 4951fa9e4066Sahrens 4952fa9e4066Sahrens vdev_config_dirty(tvd); 4953fa9e4066Sahrens 4954fa9e4066Sahrens /* 495539c23413Seschrock * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 495639c23413Seschrock * vd->vdev_detached is set and free vd's DTL object in syncing context. 495739c23413Seschrock * But first make sure we're not on any *other* txg's DTL list, to 495839c23413Seschrock * prevent vd from being accessed after it's freed. 4959fa9e4066Sahrens */ 49601195e687SMark J Musante vdpath = spa_strdup(vd->vdev_path); 49618ad4d6ddSJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 4962fa9e4066Sahrens (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4963ecc2d604Sbonwick vd->vdev_detached = B_TRUE; 4964ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, vd, txg); 4965fa9e4066Sahrens 49663d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 49673d7072f8Seschrock 4968cb04b873SMark J Musante /* hang on to the spa before we release the lock */ 4969cb04b873SMark J Musante spa_open_ref(spa, FTAG); 4970cb04b873SMark J Musante 497199653d4eSeschrock error = spa_vdev_exit(spa, vd, txg, 0); 497299653d4eSeschrock 49734445fffbSMatthew Ahrens spa_history_log_internal(spa, "detach", NULL, 49741195e687SMark J Musante "vdev=%s", vdpath); 49751195e687SMark J Musante spa_strfree(vdpath); 49761195e687SMark J Musante 497799653d4eSeschrock /* 497839c23413Seschrock * If this was the removal of the original device in a hot spare vdev, 497939c23413Seschrock * then we want to go through and remove the device from the hot spare 498039c23413Seschrock * list of every other pool. 498199653d4eSeschrock */ 498299653d4eSeschrock if (unspare) { 4983cb04b873SMark J Musante spa_t *altspa = NULL; 4984cb04b873SMark J Musante 498599653d4eSeschrock mutex_enter(&spa_namespace_lock); 4986cb04b873SMark J Musante while ((altspa = spa_next(altspa)) != NULL) { 4987cb04b873SMark J Musante if (altspa->spa_state != POOL_STATE_ACTIVE || 4988cb04b873SMark J Musante altspa == spa) 498999653d4eSeschrock continue; 4990cb04b873SMark J Musante 4991cb04b873SMark J Musante spa_open_ref(altspa, FTAG); 49929af0a4dfSJeff Bonwick mutex_exit(&spa_namespace_lock); 4993cb04b873SMark J Musante (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 49949af0a4dfSJeff Bonwick mutex_enter(&spa_namespace_lock); 4995cb04b873SMark J Musante spa_close(altspa, FTAG); 499699653d4eSeschrock } 499799653d4eSeschrock mutex_exit(&spa_namespace_lock); 4998cb04b873SMark J Musante 4999cb04b873SMark J Musante /* search the rest of the vdevs for spares to remove */ 5000cb04b873SMark J Musante spa_vdev_resilver_done(spa); 500199653d4eSeschrock } 500299653d4eSeschrock 5003cb04b873SMark J Musante /* all done with the spa; OK to release */ 5004cb04b873SMark J Musante mutex_enter(&spa_namespace_lock); 5005cb04b873SMark J Musante spa_close(spa, FTAG); 5006cb04b873SMark J Musante mutex_exit(&spa_namespace_lock); 5007cb04b873SMark J Musante 500899653d4eSeschrock return (error); 500999653d4eSeschrock } 501099653d4eSeschrock 50111195e687SMark J Musante /* 50121195e687SMark J Musante * Split a set of devices from their mirrors, and create a new pool from them. 50131195e687SMark J Musante */ 50141195e687SMark J Musante int 50151195e687SMark J Musante spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 50161195e687SMark J Musante nvlist_t *props, boolean_t exp) 50171195e687SMark J Musante { 50181195e687SMark J Musante int error = 0; 50191195e687SMark J Musante uint64_t txg, *glist; 50201195e687SMark J Musante spa_t *newspa; 50211195e687SMark J Musante uint_t c, children, lastlog; 50221195e687SMark J Musante nvlist_t **child, *nvl, *tmp; 50231195e687SMark J Musante dmu_tx_t *tx; 50241195e687SMark J Musante char *altroot = NULL; 50251195e687SMark J Musante vdev_t *rvd, **vml = NULL; /* vdev modify list */ 50261195e687SMark J Musante boolean_t activate_slog; 50271195e687SMark J Musante 5028f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 50291195e687SMark J Musante 50301195e687SMark J Musante txg = spa_vdev_enter(spa); 50311195e687SMark J Musante 50321195e687SMark J Musante /* clear the log and flush everything up to now */ 50331195e687SMark J Musante activate_slog = spa_passivate_log(spa); 50341195e687SMark J Musante (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 50351195e687SMark J Musante error = spa_offline_log(spa); 50361195e687SMark J Musante txg = spa_vdev_config_enter(spa); 50371195e687SMark J Musante 50381195e687SMark J Musante if (activate_slog) 50391195e687SMark J Musante spa_activate_log(spa); 50401195e687SMark J Musante 50411195e687SMark J Musante if (error != 0) 50421195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 50431195e687SMark J Musante 50441195e687SMark J Musante /* check new spa name before going any further */ 50451195e687SMark J Musante if (spa_lookup(newname) != NULL) 50461195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 50471195e687SMark J Musante 50481195e687SMark J Musante /* 50491195e687SMark J Musante * scan through all the children to ensure they're all mirrors 50501195e687SMark J Musante */ 50511195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 50521195e687SMark J Musante nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 50531195e687SMark J Musante &children) != 0) 50541195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 50551195e687SMark J Musante 50561195e687SMark J Musante /* first, check to ensure we've got the right child count */ 50571195e687SMark J Musante rvd = spa->spa_root_vdev; 50581195e687SMark J Musante lastlog = 0; 50591195e687SMark J Musante for (c = 0; c < rvd->vdev_children; c++) { 50601195e687SMark J Musante vdev_t *vd = rvd->vdev_child[c]; 50611195e687SMark J Musante 50621195e687SMark J Musante /* don't count the holes & logs as children */ 50631195e687SMark J Musante if (vd->vdev_islog || vd->vdev_ishole) { 50641195e687SMark J Musante if (lastlog == 0) 50651195e687SMark J Musante lastlog = c; 50661195e687SMark J Musante continue; 50671195e687SMark J Musante } 50681195e687SMark J Musante 50691195e687SMark J Musante lastlog = 0; 50701195e687SMark J Musante } 50711195e687SMark J Musante if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 50721195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 50731195e687SMark J Musante 50741195e687SMark J Musante /* next, ensure no spare or cache devices are part of the split */ 50751195e687SMark J Musante if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 50761195e687SMark J Musante nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 50771195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 50781195e687SMark J Musante 50791195e687SMark J Musante vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 50801195e687SMark J Musante glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 50811195e687SMark J Musante 50821195e687SMark J Musante /* then, loop over each vdev and validate it */ 50831195e687SMark J Musante for (c = 0; c < children; c++) { 50841195e687SMark J Musante uint64_t is_hole = 0; 50851195e687SMark J Musante 50861195e687SMark J Musante (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 50871195e687SMark J Musante &is_hole); 50881195e687SMark J Musante 50891195e687SMark J Musante if (is_hole != 0) { 50901195e687SMark J Musante if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 50911195e687SMark J Musante spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 50921195e687SMark J Musante continue; 50931195e687SMark J Musante } else { 5094be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 50951195e687SMark J Musante break; 50961195e687SMark J Musante } 50971195e687SMark J Musante } 50981195e687SMark J Musante 50991195e687SMark J Musante /* which disk is going to be split? */ 51001195e687SMark J Musante if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 51011195e687SMark J Musante &glist[c]) != 0) { 5102be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 51031195e687SMark J Musante break; 51041195e687SMark J Musante } 51051195e687SMark J Musante 51061195e687SMark J Musante /* look it up in the spa */ 51071195e687SMark J Musante vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 51081195e687SMark J Musante if (vml[c] == NULL) { 5109be6fd75aSMatthew Ahrens error = SET_ERROR(ENODEV); 51101195e687SMark J Musante break; 51111195e687SMark J Musante } 51121195e687SMark J Musante 51131195e687SMark J Musante /* make sure there's nothing stopping the split */ 51141195e687SMark J Musante if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 51151195e687SMark J Musante vml[c]->vdev_islog || 51161195e687SMark J Musante vml[c]->vdev_ishole || 51171195e687SMark J Musante vml[c]->vdev_isspare || 51181195e687SMark J Musante vml[c]->vdev_isl2cache || 51191195e687SMark J Musante !vdev_writeable(vml[c]) || 5120d41c4376SMark J Musante vml[c]->vdev_children != 0 || 51211195e687SMark J Musante vml[c]->vdev_state != VDEV_STATE_HEALTHY || 51221195e687SMark J Musante c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5123be6fd75aSMatthew Ahrens error = SET_ERROR(EINVAL); 51241195e687SMark J Musante break; 51251195e687SMark J Musante } 51261195e687SMark J Musante 51271195e687SMark J Musante if (vdev_dtl_required(vml[c])) { 5128be6fd75aSMatthew Ahrens error = SET_ERROR(EBUSY); 51291195e687SMark J Musante break; 51301195e687SMark J Musante } 51311195e687SMark J Musante 51321195e687SMark J Musante /* we need certain info from the top level */ 51331195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 51341195e687SMark J Musante vml[c]->vdev_top->vdev_ms_array) == 0); 51351195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 51361195e687SMark J Musante vml[c]->vdev_top->vdev_ms_shift) == 0); 51371195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 51381195e687SMark J Musante vml[c]->vdev_top->vdev_asize) == 0); 51391195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 51401195e687SMark J Musante vml[c]->vdev_top->vdev_ashift) == 0); 5141215198a6SJoe Stein 5142215198a6SJoe Stein /* transfer per-vdev ZAPs */ 5143215198a6SJoe Stein ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 5144215198a6SJoe Stein VERIFY0(nvlist_add_uint64(child[c], 5145215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 5146215198a6SJoe Stein 5147215198a6SJoe Stein ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 5148215198a6SJoe Stein VERIFY0(nvlist_add_uint64(child[c], 5149215198a6SJoe Stein ZPOOL_CONFIG_VDEV_TOP_ZAP, 5150215198a6SJoe Stein vml[c]->vdev_parent->vdev_top_zap)); 51511195e687SMark J Musante } 51521195e687SMark J Musante 51531195e687SMark J Musante if (error != 0) { 51541195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 51551195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 51561195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 51571195e687SMark J Musante } 51581195e687SMark J Musante 51591195e687SMark J Musante /* stop writers from using the disks */ 51601195e687SMark J Musante for (c = 0; c < children; c++) { 51611195e687SMark J Musante if (vml[c] != NULL) 51621195e687SMark J Musante vml[c]->vdev_offline = B_TRUE; 51631195e687SMark J Musante } 51641195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 51651195e687SMark J Musante 51661195e687SMark J Musante /* 51671195e687SMark J Musante * Temporarily record the splitting vdevs in the spa config. This 51681195e687SMark J Musante * will disappear once the config is regenerated. 51691195e687SMark J Musante */ 51701195e687SMark J Musante VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 51711195e687SMark J Musante VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 51721195e687SMark J Musante glist, children) == 0); 51731195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 51741195e687SMark J Musante 517598295d61SMark J Musante mutex_enter(&spa->spa_props_lock); 51761195e687SMark J Musante VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 51771195e687SMark J Musante nvl) == 0); 517898295d61SMark J Musante mutex_exit(&spa->spa_props_lock); 51791195e687SMark J Musante spa->spa_config_splitting = nvl; 51801195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 51811195e687SMark J Musante 51821195e687SMark J Musante /* configure and create the new pool */ 51831195e687SMark J Musante VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 51841195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 51851195e687SMark J Musante exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 51861195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 51871195e687SMark J Musante spa_version(spa)) == 0); 51881195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 51891195e687SMark J Musante spa->spa_config_txg) == 0); 51901195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 51911195e687SMark J Musante spa_generate_guid(NULL)) == 0); 5192215198a6SJoe Stein VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 51931195e687SMark J Musante (void) nvlist_lookup_string(props, 51941195e687SMark J Musante zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 51951195e687SMark J Musante 5196d41c4376SMark J Musante /* add the new pool to the namespace */ 51971195e687SMark J Musante newspa = spa_add(newname, config, altroot); 5198215198a6SJoe Stein newspa->spa_avz_action = AVZ_ACTION_REBUILD; 51991195e687SMark J Musante newspa->spa_config_txg = spa->spa_config_txg; 52001195e687SMark J Musante spa_set_log_state(newspa, SPA_LOG_CLEAR); 52011195e687SMark J Musante 52021195e687SMark J Musante /* release the spa config lock, retaining the namespace lock */ 52031195e687SMark J Musante spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 52041195e687SMark J Musante 52051195e687SMark J Musante if (zio_injection_enabled) 52061195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 1); 52071195e687SMark J Musante 52081195e687SMark J Musante spa_activate(newspa, spa_mode_global); 52091195e687SMark J Musante spa_async_suspend(newspa); 52101195e687SMark J Musante 52111195e687SMark J Musante /* create the new pool from the disks of the original pool */ 52121195e687SMark J Musante error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 52131195e687SMark J Musante if (error) 52141195e687SMark J Musante goto out; 52151195e687SMark J Musante 52161195e687SMark J Musante /* if that worked, generate a real config for the new pool */ 52171195e687SMark J Musante if (newspa->spa_root_vdev != NULL) { 52181195e687SMark J Musante VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 52191195e687SMark J Musante NV_UNIQUE_NAME, KM_SLEEP) == 0); 52201195e687SMark J Musante VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 52211195e687SMark J Musante ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 52221195e687SMark J Musante spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 52231195e687SMark J Musante B_TRUE)); 52241195e687SMark J Musante } 52251195e687SMark J Musante 52261195e687SMark J Musante /* set the props */ 52271195e687SMark J Musante if (props != NULL) { 52281195e687SMark J Musante spa_configfile_set(newspa, props, B_FALSE); 52291195e687SMark J Musante error = spa_prop_set(newspa, props); 52301195e687SMark J Musante if (error) 52311195e687SMark J Musante goto out; 52321195e687SMark J Musante } 52331195e687SMark J Musante 52341195e687SMark J Musante /* flush everything */ 52351195e687SMark J Musante txg = spa_vdev_config_enter(newspa); 52361195e687SMark J Musante vdev_config_dirty(newspa->spa_root_vdev); 52371195e687SMark J Musante (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 52381195e687SMark J Musante 52391195e687SMark J Musante if (zio_injection_enabled) 52401195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 2); 52411195e687SMark J Musante 52421195e687SMark J Musante spa_async_resume(newspa); 52431195e687SMark J Musante 52441195e687SMark J Musante /* finally, update the original pool's config */ 52451195e687SMark J Musante txg = spa_vdev_config_enter(spa); 52461195e687SMark J Musante tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 52471195e687SMark J Musante error = dmu_tx_assign(tx, TXG_WAIT); 52481195e687SMark J Musante if (error != 0) 52491195e687SMark J Musante dmu_tx_abort(tx); 52501195e687SMark J Musante for (c = 0; c < children; c++) { 52511195e687SMark J Musante if (vml[c] != NULL) { 52521195e687SMark J Musante vdev_split(vml[c]); 52531195e687SMark J Musante if (error == 0) 52544445fffbSMatthew Ahrens spa_history_log_internal(spa, "detach", tx, 52554445fffbSMatthew Ahrens "vdev=%s", vml[c]->vdev_path); 5256215198a6SJoe Stein 52571195e687SMark J Musante vdev_free(vml[c]); 52581195e687SMark J Musante } 52591195e687SMark J Musante } 5260215198a6SJoe Stein spa->spa_avz_action = AVZ_ACTION_REBUILD; 52611195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 52621195e687SMark J Musante spa->spa_config_splitting = NULL; 52631195e687SMark J Musante nvlist_free(nvl); 52641195e687SMark J Musante if (error == 0) 52651195e687SMark J Musante dmu_tx_commit(tx); 52661195e687SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, 0); 52671195e687SMark J Musante 52681195e687SMark J Musante if (zio_injection_enabled) 52691195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 3); 52701195e687SMark J Musante 52711195e687SMark J Musante /* split is complete; log a history record */ 52724445fffbSMatthew Ahrens spa_history_log_internal(newspa, "split", NULL, 52734445fffbSMatthew Ahrens "from pool %s", spa_name(spa)); 52741195e687SMark J Musante 52751195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 52761195e687SMark J Musante 52771195e687SMark J Musante /* if we're not going to mount the filesystems in userland, export */ 52781195e687SMark J Musante if (exp) 52791195e687SMark J Musante error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 52801195e687SMark J Musante B_FALSE, B_FALSE); 52811195e687SMark J Musante 52821195e687SMark J Musante return (error); 52831195e687SMark J Musante 52841195e687SMark J Musante out: 52851195e687SMark J Musante spa_unload(newspa); 52861195e687SMark J Musante spa_deactivate(newspa); 52871195e687SMark J Musante spa_remove(newspa); 52881195e687SMark J Musante 52891195e687SMark J Musante txg = spa_vdev_config_enter(spa); 529098295d61SMark J Musante 529198295d61SMark J Musante /* re-online all offlined disks */ 529298295d61SMark J Musante for (c = 0; c < children; c++) { 529398295d61SMark J Musante if (vml[c] != NULL) 529498295d61SMark J Musante vml[c]->vdev_offline = B_FALSE; 529598295d61SMark J Musante } 529698295d61SMark J Musante vdev_reopen(spa->spa_root_vdev); 529798295d61SMark J Musante 52981195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 52991195e687SMark J Musante spa->spa_config_splitting = NULL; 5300d41c4376SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, error); 53011195e687SMark J Musante 53021195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 53031195e687SMark J Musante return (error); 53041195e687SMark J Musante } 53051195e687SMark J Musante 5306e14bb325SJeff Bonwick static nvlist_t * 5307e14bb325SJeff Bonwick spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 530899653d4eSeschrock { 5309e14bb325SJeff Bonwick for (int i = 0; i < count; i++) { 5310e14bb325SJeff Bonwick uint64_t guid; 531199653d4eSeschrock 5312e14bb325SJeff Bonwick VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5313e14bb325SJeff Bonwick &guid) == 0); 531499653d4eSeschrock 5315e14bb325SJeff Bonwick if (guid == target_guid) 5316e14bb325SJeff Bonwick return (nvpp[i]); 531799653d4eSeschrock } 531899653d4eSeschrock 5319e14bb325SJeff Bonwick return (NULL); 5320fa94a07fSbrendan } 5321fa94a07fSbrendan 5322e14bb325SJeff Bonwick static void 5323e14bb325SJeff Bonwick spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5324e14bb325SJeff Bonwick nvlist_t *dev_to_remove) 5325fa94a07fSbrendan { 5326e14bb325SJeff Bonwick nvlist_t **newdev = NULL; 5327fa94a07fSbrendan 5328e14bb325SJeff Bonwick if (count > 1) 5329e14bb325SJeff Bonwick newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5330fa94a07fSbrendan 5331e14bb325SJeff Bonwick for (int i = 0, j = 0; i < count; i++) { 5332e14bb325SJeff Bonwick if (dev[i] == dev_to_remove) 5333e14bb325SJeff Bonwick continue; 5334e14bb325SJeff Bonwick VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5335fa94a07fSbrendan } 5336fa94a07fSbrendan 5337e14bb325SJeff Bonwick VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5338e14bb325SJeff Bonwick VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5339fa94a07fSbrendan 5340e14bb325SJeff Bonwick for (int i = 0; i < count - 1; i++) 5341e14bb325SJeff Bonwick nvlist_free(newdev[i]); 5342fa94a07fSbrendan 5343e14bb325SJeff Bonwick if (count > 1) 5344e14bb325SJeff Bonwick kmem_free(newdev, (count - 1) * sizeof (void *)); 5345fa94a07fSbrendan } 5346fa94a07fSbrendan 534788ecc943SGeorge Wilson /* 534888ecc943SGeorge Wilson * Evacuate the device. 534988ecc943SGeorge Wilson */ 53503f9d6ad7SLin Ling static int 535188ecc943SGeorge Wilson spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 535288ecc943SGeorge Wilson { 535388ecc943SGeorge Wilson uint64_t txg; 53543f9d6ad7SLin Ling int error = 0; 535588ecc943SGeorge Wilson 535688ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 535788ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5358b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 535988ecc943SGeorge Wilson 536088ecc943SGeorge Wilson /* 536188ecc943SGeorge Wilson * Evacuate the device. We don't hold the config lock as writer 536288ecc943SGeorge Wilson * since we need to do I/O but we do keep the 536388ecc943SGeorge Wilson * spa_namespace_lock held. Once this completes the device 536488ecc943SGeorge Wilson * should no longer have any blocks allocated on it. 536588ecc943SGeorge Wilson */ 536688ecc943SGeorge Wilson if (vd->vdev_islog) { 53673f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 53683f9d6ad7SLin Ling error = spa_offline_log(spa); 5369a1521560SJeff Bonwick } else { 5370be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 537188ecc943SGeorge Wilson } 537288ecc943SGeorge Wilson 5373a1521560SJeff Bonwick if (error) 5374a1521560SJeff Bonwick return (error); 5375a1521560SJeff Bonwick 537688ecc943SGeorge Wilson /* 5377a1521560SJeff Bonwick * The evacuation succeeded. Remove any remaining MOS metadata 5378a1521560SJeff Bonwick * associated with this vdev, and wait for these changes to sync. 537988ecc943SGeorge Wilson */ 5380fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 538188ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 538288ecc943SGeorge Wilson vd->vdev_removing = B_TRUE; 53830713e232SGeorge Wilson vdev_dirty_leaves(vd, VDD_DTL, txg); 538488ecc943SGeorge Wilson vdev_config_dirty(vd); 538588ecc943SGeorge Wilson spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 538688ecc943SGeorge Wilson 538788ecc943SGeorge Wilson return (0); 538888ecc943SGeorge Wilson } 538988ecc943SGeorge Wilson 539088ecc943SGeorge Wilson /* 539188ecc943SGeorge Wilson * Complete the removal by cleaning up the namespace. 539288ecc943SGeorge Wilson */ 53933f9d6ad7SLin Ling static void 5394a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 539588ecc943SGeorge Wilson { 539688ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 539788ecc943SGeorge Wilson uint64_t id = vd->vdev_id; 539888ecc943SGeorge Wilson boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 539988ecc943SGeorge Wilson 540088ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 540188ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5402b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 540388ecc943SGeorge Wilson 54043f9d6ad7SLin Ling /* 54053f9d6ad7SLin Ling * Only remove any devices which are empty. 54063f9d6ad7SLin Ling */ 54073f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 54083f9d6ad7SLin Ling return; 54093f9d6ad7SLin Ling 541088ecc943SGeorge Wilson (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5411b24ab676SJeff Bonwick 5412b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_state_dirty_node)) 5413b24ab676SJeff Bonwick vdev_state_clean(vd); 5414b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_config_dirty_node)) 5415b24ab676SJeff Bonwick vdev_config_clean(vd); 5416b24ab676SJeff Bonwick 541788ecc943SGeorge Wilson vdev_free(vd); 541888ecc943SGeorge Wilson 541988ecc943SGeorge Wilson if (last_vdev) { 542088ecc943SGeorge Wilson vdev_compact_children(rvd); 542188ecc943SGeorge Wilson } else { 542288ecc943SGeorge Wilson vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 542388ecc943SGeorge Wilson vdev_add_child(rvd, vd); 542488ecc943SGeorge Wilson } 5425fcbfa62bSLin Ling vdev_config_dirty(rvd); 5426fcbfa62bSLin Ling 5427fcbfa62bSLin Ling /* 5428fcbfa62bSLin Ling * Reassess the health of our root vdev. 5429fcbfa62bSLin Ling */ 5430fcbfa62bSLin Ling vdev_reopen(rvd); 543188ecc943SGeorge Wilson } 543288ecc943SGeorge Wilson 54333f9d6ad7SLin Ling /* 54343f9d6ad7SLin Ling * Remove a device from the pool - 54353f9d6ad7SLin Ling * 54363f9d6ad7SLin Ling * Removing a device from the vdev namespace requires several steps 54373f9d6ad7SLin Ling * and can take a significant amount of time. As a result we use 54383f9d6ad7SLin Ling * the spa_vdev_config_[enter/exit] functions which allow us to 54393f9d6ad7SLin Ling * grab and release the spa_config_lock while still holding the namespace 54403f9d6ad7SLin Ling * lock. During each step the configuration is synced out. 5441f7170741SWill Andrews * 5442f7170741SWill Andrews * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5443f7170741SWill Andrews * devices. 5444fa94a07fSbrendan */ 5445fa94a07fSbrendan int 5446fa94a07fSbrendan spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5447fa94a07fSbrendan { 5448fa94a07fSbrendan vdev_t *vd; 5449a1521560SJeff Bonwick metaslab_group_t *mg; 5450e14bb325SJeff Bonwick nvlist_t **spares, **l2cache, *nv; 54518ad4d6ddSJeff Bonwick uint64_t txg = 0; 545288ecc943SGeorge Wilson uint_t nspares, nl2cache; 5453fa94a07fSbrendan int error = 0; 54548ad4d6ddSJeff Bonwick boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5455fa94a07fSbrendan 5456f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 5457f9af39baSGeorge Wilson 54588ad4d6ddSJeff Bonwick if (!locked) 54598ad4d6ddSJeff Bonwick txg = spa_vdev_enter(spa); 5460fa94a07fSbrendan 5461c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5462fa94a07fSbrendan 5463fa94a07fSbrendan if (spa->spa_spares.sav_vdevs != NULL && 5464fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5465e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5466e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5467e14bb325SJeff Bonwick /* 5468e14bb325SJeff Bonwick * Only remove the hot spare if it's not currently in use 5469e14bb325SJeff Bonwick * in this pool. 5470e14bb325SJeff Bonwick */ 5471e14bb325SJeff Bonwick if (vd == NULL || unspare) { 5472e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_spares.sav_config, 5473e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5474e14bb325SJeff Bonwick spa_load_spares(spa); 5475e14bb325SJeff Bonwick spa->spa_spares.sav_sync = B_TRUE; 5476e14bb325SJeff Bonwick } else { 5477be6fd75aSMatthew Ahrens error = SET_ERROR(EBUSY); 5478e14bb325SJeff Bonwick } 547963364b0eSAlan Somers spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5480e14bb325SJeff Bonwick } else if (spa->spa_l2cache.sav_vdevs != NULL && 5481fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5482e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5483e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5484e14bb325SJeff Bonwick /* 5485e14bb325SJeff Bonwick * Cache devices can always be removed. 5486e14bb325SJeff Bonwick */ 5487e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5488e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5489fa94a07fSbrendan spa_load_l2cache(spa); 5490fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 549163364b0eSAlan Somers spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 549288ecc943SGeorge Wilson } else if (vd != NULL && vd->vdev_islog) { 549388ecc943SGeorge Wilson ASSERT(!locked); 5494b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 549588ecc943SGeorge Wilson 5496a1521560SJeff Bonwick mg = vd->vdev_mg; 5497a1521560SJeff Bonwick 549888ecc943SGeorge Wilson /* 5499a1521560SJeff Bonwick * Stop allocating from this vdev. 550088ecc943SGeorge Wilson */ 5501a1521560SJeff Bonwick metaslab_group_passivate(mg); 550288ecc943SGeorge Wilson 5503b24ab676SJeff Bonwick /* 5504b24ab676SJeff Bonwick * Wait for the youngest allocations and frees to sync, 5505b24ab676SJeff Bonwick * and then wait for the deferral of those frees to finish. 5506b24ab676SJeff Bonwick */ 5507b24ab676SJeff Bonwick spa_vdev_config_exit(spa, NULL, 5508b24ab676SJeff Bonwick txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5509b24ab676SJeff Bonwick 5510a1521560SJeff Bonwick /* 5511a1521560SJeff Bonwick * Attempt to evacuate the vdev. 5512a1521560SJeff Bonwick */ 5513a1521560SJeff Bonwick error = spa_vdev_remove_evacuate(spa, vd); 5514a1521560SJeff Bonwick 551588ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 551688ecc943SGeorge Wilson 5517a1521560SJeff Bonwick /* 5518a1521560SJeff Bonwick * If we couldn't evacuate the vdev, unwind. 5519a1521560SJeff Bonwick */ 5520a1521560SJeff Bonwick if (error) { 5521a1521560SJeff Bonwick metaslab_group_activate(mg); 5522a1521560SJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, error)); 5523a1521560SJeff Bonwick } 5524a1521560SJeff Bonwick 5525a1521560SJeff Bonwick /* 5526a1521560SJeff Bonwick * Clean up the vdev namespace. 5527a1521560SJeff Bonwick */ 5528a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa, vd); 552988ecc943SGeorge Wilson 553063364b0eSAlan Somers spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV); 5531e14bb325SJeff Bonwick } else if (vd != NULL) { 5532e14bb325SJeff Bonwick /* 5533e14bb325SJeff Bonwick * Normal vdevs cannot be removed (yet). 5534e14bb325SJeff Bonwick */ 5535be6fd75aSMatthew Ahrens error = SET_ERROR(ENOTSUP); 5536e14bb325SJeff Bonwick } else { 5537e14bb325SJeff Bonwick /* 5538e14bb325SJeff Bonwick * There is no vdev of any kind with the specified guid. 5539e14bb325SJeff Bonwick */ 5540be6fd75aSMatthew Ahrens error = SET_ERROR(ENOENT); 5541fa94a07fSbrendan } 554299653d4eSeschrock 55438ad4d6ddSJeff Bonwick if (!locked) 554463364b0eSAlan Somers error = spa_vdev_exit(spa, NULL, txg, error); 55458ad4d6ddSJeff Bonwick 55468ad4d6ddSJeff Bonwick return (error); 5547fa9e4066Sahrens } 5548fa9e4066Sahrens 5549fa9e4066Sahrens /* 55503d7072f8Seschrock * Find any device that's done replacing, or a vdev marked 'unspare' that's 5551f7170741SWill Andrews * currently spared, so we can detach it. 5552fa9e4066Sahrens */ 5553ea8dc4b6Seschrock static vdev_t * 55543d7072f8Seschrock spa_vdev_resilver_done_hunt(vdev_t *vd) 5555fa9e4066Sahrens { 5556ea8dc4b6Seschrock vdev_t *newvd, *oldvd; 5557fa9e4066Sahrens 5558573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 55593d7072f8Seschrock oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5560ea8dc4b6Seschrock if (oldvd != NULL) 5561ea8dc4b6Seschrock return (oldvd); 5562ea8dc4b6Seschrock } 5563fa9e4066Sahrens 55643d7072f8Seschrock /* 5565cb04b873SMark J Musante * Check for a completed replacement. We always consider the first 5566cb04b873SMark J Musante * vdev in the list to be the oldest vdev, and the last one to be 5567cb04b873SMark J Musante * the newest (see spa_vdev_attach() for how that works). In 5568cb04b873SMark J Musante * the case where the newest vdev is faulted, we will not automatically 5569cb04b873SMark J Musante * remove it after a resilver completes. This is OK as it will require 5570cb04b873SMark J Musante * user intervention to determine which disk the admin wishes to keep. 55713d7072f8Seschrock */ 5572cb04b873SMark J Musante if (vd->vdev_ops == &vdev_replacing_ops) { 5573cb04b873SMark J Musante ASSERT(vd->vdev_children > 1); 5574cb04b873SMark J Musante 5575cb04b873SMark J Musante newvd = vd->vdev_child[vd->vdev_children - 1]; 5576ea8dc4b6Seschrock oldvd = vd->vdev_child[0]; 5577ea8dc4b6Seschrock 55788ad4d6ddSJeff Bonwick if (vdev_dtl_empty(newvd, DTL_MISSING) && 5579e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 55808ad4d6ddSJeff Bonwick !vdev_dtl_required(oldvd)) 5581ea8dc4b6Seschrock return (oldvd); 5582fa9e4066Sahrens } 5583ea8dc4b6Seschrock 55843d7072f8Seschrock /* 55853d7072f8Seschrock * Check for a completed resilver with the 'unspare' flag set. 55863d7072f8Seschrock */ 5587cb04b873SMark J Musante if (vd->vdev_ops == &vdev_spare_ops) { 5588cb04b873SMark J Musante vdev_t *first = vd->vdev_child[0]; 5589cb04b873SMark J Musante vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5590cb04b873SMark J Musante 5591cb04b873SMark J Musante if (last->vdev_unspare) { 5592cb04b873SMark J Musante oldvd = first; 5593cb04b873SMark J Musante newvd = last; 5594cb04b873SMark J Musante } else if (first->vdev_unspare) { 5595cb04b873SMark J Musante oldvd = last; 5596cb04b873SMark J Musante newvd = first; 5597cb04b873SMark J Musante } else { 5598cb04b873SMark J Musante oldvd = NULL; 5599cb04b873SMark J Musante } 56003d7072f8Seschrock 5601cb04b873SMark J Musante if (oldvd != NULL && 56028ad4d6ddSJeff Bonwick vdev_dtl_empty(newvd, DTL_MISSING) && 5603e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 5604cb04b873SMark J Musante !vdev_dtl_required(oldvd)) 56053d7072f8Seschrock return (oldvd); 5606cb04b873SMark J Musante 5607cb04b873SMark J Musante /* 5608cb04b873SMark J Musante * If there are more than two spares attached to a disk, 5609cb04b873SMark J Musante * and those spares are not required, then we want to 5610cb04b873SMark J Musante * attempt to free them up now so that they can be used 5611cb04b873SMark J Musante * by other pools. Once we're back down to a single 5612cb04b873SMark J Musante * disk+spare, we stop removing them. 5613cb04b873SMark J Musante */ 5614cb04b873SMark J Musante if (vd->vdev_children > 2) { 5615cb04b873SMark J Musante newvd = vd->vdev_child[1]; 5616cb04b873SMark J Musante 5617cb04b873SMark J Musante if (newvd->vdev_isspare && last->vdev_isspare && 5618cb04b873SMark J Musante vdev_dtl_empty(last, DTL_MISSING) && 5619cb04b873SMark J Musante vdev_dtl_empty(last, DTL_OUTAGE) && 5620cb04b873SMark J Musante !vdev_dtl_required(newvd)) 5621cb04b873SMark J Musante return (newvd); 56223d7072f8Seschrock } 56233d7072f8Seschrock } 56243d7072f8Seschrock 5625ea8dc4b6Seschrock return (NULL); 5626fa9e4066Sahrens } 5627fa9e4066Sahrens 5628ea8dc4b6Seschrock static void 56293d7072f8Seschrock spa_vdev_resilver_done(spa_t *spa) 5630fa9e4066Sahrens { 56318ad4d6ddSJeff Bonwick vdev_t *vd, *pvd, *ppvd; 56328ad4d6ddSJeff Bonwick uint64_t guid, sguid, pguid, ppguid; 5633ea8dc4b6Seschrock 56348ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5635ea8dc4b6Seschrock 56363d7072f8Seschrock while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 56378ad4d6ddSJeff Bonwick pvd = vd->vdev_parent; 56388ad4d6ddSJeff Bonwick ppvd = pvd->vdev_parent; 5639ea8dc4b6Seschrock guid = vd->vdev_guid; 56408ad4d6ddSJeff Bonwick pguid = pvd->vdev_guid; 56418ad4d6ddSJeff Bonwick ppguid = ppvd->vdev_guid; 56428ad4d6ddSJeff Bonwick sguid = 0; 564399653d4eSeschrock /* 564499653d4eSeschrock * If we have just finished replacing a hot spared device, then 564599653d4eSeschrock * we need to detach the parent's first child (the original hot 564699653d4eSeschrock * spare) as well. 564799653d4eSeschrock */ 5648cb04b873SMark J Musante if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5649cb04b873SMark J Musante ppvd->vdev_children == 2) { 565099653d4eSeschrock ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 56518ad4d6ddSJeff Bonwick sguid = ppvd->vdev_child[1]->vdev_guid; 565299653d4eSeschrock } 5653b4952e17SGeorge Wilson ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5654b4952e17SGeorge Wilson 56558ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 56568ad4d6ddSJeff Bonwick if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5657ea8dc4b6Seschrock return; 56588ad4d6ddSJeff Bonwick if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 565999653d4eSeschrock return; 56608ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5661fa9e4066Sahrens } 5662fa9e4066Sahrens 56638ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 5664fa9e4066Sahrens } 5665fa9e4066Sahrens 5666c67d9675Seschrock /* 5667b3388e4fSEric Taylor * Update the stored path or FRU for this vdev. 5668c67d9675Seschrock */ 5669c67d9675Seschrock int 56706809eb4eSEric Schrock spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 56716809eb4eSEric Schrock boolean_t ispath) 5672c67d9675Seschrock { 5673c5904d13Seschrock vdev_t *vd; 5674208044b8SGeorge Wilson boolean_t sync = B_FALSE; 5675c67d9675Seschrock 5676f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 5677f9af39baSGeorge Wilson 5678b3388e4fSEric Taylor spa_vdev_state_enter(spa, SCL_ALL); 5679c67d9675Seschrock 56806809eb4eSEric Schrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5681b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5682c67d9675Seschrock 56830e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 5684b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 56850e34b6a7Sbonwick 56866809eb4eSEric Schrock if (ispath) { 5687208044b8SGeorge Wilson if (strcmp(value, vd->vdev_path) != 0) { 5688208044b8SGeorge Wilson spa_strfree(vd->vdev_path); 5689208044b8SGeorge Wilson vd->vdev_path = spa_strdup(value); 5690208044b8SGeorge Wilson sync = B_TRUE; 5691208044b8SGeorge Wilson } 56926809eb4eSEric Schrock } else { 5693208044b8SGeorge Wilson if (vd->vdev_fru == NULL) { 5694208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 5695208044b8SGeorge Wilson sync = B_TRUE; 5696208044b8SGeorge Wilson } else if (strcmp(value, vd->vdev_fru) != 0) { 56976809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 5698208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 5699208044b8SGeorge Wilson sync = B_TRUE; 5700208044b8SGeorge Wilson } 57016809eb4eSEric Schrock } 5702c67d9675Seschrock 5703208044b8SGeorge Wilson return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5704c67d9675Seschrock } 5705c67d9675Seschrock 57066809eb4eSEric Schrock int 57076809eb4eSEric Schrock spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 57086809eb4eSEric Schrock { 57096809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 57106809eb4eSEric Schrock } 57116809eb4eSEric Schrock 57126809eb4eSEric Schrock int 57136809eb4eSEric Schrock spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 57146809eb4eSEric Schrock { 57156809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 57166809eb4eSEric Schrock } 57176809eb4eSEric Schrock 5718fa9e4066Sahrens /* 5719fa9e4066Sahrens * ========================================================================== 57203f9d6ad7SLin Ling * SPA Scanning 5721fa9e4066Sahrens * ========================================================================== 5722fa9e4066Sahrens */ 5723fa9e4066Sahrens 5724ea8dc4b6Seschrock int 57253f9d6ad7SLin Ling spa_scan_stop(spa_t *spa) 5726fa9e4066Sahrens { 5727e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 57283f9d6ad7SLin Ling if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5729be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 57303f9d6ad7SLin Ling return (dsl_scan_cancel(spa->spa_dsl_pool)); 57313f9d6ad7SLin Ling } 5732bb8b5132Sek 57333f9d6ad7SLin Ling int 57343f9d6ad7SLin Ling spa_scan(spa_t *spa, pool_scan_func_t func) 57353f9d6ad7SLin Ling { 57363f9d6ad7SLin Ling ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 57373f9d6ad7SLin Ling 57383f9d6ad7SLin Ling if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5739be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 5740fa9e4066Sahrens 5741fa9e4066Sahrens /* 5742088f3894Sahrens * If a resilver was requested, but there is no DTL on a 5743088f3894Sahrens * writeable leaf device, we have nothing to do. 5744fa9e4066Sahrens */ 57453f9d6ad7SLin Ling if (func == POOL_SCAN_RESILVER && 5746088f3894Sahrens !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5747088f3894Sahrens spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5748ea8dc4b6Seschrock return (0); 5749ea8dc4b6Seschrock } 5750fa9e4066Sahrens 57513f9d6ad7SLin Ling return (dsl_scan(spa->spa_dsl_pool, func)); 5752fa9e4066Sahrens } 5753fa9e4066Sahrens 5754ea8dc4b6Seschrock /* 5755ea8dc4b6Seschrock * ========================================================================== 5756ea8dc4b6Seschrock * SPA async task processing 5757ea8dc4b6Seschrock * ========================================================================== 5758ea8dc4b6Seschrock */ 5759ea8dc4b6Seschrock 5760ea8dc4b6Seschrock static void 57613d7072f8Seschrock spa_async_remove(spa_t *spa, vdev_t *vd) 5762fa9e4066Sahrens { 576349cf58c0SBrendan Gregg - Sun Microsystems if (vd->vdev_remove_wanted) { 576498d1cbfeSGeorge Wilson vd->vdev_remove_wanted = B_FALSE; 576598d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 576649cf58c0SBrendan Gregg - Sun Microsystems vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 57671d713200SEric Schrock 57681d713200SEric Schrock /* 57691d713200SEric Schrock * We want to clear the stats, but we don't want to do a full 57701d713200SEric Schrock * vdev_clear() as that will cause us to throw away 57711d713200SEric Schrock * degraded/faulted state as well as attempt to reopen the 57721d713200SEric Schrock * device, all of which is a waste. 57731d713200SEric Schrock */ 57741d713200SEric Schrock vd->vdev_stat.vs_read_errors = 0; 57751d713200SEric Schrock vd->vdev_stat.vs_write_errors = 0; 57761d713200SEric Schrock vd->vdev_stat.vs_checksum_errors = 0; 57771d713200SEric Schrock 5778e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 5779ea8dc4b6Seschrock } 578049cf58c0SBrendan Gregg - Sun Microsystems 5781e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 578249cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, vd->vdev_child[c]); 5783ea8dc4b6Seschrock } 5784fa9e4066Sahrens 5785e14bb325SJeff Bonwick static void 5786e14bb325SJeff Bonwick spa_async_probe(spa_t *spa, vdev_t *vd) 5787e14bb325SJeff Bonwick { 5788e14bb325SJeff Bonwick if (vd->vdev_probe_wanted) { 578998d1cbfeSGeorge Wilson vd->vdev_probe_wanted = B_FALSE; 5790e14bb325SJeff Bonwick vdev_reopen(vd); /* vdev_open() does the actual probe */ 5791e14bb325SJeff Bonwick } 5792e14bb325SJeff Bonwick 5793e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 5794e14bb325SJeff Bonwick spa_async_probe(spa, vd->vdev_child[c]); 5795e14bb325SJeff Bonwick } 5796e14bb325SJeff Bonwick 5797573ca77eSGeorge Wilson static void 5798573ca77eSGeorge Wilson spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5799573ca77eSGeorge Wilson { 5800573ca77eSGeorge Wilson sysevent_id_t eid; 5801573ca77eSGeorge Wilson nvlist_t *attr; 5802573ca77eSGeorge Wilson char *physpath; 5803573ca77eSGeorge Wilson 5804573ca77eSGeorge Wilson if (!spa->spa_autoexpand) 5805573ca77eSGeorge Wilson return; 5806573ca77eSGeorge Wilson 5807573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 5808573ca77eSGeorge Wilson vdev_t *cvd = vd->vdev_child[c]; 5809573ca77eSGeorge Wilson spa_async_autoexpand(spa, cvd); 5810573ca77eSGeorge Wilson } 5811573ca77eSGeorge Wilson 5812573ca77eSGeorge Wilson if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5813573ca77eSGeorge Wilson return; 5814573ca77eSGeorge Wilson 5815573ca77eSGeorge Wilson physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5816573ca77eSGeorge Wilson (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5817573ca77eSGeorge Wilson 5818573ca77eSGeorge Wilson VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5819573ca77eSGeorge Wilson VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5820573ca77eSGeorge Wilson 5821573ca77eSGeorge Wilson (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5822573ca77eSGeorge Wilson ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 5823573ca77eSGeorge Wilson 5824573ca77eSGeorge Wilson nvlist_free(attr); 5825573ca77eSGeorge Wilson kmem_free(physpath, MAXPATHLEN); 5826573ca77eSGeorge Wilson } 5827573ca77eSGeorge Wilson 5828ea8dc4b6Seschrock static void 5829ea8dc4b6Seschrock spa_async_thread(spa_t *spa) 5830ea8dc4b6Seschrock { 5831e14bb325SJeff Bonwick int tasks; 5832ea8dc4b6Seschrock 5833ea8dc4b6Seschrock ASSERT(spa->spa_sync_on); 5834ea8dc4b6Seschrock 5835ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5836ea8dc4b6Seschrock tasks = spa->spa_async_tasks; 5837ea8dc4b6Seschrock spa->spa_async_tasks = 0; 5838ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5839ea8dc4b6Seschrock 58400373e76bSbonwick /* 58410373e76bSbonwick * See if the config needs to be updated. 58420373e76bSbonwick */ 58430373e76bSbonwick if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5844b24ab676SJeff Bonwick uint64_t old_space, new_space; 5845573ca77eSGeorge Wilson 58460373e76bSbonwick mutex_enter(&spa_namespace_lock); 5847b24ab676SJeff Bonwick old_space = metaslab_class_get_space(spa_normal_class(spa)); 58480373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5849b24ab676SJeff Bonwick new_space = metaslab_class_get_space(spa_normal_class(spa)); 58500373e76bSbonwick mutex_exit(&spa_namespace_lock); 5851573ca77eSGeorge Wilson 5852573ca77eSGeorge Wilson /* 5853573ca77eSGeorge Wilson * If the pool grew as a result of the config update, 5854573ca77eSGeorge Wilson * then log an internal history event. 5855573ca77eSGeorge Wilson */ 5856b24ab676SJeff Bonwick if (new_space != old_space) { 58574445fffbSMatthew Ahrens spa_history_log_internal(spa, "vdev online", NULL, 5858c8e1f6d2SMark J Musante "pool '%s' size: %llu(+%llu)", 5859b24ab676SJeff Bonwick spa_name(spa), new_space, new_space - old_space); 5860573ca77eSGeorge Wilson } 58610373e76bSbonwick } 58620373e76bSbonwick 5863ea8dc4b6Seschrock /* 58643d7072f8Seschrock * See if any devices need to be marked REMOVED. 5865ea8dc4b6Seschrock */ 5866e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_REMOVE) { 58678f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 58683d7072f8Seschrock spa_async_remove(spa, spa->spa_root_vdev); 5869e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 587049cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5871e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_spares.sav_count; i++) 587249cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5873e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 5874e14bb325SJeff Bonwick } 5875e14bb325SJeff Bonwick 5876573ca77eSGeorge Wilson if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5877573ca77eSGeorge Wilson spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5878573ca77eSGeorge Wilson spa_async_autoexpand(spa, spa->spa_root_vdev); 5879573ca77eSGeorge Wilson spa_config_exit(spa, SCL_CONFIG, FTAG); 5880573ca77eSGeorge Wilson } 5881573ca77eSGeorge Wilson 5882e14bb325SJeff Bonwick /* 5883e14bb325SJeff Bonwick * See if any devices need to be probed. 5884e14bb325SJeff Bonwick */ 5885e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_PROBE) { 58868f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 5887e14bb325SJeff Bonwick spa_async_probe(spa, spa->spa_root_vdev); 5888e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 58893d7072f8Seschrock } 5890ea8dc4b6Seschrock 5891ea8dc4b6Seschrock /* 5892ea8dc4b6Seschrock * If any devices are done replacing, detach them. 5893ea8dc4b6Seschrock */ 58943d7072f8Seschrock if (tasks & SPA_ASYNC_RESILVER_DONE) 58953d7072f8Seschrock spa_vdev_resilver_done(spa); 5896fa9e4066Sahrens 5897ea8dc4b6Seschrock /* 5898ea8dc4b6Seschrock * Kick off a resilver. 5899ea8dc4b6Seschrock */ 5900088f3894Sahrens if (tasks & SPA_ASYNC_RESILVER) 59013f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, 0); 5902ea8dc4b6Seschrock 5903ea8dc4b6Seschrock /* 5904ea8dc4b6Seschrock * Let the world know that we're done. 5905ea8dc4b6Seschrock */ 5906ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5907ea8dc4b6Seschrock spa->spa_async_thread = NULL; 5908ea8dc4b6Seschrock cv_broadcast(&spa->spa_async_cv); 5909ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5910ea8dc4b6Seschrock thread_exit(); 5911ea8dc4b6Seschrock } 5912ea8dc4b6Seschrock 5913ea8dc4b6Seschrock void 5914ea8dc4b6Seschrock spa_async_suspend(spa_t *spa) 5915ea8dc4b6Seschrock { 5916ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5917ea8dc4b6Seschrock spa->spa_async_suspended++; 5918ea8dc4b6Seschrock while (spa->spa_async_thread != NULL) 5919ea8dc4b6Seschrock cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5920ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5921ea8dc4b6Seschrock } 5922ea8dc4b6Seschrock 5923ea8dc4b6Seschrock void 5924ea8dc4b6Seschrock spa_async_resume(spa_t *spa) 5925ea8dc4b6Seschrock { 5926ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5927ea8dc4b6Seschrock ASSERT(spa->spa_async_suspended != 0); 5928ea8dc4b6Seschrock spa->spa_async_suspended--; 5929ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5930ea8dc4b6Seschrock } 5931ea8dc4b6Seschrock 59323cb69f73SWill Andrews static boolean_t 59333cb69f73SWill Andrews spa_async_tasks_pending(spa_t *spa) 59343cb69f73SWill Andrews { 59353cb69f73SWill Andrews uint_t non_config_tasks; 59363cb69f73SWill Andrews uint_t config_task; 59373cb69f73SWill Andrews boolean_t config_task_suspended; 59383cb69f73SWill Andrews 59393cb69f73SWill Andrews non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; 59403cb69f73SWill Andrews config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 59413cb69f73SWill Andrews if (spa->spa_ccw_fail_time == 0) { 59423cb69f73SWill Andrews config_task_suspended = B_FALSE; 59433cb69f73SWill Andrews } else { 59443cb69f73SWill Andrews config_task_suspended = 59453cb69f73SWill Andrews (gethrtime() - spa->spa_ccw_fail_time) < 59463cb69f73SWill Andrews (zfs_ccw_retry_interval * NANOSEC); 59473cb69f73SWill Andrews } 59483cb69f73SWill Andrews 59493cb69f73SWill Andrews return (non_config_tasks || (config_task && !config_task_suspended)); 59503cb69f73SWill Andrews } 59513cb69f73SWill Andrews 5952ea8dc4b6Seschrock static void 5953ea8dc4b6Seschrock spa_async_dispatch(spa_t *spa) 5954ea8dc4b6Seschrock { 5955ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 59563cb69f73SWill Andrews if (spa_async_tasks_pending(spa) && 59573cb69f73SWill Andrews !spa->spa_async_suspended && 59580373e76bSbonwick spa->spa_async_thread == NULL && 59593cb69f73SWill Andrews rootdir != NULL) 5960ea8dc4b6Seschrock spa->spa_async_thread = thread_create(NULL, 0, 5961ea8dc4b6Seschrock spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5962ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5963ea8dc4b6Seschrock } 5964ea8dc4b6Seschrock 5965ea8dc4b6Seschrock void 5966ea8dc4b6Seschrock spa_async_request(spa_t *spa, int task) 5967ea8dc4b6Seschrock { 59683f9d6ad7SLin Ling zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5969ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5970ea8dc4b6Seschrock spa->spa_async_tasks |= task; 5971ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5972fa9e4066Sahrens } 5973fa9e4066Sahrens 5974fa9e4066Sahrens /* 5975fa9e4066Sahrens * ========================================================================== 5976fa9e4066Sahrens * SPA syncing routines 5977fa9e4066Sahrens * ========================================================================== 5978fa9e4066Sahrens */ 5979fa9e4066Sahrens 5980cde58dbcSMatthew Ahrens static int 5981cde58dbcSMatthew Ahrens bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5982cde58dbcSMatthew Ahrens { 5983cde58dbcSMatthew Ahrens bpobj_t *bpo = arg; 5984cde58dbcSMatthew Ahrens bpobj_enqueue(bpo, bp, tx); 5985cde58dbcSMatthew Ahrens return (0); 5986b24ab676SJeff Bonwick } 5987b24ab676SJeff Bonwick 5988cde58dbcSMatthew Ahrens static int 5989cde58dbcSMatthew Ahrens spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5990b24ab676SJeff Bonwick { 5991b24ab676SJeff Bonwick zio_t *zio = arg; 5992b24ab676SJeff Bonwick 5993b24ab676SJeff Bonwick zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5994b24ab676SJeff Bonwick zio->io_flags)); 5995cde58dbcSMatthew Ahrens return (0); 5996fa9e4066Sahrens } 5997fa9e4066Sahrens 599869962b56SMatthew Ahrens /* 599969962b56SMatthew Ahrens * Note: this simple function is not inlined to make it easier to dtrace the 600069962b56SMatthew Ahrens * amount of time spent syncing frees. 600169962b56SMatthew Ahrens */ 600269962b56SMatthew Ahrens static void 600369962b56SMatthew Ahrens spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 600469962b56SMatthew Ahrens { 600569962b56SMatthew Ahrens zio_t *zio = zio_root(spa, NULL, NULL, 0); 600669962b56SMatthew Ahrens bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 600769962b56SMatthew Ahrens VERIFY(zio_wait(zio) == 0); 600869962b56SMatthew Ahrens } 600969962b56SMatthew Ahrens 601069962b56SMatthew Ahrens /* 601169962b56SMatthew Ahrens * Note: this simple function is not inlined to make it easier to dtrace the 601269962b56SMatthew Ahrens * amount of time spent syncing deferred frees. 601369962b56SMatthew Ahrens */ 601469962b56SMatthew Ahrens static void 601569962b56SMatthew Ahrens spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 601669962b56SMatthew Ahrens { 601769962b56SMatthew Ahrens zio_t *zio = zio_root(spa, NULL, NULL, 0); 601869962b56SMatthew Ahrens VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 601969962b56SMatthew Ahrens spa_free_sync_cb, zio, tx), ==, 0); 602069962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 602169962b56SMatthew Ahrens } 602269962b56SMatthew Ahrens 602369962b56SMatthew Ahrens 6024fa9e4066Sahrens static void 602599653d4eSeschrock spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6026fa9e4066Sahrens { 6027fa9e4066Sahrens char *packed = NULL; 6028f7991ba4STim Haley size_t bufsize; 6029fa9e4066Sahrens size_t nvsize = 0; 6030fa9e4066Sahrens dmu_buf_t *db; 6031fa9e4066Sahrens 603299653d4eSeschrock VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6033fa9e4066Sahrens 6034f7991ba4STim Haley /* 6035f7991ba4STim Haley * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 603643466aaeSMax Grossman * information. This avoids the dmu_buf_will_dirty() path and 6037f7991ba4STim Haley * saves us a pre-read to get data we don't actually care about. 6038f7991ba4STim Haley */ 6039ad135b5dSChristopher Siden bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6040f7991ba4STim Haley packed = kmem_alloc(bufsize, KM_SLEEP); 6041fa9e4066Sahrens 604299653d4eSeschrock VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6043ea8dc4b6Seschrock KM_SLEEP) == 0); 6044f7991ba4STim Haley bzero(packed + nvsize, bufsize - nvsize); 6045fa9e4066Sahrens 6046f7991ba4STim Haley dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6047fa9e4066Sahrens 6048f7991ba4STim Haley kmem_free(packed, bufsize); 6049fa9e4066Sahrens 605099653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6051fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 6052fa9e4066Sahrens *(uint64_t *)db->db_data = nvsize; 6053ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 6054fa9e4066Sahrens } 6055fa9e4066Sahrens 605699653d4eSeschrock static void 6057fa94a07fSbrendan spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6058fa94a07fSbrendan const char *config, const char *entry) 605999653d4eSeschrock { 606099653d4eSeschrock nvlist_t *nvroot; 6061fa94a07fSbrendan nvlist_t **list; 606299653d4eSeschrock int i; 606399653d4eSeschrock 6064fa94a07fSbrendan if (!sav->sav_sync) 606599653d4eSeschrock return; 606699653d4eSeschrock 606799653d4eSeschrock /* 6068fa94a07fSbrendan * Update the MOS nvlist describing the list of available devices. 6069fa94a07fSbrendan * spa_validate_aux() will have already made sure this nvlist is 60703d7072f8Seschrock * valid and the vdevs are labeled appropriately. 607199653d4eSeschrock */ 6072fa94a07fSbrendan if (sav->sav_object == 0) { 6073fa94a07fSbrendan sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6074fa94a07fSbrendan DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6075fa94a07fSbrendan sizeof (uint64_t), tx); 607699653d4eSeschrock VERIFY(zap_update(spa->spa_meta_objset, 6077fa94a07fSbrendan DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6078fa94a07fSbrendan &sav->sav_object, tx) == 0); 607999653d4eSeschrock } 608099653d4eSeschrock 608199653d4eSeschrock VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6082fa94a07fSbrendan if (sav->sav_count == 0) { 6083fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 608499653d4eSeschrock } else { 6085fa94a07fSbrendan list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6086fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 6087fa94a07fSbrendan list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 60883f9d6ad7SLin Ling B_FALSE, VDEV_CONFIG_L2CACHE); 6089fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6090fa94a07fSbrendan sav->sav_count) == 0); 6091fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 6092fa94a07fSbrendan nvlist_free(list[i]); 6093fa94a07fSbrendan kmem_free(list, sav->sav_count * sizeof (void *)); 609499653d4eSeschrock } 609599653d4eSeschrock 6096fa94a07fSbrendan spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 609706eeb2adSek nvlist_free(nvroot); 609899653d4eSeschrock 6099fa94a07fSbrendan sav->sav_sync = B_FALSE; 610099653d4eSeschrock } 610199653d4eSeschrock 6102215198a6SJoe Stein /* 6103215198a6SJoe Stein * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 6104215198a6SJoe Stein * The all-vdev ZAP must be empty. 6105215198a6SJoe Stein */ 6106215198a6SJoe Stein static void 6107215198a6SJoe Stein spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 6108215198a6SJoe Stein { 6109215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 6110215198a6SJoe Stein if (vd->vdev_top_zap != 0) { 6111215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6112215198a6SJoe Stein vd->vdev_top_zap, tx)); 6113215198a6SJoe Stein } 6114215198a6SJoe Stein if (vd->vdev_leaf_zap != 0) { 6115215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6116215198a6SJoe Stein vd->vdev_leaf_zap, tx)); 6117215198a6SJoe Stein } 6118215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 6119215198a6SJoe Stein spa_avz_build(vd->vdev_child[i], avz, tx); 6120215198a6SJoe Stein } 6121215198a6SJoe Stein } 6122215198a6SJoe Stein 612399653d4eSeschrock static void 612499653d4eSeschrock spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 612599653d4eSeschrock { 612699653d4eSeschrock nvlist_t *config; 612799653d4eSeschrock 6128215198a6SJoe Stein /* 6129215198a6SJoe Stein * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 6130215198a6SJoe Stein * its config may not be dirty but we still need to build per-vdev ZAPs. 6131215198a6SJoe Stein * Similarly, if the pool is being assembled (e.g. after a split), we 6132215198a6SJoe Stein * need to rebuild the AVZ although the config may not be dirty. 6133215198a6SJoe Stein */ 6134215198a6SJoe Stein if (list_is_empty(&spa->spa_config_dirty_list) && 6135215198a6SJoe Stein spa->spa_avz_action == AVZ_ACTION_NONE) 613699653d4eSeschrock return; 613799653d4eSeschrock 6138e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6139e14bb325SJeff Bonwick 6140215198a6SJoe Stein ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 6141215198a6SJoe Stein spa->spa_all_vdev_zaps != 0); 6142215198a6SJoe Stein 6143215198a6SJoe Stein if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 6144215198a6SJoe Stein /* Make and build the new AVZ */ 6145215198a6SJoe Stein uint64_t new_avz = zap_create(spa->spa_meta_objset, 6146215198a6SJoe Stein DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 6147215198a6SJoe Stein spa_avz_build(spa->spa_root_vdev, new_avz, tx); 6148215198a6SJoe Stein 6149215198a6SJoe Stein /* Diff old AVZ with new one */ 6150215198a6SJoe Stein zap_cursor_t zc; 6151215198a6SJoe Stein zap_attribute_t za; 6152215198a6SJoe Stein 6153215198a6SJoe Stein for (zap_cursor_init(&zc, spa->spa_meta_objset, 6154215198a6SJoe Stein spa->spa_all_vdev_zaps); 6155215198a6SJoe Stein zap_cursor_retrieve(&zc, &za) == 0; 6156215198a6SJoe Stein zap_cursor_advance(&zc)) { 6157215198a6SJoe Stein uint64_t vdzap = za.za_first_integer; 6158215198a6SJoe Stein if (zap_lookup_int(spa->spa_meta_objset, new_avz, 6159215198a6SJoe Stein vdzap) == ENOENT) { 6160215198a6SJoe Stein /* 6161215198a6SJoe Stein * ZAP is listed in old AVZ but not in new one; 6162215198a6SJoe Stein * destroy it 6163215198a6SJoe Stein */ 6164215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 6165215198a6SJoe Stein tx)); 6166215198a6SJoe Stein } 6167215198a6SJoe Stein } 6168215198a6SJoe Stein 6169215198a6SJoe Stein zap_cursor_fini(&zc); 6170215198a6SJoe Stein 6171215198a6SJoe Stein /* Destroy the old AVZ */ 6172215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, 6173215198a6SJoe Stein spa->spa_all_vdev_zaps, tx)); 6174215198a6SJoe Stein 6175215198a6SJoe Stein /* Replace the old AVZ in the dir obj with the new one */ 6176215198a6SJoe Stein VERIFY0(zap_update(spa->spa_meta_objset, 6177215198a6SJoe Stein DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 6178215198a6SJoe Stein sizeof (new_avz), 1, &new_avz, tx)); 6179215198a6SJoe Stein 6180215198a6SJoe Stein spa->spa_all_vdev_zaps = new_avz; 6181215198a6SJoe Stein } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 6182215198a6SJoe Stein zap_cursor_t zc; 6183215198a6SJoe Stein zap_attribute_t za; 6184215198a6SJoe Stein 6185215198a6SJoe Stein /* Walk through the AVZ and destroy all listed ZAPs */ 6186215198a6SJoe Stein for (zap_cursor_init(&zc, spa->spa_meta_objset, 6187215198a6SJoe Stein spa->spa_all_vdev_zaps); 6188215198a6SJoe Stein zap_cursor_retrieve(&zc, &za) == 0; 6189215198a6SJoe Stein zap_cursor_advance(&zc)) { 6190215198a6SJoe Stein uint64_t zap = za.za_first_integer; 6191215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 6192215198a6SJoe Stein } 6193215198a6SJoe Stein 6194215198a6SJoe Stein zap_cursor_fini(&zc); 6195215198a6SJoe Stein 6196215198a6SJoe Stein /* Destroy and unlink the AVZ itself */ 6197215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, 6198215198a6SJoe Stein spa->spa_all_vdev_zaps, tx)); 6199215198a6SJoe Stein VERIFY0(zap_remove(spa->spa_meta_objset, 6200215198a6SJoe Stein DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 6201215198a6SJoe Stein spa->spa_all_vdev_zaps = 0; 6202215198a6SJoe Stein } 6203215198a6SJoe Stein 6204215198a6SJoe Stein if (spa->spa_all_vdev_zaps == 0) { 6205215198a6SJoe Stein spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 6206215198a6SJoe Stein DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 6207215198a6SJoe Stein DMU_POOL_VDEV_ZAP_MAP, tx); 6208215198a6SJoe Stein } 6209215198a6SJoe Stein spa->spa_avz_action = AVZ_ACTION_NONE; 6210215198a6SJoe Stein 6211215198a6SJoe Stein /* Create ZAPs for vdevs that don't have them. */ 6212215198a6SJoe Stein vdev_construct_zaps(spa->spa_root_vdev, tx); 6213215198a6SJoe Stein 6214e14bb325SJeff Bonwick config = spa_config_generate(spa, spa->spa_root_vdev, 6215e14bb325SJeff Bonwick dmu_tx_get_txg(tx), B_FALSE); 6216e14bb325SJeff Bonwick 621725345e46SGeorge Wilson /* 621825345e46SGeorge Wilson * If we're upgrading the spa version then make sure that 621925345e46SGeorge Wilson * the config object gets updated with the correct version. 622025345e46SGeorge Wilson */ 622125345e46SGeorge Wilson if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 622225345e46SGeorge Wilson fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 622325345e46SGeorge Wilson spa->spa_uberblock.ub_version); 622425345e46SGeorge Wilson 6225e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 622699653d4eSeschrock 6227aab83bb8SJosef 'Jeff' Sipek nvlist_free(spa->spa_config_syncing); 622899653d4eSeschrock spa->spa_config_syncing = config; 622999653d4eSeschrock 623099653d4eSeschrock spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 623199653d4eSeschrock } 623299653d4eSeschrock 6233ad135b5dSChristopher Siden static void 62343b2aab18SMatthew Ahrens spa_sync_version(void *arg, dmu_tx_t *tx) 6235ad135b5dSChristopher Siden { 62363b2aab18SMatthew Ahrens uint64_t *versionp = arg; 62373b2aab18SMatthew Ahrens uint64_t version = *versionp; 62383b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6239ad135b5dSChristopher Siden 6240ad135b5dSChristopher Siden /* 6241ad135b5dSChristopher Siden * Setting the version is special cased when first creating the pool. 6242ad135b5dSChristopher Siden */ 6243ad135b5dSChristopher Siden ASSERT(tx->tx_txg != TXG_INITIAL); 6244ad135b5dSChristopher Siden 624562eae887SRichard Yao ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6246ad135b5dSChristopher Siden ASSERT(version >= spa_version(spa)); 6247ad135b5dSChristopher Siden 6248ad135b5dSChristopher Siden spa->spa_uberblock.ub_version = version; 6249ad135b5dSChristopher Siden vdev_config_dirty(spa->spa_root_vdev); 62504445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6251ad135b5dSChristopher Siden } 6252ad135b5dSChristopher Siden 6253990b4856Slling /* 6254990b4856Slling * Set zpool properties. 6255990b4856Slling */ 6256b1b8ab34Slling static void 62573b2aab18SMatthew Ahrens spa_sync_props(void *arg, dmu_tx_t *tx) 6258b1b8ab34Slling { 62593b2aab18SMatthew Ahrens nvlist_t *nvp = arg; 62603b2aab18SMatthew Ahrens spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6261b1b8ab34Slling objset_t *mos = spa->spa_meta_objset; 6262ad135b5dSChristopher Siden nvpair_t *elem = NULL; 6263b1b8ab34Slling 6264e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 6265e14bb325SJeff Bonwick 6266990b4856Slling while ((elem = nvlist_next_nvpair(nvp, elem))) { 6267ad135b5dSChristopher Siden uint64_t intval; 6268ad135b5dSChristopher Siden char *strval, *fname; 6269ad135b5dSChristopher Siden zpool_prop_t prop; 6270ad135b5dSChristopher Siden const char *propname; 6271ad135b5dSChristopher Siden zprop_type_t proptype; 62722acef22dSMatthew Ahrens spa_feature_t fid; 6273ad135b5dSChristopher Siden 6274990b4856Slling switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6275ad135b5dSChristopher Siden case ZPROP_INVAL: 6276ad135b5dSChristopher Siden /* 6277ad135b5dSChristopher Siden * We checked this earlier in spa_prop_validate(). 6278ad135b5dSChristopher Siden */ 6279ad135b5dSChristopher Siden ASSERT(zpool_prop_feature(nvpair_name(elem))); 6280ad135b5dSChristopher Siden 6281ad135b5dSChristopher Siden fname = strchr(nvpair_name(elem), '@') + 1; 62822acef22dSMatthew Ahrens VERIFY0(zfeature_lookup_name(fname, &fid)); 6283ad135b5dSChristopher Siden 62842acef22dSMatthew Ahrens spa_feature_enable(spa, fid, tx); 62854445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 62864445fffbSMatthew Ahrens "%s=enabled", nvpair_name(elem)); 6287ad135b5dSChristopher Siden break; 6288ad135b5dSChristopher Siden 6289990b4856Slling case ZPOOL_PROP_VERSION: 62900713e232SGeorge Wilson intval = fnvpair_value_uint64(elem); 6291990b4856Slling /* 6292ad135b5dSChristopher Siden * The version is synced seperatly before other 6293ad135b5dSChristopher Siden * properties and should be correct by now. 6294990b4856Slling */ 6295ad135b5dSChristopher Siden ASSERT3U(spa_version(spa), >=, intval); 6296ecd6cf80Smarks break; 6297990b4856Slling 6298990b4856Slling case ZPOOL_PROP_ALTROOT: 6299990b4856Slling /* 6300990b4856Slling * 'altroot' is a non-persistent property. It should 6301990b4856Slling * have been set temporarily at creation or import time. 6302990b4856Slling */ 6303990b4856Slling ASSERT(spa->spa_root != NULL); 6304b1b8ab34Slling break; 63053d7072f8Seschrock 6306f9af39baSGeorge Wilson case ZPOOL_PROP_READONLY: 63072f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 6308990b4856Slling /* 6309f9af39baSGeorge Wilson * 'readonly' and 'cachefile' are also non-persisitent 6310f9af39baSGeorge Wilson * properties. 6311990b4856Slling */ 63123d7072f8Seschrock break; 63138704186eSDan McDonald case ZPOOL_PROP_COMMENT: 63140713e232SGeorge Wilson strval = fnvpair_value_string(elem); 63158704186eSDan McDonald if (spa->spa_comment != NULL) 63168704186eSDan McDonald spa_strfree(spa->spa_comment); 63178704186eSDan McDonald spa->spa_comment = spa_strdup(strval); 63188704186eSDan McDonald /* 63198704186eSDan McDonald * We need to dirty the configuration on all the vdevs 63208704186eSDan McDonald * so that their labels get updated. It's unnecessary 63218704186eSDan McDonald * to do this for pool creation since the vdev's 63228704186eSDan McDonald * configuratoin has already been dirtied. 63238704186eSDan McDonald */ 63248704186eSDan McDonald if (tx->tx_txg != TXG_INITIAL) 63258704186eSDan McDonald vdev_config_dirty(spa->spa_root_vdev); 63264445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 63274445fffbSMatthew Ahrens "%s=%s", nvpair_name(elem), strval); 63288704186eSDan McDonald break; 6329990b4856Slling default: 6330990b4856Slling /* 6331990b4856Slling * Set pool property values in the poolprops mos object. 6332990b4856Slling */ 6333990b4856Slling if (spa->spa_pool_props_object == 0) { 6334ad135b5dSChristopher Siden spa->spa_pool_props_object = 6335ad135b5dSChristopher Siden zap_create_link(mos, DMU_OT_POOL_PROPS, 6336990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6337ad135b5dSChristopher Siden tx); 6338990b4856Slling } 6339990b4856Slling 6340990b4856Slling /* normalize the property name */ 6341990b4856Slling propname = zpool_prop_to_name(prop); 6342990b4856Slling proptype = zpool_prop_get_type(prop); 6343990b4856Slling 6344990b4856Slling if (nvpair_type(elem) == DATA_TYPE_STRING) { 6345990b4856Slling ASSERT(proptype == PROP_TYPE_STRING); 63460713e232SGeorge Wilson strval = fnvpair_value_string(elem); 63470713e232SGeorge Wilson VERIFY0(zap_update(mos, 6348990b4856Slling spa->spa_pool_props_object, propname, 63490713e232SGeorge Wilson 1, strlen(strval) + 1, strval, tx)); 63504445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 63514445fffbSMatthew Ahrens "%s=%s", nvpair_name(elem), strval); 6352990b4856Slling } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 63530713e232SGeorge Wilson intval = fnvpair_value_uint64(elem); 6354990b4856Slling 6355990b4856Slling if (proptype == PROP_TYPE_INDEX) { 6356990b4856Slling const char *unused; 63570713e232SGeorge Wilson VERIFY0(zpool_prop_index_to_string( 63580713e232SGeorge Wilson prop, intval, &unused)); 6359990b4856Slling } 63600713e232SGeorge Wilson VERIFY0(zap_update(mos, 6361990b4856Slling spa->spa_pool_props_object, propname, 63620713e232SGeorge Wilson 8, 1, &intval, tx)); 63634445fffbSMatthew Ahrens spa_history_log_internal(spa, "set", tx, 63644445fffbSMatthew Ahrens "%s=%lld", nvpair_name(elem), intval); 6365990b4856Slling } else { 6366990b4856Slling ASSERT(0); /* not allowed */ 6367990b4856Slling } 6368990b4856Slling 63690a4e9518Sgw switch (prop) { 63700a4e9518Sgw case ZPOOL_PROP_DELEGATION: 6371990b4856Slling spa->spa_delegation = intval; 63720a4e9518Sgw break; 63730a4e9518Sgw case ZPOOL_PROP_BOOTFS: 6374990b4856Slling spa->spa_bootfs = intval; 63750a4e9518Sgw break; 63760a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 63770a4e9518Sgw spa->spa_failmode = intval; 63780a4e9518Sgw break; 6379573ca77eSGeorge Wilson case ZPOOL_PROP_AUTOEXPAND: 6380573ca77eSGeorge Wilson spa->spa_autoexpand = intval; 6381b98131cfSEric Taylor if (tx->tx_txg != TXG_INITIAL) 6382b98131cfSEric Taylor spa_async_request(spa, 6383b98131cfSEric Taylor SPA_ASYNC_AUTOEXPAND); 6384573ca77eSGeorge Wilson break; 6385b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 6386b24ab676SJeff Bonwick spa->spa_dedup_ditto = intval; 6387b24ab676SJeff Bonwick break; 63880a4e9518Sgw default: 63890a4e9518Sgw break; 63900a4e9518Sgw } 6391990b4856Slling } 6392990b4856Slling 6393b1b8ab34Slling } 6394e14bb325SJeff Bonwick 6395e14bb325SJeff Bonwick mutex_exit(&spa->spa_props_lock); 6396b1b8ab34Slling } 6397b1b8ab34Slling 6398cde58dbcSMatthew Ahrens /* 6399cde58dbcSMatthew Ahrens * Perform one-time upgrade on-disk changes. spa_version() does not 6400cde58dbcSMatthew Ahrens * reflect the new version this txg, so there must be no changes this 6401cde58dbcSMatthew Ahrens * txg to anything that the upgrade code depends on after it executes. 6402cde58dbcSMatthew Ahrens * Therefore this must be called after dsl_pool_sync() does the sync 6403cde58dbcSMatthew Ahrens * tasks. 6404cde58dbcSMatthew Ahrens */ 6405cde58dbcSMatthew Ahrens static void 6406cde58dbcSMatthew Ahrens spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6407cde58dbcSMatthew Ahrens { 6408cde58dbcSMatthew Ahrens dsl_pool_t *dp = spa->spa_dsl_pool; 6409cde58dbcSMatthew Ahrens 6410cde58dbcSMatthew Ahrens ASSERT(spa->spa_sync_pass == 1); 6411cde58dbcSMatthew Ahrens 64123b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 64133b2aab18SMatthew Ahrens 6414cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6415cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6416cde58dbcSMatthew Ahrens dsl_pool_create_origin(dp, tx); 6417cde58dbcSMatthew Ahrens 6418cde58dbcSMatthew Ahrens /* Keeping the origin open increases spa_minref */ 6419cde58dbcSMatthew Ahrens spa->spa_minref += 3; 6420cde58dbcSMatthew Ahrens } 6421cde58dbcSMatthew Ahrens 6422cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6423cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6424cde58dbcSMatthew Ahrens dsl_pool_upgrade_clones(dp, tx); 6425cde58dbcSMatthew Ahrens } 6426cde58dbcSMatthew Ahrens 6427cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6428cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6429cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dp, tx); 6430cde58dbcSMatthew Ahrens 6431cde58dbcSMatthew Ahrens /* Keeping the freedir open increases spa_minref */ 6432cde58dbcSMatthew Ahrens spa->spa_minref += 3; 6433cde58dbcSMatthew Ahrens } 6434ad135b5dSChristopher Siden 6435ad135b5dSChristopher Siden if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6436ad135b5dSChristopher Siden spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6437ad135b5dSChristopher Siden spa_feature_create_zap_objects(spa, tx); 6438ad135b5dSChristopher Siden } 6439b8289d24SDaniil Lunev 6440b8289d24SDaniil Lunev /* 6441b8289d24SDaniil Lunev * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6442b8289d24SDaniil Lunev * when possibility to use lz4 compression for metadata was added 6443b8289d24SDaniil Lunev * Old pools that have this feature enabled must be upgraded to have 6444b8289d24SDaniil Lunev * this feature active 6445b8289d24SDaniil Lunev */ 6446b8289d24SDaniil Lunev if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6447b8289d24SDaniil Lunev boolean_t lz4_en = spa_feature_is_enabled(spa, 6448b8289d24SDaniil Lunev SPA_FEATURE_LZ4_COMPRESS); 6449b8289d24SDaniil Lunev boolean_t lz4_ac = spa_feature_is_active(spa, 6450b8289d24SDaniil Lunev SPA_FEATURE_LZ4_COMPRESS); 6451b8289d24SDaniil Lunev 6452b8289d24SDaniil Lunev if (lz4_en && !lz4_ac) 6453b8289d24SDaniil Lunev spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6454b8289d24SDaniil Lunev } 645545818ee1SMatthew Ahrens 645645818ee1SMatthew Ahrens /* 645745818ee1SMatthew Ahrens * If we haven't written the salt, do so now. Note that the 645845818ee1SMatthew Ahrens * feature may not be activated yet, but that's fine since 645945818ee1SMatthew Ahrens * the presence of this ZAP entry is backwards compatible. 646045818ee1SMatthew Ahrens */ 646145818ee1SMatthew Ahrens if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 646245818ee1SMatthew Ahrens DMU_POOL_CHECKSUM_SALT) == ENOENT) { 646345818ee1SMatthew Ahrens VERIFY0(zap_add(spa->spa_meta_objset, 646445818ee1SMatthew Ahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 646545818ee1SMatthew Ahrens sizeof (spa->spa_cksum_salt.zcs_bytes), 646645818ee1SMatthew Ahrens spa->spa_cksum_salt.zcs_bytes, tx)); 646745818ee1SMatthew Ahrens } 646845818ee1SMatthew Ahrens 64693b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 6470cde58dbcSMatthew Ahrens } 6471cde58dbcSMatthew Ahrens 6472fa9e4066Sahrens /* 6473fa9e4066Sahrens * Sync the specified transaction group. New blocks may be dirtied as 6474fa9e4066Sahrens * part of the process, so we iterate until it converges. 6475fa9e4066Sahrens */ 6476fa9e4066Sahrens void 6477fa9e4066Sahrens spa_sync(spa_t *spa, uint64_t txg) 6478fa9e4066Sahrens { 6479fa9e4066Sahrens dsl_pool_t *dp = spa->spa_dsl_pool; 6480fa9e4066Sahrens objset_t *mos = spa->spa_meta_objset; 6481b24ab676SJeff Bonwick bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 64820373e76bSbonwick vdev_t *rvd = spa->spa_root_vdev; 6483fa9e4066Sahrens vdev_t *vd; 6484fa9e4066Sahrens dmu_tx_t *tx; 6485e14bb325SJeff Bonwick int error; 6486fa9e4066Sahrens 6487f9af39baSGeorge Wilson VERIFY(spa_writeable(spa)); 6488f9af39baSGeorge Wilson 6489fa9e4066Sahrens /* 6490fa9e4066Sahrens * Lock out configuration changes. 6491fa9e4066Sahrens */ 6492e14bb325SJeff Bonwick spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6493fa9e4066Sahrens 6494fa9e4066Sahrens spa->spa_syncing_txg = txg; 6495fa9e4066Sahrens spa->spa_sync_pass = 0; 6496fa9e4066Sahrens 6497e14bb325SJeff Bonwick /* 6498e14bb325SJeff Bonwick * If there are any pending vdev state changes, convert them 6499e14bb325SJeff Bonwick * into config changes that go out with this transaction group. 6500e14bb325SJeff Bonwick */ 6501e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 65028ad4d6ddSJeff Bonwick while (list_head(&spa->spa_state_dirty_list) != NULL) { 65038ad4d6ddSJeff Bonwick /* 65048ad4d6ddSJeff Bonwick * We need the write lock here because, for aux vdevs, 65058ad4d6ddSJeff Bonwick * calling vdev_config_dirty() modifies sav_config. 65068ad4d6ddSJeff Bonwick * This is ugly and will become unnecessary when we 65078ad4d6ddSJeff Bonwick * eliminate the aux vdev wart by integrating all vdevs 65088ad4d6ddSJeff Bonwick * into the root vdev tree. 65098ad4d6ddSJeff Bonwick */ 65108ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 65118ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 65128ad4d6ddSJeff Bonwick while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 65138ad4d6ddSJeff Bonwick vdev_state_clean(vd); 65148ad4d6ddSJeff Bonwick vdev_config_dirty(vd); 65158ad4d6ddSJeff Bonwick } 65168ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 65178ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6518e14bb325SJeff Bonwick } 6519e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 6520e14bb325SJeff Bonwick 652199653d4eSeschrock tx = dmu_tx_create_assigned(dp, txg); 652299653d4eSeschrock 6523283b8460SGeorge.Wilson spa->spa_sync_starttime = gethrtime(); 6524283b8460SGeorge.Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6525283b8460SGeorge.Wilson spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6526283b8460SGeorge.Wilson 652799653d4eSeschrock /* 6528e7437265Sahrens * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 652999653d4eSeschrock * set spa_deflate if we have no raid-z vdevs. 653099653d4eSeschrock */ 6531e7437265Sahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6532e7437265Sahrens spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 653399653d4eSeschrock int i; 653499653d4eSeschrock 653599653d4eSeschrock for (i = 0; i < rvd->vdev_children; i++) { 653699653d4eSeschrock vd = rvd->vdev_child[i]; 653799653d4eSeschrock if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 653899653d4eSeschrock break; 653999653d4eSeschrock } 654099653d4eSeschrock if (i == rvd->vdev_children) { 654199653d4eSeschrock spa->spa_deflate = TRUE; 654299653d4eSeschrock VERIFY(0 == zap_add(spa->spa_meta_objset, 654399653d4eSeschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 654499653d4eSeschrock sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 654599653d4eSeschrock } 654699653d4eSeschrock } 654799653d4eSeschrock 6548fa9e4066Sahrens /* 6549fa9e4066Sahrens * Iterate to convergence. 6550fa9e4066Sahrens */ 6551fa9e4066Sahrens do { 6552b24ab676SJeff Bonwick int pass = ++spa->spa_sync_pass; 6553fa9e4066Sahrens 6554fa9e4066Sahrens spa_sync_config_object(spa, tx); 6555fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6556fa94a07fSbrendan ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6557fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6558fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6559ea8dc4b6Seschrock spa_errlog_sync(spa, txg); 6560fa9e4066Sahrens dsl_pool_sync(dp, txg); 6561fa9e4066Sahrens 656201f55e48SGeorge Wilson if (pass < zfs_sync_pass_deferred_free) { 656369962b56SMatthew Ahrens spa_sync_frees(spa, free_bpl, tx); 6564b24ab676SJeff Bonwick } else { 6565231aab85SMatthew Ahrens /* 6566231aab85SMatthew Ahrens * We can not defer frees in pass 1, because 6567231aab85SMatthew Ahrens * we sync the deferred frees later in pass 1. 6568231aab85SMatthew Ahrens */ 6569231aab85SMatthew Ahrens ASSERT3U(pass, >, 1); 6570cde58dbcSMatthew Ahrens bplist_iterate(free_bpl, bpobj_enqueue_cb, 657169962b56SMatthew Ahrens &spa->spa_deferred_bpobj, tx); 6572fa9e4066Sahrens } 6573fa9e4066Sahrens 6574b24ab676SJeff Bonwick ddt_sync(spa, txg); 65753f9d6ad7SLin Ling dsl_scan_sync(dp, tx); 6576afee20e4SGeorge Wilson 6577b24ab676SJeff Bonwick while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6578b24ab676SJeff Bonwick vdev_sync(vd, txg); 6579b24ab676SJeff Bonwick 6580231aab85SMatthew Ahrens if (pass == 1) { 6581cde58dbcSMatthew Ahrens spa_sync_upgrades(spa, tx); 6582231aab85SMatthew Ahrens ASSERT3U(txg, >=, 6583231aab85SMatthew Ahrens spa->spa_uberblock.ub_rootbp.blk_birth); 6584231aab85SMatthew Ahrens /* 6585231aab85SMatthew Ahrens * Note: We need to check if the MOS is dirty 6586231aab85SMatthew Ahrens * because we could have marked the MOS dirty 6587231aab85SMatthew Ahrens * without updating the uberblock (e.g. if we 6588231aab85SMatthew Ahrens * have sync tasks but no dirty user data). We 6589231aab85SMatthew Ahrens * need to check the uberblock's rootbp because 6590231aab85SMatthew Ahrens * it is updated if we have synced out dirty 6591231aab85SMatthew Ahrens * data (though in this case the MOS will most 6592231aab85SMatthew Ahrens * likely also be dirty due to second order 6593231aab85SMatthew Ahrens * effects, we don't want to rely on that here). 6594231aab85SMatthew Ahrens */ 6595231aab85SMatthew Ahrens if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6596231aab85SMatthew Ahrens !dmu_objset_is_dirty(mos, txg)) { 6597231aab85SMatthew Ahrens /* 6598231aab85SMatthew Ahrens * Nothing changed on the first pass, 6599231aab85SMatthew Ahrens * therefore this TXG is a no-op. Avoid 6600231aab85SMatthew Ahrens * syncing deferred frees, so that we 6601231aab85SMatthew Ahrens * can keep this TXG as a no-op. 6602231aab85SMatthew Ahrens */ 6603231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6604231aab85SMatthew Ahrens txg)); 6605231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6606231aab85SMatthew Ahrens ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6607231aab85SMatthew Ahrens break; 6608231aab85SMatthew Ahrens } 6609231aab85SMatthew Ahrens spa_sync_deferred_frees(spa, tx); 6610231aab85SMatthew Ahrens } 6611fa9e4066Sahrens 6612cde58dbcSMatthew Ahrens } while (dmu_objset_is_dirty(mos, txg)); 6613fa9e4066Sahrens 6614215198a6SJoe Stein if (!list_is_empty(&spa->spa_config_dirty_list)) { 6615215198a6SJoe Stein /* 6616215198a6SJoe Stein * Make sure that the number of ZAPs for all the vdevs matches 6617215198a6SJoe Stein * the number of ZAPs in the per-vdev ZAP list. This only gets 6618215198a6SJoe Stein * called if the config is dirty; otherwise there may be 6619215198a6SJoe Stein * outstanding AVZ operations that weren't completed in 6620215198a6SJoe Stein * spa_sync_config_object. 6621215198a6SJoe Stein */ 6622215198a6SJoe Stein uint64_t all_vdev_zap_entry_count; 6623215198a6SJoe Stein ASSERT0(zap_count(spa->spa_meta_objset, 6624215198a6SJoe Stein spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 6625215198a6SJoe Stein ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 6626215198a6SJoe Stein all_vdev_zap_entry_count); 6627215198a6SJoe Stein } 6628215198a6SJoe Stein 6629fa9e4066Sahrens /* 6630fa9e4066Sahrens * Rewrite the vdev configuration (which includes the uberblock) 6631fa9e4066Sahrens * to commit the transaction group. 66320373e76bSbonwick * 663317f17c2dSbonwick * If there are no dirty vdevs, we sync the uberblock to a few 663417f17c2dSbonwick * random top-level vdevs that are known to be visible in the 6635e14bb325SJeff Bonwick * config cache (see spa_vdev_add() for a complete description). 6636e14bb325SJeff Bonwick * If there *are* dirty vdevs, sync the uberblock to all vdevs. 66370373e76bSbonwick */ 6638e14bb325SJeff Bonwick for (;;) { 6639e14bb325SJeff Bonwick /* 6640e14bb325SJeff Bonwick * We hold SCL_STATE to prevent vdev open/close/etc. 6641e14bb325SJeff Bonwick * while we're attempting to write the vdev labels. 6642e14bb325SJeff Bonwick */ 6643e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6644e14bb325SJeff Bonwick 6645e14bb325SJeff Bonwick if (list_is_empty(&spa->spa_config_dirty_list)) { 6646e14bb325SJeff Bonwick vdev_t *svd[SPA_DVAS_PER_BP]; 6647e14bb325SJeff Bonwick int svdcount = 0; 6648e14bb325SJeff Bonwick int children = rvd->vdev_children; 6649e14bb325SJeff Bonwick int c0 = spa_get_random(children); 6650e14bb325SJeff Bonwick 6651573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 6652e14bb325SJeff Bonwick vd = rvd->vdev_child[(c0 + c) % children]; 6653e14bb325SJeff Bonwick if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6654e14bb325SJeff Bonwick continue; 6655e14bb325SJeff Bonwick svd[svdcount++] = vd; 6656e14bb325SJeff Bonwick if (svdcount == SPA_DVAS_PER_BP) 6657e14bb325SJeff Bonwick break; 6658e14bb325SJeff Bonwick } 6659eb5bb584SWill Andrews error = vdev_config_sync(svd, svdcount, txg); 6660e14bb325SJeff Bonwick } else { 6661e14bb325SJeff Bonwick error = vdev_config_sync(rvd->vdev_child, 6662eb5bb584SWill Andrews rvd->vdev_children, txg); 66630373e76bSbonwick } 6664e14bb325SJeff Bonwick 6665dfbb9432SGeorge Wilson if (error == 0) 6666dfbb9432SGeorge Wilson spa->spa_last_synced_guid = rvd->vdev_guid; 6667dfbb9432SGeorge Wilson 6668e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 6669e14bb325SJeff Bonwick 6670e14bb325SJeff Bonwick if (error == 0) 6671e14bb325SJeff Bonwick break; 6672e14bb325SJeff Bonwick zio_suspend(spa, NULL); 6673e14bb325SJeff Bonwick zio_resume_wait(spa); 66740373e76bSbonwick } 667599653d4eSeschrock dmu_tx_commit(tx); 667699653d4eSeschrock 6677283b8460SGeorge.Wilson VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6678283b8460SGeorge.Wilson 66790373e76bSbonwick /* 66800373e76bSbonwick * Clear the dirty config list. 6681fa9e4066Sahrens */ 6682e14bb325SJeff Bonwick while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 66830373e76bSbonwick vdev_config_clean(vd); 66840373e76bSbonwick 66850373e76bSbonwick /* 66860373e76bSbonwick * Now that the new config has synced transactionally, 66870373e76bSbonwick * let it become visible to the config cache. 66880373e76bSbonwick */ 66890373e76bSbonwick if (spa->spa_config_syncing != NULL) { 66900373e76bSbonwick spa_config_set(spa, spa->spa_config_syncing); 66910373e76bSbonwick spa->spa_config_txg = txg; 66920373e76bSbonwick spa->spa_config_syncing = NULL; 66930373e76bSbonwick } 6694fa9e4066Sahrens 6695fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 6696fa9e4066Sahrens 6697b24ab676SJeff Bonwick dsl_pool_sync_done(dp, txg); 6698fa9e4066Sahrens 6699fa9e4066Sahrens /* 6700fa9e4066Sahrens * Update usable space statistics. 6701fa9e4066Sahrens */ 6702fa9e4066Sahrens while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6703fa9e4066Sahrens vdev_sync_done(vd, txg); 6704fa9e4066Sahrens 6705485bbbf5SGeorge Wilson spa_update_dspace(spa); 6706485bbbf5SGeorge Wilson 6707fa9e4066Sahrens /* 6708fa9e4066Sahrens * It had better be the case that we didn't dirty anything 670999653d4eSeschrock * since vdev_config_sync(). 6710fa9e4066Sahrens */ 6711fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6712fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6713fa9e4066Sahrens ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6714b24ab676SJeff Bonwick 6715b24ab676SJeff Bonwick spa->spa_sync_pass = 0; 6716fa9e4066Sahrens 6717e14bb325SJeff Bonwick spa_config_exit(spa, SCL_CONFIG, FTAG); 6718ea8dc4b6Seschrock 6719468c413aSTim Haley spa_handle_ignored_writes(spa); 6720468c413aSTim Haley 6721ea8dc4b6Seschrock /* 6722ea8dc4b6Seschrock * If any async tasks have been requested, kick them off. 6723ea8dc4b6Seschrock */ 6724ea8dc4b6Seschrock spa_async_dispatch(spa); 6725fa9e4066Sahrens } 6726fa9e4066Sahrens 6727fa9e4066Sahrens /* 6728fa9e4066Sahrens * Sync all pools. We don't want to hold the namespace lock across these 6729fa9e4066Sahrens * operations, so we take a reference on the spa_t and drop the lock during the 6730fa9e4066Sahrens * sync. 6731fa9e4066Sahrens */ 6732fa9e4066Sahrens void 6733fa9e4066Sahrens spa_sync_allpools(void) 6734fa9e4066Sahrens { 6735fa9e4066Sahrens spa_t *spa = NULL; 6736fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6737fa9e4066Sahrens while ((spa = spa_next(spa)) != NULL) { 6738f9af39baSGeorge Wilson if (spa_state(spa) != POOL_STATE_ACTIVE || 6739f9af39baSGeorge Wilson !spa_writeable(spa) || spa_suspended(spa)) 6740fa9e4066Sahrens continue; 6741fa9e4066Sahrens spa_open_ref(spa, FTAG); 6742fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6743fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), 0); 6744fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6745fa9e4066Sahrens spa_close(spa, FTAG); 6746fa9e4066Sahrens } 6747fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6748fa9e4066Sahrens } 6749fa9e4066Sahrens 6750fa9e4066Sahrens /* 6751fa9e4066Sahrens * ========================================================================== 6752fa9e4066Sahrens * Miscellaneous routines 6753fa9e4066Sahrens * ========================================================================== 6754fa9e4066Sahrens */ 6755fa9e4066Sahrens 6756fa9e4066Sahrens /* 6757fa9e4066Sahrens * Remove all pools in the system. 6758fa9e4066Sahrens */ 6759fa9e4066Sahrens void 6760fa9e4066Sahrens spa_evict_all(void) 6761fa9e4066Sahrens { 6762fa9e4066Sahrens spa_t *spa; 6763fa9e4066Sahrens 6764fa9e4066Sahrens /* 6765fa9e4066Sahrens * Remove all cached state. All pools should be closed now, 6766fa9e4066Sahrens * so every spa in the AVL tree should be unreferenced. 6767fa9e4066Sahrens */ 6768fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6769fa9e4066Sahrens while ((spa = spa_next(NULL)) != NULL) { 6770fa9e4066Sahrens /* 6771ea8dc4b6Seschrock * Stop async tasks. The async thread may need to detach 6772ea8dc4b6Seschrock * a device that's been replaced, which requires grabbing 6773ea8dc4b6Seschrock * spa_namespace_lock, so we must drop it here. 6774fa9e4066Sahrens */ 6775fa9e4066Sahrens spa_open_ref(spa, FTAG); 6776fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6777ea8dc4b6Seschrock spa_async_suspend(spa); 6778fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 6779fa9e4066Sahrens spa_close(spa, FTAG); 6780fa9e4066Sahrens 6781fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6782fa9e4066Sahrens spa_unload(spa); 6783fa9e4066Sahrens spa_deactivate(spa); 6784fa9e4066Sahrens } 6785fa9e4066Sahrens spa_remove(spa); 6786fa9e4066Sahrens } 6787fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 6788fa9e4066Sahrens } 6789ea8dc4b6Seschrock 6790ea8dc4b6Seschrock vdev_t * 67916809eb4eSEric Schrock spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6792ea8dc4b6Seschrock { 6793c5904d13Seschrock vdev_t *vd; 6794c5904d13Seschrock int i; 6795c5904d13Seschrock 6796c5904d13Seschrock if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6797c5904d13Seschrock return (vd); 6798c5904d13Seschrock 67996809eb4eSEric Schrock if (aux) { 6800c5904d13Seschrock for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6801c5904d13Seschrock vd = spa->spa_l2cache.sav_vdevs[i]; 68026809eb4eSEric Schrock if (vd->vdev_guid == guid) 68036809eb4eSEric Schrock return (vd); 68046809eb4eSEric Schrock } 68056809eb4eSEric Schrock 68066809eb4eSEric Schrock for (i = 0; i < spa->spa_spares.sav_count; i++) { 68076809eb4eSEric Schrock vd = spa->spa_spares.sav_vdevs[i]; 6808c5904d13Seschrock if (vd->vdev_guid == guid) 6809c5904d13Seschrock return (vd); 6810c5904d13Seschrock } 6811c5904d13Seschrock } 6812c5904d13Seschrock 6813c5904d13Seschrock return (NULL); 6814ea8dc4b6Seschrock } 6815eaca9bbdSeschrock 6816eaca9bbdSeschrock void 6817990b4856Slling spa_upgrade(spa_t *spa, uint64_t version) 6818eaca9bbdSeschrock { 6819f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 6820f9af39baSGeorge Wilson 6821e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6822eaca9bbdSeschrock 6823eaca9bbdSeschrock /* 6824eaca9bbdSeschrock * This should only be called for a non-faulted pool, and since a 6825eaca9bbdSeschrock * future version would result in an unopenable pool, this shouldn't be 6826eaca9bbdSeschrock * possible. 6827eaca9bbdSeschrock */ 682862eae887SRichard Yao ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 68295d7b4d43SMatthew Ahrens ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6830eaca9bbdSeschrock 6831990b4856Slling spa->spa_uberblock.ub_version = version; 6832eaca9bbdSeschrock vdev_config_dirty(spa->spa_root_vdev); 6833eaca9bbdSeschrock 6834e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 683599653d4eSeschrock 683699653d4eSeschrock txg_wait_synced(spa_get_dsl(spa), 0); 683799653d4eSeschrock } 683899653d4eSeschrock 683999653d4eSeschrock boolean_t 684099653d4eSeschrock spa_has_spare(spa_t *spa, uint64_t guid) 684199653d4eSeschrock { 684299653d4eSeschrock int i; 684339c23413Seschrock uint64_t spareguid; 6844fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_spares; 684599653d4eSeschrock 6846fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 6847fa94a07fSbrendan if (sav->sav_vdevs[i]->vdev_guid == guid) 684899653d4eSeschrock return (B_TRUE); 684999653d4eSeschrock 6850fa94a07fSbrendan for (i = 0; i < sav->sav_npending; i++) { 6851fa94a07fSbrendan if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6852fa94a07fSbrendan &spareguid) == 0 && spareguid == guid) 685339c23413Seschrock return (B_TRUE); 685439c23413Seschrock } 685539c23413Seschrock 685699653d4eSeschrock return (B_FALSE); 6857eaca9bbdSeschrock } 6858b1b8ab34Slling 685989a89ebfSlling /* 686089a89ebfSlling * Check if a pool has an active shared spare device. 686189a89ebfSlling * Note: reference count of an active spare is 2, as a spare and as a replace 686289a89ebfSlling */ 686389a89ebfSlling static boolean_t 686489a89ebfSlling spa_has_active_shared_spare(spa_t *spa) 686589a89ebfSlling { 686689a89ebfSlling int i, refcnt; 686789a89ebfSlling uint64_t pool; 686889a89ebfSlling spa_aux_vdev_t *sav = &spa->spa_spares; 686989a89ebfSlling 687089a89ebfSlling for (i = 0; i < sav->sav_count; i++) { 687189a89ebfSlling if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 687289a89ebfSlling &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 687389a89ebfSlling refcnt > 2) 687489a89ebfSlling return (B_TRUE); 687589a89ebfSlling } 687689a89ebfSlling 687789a89ebfSlling return (B_FALSE); 687889a89ebfSlling } 687989a89ebfSlling 68803d7072f8Seschrock /* 68813d7072f8Seschrock * Post a sysevent corresponding to the given event. The 'name' must be one of 68823d7072f8Seschrock * the event definitions in sys/sysevent/eventdefs.h. The payload will be 68833d7072f8Seschrock * filled in from the spa and (optionally) the vdev. This doesn't do anything 68843d7072f8Seschrock * in the userland libzpool, as we don't want consumers to misinterpret ztest 68853d7072f8Seschrock * or zdb as real changes. 68863d7072f8Seschrock */ 68873d7072f8Seschrock void 68883d7072f8Seschrock spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 68893d7072f8Seschrock { 68903d7072f8Seschrock #ifdef _KERNEL 68913d7072f8Seschrock sysevent_t *ev; 68923d7072f8Seschrock sysevent_attr_list_t *attr = NULL; 68933d7072f8Seschrock sysevent_value_t value; 68943d7072f8Seschrock sysevent_id_t eid; 68953d7072f8Seschrock 68963d7072f8Seschrock ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 68973d7072f8Seschrock SE_SLEEP); 68983d7072f8Seschrock 68993d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 69003d7072f8Seschrock value.value.sv_string = spa_name(spa); 69013d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 69023d7072f8Seschrock goto done; 69033d7072f8Seschrock 69043d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 69053d7072f8Seschrock value.value.sv_uint64 = spa_guid(spa); 69063d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 69073d7072f8Seschrock goto done; 69083d7072f8Seschrock 69093d7072f8Seschrock if (vd) { 69103d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 69113d7072f8Seschrock value.value.sv_uint64 = vd->vdev_guid; 69123d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 69133d7072f8Seschrock SE_SLEEP) != 0) 69143d7072f8Seschrock goto done; 69153d7072f8Seschrock 69163d7072f8Seschrock if (vd->vdev_path) { 69173d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 69183d7072f8Seschrock value.value.sv_string = vd->vdev_path; 69193d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 69203d7072f8Seschrock &value, SE_SLEEP) != 0) 69213d7072f8Seschrock goto done; 69223d7072f8Seschrock } 69233d7072f8Seschrock } 69243d7072f8Seschrock 6925b01c3b58Seschrock if (sysevent_attach_attributes(ev, attr) != 0) 6926b01c3b58Seschrock goto done; 6927b01c3b58Seschrock attr = NULL; 6928b01c3b58Seschrock 69293d7072f8Seschrock (void) log_sysevent(ev, SE_SLEEP, &eid); 69303d7072f8Seschrock 69313d7072f8Seschrock done: 69323d7072f8Seschrock if (attr) 69333d7072f8Seschrock sysevent_free_attr(attr); 69343d7072f8Seschrock sysevent_free(ev); 69353d7072f8Seschrock #endif 69363d7072f8Seschrock } 6937