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. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens /* 27fa9e4066Sahrens * This file contains all the routines used when modifying on-disk SPA state. 28fa9e4066Sahrens * This includes opening, importing, destroying, exporting a pool, and syncing a 29fa9e4066Sahrens * pool. 30fa9e4066Sahrens */ 31fa9e4066Sahrens 32fa9e4066Sahrens #include <sys/zfs_context.h> 33ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 34fa9e4066Sahrens #include <sys/spa_impl.h> 35fa9e4066Sahrens #include <sys/zio.h> 36fa9e4066Sahrens #include <sys/zio_checksum.h> 37fa9e4066Sahrens #include <sys/dmu.h> 38fa9e4066Sahrens #include <sys/dmu_tx.h> 39fa9e4066Sahrens #include <sys/zap.h> 40fa9e4066Sahrens #include <sys/zil.h> 41b24ab676SJeff Bonwick #include <sys/ddt.h> 42fa9e4066Sahrens #include <sys/vdev_impl.h> 43fa9e4066Sahrens #include <sys/metaslab.h> 4488ecc943SGeorge Wilson #include <sys/metaslab_impl.h> 45fa9e4066Sahrens #include <sys/uberblock_impl.h> 46fa9e4066Sahrens #include <sys/txg.h> 47fa9e4066Sahrens #include <sys/avl.h> 48fa9e4066Sahrens #include <sys/dmu_traverse.h> 49b1b8ab34Slling #include <sys/dmu_objset.h> 50fa9e4066Sahrens #include <sys/unique.h> 51fa9e4066Sahrens #include <sys/dsl_pool.h> 52b1b8ab34Slling #include <sys/dsl_dataset.h> 53fa9e4066Sahrens #include <sys/dsl_dir.h> 54fa9e4066Sahrens #include <sys/dsl_prop.h> 55b1b8ab34Slling #include <sys/dsl_synctask.h> 56fa9e4066Sahrens #include <sys/fs/zfs.h> 57fa94a07fSbrendan #include <sys/arc.h> 58fa9e4066Sahrens #include <sys/callb.h> 5995173954Sek #include <sys/systeminfo.h> 60e7cbe64fSgw #include <sys/spa_boot.h> 61573ca77eSGeorge Wilson #include <sys/zfs_ioctl.h> 623f9d6ad7SLin Ling #include <sys/dsl_scan.h> 63fa9e4066Sahrens 645679c89fSjv #ifdef _KERNEL 65dedec472SJack Meng #include <sys/bootprops.h> 6635a5a358SJonathan Adams #include <sys/callb.h> 6735a5a358SJonathan Adams #include <sys/cpupart.h> 6835a5a358SJonathan Adams #include <sys/pool.h> 6935a5a358SJonathan Adams #include <sys/sysdc.h> 7035a5a358SJonathan Adams #include <sys/zone.h> 715679c89fSjv #endif /* _KERNEL */ 725679c89fSjv 73990b4856Slling #include "zfs_prop.h" 74b7b97454Sperrin #include "zfs_comutil.h" 75990b4856Slling 7635a5a358SJonathan Adams typedef enum zti_modes { 772e0c549eSJonathan Adams zti_mode_fixed, /* value is # of threads (min 1) */ 782e0c549eSJonathan Adams zti_mode_online_percent, /* value is % of online CPUs */ 7935a5a358SJonathan Adams zti_mode_batch, /* cpu-intensive; value is ignored */ 8080eb36f2SGeorge Wilson zti_mode_null, /* don't create a taskq */ 812e0c549eSJonathan Adams zti_nmodes 8235a5a358SJonathan Adams } zti_modes_t; 83416e0cd8Sek 8480eb36f2SGeorge Wilson #define ZTI_FIX(n) { zti_mode_fixed, (n) } 8580eb36f2SGeorge Wilson #define ZTI_PCT(n) { zti_mode_online_percent, (n) } 8635a5a358SJonathan Adams #define ZTI_BATCH { zti_mode_batch, 0 } 8780eb36f2SGeorge Wilson #define ZTI_NULL { zti_mode_null, 0 } 882e0c549eSJonathan Adams 8980eb36f2SGeorge Wilson #define ZTI_ONE ZTI_FIX(1) 902e0c549eSJonathan Adams 912e0c549eSJonathan Adams typedef struct zio_taskq_info { 9280eb36f2SGeorge Wilson enum zti_modes zti_mode; 9380eb36f2SGeorge Wilson uint_t zti_value; 942e0c549eSJonathan Adams } zio_taskq_info_t; 952e0c549eSJonathan Adams 962e0c549eSJonathan Adams static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 9735a5a358SJonathan Adams "issue", "issue_high", "intr", "intr_high" 982e0c549eSJonathan Adams }; 992e0c549eSJonathan Adams 10080eb36f2SGeorge Wilson /* 10180eb36f2SGeorge Wilson * Define the taskq threads for the following I/O types: 10280eb36f2SGeorge Wilson * NULL, READ, WRITE, FREE, CLAIM, and IOCTL 10380eb36f2SGeorge Wilson */ 10480eb36f2SGeorge Wilson const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 10580eb36f2SGeorge Wilson /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 10680eb36f2SGeorge Wilson { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 10735a5a358SJonathan Adams { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, 10835a5a358SJonathan Adams { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, 109c7cd2421SGeorge Wilson { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, 11080eb36f2SGeorge Wilson { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 11180eb36f2SGeorge Wilson { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 1122e0c549eSJonathan Adams }; 1132e0c549eSJonathan Adams 1143f9d6ad7SLin Ling static dsl_syncfunc_t spa_sync_props; 11589a89ebfSlling static boolean_t spa_has_active_shared_spare(spa_t *spa); 1161195e687SMark J Musante static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 1171195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 1181195e687SMark J Musante char **ereport); 119*cb04b873SMark J Musante static void spa_vdev_resilver_done(spa_t *spa); 120990b4856Slling 12135a5a358SJonathan Adams uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */ 12235a5a358SJonathan Adams id_t zio_taskq_psrset_bind = PS_NONE; 12335a5a358SJonathan Adams boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 12435a5a358SJonathan Adams uint_t zio_taskq_basedc = 80; /* base duty cycle */ 12535a5a358SJonathan Adams 12635a5a358SJonathan Adams boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 12735a5a358SJonathan Adams 12835a5a358SJonathan Adams /* 12935a5a358SJonathan Adams * This (illegal) pool name is used when temporarily importing a spa_t in order 13035a5a358SJonathan Adams * to get the vdev stats associated with the imported devices. 13135a5a358SJonathan Adams */ 13235a5a358SJonathan Adams #define TRYIMPORT_NAME "$import" 13335a5a358SJonathan Adams 134990b4856Slling /* 135990b4856Slling * ========================================================================== 136990b4856Slling * SPA properties routines 137990b4856Slling * ========================================================================== 138990b4856Slling */ 139990b4856Slling 140990b4856Slling /* 141990b4856Slling * Add a (source=src, propname=propval) list to an nvlist. 142990b4856Slling */ 1439d82f4f6Slling static void 144990b4856Slling spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 145990b4856Slling uint64_t intval, zprop_source_t src) 146990b4856Slling { 147990b4856Slling const char *propname = zpool_prop_to_name(prop); 148990b4856Slling nvlist_t *propval; 149990b4856Slling 1509d82f4f6Slling VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1519d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 152990b4856Slling 1539d82f4f6Slling if (strval != NULL) 1549d82f4f6Slling VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 1559d82f4f6Slling else 1569d82f4f6Slling VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 157990b4856Slling 1589d82f4f6Slling VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 159990b4856Slling nvlist_free(propval); 160990b4856Slling } 161990b4856Slling 162990b4856Slling /* 163990b4856Slling * Get property values from the spa configuration. 164990b4856Slling */ 1659d82f4f6Slling static void 166990b4856Slling spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 167990b4856Slling { 168379c004dSEric Schrock uint64_t size; 169485bbbf5SGeorge Wilson uint64_t alloc; 170990b4856Slling uint64_t cap, version; 171990b4856Slling zprop_source_t src = ZPROP_SRC_NONE; 172c5904d13Seschrock spa_config_dirent_t *dp; 173990b4856Slling 174e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 175e14bb325SJeff Bonwick 176379c004dSEric Schrock if (spa->spa_root_vdev != NULL) { 177485bbbf5SGeorge Wilson alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 178b24ab676SJeff Bonwick size = metaslab_class_get_space(spa_normal_class(spa)); 179379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 180379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 181485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 182485bbbf5SGeorge Wilson spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 183485bbbf5SGeorge Wilson size - alloc, src); 184379c004dSEric Schrock 185485bbbf5SGeorge Wilson cap = (size == 0) ? 0 : (alloc * 100 / size); 186379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 187379c004dSEric Schrock 188b24ab676SJeff Bonwick spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 189b24ab676SJeff Bonwick ddt_get_pool_dedup_ratio(spa), src); 190b24ab676SJeff Bonwick 191379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 192379c004dSEric Schrock spa->spa_root_vdev->vdev_state, src); 193379c004dSEric Schrock 194379c004dSEric Schrock version = spa_version(spa); 195379c004dSEric Schrock if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 196379c004dSEric Schrock src = ZPROP_SRC_DEFAULT; 197379c004dSEric Schrock else 198379c004dSEric Schrock src = ZPROP_SRC_LOCAL; 199379c004dSEric Schrock spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 200379c004dSEric Schrock } 201990b4856Slling 2029d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 203990b4856Slling 2049d82f4f6Slling if (spa->spa_root != NULL) 2059d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 2069d82f4f6Slling 0, ZPROP_SRC_LOCAL); 207990b4856Slling 208c5904d13Seschrock if ((dp = list_head(&spa->spa_config_list)) != NULL) { 209c5904d13Seschrock if (dp->scd_path == NULL) { 2109d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 211c5904d13Seschrock "none", 0, ZPROP_SRC_LOCAL); 212c5904d13Seschrock } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 2139d82f4f6Slling spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 214c5904d13Seschrock dp->scd_path, 0, ZPROP_SRC_LOCAL); 2152f8aaab3Seschrock } 2162f8aaab3Seschrock } 217990b4856Slling } 218990b4856Slling 219990b4856Slling /* 220990b4856Slling * Get zpool property values. 221990b4856Slling */ 222990b4856Slling int 223990b4856Slling spa_prop_get(spa_t *spa, nvlist_t **nvp) 224990b4856Slling { 225b24ab676SJeff Bonwick objset_t *mos = spa->spa_meta_objset; 226990b4856Slling zap_cursor_t zc; 227990b4856Slling zap_attribute_t za; 228990b4856Slling int err; 229990b4856Slling 2309d82f4f6Slling VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 231990b4856Slling 232e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 233e14bb325SJeff Bonwick 234990b4856Slling /* 235990b4856Slling * Get properties from the spa config. 236990b4856Slling */ 2379d82f4f6Slling spa_prop_get_config(spa, nvp); 238990b4856Slling 239990b4856Slling /* If no pool property object, no more prop to get. */ 240afee20e4SGeorge Wilson if (mos == NULL || spa->spa_pool_props_object == 0) { 241990b4856Slling mutex_exit(&spa->spa_props_lock); 242990b4856Slling return (0); 243990b4856Slling } 244990b4856Slling 245990b4856Slling /* 246990b4856Slling * Get properties from the MOS pool property object. 247990b4856Slling */ 248990b4856Slling for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 249990b4856Slling (err = zap_cursor_retrieve(&zc, &za)) == 0; 250990b4856Slling zap_cursor_advance(&zc)) { 251990b4856Slling uint64_t intval = 0; 252990b4856Slling char *strval = NULL; 253990b4856Slling zprop_source_t src = ZPROP_SRC_DEFAULT; 254990b4856Slling zpool_prop_t prop; 255990b4856Slling 256990b4856Slling if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 257990b4856Slling continue; 258990b4856Slling 259990b4856Slling switch (za.za_integer_length) { 260990b4856Slling case 8: 261990b4856Slling /* integer property */ 262990b4856Slling if (za.za_first_integer != 263990b4856Slling zpool_prop_default_numeric(prop)) 264990b4856Slling src = ZPROP_SRC_LOCAL; 265990b4856Slling 266990b4856Slling if (prop == ZPOOL_PROP_BOOTFS) { 267990b4856Slling dsl_pool_t *dp; 268990b4856Slling dsl_dataset_t *ds = NULL; 269990b4856Slling 270990b4856Slling dp = spa_get_dsl(spa); 271990b4856Slling rw_enter(&dp->dp_config_rwlock, RW_READER); 272745cd3c5Smaybee if (err = dsl_dataset_hold_obj(dp, 273745cd3c5Smaybee za.za_first_integer, FTAG, &ds)) { 274990b4856Slling rw_exit(&dp->dp_config_rwlock); 275990b4856Slling break; 276990b4856Slling } 277990b4856Slling 278990b4856Slling strval = kmem_alloc( 279990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 280990b4856Slling KM_SLEEP); 281990b4856Slling dsl_dataset_name(ds, strval); 282745cd3c5Smaybee dsl_dataset_rele(ds, FTAG); 283990b4856Slling rw_exit(&dp->dp_config_rwlock); 284990b4856Slling } else { 285990b4856Slling strval = NULL; 286990b4856Slling intval = za.za_first_integer; 287990b4856Slling } 288990b4856Slling 2899d82f4f6Slling spa_prop_add_list(*nvp, prop, strval, intval, src); 290990b4856Slling 291990b4856Slling if (strval != NULL) 292990b4856Slling kmem_free(strval, 293990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 294990b4856Slling 295990b4856Slling break; 296990b4856Slling 297990b4856Slling case 1: 298990b4856Slling /* string property */ 299990b4856Slling strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 300990b4856Slling err = zap_lookup(mos, spa->spa_pool_props_object, 301990b4856Slling za.za_name, 1, za.za_num_integers, strval); 302990b4856Slling if (err) { 303990b4856Slling kmem_free(strval, za.za_num_integers); 304990b4856Slling break; 305990b4856Slling } 3069d82f4f6Slling spa_prop_add_list(*nvp, prop, strval, 0, src); 307990b4856Slling kmem_free(strval, za.za_num_integers); 308990b4856Slling break; 309990b4856Slling 310990b4856Slling default: 311990b4856Slling break; 312990b4856Slling } 313990b4856Slling } 314990b4856Slling zap_cursor_fini(&zc); 315990b4856Slling mutex_exit(&spa->spa_props_lock); 316990b4856Slling out: 317990b4856Slling if (err && err != ENOENT) { 318990b4856Slling nvlist_free(*nvp); 3199d82f4f6Slling *nvp = NULL; 320990b4856Slling return (err); 321990b4856Slling } 322990b4856Slling 323990b4856Slling return (0); 324990b4856Slling } 325990b4856Slling 326990b4856Slling /* 327990b4856Slling * Validate the given pool properties nvlist and modify the list 328990b4856Slling * for the property values to be set. 329990b4856Slling */ 330990b4856Slling static int 331990b4856Slling spa_prop_validate(spa_t *spa, nvlist_t *props) 332990b4856Slling { 333990b4856Slling nvpair_t *elem; 334990b4856Slling int error = 0, reset_bootfs = 0; 335990b4856Slling uint64_t objnum; 336990b4856Slling 337990b4856Slling elem = NULL; 338990b4856Slling while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 339990b4856Slling zpool_prop_t prop; 340990b4856Slling char *propname, *strval; 341990b4856Slling uint64_t intval; 342990b4856Slling objset_t *os; 3432f8aaab3Seschrock char *slash; 344990b4856Slling 345990b4856Slling propname = nvpair_name(elem); 346990b4856Slling 347990b4856Slling if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) 348990b4856Slling return (EINVAL); 349990b4856Slling 350990b4856Slling switch (prop) { 351990b4856Slling case ZPOOL_PROP_VERSION: 352990b4856Slling error = nvpair_value_uint64(elem, &intval); 353990b4856Slling if (!error && 354990b4856Slling (intval < spa_version(spa) || intval > SPA_VERSION)) 355990b4856Slling error = EINVAL; 356990b4856Slling break; 357990b4856Slling 358990b4856Slling case ZPOOL_PROP_DELEGATION: 359990b4856Slling case ZPOOL_PROP_AUTOREPLACE: 360d5b5bb25SRich Morris case ZPOOL_PROP_LISTSNAPS: 361573ca77eSGeorge Wilson case ZPOOL_PROP_AUTOEXPAND: 362990b4856Slling error = nvpair_value_uint64(elem, &intval); 363990b4856Slling if (!error && intval > 1) 364990b4856Slling error = EINVAL; 365990b4856Slling break; 366990b4856Slling 367990b4856Slling case ZPOOL_PROP_BOOTFS: 36825f89ee2SJeff Bonwick /* 36925f89ee2SJeff Bonwick * If the pool version is less than SPA_VERSION_BOOTFS, 37025f89ee2SJeff Bonwick * or the pool is still being created (version == 0), 37125f89ee2SJeff Bonwick * the bootfs property cannot be set. 37225f89ee2SJeff Bonwick */ 373990b4856Slling if (spa_version(spa) < SPA_VERSION_BOOTFS) { 374990b4856Slling error = ENOTSUP; 375990b4856Slling break; 376990b4856Slling } 377990b4856Slling 378990b4856Slling /* 37915e6edf1Sgw * Make sure the vdev config is bootable 380990b4856Slling */ 38115e6edf1Sgw if (!vdev_is_bootable(spa->spa_root_vdev)) { 382990b4856Slling error = ENOTSUP; 383990b4856Slling break; 384990b4856Slling } 385990b4856Slling 386990b4856Slling reset_bootfs = 1; 387990b4856Slling 388990b4856Slling error = nvpair_value_string(elem, &strval); 389990b4856Slling 390990b4856Slling if (!error) { 39115e6edf1Sgw uint64_t compress; 39215e6edf1Sgw 393990b4856Slling if (strval == NULL || strval[0] == '\0') { 394990b4856Slling objnum = zpool_prop_default_numeric( 395990b4856Slling ZPOOL_PROP_BOOTFS); 396990b4856Slling break; 397990b4856Slling } 398990b4856Slling 399503ad85cSMatthew Ahrens if (error = dmu_objset_hold(strval, FTAG, &os)) 400990b4856Slling break; 40115e6edf1Sgw 402503ad85cSMatthew Ahrens /* Must be ZPL and not gzip compressed. */ 403503ad85cSMatthew Ahrens 404503ad85cSMatthew Ahrens if (dmu_objset_type(os) != DMU_OST_ZFS) { 405503ad85cSMatthew Ahrens error = ENOTSUP; 406503ad85cSMatthew Ahrens } else if ((error = dsl_prop_get_integer(strval, 40715e6edf1Sgw zfs_prop_to_name(ZFS_PROP_COMPRESSION), 40815e6edf1Sgw &compress, NULL)) == 0 && 40915e6edf1Sgw !BOOTFS_COMPRESS_VALID(compress)) { 41015e6edf1Sgw error = ENOTSUP; 41115e6edf1Sgw } else { 41215e6edf1Sgw objnum = dmu_objset_id(os); 41315e6edf1Sgw } 414503ad85cSMatthew Ahrens dmu_objset_rele(os, FTAG); 415990b4856Slling } 416990b4856Slling break; 417e14bb325SJeff Bonwick 4180a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 4190a4e9518Sgw error = nvpair_value_uint64(elem, &intval); 4200a4e9518Sgw if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 4210a4e9518Sgw intval > ZIO_FAILURE_MODE_PANIC)) 4220a4e9518Sgw error = EINVAL; 4230a4e9518Sgw 4240a4e9518Sgw /* 4250a4e9518Sgw * This is a special case which only occurs when 4260a4e9518Sgw * the pool has completely failed. This allows 4270a4e9518Sgw * the user to change the in-core failmode property 4280a4e9518Sgw * without syncing it out to disk (I/Os might 4290a4e9518Sgw * currently be blocked). We do this by returning 4300a4e9518Sgw * EIO to the caller (spa_prop_set) to trick it 4310a4e9518Sgw * into thinking we encountered a property validation 4320a4e9518Sgw * error. 4330a4e9518Sgw */ 434e14bb325SJeff Bonwick if (!error && spa_suspended(spa)) { 4350a4e9518Sgw spa->spa_failmode = intval; 4360a4e9518Sgw error = EIO; 4370a4e9518Sgw } 4380a4e9518Sgw break; 4392f8aaab3Seschrock 4402f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 4412f8aaab3Seschrock if ((error = nvpair_value_string(elem, &strval)) != 0) 4422f8aaab3Seschrock break; 4432f8aaab3Seschrock 4442f8aaab3Seschrock if (strval[0] == '\0') 4452f8aaab3Seschrock break; 4462f8aaab3Seschrock 4472f8aaab3Seschrock if (strcmp(strval, "none") == 0) 4482f8aaab3Seschrock break; 4492f8aaab3Seschrock 4502f8aaab3Seschrock if (strval[0] != '/') { 4512f8aaab3Seschrock error = EINVAL; 4522f8aaab3Seschrock break; 4532f8aaab3Seschrock } 4542f8aaab3Seschrock 4552f8aaab3Seschrock slash = strrchr(strval, '/'); 4562f8aaab3Seschrock ASSERT(slash != NULL); 4572f8aaab3Seschrock 4582f8aaab3Seschrock if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 4592f8aaab3Seschrock strcmp(slash, "/..") == 0) 4602f8aaab3Seschrock error = EINVAL; 4612f8aaab3Seschrock break; 462b24ab676SJeff Bonwick 463b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 464b24ab676SJeff Bonwick if (spa_version(spa) < SPA_VERSION_DEDUP) 465b24ab676SJeff Bonwick error = ENOTSUP; 466b24ab676SJeff Bonwick else 467b24ab676SJeff Bonwick error = nvpair_value_uint64(elem, &intval); 468b24ab676SJeff Bonwick if (error == 0 && 469b24ab676SJeff Bonwick intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 470b24ab676SJeff Bonwick error = EINVAL; 471b24ab676SJeff Bonwick break; 472990b4856Slling } 473990b4856Slling 474990b4856Slling if (error) 475990b4856Slling break; 476990b4856Slling } 477990b4856Slling 478990b4856Slling if (!error && reset_bootfs) { 479990b4856Slling error = nvlist_remove(props, 480990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 481990b4856Slling 482990b4856Slling if (!error) { 483990b4856Slling error = nvlist_add_uint64(props, 484990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 485990b4856Slling } 486990b4856Slling } 487990b4856Slling 488990b4856Slling return (error); 489990b4856Slling } 490990b4856Slling 491379c004dSEric Schrock void 492379c004dSEric Schrock spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 493379c004dSEric Schrock { 494379c004dSEric Schrock char *cachefile; 495379c004dSEric Schrock spa_config_dirent_t *dp; 496379c004dSEric Schrock 497379c004dSEric Schrock if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 498379c004dSEric Schrock &cachefile) != 0) 499379c004dSEric Schrock return; 500379c004dSEric Schrock 501379c004dSEric Schrock dp = kmem_alloc(sizeof (spa_config_dirent_t), 502379c004dSEric Schrock KM_SLEEP); 503379c004dSEric Schrock 504379c004dSEric Schrock if (cachefile[0] == '\0') 505379c004dSEric Schrock dp->scd_path = spa_strdup(spa_config_path); 506379c004dSEric Schrock else if (strcmp(cachefile, "none") == 0) 507379c004dSEric Schrock dp->scd_path = NULL; 508379c004dSEric Schrock else 509379c004dSEric Schrock dp->scd_path = spa_strdup(cachefile); 510379c004dSEric Schrock 511379c004dSEric Schrock list_insert_head(&spa->spa_config_list, dp); 512379c004dSEric Schrock if (need_sync) 513379c004dSEric Schrock spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 514379c004dSEric Schrock } 515379c004dSEric Schrock 516990b4856Slling int 517990b4856Slling spa_prop_set(spa_t *spa, nvlist_t *nvp) 518990b4856Slling { 519990b4856Slling int error; 520379c004dSEric Schrock nvpair_t *elem; 521379c004dSEric Schrock boolean_t need_sync = B_FALSE; 522379c004dSEric Schrock zpool_prop_t prop; 523990b4856Slling 524990b4856Slling if ((error = spa_prop_validate(spa, nvp)) != 0) 525990b4856Slling return (error); 526990b4856Slling 527379c004dSEric Schrock elem = NULL; 528379c004dSEric Schrock while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 529379c004dSEric Schrock if ((prop = zpool_name_to_prop( 530379c004dSEric Schrock nvpair_name(elem))) == ZPROP_INVAL) 531379c004dSEric Schrock return (EINVAL); 532379c004dSEric Schrock 533379c004dSEric Schrock if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT) 534379c004dSEric Schrock continue; 535379c004dSEric Schrock 536379c004dSEric Schrock need_sync = B_TRUE; 537379c004dSEric Schrock break; 538379c004dSEric Schrock } 539379c004dSEric Schrock 540379c004dSEric Schrock if (need_sync) 541379c004dSEric Schrock return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, 542379c004dSEric Schrock spa, nvp, 3)); 543379c004dSEric Schrock else 544379c004dSEric Schrock return (0); 545990b4856Slling } 546990b4856Slling 547990b4856Slling /* 548990b4856Slling * If the bootfs property value is dsobj, clear it. 549990b4856Slling */ 550990b4856Slling void 551990b4856Slling spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 552990b4856Slling { 553990b4856Slling if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 554990b4856Slling VERIFY(zap_remove(spa->spa_meta_objset, 555990b4856Slling spa->spa_pool_props_object, 556990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 557990b4856Slling spa->spa_bootfs = 0; 558990b4856Slling } 559990b4856Slling } 560990b4856Slling 561fa9e4066Sahrens /* 562fa9e4066Sahrens * ========================================================================== 563fa9e4066Sahrens * SPA state manipulation (open/create/destroy/import/export) 564fa9e4066Sahrens * ========================================================================== 565fa9e4066Sahrens */ 566fa9e4066Sahrens 567ea8dc4b6Seschrock static int 568ea8dc4b6Seschrock spa_error_entry_compare(const void *a, const void *b) 569ea8dc4b6Seschrock { 570ea8dc4b6Seschrock spa_error_entry_t *sa = (spa_error_entry_t *)a; 571ea8dc4b6Seschrock spa_error_entry_t *sb = (spa_error_entry_t *)b; 572ea8dc4b6Seschrock int ret; 573ea8dc4b6Seschrock 574ea8dc4b6Seschrock ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 575ea8dc4b6Seschrock sizeof (zbookmark_t)); 576ea8dc4b6Seschrock 577ea8dc4b6Seschrock if (ret < 0) 578ea8dc4b6Seschrock return (-1); 579ea8dc4b6Seschrock else if (ret > 0) 580ea8dc4b6Seschrock return (1); 581ea8dc4b6Seschrock else 582ea8dc4b6Seschrock return (0); 583ea8dc4b6Seschrock } 584ea8dc4b6Seschrock 585ea8dc4b6Seschrock /* 586ea8dc4b6Seschrock * Utility function which retrieves copies of the current logs and 587ea8dc4b6Seschrock * re-initializes them in the process. 588ea8dc4b6Seschrock */ 589ea8dc4b6Seschrock void 590ea8dc4b6Seschrock spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 591ea8dc4b6Seschrock { 592ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 593ea8dc4b6Seschrock 594ea8dc4b6Seschrock bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 595ea8dc4b6Seschrock bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 596ea8dc4b6Seschrock 597ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 598ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 599ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 600ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 601ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 602ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 603ea8dc4b6Seschrock } 604ea8dc4b6Seschrock 60535a5a358SJonathan Adams static taskq_t * 60635a5a358SJonathan Adams spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode, 60735a5a358SJonathan Adams uint_t value) 608fa9e4066Sahrens { 60935a5a358SJonathan Adams uint_t flags = TASKQ_PREPOPULATE; 61035a5a358SJonathan Adams boolean_t batch = B_FALSE; 611fa9e4066Sahrens 61235a5a358SJonathan Adams switch (mode) { 61335a5a358SJonathan Adams case zti_mode_null: 61435a5a358SJonathan Adams return (NULL); /* no taskq needed */ 615fa9e4066Sahrens 61635a5a358SJonathan Adams case zti_mode_fixed: 61735a5a358SJonathan Adams ASSERT3U(value, >=, 1); 61835a5a358SJonathan Adams value = MAX(value, 1); 61935a5a358SJonathan Adams break; 620fa9e4066Sahrens 62135a5a358SJonathan Adams case zti_mode_batch: 62235a5a358SJonathan Adams batch = B_TRUE; 62335a5a358SJonathan Adams flags |= TASKQ_THREADS_CPU_PCT; 62435a5a358SJonathan Adams value = zio_taskq_batch_pct; 62535a5a358SJonathan Adams break; 62635a5a358SJonathan Adams 62735a5a358SJonathan Adams case zti_mode_online_percent: 62835a5a358SJonathan Adams flags |= TASKQ_THREADS_CPU_PCT; 62935a5a358SJonathan Adams break; 63035a5a358SJonathan Adams 63135a5a358SJonathan Adams default: 63235a5a358SJonathan Adams panic("unrecognized mode for %s taskq (%u:%u) in " 63335a5a358SJonathan Adams "spa_activate()", 63435a5a358SJonathan Adams name, mode, value); 63535a5a358SJonathan Adams break; 63635a5a358SJonathan Adams } 63735a5a358SJonathan Adams 63835a5a358SJonathan Adams if (zio_taskq_sysdc && spa->spa_proc != &p0) { 63935a5a358SJonathan Adams if (batch) 64035a5a358SJonathan Adams flags |= TASKQ_DC_BATCH; 64135a5a358SJonathan Adams 64235a5a358SJonathan Adams return (taskq_create_sysdc(name, value, 50, INT_MAX, 64335a5a358SJonathan Adams spa->spa_proc, zio_taskq_basedc, flags)); 64435a5a358SJonathan Adams } 64535a5a358SJonathan Adams return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX, 64635a5a358SJonathan Adams spa->spa_proc, flags)); 64735a5a358SJonathan Adams } 64835a5a358SJonathan Adams 64935a5a358SJonathan Adams static void 65035a5a358SJonathan Adams spa_create_zio_taskqs(spa_t *spa) 65135a5a358SJonathan Adams { 652e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 653e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 65480eb36f2SGeorge Wilson const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 65580eb36f2SGeorge Wilson enum zti_modes mode = ztip->zti_mode; 65680eb36f2SGeorge Wilson uint_t value = ztip->zti_value; 6572e0c549eSJonathan Adams char name[32]; 6582e0c549eSJonathan Adams 6592e0c549eSJonathan Adams (void) snprintf(name, sizeof (name), 66080eb36f2SGeorge Wilson "%s_%s", zio_type_name[t], zio_taskq_types[q]); 6612e0c549eSJonathan Adams 66235a5a358SJonathan Adams spa->spa_zio_taskq[t][q] = 66335a5a358SJonathan Adams spa_taskq_create(spa, name, mode, value); 66435a5a358SJonathan Adams } 66535a5a358SJonathan Adams } 66635a5a358SJonathan Adams } 66735a5a358SJonathan Adams 66835a5a358SJonathan Adams #ifdef _KERNEL 66935a5a358SJonathan Adams static void 67035a5a358SJonathan Adams spa_thread(void *arg) 67135a5a358SJonathan Adams { 67235a5a358SJonathan Adams callb_cpr_t cprinfo; 6732e0c549eSJonathan Adams 67435a5a358SJonathan Adams spa_t *spa = arg; 67535a5a358SJonathan Adams user_t *pu = PTOU(curproc); 6762e0c549eSJonathan Adams 67735a5a358SJonathan Adams CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 67835a5a358SJonathan Adams spa->spa_name); 6792e0c549eSJonathan Adams 68035a5a358SJonathan Adams ASSERT(curproc != &p0); 68135a5a358SJonathan Adams (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 68235a5a358SJonathan Adams "zpool-%s", spa->spa_name); 68335a5a358SJonathan Adams (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 6842e0c549eSJonathan Adams 68535a5a358SJonathan Adams /* bind this thread to the requested psrset */ 68635a5a358SJonathan Adams if (zio_taskq_psrset_bind != PS_NONE) { 68735a5a358SJonathan Adams pool_lock(); 68835a5a358SJonathan Adams mutex_enter(&cpu_lock); 68935a5a358SJonathan Adams mutex_enter(&pidlock); 69035a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 69180eb36f2SGeorge Wilson 69235a5a358SJonathan Adams if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 69335a5a358SJonathan Adams 0, NULL, NULL) == 0) { 69435a5a358SJonathan Adams curthread->t_bind_pset = zio_taskq_psrset_bind; 69535a5a358SJonathan Adams } else { 69635a5a358SJonathan Adams cmn_err(CE_WARN, 69735a5a358SJonathan Adams "Couldn't bind process for zfs pool \"%s\" to " 69835a5a358SJonathan Adams "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 69935a5a358SJonathan Adams } 70035a5a358SJonathan Adams 70135a5a358SJonathan Adams mutex_exit(&curproc->p_lock); 70235a5a358SJonathan Adams mutex_exit(&pidlock); 70335a5a358SJonathan Adams mutex_exit(&cpu_lock); 70435a5a358SJonathan Adams pool_unlock(); 70535a5a358SJonathan Adams } 70635a5a358SJonathan Adams 70735a5a358SJonathan Adams if (zio_taskq_sysdc) { 70835a5a358SJonathan Adams sysdc_thread_enter(curthread, 100, 0); 70935a5a358SJonathan Adams } 71035a5a358SJonathan Adams 71135a5a358SJonathan Adams spa->spa_proc = curproc; 71235a5a358SJonathan Adams spa->spa_did = curthread->t_did; 71335a5a358SJonathan Adams 71435a5a358SJonathan Adams spa_create_zio_taskqs(spa); 71535a5a358SJonathan Adams 71635a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 71735a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 71835a5a358SJonathan Adams 71935a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_ACTIVE; 72035a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 72135a5a358SJonathan Adams 72235a5a358SJonathan Adams CALLB_CPR_SAFE_BEGIN(&cprinfo); 72335a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_ACTIVE) 72435a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 72535a5a358SJonathan Adams CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 72635a5a358SJonathan Adams 72735a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 72835a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_GONE; 72935a5a358SJonathan Adams spa->spa_proc = &p0; 73035a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 73135a5a358SJonathan Adams CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 73235a5a358SJonathan Adams 73335a5a358SJonathan Adams mutex_enter(&curproc->p_lock); 73435a5a358SJonathan Adams lwp_exit(); 73535a5a358SJonathan Adams } 73635a5a358SJonathan Adams #endif 73735a5a358SJonathan Adams 73835a5a358SJonathan Adams /* 73935a5a358SJonathan Adams * Activate an uninitialized pool. 74035a5a358SJonathan Adams */ 74135a5a358SJonathan Adams static void 74235a5a358SJonathan Adams spa_activate(spa_t *spa, int mode) 74335a5a358SJonathan Adams { 74435a5a358SJonathan Adams ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 74535a5a358SJonathan Adams 74635a5a358SJonathan Adams spa->spa_state = POOL_STATE_ACTIVE; 74735a5a358SJonathan Adams spa->spa_mode = mode; 74835a5a358SJonathan Adams 74935a5a358SJonathan Adams spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 75035a5a358SJonathan Adams spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 75135a5a358SJonathan Adams 75235a5a358SJonathan Adams /* Try to create a covering process */ 75335a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 75435a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 75535a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 75635a5a358SJonathan Adams spa->spa_did = 0; 75735a5a358SJonathan Adams 75835a5a358SJonathan Adams /* Only create a process if we're going to be around a while. */ 75935a5a358SJonathan Adams if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 76035a5a358SJonathan Adams if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 76135a5a358SJonathan Adams NULL, 0) == 0) { 76235a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_CREATED; 76335a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_CREATED) { 76435a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, 76535a5a358SJonathan Adams &spa->spa_proc_lock); 7662e0c549eSJonathan Adams } 76735a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 76835a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 76935a5a358SJonathan Adams ASSERT(spa->spa_did != 0); 77035a5a358SJonathan Adams } else { 77135a5a358SJonathan Adams #ifdef _KERNEL 77235a5a358SJonathan Adams cmn_err(CE_WARN, 77335a5a358SJonathan Adams "Couldn't create process for zfs pool \"%s\"\n", 77435a5a358SJonathan Adams spa->spa_name); 77535a5a358SJonathan Adams #endif 776e14bb325SJeff Bonwick } 777fa9e4066Sahrens } 77835a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 77935a5a358SJonathan Adams 78035a5a358SJonathan Adams /* If we didn't create a process, we need to create our taskqs. */ 78135a5a358SJonathan Adams if (spa->spa_proc == &p0) { 78235a5a358SJonathan Adams spa_create_zio_taskqs(spa); 78335a5a358SJonathan Adams } 784fa9e4066Sahrens 785e14bb325SJeff Bonwick list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 786e14bb325SJeff Bonwick offsetof(vdev_t, vdev_config_dirty_node)); 787e14bb325SJeff Bonwick list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 788e14bb325SJeff Bonwick offsetof(vdev_t, vdev_state_dirty_node)); 789fa9e4066Sahrens 790fa9e4066Sahrens txg_list_create(&spa->spa_vdev_txg_list, 791fa9e4066Sahrens offsetof(struct vdev, vdev_txg_node)); 792ea8dc4b6Seschrock 793ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 794ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 795ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 796ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 797ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 798ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 799fa9e4066Sahrens } 800fa9e4066Sahrens 801fa9e4066Sahrens /* 802fa9e4066Sahrens * Opposite of spa_activate(). 803fa9e4066Sahrens */ 804fa9e4066Sahrens static void 805fa9e4066Sahrens spa_deactivate(spa_t *spa) 806fa9e4066Sahrens { 807fa9e4066Sahrens ASSERT(spa->spa_sync_on == B_FALSE); 808fa9e4066Sahrens ASSERT(spa->spa_dsl_pool == NULL); 809fa9e4066Sahrens ASSERT(spa->spa_root_vdev == NULL); 81025f89ee2SJeff Bonwick ASSERT(spa->spa_async_zio_root == NULL); 811fa9e4066Sahrens ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 812fa9e4066Sahrens 813fa9e4066Sahrens txg_list_destroy(&spa->spa_vdev_txg_list); 814fa9e4066Sahrens 815e14bb325SJeff Bonwick list_destroy(&spa->spa_config_dirty_list); 816e14bb325SJeff Bonwick list_destroy(&spa->spa_state_dirty_list); 817fa9e4066Sahrens 818e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 819e14bb325SJeff Bonwick for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 82080eb36f2SGeorge Wilson if (spa->spa_zio_taskq[t][q] != NULL) 82180eb36f2SGeorge Wilson taskq_destroy(spa->spa_zio_taskq[t][q]); 822e14bb325SJeff Bonwick spa->spa_zio_taskq[t][q] = NULL; 823e14bb325SJeff Bonwick } 824fa9e4066Sahrens } 825fa9e4066Sahrens 826fa9e4066Sahrens metaslab_class_destroy(spa->spa_normal_class); 827fa9e4066Sahrens spa->spa_normal_class = NULL; 828fa9e4066Sahrens 8298654d025Sperrin metaslab_class_destroy(spa->spa_log_class); 8308654d025Sperrin spa->spa_log_class = NULL; 8318654d025Sperrin 832ea8dc4b6Seschrock /* 833ea8dc4b6Seschrock * If this was part of an import or the open otherwise failed, we may 834ea8dc4b6Seschrock * still have errors left in the queues. Empty them just in case. 835ea8dc4b6Seschrock */ 836ea8dc4b6Seschrock spa_errlog_drain(spa); 837ea8dc4b6Seschrock 838ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_scrub); 839ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_last); 840ea8dc4b6Seschrock 841fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 84235a5a358SJonathan Adams 84335a5a358SJonathan Adams mutex_enter(&spa->spa_proc_lock); 84435a5a358SJonathan Adams if (spa->spa_proc_state != SPA_PROC_NONE) { 84535a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 84635a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_DEACTIVATE; 84735a5a358SJonathan Adams cv_broadcast(&spa->spa_proc_cv); 84835a5a358SJonathan Adams while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 84935a5a358SJonathan Adams ASSERT(spa->spa_proc != &p0); 85035a5a358SJonathan Adams cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 85135a5a358SJonathan Adams } 85235a5a358SJonathan Adams ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 85335a5a358SJonathan Adams spa->spa_proc_state = SPA_PROC_NONE; 85435a5a358SJonathan Adams } 85535a5a358SJonathan Adams ASSERT(spa->spa_proc == &p0); 85635a5a358SJonathan Adams mutex_exit(&spa->spa_proc_lock); 85735a5a358SJonathan Adams 85835a5a358SJonathan Adams /* 85935a5a358SJonathan Adams * We want to make sure spa_thread() has actually exited the ZFS 86035a5a358SJonathan Adams * module, so that the module can't be unloaded out from underneath 86135a5a358SJonathan Adams * it. 86235a5a358SJonathan Adams */ 86335a5a358SJonathan Adams if (spa->spa_did != 0) { 86435a5a358SJonathan Adams thread_join(spa->spa_did); 86535a5a358SJonathan Adams spa->spa_did = 0; 86635a5a358SJonathan Adams } 867fa9e4066Sahrens } 868fa9e4066Sahrens 869fa9e4066Sahrens /* 870fa9e4066Sahrens * Verify a pool configuration, and construct the vdev tree appropriately. This 871fa9e4066Sahrens * will create all the necessary vdevs in the appropriate layout, with each vdev 872fa9e4066Sahrens * in the CLOSED state. This will prep the pool before open/creation/import. 873fa9e4066Sahrens * All vdev validation is done by the vdev_alloc() routine. 874fa9e4066Sahrens */ 87599653d4eSeschrock static int 87699653d4eSeschrock spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 87799653d4eSeschrock uint_t id, int atype) 878fa9e4066Sahrens { 879fa9e4066Sahrens nvlist_t **child; 880573ca77eSGeorge Wilson uint_t children; 88199653d4eSeschrock int error; 882fa9e4066Sahrens 88399653d4eSeschrock if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 88499653d4eSeschrock return (error); 885fa9e4066Sahrens 88699653d4eSeschrock if ((*vdp)->vdev_ops->vdev_op_leaf) 88799653d4eSeschrock return (0); 888fa9e4066Sahrens 889e14bb325SJeff Bonwick error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 890e14bb325SJeff Bonwick &child, &children); 891e14bb325SJeff Bonwick 892e14bb325SJeff Bonwick if (error == ENOENT) 893e14bb325SJeff Bonwick return (0); 894e14bb325SJeff Bonwick 895e14bb325SJeff Bonwick if (error) { 89699653d4eSeschrock vdev_free(*vdp); 89799653d4eSeschrock *vdp = NULL; 89899653d4eSeschrock return (EINVAL); 899fa9e4066Sahrens } 900fa9e4066Sahrens 901573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 90299653d4eSeschrock vdev_t *vd; 90399653d4eSeschrock if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 90499653d4eSeschrock atype)) != 0) { 90599653d4eSeschrock vdev_free(*vdp); 90699653d4eSeschrock *vdp = NULL; 90799653d4eSeschrock return (error); 908fa9e4066Sahrens } 909fa9e4066Sahrens } 910fa9e4066Sahrens 91199653d4eSeschrock ASSERT(*vdp != NULL); 91299653d4eSeschrock 91399653d4eSeschrock return (0); 914fa9e4066Sahrens } 915fa9e4066Sahrens 916fa9e4066Sahrens /* 917fa9e4066Sahrens * Opposite of spa_load(). 918fa9e4066Sahrens */ 919fa9e4066Sahrens static void 920fa9e4066Sahrens spa_unload(spa_t *spa) 921fa9e4066Sahrens { 92299653d4eSeschrock int i; 92399653d4eSeschrock 924e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 925e14bb325SJeff Bonwick 926ea8dc4b6Seschrock /* 927ea8dc4b6Seschrock * Stop async tasks. 928ea8dc4b6Seschrock */ 929ea8dc4b6Seschrock spa_async_suspend(spa); 930ea8dc4b6Seschrock 931fa9e4066Sahrens /* 932fa9e4066Sahrens * Stop syncing. 933fa9e4066Sahrens */ 934fa9e4066Sahrens if (spa->spa_sync_on) { 935fa9e4066Sahrens txg_sync_stop(spa->spa_dsl_pool); 936fa9e4066Sahrens spa->spa_sync_on = B_FALSE; 937fa9e4066Sahrens } 938fa9e4066Sahrens 939fa9e4066Sahrens /* 940e14bb325SJeff Bonwick * Wait for any outstanding async I/O to complete. 941fa9e4066Sahrens */ 94254d692b7SGeorge Wilson if (spa->spa_async_zio_root != NULL) { 94354d692b7SGeorge Wilson (void) zio_wait(spa->spa_async_zio_root); 94454d692b7SGeorge Wilson spa->spa_async_zio_root = NULL; 94554d692b7SGeorge Wilson } 946fa9e4066Sahrens 947cde58dbcSMatthew Ahrens bpobj_close(&spa->spa_deferred_bpobj); 948cde58dbcSMatthew Ahrens 949fa9e4066Sahrens /* 950fa9e4066Sahrens * Close the dsl pool. 951fa9e4066Sahrens */ 952fa9e4066Sahrens if (spa->spa_dsl_pool) { 953fa9e4066Sahrens dsl_pool_close(spa->spa_dsl_pool); 954fa9e4066Sahrens spa->spa_dsl_pool = NULL; 955afee20e4SGeorge Wilson spa->spa_meta_objset = NULL; 956fa9e4066Sahrens } 957fa9e4066Sahrens 958b24ab676SJeff Bonwick ddt_unload(spa); 959b24ab676SJeff Bonwick 9608ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 9618ad4d6ddSJeff Bonwick 9628ad4d6ddSJeff Bonwick /* 9638ad4d6ddSJeff Bonwick * Drop and purge level 2 cache 9648ad4d6ddSJeff Bonwick */ 9658ad4d6ddSJeff Bonwick spa_l2cache_drop(spa); 9668ad4d6ddSJeff Bonwick 967fa9e4066Sahrens /* 968fa9e4066Sahrens * Close all vdevs. 969fa9e4066Sahrens */ 9700e34b6a7Sbonwick if (spa->spa_root_vdev) 971fa9e4066Sahrens vdev_free(spa->spa_root_vdev); 9720e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == NULL); 973ea8dc4b6Seschrock 974fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 975fa94a07fSbrendan vdev_free(spa->spa_spares.sav_vdevs[i]); 976fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) { 977fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 978fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 979fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 98099653d4eSeschrock } 981fa94a07fSbrendan if (spa->spa_spares.sav_config) { 982fa94a07fSbrendan nvlist_free(spa->spa_spares.sav_config); 983fa94a07fSbrendan spa->spa_spares.sav_config = NULL; 984fa94a07fSbrendan } 9852ce8af81SEric Schrock spa->spa_spares.sav_count = 0; 986fa94a07fSbrendan 987fa94a07fSbrendan for (i = 0; i < spa->spa_l2cache.sav_count; i++) 988fa94a07fSbrendan vdev_free(spa->spa_l2cache.sav_vdevs[i]); 989fa94a07fSbrendan if (spa->spa_l2cache.sav_vdevs) { 990fa94a07fSbrendan kmem_free(spa->spa_l2cache.sav_vdevs, 991fa94a07fSbrendan spa->spa_l2cache.sav_count * sizeof (void *)); 992fa94a07fSbrendan spa->spa_l2cache.sav_vdevs = NULL; 993fa94a07fSbrendan } 994fa94a07fSbrendan if (spa->spa_l2cache.sav_config) { 995fa94a07fSbrendan nvlist_free(spa->spa_l2cache.sav_config); 996fa94a07fSbrendan spa->spa_l2cache.sav_config = NULL; 99799653d4eSeschrock } 9982ce8af81SEric Schrock spa->spa_l2cache.sav_count = 0; 99999653d4eSeschrock 1000ea8dc4b6Seschrock spa->spa_async_suspended = 0; 10018ad4d6ddSJeff Bonwick 10028ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1003fa9e4066Sahrens } 1004fa9e4066Sahrens 100599653d4eSeschrock /* 100699653d4eSeschrock * Load (or re-load) the current list of vdevs describing the active spares for 100799653d4eSeschrock * this pool. When this is called, we have some form of basic information in 1008fa94a07fSbrendan * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1009fa94a07fSbrendan * then re-generate a more complete list including status information. 101099653d4eSeschrock */ 101199653d4eSeschrock static void 101299653d4eSeschrock spa_load_spares(spa_t *spa) 101399653d4eSeschrock { 101499653d4eSeschrock nvlist_t **spares; 101599653d4eSeschrock uint_t nspares; 101699653d4eSeschrock int i; 101739c23413Seschrock vdev_t *vd, *tvd; 101899653d4eSeschrock 1019e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1020e14bb325SJeff Bonwick 102199653d4eSeschrock /* 102299653d4eSeschrock * First, close and free any existing spare vdevs. 102399653d4eSeschrock */ 1024fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 1025fa94a07fSbrendan vd = spa->spa_spares.sav_vdevs[i]; 102639c23413Seschrock 102739c23413Seschrock /* Undo the call to spa_activate() below */ 1028c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1029c5904d13Seschrock B_FALSE)) != NULL && tvd->vdev_isspare) 103039c23413Seschrock spa_spare_remove(tvd); 103139c23413Seschrock vdev_close(vd); 103239c23413Seschrock vdev_free(vd); 103399653d4eSeschrock } 103439c23413Seschrock 1035fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) 1036fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 1037fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 103899653d4eSeschrock 1039fa94a07fSbrendan if (spa->spa_spares.sav_config == NULL) 104099653d4eSeschrock nspares = 0; 104199653d4eSeschrock else 1042fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 104399653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 104499653d4eSeschrock 1045fa94a07fSbrendan spa->spa_spares.sav_count = (int)nspares; 1046fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 104799653d4eSeschrock 104899653d4eSeschrock if (nspares == 0) 104999653d4eSeschrock return; 105099653d4eSeschrock 105199653d4eSeschrock /* 105299653d4eSeschrock * Construct the array of vdevs, opening them to get status in the 105339c23413Seschrock * process. For each spare, there is potentially two different vdev_t 105439c23413Seschrock * structures associated with it: one in the list of spares (used only 105539c23413Seschrock * for basic validation purposes) and one in the active vdev 105639c23413Seschrock * configuration (if it's spared in). During this phase we open and 105739c23413Seschrock * validate each vdev on the spare list. If the vdev also exists in the 105839c23413Seschrock * active configuration, then we also mark this vdev as an active spare. 105999653d4eSeschrock */ 1060fa94a07fSbrendan spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1061fa94a07fSbrendan KM_SLEEP); 1062fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 106399653d4eSeschrock VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 106499653d4eSeschrock VDEV_ALLOC_SPARE) == 0); 106599653d4eSeschrock ASSERT(vd != NULL); 106699653d4eSeschrock 1067fa94a07fSbrendan spa->spa_spares.sav_vdevs[i] = vd; 106899653d4eSeschrock 1069c5904d13Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1070c5904d13Seschrock B_FALSE)) != NULL) { 107139c23413Seschrock if (!tvd->vdev_isspare) 107239c23413Seschrock spa_spare_add(tvd); 107339c23413Seschrock 107439c23413Seschrock /* 107539c23413Seschrock * We only mark the spare active if we were successfully 107639c23413Seschrock * able to load the vdev. Otherwise, importing a pool 107739c23413Seschrock * with a bad active spare would result in strange 107839c23413Seschrock * behavior, because multiple pool would think the spare 107939c23413Seschrock * is actively in use. 108039c23413Seschrock * 108139c23413Seschrock * There is a vulnerability here to an equally bizarre 108239c23413Seschrock * circumstance, where a dead active spare is later 108339c23413Seschrock * brought back to life (onlined or otherwise). Given 108439c23413Seschrock * the rarity of this scenario, and the extra complexity 108539c23413Seschrock * it adds, we ignore the possibility. 108639c23413Seschrock */ 108739c23413Seschrock if (!vdev_is_dead(tvd)) 108839c23413Seschrock spa_spare_activate(tvd); 108939c23413Seschrock } 109039c23413Seschrock 1091e14bb325SJeff Bonwick vd->vdev_top = vd; 10926809eb4eSEric Schrock vd->vdev_aux = &spa->spa_spares; 1093e14bb325SJeff Bonwick 109499653d4eSeschrock if (vdev_open(vd) != 0) 109599653d4eSeschrock continue; 109699653d4eSeschrock 1097fa94a07fSbrendan if (vdev_validate_aux(vd) == 0) 1098fa94a07fSbrendan spa_spare_add(vd); 109999653d4eSeschrock } 110099653d4eSeschrock 110199653d4eSeschrock /* 110299653d4eSeschrock * Recompute the stashed list of spares, with status information 110399653d4eSeschrock * this time. 110499653d4eSeschrock */ 1105fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 110699653d4eSeschrock DATA_TYPE_NVLIST_ARRAY) == 0); 110799653d4eSeschrock 1108fa94a07fSbrendan spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1109fa94a07fSbrendan KM_SLEEP); 1110fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 1111fa94a07fSbrendan spares[i] = vdev_config_generate(spa, 11123f9d6ad7SLin Ling spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1113fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1114fa94a07fSbrendan ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1115fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 111699653d4eSeschrock nvlist_free(spares[i]); 1117fa94a07fSbrendan kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1118fa94a07fSbrendan } 1119fa94a07fSbrendan 1120fa94a07fSbrendan /* 1121fa94a07fSbrendan * Load (or re-load) the current list of vdevs describing the active l2cache for 1122fa94a07fSbrendan * this pool. When this is called, we have some form of basic information in 1123fa94a07fSbrendan * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1124fa94a07fSbrendan * then re-generate a more complete list including status information. 1125fa94a07fSbrendan * Devices which are already active have their details maintained, and are 1126fa94a07fSbrendan * not re-opened. 1127fa94a07fSbrendan */ 1128fa94a07fSbrendan static void 1129fa94a07fSbrendan spa_load_l2cache(spa_t *spa) 1130fa94a07fSbrendan { 1131fa94a07fSbrendan nvlist_t **l2cache; 1132fa94a07fSbrendan uint_t nl2cache; 1133fa94a07fSbrendan int i, j, oldnvdevs; 1134573ca77eSGeorge Wilson uint64_t guid; 1135fa94a07fSbrendan vdev_t *vd, **oldvdevs, **newvdevs; 1136fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 1137fa94a07fSbrendan 1138e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1139e14bb325SJeff Bonwick 1140fa94a07fSbrendan if (sav->sav_config != NULL) { 1141fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1142fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1143fa94a07fSbrendan newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1144fa94a07fSbrendan } else { 1145fa94a07fSbrendan nl2cache = 0; 1146fa94a07fSbrendan } 1147fa94a07fSbrendan 1148fa94a07fSbrendan oldvdevs = sav->sav_vdevs; 1149fa94a07fSbrendan oldnvdevs = sav->sav_count; 1150fa94a07fSbrendan sav->sav_vdevs = NULL; 1151fa94a07fSbrendan sav->sav_count = 0; 1152fa94a07fSbrendan 1153fa94a07fSbrendan /* 1154fa94a07fSbrendan * Process new nvlist of vdevs. 1155fa94a07fSbrendan */ 1156fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 1157fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1158fa94a07fSbrendan &guid) == 0); 1159fa94a07fSbrendan 1160fa94a07fSbrendan newvdevs[i] = NULL; 1161fa94a07fSbrendan for (j = 0; j < oldnvdevs; j++) { 1162fa94a07fSbrendan vd = oldvdevs[j]; 1163fa94a07fSbrendan if (vd != NULL && guid == vd->vdev_guid) { 1164fa94a07fSbrendan /* 1165fa94a07fSbrendan * Retain previous vdev for add/remove ops. 1166fa94a07fSbrendan */ 1167fa94a07fSbrendan newvdevs[i] = vd; 1168fa94a07fSbrendan oldvdevs[j] = NULL; 1169fa94a07fSbrendan break; 1170fa94a07fSbrendan } 1171fa94a07fSbrendan } 1172fa94a07fSbrendan 1173fa94a07fSbrendan if (newvdevs[i] == NULL) { 1174fa94a07fSbrendan /* 1175fa94a07fSbrendan * Create new vdev 1176fa94a07fSbrendan */ 1177fa94a07fSbrendan VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1178fa94a07fSbrendan VDEV_ALLOC_L2CACHE) == 0); 1179fa94a07fSbrendan ASSERT(vd != NULL); 1180fa94a07fSbrendan newvdevs[i] = vd; 1181fa94a07fSbrendan 1182fa94a07fSbrendan /* 1183fa94a07fSbrendan * Commit this vdev as an l2cache device, 1184fa94a07fSbrendan * even if it fails to open. 1185fa94a07fSbrendan */ 1186fa94a07fSbrendan spa_l2cache_add(vd); 1187fa94a07fSbrendan 1188c5904d13Seschrock vd->vdev_top = vd; 1189c5904d13Seschrock vd->vdev_aux = sav; 1190c5904d13Seschrock 1191c5904d13Seschrock spa_l2cache_activate(vd); 1192c5904d13Seschrock 1193fa94a07fSbrendan if (vdev_open(vd) != 0) 1194fa94a07fSbrendan continue; 1195fa94a07fSbrendan 1196fa94a07fSbrendan (void) vdev_validate_aux(vd); 1197fa94a07fSbrendan 1198573ca77eSGeorge Wilson if (!vdev_is_dead(vd)) 1199573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1200fa94a07fSbrendan } 1201fa94a07fSbrendan } 1202fa94a07fSbrendan 1203fa94a07fSbrendan /* 1204fa94a07fSbrendan * Purge vdevs that were dropped 1205fa94a07fSbrendan */ 1206fa94a07fSbrendan for (i = 0; i < oldnvdevs; i++) { 1207fa94a07fSbrendan uint64_t pool; 1208fa94a07fSbrendan 1209fa94a07fSbrendan vd = oldvdevs[i]; 1210fa94a07fSbrendan if (vd != NULL) { 12118ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 12128ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 1213fa94a07fSbrendan l2arc_remove_vdev(vd); 1214fa94a07fSbrendan (void) vdev_close(vd); 1215fa94a07fSbrendan spa_l2cache_remove(vd); 1216fa94a07fSbrendan } 1217fa94a07fSbrendan } 1218fa94a07fSbrendan 1219fa94a07fSbrendan if (oldvdevs) 1220fa94a07fSbrendan kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1221fa94a07fSbrendan 1222fa94a07fSbrendan if (sav->sav_config == NULL) 1223fa94a07fSbrendan goto out; 1224fa94a07fSbrendan 1225fa94a07fSbrendan sav->sav_vdevs = newvdevs; 1226fa94a07fSbrendan sav->sav_count = (int)nl2cache; 1227fa94a07fSbrendan 1228fa94a07fSbrendan /* 1229fa94a07fSbrendan * Recompute the stashed list of l2cache devices, with status 1230fa94a07fSbrendan * information this time. 1231fa94a07fSbrendan */ 1232fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1233fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 1234fa94a07fSbrendan 1235fa94a07fSbrendan l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1236fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1237fa94a07fSbrendan l2cache[i] = vdev_config_generate(spa, 12383f9d6ad7SLin Ling sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1239fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1240fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1241fa94a07fSbrendan out: 1242fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 1243fa94a07fSbrendan nvlist_free(l2cache[i]); 1244fa94a07fSbrendan if (sav->sav_count) 1245fa94a07fSbrendan kmem_free(l2cache, sav->sav_count * sizeof (void *)); 124699653d4eSeschrock } 124799653d4eSeschrock 124899653d4eSeschrock static int 124999653d4eSeschrock load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 125099653d4eSeschrock { 125199653d4eSeschrock dmu_buf_t *db; 125299653d4eSeschrock char *packed = NULL; 125399653d4eSeschrock size_t nvsize = 0; 125499653d4eSeschrock int error; 125599653d4eSeschrock *value = NULL; 125699653d4eSeschrock 125799653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 125899653d4eSeschrock nvsize = *(uint64_t *)db->db_data; 125999653d4eSeschrock dmu_buf_rele(db, FTAG); 126099653d4eSeschrock 126199653d4eSeschrock packed = kmem_alloc(nvsize, KM_SLEEP); 12627bfdf011SNeil Perrin error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 12637bfdf011SNeil Perrin DMU_READ_PREFETCH); 126499653d4eSeschrock if (error == 0) 126599653d4eSeschrock error = nvlist_unpack(packed, nvsize, value, 0); 126699653d4eSeschrock kmem_free(packed, nvsize); 126799653d4eSeschrock 126899653d4eSeschrock return (error); 126999653d4eSeschrock } 127099653d4eSeschrock 12713d7072f8Seschrock /* 12723d7072f8Seschrock * Checks to see if the given vdev could not be opened, in which case we post a 12733d7072f8Seschrock * sysevent to notify the autoreplace code that the device has been removed. 12743d7072f8Seschrock */ 12753d7072f8Seschrock static void 12763d7072f8Seschrock spa_check_removed(vdev_t *vd) 12773d7072f8Seschrock { 1278573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 12793d7072f8Seschrock spa_check_removed(vd->vdev_child[c]); 12803d7072f8Seschrock 12813d7072f8Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 12823d7072f8Seschrock zfs_post_autoreplace(vd->vdev_spa, vd); 12833d7072f8Seschrock spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 12843d7072f8Seschrock } 12853d7072f8Seschrock } 12863d7072f8Seschrock 1287e6ca193dSGeorge Wilson /* 12884b964adaSGeorge Wilson * Validate the current config against the MOS config 1289e6ca193dSGeorge Wilson */ 12904b964adaSGeorge Wilson static boolean_t 12914b964adaSGeorge Wilson spa_config_valid(spa_t *spa, nvlist_t *config) 1292e6ca193dSGeorge Wilson { 12934b964adaSGeorge Wilson vdev_t *mrvd, *rvd = spa->spa_root_vdev; 12944b964adaSGeorge Wilson nvlist_t *nv; 12954b964adaSGeorge Wilson 12964b964adaSGeorge Wilson VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 12974b964adaSGeorge Wilson 12984b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 12994b964adaSGeorge Wilson VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 13004b964adaSGeorge Wilson 13014b964adaSGeorge Wilson ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1302e6ca193dSGeorge Wilson 130388ecc943SGeorge Wilson /* 13044b964adaSGeorge Wilson * If we're doing a normal import, then build up any additional 13054b964adaSGeorge Wilson * diagnostic information about missing devices in this config. 13064b964adaSGeorge Wilson * We'll pass this up to the user for further processing. 130788ecc943SGeorge Wilson */ 13084b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 13094b964adaSGeorge Wilson nvlist_t **child, *nv; 13104b964adaSGeorge Wilson uint64_t idx = 0; 13114b964adaSGeorge Wilson 13124b964adaSGeorge Wilson child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 13134b964adaSGeorge Wilson KM_SLEEP); 13144b964adaSGeorge Wilson VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1315e6ca193dSGeorge Wilson 13164b964adaSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 13174b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 13184b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 13194b964adaSGeorge Wilson 13204b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 13214b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops && 13224b964adaSGeorge Wilson mtvd->vdev_islog) 13234b964adaSGeorge Wilson child[idx++] = vdev_config_generate(spa, mtvd, 13244b964adaSGeorge Wilson B_FALSE, 0); 13254b964adaSGeorge Wilson } 13264b964adaSGeorge Wilson 13274b964adaSGeorge Wilson if (idx) { 13284b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist_array(nv, 13294b964adaSGeorge Wilson ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 13304b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(spa->spa_load_info, 13314b964adaSGeorge Wilson ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 13324b964adaSGeorge Wilson 13334b964adaSGeorge Wilson for (int i = 0; i < idx; i++) 13344b964adaSGeorge Wilson nvlist_free(child[i]); 13354b964adaSGeorge Wilson } 13364b964adaSGeorge Wilson nvlist_free(nv); 13374b964adaSGeorge Wilson kmem_free(child, rvd->vdev_children * sizeof (char **)); 13384b964adaSGeorge Wilson } 13394b964adaSGeorge Wilson 13404b964adaSGeorge Wilson /* 13414b964adaSGeorge Wilson * Compare the root vdev tree with the information we have 13424b964adaSGeorge Wilson * from the MOS config (mrvd). Check each top-level vdev 13434b964adaSGeorge Wilson * with the corresponding MOS config top-level (mtvd). 13444b964adaSGeorge Wilson */ 134588ecc943SGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 13464b964adaSGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 13474b964adaSGeorge Wilson vdev_t *mtvd = mrvd->vdev_child[c]; 13484b964adaSGeorge Wilson 13494b964adaSGeorge Wilson /* 13504b964adaSGeorge Wilson * Resolve any "missing" vdevs in the current configuration. 13514b964adaSGeorge Wilson * If we find that the MOS config has more accurate information 13524b964adaSGeorge Wilson * about the top-level vdev then use that vdev instead. 13534b964adaSGeorge Wilson */ 13544b964adaSGeorge Wilson if (tvd->vdev_ops == &vdev_missing_ops && 13554b964adaSGeorge Wilson mtvd->vdev_ops != &vdev_missing_ops) { 13564b964adaSGeorge Wilson 13574b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 13584b964adaSGeorge Wilson continue; 13594b964adaSGeorge Wilson 13604b964adaSGeorge Wilson /* 13614b964adaSGeorge Wilson * Device specific actions. 13624b964adaSGeorge Wilson */ 13634b964adaSGeorge Wilson if (mtvd->vdev_islog) { 13644b964adaSGeorge Wilson spa_set_log_state(spa, SPA_LOG_CLEAR); 13654b964adaSGeorge Wilson } else { 13664b964adaSGeorge Wilson /* 13674b964adaSGeorge Wilson * XXX - once we have 'readonly' pool 13684b964adaSGeorge Wilson * support we should be able to handle 13694b964adaSGeorge Wilson * missing data devices by transitioning 13704b964adaSGeorge Wilson * the pool to readonly. 13714b964adaSGeorge Wilson */ 13724b964adaSGeorge Wilson continue; 13734b964adaSGeorge Wilson } 13744b964adaSGeorge Wilson 13754b964adaSGeorge Wilson /* 13764b964adaSGeorge Wilson * Swap the missing vdev with the data we were 13774b964adaSGeorge Wilson * able to obtain from the MOS config. 13784b964adaSGeorge Wilson */ 13794b964adaSGeorge Wilson vdev_remove_child(rvd, tvd); 13804b964adaSGeorge Wilson vdev_remove_child(mrvd, mtvd); 13814b964adaSGeorge Wilson 13824b964adaSGeorge Wilson vdev_add_child(rvd, mtvd); 13834b964adaSGeorge Wilson vdev_add_child(mrvd, tvd); 13844b964adaSGeorge Wilson 13854b964adaSGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 13864b964adaSGeorge Wilson vdev_load(mtvd); 13874b964adaSGeorge Wilson spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 13884b964adaSGeorge Wilson 13894b964adaSGeorge Wilson vdev_reopen(rvd); 13904b964adaSGeorge Wilson } else if (mtvd->vdev_islog) { 13914b964adaSGeorge Wilson /* 13924b964adaSGeorge Wilson * Load the slog device's state from the MOS config 13934b964adaSGeorge Wilson * since it's possible that the label does not 13944b964adaSGeorge Wilson * contain the most up-to-date information. 13954b964adaSGeorge Wilson */ 13964b964adaSGeorge Wilson vdev_load_log_state(tvd, mtvd); 13974b964adaSGeorge Wilson vdev_reopen(tvd); 13984b964adaSGeorge Wilson } 1399e6ca193dSGeorge Wilson } 14004b964adaSGeorge Wilson vdev_free(mrvd); 140188ecc943SGeorge Wilson spa_config_exit(spa, SCL_ALL, FTAG); 14024b964adaSGeorge Wilson 14034b964adaSGeorge Wilson /* 14044b964adaSGeorge Wilson * Ensure we were able to validate the config. 14054b964adaSGeorge Wilson */ 14064b964adaSGeorge Wilson return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1407e6ca193dSGeorge Wilson } 1408e6ca193dSGeorge Wilson 1409b87f3af3Sperrin /* 1410b87f3af3Sperrin * Check for missing log devices 1411b87f3af3Sperrin */ 14124b964adaSGeorge Wilson static int 1413b87f3af3Sperrin spa_check_logs(spa_t *spa) 1414b87f3af3Sperrin { 1415b87f3af3Sperrin switch (spa->spa_log_state) { 1416b87f3af3Sperrin case SPA_LOG_MISSING: 1417b87f3af3Sperrin /* need to recheck in case slog has been restored */ 1418b87f3af3Sperrin case SPA_LOG_UNKNOWN: 1419b87f3af3Sperrin if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, 1420b87f3af3Sperrin DS_FIND_CHILDREN)) { 14211195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_MISSING); 1422b87f3af3Sperrin return (1); 1423b87f3af3Sperrin } 1424b87f3af3Sperrin break; 1425b87f3af3Sperrin } 1426b87f3af3Sperrin return (0); 1427b87f3af3Sperrin } 1428b87f3af3Sperrin 14291195e687SMark J Musante static boolean_t 14301195e687SMark J Musante spa_passivate_log(spa_t *spa) 14311195e687SMark J Musante { 14321195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 14331195e687SMark J Musante boolean_t slog_found = B_FALSE; 14341195e687SMark J Musante 14351195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 14361195e687SMark J Musante 14371195e687SMark J Musante if (!spa_has_slogs(spa)) 14381195e687SMark J Musante return (B_FALSE); 14391195e687SMark J Musante 14401195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 14411195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 14421195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 14431195e687SMark J Musante 14441195e687SMark J Musante if (tvd->vdev_islog) { 14451195e687SMark J Musante metaslab_group_passivate(mg); 14461195e687SMark J Musante slog_found = B_TRUE; 14471195e687SMark J Musante } 14481195e687SMark J Musante } 14491195e687SMark J Musante 14501195e687SMark J Musante return (slog_found); 14511195e687SMark J Musante } 14521195e687SMark J Musante 14531195e687SMark J Musante static void 14541195e687SMark J Musante spa_activate_log(spa_t *spa) 14551195e687SMark J Musante { 14561195e687SMark J Musante vdev_t *rvd = spa->spa_root_vdev; 14571195e687SMark J Musante 14581195e687SMark J Musante ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 14591195e687SMark J Musante 14601195e687SMark J Musante for (int c = 0; c < rvd->vdev_children; c++) { 14611195e687SMark J Musante vdev_t *tvd = rvd->vdev_child[c]; 14621195e687SMark J Musante metaslab_group_t *mg = tvd->vdev_mg; 14631195e687SMark J Musante 14641195e687SMark J Musante if (tvd->vdev_islog) 14651195e687SMark J Musante metaslab_group_activate(mg); 14661195e687SMark J Musante } 14671195e687SMark J Musante } 14681195e687SMark J Musante 14691195e687SMark J Musante int 14701195e687SMark J Musante spa_offline_log(spa_t *spa) 14711195e687SMark J Musante { 14721195e687SMark J Musante int error = 0; 14731195e687SMark J Musante 14741195e687SMark J Musante if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 14751195e687SMark J Musante NULL, DS_FIND_CHILDREN)) == 0) { 14761195e687SMark J Musante 14771195e687SMark J Musante /* 14781195e687SMark J Musante * We successfully offlined the log device, sync out the 14791195e687SMark J Musante * current txg so that the "stubby" block can be removed 14801195e687SMark J Musante * by zil_sync(). 14811195e687SMark J Musante */ 14821195e687SMark J Musante txg_wait_synced(spa->spa_dsl_pool, 0); 14831195e687SMark J Musante } 14841195e687SMark J Musante return (error); 14851195e687SMark J Musante } 14861195e687SMark J Musante 1487b693757aSEric Schrock static void 1488b693757aSEric Schrock spa_aux_check_removed(spa_aux_vdev_t *sav) 1489b693757aSEric Schrock { 1490b24ab676SJeff Bonwick for (int i = 0; i < sav->sav_count; i++) 1491b693757aSEric Schrock spa_check_removed(sav->sav_vdevs[i]); 1492b693757aSEric Schrock } 1493b693757aSEric Schrock 1494b24ab676SJeff Bonwick void 1495b24ab676SJeff Bonwick spa_claim_notify(zio_t *zio) 1496b24ab676SJeff Bonwick { 1497b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 1498b24ab676SJeff Bonwick 1499b24ab676SJeff Bonwick if (zio->io_error) 1500b24ab676SJeff Bonwick return; 1501b24ab676SJeff Bonwick 1502b24ab676SJeff Bonwick mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1503b24ab676SJeff Bonwick if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1504b24ab676SJeff Bonwick spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1505b24ab676SJeff Bonwick mutex_exit(&spa->spa_props_lock); 1506b24ab676SJeff Bonwick } 1507b24ab676SJeff Bonwick 1508468c413aSTim Haley typedef struct spa_load_error { 1509c8ee1847SVictor Latushkin uint64_t sle_meta_count; 1510468c413aSTim Haley uint64_t sle_data_count; 1511468c413aSTim Haley } spa_load_error_t; 1512468c413aSTim Haley 1513468c413aSTim Haley static void 1514468c413aSTim Haley spa_load_verify_done(zio_t *zio) 1515468c413aSTim Haley { 1516468c413aSTim Haley blkptr_t *bp = zio->io_bp; 1517468c413aSTim Haley spa_load_error_t *sle = zio->io_private; 1518468c413aSTim Haley dmu_object_type_t type = BP_GET_TYPE(bp); 1519468c413aSTim Haley int error = zio->io_error; 1520468c413aSTim Haley 1521468c413aSTim Haley if (error) { 1522468c413aSTim Haley if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) && 1523468c413aSTim Haley type != DMU_OT_INTENT_LOG) 1524c8ee1847SVictor Latushkin atomic_add_64(&sle->sle_meta_count, 1); 1525468c413aSTim Haley else 1526468c413aSTim Haley atomic_add_64(&sle->sle_data_count, 1); 1527468c413aSTim Haley } 1528468c413aSTim Haley zio_data_buf_free(zio->io_data, zio->io_size); 1529468c413aSTim Haley } 1530468c413aSTim Haley 1531468c413aSTim Haley /*ARGSUSED*/ 1532468c413aSTim Haley static int 1533b24ab676SJeff Bonwick spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 15343f9d6ad7SLin Ling arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1535468c413aSTim Haley { 1536468c413aSTim Haley if (bp != NULL) { 1537468c413aSTim Haley zio_t *rio = arg; 1538468c413aSTim Haley size_t size = BP_GET_PSIZE(bp); 1539468c413aSTim Haley void *data = zio_data_buf_alloc(size); 1540468c413aSTim Haley 1541468c413aSTim Haley zio_nowait(zio_read(rio, spa, bp, data, size, 1542468c413aSTim Haley spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1543468c413aSTim Haley ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1544468c413aSTim Haley ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1545468c413aSTim Haley } 1546468c413aSTim Haley return (0); 1547468c413aSTim Haley } 1548468c413aSTim Haley 1549468c413aSTim Haley static int 1550468c413aSTim Haley spa_load_verify(spa_t *spa) 1551468c413aSTim Haley { 1552468c413aSTim Haley zio_t *rio; 1553468c413aSTim Haley spa_load_error_t sle = { 0 }; 1554468c413aSTim Haley zpool_rewind_policy_t policy; 1555468c413aSTim Haley boolean_t verify_ok = B_FALSE; 1556468c413aSTim Haley int error; 1557468c413aSTim Haley 1558c8ee1847SVictor Latushkin zpool_get_rewind_policy(spa->spa_config, &policy); 1559c8ee1847SVictor Latushkin 1560c8ee1847SVictor Latushkin if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1561c8ee1847SVictor Latushkin return (0); 1562c8ee1847SVictor Latushkin 1563468c413aSTim Haley rio = zio_root(spa, NULL, &sle, 1564468c413aSTim Haley ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1565468c413aSTim Haley 1566bbfd46c4SJeff Bonwick error = traverse_pool(spa, spa->spa_verify_min_txg, 1567bbfd46c4SJeff Bonwick TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1568468c413aSTim Haley 1569468c413aSTim Haley (void) zio_wait(rio); 1570468c413aSTim Haley 1571c8ee1847SVictor Latushkin spa->spa_load_meta_errors = sle.sle_meta_count; 1572468c413aSTim Haley spa->spa_load_data_errors = sle.sle_data_count; 1573468c413aSTim Haley 1574c8ee1847SVictor Latushkin if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1575468c413aSTim Haley sle.sle_data_count <= policy.zrp_maxdata) { 15764b964adaSGeorge Wilson int64_t loss = 0; 15774b964adaSGeorge Wilson 1578468c413aSTim Haley verify_ok = B_TRUE; 1579468c413aSTim Haley spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1580468c413aSTim Haley spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 15814b964adaSGeorge Wilson 15824b964adaSGeorge Wilson loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 15834b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 15844b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 15854b964adaSGeorge Wilson VERIFY(nvlist_add_int64(spa->spa_load_info, 15864b964adaSGeorge Wilson ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 15874b964adaSGeorge Wilson VERIFY(nvlist_add_uint64(spa->spa_load_info, 15884b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1589a33cae98STim Haley } else { 1590a33cae98STim Haley spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1591468c413aSTim Haley } 1592468c413aSTim Haley 1593468c413aSTim Haley if (error) { 1594468c413aSTim Haley if (error != ENXIO && error != EIO) 1595468c413aSTim Haley error = EIO; 1596468c413aSTim Haley return (error); 1597468c413aSTim Haley } 1598468c413aSTim Haley 1599468c413aSTim Haley return (verify_ok ? 0 : EIO); 1600468c413aSTim Haley } 1601468c413aSTim Haley 16021195e687SMark J Musante /* 16031195e687SMark J Musante * Find a value in the pool props object. 16041195e687SMark J Musante */ 16051195e687SMark J Musante static void 16061195e687SMark J Musante spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 16071195e687SMark J Musante { 16081195e687SMark J Musante (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 16091195e687SMark J Musante zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 16101195e687SMark J Musante } 16111195e687SMark J Musante 16121195e687SMark J Musante /* 16131195e687SMark J Musante * Find a value in the pool directory object. 16141195e687SMark J Musante */ 16151195e687SMark J Musante static int 16161195e687SMark J Musante spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 16171195e687SMark J Musante { 16181195e687SMark J Musante return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 16191195e687SMark J Musante name, sizeof (uint64_t), 1, val)); 16201195e687SMark J Musante } 16211195e687SMark J Musante 16221195e687SMark J Musante static int 16231195e687SMark J Musante spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 16241195e687SMark J Musante { 16251195e687SMark J Musante vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 16261195e687SMark J Musante return (err); 16271195e687SMark J Musante } 16281195e687SMark J Musante 16291195e687SMark J Musante /* 16301195e687SMark J Musante * Fix up config after a partly-completed split. This is done with the 16311195e687SMark J Musante * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 16321195e687SMark J Musante * pool have that entry in their config, but only the splitting one contains 16331195e687SMark J Musante * a list of all the guids of the vdevs that are being split off. 16341195e687SMark J Musante * 16351195e687SMark J Musante * This function determines what to do with that list: either rejoin 16361195e687SMark J Musante * all the disks to the pool, or complete the splitting process. To attempt 16371195e687SMark J Musante * the rejoin, each disk that is offlined is marked online again, and 16381195e687SMark J Musante * we do a reopen() call. If the vdev label for every disk that was 16391195e687SMark J Musante * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 16401195e687SMark J Musante * then we call vdev_split() on each disk, and complete the split. 16411195e687SMark J Musante * 1642d41c4376SMark J Musante * Otherwise we leave the config alone, with all the vdevs in place in 1643d41c4376SMark J Musante * the original pool. 16441195e687SMark J Musante */ 16451195e687SMark J Musante static void 16461195e687SMark J Musante spa_try_repair(spa_t *spa, nvlist_t *config) 16471195e687SMark J Musante { 16481195e687SMark J Musante uint_t extracted; 16491195e687SMark J Musante uint64_t *glist; 16501195e687SMark J Musante uint_t i, gcount; 16511195e687SMark J Musante nvlist_t *nvl; 16521195e687SMark J Musante vdev_t **vd; 16531195e687SMark J Musante boolean_t attempt_reopen; 16541195e687SMark J Musante 16551195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 16561195e687SMark J Musante return; 16571195e687SMark J Musante 16581195e687SMark J Musante /* check that the config is complete */ 16591195e687SMark J Musante if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 16601195e687SMark J Musante &glist, &gcount) != 0) 16611195e687SMark J Musante return; 16621195e687SMark J Musante 16631195e687SMark J Musante vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 16641195e687SMark J Musante 16651195e687SMark J Musante /* attempt to online all the vdevs & validate */ 16661195e687SMark J Musante attempt_reopen = B_TRUE; 16671195e687SMark J Musante for (i = 0; i < gcount; i++) { 16681195e687SMark J Musante if (glist[i] == 0) /* vdev is hole */ 16691195e687SMark J Musante continue; 16701195e687SMark J Musante 16711195e687SMark J Musante vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 16721195e687SMark J Musante if (vd[i] == NULL) { 16731195e687SMark J Musante /* 16741195e687SMark J Musante * Don't bother attempting to reopen the disks; 16751195e687SMark J Musante * just do the split. 16761195e687SMark J Musante */ 16771195e687SMark J Musante attempt_reopen = B_FALSE; 16781195e687SMark J Musante } else { 16791195e687SMark J Musante /* attempt to re-online it */ 16801195e687SMark J Musante vd[i]->vdev_offline = B_FALSE; 16811195e687SMark J Musante } 16821195e687SMark J Musante } 16831195e687SMark J Musante 16841195e687SMark J Musante if (attempt_reopen) { 16851195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 16861195e687SMark J Musante 16871195e687SMark J Musante /* check each device to see what state it's in */ 16881195e687SMark J Musante for (extracted = 0, i = 0; i < gcount; i++) { 16891195e687SMark J Musante if (vd[i] != NULL && 16901195e687SMark J Musante vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 16911195e687SMark J Musante break; 16921195e687SMark J Musante ++extracted; 16931195e687SMark J Musante } 16941195e687SMark J Musante } 16951195e687SMark J Musante 16961195e687SMark J Musante /* 16971195e687SMark J Musante * If every disk has been moved to the new pool, or if we never 16981195e687SMark J Musante * even attempted to look at them, then we split them off for 16991195e687SMark J Musante * good. 17001195e687SMark J Musante */ 17011195e687SMark J Musante if (!attempt_reopen || gcount == extracted) { 17021195e687SMark J Musante for (i = 0; i < gcount; i++) 17031195e687SMark J Musante if (vd[i] != NULL) 17041195e687SMark J Musante vdev_split(vd[i]); 17051195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 17061195e687SMark J Musante } 17071195e687SMark J Musante 17081195e687SMark J Musante kmem_free(vd, gcount * sizeof (vdev_t *)); 17091195e687SMark J Musante } 17101195e687SMark J Musante 17111195e687SMark J Musante static int 17121195e687SMark J Musante spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 17131195e687SMark J Musante boolean_t mosconfig) 17141195e687SMark J Musante { 17151195e687SMark J Musante nvlist_t *config = spa->spa_config; 17161195e687SMark J Musante char *ereport = FM_EREPORT_ZFS_POOL; 17171195e687SMark J Musante int error; 17181195e687SMark J Musante uint64_t pool_guid; 17191195e687SMark J Musante nvlist_t *nvl; 17201195e687SMark J Musante 17211195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 17221195e687SMark J Musante return (EINVAL); 17231195e687SMark J Musante 17241195e687SMark J Musante /* 17251195e687SMark J Musante * Versioning wasn't explicitly added to the label until later, so if 17261195e687SMark J Musante * it's not present treat it as the initial version. 17271195e687SMark J Musante */ 17281195e687SMark J Musante if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 17291195e687SMark J Musante &spa->spa_ubsync.ub_version) != 0) 17301195e687SMark J Musante spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 17311195e687SMark J Musante 17321195e687SMark J Musante (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 17331195e687SMark J Musante &spa->spa_config_txg); 17341195e687SMark J Musante 17351195e687SMark J Musante if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 17361195e687SMark J Musante spa_guid_exists(pool_guid, 0)) { 17371195e687SMark J Musante error = EEXIST; 17381195e687SMark J Musante } else { 17391195e687SMark J Musante spa->spa_load_guid = pool_guid; 17401195e687SMark J Musante 17411195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 17421195e687SMark J Musante &nvl) == 0) { 17431195e687SMark J Musante VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 17441195e687SMark J Musante KM_SLEEP) == 0); 17451195e687SMark J Musante } 17461195e687SMark J Musante 174711027bc7STim Haley gethrestime(&spa->spa_loaded_ts); 17481195e687SMark J Musante error = spa_load_impl(spa, pool_guid, config, state, type, 17491195e687SMark J Musante mosconfig, &ereport); 17501195e687SMark J Musante } 17511195e687SMark J Musante 17521195e687SMark J Musante spa->spa_minref = refcount_count(&spa->spa_refcount); 175311027bc7STim Haley if (error) { 175411027bc7STim Haley if (error != EEXIST) { 175511027bc7STim Haley spa->spa_loaded_ts.tv_sec = 0; 175611027bc7STim Haley spa->spa_loaded_ts.tv_nsec = 0; 175711027bc7STim Haley } 175811027bc7STim Haley if (error != EBADF) { 175911027bc7STim Haley zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 176011027bc7STim Haley } 176111027bc7STim Haley } 17621195e687SMark J Musante spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 17631195e687SMark J Musante spa->spa_ena = 0; 17641195e687SMark J Musante 17651195e687SMark J Musante return (error); 17661195e687SMark J Musante } 17671195e687SMark J Musante 1768fa9e4066Sahrens /* 1769fa9e4066Sahrens * Load an existing storage pool, using the pool's builtin spa_config as a 1770ea8dc4b6Seschrock * source of configuration information. 1771fa9e4066Sahrens */ 1772fa9e4066Sahrens static int 17731195e687SMark J Musante spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 17741195e687SMark J Musante spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 17751195e687SMark J Musante char **ereport) 1776fa9e4066Sahrens { 1777fa9e4066Sahrens int error = 0; 1778871a9500SMark J Musante nvlist_t *nvroot = NULL; 1779fa9e4066Sahrens vdev_t *rvd; 1780fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 17814b964adaSGeorge Wilson uint64_t children, config_cache_txg = spa->spa_config_txg; 17828ad4d6ddSJeff Bonwick int orig_mode = spa->spa_mode; 17831195e687SMark J Musante int parse; 1784cde58dbcSMatthew Ahrens uint64_t obj; 1785fa9e4066Sahrens 17868ad4d6ddSJeff Bonwick /* 17878ad4d6ddSJeff Bonwick * If this is an untrusted config, access the pool in read-only mode. 17888ad4d6ddSJeff Bonwick * This prevents things like resilvering recently removed devices. 17898ad4d6ddSJeff Bonwick */ 17908ad4d6ddSJeff Bonwick if (!mosconfig) 17918ad4d6ddSJeff Bonwick spa->spa_mode = FREAD; 17928ad4d6ddSJeff Bonwick 1793e14bb325SJeff Bonwick ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1794e14bb325SJeff Bonwick 1795ea8dc4b6Seschrock spa->spa_load_state = state; 17960373e76bSbonwick 17971195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 17981195e687SMark J Musante return (EINVAL); 1799fa9e4066Sahrens 18001195e687SMark J Musante parse = (type == SPA_IMPORT_EXISTING ? 18011195e687SMark J Musante VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 1802b5989ec7Seschrock 180354d692b7SGeorge Wilson /* 180454d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 180554d692b7SGeorge Wilson */ 180625f89ee2SJeff Bonwick spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 180725f89ee2SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 180854d692b7SGeorge Wilson 1809fa9e4066Sahrens /* 181099653d4eSeschrock * Parse the configuration into a vdev tree. We explicitly set the 181199653d4eSeschrock * value that will be returned by spa_version() since parsing the 181299653d4eSeschrock * configuration requires knowing the version number. 1813fa9e4066Sahrens */ 1814e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 18151195e687SMark J Musante error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 1816e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 1817fa9e4066Sahrens 181899653d4eSeschrock if (error != 0) 18191195e687SMark J Musante return (error); 1820fa9e4066Sahrens 18210e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == rvd); 18221195e687SMark J Musante 18231195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 18241195e687SMark J Musante ASSERT(spa_guid(spa) == pool_guid); 18251195e687SMark J Musante } 1826fa9e4066Sahrens 1827fa9e4066Sahrens /* 1828fa9e4066Sahrens * Try to open all vdevs, loading each label in the process. 1829fa9e4066Sahrens */ 1830e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 18310bf246f5Smc error = vdev_open(rvd); 1832e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 18330bf246f5Smc if (error != 0) 18341195e687SMark J Musante return (error); 1835fa9e4066Sahrens 1836560e6e96Seschrock /* 183777e3a39cSMark J Musante * We need to validate the vdev labels against the configuration that 183877e3a39cSMark J Musante * we have in hand, which is dependent on the setting of mosconfig. If 183977e3a39cSMark J Musante * mosconfig is true then we're validating the vdev labels based on 18401195e687SMark J Musante * that config. Otherwise, we're validating against the cached config 184177e3a39cSMark J Musante * (zpool.cache) that was read when we loaded the zfs module, and then 184277e3a39cSMark J Musante * later we will recursively call spa_load() and validate against 184377e3a39cSMark J Musante * the vdev config. 18441195e687SMark J Musante * 18451195e687SMark J Musante * If we're assembling a new pool that's been split off from an 18461195e687SMark J Musante * existing pool, the labels haven't yet been updated so we skip 18471195e687SMark J Musante * validation for now. 1848560e6e96Seschrock */ 18491195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 18501195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 18511195e687SMark J Musante error = vdev_validate(rvd); 18521195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 1853560e6e96Seschrock 18541195e687SMark J Musante if (error != 0) 18551195e687SMark J Musante return (error); 18561195e687SMark J Musante 18571195e687SMark J Musante if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 18581195e687SMark J Musante return (ENXIO); 1859560e6e96Seschrock } 1860560e6e96Seschrock 1861fa9e4066Sahrens /* 1862fa9e4066Sahrens * Find the best uberblock. 1863fa9e4066Sahrens */ 1864e14bb325SJeff Bonwick vdev_uberblock_load(NULL, rvd, ub); 1865fa9e4066Sahrens 1866fa9e4066Sahrens /* 1867fa9e4066Sahrens * If we weren't able to find a single valid uberblock, return failure. 1868fa9e4066Sahrens */ 18691195e687SMark J Musante if (ub->ub_txg == 0) 18701195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 1871ea8dc4b6Seschrock 1872ea8dc4b6Seschrock /* 1873ea8dc4b6Seschrock * If the pool is newer than the code, we can't open it. 1874ea8dc4b6Seschrock */ 18751195e687SMark J Musante if (ub->ub_version > SPA_VERSION) 18761195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 1877fa9e4066Sahrens 1878fa9e4066Sahrens /* 1879fa9e4066Sahrens * If the vdev guid sum doesn't match the uberblock, we have an 18804b964adaSGeorge Wilson * incomplete configuration. We first check to see if the pool 18814b964adaSGeorge Wilson * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 18824b964adaSGeorge Wilson * If it is, defer the vdev_guid_sum check till later so we 18834b964adaSGeorge Wilson * can handle missing vdevs. 1884fa9e4066Sahrens */ 18854b964adaSGeorge Wilson if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 18864b964adaSGeorge Wilson &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 18871195e687SMark J Musante rvd->vdev_guid_sum != ub->ub_guid_sum) 18881195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 18891195e687SMark J Musante 18901195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 18911195e687SMark J Musante spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 18921195e687SMark J Musante spa_try_repair(spa, config); 18931195e687SMark J Musante spa_config_exit(spa, SCL_ALL, FTAG); 18941195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 18951195e687SMark J Musante spa->spa_config_splitting = NULL; 1896fa9e4066Sahrens } 1897fa9e4066Sahrens 1898fa9e4066Sahrens /* 1899fa9e4066Sahrens * Initialize internal SPA structures. 1900fa9e4066Sahrens */ 1901fa9e4066Sahrens spa->spa_state = POOL_STATE_ACTIVE; 1902fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 1903468c413aSTim Haley spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 1904c8ee1847SVictor Latushkin TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 1905468c413aSTim Haley spa->spa_first_txg = spa->spa_last_ubsync_txg ? 1906468c413aSTim Haley spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 1907b24ab676SJeff Bonwick spa->spa_claim_max_txg = spa->spa_first_txg; 19083f9d6ad7SLin Ling spa->spa_prev_software_version = ub->ub_software_version; 1909b24ab676SJeff Bonwick 1910ea8dc4b6Seschrock error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 19111195e687SMark J Musante if (error) 19121195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1913fa9e4066Sahrens spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 1914fa9e4066Sahrens 19151195e687SMark J Musante if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 19161195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1917fa9e4066Sahrens 1918fa9e4066Sahrens if (!mosconfig) { 191995173954Sek uint64_t hostid; 1920871a9500SMark J Musante nvlist_t *policy = NULL, *nvconfig; 1921871a9500SMark J Musante 1922871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 1923871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1924fa9e4066Sahrens 192588ecc943SGeorge Wilson if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 192677650510SLin Ling ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 192795173954Sek char *hostname; 192895173954Sek unsigned long myhostid = 0; 192995173954Sek 193088ecc943SGeorge Wilson VERIFY(nvlist_lookup_string(nvconfig, 193195173954Sek ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 193295173954Sek 19335679c89fSjv #ifdef _KERNEL 19345679c89fSjv myhostid = zone_get_hostid(NULL); 19355679c89fSjv #else /* _KERNEL */ 19365679c89fSjv /* 19375679c89fSjv * We're emulating the system's hostid in userland, so 19385679c89fSjv * we can't use zone_get_hostid(). 19395679c89fSjv */ 194095173954Sek (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 19415679c89fSjv #endif /* _KERNEL */ 194217194a52Slling if (hostid != 0 && myhostid != 0 && 19435679c89fSjv hostid != myhostid) { 1944871a9500SMark J Musante nvlist_free(nvconfig); 194595173954Sek cmn_err(CE_WARN, "pool '%s' could not be " 194695173954Sek "loaded as it was last accessed by " 194777650510SLin Ling "another system (host: %s hostid: 0x%lx). " 194895173954Sek "See: http://www.sun.com/msg/ZFS-8000-EY", 1949e14bb325SJeff Bonwick spa_name(spa), hostname, 195095173954Sek (unsigned long)hostid); 19511195e687SMark J Musante return (EBADF); 195295173954Sek } 195395173954Sek } 1954c8ee1847SVictor Latushkin if (nvlist_lookup_nvlist(spa->spa_config, 1955c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, &policy) == 0) 1956c8ee1847SVictor Latushkin VERIFY(nvlist_add_nvlist(nvconfig, 1957c8ee1847SVictor Latushkin ZPOOL_REWIND_POLICY, policy) == 0); 195895173954Sek 195988ecc943SGeorge Wilson spa_config_set(spa, nvconfig); 1960fa9e4066Sahrens spa_unload(spa); 1961fa9e4066Sahrens spa_deactivate(spa); 19628ad4d6ddSJeff Bonwick spa_activate(spa, orig_mode); 1963fa9e4066Sahrens 19641195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 1965fa9e4066Sahrens } 1966fa9e4066Sahrens 1967cde58dbcSMatthew Ahrens if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 1968cde58dbcSMatthew Ahrens return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1969cde58dbcSMatthew Ahrens error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 1970cde58dbcSMatthew Ahrens if (error != 0) 19711195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1972fa9e4066Sahrens 197399653d4eSeschrock /* 197499653d4eSeschrock * Load the bit that tells us to use the new accounting function 197599653d4eSeschrock * (raid-z deflation). If we have an older pool, this will not 197699653d4eSeschrock * be present. 197799653d4eSeschrock */ 19781195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 19791195e687SMark J Musante if (error != 0 && error != ENOENT) 19801195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 198199653d4eSeschrock 19823f9d6ad7SLin Ling error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 19833f9d6ad7SLin Ling &spa->spa_creation_version); 19843f9d6ad7SLin Ling if (error != 0 && error != ENOENT) 19853f9d6ad7SLin Ling return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 19863f9d6ad7SLin Ling 1987fa9e4066Sahrens /* 1988ea8dc4b6Seschrock * Load the persistent error log. If we have an older pool, this will 1989ea8dc4b6Seschrock * not be present. 1990fa9e4066Sahrens */ 19911195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 19921195e687SMark J Musante if (error != 0 && error != ENOENT) 19931195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1994ea8dc4b6Seschrock 19951195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 19961195e687SMark J Musante &spa->spa_errlog_scrub); 19971195e687SMark J Musante if (error != 0 && error != ENOENT) 19981195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 1999ea8dc4b6Seschrock 200006eeb2adSek /* 200106eeb2adSek * Load the history object. If we have an older pool, this 200206eeb2adSek * will not be present. 200306eeb2adSek */ 20041195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 20051195e687SMark J Musante if (error != 0 && error != ENOENT) 20061195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 20071195e687SMark J Musante 20081195e687SMark J Musante /* 20091195e687SMark J Musante * If we're assembling the pool from the split-off vdevs of 20101195e687SMark J Musante * an existing pool, we don't want to attach the spares & cache 20111195e687SMark J Musante * devices. 20121195e687SMark J Musante */ 201306eeb2adSek 201499653d4eSeschrock /* 201599653d4eSeschrock * Load any hot spares for this pool. 201699653d4eSeschrock */ 20171195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 20181195e687SMark J Musante if (error != 0 && error != ENOENT) 20191195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 20201195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2021e7437265Sahrens ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2022fa94a07fSbrendan if (load_nvlist(spa, spa->spa_spares.sav_object, 20231195e687SMark J Musante &spa->spa_spares.sav_config) != 0) 20241195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 202599653d4eSeschrock 2026e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 202799653d4eSeschrock spa_load_spares(spa); 2028e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 20291195e687SMark J Musante } else if (error == 0) { 20301195e687SMark J Musante spa->spa_spares.sav_sync = B_TRUE; 203199653d4eSeschrock } 203299653d4eSeschrock 2033fa94a07fSbrendan /* 2034fa94a07fSbrendan * Load any level 2 ARC devices for this pool. 2035fa94a07fSbrendan */ 20361195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2037fa94a07fSbrendan &spa->spa_l2cache.sav_object); 20381195e687SMark J Musante if (error != 0 && error != ENOENT) 20391195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 20401195e687SMark J Musante if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2041fa94a07fSbrendan ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2042fa94a07fSbrendan if (load_nvlist(spa, spa->spa_l2cache.sav_object, 20431195e687SMark J Musante &spa->spa_l2cache.sav_config) != 0) 20441195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2045fa94a07fSbrendan 2046e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2047fa94a07fSbrendan spa_load_l2cache(spa); 2048e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 20491195e687SMark J Musante } else if (error == 0) { 20501195e687SMark J Musante spa->spa_l2cache.sav_sync = B_TRUE; 2051fa94a07fSbrendan } 2052fa94a07fSbrendan 2053990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2054ecd6cf80Smarks 20551195e687SMark J Musante error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 20561195e687SMark J Musante if (error && error != ENOENT) 20571195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2058b1b8ab34Slling 2059b1b8ab34Slling if (error == 0) { 20601195e687SMark J Musante uint64_t autoreplace; 20611195e687SMark J Musante 20621195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 20631195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 20641195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 20651195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 20661195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 20671195e687SMark J Musante spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 20681195e687SMark J Musante &spa->spa_dedup_ditto); 20691195e687SMark J Musante 2070b693757aSEric Schrock spa->spa_autoreplace = (autoreplace != 0); 2071b1b8ab34Slling } 2072b1b8ab34Slling 20733d7072f8Seschrock /* 20743d7072f8Seschrock * If the 'autoreplace' property is set, then post a resource notifying 20753d7072f8Seschrock * the ZFS DE that it should not issue any faults for unopenable 20763d7072f8Seschrock * devices. We also iterate over the vdevs, and post a sysevent for any 20773d7072f8Seschrock * unopenable vdevs so that the normal autoreplace handler can take 20783d7072f8Seschrock * over. 20793d7072f8Seschrock */ 2080b693757aSEric Schrock if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 20813d7072f8Seschrock spa_check_removed(spa->spa_root_vdev); 2082b693757aSEric Schrock /* 2083b693757aSEric Schrock * For the import case, this is done in spa_import(), because 2084b693757aSEric Schrock * at this point we're using the spare definitions from 2085b693757aSEric Schrock * the MOS config, not necessarily from the userland config. 2086b693757aSEric Schrock */ 2087b693757aSEric Schrock if (state != SPA_LOAD_IMPORT) { 2088b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 2089b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 2090b693757aSEric Schrock } 2091b693757aSEric Schrock } 20923d7072f8Seschrock 2093ea8dc4b6Seschrock /* 2094560e6e96Seschrock * Load the vdev state for all toplevel vdevs. 2095ea8dc4b6Seschrock */ 2096560e6e96Seschrock vdev_load(rvd); 20970373e76bSbonwick 2098fa9e4066Sahrens /* 2099fa9e4066Sahrens * Propagate the leaf DTLs we just loaded all the way up the tree. 2100fa9e4066Sahrens */ 2101e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2102fa9e4066Sahrens vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2103e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2104fa9e4066Sahrens 2105b24ab676SJeff Bonwick /* 2106b24ab676SJeff Bonwick * Load the DDTs (dedup tables). 2107b24ab676SJeff Bonwick */ 2108b24ab676SJeff Bonwick error = ddt_load(spa); 21091195e687SMark J Musante if (error != 0) 21101195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2111b24ab676SJeff Bonwick 2112485bbbf5SGeorge Wilson spa_update_dspace(spa); 2113485bbbf5SGeorge Wilson 2114b24ab676SJeff Bonwick /* 21154b964adaSGeorge Wilson * Validate the config, using the MOS config to fill in any 21164b964adaSGeorge Wilson * information which might be missing. If we fail to validate 21174b964adaSGeorge Wilson * the config then declare the pool unfit for use. If we're 21184b964adaSGeorge Wilson * assembling a pool from a split, the log is not transferred 21194b964adaSGeorge Wilson * over. 2120b24ab676SJeff Bonwick */ 21211195e687SMark J Musante if (type != SPA_IMPORT_ASSEMBLE) { 2122871a9500SMark J Musante nvlist_t *nvconfig; 2123871a9500SMark J Musante 2124871a9500SMark J Musante if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2125871a9500SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2126871a9500SMark J Musante 21274b964adaSGeorge Wilson if (!spa_config_valid(spa, nvconfig)) { 21284b964adaSGeorge Wilson nvlist_free(nvconfig); 21294b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 21304b964adaSGeorge Wilson ENXIO)); 21314b964adaSGeorge Wilson } 21321195e687SMark J Musante nvlist_free(nvconfig); 21331195e687SMark J Musante 21344b964adaSGeorge Wilson /* 21354b964adaSGeorge Wilson * Now that we've validate the config, check the state of the 21364b964adaSGeorge Wilson * root vdev. If it can't be opened, it indicates one or 21374b964adaSGeorge Wilson * more toplevel vdevs are faulted. 21384b964adaSGeorge Wilson */ 21394b964adaSGeorge Wilson if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 21404b964adaSGeorge Wilson return (ENXIO); 21414b964adaSGeorge Wilson 21421195e687SMark J Musante if (spa_check_logs(spa)) { 21431195e687SMark J Musante *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 21441195e687SMark J Musante return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 21451195e687SMark J Musante } 2146b24ab676SJeff Bonwick } 2147b24ab676SJeff Bonwick 21484b964adaSGeorge Wilson /* 21494b964adaSGeorge Wilson * We've successfully opened the pool, verify that we're ready 21504b964adaSGeorge Wilson * to start pushing transactions. 21514b964adaSGeorge Wilson */ 21524b964adaSGeorge Wilson if (state != SPA_LOAD_TRYIMPORT) { 21534b964adaSGeorge Wilson if (error = spa_load_verify(spa)) 21544b964adaSGeorge Wilson return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 21554b964adaSGeorge Wilson error)); 21564b964adaSGeorge Wilson } 21574b964adaSGeorge Wilson 2158468c413aSTim Haley if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2159468c413aSTim Haley spa->spa_load_max_txg == UINT64_MAX)) { 21605dabedeeSbonwick dmu_tx_t *tx; 21610373e76bSbonwick int need_update = B_FALSE; 21628ad4d6ddSJeff Bonwick 21638ad4d6ddSJeff Bonwick ASSERT(state != SPA_LOAD_TRYIMPORT); 21645dabedeeSbonwick 21650373e76bSbonwick /* 21660373e76bSbonwick * Claim log blocks that haven't been committed yet. 21670373e76bSbonwick * This must all happen in a single txg. 2168b24ab676SJeff Bonwick * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2169b24ab676SJeff Bonwick * invoked from zil_claim_log_block()'s i/o done callback. 2170468c413aSTim Haley * Price of rollback is that we abandon the log. 21710373e76bSbonwick */ 2172b24ab676SJeff Bonwick spa->spa_claiming = B_TRUE; 2173b24ab676SJeff Bonwick 21745dabedeeSbonwick tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2175fa9e4066Sahrens spa_first_txg(spa)); 2176e14bb325SJeff Bonwick (void) dmu_objset_find(spa_name(spa), 21770b69c2f0Sahrens zil_claim, tx, DS_FIND_CHILDREN); 2178fa9e4066Sahrens dmu_tx_commit(tx); 2179fa9e4066Sahrens 2180b24ab676SJeff Bonwick spa->spa_claiming = B_FALSE; 2181b24ab676SJeff Bonwick 21821195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_GOOD); 2183fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 2184fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 2185fa9e4066Sahrens 2186fa9e4066Sahrens /* 2187b24ab676SJeff Bonwick * Wait for all claims to sync. We sync up to the highest 2188b24ab676SJeff Bonwick * claimed log block birth time so that claimed log blocks 2189b24ab676SJeff Bonwick * don't appear to be from the future. spa_claim_max_txg 2190b24ab676SJeff Bonwick * will have been set for us by either zil_check_log_chain() 2191b24ab676SJeff Bonwick * (invoked from spa_check_logs()) or zil_claim() above. 2192fa9e4066Sahrens */ 2193b24ab676SJeff Bonwick txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 21940e34b6a7Sbonwick 21950e34b6a7Sbonwick /* 21960373e76bSbonwick * If the config cache is stale, or we have uninitialized 21970373e76bSbonwick * metaslabs (see spa_vdev_add()), then update the config. 2198bc758434SLin Ling * 21994b964adaSGeorge Wilson * If this is a verbatim import, trust the current 2200bc758434SLin Ling * in-core spa_config and update the disk labels. 22010e34b6a7Sbonwick */ 22020373e76bSbonwick if (config_cache_txg != spa->spa_config_txg || 22034b964adaSGeorge Wilson state == SPA_LOAD_IMPORT || 22044b964adaSGeorge Wilson state == SPA_LOAD_RECOVER || 22054b964adaSGeorge Wilson (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 22060373e76bSbonwick need_update = B_TRUE; 22070373e76bSbonwick 22088ad4d6ddSJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) 22090373e76bSbonwick if (rvd->vdev_child[c]->vdev_ms_array == 0) 22100373e76bSbonwick need_update = B_TRUE; 22110e34b6a7Sbonwick 22120e34b6a7Sbonwick /* 22130373e76bSbonwick * Update the config cache asychronously in case we're the 22140373e76bSbonwick * root pool, in which case the config cache isn't writable yet. 22150e34b6a7Sbonwick */ 22160373e76bSbonwick if (need_update) 22170373e76bSbonwick spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 22188ad4d6ddSJeff Bonwick 22198ad4d6ddSJeff Bonwick /* 22208ad4d6ddSJeff Bonwick * Check all DTLs to see if anything needs resilvering. 22218ad4d6ddSJeff Bonwick */ 22223f9d6ad7SLin Ling if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 22233f9d6ad7SLin Ling vdev_resilver_needed(rvd, NULL, NULL)) 22248ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 2225503ad85cSMatthew Ahrens 2226503ad85cSMatthew Ahrens /* 2227503ad85cSMatthew Ahrens * Delete any inconsistent datasets. 2228503ad85cSMatthew Ahrens */ 2229503ad85cSMatthew Ahrens (void) dmu_objset_find(spa_name(spa), 2230503ad85cSMatthew Ahrens dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2231ca45db41SChris Kirby 2232ca45db41SChris Kirby /* 2233ca45db41SChris Kirby * Clean up any stale temporary dataset userrefs. 2234ca45db41SChris Kirby */ 2235ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2236fa9e4066Sahrens } 2237fa9e4066Sahrens 22381195e687SMark J Musante return (0); 2239fa9e4066Sahrens } 2240fa9e4066Sahrens 2241468c413aSTim Haley static int 2242468c413aSTim Haley spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2243468c413aSTim Haley { 2244468c413aSTim Haley spa_unload(spa); 2245468c413aSTim Haley spa_deactivate(spa); 2246468c413aSTim Haley 2247468c413aSTim Haley spa->spa_load_max_txg--; 2248468c413aSTim Haley 2249468c413aSTim Haley spa_activate(spa, spa_mode_global); 2250468c413aSTim Haley spa_async_suspend(spa); 2251468c413aSTim Haley 22521195e687SMark J Musante return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2253468c413aSTim Haley } 2254468c413aSTim Haley 2255468c413aSTim Haley static int 2256468c413aSTim Haley spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2257c8ee1847SVictor Latushkin uint64_t max_request, int rewind_flags) 2258468c413aSTim Haley { 2259468c413aSTim Haley nvlist_t *config = NULL; 2260468c413aSTim Haley int load_error, rewind_error; 2261c8ee1847SVictor Latushkin uint64_t safe_rewind_txg; 2262468c413aSTim Haley uint64_t min_txg; 2263468c413aSTim Haley 2264a33cae98STim Haley if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2265468c413aSTim Haley spa->spa_load_max_txg = spa->spa_load_txg; 22661195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2267a33cae98STim Haley } else { 2268468c413aSTim Haley spa->spa_load_max_txg = max_request; 2269a33cae98STim Haley } 2270468c413aSTim Haley 22711195e687SMark J Musante load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 22721195e687SMark J Musante mosconfig); 2273468c413aSTim Haley if (load_error == 0) 2274468c413aSTim Haley return (0); 2275468c413aSTim Haley 2276468c413aSTim Haley if (spa->spa_root_vdev != NULL) 2277468c413aSTim Haley config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2278468c413aSTim Haley 2279468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2280468c413aSTim Haley spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2281468c413aSTim Haley 2282c8ee1847SVictor Latushkin if (rewind_flags & ZPOOL_NEVER_REWIND) { 2283468c413aSTim Haley nvlist_free(config); 2284468c413aSTim Haley return (load_error); 2285468c413aSTim Haley } 2286468c413aSTim Haley 2287468c413aSTim Haley /* Price of rolling back is discarding txgs, including log */ 2288468c413aSTim Haley if (state == SPA_LOAD_RECOVER) 22891195e687SMark J Musante spa_set_log_state(spa, SPA_LOG_CLEAR); 2290468c413aSTim Haley 2291c8ee1847SVictor Latushkin spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2292c8ee1847SVictor Latushkin safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2293c8ee1847SVictor Latushkin min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2294c8ee1847SVictor Latushkin TXG_INITIAL : safe_rewind_txg; 2295468c413aSTim Haley 2296c8ee1847SVictor Latushkin /* 2297c8ee1847SVictor Latushkin * Continue as long as we're finding errors, we're still within 2298c8ee1847SVictor Latushkin * the acceptable rewind range, and we're still finding uberblocks 2299c8ee1847SVictor Latushkin */ 2300c8ee1847SVictor Latushkin while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2301c8ee1847SVictor Latushkin spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2302c8ee1847SVictor Latushkin if (spa->spa_load_max_txg < safe_rewind_txg) 2303468c413aSTim Haley spa->spa_extreme_rewind = B_TRUE; 2304468c413aSTim Haley rewind_error = spa_load_retry(spa, state, mosconfig); 2305468c413aSTim Haley } 2306468c413aSTim Haley 2307468c413aSTim Haley spa->spa_extreme_rewind = B_FALSE; 2308468c413aSTim Haley spa->spa_load_max_txg = UINT64_MAX; 2309468c413aSTim Haley 2310468c413aSTim Haley if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2311468c413aSTim Haley spa_config_set(spa, config); 2312468c413aSTim Haley 2313468c413aSTim Haley return (state == SPA_LOAD_RECOVER ? rewind_error : load_error); 2314468c413aSTim Haley } 2315468c413aSTim Haley 2316fa9e4066Sahrens /* 2317fa9e4066Sahrens * Pool Open/Import 2318fa9e4066Sahrens * 2319fa9e4066Sahrens * The import case is identical to an open except that the configuration is sent 2320fa9e4066Sahrens * down from userland, instead of grabbed from the configuration cache. For the 2321fa9e4066Sahrens * case of an open, the pool configuration will exist in the 23223d7072f8Seschrock * POOL_STATE_UNINITIALIZED state. 2323fa9e4066Sahrens * 2324fa9e4066Sahrens * The stats information (gen/count/ustats) is used to gather vdev statistics at 2325fa9e4066Sahrens * the same time open the pool, without having to keep around the spa_t in some 2326fa9e4066Sahrens * ambiguous state. 2327fa9e4066Sahrens */ 2328fa9e4066Sahrens static int 2329468c413aSTim Haley spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2330468c413aSTim Haley nvlist_t **config) 2331fa9e4066Sahrens { 2332fa9e4066Sahrens spa_t *spa; 23334b964adaSGeorge Wilson spa_load_state_t state = SPA_LOAD_OPEN; 2334fa9e4066Sahrens int error; 2335fa9e4066Sahrens int locked = B_FALSE; 2336fa9e4066Sahrens 2337fa9e4066Sahrens *spapp = NULL; 2338fa9e4066Sahrens 2339fa9e4066Sahrens /* 2340fa9e4066Sahrens * As disgusting as this is, we need to support recursive calls to this 2341fa9e4066Sahrens * function because dsl_dir_open() is called during spa_load(), and ends 2342fa9e4066Sahrens * up calling spa_open() again. The real fix is to figure out how to 2343fa9e4066Sahrens * avoid dsl_dir_open() calling this in the first place. 2344fa9e4066Sahrens */ 2345fa9e4066Sahrens if (mutex_owner(&spa_namespace_lock) != curthread) { 2346fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 2347fa9e4066Sahrens locked = B_TRUE; 2348fa9e4066Sahrens } 2349fa9e4066Sahrens 2350fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 2351fa9e4066Sahrens if (locked) 2352fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2353fa9e4066Sahrens return (ENOENT); 2354fa9e4066Sahrens } 2355468c413aSTim Haley 2356fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 23574b44c88cSTim Haley zpool_rewind_policy_t policy; 23584b44c88cSTim Haley 23594b44c88cSTim Haley zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 23604b44c88cSTim Haley &policy); 23614b44c88cSTim Haley if (policy.zrp_request & ZPOOL_DO_REWIND) 23624b44c88cSTim Haley state = SPA_LOAD_RECOVER; 2363fa9e4066Sahrens 23648ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 2365fa9e4066Sahrens 2366468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 2367468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2368468c413aSTim Haley 2369468c413aSTim Haley error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2370c8ee1847SVictor Latushkin policy.zrp_request); 2371fa9e4066Sahrens 2372fa9e4066Sahrens if (error == EBADF) { 2373fa9e4066Sahrens /* 2374560e6e96Seschrock * If vdev_validate() returns failure (indicated by 2375560e6e96Seschrock * EBADF), it indicates that one of the vdevs indicates 2376560e6e96Seschrock * that the pool has been exported or destroyed. If 2377560e6e96Seschrock * this is the case, the config cache is out of sync and 2378560e6e96Seschrock * we should remove the pool from the namespace. 2379fa9e4066Sahrens */ 2380fa9e4066Sahrens spa_unload(spa); 2381fa9e4066Sahrens spa_deactivate(spa); 2382c5904d13Seschrock spa_config_sync(spa, B_TRUE, B_TRUE); 2383fa9e4066Sahrens spa_remove(spa); 2384fa9e4066Sahrens if (locked) 2385fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2386fa9e4066Sahrens return (ENOENT); 2387ea8dc4b6Seschrock } 2388ea8dc4b6Seschrock 2389ea8dc4b6Seschrock if (error) { 2390fa9e4066Sahrens /* 2391fa9e4066Sahrens * We can't open the pool, but we still have useful 2392fa9e4066Sahrens * information: the state of each vdev after the 2393fa9e4066Sahrens * attempted vdev_open(). Return this to the user. 2394fa9e4066Sahrens */ 23954b964adaSGeorge Wilson if (config != NULL && spa->spa_config) { 2396468c413aSTim Haley VERIFY(nvlist_dup(spa->spa_config, config, 2397468c413aSTim Haley KM_SLEEP) == 0); 23984b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, 23994b964adaSGeorge Wilson ZPOOL_CONFIG_LOAD_INFO, 24004b964adaSGeorge Wilson spa->spa_load_info) == 0); 24014b964adaSGeorge Wilson } 2402fa9e4066Sahrens spa_unload(spa); 2403fa9e4066Sahrens spa_deactivate(spa); 2404468c413aSTim Haley spa->spa_last_open_failed = error; 2405fa9e4066Sahrens if (locked) 2406fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2407fa9e4066Sahrens *spapp = NULL; 2408fa9e4066Sahrens return (error); 2409fa9e4066Sahrens } 2410fa9e4066Sahrens } 2411fa9e4066Sahrens 2412fa9e4066Sahrens spa_open_ref(spa, tag); 24133d7072f8Seschrock 2414468c413aSTim Haley if (config != NULL) 2415468c413aSTim Haley *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2416468c413aSTim Haley 24174b964adaSGeorge Wilson /* 24184b964adaSGeorge Wilson * If we've recovered the pool, pass back any information we 24194b964adaSGeorge Wilson * gathered while doing the load. 24204b964adaSGeorge Wilson */ 24214b964adaSGeorge Wilson if (state == SPA_LOAD_RECOVER) { 24224b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 24234b964adaSGeorge Wilson spa->spa_load_info) == 0); 24244b964adaSGeorge Wilson } 24254b964adaSGeorge Wilson 2426a33cae98STim Haley if (locked) { 2427a33cae98STim Haley spa->spa_last_open_failed = 0; 2428a33cae98STim Haley spa->spa_last_ubsync_txg = 0; 2429a33cae98STim Haley spa->spa_load_txg = 0; 2430fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2431a33cae98STim Haley } 2432fa9e4066Sahrens 2433fa9e4066Sahrens *spapp = spa; 2434fa9e4066Sahrens 2435fa9e4066Sahrens return (0); 2436fa9e4066Sahrens } 2437fa9e4066Sahrens 2438468c413aSTim Haley int 2439468c413aSTim Haley spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2440468c413aSTim Haley nvlist_t **config) 2441468c413aSTim Haley { 2442468c413aSTim Haley return (spa_open_common(name, spapp, tag, policy, config)); 2443468c413aSTim Haley } 2444468c413aSTim Haley 2445fa9e4066Sahrens int 2446fa9e4066Sahrens spa_open(const char *name, spa_t **spapp, void *tag) 2447fa9e4066Sahrens { 2448468c413aSTim Haley return (spa_open_common(name, spapp, tag, NULL, NULL)); 2449fa9e4066Sahrens } 2450fa9e4066Sahrens 2451ea8dc4b6Seschrock /* 2452ea8dc4b6Seschrock * Lookup the given spa_t, incrementing the inject count in the process, 2453ea8dc4b6Seschrock * preventing it from being exported or destroyed. 2454ea8dc4b6Seschrock */ 2455ea8dc4b6Seschrock spa_t * 2456ea8dc4b6Seschrock spa_inject_addref(char *name) 2457ea8dc4b6Seschrock { 2458ea8dc4b6Seschrock spa_t *spa; 2459ea8dc4b6Seschrock 2460ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 2461ea8dc4b6Seschrock if ((spa = spa_lookup(name)) == NULL) { 2462ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 2463ea8dc4b6Seschrock return (NULL); 2464ea8dc4b6Seschrock } 2465ea8dc4b6Seschrock spa->spa_inject_ref++; 2466ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 2467ea8dc4b6Seschrock 2468ea8dc4b6Seschrock return (spa); 2469ea8dc4b6Seschrock } 2470ea8dc4b6Seschrock 2471ea8dc4b6Seschrock void 2472ea8dc4b6Seschrock spa_inject_delref(spa_t *spa) 2473ea8dc4b6Seschrock { 2474ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 2475ea8dc4b6Seschrock spa->spa_inject_ref--; 2476ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 2477ea8dc4b6Seschrock } 2478ea8dc4b6Seschrock 2479fa94a07fSbrendan /* 2480fa94a07fSbrendan * Add spares device information to the nvlist. 2481fa94a07fSbrendan */ 248299653d4eSeschrock static void 248399653d4eSeschrock spa_add_spares(spa_t *spa, nvlist_t *config) 248499653d4eSeschrock { 248599653d4eSeschrock nvlist_t **spares; 248699653d4eSeschrock uint_t i, nspares; 248799653d4eSeschrock nvlist_t *nvroot; 248899653d4eSeschrock uint64_t guid; 248999653d4eSeschrock vdev_stat_t *vs; 249099653d4eSeschrock uint_t vsc; 249139c23413Seschrock uint64_t pool; 249299653d4eSeschrock 24936809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 24946809eb4eSEric Schrock 2495fa94a07fSbrendan if (spa->spa_spares.sav_count == 0) 249699653d4eSeschrock return; 249799653d4eSeschrock 249899653d4eSeschrock VERIFY(nvlist_lookup_nvlist(config, 249999653d4eSeschrock ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2500fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 250199653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 250299653d4eSeschrock if (nspares != 0) { 250399653d4eSeschrock VERIFY(nvlist_add_nvlist_array(nvroot, 250499653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 250599653d4eSeschrock VERIFY(nvlist_lookup_nvlist_array(nvroot, 250699653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 250799653d4eSeschrock 250899653d4eSeschrock /* 250999653d4eSeschrock * Go through and find any spares which have since been 251099653d4eSeschrock * repurposed as an active spare. If this is the case, update 251199653d4eSeschrock * their status appropriately. 251299653d4eSeschrock */ 251399653d4eSeschrock for (i = 0; i < nspares; i++) { 251499653d4eSeschrock VERIFY(nvlist_lookup_uint64(spares[i], 251599653d4eSeschrock ZPOOL_CONFIG_GUID, &guid) == 0); 251689a89ebfSlling if (spa_spare_exists(guid, &pool, NULL) && 251789a89ebfSlling pool != 0ULL) { 251899653d4eSeschrock VERIFY(nvlist_lookup_uint64_array( 25193f9d6ad7SLin Ling spares[i], ZPOOL_CONFIG_VDEV_STATS, 252099653d4eSeschrock (uint64_t **)&vs, &vsc) == 0); 252199653d4eSeschrock vs->vs_state = VDEV_STATE_CANT_OPEN; 252299653d4eSeschrock vs->vs_aux = VDEV_AUX_SPARED; 252399653d4eSeschrock } 252499653d4eSeschrock } 252599653d4eSeschrock } 252699653d4eSeschrock } 252799653d4eSeschrock 2528fa94a07fSbrendan /* 2529fa94a07fSbrendan * Add l2cache device information to the nvlist, including vdev stats. 2530fa94a07fSbrendan */ 2531fa94a07fSbrendan static void 2532fa94a07fSbrendan spa_add_l2cache(spa_t *spa, nvlist_t *config) 2533fa94a07fSbrendan { 2534fa94a07fSbrendan nvlist_t **l2cache; 2535fa94a07fSbrendan uint_t i, j, nl2cache; 2536fa94a07fSbrendan nvlist_t *nvroot; 2537fa94a07fSbrendan uint64_t guid; 2538fa94a07fSbrendan vdev_t *vd; 2539fa94a07fSbrendan vdev_stat_t *vs; 2540fa94a07fSbrendan uint_t vsc; 2541fa94a07fSbrendan 25426809eb4eSEric Schrock ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 25436809eb4eSEric Schrock 2544fa94a07fSbrendan if (spa->spa_l2cache.sav_count == 0) 2545fa94a07fSbrendan return; 2546fa94a07fSbrendan 2547fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist(config, 2548fa94a07fSbrendan ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2549fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 2550fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 2551fa94a07fSbrendan if (nl2cache != 0) { 2552fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, 2553fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2554fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(nvroot, 2555fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 2556fa94a07fSbrendan 2557fa94a07fSbrendan /* 2558fa94a07fSbrendan * Update level 2 cache device stats. 2559fa94a07fSbrendan */ 2560fa94a07fSbrendan 2561fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 2562fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 2563fa94a07fSbrendan ZPOOL_CONFIG_GUID, &guid) == 0); 2564fa94a07fSbrendan 2565fa94a07fSbrendan vd = NULL; 2566fa94a07fSbrendan for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 2567fa94a07fSbrendan if (guid == 2568fa94a07fSbrendan spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 2569fa94a07fSbrendan vd = spa->spa_l2cache.sav_vdevs[j]; 2570fa94a07fSbrendan break; 2571fa94a07fSbrendan } 2572fa94a07fSbrendan } 2573fa94a07fSbrendan ASSERT(vd != NULL); 2574fa94a07fSbrendan 2575fa94a07fSbrendan VERIFY(nvlist_lookup_uint64_array(l2cache[i], 25763f9d6ad7SLin Ling ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 25773f9d6ad7SLin Ling == 0); 2578fa94a07fSbrendan vdev_get_stats(vd, vs); 2579fa94a07fSbrendan } 2580fa94a07fSbrendan } 2581fa94a07fSbrendan } 2582fa94a07fSbrendan 2583fa9e4066Sahrens int 2584ea8dc4b6Seschrock spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) 2585fa9e4066Sahrens { 2586fa9e4066Sahrens int error; 2587fa9e4066Sahrens spa_t *spa; 2588fa9e4066Sahrens 2589fa9e4066Sahrens *config = NULL; 2590468c413aSTim Haley error = spa_open_common(name, &spa, FTAG, NULL, config); 2591fa9e4066Sahrens 25926809eb4eSEric Schrock if (spa != NULL) { 25936809eb4eSEric Schrock /* 25946809eb4eSEric Schrock * This still leaves a window of inconsistency where the spares 25956809eb4eSEric Schrock * or l2cache devices could change and the config would be 25966809eb4eSEric Schrock * self-inconsistent. 25976809eb4eSEric Schrock */ 25986809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 2599ea8dc4b6Seschrock 26006809eb4eSEric Schrock if (*config != NULL) { 260111027bc7STim Haley uint64_t loadtimes[2]; 260211027bc7STim Haley 260311027bc7STim Haley loadtimes[0] = spa->spa_loaded_ts.tv_sec; 260411027bc7STim Haley loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 260511027bc7STim Haley VERIFY(nvlist_add_uint64_array(*config, 260611027bc7STim Haley ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 260711027bc7STim Haley 2608e14bb325SJeff Bonwick VERIFY(nvlist_add_uint64(*config, 26096809eb4eSEric Schrock ZPOOL_CONFIG_ERRCOUNT, 26106809eb4eSEric Schrock spa_get_errlog_size(spa)) == 0); 2611e14bb325SJeff Bonwick 26126809eb4eSEric Schrock if (spa_suspended(spa)) 26136809eb4eSEric Schrock VERIFY(nvlist_add_uint64(*config, 26146809eb4eSEric Schrock ZPOOL_CONFIG_SUSPENDED, 26156809eb4eSEric Schrock spa->spa_failmode) == 0); 26166809eb4eSEric Schrock 26176809eb4eSEric Schrock spa_add_spares(spa, *config); 26186809eb4eSEric Schrock spa_add_l2cache(spa, *config); 26196809eb4eSEric Schrock } 262099653d4eSeschrock } 262199653d4eSeschrock 2622ea8dc4b6Seschrock /* 2623ea8dc4b6Seschrock * We want to get the alternate root even for faulted pools, so we cheat 2624ea8dc4b6Seschrock * and call spa_lookup() directly. 2625ea8dc4b6Seschrock */ 2626ea8dc4b6Seschrock if (altroot) { 2627ea8dc4b6Seschrock if (spa == NULL) { 2628ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 2629ea8dc4b6Seschrock spa = spa_lookup(name); 2630ea8dc4b6Seschrock if (spa) 2631ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 2632ea8dc4b6Seschrock else 2633ea8dc4b6Seschrock altroot[0] = '\0'; 2634ea8dc4b6Seschrock spa = NULL; 2635ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 2636ea8dc4b6Seschrock } else { 2637ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 2638ea8dc4b6Seschrock } 2639ea8dc4b6Seschrock } 2640ea8dc4b6Seschrock 26416809eb4eSEric Schrock if (spa != NULL) { 26426809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 2643fa9e4066Sahrens spa_close(spa, FTAG); 26446809eb4eSEric Schrock } 2645fa9e4066Sahrens 2646fa9e4066Sahrens return (error); 2647fa9e4066Sahrens } 2648fa9e4066Sahrens 264999653d4eSeschrock /* 2650fa94a07fSbrendan * Validate that the auxiliary device array is well formed. We must have an 2651fa94a07fSbrendan * array of nvlists, each which describes a valid leaf vdev. If this is an 2652fa94a07fSbrendan * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 2653fa94a07fSbrendan * specified, as long as they are well-formed. 265499653d4eSeschrock */ 265599653d4eSeschrock static int 2656fa94a07fSbrendan spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 2657fa94a07fSbrendan spa_aux_vdev_t *sav, const char *config, uint64_t version, 2658fa94a07fSbrendan vdev_labeltype_t label) 265999653d4eSeschrock { 2660fa94a07fSbrendan nvlist_t **dev; 2661fa94a07fSbrendan uint_t i, ndev; 266299653d4eSeschrock vdev_t *vd; 266399653d4eSeschrock int error; 266499653d4eSeschrock 2665e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 2666e14bb325SJeff Bonwick 266799653d4eSeschrock /* 2668fa94a07fSbrendan * It's acceptable to have no devs specified. 266999653d4eSeschrock */ 2670fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 267199653d4eSeschrock return (0); 267299653d4eSeschrock 2673fa94a07fSbrendan if (ndev == 0) 267499653d4eSeschrock return (EINVAL); 267599653d4eSeschrock 267699653d4eSeschrock /* 2677fa94a07fSbrendan * Make sure the pool is formatted with a version that supports this 2678fa94a07fSbrendan * device type. 267999653d4eSeschrock */ 2680fa94a07fSbrendan if (spa_version(spa) < version) 268199653d4eSeschrock return (ENOTSUP); 268299653d4eSeschrock 268339c23413Seschrock /* 2684fa94a07fSbrendan * Set the pending device list so we correctly handle device in-use 268539c23413Seschrock * checking. 268639c23413Seschrock */ 2687fa94a07fSbrendan sav->sav_pending = dev; 2688fa94a07fSbrendan sav->sav_npending = ndev; 268939c23413Seschrock 2690fa94a07fSbrendan for (i = 0; i < ndev; i++) { 2691fa94a07fSbrendan if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 269299653d4eSeschrock mode)) != 0) 269339c23413Seschrock goto out; 269499653d4eSeschrock 269599653d4eSeschrock if (!vd->vdev_ops->vdev_op_leaf) { 269699653d4eSeschrock vdev_free(vd); 269739c23413Seschrock error = EINVAL; 269839c23413Seschrock goto out; 269999653d4eSeschrock } 270099653d4eSeschrock 2701fa94a07fSbrendan /* 2702e14bb325SJeff Bonwick * The L2ARC currently only supports disk devices in 2703e14bb325SJeff Bonwick * kernel context. For user-level testing, we allow it. 2704fa94a07fSbrendan */ 2705e14bb325SJeff Bonwick #ifdef _KERNEL 2706fa94a07fSbrendan if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 2707fa94a07fSbrendan strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 2708fa94a07fSbrendan error = ENOTBLK; 2709fa94a07fSbrendan goto out; 2710fa94a07fSbrendan } 2711e14bb325SJeff Bonwick #endif 271299653d4eSeschrock vd->vdev_top = vd; 271399653d4eSeschrock 271439c23413Seschrock if ((error = vdev_open(vd)) == 0 && 2715fa94a07fSbrendan (error = vdev_label_init(vd, crtxg, label)) == 0) { 2716fa94a07fSbrendan VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 271739c23413Seschrock vd->vdev_guid) == 0); 271839c23413Seschrock } 271999653d4eSeschrock 272099653d4eSeschrock vdev_free(vd); 272139c23413Seschrock 2722fa94a07fSbrendan if (error && 2723fa94a07fSbrendan (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 272439c23413Seschrock goto out; 272539c23413Seschrock else 272639c23413Seschrock error = 0; 272799653d4eSeschrock } 272899653d4eSeschrock 272939c23413Seschrock out: 2730fa94a07fSbrendan sav->sav_pending = NULL; 2731fa94a07fSbrendan sav->sav_npending = 0; 273239c23413Seschrock return (error); 273399653d4eSeschrock } 273499653d4eSeschrock 2735fa94a07fSbrendan static int 2736fa94a07fSbrendan spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 2737fa94a07fSbrendan { 2738fa94a07fSbrendan int error; 2739fa94a07fSbrendan 2740e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 2741e14bb325SJeff Bonwick 2742fa94a07fSbrendan if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 2743fa94a07fSbrendan &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 2744fa94a07fSbrendan VDEV_LABEL_SPARE)) != 0) { 2745fa94a07fSbrendan return (error); 2746fa94a07fSbrendan } 2747fa94a07fSbrendan 2748fa94a07fSbrendan return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 2749fa94a07fSbrendan &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 2750fa94a07fSbrendan VDEV_LABEL_L2CACHE)); 2751fa94a07fSbrendan } 2752fa94a07fSbrendan 2753fa94a07fSbrendan static void 2754fa94a07fSbrendan spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 2755fa94a07fSbrendan const char *config) 2756fa94a07fSbrendan { 2757fa94a07fSbrendan int i; 2758fa94a07fSbrendan 2759fa94a07fSbrendan if (sav->sav_config != NULL) { 2760fa94a07fSbrendan nvlist_t **olddevs; 2761fa94a07fSbrendan uint_t oldndevs; 2762fa94a07fSbrendan nvlist_t **newdevs; 2763fa94a07fSbrendan 2764fa94a07fSbrendan /* 2765fa94a07fSbrendan * Generate new dev list by concatentating with the 2766fa94a07fSbrendan * current dev list. 2767fa94a07fSbrendan */ 2768fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 2769fa94a07fSbrendan &olddevs, &oldndevs) == 0); 2770fa94a07fSbrendan 2771fa94a07fSbrendan newdevs = kmem_alloc(sizeof (void *) * 2772fa94a07fSbrendan (ndevs + oldndevs), KM_SLEEP); 2773fa94a07fSbrendan for (i = 0; i < oldndevs; i++) 2774fa94a07fSbrendan VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 2775fa94a07fSbrendan KM_SLEEP) == 0); 2776fa94a07fSbrendan for (i = 0; i < ndevs; i++) 2777fa94a07fSbrendan VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 2778fa94a07fSbrendan KM_SLEEP) == 0); 2779fa94a07fSbrendan 2780fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, config, 2781fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 2782fa94a07fSbrendan 2783fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 2784fa94a07fSbrendan config, newdevs, ndevs + oldndevs) == 0); 2785fa94a07fSbrendan for (i = 0; i < oldndevs + ndevs; i++) 2786fa94a07fSbrendan nvlist_free(newdevs[i]); 2787fa94a07fSbrendan kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 2788fa94a07fSbrendan } else { 2789fa94a07fSbrendan /* 2790fa94a07fSbrendan * Generate a new dev list. 2791fa94a07fSbrendan */ 2792fa94a07fSbrendan VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 2793fa94a07fSbrendan KM_SLEEP) == 0); 2794fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 2795fa94a07fSbrendan devs, ndevs) == 0); 2796fa94a07fSbrendan } 2797fa94a07fSbrendan } 2798fa94a07fSbrendan 2799fa94a07fSbrendan /* 2800fa94a07fSbrendan * Stop and drop level 2 ARC devices 2801fa94a07fSbrendan */ 2802fa94a07fSbrendan void 2803fa94a07fSbrendan spa_l2cache_drop(spa_t *spa) 2804fa94a07fSbrendan { 2805fa94a07fSbrendan vdev_t *vd; 2806fa94a07fSbrendan int i; 2807fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 2808fa94a07fSbrendan 2809fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) { 2810fa94a07fSbrendan uint64_t pool; 2811fa94a07fSbrendan 2812fa94a07fSbrendan vd = sav->sav_vdevs[i]; 2813fa94a07fSbrendan ASSERT(vd != NULL); 2814fa94a07fSbrendan 28158ad4d6ddSJeff Bonwick if (spa_l2cache_exists(vd->vdev_guid, &pool) && 28168ad4d6ddSJeff Bonwick pool != 0ULL && l2arc_vdev_present(vd)) 2817fa94a07fSbrendan l2arc_remove_vdev(vd); 2818fa94a07fSbrendan if (vd->vdev_isl2cache) 2819fa94a07fSbrendan spa_l2cache_remove(vd); 2820fa94a07fSbrendan vdev_clear_stats(vd); 2821fa94a07fSbrendan (void) vdev_close(vd); 2822fa94a07fSbrendan } 2823fa94a07fSbrendan } 2824fa94a07fSbrendan 2825fa9e4066Sahrens /* 2826fa9e4066Sahrens * Pool Creation 2827fa9e4066Sahrens */ 2828fa9e4066Sahrens int 2829990b4856Slling spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 28300a48a24eStimh const char *history_str, nvlist_t *zplprops) 2831fa9e4066Sahrens { 2832fa9e4066Sahrens spa_t *spa; 2833990b4856Slling char *altroot = NULL; 28340373e76bSbonwick vdev_t *rvd; 2835fa9e4066Sahrens dsl_pool_t *dp; 2836fa9e4066Sahrens dmu_tx_t *tx; 2837573ca77eSGeorge Wilson int error = 0; 2838fa9e4066Sahrens uint64_t txg = TXG_INITIAL; 2839fa94a07fSbrendan nvlist_t **spares, **l2cache; 2840fa94a07fSbrendan uint_t nspares, nl2cache; 2841cde58dbcSMatthew Ahrens uint64_t version, obj; 2842fa9e4066Sahrens 2843fa9e4066Sahrens /* 2844fa9e4066Sahrens * If this pool already exists, return failure. 2845fa9e4066Sahrens */ 2846fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 2847fa9e4066Sahrens if (spa_lookup(pool) != NULL) { 2848fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2849fa9e4066Sahrens return (EEXIST); 2850fa9e4066Sahrens } 2851fa9e4066Sahrens 2852fa9e4066Sahrens /* 2853fa9e4066Sahrens * Allocate a new spa_t structure. 2854fa9e4066Sahrens */ 2855990b4856Slling (void) nvlist_lookup_string(props, 2856990b4856Slling zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 2857468c413aSTim Haley spa = spa_add(pool, NULL, altroot); 28588ad4d6ddSJeff Bonwick spa_activate(spa, spa_mode_global); 2859fa9e4066Sahrens 2860990b4856Slling if (props && (error = spa_prop_validate(spa, props))) { 2861990b4856Slling spa_deactivate(spa); 2862990b4856Slling spa_remove(spa); 2863c5904d13Seschrock mutex_exit(&spa_namespace_lock); 2864990b4856Slling return (error); 2865990b4856Slling } 2866990b4856Slling 2867990b4856Slling if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), 2868990b4856Slling &version) != 0) 2869990b4856Slling version = SPA_VERSION; 2870990b4856Slling ASSERT(version <= SPA_VERSION); 2871b24ab676SJeff Bonwick 2872b24ab676SJeff Bonwick spa->spa_first_txg = txg; 2873b24ab676SJeff Bonwick spa->spa_uberblock.ub_txg = txg - 1; 2874990b4856Slling spa->spa_uberblock.ub_version = version; 2875fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 2876fa9e4066Sahrens 287754d692b7SGeorge Wilson /* 287854d692b7SGeorge Wilson * Create "The Godfather" zio to hold all async IOs 287954d692b7SGeorge Wilson */ 288025f89ee2SJeff Bonwick spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 288125f89ee2SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 288254d692b7SGeorge Wilson 28830373e76bSbonwick /* 28840373e76bSbonwick * Create the root vdev. 28850373e76bSbonwick */ 2886e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 28870373e76bSbonwick 288899653d4eSeschrock error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 28890373e76bSbonwick 289099653d4eSeschrock ASSERT(error != 0 || rvd != NULL); 289199653d4eSeschrock ASSERT(error != 0 || spa->spa_root_vdev == rvd); 28920373e76bSbonwick 2893b7b97454Sperrin if (error == 0 && !zfs_allocatable_devs(nvroot)) 28940373e76bSbonwick error = EINVAL; 289599653d4eSeschrock 289699653d4eSeschrock if (error == 0 && 289799653d4eSeschrock (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 2898fa94a07fSbrendan (error = spa_validate_aux(spa, nvroot, txg, 289999653d4eSeschrock VDEV_ALLOC_ADD)) == 0) { 2900573ca77eSGeorge Wilson for (int c = 0; c < rvd->vdev_children; c++) { 2901573ca77eSGeorge Wilson vdev_metaslab_set_size(rvd->vdev_child[c]); 2902573ca77eSGeorge Wilson vdev_expand(rvd->vdev_child[c], txg); 2903573ca77eSGeorge Wilson } 29040373e76bSbonwick } 29050373e76bSbonwick 2906e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2907fa9e4066Sahrens 290899653d4eSeschrock if (error != 0) { 2909fa9e4066Sahrens spa_unload(spa); 2910fa9e4066Sahrens spa_deactivate(spa); 2911fa9e4066Sahrens spa_remove(spa); 2912fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2913fa9e4066Sahrens return (error); 2914fa9e4066Sahrens } 2915fa9e4066Sahrens 291699653d4eSeschrock /* 291799653d4eSeschrock * Get the list of spares, if specified. 291899653d4eSeschrock */ 291999653d4eSeschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 292099653d4eSeschrock &spares, &nspares) == 0) { 2921fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 292299653d4eSeschrock KM_SLEEP) == 0); 2923fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 292499653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2925e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 292699653d4eSeschrock spa_load_spares(spa); 2927e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2928fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 2929fa94a07fSbrendan } 2930fa94a07fSbrendan 2931fa94a07fSbrendan /* 2932fa94a07fSbrendan * Get the list of level 2 cache devices, if specified. 2933fa94a07fSbrendan */ 2934fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 2935fa94a07fSbrendan &l2cache, &nl2cache) == 0) { 2936fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 2937fa94a07fSbrendan NV_UNIQUE_NAME, KM_SLEEP) == 0); 2938fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 2939fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2940e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2941fa94a07fSbrendan spa_load_l2cache(spa); 2942e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 2943fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 294499653d4eSeschrock } 294599653d4eSeschrock 29460a48a24eStimh spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 2947fa9e4066Sahrens spa->spa_meta_objset = dp->dp_meta_objset; 2948fa9e4066Sahrens 2949485bbbf5SGeorge Wilson /* 2950485bbbf5SGeorge Wilson * Create DDTs (dedup tables). 2951485bbbf5SGeorge Wilson */ 2952485bbbf5SGeorge Wilson ddt_create(spa); 2953485bbbf5SGeorge Wilson 2954485bbbf5SGeorge Wilson spa_update_dspace(spa); 2955485bbbf5SGeorge Wilson 2956fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 2957fa9e4066Sahrens 2958fa9e4066Sahrens /* 2959fa9e4066Sahrens * Create the pool config object. 2960fa9e4066Sahrens */ 2961fa9e4066Sahrens spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 2962f7991ba4STim Haley DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 2963fa9e4066Sahrens DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 2964fa9e4066Sahrens 2965ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 2966fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 2967ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 2968ea8dc4b6Seschrock cmn_err(CE_PANIC, "failed to add pool config"); 2969ea8dc4b6Seschrock } 2970fa9e4066Sahrens 29713f9d6ad7SLin Ling if (zap_add(spa->spa_meta_objset, 29723f9d6ad7SLin Ling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 29733f9d6ad7SLin Ling sizeof (uint64_t), 1, &version, tx) != 0) { 29743f9d6ad7SLin Ling cmn_err(CE_PANIC, "failed to add pool version"); 29753f9d6ad7SLin Ling } 29763f9d6ad7SLin Ling 2977990b4856Slling /* Newly created pools with the right version are always deflated. */ 2978990b4856Slling if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 2979990b4856Slling spa->spa_deflate = TRUE; 2980990b4856Slling if (zap_add(spa->spa_meta_objset, 2981990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 2982990b4856Slling sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 2983990b4856Slling cmn_err(CE_PANIC, "failed to add deflate"); 2984990b4856Slling } 298599653d4eSeschrock } 298699653d4eSeschrock 2987fa9e4066Sahrens /* 2988cde58dbcSMatthew Ahrens * Create the deferred-free bpobj. Turn off compression 2989fa9e4066Sahrens * because sync-to-convergence takes longer if the blocksize 2990fa9e4066Sahrens * keeps changing. 2991fa9e4066Sahrens */ 2992cde58dbcSMatthew Ahrens obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 2993cde58dbcSMatthew Ahrens dmu_object_set_compress(spa->spa_meta_objset, obj, 2994cde58dbcSMatthew Ahrens ZIO_COMPRESS_OFF, tx); 2995ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 2996cde58dbcSMatthew Ahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 2997cde58dbcSMatthew Ahrens sizeof (uint64_t), 1, &obj, tx) != 0) { 2998cde58dbcSMatthew Ahrens cmn_err(CE_PANIC, "failed to add bpobj"); 2999ea8dc4b6Seschrock } 3000cde58dbcSMatthew Ahrens VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3001cde58dbcSMatthew Ahrens spa->spa_meta_objset, obj)); 3002fa9e4066Sahrens 300306eeb2adSek /* 300406eeb2adSek * Create the pool's history object. 300506eeb2adSek */ 3006990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY) 3007990b4856Slling spa_history_create_obj(spa, tx); 3008990b4856Slling 3009990b4856Slling /* 3010990b4856Slling * Set pool properties. 3011990b4856Slling */ 3012990b4856Slling spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3013990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 30140a4e9518Sgw spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3015573ca77eSGeorge Wilson spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3016b24ab676SJeff Bonwick 3017379c004dSEric Schrock if (props != NULL) { 3018379c004dSEric Schrock spa_configfile_set(spa, props, B_FALSE); 30193f9d6ad7SLin Ling spa_sync_props(spa, props, tx); 3020379c004dSEric Schrock } 302106eeb2adSek 3022fa9e4066Sahrens dmu_tx_commit(tx); 3023fa9e4066Sahrens 3024fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 3025fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 3026fa9e4066Sahrens 3027fa9e4066Sahrens /* 3028fa9e4066Sahrens * We explicitly wait for the first transaction to complete so that our 3029fa9e4066Sahrens * bean counters are appropriately updated. 3030fa9e4066Sahrens */ 3031fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 3032fa9e4066Sahrens 3033c5904d13Seschrock spa_config_sync(spa, B_FALSE, B_TRUE); 3034fa9e4066Sahrens 3035990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL) 3036228975ccSek (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE); 3037c8e1f6d2SMark J Musante spa_history_log_version(spa, LOG_POOL_CREATE); 3038228975ccSek 3039088f3894Sahrens spa->spa_minref = refcount_count(&spa->spa_refcount); 3040088f3894Sahrens 3041daaa36a7SGeorge Wilson mutex_exit(&spa_namespace_lock); 3042daaa36a7SGeorge Wilson 3043fa9e4066Sahrens return (0); 3044fa9e4066Sahrens } 3045fa9e4066Sahrens 3046e7cbe64fSgw #ifdef _KERNEL 3047e7cbe64fSgw /* 304821ecdf64SLin Ling * Get the root pool information from the root disk, then import the root pool 304921ecdf64SLin Ling * during the system boot up time. 3050e7cbe64fSgw */ 305121ecdf64SLin Ling extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 305221ecdf64SLin Ling 305321ecdf64SLin Ling static nvlist_t * 305421ecdf64SLin Ling spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3055e7cbe64fSgw { 305621ecdf64SLin Ling nvlist_t *config; 3057e7cbe64fSgw nvlist_t *nvtop, *nvroot; 3058e7cbe64fSgw uint64_t pgid; 3059e7cbe64fSgw 306021ecdf64SLin Ling if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 306121ecdf64SLin Ling return (NULL); 306221ecdf64SLin Ling 3063e7cbe64fSgw /* 3064e7cbe64fSgw * Add this top-level vdev to the child array. 3065e7cbe64fSgw */ 306621ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 306721ecdf64SLin Ling &nvtop) == 0); 306821ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 306921ecdf64SLin Ling &pgid) == 0); 307021ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3071e7cbe64fSgw 3072e7cbe64fSgw /* 3073e7cbe64fSgw * Put this pool's top-level vdevs into a root vdev. 3074e7cbe64fSgw */ 3075e7cbe64fSgw VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 307621ecdf64SLin Ling VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 307721ecdf64SLin Ling VDEV_TYPE_ROOT) == 0); 3078e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3079e7cbe64fSgw VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3080e7cbe64fSgw VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3081e7cbe64fSgw &nvtop, 1) == 0); 3082e7cbe64fSgw 3083e7cbe64fSgw /* 3084e7cbe64fSgw * Replace the existing vdev_tree with the new root vdev in 3085e7cbe64fSgw * this pool's configuration (remove the old, add the new). 3086e7cbe64fSgw */ 3087e7cbe64fSgw VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3088e7cbe64fSgw nvlist_free(nvroot); 308921ecdf64SLin Ling return (config); 3090e7cbe64fSgw } 3091e7cbe64fSgw 3092e7cbe64fSgw /* 309321ecdf64SLin Ling * Walk the vdev tree and see if we can find a device with "better" 309421ecdf64SLin Ling * configuration. A configuration is "better" if the label on that 309521ecdf64SLin Ling * device has a more recent txg. 3096051aabe6Staylor */ 309721ecdf64SLin Ling static void 309821ecdf64SLin Ling spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3099051aabe6Staylor { 3100573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 310121ecdf64SLin Ling spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3102051aabe6Staylor 310321ecdf64SLin Ling if (vd->vdev_ops->vdev_op_leaf) { 310421ecdf64SLin Ling nvlist_t *label; 310521ecdf64SLin Ling uint64_t label_txg; 3106051aabe6Staylor 310721ecdf64SLin Ling if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 310821ecdf64SLin Ling &label) != 0) 310921ecdf64SLin Ling return; 3110051aabe6Staylor 311121ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 311221ecdf64SLin Ling &label_txg) == 0); 3113051aabe6Staylor 311421ecdf64SLin Ling /* 311521ecdf64SLin Ling * Do we have a better boot device? 311621ecdf64SLin Ling */ 311721ecdf64SLin Ling if (label_txg > *txg) { 311821ecdf64SLin Ling *txg = label_txg; 311921ecdf64SLin Ling *avd = vd; 3120051aabe6Staylor } 312121ecdf64SLin Ling nvlist_free(label); 3122051aabe6Staylor } 3123051aabe6Staylor } 3124051aabe6Staylor 3125e7cbe64fSgw /* 3126e7cbe64fSgw * Import a root pool. 3127e7cbe64fSgw * 3128051aabe6Staylor * For x86. devpath_list will consist of devid and/or physpath name of 3129051aabe6Staylor * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3130051aabe6Staylor * The GRUB "findroot" command will return the vdev we should boot. 3131e7cbe64fSgw * 3132e7cbe64fSgw * For Sparc, devpath_list consists the physpath name of the booting device 3133e7cbe64fSgw * no matter the rootpool is a single device pool or a mirrored pool. 3134e7cbe64fSgw * e.g. 3135e7cbe64fSgw * "/pci@1f,0/ide@d/disk@0,0:a" 3136e7cbe64fSgw */ 3137e7cbe64fSgw int 3138051aabe6Staylor spa_import_rootpool(char *devpath, char *devid) 3139e7cbe64fSgw { 314021ecdf64SLin Ling spa_t *spa; 314121ecdf64SLin Ling vdev_t *rvd, *bvd, *avd = NULL; 314221ecdf64SLin Ling nvlist_t *config, *nvtop; 314321ecdf64SLin Ling uint64_t guid, txg; 3144e7cbe64fSgw char *pname; 3145e7cbe64fSgw int error; 3146e7cbe64fSgw 3147e7cbe64fSgw /* 314821ecdf64SLin Ling * Read the label from the boot device and generate a configuration. 3149e7cbe64fSgw */ 3150dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3151dedec472SJack Meng #if defined(_OBP) && defined(_KERNEL) 3152dedec472SJack Meng if (config == NULL) { 3153dedec472SJack Meng if (strstr(devpath, "/iscsi/ssd") != NULL) { 3154dedec472SJack Meng /* iscsi boot */ 3155dedec472SJack Meng get_iscsi_bootpath_phy(devpath); 3156dedec472SJack Meng config = spa_generate_rootconf(devpath, devid, &guid); 3157dedec472SJack Meng } 3158dedec472SJack Meng } 3159dedec472SJack Meng #endif 3160dedec472SJack Meng if (config == NULL) { 316121ecdf64SLin Ling cmn_err(CE_NOTE, "Can not read the pool label from '%s'", 316221ecdf64SLin Ling devpath); 316321ecdf64SLin Ling return (EIO); 316421ecdf64SLin Ling } 3165e7cbe64fSgw 316621ecdf64SLin Ling VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 316721ecdf64SLin Ling &pname) == 0); 316821ecdf64SLin Ling VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3169e7cbe64fSgw 31706809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 31716809eb4eSEric Schrock if ((spa = spa_lookup(pname)) != NULL) { 31726809eb4eSEric Schrock /* 31736809eb4eSEric Schrock * Remove the existing root pool from the namespace so that we 31746809eb4eSEric Schrock * can replace it with the correct config we just read in. 31756809eb4eSEric Schrock */ 31766809eb4eSEric Schrock spa_remove(spa); 31776809eb4eSEric Schrock } 31786809eb4eSEric Schrock 3179468c413aSTim Haley spa = spa_add(pname, config, NULL); 31806809eb4eSEric Schrock spa->spa_is_root = B_TRUE; 31814b964adaSGeorge Wilson spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3182e7cbe64fSgw 318321ecdf64SLin Ling /* 318421ecdf64SLin Ling * Build up a vdev tree based on the boot device's label config. 318521ecdf64SLin Ling */ 318621ecdf64SLin Ling VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 318721ecdf64SLin Ling &nvtop) == 0); 318821ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 318921ecdf64SLin Ling error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 319021ecdf64SLin Ling VDEV_ALLOC_ROOTPOOL); 319121ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 319221ecdf64SLin Ling if (error) { 319321ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 319421ecdf64SLin Ling nvlist_free(config); 319521ecdf64SLin Ling cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 319621ecdf64SLin Ling pname); 319721ecdf64SLin Ling return (error); 319821ecdf64SLin Ling } 319921ecdf64SLin Ling 320021ecdf64SLin Ling /* 320121ecdf64SLin Ling * Get the boot vdev. 320221ecdf64SLin Ling */ 320321ecdf64SLin Ling if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 320421ecdf64SLin Ling cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 320521ecdf64SLin Ling (u_longlong_t)guid); 320621ecdf64SLin Ling error = ENOENT; 320721ecdf64SLin Ling goto out; 320821ecdf64SLin Ling } 3209e7cbe64fSgw 321021ecdf64SLin Ling /* 321121ecdf64SLin Ling * Determine if there is a better boot device. 321221ecdf64SLin Ling */ 321321ecdf64SLin Ling avd = bvd; 321421ecdf64SLin Ling spa_alt_rootvdev(rvd, &avd, &txg); 321521ecdf64SLin Ling if (avd != bvd) { 321621ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 321721ecdf64SLin Ling "try booting from '%s'", avd->vdev_path); 321821ecdf64SLin Ling error = EINVAL; 321921ecdf64SLin Ling goto out; 322021ecdf64SLin Ling } 3221e7cbe64fSgw 322221ecdf64SLin Ling /* 322321ecdf64SLin Ling * If the boot device is part of a spare vdev then ensure that 322421ecdf64SLin Ling * we're booting off the active spare. 322521ecdf64SLin Ling */ 322621ecdf64SLin Ling if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 322721ecdf64SLin Ling !bvd->vdev_isspare) { 322821ecdf64SLin Ling cmn_err(CE_NOTE, "The boot device is currently spared. Please " 322921ecdf64SLin Ling "try booting from '%s'", 3230*cb04b873SMark J Musante bvd->vdev_parent-> 3231*cb04b873SMark J Musante vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 323221ecdf64SLin Ling error = EINVAL; 323321ecdf64SLin Ling goto out; 323421ecdf64SLin Ling } 323521ecdf64SLin Ling 323621ecdf64SLin Ling error = 0; 3237c8e1f6d2SMark J Musante spa_history_log_version(spa, LOG_POOL_IMPORT); 323821ecdf64SLin Ling out: 323921ecdf64SLin Ling spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 324021ecdf64SLin Ling vdev_free(rvd); 324121ecdf64SLin Ling spa_config_exit(spa, SCL_ALL, FTAG); 324221ecdf64SLin Ling mutex_exit(&spa_namespace_lock); 324321ecdf64SLin Ling 324421ecdf64SLin Ling nvlist_free(config); 3245e7cbe64fSgw return (error); 3246e7cbe64fSgw } 324721ecdf64SLin Ling 3248e7cbe64fSgw #endif 3249e7cbe64fSgw 32506809eb4eSEric Schrock /* 32516809eb4eSEric Schrock * Import a non-root pool into the system. 32526809eb4eSEric Schrock */ 3253c5904d13Seschrock int 32544b964adaSGeorge Wilson spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3255c5904d13Seschrock { 32566809eb4eSEric Schrock spa_t *spa; 32576809eb4eSEric Schrock char *altroot = NULL; 3258468c413aSTim Haley spa_load_state_t state = SPA_LOAD_IMPORT; 3259468c413aSTim Haley zpool_rewind_policy_t policy; 32606809eb4eSEric Schrock int error; 32616809eb4eSEric Schrock nvlist_t *nvroot; 32626809eb4eSEric Schrock nvlist_t **spares, **l2cache; 32636809eb4eSEric Schrock uint_t nspares, nl2cache; 32646809eb4eSEric Schrock 32656809eb4eSEric Schrock /* 32666809eb4eSEric Schrock * If a pool with this name exists, return failure. 32676809eb4eSEric Schrock */ 32686809eb4eSEric Schrock mutex_enter(&spa_namespace_lock); 32691195e687SMark J Musante if (spa_lookup(pool) != NULL) { 32706809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 32716809eb4eSEric Schrock return (EEXIST); 32726809eb4eSEric Schrock } 32736809eb4eSEric Schrock 32746809eb4eSEric Schrock /* 32756809eb4eSEric Schrock * Create and initialize the spa structure. 32766809eb4eSEric Schrock */ 32776809eb4eSEric Schrock (void) nvlist_lookup_string(props, 32786809eb4eSEric Schrock zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3279468c413aSTim Haley spa = spa_add(pool, config, altroot); 32804b964adaSGeorge Wilson spa->spa_import_flags = flags; 32814b964adaSGeorge Wilson 32824b964adaSGeorge Wilson /* 32834b964adaSGeorge Wilson * Verbatim import - Take a pool and insert it into the namespace 32844b964adaSGeorge Wilson * as if it had been loaded at boot. 32854b964adaSGeorge Wilson */ 32864b964adaSGeorge Wilson if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 32874b964adaSGeorge Wilson if (props != NULL) 32884b964adaSGeorge Wilson spa_configfile_set(spa, props, B_FALSE); 32894b964adaSGeorge Wilson 32904b964adaSGeorge Wilson spa_config_sync(spa, B_FALSE, B_TRUE); 32914b964adaSGeorge Wilson 32924b964adaSGeorge Wilson mutex_exit(&spa_namespace_lock); 32934b964adaSGeorge Wilson spa_history_log_version(spa, LOG_POOL_IMPORT); 32944b964adaSGeorge Wilson 32954b964adaSGeorge Wilson return (0); 32964b964adaSGeorge Wilson } 32974b964adaSGeorge Wilson 32986809eb4eSEric Schrock spa_activate(spa, spa_mode_global); 32996809eb4eSEric Schrock 330025f89ee2SJeff Bonwick /* 330125f89ee2SJeff Bonwick * Don't start async tasks until we know everything is healthy. 330225f89ee2SJeff Bonwick */ 330325f89ee2SJeff Bonwick spa_async_suspend(spa); 330425f89ee2SJeff Bonwick 33054b964adaSGeorge Wilson zpool_get_rewind_policy(config, &policy); 33064b964adaSGeorge Wilson if (policy.zrp_request & ZPOOL_DO_REWIND) 33074b964adaSGeorge Wilson state = SPA_LOAD_RECOVER; 33084b964adaSGeorge Wilson 33096809eb4eSEric Schrock /* 33106809eb4eSEric Schrock * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 33116809eb4eSEric Schrock * because the user-supplied config is actually the one to trust when 33126809eb4eSEric Schrock * doing an import. 33136809eb4eSEric Schrock */ 3314468c413aSTim Haley if (state != SPA_LOAD_RECOVER) 3315468c413aSTim Haley spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 33164b964adaSGeorge Wilson 3317468c413aSTim Haley error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 3318c8ee1847SVictor Latushkin policy.zrp_request); 3319468c413aSTim Haley 3320468c413aSTim Haley /* 33214b964adaSGeorge Wilson * Propagate anything learned while loading the pool and pass it 33224b964adaSGeorge Wilson * back to caller (i.e. rewind info, missing devices, etc). 3323468c413aSTim Haley */ 33244b964adaSGeorge Wilson VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 33254b964adaSGeorge Wilson spa->spa_load_info) == 0); 33266809eb4eSEric Schrock 33276809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 33286809eb4eSEric Schrock /* 33296809eb4eSEric Schrock * Toss any existing sparelist, as it doesn't have any validity 33306809eb4eSEric Schrock * anymore, and conflicts with spa_has_spare(). 33316809eb4eSEric Schrock */ 33326809eb4eSEric Schrock if (spa->spa_spares.sav_config) { 33336809eb4eSEric Schrock nvlist_free(spa->spa_spares.sav_config); 33346809eb4eSEric Schrock spa->spa_spares.sav_config = NULL; 33356809eb4eSEric Schrock spa_load_spares(spa); 33366809eb4eSEric Schrock } 33376809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) { 33386809eb4eSEric Schrock nvlist_free(spa->spa_l2cache.sav_config); 33396809eb4eSEric Schrock spa->spa_l2cache.sav_config = NULL; 33406809eb4eSEric Schrock spa_load_l2cache(spa); 33416809eb4eSEric Schrock } 33426809eb4eSEric Schrock 33436809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 33446809eb4eSEric Schrock &nvroot) == 0); 33456809eb4eSEric Schrock if (error == 0) 33466809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 33476809eb4eSEric Schrock VDEV_ALLOC_SPARE); 33486809eb4eSEric Schrock if (error == 0) 33496809eb4eSEric Schrock error = spa_validate_aux(spa, nvroot, -1ULL, 33506809eb4eSEric Schrock VDEV_ALLOC_L2CACHE); 33516809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 33526809eb4eSEric Schrock 33536809eb4eSEric Schrock if (props != NULL) 33546809eb4eSEric Schrock spa_configfile_set(spa, props, B_FALSE); 33556809eb4eSEric Schrock 33566809eb4eSEric Schrock if (error != 0 || (props && spa_writeable(spa) && 33576809eb4eSEric Schrock (error = spa_prop_set(spa, props)))) { 33586809eb4eSEric Schrock spa_unload(spa); 33596809eb4eSEric Schrock spa_deactivate(spa); 33606809eb4eSEric Schrock spa_remove(spa); 33616809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 33626809eb4eSEric Schrock return (error); 33636809eb4eSEric Schrock } 33646809eb4eSEric Schrock 3365955ef359SLin Ling spa_async_resume(spa); 3366955ef359SLin Ling 33676809eb4eSEric Schrock /* 33686809eb4eSEric Schrock * Override any spares and level 2 cache devices as specified by 33696809eb4eSEric Schrock * the user, as these may have correct device names/devids, etc. 33706809eb4eSEric Schrock */ 33716809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 33726809eb4eSEric Schrock &spares, &nspares) == 0) { 33736809eb4eSEric Schrock if (spa->spa_spares.sav_config) 33746809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_spares.sav_config, 33756809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 33766809eb4eSEric Schrock else 33776809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 33786809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 33796809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 33806809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 33816809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 33826809eb4eSEric Schrock spa_load_spares(spa); 33836809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 33846809eb4eSEric Schrock spa->spa_spares.sav_sync = B_TRUE; 33856809eb4eSEric Schrock } 33866809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 33876809eb4eSEric Schrock &l2cache, &nl2cache) == 0) { 33886809eb4eSEric Schrock if (spa->spa_l2cache.sav_config) 33896809eb4eSEric Schrock VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 33906809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 33916809eb4eSEric Schrock else 33926809eb4eSEric Schrock VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 33936809eb4eSEric Schrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 33946809eb4eSEric Schrock VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 33956809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 33966809eb4eSEric Schrock spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 33976809eb4eSEric Schrock spa_load_l2cache(spa); 33986809eb4eSEric Schrock spa_config_exit(spa, SCL_ALL, FTAG); 33996809eb4eSEric Schrock spa->spa_l2cache.sav_sync = B_TRUE; 34006809eb4eSEric Schrock } 34016809eb4eSEric Schrock 3402b693757aSEric Schrock /* 3403b693757aSEric Schrock * Check for any removed devices. 3404b693757aSEric Schrock */ 3405b693757aSEric Schrock if (spa->spa_autoreplace) { 3406b693757aSEric Schrock spa_aux_check_removed(&spa->spa_spares); 3407b693757aSEric Schrock spa_aux_check_removed(&spa->spa_l2cache); 3408b693757aSEric Schrock } 3409b693757aSEric Schrock 34106809eb4eSEric Schrock if (spa_writeable(spa)) { 34116809eb4eSEric Schrock /* 34126809eb4eSEric Schrock * Update the config cache to include the newly-imported pool. 34136809eb4eSEric Schrock */ 3414bc758434SLin Ling spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 34156809eb4eSEric Schrock } 34166809eb4eSEric Schrock 3417573ca77eSGeorge Wilson /* 3418573ca77eSGeorge Wilson * It's possible that the pool was expanded while it was exported. 3419573ca77eSGeorge Wilson * We kick off an async task to handle this for us. 3420573ca77eSGeorge Wilson */ 3421573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 3422573ca77eSGeorge Wilson 34236809eb4eSEric Schrock mutex_exit(&spa_namespace_lock); 3424c8e1f6d2SMark J Musante spa_history_log_version(spa, LOG_POOL_IMPORT); 34256809eb4eSEric Schrock 34266809eb4eSEric Schrock return (0); 3427c5904d13Seschrock } 3428c5904d13Seschrock 3429fa9e4066Sahrens nvlist_t * 3430fa9e4066Sahrens spa_tryimport(nvlist_t *tryconfig) 3431fa9e4066Sahrens { 3432fa9e4066Sahrens nvlist_t *config = NULL; 3433fa9e4066Sahrens char *poolname; 3434fa9e4066Sahrens spa_t *spa; 3435fa9e4066Sahrens uint64_t state; 34367b7154beSLin Ling int error; 3437fa9e4066Sahrens 3438fa9e4066Sahrens if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 3439fa9e4066Sahrens return (NULL); 3440fa9e4066Sahrens 3441fa9e4066Sahrens if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 3442fa9e4066Sahrens return (NULL); 3443fa9e4066Sahrens 3444fa9e4066Sahrens /* 34450373e76bSbonwick * Create and initialize the spa structure. 3446fa9e4066Sahrens */ 34470373e76bSbonwick mutex_enter(&spa_namespace_lock); 3448468c413aSTim Haley spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 34498ad4d6ddSJeff Bonwick spa_activate(spa, FREAD); 3450fa9e4066Sahrens 3451fa9e4066Sahrens /* 34520373e76bSbonwick * Pass off the heavy lifting to spa_load(). 3453ecc2d604Sbonwick * Pass TRUE for mosconfig because the user-supplied config 3454ecc2d604Sbonwick * is actually the one to trust when doing an import. 3455fa9e4066Sahrens */ 34561195e687SMark J Musante error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 3457fa9e4066Sahrens 3458fa9e4066Sahrens /* 3459fa9e4066Sahrens * If 'tryconfig' was at least parsable, return the current config. 3460fa9e4066Sahrens */ 3461fa9e4066Sahrens if (spa->spa_root_vdev != NULL) { 3462fa9e4066Sahrens config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3463fa9e4066Sahrens VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 3464fa9e4066Sahrens poolname) == 0); 3465fa9e4066Sahrens VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 3466fa9e4066Sahrens state) == 0); 346795173954Sek VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 346895173954Sek spa->spa_uberblock.ub_timestamp) == 0); 346999653d4eSeschrock 3470e7cbe64fSgw /* 3471e7cbe64fSgw * If the bootfs property exists on this pool then we 3472e7cbe64fSgw * copy it out so that external consumers can tell which 3473e7cbe64fSgw * pools are bootable. 3474e7cbe64fSgw */ 34757b7154beSLin Ling if ((!error || error == EEXIST) && spa->spa_bootfs) { 3476e7cbe64fSgw char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3477e7cbe64fSgw 3478e7cbe64fSgw /* 3479e7cbe64fSgw * We have to play games with the name since the 3480e7cbe64fSgw * pool was opened as TRYIMPORT_NAME. 3481e7cbe64fSgw */ 3482e14bb325SJeff Bonwick if (dsl_dsobj_to_dsname(spa_name(spa), 3483e7cbe64fSgw spa->spa_bootfs, tmpname) == 0) { 3484e7cbe64fSgw char *cp; 3485e7cbe64fSgw char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3486e7cbe64fSgw 3487e7cbe64fSgw cp = strchr(tmpname, '/'); 3488e7cbe64fSgw if (cp == NULL) { 3489e7cbe64fSgw (void) strlcpy(dsname, tmpname, 3490e7cbe64fSgw MAXPATHLEN); 3491e7cbe64fSgw } else { 3492e7cbe64fSgw (void) snprintf(dsname, MAXPATHLEN, 3493e7cbe64fSgw "%s/%s", poolname, ++cp); 3494e7cbe64fSgw } 3495e7cbe64fSgw VERIFY(nvlist_add_string(config, 3496e7cbe64fSgw ZPOOL_CONFIG_BOOTFS, dsname) == 0); 3497e7cbe64fSgw kmem_free(dsname, MAXPATHLEN); 3498e7cbe64fSgw } 3499e7cbe64fSgw kmem_free(tmpname, MAXPATHLEN); 3500e7cbe64fSgw } 3501e7cbe64fSgw 350299653d4eSeschrock /* 3503fa94a07fSbrendan * Add the list of hot spares and level 2 cache devices. 350499653d4eSeschrock */ 35056809eb4eSEric Schrock spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 350699653d4eSeschrock spa_add_spares(spa, config); 3507fa94a07fSbrendan spa_add_l2cache(spa, config); 35086809eb4eSEric Schrock spa_config_exit(spa, SCL_CONFIG, FTAG); 3509fa9e4066Sahrens } 3510fa9e4066Sahrens 3511fa9e4066Sahrens spa_unload(spa); 3512fa9e4066Sahrens spa_deactivate(spa); 3513fa9e4066Sahrens spa_remove(spa); 3514fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3515fa9e4066Sahrens 3516fa9e4066Sahrens return (config); 3517fa9e4066Sahrens } 3518fa9e4066Sahrens 3519fa9e4066Sahrens /* 3520fa9e4066Sahrens * Pool export/destroy 3521fa9e4066Sahrens * 3522fa9e4066Sahrens * The act of destroying or exporting a pool is very simple. We make sure there 3523fa9e4066Sahrens * is no more pending I/O and any references to the pool are gone. Then, we 3524fa9e4066Sahrens * update the pool state and sync all the labels to disk, removing the 3525394ab0cbSGeorge Wilson * configuration from the cache afterwards. If the 'hardforce' flag is set, then 3526394ab0cbSGeorge Wilson * we don't sync the labels or remove the configuration cache. 3527fa9e4066Sahrens */ 3528fa9e4066Sahrens static int 352989a89ebfSlling spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 3530394ab0cbSGeorge Wilson boolean_t force, boolean_t hardforce) 3531fa9e4066Sahrens { 3532fa9e4066Sahrens spa_t *spa; 3533fa9e4066Sahrens 353444cd46caSbillm if (oldconfig) 353544cd46caSbillm *oldconfig = NULL; 353644cd46caSbillm 35378ad4d6ddSJeff Bonwick if (!(spa_mode_global & FWRITE)) 3538fa9e4066Sahrens return (EROFS); 3539fa9e4066Sahrens 3540fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 3541fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 3542fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3543fa9e4066Sahrens return (ENOENT); 3544fa9e4066Sahrens } 3545fa9e4066Sahrens 3546ea8dc4b6Seschrock /* 3547ea8dc4b6Seschrock * Put a hold on the pool, drop the namespace lock, stop async tasks, 3548ea8dc4b6Seschrock * reacquire the namespace lock, and see if we can export. 3549ea8dc4b6Seschrock */ 3550ea8dc4b6Seschrock spa_open_ref(spa, FTAG); 3551ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 3552ea8dc4b6Seschrock spa_async_suspend(spa); 3553ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 3554ea8dc4b6Seschrock spa_close(spa, FTAG); 3555ea8dc4b6Seschrock 3556fa9e4066Sahrens /* 3557fa9e4066Sahrens * The pool will be in core if it's openable, 3558fa9e4066Sahrens * in which case we can modify its state. 3559fa9e4066Sahrens */ 3560fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 3561fa9e4066Sahrens /* 3562fa9e4066Sahrens * Objsets may be open only because they're dirty, so we 3563fa9e4066Sahrens * have to force it to sync before checking spa_refcnt. 3564fa9e4066Sahrens */ 3565fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, 0); 3566fa9e4066Sahrens 3567ea8dc4b6Seschrock /* 3568ea8dc4b6Seschrock * A pool cannot be exported or destroyed if there are active 3569ea8dc4b6Seschrock * references. If we are resetting a pool, allow references by 3570ea8dc4b6Seschrock * fault injection handlers. 3571ea8dc4b6Seschrock */ 3572ea8dc4b6Seschrock if (!spa_refcount_zero(spa) || 3573ea8dc4b6Seschrock (spa->spa_inject_ref != 0 && 3574ea8dc4b6Seschrock new_state != POOL_STATE_UNINITIALIZED)) { 3575ea8dc4b6Seschrock spa_async_resume(spa); 3576fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3577fa9e4066Sahrens return (EBUSY); 3578fa9e4066Sahrens } 3579fa9e4066Sahrens 358089a89ebfSlling /* 358189a89ebfSlling * A pool cannot be exported if it has an active shared spare. 358289a89ebfSlling * This is to prevent other pools stealing the active spare 358389a89ebfSlling * from an exported pool. At user's own will, such pool can 358489a89ebfSlling * be forcedly exported. 358589a89ebfSlling */ 358689a89ebfSlling if (!force && new_state == POOL_STATE_EXPORTED && 358789a89ebfSlling spa_has_active_shared_spare(spa)) { 358889a89ebfSlling spa_async_resume(spa); 358989a89ebfSlling mutex_exit(&spa_namespace_lock); 359089a89ebfSlling return (EXDEV); 359189a89ebfSlling } 359289a89ebfSlling 3593fa9e4066Sahrens /* 3594fa9e4066Sahrens * We want this to be reflected on every label, 3595fa9e4066Sahrens * so mark them all dirty. spa_unload() will do the 3596fa9e4066Sahrens * final sync that pushes these changes out. 3597fa9e4066Sahrens */ 3598394ab0cbSGeorge Wilson if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 3599e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3600ea8dc4b6Seschrock spa->spa_state = new_state; 36013f9d6ad7SLin Ling spa->spa_final_txg = spa_last_synced_txg(spa) + 36023f9d6ad7SLin Ling TXG_DEFER_SIZE + 1; 3603ea8dc4b6Seschrock vdev_config_dirty(spa->spa_root_vdev); 3604e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 3605ea8dc4b6Seschrock } 3606fa9e4066Sahrens } 3607fa9e4066Sahrens 36083d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 36093d7072f8Seschrock 3610fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 3611fa9e4066Sahrens spa_unload(spa); 3612fa9e4066Sahrens spa_deactivate(spa); 3613fa9e4066Sahrens } 3614fa9e4066Sahrens 361544cd46caSbillm if (oldconfig && spa->spa_config) 361644cd46caSbillm VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 361744cd46caSbillm 3618ea8dc4b6Seschrock if (new_state != POOL_STATE_UNINITIALIZED) { 3619394ab0cbSGeorge Wilson if (!hardforce) 3620394ab0cbSGeorge Wilson spa_config_sync(spa, B_TRUE, B_TRUE); 3621ea8dc4b6Seschrock spa_remove(spa); 3622ea8dc4b6Seschrock } 3623fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 3624fa9e4066Sahrens 3625fa9e4066Sahrens return (0); 3626fa9e4066Sahrens } 3627fa9e4066Sahrens 3628fa9e4066Sahrens /* 3629fa9e4066Sahrens * Destroy a storage pool. 3630fa9e4066Sahrens */ 3631fa9e4066Sahrens int 3632fa9e4066Sahrens spa_destroy(char *pool) 3633fa9e4066Sahrens { 3634394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 3635394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 3636fa9e4066Sahrens } 3637fa9e4066Sahrens 3638fa9e4066Sahrens /* 3639fa9e4066Sahrens * Export a storage pool. 3640fa9e4066Sahrens */ 3641fa9e4066Sahrens int 3642394ab0cbSGeorge Wilson spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 3643394ab0cbSGeorge Wilson boolean_t hardforce) 3644fa9e4066Sahrens { 3645394ab0cbSGeorge Wilson return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 3646394ab0cbSGeorge Wilson force, hardforce)); 3647fa9e4066Sahrens } 3648fa9e4066Sahrens 3649ea8dc4b6Seschrock /* 3650ea8dc4b6Seschrock * Similar to spa_export(), this unloads the spa_t without actually removing it 3651ea8dc4b6Seschrock * from the namespace in any way. 3652ea8dc4b6Seschrock */ 3653ea8dc4b6Seschrock int 3654ea8dc4b6Seschrock spa_reset(char *pool) 3655ea8dc4b6Seschrock { 365689a89ebfSlling return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 3657394ab0cbSGeorge Wilson B_FALSE, B_FALSE)); 3658ea8dc4b6Seschrock } 3659ea8dc4b6Seschrock 3660fa9e4066Sahrens /* 3661fa9e4066Sahrens * ========================================================================== 3662fa9e4066Sahrens * Device manipulation 3663fa9e4066Sahrens * ========================================================================== 3664fa9e4066Sahrens */ 3665fa9e4066Sahrens 3666fa9e4066Sahrens /* 36678654d025Sperrin * Add a device to a storage pool. 3668fa9e4066Sahrens */ 3669fa9e4066Sahrens int 3670fa9e4066Sahrens spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 3671fa9e4066Sahrens { 367288ecc943SGeorge Wilson uint64_t txg, id; 36738ad4d6ddSJeff Bonwick int error; 3674fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 36750e34b6a7Sbonwick vdev_t *vd, *tvd; 3676fa94a07fSbrendan nvlist_t **spares, **l2cache; 3677fa94a07fSbrendan uint_t nspares, nl2cache; 3678fa9e4066Sahrens 3679fa9e4066Sahrens txg = spa_vdev_enter(spa); 3680fa9e4066Sahrens 368199653d4eSeschrock if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 368299653d4eSeschrock VDEV_ALLOC_ADD)) != 0) 368399653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, error)); 3684fa9e4066Sahrens 3685e14bb325SJeff Bonwick spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 368699653d4eSeschrock 3687fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 3688fa94a07fSbrendan &nspares) != 0) 368999653d4eSeschrock nspares = 0; 369099653d4eSeschrock 3691fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 3692fa94a07fSbrendan &nl2cache) != 0) 3693fa94a07fSbrendan nl2cache = 0; 3694fa94a07fSbrendan 3695e14bb325SJeff Bonwick if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 3696fa9e4066Sahrens return (spa_vdev_exit(spa, vd, txg, EINVAL)); 3697fa9e4066Sahrens 3698e14bb325SJeff Bonwick if (vd->vdev_children != 0 && 3699e14bb325SJeff Bonwick (error = vdev_create(vd, txg, B_FALSE)) != 0) 3700e14bb325SJeff Bonwick return (spa_vdev_exit(spa, vd, txg, error)); 370199653d4eSeschrock 370239c23413Seschrock /* 3703fa94a07fSbrendan * We must validate the spares and l2cache devices after checking the 3704fa94a07fSbrendan * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 370539c23413Seschrock */ 3706e14bb325SJeff Bonwick if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 370739c23413Seschrock return (spa_vdev_exit(spa, vd, txg, error)); 370839c23413Seschrock 370939c23413Seschrock /* 371039c23413Seschrock * Transfer each new top-level vdev from vd to rvd. 371139c23413Seschrock */ 37128ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 371388ecc943SGeorge Wilson 371488ecc943SGeorge Wilson /* 371588ecc943SGeorge Wilson * Set the vdev id to the first hole, if one exists. 371688ecc943SGeorge Wilson */ 371788ecc943SGeorge Wilson for (id = 0; id < rvd->vdev_children; id++) { 371888ecc943SGeorge Wilson if (rvd->vdev_child[id]->vdev_ishole) { 371988ecc943SGeorge Wilson vdev_free(rvd->vdev_child[id]); 372088ecc943SGeorge Wilson break; 372188ecc943SGeorge Wilson } 372288ecc943SGeorge Wilson } 372339c23413Seschrock tvd = vd->vdev_child[c]; 372439c23413Seschrock vdev_remove_child(vd, tvd); 372588ecc943SGeorge Wilson tvd->vdev_id = id; 372639c23413Seschrock vdev_add_child(rvd, tvd); 372739c23413Seschrock vdev_config_dirty(tvd); 372839c23413Seschrock } 372939c23413Seschrock 373099653d4eSeschrock if (nspares != 0) { 3731fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 3732fa94a07fSbrendan ZPOOL_CONFIG_SPARES); 373399653d4eSeschrock spa_load_spares(spa); 3734fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 3735fa94a07fSbrendan } 3736fa94a07fSbrendan 3737fa94a07fSbrendan if (nl2cache != 0) { 3738fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 3739fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE); 3740fa94a07fSbrendan spa_load_l2cache(spa); 3741fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 3742fa9e4066Sahrens } 3743fa9e4066Sahrens 3744fa9e4066Sahrens /* 37450e34b6a7Sbonwick * We have to be careful when adding new vdevs to an existing pool. 37460e34b6a7Sbonwick * If other threads start allocating from these vdevs before we 37470e34b6a7Sbonwick * sync the config cache, and we lose power, then upon reboot we may 37480e34b6a7Sbonwick * fail to open the pool because there are DVAs that the config cache 37490e34b6a7Sbonwick * can't translate. Therefore, we first add the vdevs without 37500e34b6a7Sbonwick * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 37510373e76bSbonwick * and then let spa_config_update() initialize the new metaslabs. 37520e34b6a7Sbonwick * 37530e34b6a7Sbonwick * spa_load() checks for added-but-not-initialized vdevs, so that 37540e34b6a7Sbonwick * if we lose power at any point in this sequence, the remaining 37550e34b6a7Sbonwick * steps will be completed the next time we load the pool. 37560e34b6a7Sbonwick */ 37570373e76bSbonwick (void) spa_vdev_exit(spa, vd, txg, 0); 37580e34b6a7Sbonwick 37590373e76bSbonwick mutex_enter(&spa_namespace_lock); 37600373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 37610373e76bSbonwick mutex_exit(&spa_namespace_lock); 3762fa9e4066Sahrens 37630373e76bSbonwick return (0); 3764fa9e4066Sahrens } 3765fa9e4066Sahrens 3766fa9e4066Sahrens /* 3767fa9e4066Sahrens * Attach a device to a mirror. The arguments are the path to any device 3768fa9e4066Sahrens * in the mirror, and the nvroot for the new device. If the path specifies 3769fa9e4066Sahrens * a device that is not mirrored, we automatically insert the mirror vdev. 3770fa9e4066Sahrens * 3771fa9e4066Sahrens * If 'replacing' is specified, the new device is intended to replace the 3772fa9e4066Sahrens * existing device; in this case the two devices are made into their own 37733d7072f8Seschrock * mirror using the 'replacing' vdev, which is functionally identical to 3774fa9e4066Sahrens * the mirror vdev (it actually reuses all the same ops) but has a few 3775fa9e4066Sahrens * extra rules: you can't attach to it after it's been created, and upon 3776fa9e4066Sahrens * completion of resilvering, the first disk (the one being replaced) 3777fa9e4066Sahrens * is automatically detached. 3778fa9e4066Sahrens */ 3779fa9e4066Sahrens int 3780ea8dc4b6Seschrock spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 3781fa9e4066Sahrens { 37823f9d6ad7SLin Ling uint64_t txg, dtl_max_txg; 3783fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3784fa9e4066Sahrens vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 378599653d4eSeschrock vdev_ops_t *pvops; 37869b3f6b42SEric Kustarz char *oldvdpath, *newvdpath; 37879b3f6b42SEric Kustarz int newvd_isspare; 37889b3f6b42SEric Kustarz int error; 3789fa9e4066Sahrens 3790fa9e4066Sahrens txg = spa_vdev_enter(spa); 3791fa9e4066Sahrens 3792c5904d13Seschrock oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 3793fa9e4066Sahrens 3794fa9e4066Sahrens if (oldvd == NULL) 3795fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 3796fa9e4066Sahrens 37970e34b6a7Sbonwick if (!oldvd->vdev_ops->vdev_op_leaf) 37980e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 37990e34b6a7Sbonwick 3800fa9e4066Sahrens pvd = oldvd->vdev_parent; 3801fa9e4066Sahrens 380299653d4eSeschrock if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 38033d7072f8Seschrock VDEV_ALLOC_ADD)) != 0) 38043d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 38053d7072f8Seschrock 38063d7072f8Seschrock if (newrootvd->vdev_children != 1) 3807fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 3808fa9e4066Sahrens 3809fa9e4066Sahrens newvd = newrootvd->vdev_child[0]; 3810fa9e4066Sahrens 3811fa9e4066Sahrens if (!newvd->vdev_ops->vdev_op_leaf) 3812fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 3813fa9e4066Sahrens 381499653d4eSeschrock if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 3815fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, error)); 3816fa9e4066Sahrens 38178654d025Sperrin /* 38188654d025Sperrin * Spares can't replace logs 38198654d025Sperrin */ 3820ee0eb9f2SEric Schrock if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 38218654d025Sperrin return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 38228654d025Sperrin 382399653d4eSeschrock if (!replacing) { 382499653d4eSeschrock /* 382599653d4eSeschrock * For attach, the only allowable parent is a mirror or the root 382699653d4eSeschrock * vdev. 382799653d4eSeschrock */ 382899653d4eSeschrock if (pvd->vdev_ops != &vdev_mirror_ops && 382999653d4eSeschrock pvd->vdev_ops != &vdev_root_ops) 383099653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 383199653d4eSeschrock 383299653d4eSeschrock pvops = &vdev_mirror_ops; 383399653d4eSeschrock } else { 383499653d4eSeschrock /* 383599653d4eSeschrock * Active hot spares can only be replaced by inactive hot 383699653d4eSeschrock * spares. 383799653d4eSeschrock */ 383899653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 3839*cb04b873SMark J Musante oldvd->vdev_isspare && 384099653d4eSeschrock !spa_has_spare(spa, newvd->vdev_guid)) 384199653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 384299653d4eSeschrock 384399653d4eSeschrock /* 384499653d4eSeschrock * If the source is a hot spare, and the parent isn't already a 384599653d4eSeschrock * spare, then we want to create a new hot spare. Otherwise, we 384639c23413Seschrock * want to create a replacing vdev. The user is not allowed to 384739c23413Seschrock * attach to a spared vdev child unless the 'isspare' state is 384839c23413Seschrock * the same (spare replaces spare, non-spare replaces 384939c23413Seschrock * non-spare). 385099653d4eSeschrock */ 3851*cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && 3852*cb04b873SMark J Musante spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 385399653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 3854*cb04b873SMark J Musante } else if (pvd->vdev_ops == &vdev_spare_ops && 3855*cb04b873SMark J Musante newvd->vdev_isspare != oldvd->vdev_isspare) { 385639c23413Seschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 3857*cb04b873SMark J Musante } 3858*cb04b873SMark J Musante 3859*cb04b873SMark J Musante if (newvd->vdev_isspare) 386099653d4eSeschrock pvops = &vdev_spare_ops; 386199653d4eSeschrock else 386299653d4eSeschrock pvops = &vdev_replacing_ops; 386399653d4eSeschrock } 386499653d4eSeschrock 38652a79c5feSlling /* 3866573ca77eSGeorge Wilson * Make sure the new device is big enough. 38672a79c5feSlling */ 3868573ca77eSGeorge Wilson if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 3869fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 3870fa9e4066Sahrens 3871ecc2d604Sbonwick /* 3872ecc2d604Sbonwick * The new device cannot have a higher alignment requirement 3873ecc2d604Sbonwick * than the top-level vdev. 3874ecc2d604Sbonwick */ 3875ecc2d604Sbonwick if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 3876fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 3877fa9e4066Sahrens 3878fa9e4066Sahrens /* 3879fa9e4066Sahrens * If this is an in-place replacement, update oldvd's path and devid 3880fa9e4066Sahrens * to make it distinguishable from newvd, and unopenable from now on. 3881fa9e4066Sahrens */ 3882fa9e4066Sahrens if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 3883fa9e4066Sahrens spa_strfree(oldvd->vdev_path); 3884fa9e4066Sahrens oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 3885fa9e4066Sahrens KM_SLEEP); 3886fa9e4066Sahrens (void) sprintf(oldvd->vdev_path, "%s/%s", 3887fa9e4066Sahrens newvd->vdev_path, "old"); 3888fa9e4066Sahrens if (oldvd->vdev_devid != NULL) { 3889fa9e4066Sahrens spa_strfree(oldvd->vdev_devid); 3890fa9e4066Sahrens oldvd->vdev_devid = NULL; 3891fa9e4066Sahrens } 3892fa9e4066Sahrens } 3893fa9e4066Sahrens 3894*cb04b873SMark J Musante /* mark the device being resilvered */ 3895*cb04b873SMark J Musante newvd->vdev_resilvering = B_TRUE; 3896*cb04b873SMark J Musante 3897fa9e4066Sahrens /* 389899653d4eSeschrock * If the parent is not a mirror, or if we're replacing, insert the new 389999653d4eSeschrock * mirror/replacing/spare vdev above oldvd. 3900fa9e4066Sahrens */ 3901fa9e4066Sahrens if (pvd->vdev_ops != pvops) 3902fa9e4066Sahrens pvd = vdev_add_parent(oldvd, pvops); 3903fa9e4066Sahrens 3904fa9e4066Sahrens ASSERT(pvd->vdev_top->vdev_parent == rvd); 3905fa9e4066Sahrens ASSERT(pvd->vdev_ops == pvops); 3906fa9e4066Sahrens ASSERT(oldvd->vdev_parent == pvd); 3907fa9e4066Sahrens 3908fa9e4066Sahrens /* 3909fa9e4066Sahrens * Extract the new device from its root and add it to pvd. 3910fa9e4066Sahrens */ 3911fa9e4066Sahrens vdev_remove_child(newrootvd, newvd); 3912fa9e4066Sahrens newvd->vdev_id = pvd->vdev_children; 391388ecc943SGeorge Wilson newvd->vdev_crtxg = oldvd->vdev_crtxg; 3914fa9e4066Sahrens vdev_add_child(pvd, newvd); 3915fa9e4066Sahrens 3916fa9e4066Sahrens tvd = newvd->vdev_top; 3917fa9e4066Sahrens ASSERT(pvd->vdev_top == tvd); 3918fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 3919fa9e4066Sahrens 3920fa9e4066Sahrens vdev_config_dirty(tvd); 3921fa9e4066Sahrens 3922fa9e4066Sahrens /* 39233f9d6ad7SLin Ling * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 39243f9d6ad7SLin Ling * for any dmu_sync-ed blocks. It will propagate upward when 39253f9d6ad7SLin Ling * spa_vdev_exit() calls vdev_dtl_reassess(). 3926fa9e4066Sahrens */ 39273f9d6ad7SLin Ling dtl_max_txg = txg + TXG_CONCURRENT_STATES; 3928fa9e4066Sahrens 39293f9d6ad7SLin Ling vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 39303f9d6ad7SLin Ling dtl_max_txg - TXG_INITIAL); 3931fa9e4066Sahrens 39326809eb4eSEric Schrock if (newvd->vdev_isspare) { 393339c23413Seschrock spa_spare_activate(newvd); 39346809eb4eSEric Schrock spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 39356809eb4eSEric Schrock } 39366809eb4eSEric Schrock 3937e14bb325SJeff Bonwick oldvdpath = spa_strdup(oldvd->vdev_path); 3938e14bb325SJeff Bonwick newvdpath = spa_strdup(newvd->vdev_path); 39399b3f6b42SEric Kustarz newvd_isspare = newvd->vdev_isspare; 3940ea8dc4b6Seschrock 3941fa9e4066Sahrens /* 3942fa9e4066Sahrens * Mark newvd's DTL dirty in this txg. 3943fa9e4066Sahrens */ 3944ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, newvd, txg); 3945fa9e4066Sahrens 39463f9d6ad7SLin Ling /* 39473f9d6ad7SLin Ling * Restart the resilver 39483f9d6ad7SLin Ling */ 39493f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 39503f9d6ad7SLin Ling 39513f9d6ad7SLin Ling /* 39523f9d6ad7SLin Ling * Commit the config 39533f9d6ad7SLin Ling */ 39543f9d6ad7SLin Ling (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 3955fa9e4066Sahrens 39563f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL, 39573f9d6ad7SLin Ling "%s vdev=%s %s vdev=%s", 3958c8e1f6d2SMark J Musante replacing && newvd_isspare ? "spare in" : 3959c8e1f6d2SMark J Musante replacing ? "replace" : "attach", newvdpath, 3960c8e1f6d2SMark J Musante replacing ? "for" : "to", oldvdpath); 39619b3f6b42SEric Kustarz 39629b3f6b42SEric Kustarz spa_strfree(oldvdpath); 39639b3f6b42SEric Kustarz spa_strfree(newvdpath); 39649b3f6b42SEric Kustarz 3965943e9869SLori Alt if (spa->spa_bootfs) 3966943e9869SLori Alt spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 3967943e9869SLori Alt 3968fa9e4066Sahrens return (0); 3969fa9e4066Sahrens } 3970fa9e4066Sahrens 3971fa9e4066Sahrens /* 3972fa9e4066Sahrens * Detach a device from a mirror or replacing vdev. 3973fa9e4066Sahrens * If 'replace_done' is specified, only detach if the parent 3974fa9e4066Sahrens * is a replacing vdev. 3975fa9e4066Sahrens */ 3976fa9e4066Sahrens int 39778ad4d6ddSJeff Bonwick spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 3978fa9e4066Sahrens { 3979fa9e4066Sahrens uint64_t txg; 39808ad4d6ddSJeff Bonwick int error; 3981fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3982fa9e4066Sahrens vdev_t *vd, *pvd, *cvd, *tvd; 398399653d4eSeschrock boolean_t unspare = B_FALSE; 398499653d4eSeschrock uint64_t unspare_guid; 39851195e687SMark J Musante char *vdpath; 3986fa9e4066Sahrens 3987fa9e4066Sahrens txg = spa_vdev_enter(spa); 3988fa9e4066Sahrens 3989c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 3990fa9e4066Sahrens 3991fa9e4066Sahrens if (vd == NULL) 3992fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 3993fa9e4066Sahrens 39940e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 39950e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 39960e34b6a7Sbonwick 3997fa9e4066Sahrens pvd = vd->vdev_parent; 3998fa9e4066Sahrens 39998ad4d6ddSJeff Bonwick /* 40008ad4d6ddSJeff Bonwick * If the parent/child relationship is not as expected, don't do it. 40018ad4d6ddSJeff Bonwick * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 40028ad4d6ddSJeff Bonwick * vdev that's replacing B with C. The user's intent in replacing 40038ad4d6ddSJeff Bonwick * is to go from M(A,B) to M(A,C). If the user decides to cancel 40048ad4d6ddSJeff Bonwick * the replace by detaching C, the expected behavior is to end up 40058ad4d6ddSJeff Bonwick * M(A,B). But suppose that right after deciding to detach C, 40068ad4d6ddSJeff Bonwick * the replacement of B completes. We would have M(A,C), and then 40078ad4d6ddSJeff Bonwick * ask to detach C, which would leave us with just A -- not what 40088ad4d6ddSJeff Bonwick * the user wanted. To prevent this, we make sure that the 40098ad4d6ddSJeff Bonwick * parent/child relationship hasn't changed -- in this example, 40108ad4d6ddSJeff Bonwick * that C's parent is still the replacing vdev R. 40118ad4d6ddSJeff Bonwick */ 40128ad4d6ddSJeff Bonwick if (pvd->vdev_guid != pguid && pguid != 0) 40138ad4d6ddSJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 40148ad4d6ddSJeff Bonwick 4015fa9e4066Sahrens /* 4016*cb04b873SMark J Musante * Only 'replacing' or 'spare' vdevs can be replaced. 401799653d4eSeschrock */ 4018*cb04b873SMark J Musante if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4019*cb04b873SMark J Musante pvd->vdev_ops != &vdev_spare_ops) 4020*cb04b873SMark J Musante return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 402199653d4eSeschrock 402299653d4eSeschrock ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4023e7437265Sahrens spa_version(spa) >= SPA_VERSION_SPARES); 4024fa9e4066Sahrens 4025fa9e4066Sahrens /* 402699653d4eSeschrock * Only mirror, replacing, and spare vdevs support detach. 4027fa9e4066Sahrens */ 4028fa9e4066Sahrens if (pvd->vdev_ops != &vdev_replacing_ops && 402999653d4eSeschrock pvd->vdev_ops != &vdev_mirror_ops && 403099653d4eSeschrock pvd->vdev_ops != &vdev_spare_ops) 4031fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4032fa9e4066Sahrens 4033fa9e4066Sahrens /* 40348ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 40358ad4d6ddSJeff Bonwick * we cannot safely detach it. 4036fa9e4066Sahrens */ 40378ad4d6ddSJeff Bonwick if (vdev_dtl_required(vd)) 4038fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4039fa9e4066Sahrens 40408ad4d6ddSJeff Bonwick ASSERT(pvd->vdev_children >= 2); 4041fa9e4066Sahrens 4042bf82a41bSeschrock /* 4043bf82a41bSeschrock * If we are detaching the second disk from a replacing vdev, then 4044bf82a41bSeschrock * check to see if we changed the original vdev's path to have "/old" 4045bf82a41bSeschrock * at the end in spa_vdev_attach(). If so, undo that change now. 4046bf82a41bSeschrock */ 4047*cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4048*cb04b873SMark J Musante vd->vdev_path != NULL) { 4049*cb04b873SMark J Musante size_t len = strlen(vd->vdev_path); 4050*cb04b873SMark J Musante 4051*cb04b873SMark J Musante for (int c = 0; c < pvd->vdev_children; c++) { 4052*cb04b873SMark J Musante cvd = pvd->vdev_child[c]; 4053*cb04b873SMark J Musante 4054*cb04b873SMark J Musante if (cvd == vd || cvd->vdev_path == NULL) 4055*cb04b873SMark J Musante continue; 4056*cb04b873SMark J Musante 4057*cb04b873SMark J Musante if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4058*cb04b873SMark J Musante strcmp(cvd->vdev_path + len, "/old") == 0) { 4059*cb04b873SMark J Musante spa_strfree(cvd->vdev_path); 4060*cb04b873SMark J Musante cvd->vdev_path = spa_strdup(vd->vdev_path); 4061*cb04b873SMark J Musante break; 4062*cb04b873SMark J Musante } 4063bf82a41bSeschrock } 4064bf82a41bSeschrock } 4065bf82a41bSeschrock 406699653d4eSeschrock /* 406799653d4eSeschrock * If we are detaching the original disk from a spare, then it implies 406899653d4eSeschrock * that the spare should become a real disk, and be removed from the 406999653d4eSeschrock * active spare list for the pool. 407099653d4eSeschrock */ 407199653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 4072*cb04b873SMark J Musante vd->vdev_id == 0 && 4073*cb04b873SMark J Musante pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 407499653d4eSeschrock unspare = B_TRUE; 407599653d4eSeschrock 4076fa9e4066Sahrens /* 4077fa9e4066Sahrens * Erase the disk labels so the disk can be used for other things. 4078fa9e4066Sahrens * This must be done after all other error cases are handled, 4079fa9e4066Sahrens * but before we disembowel vd (so we can still do I/O to it). 4080fa9e4066Sahrens * But if we can't do it, don't treat the error as fatal -- 4081fa9e4066Sahrens * it may be that the unwritability of the disk is the reason 4082fa9e4066Sahrens * it's being detached! 4083fa9e4066Sahrens */ 408439c23413Seschrock error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4085fa9e4066Sahrens 4086fa9e4066Sahrens /* 4087fa9e4066Sahrens * Remove vd from its parent and compact the parent's children. 4088fa9e4066Sahrens */ 4089fa9e4066Sahrens vdev_remove_child(pvd, vd); 4090fa9e4066Sahrens vdev_compact_children(pvd); 4091fa9e4066Sahrens 4092fa9e4066Sahrens /* 4093fa9e4066Sahrens * Remember one of the remaining children so we can get tvd below. 4094fa9e4066Sahrens */ 4095*cb04b873SMark J Musante cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4096fa9e4066Sahrens 409799653d4eSeschrock /* 409899653d4eSeschrock * If we need to remove the remaining child from the list of hot spares, 40998ad4d6ddSJeff Bonwick * do it now, marking the vdev as no longer a spare in the process. 41008ad4d6ddSJeff Bonwick * We must do this before vdev_remove_parent(), because that can 41018ad4d6ddSJeff Bonwick * change the GUID if it creates a new toplevel GUID. For a similar 41028ad4d6ddSJeff Bonwick * reason, we must remove the spare now, in the same txg as the detach; 41038ad4d6ddSJeff Bonwick * otherwise someone could attach a new sibling, change the GUID, and 41048ad4d6ddSJeff Bonwick * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 410599653d4eSeschrock */ 410699653d4eSeschrock if (unspare) { 410799653d4eSeschrock ASSERT(cvd->vdev_isspare); 410839c23413Seschrock spa_spare_remove(cvd); 410999653d4eSeschrock unspare_guid = cvd->vdev_guid; 41108ad4d6ddSJeff Bonwick (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4111*cb04b873SMark J Musante cvd->vdev_unspare = B_TRUE; 411299653d4eSeschrock } 411399653d4eSeschrock 4114fa9e4066Sahrens /* 4115fa9e4066Sahrens * If the parent mirror/replacing vdev only has one child, 4116fa9e4066Sahrens * the parent is no longer needed. Remove it from the tree. 4117fa9e4066Sahrens */ 4118*cb04b873SMark J Musante if (pvd->vdev_children == 1) { 4119*cb04b873SMark J Musante if (pvd->vdev_ops == &vdev_spare_ops) 4120*cb04b873SMark J Musante cvd->vdev_unspare = B_FALSE; 4121fa9e4066Sahrens vdev_remove_parent(cvd); 4122*cb04b873SMark J Musante cvd->vdev_resilvering = B_FALSE; 4123*cb04b873SMark J Musante } 4124*cb04b873SMark J Musante 4125fa9e4066Sahrens 4126fa9e4066Sahrens /* 4127fa9e4066Sahrens * We don't set tvd until now because the parent we just removed 4128fa9e4066Sahrens * may have been the previous top-level vdev. 4129fa9e4066Sahrens */ 4130fa9e4066Sahrens tvd = cvd->vdev_top; 4131fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 4132fa9e4066Sahrens 4133fa9e4066Sahrens /* 413439c23413Seschrock * Reevaluate the parent vdev state. 4135fa9e4066Sahrens */ 41363d7072f8Seschrock vdev_propagate_state(cvd); 4137fa9e4066Sahrens 4138fa9e4066Sahrens /* 4139573ca77eSGeorge Wilson * If the 'autoexpand' property is set on the pool then automatically 4140573ca77eSGeorge Wilson * try to expand the size of the pool. For example if the device we 4141573ca77eSGeorge Wilson * just detached was smaller than the others, it may be possible to 4142573ca77eSGeorge Wilson * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4143573ca77eSGeorge Wilson * first so that we can obtain the updated sizes of the leaf vdevs. 4144fa9e4066Sahrens */ 4145573ca77eSGeorge Wilson if (spa->spa_autoexpand) { 4146573ca77eSGeorge Wilson vdev_reopen(tvd); 4147573ca77eSGeorge Wilson vdev_expand(tvd, txg); 4148573ca77eSGeorge Wilson } 4149fa9e4066Sahrens 4150fa9e4066Sahrens vdev_config_dirty(tvd); 4151fa9e4066Sahrens 4152fa9e4066Sahrens /* 415339c23413Seschrock * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 415439c23413Seschrock * vd->vdev_detached is set and free vd's DTL object in syncing context. 415539c23413Seschrock * But first make sure we're not on any *other* txg's DTL list, to 415639c23413Seschrock * prevent vd from being accessed after it's freed. 4157fa9e4066Sahrens */ 41581195e687SMark J Musante vdpath = spa_strdup(vd->vdev_path); 41598ad4d6ddSJeff Bonwick for (int t = 0; t < TXG_SIZE; t++) 4160fa9e4066Sahrens (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4161ecc2d604Sbonwick vd->vdev_detached = B_TRUE; 4162ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, vd, txg); 4163fa9e4066Sahrens 41643d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 41653d7072f8Seschrock 4166*cb04b873SMark J Musante /* hang on to the spa before we release the lock */ 4167*cb04b873SMark J Musante spa_open_ref(spa, FTAG); 4168*cb04b873SMark J Musante 416999653d4eSeschrock error = spa_vdev_exit(spa, vd, txg, 0); 417099653d4eSeschrock 41713f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL, 41721195e687SMark J Musante "vdev=%s", vdpath); 41731195e687SMark J Musante spa_strfree(vdpath); 41741195e687SMark J Musante 417599653d4eSeschrock /* 417639c23413Seschrock * If this was the removal of the original device in a hot spare vdev, 417739c23413Seschrock * then we want to go through and remove the device from the hot spare 417839c23413Seschrock * list of every other pool. 417999653d4eSeschrock */ 418099653d4eSeschrock if (unspare) { 4181*cb04b873SMark J Musante spa_t *altspa = NULL; 4182*cb04b873SMark J Musante 418399653d4eSeschrock mutex_enter(&spa_namespace_lock); 4184*cb04b873SMark J Musante while ((altspa = spa_next(altspa)) != NULL) { 4185*cb04b873SMark J Musante if (altspa->spa_state != POOL_STATE_ACTIVE || 4186*cb04b873SMark J Musante altspa == spa) 418799653d4eSeschrock continue; 4188*cb04b873SMark J Musante 4189*cb04b873SMark J Musante spa_open_ref(altspa, FTAG); 41909af0a4dfSJeff Bonwick mutex_exit(&spa_namespace_lock); 4191*cb04b873SMark J Musante (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 41929af0a4dfSJeff Bonwick mutex_enter(&spa_namespace_lock); 4193*cb04b873SMark J Musante spa_close(altspa, FTAG); 419499653d4eSeschrock } 419599653d4eSeschrock mutex_exit(&spa_namespace_lock); 4196*cb04b873SMark J Musante 4197*cb04b873SMark J Musante /* search the rest of the vdevs for spares to remove */ 4198*cb04b873SMark J Musante spa_vdev_resilver_done(spa); 419999653d4eSeschrock } 420099653d4eSeschrock 4201*cb04b873SMark J Musante /* all done with the spa; OK to release */ 4202*cb04b873SMark J Musante mutex_enter(&spa_namespace_lock); 4203*cb04b873SMark J Musante spa_close(spa, FTAG); 4204*cb04b873SMark J Musante mutex_exit(&spa_namespace_lock); 4205*cb04b873SMark J Musante 420699653d4eSeschrock return (error); 420799653d4eSeschrock } 420899653d4eSeschrock 42091195e687SMark J Musante /* 42101195e687SMark J Musante * Split a set of devices from their mirrors, and create a new pool from them. 42111195e687SMark J Musante */ 42121195e687SMark J Musante int 42131195e687SMark J Musante spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 42141195e687SMark J Musante nvlist_t *props, boolean_t exp) 42151195e687SMark J Musante { 42161195e687SMark J Musante int error = 0; 42171195e687SMark J Musante uint64_t txg, *glist; 42181195e687SMark J Musante spa_t *newspa; 42191195e687SMark J Musante uint_t c, children, lastlog; 42201195e687SMark J Musante nvlist_t **child, *nvl, *tmp; 42211195e687SMark J Musante dmu_tx_t *tx; 42221195e687SMark J Musante char *altroot = NULL; 42231195e687SMark J Musante vdev_t *rvd, **vml = NULL; /* vdev modify list */ 42241195e687SMark J Musante boolean_t activate_slog; 42251195e687SMark J Musante 42261195e687SMark J Musante if (!spa_writeable(spa)) 42271195e687SMark J Musante return (EROFS); 42281195e687SMark J Musante 42291195e687SMark J Musante txg = spa_vdev_enter(spa); 42301195e687SMark J Musante 42311195e687SMark J Musante /* clear the log and flush everything up to now */ 42321195e687SMark J Musante activate_slog = spa_passivate_log(spa); 42331195e687SMark J Musante (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 42341195e687SMark J Musante error = spa_offline_log(spa); 42351195e687SMark J Musante txg = spa_vdev_config_enter(spa); 42361195e687SMark J Musante 42371195e687SMark J Musante if (activate_slog) 42381195e687SMark J Musante spa_activate_log(spa); 42391195e687SMark J Musante 42401195e687SMark J Musante if (error != 0) 42411195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 42421195e687SMark J Musante 42431195e687SMark J Musante /* check new spa name before going any further */ 42441195e687SMark J Musante if (spa_lookup(newname) != NULL) 42451195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 42461195e687SMark J Musante 42471195e687SMark J Musante /* 42481195e687SMark J Musante * scan through all the children to ensure they're all mirrors 42491195e687SMark J Musante */ 42501195e687SMark J Musante if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 42511195e687SMark J Musante nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 42521195e687SMark J Musante &children) != 0) 42531195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 42541195e687SMark J Musante 42551195e687SMark J Musante /* first, check to ensure we've got the right child count */ 42561195e687SMark J Musante rvd = spa->spa_root_vdev; 42571195e687SMark J Musante lastlog = 0; 42581195e687SMark J Musante for (c = 0; c < rvd->vdev_children; c++) { 42591195e687SMark J Musante vdev_t *vd = rvd->vdev_child[c]; 42601195e687SMark J Musante 42611195e687SMark J Musante /* don't count the holes & logs as children */ 42621195e687SMark J Musante if (vd->vdev_islog || vd->vdev_ishole) { 42631195e687SMark J Musante if (lastlog == 0) 42641195e687SMark J Musante lastlog = c; 42651195e687SMark J Musante continue; 42661195e687SMark J Musante } 42671195e687SMark J Musante 42681195e687SMark J Musante lastlog = 0; 42691195e687SMark J Musante } 42701195e687SMark J Musante if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 42711195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 42721195e687SMark J Musante 42731195e687SMark J Musante /* next, ensure no spare or cache devices are part of the split */ 42741195e687SMark J Musante if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 42751195e687SMark J Musante nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 42761195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 42771195e687SMark J Musante 42781195e687SMark J Musante vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 42791195e687SMark J Musante glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 42801195e687SMark J Musante 42811195e687SMark J Musante /* then, loop over each vdev and validate it */ 42821195e687SMark J Musante for (c = 0; c < children; c++) { 42831195e687SMark J Musante uint64_t is_hole = 0; 42841195e687SMark J Musante 42851195e687SMark J Musante (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 42861195e687SMark J Musante &is_hole); 42871195e687SMark J Musante 42881195e687SMark J Musante if (is_hole != 0) { 42891195e687SMark J Musante if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 42901195e687SMark J Musante spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 42911195e687SMark J Musante continue; 42921195e687SMark J Musante } else { 42931195e687SMark J Musante error = EINVAL; 42941195e687SMark J Musante break; 42951195e687SMark J Musante } 42961195e687SMark J Musante } 42971195e687SMark J Musante 42981195e687SMark J Musante /* which disk is going to be split? */ 42991195e687SMark J Musante if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 43001195e687SMark J Musante &glist[c]) != 0) { 43011195e687SMark J Musante error = EINVAL; 43021195e687SMark J Musante break; 43031195e687SMark J Musante } 43041195e687SMark J Musante 43051195e687SMark J Musante /* look it up in the spa */ 43061195e687SMark J Musante vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 43071195e687SMark J Musante if (vml[c] == NULL) { 43081195e687SMark J Musante error = ENODEV; 43091195e687SMark J Musante break; 43101195e687SMark J Musante } 43111195e687SMark J Musante 43121195e687SMark J Musante /* make sure there's nothing stopping the split */ 43131195e687SMark J Musante if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 43141195e687SMark J Musante vml[c]->vdev_islog || 43151195e687SMark J Musante vml[c]->vdev_ishole || 43161195e687SMark J Musante vml[c]->vdev_isspare || 43171195e687SMark J Musante vml[c]->vdev_isl2cache || 43181195e687SMark J Musante !vdev_writeable(vml[c]) || 4319d41c4376SMark J Musante vml[c]->vdev_children != 0 || 43201195e687SMark J Musante vml[c]->vdev_state != VDEV_STATE_HEALTHY || 43211195e687SMark J Musante c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 43221195e687SMark J Musante error = EINVAL; 43231195e687SMark J Musante break; 43241195e687SMark J Musante } 43251195e687SMark J Musante 43261195e687SMark J Musante if (vdev_dtl_required(vml[c])) { 43271195e687SMark J Musante error = EBUSY; 43281195e687SMark J Musante break; 43291195e687SMark J Musante } 43301195e687SMark J Musante 43311195e687SMark J Musante /* we need certain info from the top level */ 43321195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 43331195e687SMark J Musante vml[c]->vdev_top->vdev_ms_array) == 0); 43341195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 43351195e687SMark J Musante vml[c]->vdev_top->vdev_ms_shift) == 0); 43361195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 43371195e687SMark J Musante vml[c]->vdev_top->vdev_asize) == 0); 43381195e687SMark J Musante VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 43391195e687SMark J Musante vml[c]->vdev_top->vdev_ashift) == 0); 43401195e687SMark J Musante } 43411195e687SMark J Musante 43421195e687SMark J Musante if (error != 0) { 43431195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 43441195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 43451195e687SMark J Musante return (spa_vdev_exit(spa, NULL, txg, error)); 43461195e687SMark J Musante } 43471195e687SMark J Musante 43481195e687SMark J Musante /* stop writers from using the disks */ 43491195e687SMark J Musante for (c = 0; c < children; c++) { 43501195e687SMark J Musante if (vml[c] != NULL) 43511195e687SMark J Musante vml[c]->vdev_offline = B_TRUE; 43521195e687SMark J Musante } 43531195e687SMark J Musante vdev_reopen(spa->spa_root_vdev); 43541195e687SMark J Musante 43551195e687SMark J Musante /* 43561195e687SMark J Musante * Temporarily record the splitting vdevs in the spa config. This 43571195e687SMark J Musante * will disappear once the config is regenerated. 43581195e687SMark J Musante */ 43591195e687SMark J Musante VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 43601195e687SMark J Musante VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 43611195e687SMark J Musante glist, children) == 0); 43621195e687SMark J Musante kmem_free(glist, children * sizeof (uint64_t)); 43631195e687SMark J Musante 436498295d61SMark J Musante mutex_enter(&spa->spa_props_lock); 43651195e687SMark J Musante VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 43661195e687SMark J Musante nvl) == 0); 436798295d61SMark J Musante mutex_exit(&spa->spa_props_lock); 43681195e687SMark J Musante spa->spa_config_splitting = nvl; 43691195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 43701195e687SMark J Musante 43711195e687SMark J Musante /* configure and create the new pool */ 43721195e687SMark J Musante VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 43731195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 43741195e687SMark J Musante exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 43751195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 43761195e687SMark J Musante spa_version(spa)) == 0); 43771195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 43781195e687SMark J Musante spa->spa_config_txg) == 0); 43791195e687SMark J Musante VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 43801195e687SMark J Musante spa_generate_guid(NULL)) == 0); 43811195e687SMark J Musante (void) nvlist_lookup_string(props, 43821195e687SMark J Musante zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 43831195e687SMark J Musante 4384d41c4376SMark J Musante /* add the new pool to the namespace */ 43851195e687SMark J Musante newspa = spa_add(newname, config, altroot); 43861195e687SMark J Musante newspa->spa_config_txg = spa->spa_config_txg; 43871195e687SMark J Musante spa_set_log_state(newspa, SPA_LOG_CLEAR); 43881195e687SMark J Musante 43891195e687SMark J Musante /* release the spa config lock, retaining the namespace lock */ 43901195e687SMark J Musante spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 43911195e687SMark J Musante 43921195e687SMark J Musante if (zio_injection_enabled) 43931195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 1); 43941195e687SMark J Musante 43951195e687SMark J Musante spa_activate(newspa, spa_mode_global); 43961195e687SMark J Musante spa_async_suspend(newspa); 43971195e687SMark J Musante 43981195e687SMark J Musante /* create the new pool from the disks of the original pool */ 43991195e687SMark J Musante error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 44001195e687SMark J Musante if (error) 44011195e687SMark J Musante goto out; 44021195e687SMark J Musante 44031195e687SMark J Musante /* if that worked, generate a real config for the new pool */ 44041195e687SMark J Musante if (newspa->spa_root_vdev != NULL) { 44051195e687SMark J Musante VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 44061195e687SMark J Musante NV_UNIQUE_NAME, KM_SLEEP) == 0); 44071195e687SMark J Musante VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 44081195e687SMark J Musante ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 44091195e687SMark J Musante spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 44101195e687SMark J Musante B_TRUE)); 44111195e687SMark J Musante } 44121195e687SMark J Musante 44131195e687SMark J Musante /* set the props */ 44141195e687SMark J Musante if (props != NULL) { 44151195e687SMark J Musante spa_configfile_set(newspa, props, B_FALSE); 44161195e687SMark J Musante error = spa_prop_set(newspa, props); 44171195e687SMark J Musante if (error) 44181195e687SMark J Musante goto out; 44191195e687SMark J Musante } 44201195e687SMark J Musante 44211195e687SMark J Musante /* flush everything */ 44221195e687SMark J Musante txg = spa_vdev_config_enter(newspa); 44231195e687SMark J Musante vdev_config_dirty(newspa->spa_root_vdev); 44241195e687SMark J Musante (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 44251195e687SMark J Musante 44261195e687SMark J Musante if (zio_injection_enabled) 44271195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 2); 44281195e687SMark J Musante 44291195e687SMark J Musante spa_async_resume(newspa); 44301195e687SMark J Musante 44311195e687SMark J Musante /* finally, update the original pool's config */ 44321195e687SMark J Musante txg = spa_vdev_config_enter(spa); 44331195e687SMark J Musante tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 44341195e687SMark J Musante error = dmu_tx_assign(tx, TXG_WAIT); 44351195e687SMark J Musante if (error != 0) 44361195e687SMark J Musante dmu_tx_abort(tx); 44371195e687SMark J Musante for (c = 0; c < children; c++) { 44381195e687SMark J Musante if (vml[c] != NULL) { 44391195e687SMark J Musante vdev_split(vml[c]); 44401195e687SMark J Musante if (error == 0) 44413f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_VDEV_DETACH, 44423f9d6ad7SLin Ling spa, tx, "vdev=%s", 44431195e687SMark J Musante vml[c]->vdev_path); 44441195e687SMark J Musante vdev_free(vml[c]); 44451195e687SMark J Musante } 44461195e687SMark J Musante } 44471195e687SMark J Musante vdev_config_dirty(spa->spa_root_vdev); 44481195e687SMark J Musante spa->spa_config_splitting = NULL; 44491195e687SMark J Musante nvlist_free(nvl); 44501195e687SMark J Musante if (error == 0) 44511195e687SMark J Musante dmu_tx_commit(tx); 44521195e687SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, 0); 44531195e687SMark J Musante 44541195e687SMark J Musante if (zio_injection_enabled) 44551195e687SMark J Musante zio_handle_panic_injection(spa, FTAG, 3); 44561195e687SMark J Musante 44571195e687SMark J Musante /* split is complete; log a history record */ 44583f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL, 44591195e687SMark J Musante "split new pool %s from pool %s", newname, spa_name(spa)); 44601195e687SMark J Musante 44611195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 44621195e687SMark J Musante 44631195e687SMark J Musante /* if we're not going to mount the filesystems in userland, export */ 44641195e687SMark J Musante if (exp) 44651195e687SMark J Musante error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 44661195e687SMark J Musante B_FALSE, B_FALSE); 44671195e687SMark J Musante 44681195e687SMark J Musante return (error); 44691195e687SMark J Musante 44701195e687SMark J Musante out: 44711195e687SMark J Musante spa_unload(newspa); 44721195e687SMark J Musante spa_deactivate(newspa); 44731195e687SMark J Musante spa_remove(newspa); 44741195e687SMark J Musante 44751195e687SMark J Musante txg = spa_vdev_config_enter(spa); 447698295d61SMark J Musante 447798295d61SMark J Musante /* re-online all offlined disks */ 447898295d61SMark J Musante for (c = 0; c < children; c++) { 447998295d61SMark J Musante if (vml[c] != NULL) 448098295d61SMark J Musante vml[c]->vdev_offline = B_FALSE; 448198295d61SMark J Musante } 448298295d61SMark J Musante vdev_reopen(spa->spa_root_vdev); 448398295d61SMark J Musante 44841195e687SMark J Musante nvlist_free(spa->spa_config_splitting); 44851195e687SMark J Musante spa->spa_config_splitting = NULL; 4486d41c4376SMark J Musante (void) spa_vdev_exit(spa, NULL, txg, error); 44871195e687SMark J Musante 44881195e687SMark J Musante kmem_free(vml, children * sizeof (vdev_t *)); 44891195e687SMark J Musante return (error); 44901195e687SMark J Musante } 44911195e687SMark J Musante 4492e14bb325SJeff Bonwick static nvlist_t * 4493e14bb325SJeff Bonwick spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 449499653d4eSeschrock { 4495e14bb325SJeff Bonwick for (int i = 0; i < count; i++) { 4496e14bb325SJeff Bonwick uint64_t guid; 449799653d4eSeschrock 4498e14bb325SJeff Bonwick VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 4499e14bb325SJeff Bonwick &guid) == 0); 450099653d4eSeschrock 4501e14bb325SJeff Bonwick if (guid == target_guid) 4502e14bb325SJeff Bonwick return (nvpp[i]); 450399653d4eSeschrock } 450499653d4eSeschrock 4505e14bb325SJeff Bonwick return (NULL); 4506fa94a07fSbrendan } 4507fa94a07fSbrendan 4508e14bb325SJeff Bonwick static void 4509e14bb325SJeff Bonwick spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 4510e14bb325SJeff Bonwick nvlist_t *dev_to_remove) 4511fa94a07fSbrendan { 4512e14bb325SJeff Bonwick nvlist_t **newdev = NULL; 4513fa94a07fSbrendan 4514e14bb325SJeff Bonwick if (count > 1) 4515e14bb325SJeff Bonwick newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 4516fa94a07fSbrendan 4517e14bb325SJeff Bonwick for (int i = 0, j = 0; i < count; i++) { 4518e14bb325SJeff Bonwick if (dev[i] == dev_to_remove) 4519e14bb325SJeff Bonwick continue; 4520e14bb325SJeff Bonwick VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 4521fa94a07fSbrendan } 4522fa94a07fSbrendan 4523e14bb325SJeff Bonwick VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 4524e14bb325SJeff Bonwick VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 4525fa94a07fSbrendan 4526e14bb325SJeff Bonwick for (int i = 0; i < count - 1; i++) 4527e14bb325SJeff Bonwick nvlist_free(newdev[i]); 4528fa94a07fSbrendan 4529e14bb325SJeff Bonwick if (count > 1) 4530e14bb325SJeff Bonwick kmem_free(newdev, (count - 1) * sizeof (void *)); 4531fa94a07fSbrendan } 4532fa94a07fSbrendan 453388ecc943SGeorge Wilson /* 453488ecc943SGeorge Wilson * Evacuate the device. 453588ecc943SGeorge Wilson */ 45363f9d6ad7SLin Ling static int 453788ecc943SGeorge Wilson spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 453888ecc943SGeorge Wilson { 453988ecc943SGeorge Wilson uint64_t txg; 45403f9d6ad7SLin Ling int error = 0; 454188ecc943SGeorge Wilson 454288ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 454388ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 4544b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 454588ecc943SGeorge Wilson 454688ecc943SGeorge Wilson /* 454788ecc943SGeorge Wilson * Evacuate the device. We don't hold the config lock as writer 454888ecc943SGeorge Wilson * since we need to do I/O but we do keep the 454988ecc943SGeorge Wilson * spa_namespace_lock held. Once this completes the device 455088ecc943SGeorge Wilson * should no longer have any blocks allocated on it. 455188ecc943SGeorge Wilson */ 455288ecc943SGeorge Wilson if (vd->vdev_islog) { 45533f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 45543f9d6ad7SLin Ling error = spa_offline_log(spa); 4555a1521560SJeff Bonwick } else { 45563f9d6ad7SLin Ling error = ENOTSUP; 455788ecc943SGeorge Wilson } 455888ecc943SGeorge Wilson 4559a1521560SJeff Bonwick if (error) 4560a1521560SJeff Bonwick return (error); 4561a1521560SJeff Bonwick 456288ecc943SGeorge Wilson /* 4563a1521560SJeff Bonwick * The evacuation succeeded. Remove any remaining MOS metadata 4564a1521560SJeff Bonwick * associated with this vdev, and wait for these changes to sync. 456588ecc943SGeorge Wilson */ 45663f9d6ad7SLin Ling ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 456788ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 456888ecc943SGeorge Wilson vd->vdev_removing = B_TRUE; 456988ecc943SGeorge Wilson vdev_dirty(vd, 0, NULL, txg); 457088ecc943SGeorge Wilson vdev_config_dirty(vd); 457188ecc943SGeorge Wilson spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 457288ecc943SGeorge Wilson 457388ecc943SGeorge Wilson return (0); 457488ecc943SGeorge Wilson } 457588ecc943SGeorge Wilson 457688ecc943SGeorge Wilson /* 457788ecc943SGeorge Wilson * Complete the removal by cleaning up the namespace. 457888ecc943SGeorge Wilson */ 45793f9d6ad7SLin Ling static void 4580a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 458188ecc943SGeorge Wilson { 458288ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 458388ecc943SGeorge Wilson uint64_t id = vd->vdev_id; 458488ecc943SGeorge Wilson boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 458588ecc943SGeorge Wilson 458688ecc943SGeorge Wilson ASSERT(MUTEX_HELD(&spa_namespace_lock)); 458788ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4588b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 458988ecc943SGeorge Wilson 45903f9d6ad7SLin Ling /* 45913f9d6ad7SLin Ling * Only remove any devices which are empty. 45923f9d6ad7SLin Ling */ 45933f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc != 0) 45943f9d6ad7SLin Ling return; 45953f9d6ad7SLin Ling 459688ecc943SGeorge Wilson (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4597b24ab676SJeff Bonwick 4598b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_state_dirty_node)) 4599b24ab676SJeff Bonwick vdev_state_clean(vd); 4600b24ab676SJeff Bonwick if (list_link_active(&vd->vdev_config_dirty_node)) 4601b24ab676SJeff Bonwick vdev_config_clean(vd); 4602b24ab676SJeff Bonwick 460388ecc943SGeorge Wilson vdev_free(vd); 460488ecc943SGeorge Wilson 460588ecc943SGeorge Wilson if (last_vdev) { 460688ecc943SGeorge Wilson vdev_compact_children(rvd); 460788ecc943SGeorge Wilson } else { 460888ecc943SGeorge Wilson vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 460988ecc943SGeorge Wilson vdev_add_child(rvd, vd); 461088ecc943SGeorge Wilson } 4611fcbfa62bSLin Ling vdev_config_dirty(rvd); 4612fcbfa62bSLin Ling 4613fcbfa62bSLin Ling /* 4614fcbfa62bSLin Ling * Reassess the health of our root vdev. 4615fcbfa62bSLin Ling */ 4616fcbfa62bSLin Ling vdev_reopen(rvd); 461788ecc943SGeorge Wilson } 461888ecc943SGeorge Wilson 46193f9d6ad7SLin Ling /* 46203f9d6ad7SLin Ling * Remove a device from the pool - 46213f9d6ad7SLin Ling * 46223f9d6ad7SLin Ling * Removing a device from the vdev namespace requires several steps 46233f9d6ad7SLin Ling * and can take a significant amount of time. As a result we use 46243f9d6ad7SLin Ling * the spa_vdev_config_[enter/exit] functions which allow us to 46253f9d6ad7SLin Ling * grab and release the spa_config_lock while still holding the namespace 46263f9d6ad7SLin Ling * lock. During each step the configuration is synced out. 46273f9d6ad7SLin Ling */ 46283f9d6ad7SLin Ling 4629fa94a07fSbrendan /* 4630fa94a07fSbrendan * Remove a device from the pool. Currently, this supports removing only hot 463188ecc943SGeorge Wilson * spares, slogs, and level 2 ARC devices. 4632fa94a07fSbrendan */ 4633fa94a07fSbrendan int 4634fa94a07fSbrendan spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 4635fa94a07fSbrendan { 4636fa94a07fSbrendan vdev_t *vd; 4637a1521560SJeff Bonwick metaslab_group_t *mg; 4638e14bb325SJeff Bonwick nvlist_t **spares, **l2cache, *nv; 46398ad4d6ddSJeff Bonwick uint64_t txg = 0; 464088ecc943SGeorge Wilson uint_t nspares, nl2cache; 4641fa94a07fSbrendan int error = 0; 46428ad4d6ddSJeff Bonwick boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 4643fa94a07fSbrendan 46448ad4d6ddSJeff Bonwick if (!locked) 46458ad4d6ddSJeff Bonwick txg = spa_vdev_enter(spa); 4646fa94a07fSbrendan 4647c5904d13Seschrock vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4648fa94a07fSbrendan 4649fa94a07fSbrendan if (spa->spa_spares.sav_vdevs != NULL && 4650fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 4651e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 4652e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 4653e14bb325SJeff Bonwick /* 4654e14bb325SJeff Bonwick * Only remove the hot spare if it's not currently in use 4655e14bb325SJeff Bonwick * in this pool. 4656e14bb325SJeff Bonwick */ 4657e14bb325SJeff Bonwick if (vd == NULL || unspare) { 4658e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_spares.sav_config, 4659e14bb325SJeff Bonwick ZPOOL_CONFIG_SPARES, spares, nspares, nv); 4660e14bb325SJeff Bonwick spa_load_spares(spa); 4661e14bb325SJeff Bonwick spa->spa_spares.sav_sync = B_TRUE; 4662e14bb325SJeff Bonwick } else { 4663e14bb325SJeff Bonwick error = EBUSY; 4664e14bb325SJeff Bonwick } 4665e14bb325SJeff Bonwick } else if (spa->spa_l2cache.sav_vdevs != NULL && 4666fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 4667e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 4668e14bb325SJeff Bonwick (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 4669e14bb325SJeff Bonwick /* 4670e14bb325SJeff Bonwick * Cache devices can always be removed. 4671e14bb325SJeff Bonwick */ 4672e14bb325SJeff Bonwick spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 4673e14bb325SJeff Bonwick ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 4674fa94a07fSbrendan spa_load_l2cache(spa); 4675fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 467688ecc943SGeorge Wilson } else if (vd != NULL && vd->vdev_islog) { 467788ecc943SGeorge Wilson ASSERT(!locked); 4678b24ab676SJeff Bonwick ASSERT(vd == vd->vdev_top); 467988ecc943SGeorge Wilson 468088ecc943SGeorge Wilson /* 468188ecc943SGeorge Wilson * XXX - Once we have bp-rewrite this should 468288ecc943SGeorge Wilson * become the common case. 468388ecc943SGeorge Wilson */ 468488ecc943SGeorge Wilson 4685a1521560SJeff Bonwick mg = vd->vdev_mg; 4686a1521560SJeff Bonwick 468788ecc943SGeorge Wilson /* 4688a1521560SJeff Bonwick * Stop allocating from this vdev. 468988ecc943SGeorge Wilson */ 4690a1521560SJeff Bonwick metaslab_group_passivate(mg); 469188ecc943SGeorge Wilson 4692b24ab676SJeff Bonwick /* 4693b24ab676SJeff Bonwick * Wait for the youngest allocations and frees to sync, 4694b24ab676SJeff Bonwick * and then wait for the deferral of those frees to finish. 4695b24ab676SJeff Bonwick */ 4696b24ab676SJeff Bonwick spa_vdev_config_exit(spa, NULL, 4697b24ab676SJeff Bonwick txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 4698b24ab676SJeff Bonwick 4699a1521560SJeff Bonwick /* 4700a1521560SJeff Bonwick * Attempt to evacuate the vdev. 4701a1521560SJeff Bonwick */ 4702a1521560SJeff Bonwick error = spa_vdev_remove_evacuate(spa, vd); 4703a1521560SJeff Bonwick 470488ecc943SGeorge Wilson txg = spa_vdev_config_enter(spa); 470588ecc943SGeorge Wilson 4706a1521560SJeff Bonwick /* 4707a1521560SJeff Bonwick * If we couldn't evacuate the vdev, unwind. 4708a1521560SJeff Bonwick */ 4709a1521560SJeff Bonwick if (error) { 4710a1521560SJeff Bonwick metaslab_group_activate(mg); 4711a1521560SJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, error)); 4712a1521560SJeff Bonwick } 4713a1521560SJeff Bonwick 4714a1521560SJeff Bonwick /* 4715a1521560SJeff Bonwick * Clean up the vdev namespace. 4716a1521560SJeff Bonwick */ 4717a1521560SJeff Bonwick spa_vdev_remove_from_namespace(spa, vd); 471888ecc943SGeorge Wilson 4719e14bb325SJeff Bonwick } else if (vd != NULL) { 4720e14bb325SJeff Bonwick /* 4721e14bb325SJeff Bonwick * Normal vdevs cannot be removed (yet). 4722e14bb325SJeff Bonwick */ 4723e14bb325SJeff Bonwick error = ENOTSUP; 4724e14bb325SJeff Bonwick } else { 4725e14bb325SJeff Bonwick /* 4726e14bb325SJeff Bonwick * There is no vdev of any kind with the specified guid. 4727e14bb325SJeff Bonwick */ 4728e14bb325SJeff Bonwick error = ENOENT; 4729fa94a07fSbrendan } 473099653d4eSeschrock 47318ad4d6ddSJeff Bonwick if (!locked) 47328ad4d6ddSJeff Bonwick return (spa_vdev_exit(spa, NULL, txg, error)); 47338ad4d6ddSJeff Bonwick 47348ad4d6ddSJeff Bonwick return (error); 4735fa9e4066Sahrens } 4736fa9e4066Sahrens 4737fa9e4066Sahrens /* 47383d7072f8Seschrock * Find any device that's done replacing, or a vdev marked 'unspare' that's 47393d7072f8Seschrock * current spared, so we can detach it. 4740fa9e4066Sahrens */ 4741ea8dc4b6Seschrock static vdev_t * 47423d7072f8Seschrock spa_vdev_resilver_done_hunt(vdev_t *vd) 4743fa9e4066Sahrens { 4744ea8dc4b6Seschrock vdev_t *newvd, *oldvd; 4745fa9e4066Sahrens 4746573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 47473d7072f8Seschrock oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 4748ea8dc4b6Seschrock if (oldvd != NULL) 4749ea8dc4b6Seschrock return (oldvd); 4750ea8dc4b6Seschrock } 4751fa9e4066Sahrens 47523d7072f8Seschrock /* 4753*cb04b873SMark J Musante * Check for a completed replacement. We always consider the first 4754*cb04b873SMark J Musante * vdev in the list to be the oldest vdev, and the last one to be 4755*cb04b873SMark J Musante * the newest (see spa_vdev_attach() for how that works). In 4756*cb04b873SMark J Musante * the case where the newest vdev is faulted, we will not automatically 4757*cb04b873SMark J Musante * remove it after a resilver completes. This is OK as it will require 4758*cb04b873SMark J Musante * user intervention to determine which disk the admin wishes to keep. 47593d7072f8Seschrock */ 4760*cb04b873SMark J Musante if (vd->vdev_ops == &vdev_replacing_ops) { 4761*cb04b873SMark J Musante ASSERT(vd->vdev_children > 1); 4762*cb04b873SMark J Musante 4763*cb04b873SMark J Musante newvd = vd->vdev_child[vd->vdev_children - 1]; 4764ea8dc4b6Seschrock oldvd = vd->vdev_child[0]; 4765ea8dc4b6Seschrock 47668ad4d6ddSJeff Bonwick if (vdev_dtl_empty(newvd, DTL_MISSING) && 4767e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 47688ad4d6ddSJeff Bonwick !vdev_dtl_required(oldvd)) 4769ea8dc4b6Seschrock return (oldvd); 4770fa9e4066Sahrens } 4771ea8dc4b6Seschrock 47723d7072f8Seschrock /* 47733d7072f8Seschrock * Check for a completed resilver with the 'unspare' flag set. 47743d7072f8Seschrock */ 4775*cb04b873SMark J Musante if (vd->vdev_ops == &vdev_spare_ops) { 4776*cb04b873SMark J Musante vdev_t *first = vd->vdev_child[0]; 4777*cb04b873SMark J Musante vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 4778*cb04b873SMark J Musante 4779*cb04b873SMark J Musante if (last->vdev_unspare) { 4780*cb04b873SMark J Musante oldvd = first; 4781*cb04b873SMark J Musante newvd = last; 4782*cb04b873SMark J Musante } else if (first->vdev_unspare) { 4783*cb04b873SMark J Musante oldvd = last; 4784*cb04b873SMark J Musante newvd = first; 4785*cb04b873SMark J Musante } else { 4786*cb04b873SMark J Musante oldvd = NULL; 4787*cb04b873SMark J Musante } 47883d7072f8Seschrock 4789*cb04b873SMark J Musante if (oldvd != NULL && 47908ad4d6ddSJeff Bonwick vdev_dtl_empty(newvd, DTL_MISSING) && 4791e69acc92SVictor Latushkin vdev_dtl_empty(newvd, DTL_OUTAGE) && 4792*cb04b873SMark J Musante !vdev_dtl_required(oldvd)) 47933d7072f8Seschrock return (oldvd); 4794*cb04b873SMark J Musante 4795*cb04b873SMark J Musante /* 4796*cb04b873SMark J Musante * If there are more than two spares attached to a disk, 4797*cb04b873SMark J Musante * and those spares are not required, then we want to 4798*cb04b873SMark J Musante * attempt to free them up now so that they can be used 4799*cb04b873SMark J Musante * by other pools. Once we're back down to a single 4800*cb04b873SMark J Musante * disk+spare, we stop removing them. 4801*cb04b873SMark J Musante */ 4802*cb04b873SMark J Musante if (vd->vdev_children > 2) { 4803*cb04b873SMark J Musante newvd = vd->vdev_child[1]; 4804*cb04b873SMark J Musante 4805*cb04b873SMark J Musante if (newvd->vdev_isspare && last->vdev_isspare && 4806*cb04b873SMark J Musante vdev_dtl_empty(last, DTL_MISSING) && 4807*cb04b873SMark J Musante vdev_dtl_empty(last, DTL_OUTAGE) && 4808*cb04b873SMark J Musante !vdev_dtl_required(newvd)) 4809*cb04b873SMark J Musante return (newvd); 48103d7072f8Seschrock } 48113d7072f8Seschrock } 48123d7072f8Seschrock 4813ea8dc4b6Seschrock return (NULL); 4814fa9e4066Sahrens } 4815fa9e4066Sahrens 4816ea8dc4b6Seschrock static void 48173d7072f8Seschrock spa_vdev_resilver_done(spa_t *spa) 4818fa9e4066Sahrens { 48198ad4d6ddSJeff Bonwick vdev_t *vd, *pvd, *ppvd; 48208ad4d6ddSJeff Bonwick uint64_t guid, sguid, pguid, ppguid; 4821ea8dc4b6Seschrock 48228ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4823ea8dc4b6Seschrock 48243d7072f8Seschrock while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 48258ad4d6ddSJeff Bonwick pvd = vd->vdev_parent; 48268ad4d6ddSJeff Bonwick ppvd = pvd->vdev_parent; 4827ea8dc4b6Seschrock guid = vd->vdev_guid; 48288ad4d6ddSJeff Bonwick pguid = pvd->vdev_guid; 48298ad4d6ddSJeff Bonwick ppguid = ppvd->vdev_guid; 48308ad4d6ddSJeff Bonwick sguid = 0; 483199653d4eSeschrock /* 483299653d4eSeschrock * If we have just finished replacing a hot spared device, then 483399653d4eSeschrock * we need to detach the parent's first child (the original hot 483499653d4eSeschrock * spare) as well. 483599653d4eSeschrock */ 4836*cb04b873SMark J Musante if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 4837*cb04b873SMark J Musante ppvd->vdev_children == 2) { 483899653d4eSeschrock ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 48398ad4d6ddSJeff Bonwick sguid = ppvd->vdev_child[1]->vdev_guid; 484099653d4eSeschrock } 48418ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 48428ad4d6ddSJeff Bonwick if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 4843ea8dc4b6Seschrock return; 48448ad4d6ddSJeff Bonwick if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 484599653d4eSeschrock return; 48468ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4847fa9e4066Sahrens } 4848fa9e4066Sahrens 48498ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 4850fa9e4066Sahrens } 4851fa9e4066Sahrens 4852c67d9675Seschrock /* 4853b3388e4fSEric Taylor * Update the stored path or FRU for this vdev. 4854c67d9675Seschrock */ 4855c67d9675Seschrock int 48566809eb4eSEric Schrock spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 48576809eb4eSEric Schrock boolean_t ispath) 4858c67d9675Seschrock { 4859c5904d13Seschrock vdev_t *vd; 4860208044b8SGeorge Wilson boolean_t sync = B_FALSE; 4861c67d9675Seschrock 4862b3388e4fSEric Taylor spa_vdev_state_enter(spa, SCL_ALL); 4863c67d9675Seschrock 48646809eb4eSEric Schrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 4865b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOENT)); 4866c67d9675Seschrock 48670e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 4868b3388e4fSEric Taylor return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 48690e34b6a7Sbonwick 48706809eb4eSEric Schrock if (ispath) { 4871208044b8SGeorge Wilson if (strcmp(value, vd->vdev_path) != 0) { 4872208044b8SGeorge Wilson spa_strfree(vd->vdev_path); 4873208044b8SGeorge Wilson vd->vdev_path = spa_strdup(value); 4874208044b8SGeorge Wilson sync = B_TRUE; 4875208044b8SGeorge Wilson } 48766809eb4eSEric Schrock } else { 4877208044b8SGeorge Wilson if (vd->vdev_fru == NULL) { 4878208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 4879208044b8SGeorge Wilson sync = B_TRUE; 4880208044b8SGeorge Wilson } else if (strcmp(value, vd->vdev_fru) != 0) { 48816809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 4882208044b8SGeorge Wilson vd->vdev_fru = spa_strdup(value); 4883208044b8SGeorge Wilson sync = B_TRUE; 4884208044b8SGeorge Wilson } 48856809eb4eSEric Schrock } 4886c67d9675Seschrock 4887208044b8SGeorge Wilson return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 4888c67d9675Seschrock } 4889c67d9675Seschrock 48906809eb4eSEric Schrock int 48916809eb4eSEric Schrock spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 48926809eb4eSEric Schrock { 48936809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 48946809eb4eSEric Schrock } 48956809eb4eSEric Schrock 48966809eb4eSEric Schrock int 48976809eb4eSEric Schrock spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 48986809eb4eSEric Schrock { 48996809eb4eSEric Schrock return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 49006809eb4eSEric Schrock } 49016809eb4eSEric Schrock 4902fa9e4066Sahrens /* 4903fa9e4066Sahrens * ========================================================================== 49043f9d6ad7SLin Ling * SPA Scanning 4905fa9e4066Sahrens * ========================================================================== 4906fa9e4066Sahrens */ 4907fa9e4066Sahrens 4908ea8dc4b6Seschrock int 49093f9d6ad7SLin Ling spa_scan_stop(spa_t *spa) 4910fa9e4066Sahrens { 4911e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 49123f9d6ad7SLin Ling if (dsl_scan_resilvering(spa->spa_dsl_pool)) 49133f9d6ad7SLin Ling return (EBUSY); 49143f9d6ad7SLin Ling return (dsl_scan_cancel(spa->spa_dsl_pool)); 49153f9d6ad7SLin Ling } 4916bb8b5132Sek 49173f9d6ad7SLin Ling int 49183f9d6ad7SLin Ling spa_scan(spa_t *spa, pool_scan_func_t func) 49193f9d6ad7SLin Ling { 49203f9d6ad7SLin Ling ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 49213f9d6ad7SLin Ling 49223f9d6ad7SLin Ling if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 4923fa9e4066Sahrens return (ENOTSUP); 4924fa9e4066Sahrens 4925fa9e4066Sahrens /* 4926088f3894Sahrens * If a resilver was requested, but there is no DTL on a 4927088f3894Sahrens * writeable leaf device, we have nothing to do. 4928fa9e4066Sahrens */ 49293f9d6ad7SLin Ling if (func == POOL_SCAN_RESILVER && 4930088f3894Sahrens !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 4931088f3894Sahrens spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 4932ea8dc4b6Seschrock return (0); 4933ea8dc4b6Seschrock } 4934fa9e4066Sahrens 49353f9d6ad7SLin Ling return (dsl_scan(spa->spa_dsl_pool, func)); 4936fa9e4066Sahrens } 4937fa9e4066Sahrens 4938ea8dc4b6Seschrock /* 4939ea8dc4b6Seschrock * ========================================================================== 4940ea8dc4b6Seschrock * SPA async task processing 4941ea8dc4b6Seschrock * ========================================================================== 4942ea8dc4b6Seschrock */ 4943ea8dc4b6Seschrock 4944ea8dc4b6Seschrock static void 49453d7072f8Seschrock spa_async_remove(spa_t *spa, vdev_t *vd) 4946fa9e4066Sahrens { 494749cf58c0SBrendan Gregg - Sun Microsystems if (vd->vdev_remove_wanted) { 494898d1cbfeSGeorge Wilson vd->vdev_remove_wanted = B_FALSE; 494998d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 495049cf58c0SBrendan Gregg - Sun Microsystems vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 49511d713200SEric Schrock 49521d713200SEric Schrock /* 49531d713200SEric Schrock * We want to clear the stats, but we don't want to do a full 49541d713200SEric Schrock * vdev_clear() as that will cause us to throw away 49551d713200SEric Schrock * degraded/faulted state as well as attempt to reopen the 49561d713200SEric Schrock * device, all of which is a waste. 49571d713200SEric Schrock */ 49581d713200SEric Schrock vd->vdev_stat.vs_read_errors = 0; 49591d713200SEric Schrock vd->vdev_stat.vs_write_errors = 0; 49601d713200SEric Schrock vd->vdev_stat.vs_checksum_errors = 0; 49611d713200SEric Schrock 4962e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 4963ea8dc4b6Seschrock } 496449cf58c0SBrendan Gregg - Sun Microsystems 4965e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 496649cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, vd->vdev_child[c]); 4967ea8dc4b6Seschrock } 4968fa9e4066Sahrens 4969e14bb325SJeff Bonwick static void 4970e14bb325SJeff Bonwick spa_async_probe(spa_t *spa, vdev_t *vd) 4971e14bb325SJeff Bonwick { 4972e14bb325SJeff Bonwick if (vd->vdev_probe_wanted) { 497398d1cbfeSGeorge Wilson vd->vdev_probe_wanted = B_FALSE; 4974e14bb325SJeff Bonwick vdev_reopen(vd); /* vdev_open() does the actual probe */ 4975e14bb325SJeff Bonwick } 4976e14bb325SJeff Bonwick 4977e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 4978e14bb325SJeff Bonwick spa_async_probe(spa, vd->vdev_child[c]); 4979e14bb325SJeff Bonwick } 4980e14bb325SJeff Bonwick 4981573ca77eSGeorge Wilson static void 4982573ca77eSGeorge Wilson spa_async_autoexpand(spa_t *spa, vdev_t *vd) 4983573ca77eSGeorge Wilson { 4984573ca77eSGeorge Wilson sysevent_id_t eid; 4985573ca77eSGeorge Wilson nvlist_t *attr; 4986573ca77eSGeorge Wilson char *physpath; 4987573ca77eSGeorge Wilson 4988573ca77eSGeorge Wilson if (!spa->spa_autoexpand) 4989573ca77eSGeorge Wilson return; 4990573ca77eSGeorge Wilson 4991573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 4992573ca77eSGeorge Wilson vdev_t *cvd = vd->vdev_child[c]; 4993573ca77eSGeorge Wilson spa_async_autoexpand(spa, cvd); 4994573ca77eSGeorge Wilson } 4995573ca77eSGeorge Wilson 4996573ca77eSGeorge Wilson if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 4997573ca77eSGeorge Wilson return; 4998573ca77eSGeorge Wilson 4999573ca77eSGeorge Wilson physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5000573ca77eSGeorge Wilson (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5001573ca77eSGeorge Wilson 5002573ca77eSGeorge Wilson VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5003573ca77eSGeorge Wilson VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5004573ca77eSGeorge Wilson 5005573ca77eSGeorge Wilson (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5006573ca77eSGeorge Wilson ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 5007573ca77eSGeorge Wilson 5008573ca77eSGeorge Wilson nvlist_free(attr); 5009573ca77eSGeorge Wilson kmem_free(physpath, MAXPATHLEN); 5010573ca77eSGeorge Wilson } 5011573ca77eSGeorge Wilson 5012ea8dc4b6Seschrock static void 5013ea8dc4b6Seschrock spa_async_thread(spa_t *spa) 5014ea8dc4b6Seschrock { 5015e14bb325SJeff Bonwick int tasks; 5016ea8dc4b6Seschrock 5017ea8dc4b6Seschrock ASSERT(spa->spa_sync_on); 5018ea8dc4b6Seschrock 5019ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5020ea8dc4b6Seschrock tasks = spa->spa_async_tasks; 5021ea8dc4b6Seschrock spa->spa_async_tasks = 0; 5022ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5023ea8dc4b6Seschrock 50240373e76bSbonwick /* 50250373e76bSbonwick * See if the config needs to be updated. 50260373e76bSbonwick */ 50270373e76bSbonwick if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5028b24ab676SJeff Bonwick uint64_t old_space, new_space; 5029573ca77eSGeorge Wilson 50300373e76bSbonwick mutex_enter(&spa_namespace_lock); 5031b24ab676SJeff Bonwick old_space = metaslab_class_get_space(spa_normal_class(spa)); 50320373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5033b24ab676SJeff Bonwick new_space = metaslab_class_get_space(spa_normal_class(spa)); 50340373e76bSbonwick mutex_exit(&spa_namespace_lock); 5035573ca77eSGeorge Wilson 5036573ca77eSGeorge Wilson /* 5037573ca77eSGeorge Wilson * If the pool grew as a result of the config update, 5038573ca77eSGeorge Wilson * then log an internal history event. 5039573ca77eSGeorge Wilson */ 5040b24ab676SJeff Bonwick if (new_space != old_space) { 50413f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_VDEV_ONLINE, 50423f9d6ad7SLin Ling spa, NULL, 5043c8e1f6d2SMark J Musante "pool '%s' size: %llu(+%llu)", 5044b24ab676SJeff Bonwick spa_name(spa), new_space, new_space - old_space); 5045573ca77eSGeorge Wilson } 50460373e76bSbonwick } 50470373e76bSbonwick 5048ea8dc4b6Seschrock /* 50493d7072f8Seschrock * See if any devices need to be marked REMOVED. 5050ea8dc4b6Seschrock */ 5051e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_REMOVE) { 50528f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 50533d7072f8Seschrock spa_async_remove(spa, spa->spa_root_vdev); 5054e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 505549cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5056e14bb325SJeff Bonwick for (int i = 0; i < spa->spa_spares.sav_count; i++) 505749cf58c0SBrendan Gregg - Sun Microsystems spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5058e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 5059e14bb325SJeff Bonwick } 5060e14bb325SJeff Bonwick 5061573ca77eSGeorge Wilson if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5062573ca77eSGeorge Wilson spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5063573ca77eSGeorge Wilson spa_async_autoexpand(spa, spa->spa_root_vdev); 5064573ca77eSGeorge Wilson spa_config_exit(spa, SCL_CONFIG, FTAG); 5065573ca77eSGeorge Wilson } 5066573ca77eSGeorge Wilson 5067e14bb325SJeff Bonwick /* 5068e14bb325SJeff Bonwick * See if any devices need to be probed. 5069e14bb325SJeff Bonwick */ 5070e14bb325SJeff Bonwick if (tasks & SPA_ASYNC_PROBE) { 50718f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 5072e14bb325SJeff Bonwick spa_async_probe(spa, spa->spa_root_vdev); 5073e14bb325SJeff Bonwick (void) spa_vdev_state_exit(spa, NULL, 0); 50743d7072f8Seschrock } 5075ea8dc4b6Seschrock 5076ea8dc4b6Seschrock /* 5077ea8dc4b6Seschrock * If any devices are done replacing, detach them. 5078ea8dc4b6Seschrock */ 50793d7072f8Seschrock if (tasks & SPA_ASYNC_RESILVER_DONE) 50803d7072f8Seschrock spa_vdev_resilver_done(spa); 5081fa9e4066Sahrens 5082ea8dc4b6Seschrock /* 5083ea8dc4b6Seschrock * Kick off a resilver. 5084ea8dc4b6Seschrock */ 5085088f3894Sahrens if (tasks & SPA_ASYNC_RESILVER) 50863f9d6ad7SLin Ling dsl_resilver_restart(spa->spa_dsl_pool, 0); 5087ea8dc4b6Seschrock 5088ea8dc4b6Seschrock /* 5089ea8dc4b6Seschrock * Let the world know that we're done. 5090ea8dc4b6Seschrock */ 5091ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5092ea8dc4b6Seschrock spa->spa_async_thread = NULL; 5093ea8dc4b6Seschrock cv_broadcast(&spa->spa_async_cv); 5094ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5095ea8dc4b6Seschrock thread_exit(); 5096ea8dc4b6Seschrock } 5097ea8dc4b6Seschrock 5098ea8dc4b6Seschrock void 5099ea8dc4b6Seschrock spa_async_suspend(spa_t *spa) 5100ea8dc4b6Seschrock { 5101ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5102ea8dc4b6Seschrock spa->spa_async_suspended++; 5103ea8dc4b6Seschrock while (spa->spa_async_thread != NULL) 5104ea8dc4b6Seschrock cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5105ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5106ea8dc4b6Seschrock } 5107ea8dc4b6Seschrock 5108ea8dc4b6Seschrock void 5109ea8dc4b6Seschrock spa_async_resume(spa_t *spa) 5110ea8dc4b6Seschrock { 5111ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5112ea8dc4b6Seschrock ASSERT(spa->spa_async_suspended != 0); 5113ea8dc4b6Seschrock spa->spa_async_suspended--; 5114ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5115ea8dc4b6Seschrock } 5116ea8dc4b6Seschrock 5117ea8dc4b6Seschrock static void 5118ea8dc4b6Seschrock spa_async_dispatch(spa_t *spa) 5119ea8dc4b6Seschrock { 5120ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5121ea8dc4b6Seschrock if (spa->spa_async_tasks && !spa->spa_async_suspended && 51220373e76bSbonwick spa->spa_async_thread == NULL && 51230373e76bSbonwick rootdir != NULL && !vn_is_readonly(rootdir)) 5124ea8dc4b6Seschrock spa->spa_async_thread = thread_create(NULL, 0, 5125ea8dc4b6Seschrock spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5126ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5127ea8dc4b6Seschrock } 5128ea8dc4b6Seschrock 5129ea8dc4b6Seschrock void 5130ea8dc4b6Seschrock spa_async_request(spa_t *spa, int task) 5131ea8dc4b6Seschrock { 51323f9d6ad7SLin Ling zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5133ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 5134ea8dc4b6Seschrock spa->spa_async_tasks |= task; 5135ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 5136fa9e4066Sahrens } 5137fa9e4066Sahrens 5138fa9e4066Sahrens /* 5139fa9e4066Sahrens * ========================================================================== 5140fa9e4066Sahrens * SPA syncing routines 5141fa9e4066Sahrens * ========================================================================== 5142fa9e4066Sahrens */ 5143fa9e4066Sahrens 5144cde58dbcSMatthew Ahrens static int 5145cde58dbcSMatthew Ahrens bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5146cde58dbcSMatthew Ahrens { 5147cde58dbcSMatthew Ahrens bpobj_t *bpo = arg; 5148cde58dbcSMatthew Ahrens bpobj_enqueue(bpo, bp, tx); 5149cde58dbcSMatthew Ahrens return (0); 5150b24ab676SJeff Bonwick } 5151b24ab676SJeff Bonwick 5152cde58dbcSMatthew Ahrens static int 5153cde58dbcSMatthew Ahrens spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5154b24ab676SJeff Bonwick { 5155b24ab676SJeff Bonwick zio_t *zio = arg; 5156b24ab676SJeff Bonwick 5157b24ab676SJeff Bonwick zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5158b24ab676SJeff Bonwick zio->io_flags)); 5159cde58dbcSMatthew Ahrens return (0); 5160fa9e4066Sahrens } 5161fa9e4066Sahrens 5162fa9e4066Sahrens static void 516399653d4eSeschrock spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5164fa9e4066Sahrens { 5165fa9e4066Sahrens char *packed = NULL; 5166f7991ba4STim Haley size_t bufsize; 5167fa9e4066Sahrens size_t nvsize = 0; 5168fa9e4066Sahrens dmu_buf_t *db; 5169fa9e4066Sahrens 517099653d4eSeschrock VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5171fa9e4066Sahrens 5172f7991ba4STim Haley /* 5173f7991ba4STim Haley * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5174f7991ba4STim Haley * information. This avoids the dbuf_will_dirty() path and 5175f7991ba4STim Haley * saves us a pre-read to get data we don't actually care about. 5176f7991ba4STim Haley */ 5177f7991ba4STim Haley bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE); 5178f7991ba4STim Haley packed = kmem_alloc(bufsize, KM_SLEEP); 5179fa9e4066Sahrens 518099653d4eSeschrock VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5181ea8dc4b6Seschrock KM_SLEEP) == 0); 5182f7991ba4STim Haley bzero(packed + nvsize, bufsize - nvsize); 5183fa9e4066Sahrens 5184f7991ba4STim Haley dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5185fa9e4066Sahrens 5186f7991ba4STim Haley kmem_free(packed, bufsize); 5187fa9e4066Sahrens 518899653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5189fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 5190fa9e4066Sahrens *(uint64_t *)db->db_data = nvsize; 5191ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 5192fa9e4066Sahrens } 5193fa9e4066Sahrens 519499653d4eSeschrock static void 5195fa94a07fSbrendan spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5196fa94a07fSbrendan const char *config, const char *entry) 519799653d4eSeschrock { 519899653d4eSeschrock nvlist_t *nvroot; 5199fa94a07fSbrendan nvlist_t **list; 520099653d4eSeschrock int i; 520199653d4eSeschrock 5202fa94a07fSbrendan if (!sav->sav_sync) 520399653d4eSeschrock return; 520499653d4eSeschrock 520599653d4eSeschrock /* 5206fa94a07fSbrendan * Update the MOS nvlist describing the list of available devices. 5207fa94a07fSbrendan * spa_validate_aux() will have already made sure this nvlist is 52083d7072f8Seschrock * valid and the vdevs are labeled appropriately. 520999653d4eSeschrock */ 5210fa94a07fSbrendan if (sav->sav_object == 0) { 5211fa94a07fSbrendan sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5212fa94a07fSbrendan DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5213fa94a07fSbrendan sizeof (uint64_t), tx); 521499653d4eSeschrock VERIFY(zap_update(spa->spa_meta_objset, 5215fa94a07fSbrendan DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5216fa94a07fSbrendan &sav->sav_object, tx) == 0); 521799653d4eSeschrock } 521899653d4eSeschrock 521999653d4eSeschrock VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5220fa94a07fSbrendan if (sav->sav_count == 0) { 5221fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 522299653d4eSeschrock } else { 5223fa94a07fSbrendan list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5224fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 5225fa94a07fSbrendan list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 52263f9d6ad7SLin Ling B_FALSE, VDEV_CONFIG_L2CACHE); 5227fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5228fa94a07fSbrendan sav->sav_count) == 0); 5229fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 5230fa94a07fSbrendan nvlist_free(list[i]); 5231fa94a07fSbrendan kmem_free(list, sav->sav_count * sizeof (void *)); 523299653d4eSeschrock } 523399653d4eSeschrock 5234fa94a07fSbrendan spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 523506eeb2adSek nvlist_free(nvroot); 523699653d4eSeschrock 5237fa94a07fSbrendan sav->sav_sync = B_FALSE; 523899653d4eSeschrock } 523999653d4eSeschrock 524099653d4eSeschrock static void 524199653d4eSeschrock spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 524299653d4eSeschrock { 524399653d4eSeschrock nvlist_t *config; 524499653d4eSeschrock 5245e14bb325SJeff Bonwick if (list_is_empty(&spa->spa_config_dirty_list)) 524699653d4eSeschrock return; 524799653d4eSeschrock 5248e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5249e14bb325SJeff Bonwick 5250e14bb325SJeff Bonwick config = spa_config_generate(spa, spa->spa_root_vdev, 5251e14bb325SJeff Bonwick dmu_tx_get_txg(tx), B_FALSE); 5252e14bb325SJeff Bonwick 5253e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 525499653d4eSeschrock 525599653d4eSeschrock if (spa->spa_config_syncing) 525699653d4eSeschrock nvlist_free(spa->spa_config_syncing); 525799653d4eSeschrock spa->spa_config_syncing = config; 525899653d4eSeschrock 525999653d4eSeschrock spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 526099653d4eSeschrock } 526199653d4eSeschrock 5262990b4856Slling /* 5263990b4856Slling * Set zpool properties. 5264990b4856Slling */ 5265b1b8ab34Slling static void 52663f9d6ad7SLin Ling spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) 5267b1b8ab34Slling { 5268b1b8ab34Slling spa_t *spa = arg1; 5269b1b8ab34Slling objset_t *mos = spa->spa_meta_objset; 5270990b4856Slling nvlist_t *nvp = arg2; 5271990b4856Slling nvpair_t *elem; 52723d7072f8Seschrock uint64_t intval; 5273c5904d13Seschrock char *strval; 5274990b4856Slling zpool_prop_t prop; 5275990b4856Slling const char *propname; 5276990b4856Slling zprop_type_t proptype; 5277b1b8ab34Slling 5278e14bb325SJeff Bonwick mutex_enter(&spa->spa_props_lock); 5279e14bb325SJeff Bonwick 5280990b4856Slling elem = NULL; 5281990b4856Slling while ((elem = nvlist_next_nvpair(nvp, elem))) { 5282990b4856Slling switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 5283990b4856Slling case ZPOOL_PROP_VERSION: 5284990b4856Slling /* 5285990b4856Slling * Only set version for non-zpool-creation cases 5286990b4856Slling * (set/import). spa_create() needs special care 5287990b4856Slling * for version setting. 5288990b4856Slling */ 5289990b4856Slling if (tx->tx_txg != TXG_INITIAL) { 5290990b4856Slling VERIFY(nvpair_value_uint64(elem, 5291990b4856Slling &intval) == 0); 5292990b4856Slling ASSERT(intval <= SPA_VERSION); 5293990b4856Slling ASSERT(intval >= spa_version(spa)); 5294990b4856Slling spa->spa_uberblock.ub_version = intval; 5295990b4856Slling vdev_config_dirty(spa->spa_root_vdev); 5296990b4856Slling } 5297ecd6cf80Smarks break; 5298990b4856Slling 5299990b4856Slling case ZPOOL_PROP_ALTROOT: 5300990b4856Slling /* 5301990b4856Slling * 'altroot' is a non-persistent property. It should 5302990b4856Slling * have been set temporarily at creation or import time. 5303990b4856Slling */ 5304990b4856Slling ASSERT(spa->spa_root != NULL); 5305b1b8ab34Slling break; 53063d7072f8Seschrock 53072f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 5308990b4856Slling /* 5309379c004dSEric Schrock * 'cachefile' is also a non-persisitent property. 5310990b4856Slling */ 53113d7072f8Seschrock break; 5312990b4856Slling default: 5313990b4856Slling /* 5314990b4856Slling * Set pool property values in the poolprops mos object. 5315990b4856Slling */ 5316990b4856Slling if (spa->spa_pool_props_object == 0) { 5317990b4856Slling VERIFY((spa->spa_pool_props_object = 5318990b4856Slling zap_create(mos, DMU_OT_POOL_PROPS, 5319990b4856Slling DMU_OT_NONE, 0, tx)) > 0); 5320990b4856Slling 5321990b4856Slling VERIFY(zap_update(mos, 5322990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 5323990b4856Slling 8, 1, &spa->spa_pool_props_object, tx) 5324990b4856Slling == 0); 5325990b4856Slling } 5326990b4856Slling 5327990b4856Slling /* normalize the property name */ 5328990b4856Slling propname = zpool_prop_to_name(prop); 5329990b4856Slling proptype = zpool_prop_get_type(prop); 5330990b4856Slling 5331990b4856Slling if (nvpair_type(elem) == DATA_TYPE_STRING) { 5332990b4856Slling ASSERT(proptype == PROP_TYPE_STRING); 5333990b4856Slling VERIFY(nvpair_value_string(elem, &strval) == 0); 5334990b4856Slling VERIFY(zap_update(mos, 5335990b4856Slling spa->spa_pool_props_object, propname, 5336990b4856Slling 1, strlen(strval) + 1, strval, tx) == 0); 5337990b4856Slling 5338990b4856Slling } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 5339990b4856Slling VERIFY(nvpair_value_uint64(elem, &intval) == 0); 5340990b4856Slling 5341990b4856Slling if (proptype == PROP_TYPE_INDEX) { 5342990b4856Slling const char *unused; 5343990b4856Slling VERIFY(zpool_prop_index_to_string( 5344990b4856Slling prop, intval, &unused) == 0); 5345990b4856Slling } 5346990b4856Slling VERIFY(zap_update(mos, 5347990b4856Slling spa->spa_pool_props_object, propname, 5348990b4856Slling 8, 1, &intval, tx) == 0); 5349990b4856Slling } else { 5350990b4856Slling ASSERT(0); /* not allowed */ 5351990b4856Slling } 5352990b4856Slling 53530a4e9518Sgw switch (prop) { 53540a4e9518Sgw case ZPOOL_PROP_DELEGATION: 5355990b4856Slling spa->spa_delegation = intval; 53560a4e9518Sgw break; 53570a4e9518Sgw case ZPOOL_PROP_BOOTFS: 5358990b4856Slling spa->spa_bootfs = intval; 53590a4e9518Sgw break; 53600a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 53610a4e9518Sgw spa->spa_failmode = intval; 53620a4e9518Sgw break; 5363573ca77eSGeorge Wilson case ZPOOL_PROP_AUTOEXPAND: 5364573ca77eSGeorge Wilson spa->spa_autoexpand = intval; 5365b98131cfSEric Taylor if (tx->tx_txg != TXG_INITIAL) 5366b98131cfSEric Taylor spa_async_request(spa, 5367b98131cfSEric Taylor SPA_ASYNC_AUTOEXPAND); 5368573ca77eSGeorge Wilson break; 5369b24ab676SJeff Bonwick case ZPOOL_PROP_DEDUPDITTO: 5370b24ab676SJeff Bonwick spa->spa_dedup_ditto = intval; 5371b24ab676SJeff Bonwick break; 53720a4e9518Sgw default: 53730a4e9518Sgw break; 53740a4e9518Sgw } 5375990b4856Slling } 5376990b4856Slling 5377990b4856Slling /* log internal history if this is not a zpool create */ 5378990b4856Slling if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY && 5379990b4856Slling tx->tx_txg != TXG_INITIAL) { 53803f9d6ad7SLin Ling spa_history_log_internal(LOG_POOL_PROPSET, 53813f9d6ad7SLin Ling spa, tx, "%s %lld %s", 5382e14bb325SJeff Bonwick nvpair_name(elem), intval, spa_name(spa)); 5383b1b8ab34Slling } 5384b1b8ab34Slling } 5385e14bb325SJeff Bonwick 5386e14bb325SJeff Bonwick mutex_exit(&spa->spa_props_lock); 5387b1b8ab34Slling } 5388b1b8ab34Slling 5389cde58dbcSMatthew Ahrens /* 5390cde58dbcSMatthew Ahrens * Perform one-time upgrade on-disk changes. spa_version() does not 5391cde58dbcSMatthew Ahrens * reflect the new version this txg, so there must be no changes this 5392cde58dbcSMatthew Ahrens * txg to anything that the upgrade code depends on after it executes. 5393cde58dbcSMatthew Ahrens * Therefore this must be called after dsl_pool_sync() does the sync 5394cde58dbcSMatthew Ahrens * tasks. 5395cde58dbcSMatthew Ahrens */ 5396cde58dbcSMatthew Ahrens static void 5397cde58dbcSMatthew Ahrens spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 5398cde58dbcSMatthew Ahrens { 5399cde58dbcSMatthew Ahrens dsl_pool_t *dp = spa->spa_dsl_pool; 5400cde58dbcSMatthew Ahrens 5401cde58dbcSMatthew Ahrens ASSERT(spa->spa_sync_pass == 1); 5402cde58dbcSMatthew Ahrens 5403cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 5404cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 5405cde58dbcSMatthew Ahrens dsl_pool_create_origin(dp, tx); 5406cde58dbcSMatthew Ahrens 5407cde58dbcSMatthew Ahrens /* Keeping the origin open increases spa_minref */ 5408cde58dbcSMatthew Ahrens spa->spa_minref += 3; 5409cde58dbcSMatthew Ahrens } 5410cde58dbcSMatthew Ahrens 5411cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 5412cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 5413cde58dbcSMatthew Ahrens dsl_pool_upgrade_clones(dp, tx); 5414cde58dbcSMatthew Ahrens } 5415cde58dbcSMatthew Ahrens 5416cde58dbcSMatthew Ahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 5417cde58dbcSMatthew Ahrens spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 5418cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dp, tx); 5419cde58dbcSMatthew Ahrens 5420cde58dbcSMatthew Ahrens /* Keeping the freedir open increases spa_minref */ 5421cde58dbcSMatthew Ahrens spa->spa_minref += 3; 5422cde58dbcSMatthew Ahrens } 5423cde58dbcSMatthew Ahrens } 5424cde58dbcSMatthew Ahrens 5425fa9e4066Sahrens /* 5426fa9e4066Sahrens * Sync the specified transaction group. New blocks may be dirtied as 5427fa9e4066Sahrens * part of the process, so we iterate until it converges. 5428fa9e4066Sahrens */ 5429fa9e4066Sahrens void 5430fa9e4066Sahrens spa_sync(spa_t *spa, uint64_t txg) 5431fa9e4066Sahrens { 5432fa9e4066Sahrens dsl_pool_t *dp = spa->spa_dsl_pool; 5433fa9e4066Sahrens objset_t *mos = spa->spa_meta_objset; 5434cde58dbcSMatthew Ahrens bpobj_t *defer_bpo = &spa->spa_deferred_bpobj; 5435b24ab676SJeff Bonwick bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 54360373e76bSbonwick vdev_t *rvd = spa->spa_root_vdev; 5437fa9e4066Sahrens vdev_t *vd; 5438fa9e4066Sahrens dmu_tx_t *tx; 5439e14bb325SJeff Bonwick int error; 5440fa9e4066Sahrens 5441fa9e4066Sahrens /* 5442fa9e4066Sahrens * Lock out configuration changes. 5443fa9e4066Sahrens */ 5444e14bb325SJeff Bonwick spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5445fa9e4066Sahrens 5446fa9e4066Sahrens spa->spa_syncing_txg = txg; 5447fa9e4066Sahrens spa->spa_sync_pass = 0; 5448fa9e4066Sahrens 5449e14bb325SJeff Bonwick /* 5450e14bb325SJeff Bonwick * If there are any pending vdev state changes, convert them 5451e14bb325SJeff Bonwick * into config changes that go out with this transaction group. 5452e14bb325SJeff Bonwick */ 5453e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 54548ad4d6ddSJeff Bonwick while (list_head(&spa->spa_state_dirty_list) != NULL) { 54558ad4d6ddSJeff Bonwick /* 54568ad4d6ddSJeff Bonwick * We need the write lock here because, for aux vdevs, 54578ad4d6ddSJeff Bonwick * calling vdev_config_dirty() modifies sav_config. 54588ad4d6ddSJeff Bonwick * This is ugly and will become unnecessary when we 54598ad4d6ddSJeff Bonwick * eliminate the aux vdev wart by integrating all vdevs 54608ad4d6ddSJeff Bonwick * into the root vdev tree. 54618ad4d6ddSJeff Bonwick */ 54628ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 54638ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 54648ad4d6ddSJeff Bonwick while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 54658ad4d6ddSJeff Bonwick vdev_state_clean(vd); 54668ad4d6ddSJeff Bonwick vdev_config_dirty(vd); 54678ad4d6ddSJeff Bonwick } 54688ad4d6ddSJeff Bonwick spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 54698ad4d6ddSJeff Bonwick spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 5470e14bb325SJeff Bonwick } 5471e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 5472e14bb325SJeff Bonwick 547399653d4eSeschrock tx = dmu_tx_create_assigned(dp, txg); 547499653d4eSeschrock 547599653d4eSeschrock /* 5476e7437265Sahrens * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 547799653d4eSeschrock * set spa_deflate if we have no raid-z vdevs. 547899653d4eSeschrock */ 5479e7437265Sahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 5480e7437265Sahrens spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 548199653d4eSeschrock int i; 548299653d4eSeschrock 548399653d4eSeschrock for (i = 0; i < rvd->vdev_children; i++) { 548499653d4eSeschrock vd = rvd->vdev_child[i]; 548599653d4eSeschrock if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 548699653d4eSeschrock break; 548799653d4eSeschrock } 548899653d4eSeschrock if (i == rvd->vdev_children) { 548999653d4eSeschrock spa->spa_deflate = TRUE; 549099653d4eSeschrock VERIFY(0 == zap_add(spa->spa_meta_objset, 549199653d4eSeschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 549299653d4eSeschrock sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 549399653d4eSeschrock } 549499653d4eSeschrock } 549599653d4eSeschrock 5496fa9e4066Sahrens /* 54973f9d6ad7SLin Ling * If anything has changed in this txg, or if someone is waiting 54983f9d6ad7SLin Ling * for this txg to sync (eg, spa_vdev_remove()), push the 54993f9d6ad7SLin Ling * deferred frees from the previous txg. If not, leave them 55003f9d6ad7SLin Ling * alone so that we don't generate work on an otherwise idle 55013f9d6ad7SLin Ling * system. 5502fa9e4066Sahrens */ 5503fa9e4066Sahrens if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 55041615a317Sek !txg_list_empty(&dp->dp_dirty_dirs, txg) || 55053f9d6ad7SLin Ling !txg_list_empty(&dp->dp_sync_tasks, txg) || 5506cde58dbcSMatthew Ahrens ((dsl_scan_active(dp->dp_scan) || 5507cde58dbcSMatthew Ahrens txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 5508cde58dbcSMatthew Ahrens zio_t *zio = zio_root(spa, NULL, NULL, 0); 5509cde58dbcSMatthew Ahrens VERIFY3U(bpobj_iterate(defer_bpo, 5510cde58dbcSMatthew Ahrens spa_free_sync_cb, zio, tx), ==, 0); 5511cde58dbcSMatthew Ahrens VERIFY3U(zio_wait(zio), ==, 0); 5512cde58dbcSMatthew Ahrens } 5513fa9e4066Sahrens 5514fa9e4066Sahrens /* 5515fa9e4066Sahrens * Iterate to convergence. 5516fa9e4066Sahrens */ 5517fa9e4066Sahrens do { 5518b24ab676SJeff Bonwick int pass = ++spa->spa_sync_pass; 5519fa9e4066Sahrens 5520fa9e4066Sahrens spa_sync_config_object(spa, tx); 5521fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_spares, tx, 5522fa94a07fSbrendan ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 5523fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 5524fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 5525ea8dc4b6Seschrock spa_errlog_sync(spa, txg); 5526fa9e4066Sahrens dsl_pool_sync(dp, txg); 5527fa9e4066Sahrens 5528b24ab676SJeff Bonwick if (pass <= SYNC_PASS_DEFERRED_FREE) { 5529b24ab676SJeff Bonwick zio_t *zio = zio_root(spa, NULL, NULL, 0); 5530cde58dbcSMatthew Ahrens bplist_iterate(free_bpl, spa_free_sync_cb, 5531cde58dbcSMatthew Ahrens zio, tx); 5532b24ab676SJeff Bonwick VERIFY(zio_wait(zio) == 0); 5533b24ab676SJeff Bonwick } else { 5534cde58dbcSMatthew Ahrens bplist_iterate(free_bpl, bpobj_enqueue_cb, 5535cde58dbcSMatthew Ahrens defer_bpo, tx); 5536fa9e4066Sahrens } 5537fa9e4066Sahrens 5538b24ab676SJeff Bonwick ddt_sync(spa, txg); 55393f9d6ad7SLin Ling dsl_scan_sync(dp, tx); 5540afee20e4SGeorge Wilson 5541b24ab676SJeff Bonwick while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 5542b24ab676SJeff Bonwick vdev_sync(vd, txg); 5543b24ab676SJeff Bonwick 5544cde58dbcSMatthew Ahrens if (pass == 1) 5545cde58dbcSMatthew Ahrens spa_sync_upgrades(spa, tx); 5546fa9e4066Sahrens 5547cde58dbcSMatthew Ahrens } while (dmu_objset_is_dirty(mos, txg)); 5548fa9e4066Sahrens 5549fa9e4066Sahrens /* 5550fa9e4066Sahrens * Rewrite the vdev configuration (which includes the uberblock) 5551fa9e4066Sahrens * to commit the transaction group. 55520373e76bSbonwick * 555317f17c2dSbonwick * If there are no dirty vdevs, we sync the uberblock to a few 555417f17c2dSbonwick * random top-level vdevs that are known to be visible in the 5555e14bb325SJeff Bonwick * config cache (see spa_vdev_add() for a complete description). 5556e14bb325SJeff Bonwick * If there *are* dirty vdevs, sync the uberblock to all vdevs. 55570373e76bSbonwick */ 5558e14bb325SJeff Bonwick for (;;) { 5559e14bb325SJeff Bonwick /* 5560e14bb325SJeff Bonwick * We hold SCL_STATE to prevent vdev open/close/etc. 5561e14bb325SJeff Bonwick * while we're attempting to write the vdev labels. 5562e14bb325SJeff Bonwick */ 5563e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5564e14bb325SJeff Bonwick 5565e14bb325SJeff Bonwick if (list_is_empty(&spa->spa_config_dirty_list)) { 5566e14bb325SJeff Bonwick vdev_t *svd[SPA_DVAS_PER_BP]; 5567e14bb325SJeff Bonwick int svdcount = 0; 5568e14bb325SJeff Bonwick int children = rvd->vdev_children; 5569e14bb325SJeff Bonwick int c0 = spa_get_random(children); 5570e14bb325SJeff Bonwick 5571573ca77eSGeorge Wilson for (int c = 0; c < children; c++) { 5572e14bb325SJeff Bonwick vd = rvd->vdev_child[(c0 + c) % children]; 5573e14bb325SJeff Bonwick if (vd->vdev_ms_array == 0 || vd->vdev_islog) 5574e14bb325SJeff Bonwick continue; 5575e14bb325SJeff Bonwick svd[svdcount++] = vd; 5576e14bb325SJeff Bonwick if (svdcount == SPA_DVAS_PER_BP) 5577e14bb325SJeff Bonwick break; 5578e14bb325SJeff Bonwick } 55798956713aSEric Schrock error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 55808956713aSEric Schrock if (error != 0) 55818956713aSEric Schrock error = vdev_config_sync(svd, svdcount, txg, 55828956713aSEric Schrock B_TRUE); 5583e14bb325SJeff Bonwick } else { 5584e14bb325SJeff Bonwick error = vdev_config_sync(rvd->vdev_child, 55858956713aSEric Schrock rvd->vdev_children, txg, B_FALSE); 55868956713aSEric Schrock if (error != 0) 55878956713aSEric Schrock error = vdev_config_sync(rvd->vdev_child, 55888956713aSEric Schrock rvd->vdev_children, txg, B_TRUE); 55890373e76bSbonwick } 5590e14bb325SJeff Bonwick 5591e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 5592e14bb325SJeff Bonwick 5593e14bb325SJeff Bonwick if (error == 0) 5594e14bb325SJeff Bonwick break; 5595e14bb325SJeff Bonwick zio_suspend(spa, NULL); 5596e14bb325SJeff Bonwick zio_resume_wait(spa); 55970373e76bSbonwick } 559899653d4eSeschrock dmu_tx_commit(tx); 559999653d4eSeschrock 56000373e76bSbonwick /* 56010373e76bSbonwick * Clear the dirty config list. 5602fa9e4066Sahrens */ 5603e14bb325SJeff Bonwick while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 56040373e76bSbonwick vdev_config_clean(vd); 56050373e76bSbonwick 56060373e76bSbonwick /* 56070373e76bSbonwick * Now that the new config has synced transactionally, 56080373e76bSbonwick * let it become visible to the config cache. 56090373e76bSbonwick */ 56100373e76bSbonwick if (spa->spa_config_syncing != NULL) { 56110373e76bSbonwick spa_config_set(spa, spa->spa_config_syncing); 56120373e76bSbonwick spa->spa_config_txg = txg; 56130373e76bSbonwick spa->spa_config_syncing = NULL; 56140373e76bSbonwick } 5615fa9e4066Sahrens 5616fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 5617fa9e4066Sahrens 5618b24ab676SJeff Bonwick dsl_pool_sync_done(dp, txg); 5619fa9e4066Sahrens 5620fa9e4066Sahrens /* 5621fa9e4066Sahrens * Update usable space statistics. 5622fa9e4066Sahrens */ 5623fa9e4066Sahrens while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 5624fa9e4066Sahrens vdev_sync_done(vd, txg); 5625fa9e4066Sahrens 5626485bbbf5SGeorge Wilson spa_update_dspace(spa); 5627485bbbf5SGeorge Wilson 5628fa9e4066Sahrens /* 5629fa9e4066Sahrens * It had better be the case that we didn't dirty anything 563099653d4eSeschrock * since vdev_config_sync(). 5631fa9e4066Sahrens */ 5632fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 5633fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 5634fa9e4066Sahrens ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 5635b24ab676SJeff Bonwick 5636b24ab676SJeff Bonwick spa->spa_sync_pass = 0; 5637fa9e4066Sahrens 5638e14bb325SJeff Bonwick spa_config_exit(spa, SCL_CONFIG, FTAG); 5639ea8dc4b6Seschrock 5640468c413aSTim Haley spa_handle_ignored_writes(spa); 5641468c413aSTim Haley 5642ea8dc4b6Seschrock /* 5643ea8dc4b6Seschrock * If any async tasks have been requested, kick them off. 5644ea8dc4b6Seschrock */ 5645ea8dc4b6Seschrock spa_async_dispatch(spa); 5646fa9e4066Sahrens } 5647fa9e4066Sahrens 5648fa9e4066Sahrens /* 5649fa9e4066Sahrens * Sync all pools. We don't want to hold the namespace lock across these 5650fa9e4066Sahrens * operations, so we take a reference on the spa_t and drop the lock during the 5651fa9e4066Sahrens * sync. 5652fa9e4066Sahrens */ 5653fa9e4066Sahrens void 5654fa9e4066Sahrens spa_sync_allpools(void) 5655fa9e4066Sahrens { 5656fa9e4066Sahrens spa_t *spa = NULL; 5657fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 5658fa9e4066Sahrens while ((spa = spa_next(spa)) != NULL) { 5659e14bb325SJeff Bonwick if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa)) 5660fa9e4066Sahrens continue; 5661fa9e4066Sahrens spa_open_ref(spa, FTAG); 5662fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 5663fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), 0); 5664fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 5665fa9e4066Sahrens spa_close(spa, FTAG); 5666fa9e4066Sahrens } 5667fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 5668fa9e4066Sahrens } 5669fa9e4066Sahrens 5670fa9e4066Sahrens /* 5671fa9e4066Sahrens * ========================================================================== 5672fa9e4066Sahrens * Miscellaneous routines 5673fa9e4066Sahrens * ========================================================================== 5674fa9e4066Sahrens */ 5675fa9e4066Sahrens 5676fa9e4066Sahrens /* 5677fa9e4066Sahrens * Remove all pools in the system. 5678fa9e4066Sahrens */ 5679fa9e4066Sahrens void 5680fa9e4066Sahrens spa_evict_all(void) 5681fa9e4066Sahrens { 5682fa9e4066Sahrens spa_t *spa; 5683fa9e4066Sahrens 5684fa9e4066Sahrens /* 5685fa9e4066Sahrens * Remove all cached state. All pools should be closed now, 5686fa9e4066Sahrens * so every spa in the AVL tree should be unreferenced. 5687fa9e4066Sahrens */ 5688fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 5689fa9e4066Sahrens while ((spa = spa_next(NULL)) != NULL) { 5690fa9e4066Sahrens /* 5691ea8dc4b6Seschrock * Stop async tasks. The async thread may need to detach 5692ea8dc4b6Seschrock * a device that's been replaced, which requires grabbing 5693ea8dc4b6Seschrock * spa_namespace_lock, so we must drop it here. 5694fa9e4066Sahrens */ 5695fa9e4066Sahrens spa_open_ref(spa, FTAG); 5696fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 5697ea8dc4b6Seschrock spa_async_suspend(spa); 5698fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 5699fa9e4066Sahrens spa_close(spa, FTAG); 5700fa9e4066Sahrens 5701fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 5702fa9e4066Sahrens spa_unload(spa); 5703fa9e4066Sahrens spa_deactivate(spa); 5704fa9e4066Sahrens } 5705fa9e4066Sahrens spa_remove(spa); 5706fa9e4066Sahrens } 5707fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 5708fa9e4066Sahrens } 5709ea8dc4b6Seschrock 5710ea8dc4b6Seschrock vdev_t * 57116809eb4eSEric Schrock spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 5712ea8dc4b6Seschrock { 5713c5904d13Seschrock vdev_t *vd; 5714c5904d13Seschrock int i; 5715c5904d13Seschrock 5716c5904d13Seschrock if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 5717c5904d13Seschrock return (vd); 5718c5904d13Seschrock 57196809eb4eSEric Schrock if (aux) { 5720c5904d13Seschrock for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 5721c5904d13Seschrock vd = spa->spa_l2cache.sav_vdevs[i]; 57226809eb4eSEric Schrock if (vd->vdev_guid == guid) 57236809eb4eSEric Schrock return (vd); 57246809eb4eSEric Schrock } 57256809eb4eSEric Schrock 57266809eb4eSEric Schrock for (i = 0; i < spa->spa_spares.sav_count; i++) { 57276809eb4eSEric Schrock vd = spa->spa_spares.sav_vdevs[i]; 5728c5904d13Seschrock if (vd->vdev_guid == guid) 5729c5904d13Seschrock return (vd); 5730c5904d13Seschrock } 5731c5904d13Seschrock } 5732c5904d13Seschrock 5733c5904d13Seschrock return (NULL); 5734ea8dc4b6Seschrock } 5735eaca9bbdSeschrock 5736eaca9bbdSeschrock void 5737990b4856Slling spa_upgrade(spa_t *spa, uint64_t version) 5738eaca9bbdSeschrock { 5739e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5740eaca9bbdSeschrock 5741eaca9bbdSeschrock /* 5742eaca9bbdSeschrock * This should only be called for a non-faulted pool, and since a 5743eaca9bbdSeschrock * future version would result in an unopenable pool, this shouldn't be 5744eaca9bbdSeschrock * possible. 5745eaca9bbdSeschrock */ 5746e7437265Sahrens ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION); 5747990b4856Slling ASSERT(version >= spa->spa_uberblock.ub_version); 5748eaca9bbdSeschrock 5749990b4856Slling spa->spa_uberblock.ub_version = version; 5750eaca9bbdSeschrock vdev_config_dirty(spa->spa_root_vdev); 5751eaca9bbdSeschrock 5752e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 575399653d4eSeschrock 575499653d4eSeschrock txg_wait_synced(spa_get_dsl(spa), 0); 575599653d4eSeschrock } 575699653d4eSeschrock 575799653d4eSeschrock boolean_t 575899653d4eSeschrock spa_has_spare(spa_t *spa, uint64_t guid) 575999653d4eSeschrock { 576099653d4eSeschrock int i; 576139c23413Seschrock uint64_t spareguid; 5762fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_spares; 576399653d4eSeschrock 5764fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 5765fa94a07fSbrendan if (sav->sav_vdevs[i]->vdev_guid == guid) 576699653d4eSeschrock return (B_TRUE); 576799653d4eSeschrock 5768fa94a07fSbrendan for (i = 0; i < sav->sav_npending; i++) { 5769fa94a07fSbrendan if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 5770fa94a07fSbrendan &spareguid) == 0 && spareguid == guid) 577139c23413Seschrock return (B_TRUE); 577239c23413Seschrock } 577339c23413Seschrock 577499653d4eSeschrock return (B_FALSE); 5775eaca9bbdSeschrock } 5776b1b8ab34Slling 577789a89ebfSlling /* 577889a89ebfSlling * Check if a pool has an active shared spare device. 577989a89ebfSlling * Note: reference count of an active spare is 2, as a spare and as a replace 578089a89ebfSlling */ 578189a89ebfSlling static boolean_t 578289a89ebfSlling spa_has_active_shared_spare(spa_t *spa) 578389a89ebfSlling { 578489a89ebfSlling int i, refcnt; 578589a89ebfSlling uint64_t pool; 578689a89ebfSlling spa_aux_vdev_t *sav = &spa->spa_spares; 578789a89ebfSlling 578889a89ebfSlling for (i = 0; i < sav->sav_count; i++) { 578989a89ebfSlling if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 579089a89ebfSlling &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 579189a89ebfSlling refcnt > 2) 579289a89ebfSlling return (B_TRUE); 579389a89ebfSlling } 579489a89ebfSlling 579589a89ebfSlling return (B_FALSE); 579689a89ebfSlling } 579789a89ebfSlling 57983d7072f8Seschrock /* 57993d7072f8Seschrock * Post a sysevent corresponding to the given event. The 'name' must be one of 58003d7072f8Seschrock * the event definitions in sys/sysevent/eventdefs.h. The payload will be 58013d7072f8Seschrock * filled in from the spa and (optionally) the vdev. This doesn't do anything 58023d7072f8Seschrock * in the userland libzpool, as we don't want consumers to misinterpret ztest 58033d7072f8Seschrock * or zdb as real changes. 58043d7072f8Seschrock */ 58053d7072f8Seschrock void 58063d7072f8Seschrock spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 58073d7072f8Seschrock { 58083d7072f8Seschrock #ifdef _KERNEL 58093d7072f8Seschrock sysevent_t *ev; 58103d7072f8Seschrock sysevent_attr_list_t *attr = NULL; 58113d7072f8Seschrock sysevent_value_t value; 58123d7072f8Seschrock sysevent_id_t eid; 58133d7072f8Seschrock 58143d7072f8Seschrock ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 58153d7072f8Seschrock SE_SLEEP); 58163d7072f8Seschrock 58173d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 58183d7072f8Seschrock value.value.sv_string = spa_name(spa); 58193d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 58203d7072f8Seschrock goto done; 58213d7072f8Seschrock 58223d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 58233d7072f8Seschrock value.value.sv_uint64 = spa_guid(spa); 58243d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 58253d7072f8Seschrock goto done; 58263d7072f8Seschrock 58273d7072f8Seschrock if (vd) { 58283d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 58293d7072f8Seschrock value.value.sv_uint64 = vd->vdev_guid; 58303d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 58313d7072f8Seschrock SE_SLEEP) != 0) 58323d7072f8Seschrock goto done; 58333d7072f8Seschrock 58343d7072f8Seschrock if (vd->vdev_path) { 58353d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 58363d7072f8Seschrock value.value.sv_string = vd->vdev_path; 58373d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 58383d7072f8Seschrock &value, SE_SLEEP) != 0) 58393d7072f8Seschrock goto done; 58403d7072f8Seschrock } 58413d7072f8Seschrock } 58423d7072f8Seschrock 5843b01c3b58Seschrock if (sysevent_attach_attributes(ev, attr) != 0) 5844b01c3b58Seschrock goto done; 5845b01c3b58Seschrock attr = NULL; 5846b01c3b58Seschrock 58473d7072f8Seschrock (void) log_sysevent(ev, SE_SLEEP, &eid); 58483d7072f8Seschrock 58493d7072f8Seschrock done: 58503d7072f8Seschrock if (attr) 58513d7072f8Seschrock sysevent_free_attr(attr); 58523d7072f8Seschrock sysevent_free(ev); 58533d7072f8Seschrock #endif 58543d7072f8Seschrock } 5855