1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * 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. 24e9103aaeSGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25bf16b11eSMatthew Ahrens * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/zfs_context.h> 29ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 30fa9e4066Sahrens #include <sys/spa.h> 31fa9e4066Sahrens #include <sys/spa_impl.h> 32fa9e4066Sahrens #include <sys/dmu.h> 33fa9e4066Sahrens #include <sys/dmu_tx.h> 34fa9e4066Sahrens #include <sys/vdev_impl.h> 35fa9e4066Sahrens #include <sys/uberblock_impl.h> 36fa9e4066Sahrens #include <sys/metaslab.h> 37fa9e4066Sahrens #include <sys/metaslab_impl.h> 38fa9e4066Sahrens #include <sys/space_map.h> 390713e232SGeorge Wilson #include <sys/space_reftree.h> 40fa9e4066Sahrens #include <sys/zio.h> 41fa9e4066Sahrens #include <sys/zap.h> 42fa9e4066Sahrens #include <sys/fs/zfs.h> 43c5904d13Seschrock #include <sys/arc.h> 44e6ca193dSGeorge Wilson #include <sys/zil.h> 453f9d6ad7SLin Ling #include <sys/dsl_scan.h> 46fa9e4066Sahrens 47fa9e4066Sahrens /* 48fa9e4066Sahrens * Virtual device management. 49fa9e4066Sahrens */ 50fa9e4066Sahrens 51fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 52fa9e4066Sahrens &vdev_root_ops, 53fa9e4066Sahrens &vdev_raidz_ops, 54fa9e4066Sahrens &vdev_mirror_ops, 55fa9e4066Sahrens &vdev_replacing_ops, 5699653d4eSeschrock &vdev_spare_ops, 57fa9e4066Sahrens &vdev_disk_ops, 58fa9e4066Sahrens &vdev_file_ops, 59fa9e4066Sahrens &vdev_missing_ops, 6088ecc943SGeorge Wilson &vdev_hole_ops, 61fa9e4066Sahrens NULL 62fa9e4066Sahrens }; 63fa9e4066Sahrens 64088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 65088f3894Sahrens int zfs_scrub_limit = 10; 6605b2b3b8Smishra 67*bf3e216cSMatthew Ahrens /* 68*bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 69*bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 70*bf3e216cSMatthew Ahrens */ 71*bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 72*bf3e216cSMatthew Ahrens 73fa9e4066Sahrens /* 74fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 75fa9e4066Sahrens */ 76fa9e4066Sahrens static vdev_ops_t * 77fa9e4066Sahrens vdev_getops(const char *type) 78fa9e4066Sahrens { 79fa9e4066Sahrens vdev_ops_t *ops, **opspp; 80fa9e4066Sahrens 81fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 82fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 83fa9e4066Sahrens break; 84fa9e4066Sahrens 85fa9e4066Sahrens return (ops); 86fa9e4066Sahrens } 87fa9e4066Sahrens 88fa9e4066Sahrens /* 89fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 90fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 91fa9e4066Sahrens */ 92fa9e4066Sahrens uint64_t 93fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 94fa9e4066Sahrens { 95ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 96fa9e4066Sahrens uint64_t csize; 97fa9e4066Sahrens 98573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 99fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 100fa9e4066Sahrens asize = MAX(asize, csize); 101fa9e4066Sahrens } 102fa9e4066Sahrens 103fa9e4066Sahrens return (asize); 104fa9e4066Sahrens } 105fa9e4066Sahrens 1062a79c5feSlling /* 107573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 108573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 109573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 110573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1112a79c5feSlling */ 1122a79c5feSlling uint64_t 113573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1142a79c5feSlling { 115573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1162a79c5feSlling 117573ca77eSGeorge Wilson /* 1184263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 119573ca77eSGeorge Wilson * just return our own asize. 120573ca77eSGeorge Wilson */ 121573ca77eSGeorge Wilson if (pvd == NULL) 122573ca77eSGeorge Wilson return (vd->vdev_asize); 1232a79c5feSlling 1242a79c5feSlling /* 125573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 126573ca77eSGeorge Wilson * to the nearest metaslab. 1272a79c5feSlling */ 128573ca77eSGeorge Wilson if (vd == vd->vdev_top) 129573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1302a79c5feSlling 131573ca77eSGeorge Wilson /* 132573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 133573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 134573ca77eSGeorge Wilson */ 135573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 136573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1372a79c5feSlling 138573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 139573ca77eSGeorge Wilson } 1402a79c5feSlling 141573ca77eSGeorge Wilson void 142573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 143573ca77eSGeorge Wilson { 144573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 145573ca77eSGeorge Wilson 146573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 147573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1482a79c5feSlling } 1492a79c5feSlling 150fa9e4066Sahrens vdev_t * 151fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 152fa9e4066Sahrens { 153fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 154fa9e4066Sahrens 155e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 156e05725b1Sbonwick 157088f3894Sahrens if (vdev < rvd->vdev_children) { 158088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 159fa9e4066Sahrens return (rvd->vdev_child[vdev]); 160088f3894Sahrens } 161fa9e4066Sahrens 162fa9e4066Sahrens return (NULL); 163fa9e4066Sahrens } 164fa9e4066Sahrens 165fa9e4066Sahrens vdev_t * 166fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 167fa9e4066Sahrens { 168fa9e4066Sahrens vdev_t *mvd; 169fa9e4066Sahrens 1700e34b6a7Sbonwick if (vd->vdev_guid == guid) 171fa9e4066Sahrens return (vd); 172fa9e4066Sahrens 173573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 174fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 175fa9e4066Sahrens NULL) 176fa9e4066Sahrens return (mvd); 177fa9e4066Sahrens 178fa9e4066Sahrens return (NULL); 179fa9e4066Sahrens } 180fa9e4066Sahrens 181fa9e4066Sahrens void 182fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 183fa9e4066Sahrens { 184fa9e4066Sahrens size_t oldsize, newsize; 185fa9e4066Sahrens uint64_t id = cvd->vdev_id; 186fa9e4066Sahrens vdev_t **newchild; 187fa9e4066Sahrens 188e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 189fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 190fa9e4066Sahrens 191fa9e4066Sahrens cvd->vdev_parent = pvd; 192fa9e4066Sahrens 193fa9e4066Sahrens if (pvd == NULL) 194fa9e4066Sahrens return; 195fa9e4066Sahrens 196fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 197fa9e4066Sahrens 198fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 199fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 200fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 201fa9e4066Sahrens 202fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 203fa9e4066Sahrens if (pvd->vdev_child != NULL) { 204fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 205fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 206fa9e4066Sahrens } 207fa9e4066Sahrens 208fa9e4066Sahrens pvd->vdev_child = newchild; 209fa9e4066Sahrens pvd->vdev_child[id] = cvd; 210fa9e4066Sahrens 211fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 212fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 213fa9e4066Sahrens 214fa9e4066Sahrens /* 215fa9e4066Sahrens * Walk up all ancestors to update guid sum. 216fa9e4066Sahrens */ 217fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 218fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 219fa9e4066Sahrens } 220fa9e4066Sahrens 221fa9e4066Sahrens void 222fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 223fa9e4066Sahrens { 224fa9e4066Sahrens int c; 225fa9e4066Sahrens uint_t id = cvd->vdev_id; 226fa9e4066Sahrens 227fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 228fa9e4066Sahrens 229fa9e4066Sahrens if (pvd == NULL) 230fa9e4066Sahrens return; 231fa9e4066Sahrens 232fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 233fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 234fa9e4066Sahrens 235fa9e4066Sahrens pvd->vdev_child[id] = NULL; 236fa9e4066Sahrens cvd->vdev_parent = NULL; 237fa9e4066Sahrens 238fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 239fa9e4066Sahrens if (pvd->vdev_child[c]) 240fa9e4066Sahrens break; 241fa9e4066Sahrens 242fa9e4066Sahrens if (c == pvd->vdev_children) { 243fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 244fa9e4066Sahrens pvd->vdev_child = NULL; 245fa9e4066Sahrens pvd->vdev_children = 0; 246fa9e4066Sahrens } 247fa9e4066Sahrens 248fa9e4066Sahrens /* 249fa9e4066Sahrens * Walk up all ancestors to update guid sum. 250fa9e4066Sahrens */ 251fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 252fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 253fa9e4066Sahrens } 254fa9e4066Sahrens 255fa9e4066Sahrens /* 256fa9e4066Sahrens * Remove any holes in the child array. 257fa9e4066Sahrens */ 258fa9e4066Sahrens void 259fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 260fa9e4066Sahrens { 261fa9e4066Sahrens vdev_t **newchild, *cvd; 262fa9e4066Sahrens int oldc = pvd->vdev_children; 263573ca77eSGeorge Wilson int newc; 264fa9e4066Sahrens 265e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 266fa9e4066Sahrens 267573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 268fa9e4066Sahrens if (pvd->vdev_child[c]) 269fa9e4066Sahrens newc++; 270fa9e4066Sahrens 271fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 272fa9e4066Sahrens 273573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 274fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 275fa9e4066Sahrens newchild[newc] = cvd; 276fa9e4066Sahrens cvd->vdev_id = newc++; 277fa9e4066Sahrens } 278fa9e4066Sahrens } 279fa9e4066Sahrens 280fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 281fa9e4066Sahrens pvd->vdev_child = newchild; 282fa9e4066Sahrens pvd->vdev_children = newc; 283fa9e4066Sahrens } 284fa9e4066Sahrens 285fa9e4066Sahrens /* 286fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 287fa9e4066Sahrens */ 28888ecc943SGeorge Wilson vdev_t * 289fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 290fa9e4066Sahrens { 291fa9e4066Sahrens vdev_t *vd; 292fa9e4066Sahrens 293fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 294fa9e4066Sahrens 2950e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2960e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2970e34b6a7Sbonwick spa->spa_root_vdev = vd; 298e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 2990e34b6a7Sbonwick } 3000e34b6a7Sbonwick 30188ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 3020e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 3030e34b6a7Sbonwick /* 3040e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3050e34b6a7Sbonwick * which must be unique among all pools. 3060e34b6a7Sbonwick */ 3071195e687SMark J Musante guid = spa_generate_guid(NULL); 3080e34b6a7Sbonwick } else { 3090e34b6a7Sbonwick /* 3100e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3110e34b6a7Sbonwick */ 3121195e687SMark J Musante guid = spa_generate_guid(spa); 3130e34b6a7Sbonwick } 3140e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3150e34b6a7Sbonwick } 3160e34b6a7Sbonwick 317fa9e4066Sahrens vd->vdev_spa = spa; 318fa9e4066Sahrens vd->vdev_id = id; 319fa9e4066Sahrens vd->vdev_guid = guid; 320fa9e4066Sahrens vd->vdev_guid_sum = guid; 321fa9e4066Sahrens vd->vdev_ops = ops; 322fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 32388ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 324fa9e4066Sahrens 325fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3265ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 327e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3288ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3290713e232SGeorge Wilson vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 3308ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3318ad4d6ddSJeff Bonwick } 332fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 333fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 334fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 335fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 336fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3373d7072f8Seschrock vdev_queue_init(vd); 3383d7072f8Seschrock vdev_cache_init(vd); 339fa9e4066Sahrens 340fa9e4066Sahrens return (vd); 341fa9e4066Sahrens } 342fa9e4066Sahrens 343fa9e4066Sahrens /* 344fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 345fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 346fa9e4066Sahrens * different for each case. 347fa9e4066Sahrens */ 34899653d4eSeschrock int 34999653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 35099653d4eSeschrock int alloctype) 351fa9e4066Sahrens { 352fa9e4066Sahrens vdev_ops_t *ops; 353fa9e4066Sahrens char *type; 3548654d025Sperrin uint64_t guid = 0, islog, nparity; 355fa9e4066Sahrens vdev_t *vd; 356fa9e4066Sahrens 357e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 358fa9e4066Sahrens 359fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 360be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 361fa9e4066Sahrens 362fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 363be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 364fa9e4066Sahrens 365fa9e4066Sahrens /* 366fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 367fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 368fa9e4066Sahrens */ 369fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 370fa9e4066Sahrens uint64_t label_id; 371fa9e4066Sahrens 372fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 373fa9e4066Sahrens label_id != id) 374be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 375fa9e4066Sahrens 376fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 377be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 37899653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 37999653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 380be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 381fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 382fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 383be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 38421ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 38521ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 386be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 387fa9e4066Sahrens } 388fa9e4066Sahrens 38999653d4eSeschrock /* 39099653d4eSeschrock * The first allocated vdev must be of type 'root'. 39199653d4eSeschrock */ 39299653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 393be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 39499653d4eSeschrock 3958654d025Sperrin /* 3968654d025Sperrin * Determine whether we're a log vdev. 3978654d025Sperrin */ 3988654d025Sperrin islog = 0; 3998654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 400990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 401be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 402fa9e4066Sahrens 40388ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 404be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 40588ecc943SGeorge Wilson 40699653d4eSeschrock /* 4078654d025Sperrin * Set the nparity property for RAID-Z vdevs. 40899653d4eSeschrock */ 4098654d025Sperrin nparity = -1ULL; 41099653d4eSeschrock if (ops == &vdev_raidz_ops) { 41199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4128654d025Sperrin &nparity) == 0) { 413b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 414be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 41599653d4eSeschrock /* 416f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 417f94275ceSAdam Leventhal * device. 41899653d4eSeschrock */ 419f94275ceSAdam Leventhal if (nparity > 1 && 420f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 421be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 422f94275ceSAdam Leventhal if (nparity > 2 && 423f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 424be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 42599653d4eSeschrock } else { 42699653d4eSeschrock /* 42799653d4eSeschrock * We require the parity to be specified for SPAs that 42899653d4eSeschrock * support multiple parity levels. 42999653d4eSeschrock */ 430f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 431be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 43299653d4eSeschrock /* 43399653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 43499653d4eSeschrock */ 4358654d025Sperrin nparity = 1; 43699653d4eSeschrock } 43799653d4eSeschrock } else { 4388654d025Sperrin nparity = 0; 43999653d4eSeschrock } 4408654d025Sperrin ASSERT(nparity != -1ULL); 4418654d025Sperrin 4428654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4438654d025Sperrin 4448654d025Sperrin vd->vdev_islog = islog; 4458654d025Sperrin vd->vdev_nparity = nparity; 4468654d025Sperrin 4478654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4488654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4498654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4508654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4518654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4528654d025Sperrin &vd->vdev_physpath) == 0) 4538654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4546809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4556809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 45699653d4eSeschrock 457afefbcddSeschrock /* 458afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 459afefbcddSeschrock * as -1. 460afefbcddSeschrock */ 461afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 462afefbcddSeschrock &vd->vdev_wholedisk) != 0) 463afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 464afefbcddSeschrock 465ea8dc4b6Seschrock /* 466ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 467ea8dc4b6Seschrock * was not present at the time of import. 468ea8dc4b6Seschrock */ 4696809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4706809eb4eSEric Schrock &vd->vdev_not_present); 471ea8dc4b6Seschrock 472ecc2d604Sbonwick /* 473ecc2d604Sbonwick * Get the alignment requirement. 474ecc2d604Sbonwick */ 475ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 476ecc2d604Sbonwick 47788ecc943SGeorge Wilson /* 47888ecc943SGeorge Wilson * Retrieve the vdev creation time. 47988ecc943SGeorge Wilson */ 48088ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 48188ecc943SGeorge Wilson &vd->vdev_crtxg); 48288ecc943SGeorge Wilson 483fa9e4066Sahrens /* 484fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 485fa9e4066Sahrens */ 4861195e687SMark J Musante if (parent && !parent->vdev_parent && 4871195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 488fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 489fa9e4066Sahrens &vd->vdev_ms_array); 490fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 491fa9e4066Sahrens &vd->vdev_ms_shift); 492fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 493fa9e4066Sahrens &vd->vdev_asize); 4943f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 4953f9d6ad7SLin Ling &vd->vdev_removing); 496fa9e4066Sahrens } 497fa9e4066Sahrens 498cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 499a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 5009f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 5011195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 5029f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 503a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 504a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 505a1521560SJeff Bonwick } 506a1521560SJeff Bonwick 507fa9e4066Sahrens /* 5083d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 509fa9e4066Sahrens */ 510c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 51121ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 51221ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 513c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 514c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5150713e232SGeorge Wilson &vd->vdev_dtl_object); 516c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 517c5904d13Seschrock &vd->vdev_unspare); 518c5904d13Seschrock } 51921ecdf64SLin Ling 52021ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 52121ecdf64SLin Ling uint64_t spare = 0; 52221ecdf64SLin Ling 52321ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 52421ecdf64SLin Ling &spare) == 0 && spare) 52521ecdf64SLin Ling spa_spare_add(vd); 52621ecdf64SLin Ling } 52721ecdf64SLin Ling 528ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 529ecc2d604Sbonwick &vd->vdev_offline); 530c5904d13Seschrock 531b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 532b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 533cb04b873SMark J Musante 5343d7072f8Seschrock /* 5353d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5363d7072f8Seschrock * state, as the diagnosis made on another system may not be 537069f55e2SEric Schrock * valid in the current context. Local vdevs will 538069f55e2SEric Schrock * remain in the faulted state. 5393d7072f8Seschrock */ 540b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5413d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5423d7072f8Seschrock &vd->vdev_faulted); 5433d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5443d7072f8Seschrock &vd->vdev_degraded); 5453d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5463d7072f8Seschrock &vd->vdev_removed); 547069f55e2SEric Schrock 548069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 549069f55e2SEric Schrock char *aux; 550069f55e2SEric Schrock 551069f55e2SEric Schrock vd->vdev_label_aux = 552069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 553069f55e2SEric Schrock if (nvlist_lookup_string(nv, 554069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 555069f55e2SEric Schrock strcmp(aux, "external") == 0) 556069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 557069f55e2SEric Schrock } 5583d7072f8Seschrock } 559fa9e4066Sahrens } 560fa9e4066Sahrens 561fa9e4066Sahrens /* 562fa9e4066Sahrens * Add ourselves to the parent's list of children. 563fa9e4066Sahrens */ 564fa9e4066Sahrens vdev_add_child(parent, vd); 565fa9e4066Sahrens 56699653d4eSeschrock *vdp = vd; 56799653d4eSeschrock 56899653d4eSeschrock return (0); 569fa9e4066Sahrens } 570fa9e4066Sahrens 571fa9e4066Sahrens void 572fa9e4066Sahrens vdev_free(vdev_t *vd) 573fa9e4066Sahrens { 5743d7072f8Seschrock spa_t *spa = vd->vdev_spa; 575fa9e4066Sahrens 576fa9e4066Sahrens /* 577fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 578fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 579fa9e4066Sahrens */ 580fa9e4066Sahrens vdev_close(vd); 581fa9e4066Sahrens 582e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 583b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 584fa9e4066Sahrens 585fa9e4066Sahrens /* 586fa9e4066Sahrens * Free all children. 587fa9e4066Sahrens */ 588573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 589fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 590fa9e4066Sahrens 591fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 592fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 593fa9e4066Sahrens 594fa9e4066Sahrens /* 595fa9e4066Sahrens * Discard allocation state. 596fa9e4066Sahrens */ 597a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 598fa9e4066Sahrens vdev_metaslab_fini(vd); 599a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 600a1521560SJeff Bonwick } 601fa9e4066Sahrens 602fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 603fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 604fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 605fa9e4066Sahrens 606fa9e4066Sahrens /* 607fa9e4066Sahrens * Remove this vdev from its parent's child list. 608fa9e4066Sahrens */ 609fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 610fa9e4066Sahrens 611fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 612fa9e4066Sahrens 6133d7072f8Seschrock /* 6143d7072f8Seschrock * Clean up vdev structure. 6153d7072f8Seschrock */ 6163d7072f8Seschrock vdev_queue_fini(vd); 6173d7072f8Seschrock vdev_cache_fini(vd); 6183d7072f8Seschrock 6193d7072f8Seschrock if (vd->vdev_path) 6203d7072f8Seschrock spa_strfree(vd->vdev_path); 6213d7072f8Seschrock if (vd->vdev_devid) 6223d7072f8Seschrock spa_strfree(vd->vdev_devid); 6233d7072f8Seschrock if (vd->vdev_physpath) 6243d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6256809eb4eSEric Schrock if (vd->vdev_fru) 6266809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6273d7072f8Seschrock 6283d7072f8Seschrock if (vd->vdev_isspare) 6293d7072f8Seschrock spa_spare_remove(vd); 630fa94a07fSbrendan if (vd->vdev_isl2cache) 631fa94a07fSbrendan spa_l2cache_remove(vd); 6323d7072f8Seschrock 6333d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6343d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6358ad4d6ddSJeff Bonwick 6363d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6370713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 6388ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6390713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 6400713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 6418ad4d6ddSJeff Bonwick } 6423d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6438ad4d6ddSJeff Bonwick 6443d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6453d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 646e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6473d7072f8Seschrock 6483d7072f8Seschrock if (vd == spa->spa_root_vdev) 6493d7072f8Seschrock spa->spa_root_vdev = NULL; 6503d7072f8Seschrock 6513d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 652fa9e4066Sahrens } 653fa9e4066Sahrens 654fa9e4066Sahrens /* 655fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 656fa9e4066Sahrens */ 657fa9e4066Sahrens static void 658fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 659fa9e4066Sahrens { 660fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 661fa9e4066Sahrens metaslab_t *msp; 662fa9e4066Sahrens vdev_t *vd; 663fa9e4066Sahrens int t; 664fa9e4066Sahrens 665fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 666fa9e4066Sahrens 667fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 668fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 669fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 670fa9e4066Sahrens 671fa9e4066Sahrens svd->vdev_ms_array = 0; 672fa9e4066Sahrens svd->vdev_ms_shift = 0; 673fa9e4066Sahrens svd->vdev_ms_count = 0; 674fa9e4066Sahrens 675cd0837ccSGeorge Wilson if (tvd->vdev_mg) 676cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 677fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 678fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 679fa9e4066Sahrens 680fa9e4066Sahrens svd->vdev_mg = NULL; 681fa9e4066Sahrens svd->vdev_ms = NULL; 682ecc2d604Sbonwick 683ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 684ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 685fa9e4066Sahrens 686fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 687fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 68899653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 689fa9e4066Sahrens 690fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 691fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 69299653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 693fa9e4066Sahrens 694fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 695fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 696fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 697fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 698fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 699fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 700fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 701fa9e4066Sahrens } 702fa9e4066Sahrens 703e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 704fa9e4066Sahrens vdev_config_clean(svd); 705fa9e4066Sahrens vdev_config_dirty(tvd); 706fa9e4066Sahrens } 707fa9e4066Sahrens 708e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 709e14bb325SJeff Bonwick vdev_state_clean(svd); 710e14bb325SJeff Bonwick vdev_state_dirty(tvd); 711e14bb325SJeff Bonwick } 712e14bb325SJeff Bonwick 71399653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 71499653d4eSeschrock svd->vdev_deflate_ratio = 0; 7158654d025Sperrin 7168654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7178654d025Sperrin svd->vdev_islog = 0; 718fa9e4066Sahrens } 719fa9e4066Sahrens 720fa9e4066Sahrens static void 721fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 722fa9e4066Sahrens { 723fa9e4066Sahrens if (vd == NULL) 724fa9e4066Sahrens return; 725fa9e4066Sahrens 726fa9e4066Sahrens vd->vdev_top = tvd; 727fa9e4066Sahrens 728573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 729fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 730fa9e4066Sahrens } 731fa9e4066Sahrens 732fa9e4066Sahrens /* 733fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 734fa9e4066Sahrens */ 735fa9e4066Sahrens vdev_t * 736fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 737fa9e4066Sahrens { 738fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 739fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 740fa9e4066Sahrens vdev_t *mvd; 741fa9e4066Sahrens 742e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 743fa9e4066Sahrens 744fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 745ecc2d604Sbonwick 746ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 747573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 7484263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 749ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 750ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 75188ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 752ecc2d604Sbonwick 753fa9e4066Sahrens vdev_remove_child(pvd, cvd); 754fa9e4066Sahrens vdev_add_child(pvd, mvd); 755fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 756fa9e4066Sahrens vdev_add_child(mvd, cvd); 757fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 758fa9e4066Sahrens 759fa9e4066Sahrens if (mvd == mvd->vdev_top) 760fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 761fa9e4066Sahrens 762fa9e4066Sahrens return (mvd); 763fa9e4066Sahrens } 764fa9e4066Sahrens 765fa9e4066Sahrens /* 766fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 767fa9e4066Sahrens */ 768fa9e4066Sahrens void 769fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 770fa9e4066Sahrens { 771fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 772fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 773fa9e4066Sahrens 774e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 775fa9e4066Sahrens 776fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 777fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 77899653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 77999653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 780ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 781fa9e4066Sahrens 782fa9e4066Sahrens vdev_remove_child(mvd, cvd); 783fa9e4066Sahrens vdev_remove_child(pvd, mvd); 7848ad4d6ddSJeff Bonwick 78599653d4eSeschrock /* 786e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 787e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 788e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 789e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 79099653d4eSeschrock */ 7918ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 7928ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 7931195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 7948ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 7958ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 7968ad4d6ddSJeff Bonwick } 797e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 798e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 799fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 800fa9e4066Sahrens 801fa9e4066Sahrens if (cvd == cvd->vdev_top) 802fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 803fa9e4066Sahrens 804fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 805fa9e4066Sahrens vdev_free(mvd); 806fa9e4066Sahrens } 807fa9e4066Sahrens 808ea8dc4b6Seschrock int 809fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 810fa9e4066Sahrens { 811fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 812ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 813ecc2d604Sbonwick uint64_t m; 814fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 815fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 816ecc2d604Sbonwick metaslab_t **mspp; 817ecc2d604Sbonwick int error; 818fa9e4066Sahrens 819a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 820a1521560SJeff Bonwick 82188ecc943SGeorge Wilson /* 82288ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 82388ecc943SGeorge Wilson */ 82488ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8250e34b6a7Sbonwick return (0); 8260e34b6a7Sbonwick 82788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 82888ecc943SGeorge Wilson 829e6ca193dSGeorge Wilson /* 830e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 831e6ca193dSGeorge Wilson * in 128k (1 << 17) because it is the current "typical" blocksize. 832e6ca193dSGeorge Wilson * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 833e6ca193dSGeorge Wilson * or we will inconsistently account for existing bp's. 834e6ca193dSGeorge Wilson */ 835e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 836e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 837e6ca193dSGeorge Wilson 838fa9e4066Sahrens ASSERT(oldc <= newc); 839fa9e4066Sahrens 840ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 841fa9e4066Sahrens 842ecc2d604Sbonwick if (oldc != 0) { 843ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 844ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 845ecc2d604Sbonwick } 846fa9e4066Sahrens 847ecc2d604Sbonwick vd->vdev_ms = mspp; 848ecc2d604Sbonwick vd->vdev_ms_count = newc; 849fa9e4066Sahrens 850ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 8510713e232SGeorge Wilson uint64_t object = 0; 8520713e232SGeorge Wilson 853ecc2d604Sbonwick if (txg == 0) { 854ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8557bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8567bfdf011SNeil Perrin DMU_READ_PREFETCH); 857ecc2d604Sbonwick if (error) 858ecc2d604Sbonwick return (error); 859fa9e4066Sahrens } 8600713e232SGeorge Wilson vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg); 861fa9e4066Sahrens } 862fa9e4066Sahrens 863a1521560SJeff Bonwick if (txg == 0) 864a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 865a1521560SJeff Bonwick 8663f9d6ad7SLin Ling /* 8673f9d6ad7SLin Ling * If the vdev is being removed we don't activate 8683f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 8693f9d6ad7SLin Ling * allocations are performed on this device. 8703f9d6ad7SLin Ling */ 8713f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 872a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 873a1521560SJeff Bonwick 874a1521560SJeff Bonwick if (txg == 0) 875a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 876a1521560SJeff Bonwick 877ea8dc4b6Seschrock return (0); 878fa9e4066Sahrens } 879fa9e4066Sahrens 880fa9e4066Sahrens void 881fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 882fa9e4066Sahrens { 883fa9e4066Sahrens uint64_t m; 884fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 885fa9e4066Sahrens 886fa9e4066Sahrens if (vd->vdev_ms != NULL) { 887a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 8880713e232SGeorge Wilson for (m = 0; m < count; m++) { 8890713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 8900713e232SGeorge Wilson 8910713e232SGeorge Wilson if (msp != NULL) 8920713e232SGeorge Wilson metaslab_fini(msp); 8930713e232SGeorge Wilson } 894fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 895fa9e4066Sahrens vd->vdev_ms = NULL; 896fa9e4066Sahrens } 897fa9e4066Sahrens } 898fa9e4066Sahrens 899e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 900e14bb325SJeff Bonwick boolean_t vps_readable; 901e14bb325SJeff Bonwick boolean_t vps_writeable; 902e14bb325SJeff Bonwick int vps_flags; 903e14bb325SJeff Bonwick } vdev_probe_stats_t; 904e14bb325SJeff Bonwick 905e14bb325SJeff Bonwick static void 906e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9070a4e9518Sgw { 9088ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 909a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 910e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 911a3f829aeSBill Moore 912a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 913e14bb325SJeff Bonwick 914e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 915e14bb325SJeff Bonwick if (zio->io_error == 0) 916e14bb325SJeff Bonwick vps->vps_readable = 1; 9178ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 918a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 919e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 920e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 921e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 922e14bb325SJeff Bonwick } else { 923e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 924e14bb325SJeff Bonwick } 925e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 926e14bb325SJeff Bonwick if (zio->io_error == 0) 927e14bb325SJeff Bonwick vps->vps_writeable = 1; 928e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 929e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 930a3f829aeSBill Moore zio_t *pio; 931e14bb325SJeff Bonwick 932e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 933e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 934e14bb325SJeff Bonwick 935e14bb325SJeff Bonwick if (vdev_readable(vd) && 9368ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 937e14bb325SJeff Bonwick zio->io_error = 0; 938e14bb325SJeff Bonwick } else { 939e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 940e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9418ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 942be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 943e14bb325SJeff Bonwick } 944a3f829aeSBill Moore 945a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 946a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 947a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 948a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 949a3f829aeSBill Moore 950a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 951a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 952be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 953a3f829aeSBill Moore 954e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 955e14bb325SJeff Bonwick } 956e14bb325SJeff Bonwick } 9570a4e9518Sgw 958e14bb325SJeff Bonwick /* 959f7170741SWill Andrews * Determine whether this device is accessible. 960f7170741SWill Andrews * 961f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 962f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 963f7170741SWill Andrews * a VTOC. 964e14bb325SJeff Bonwick */ 965e14bb325SJeff Bonwick zio_t * 966a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 967e14bb325SJeff Bonwick { 968e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 969a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 970a3f829aeSBill Moore zio_t *pio; 971a3f829aeSBill Moore 972a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9730a4e9518Sgw 974a3f829aeSBill Moore /* 975a3f829aeSBill Moore * Don't probe the probe. 976a3f829aeSBill Moore */ 977a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 978a3f829aeSBill Moore return (NULL); 979e14bb325SJeff Bonwick 980a3f829aeSBill Moore /* 981a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 982a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 983a3f829aeSBill Moore * this vdev will become parents of the probe io. 984a3f829aeSBill Moore */ 985a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 986e14bb325SJeff Bonwick 987a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 988a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 989a3f829aeSBill Moore 990a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 991a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 9928956713aSEric Schrock ZIO_FLAG_TRYHARD; 993a3f829aeSBill Moore 994a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 995a3f829aeSBill Moore /* 996a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 997a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 998a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 999a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1000a3f829aeSBill Moore * any zio looking at these values can assume that 1001a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1002a3f829aeSBill Moore * important because when a device has intermittent 1003a3f829aeSBill Moore * connectivity problems, we want to ensure that 1004a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1005a3f829aeSBill Moore * the zio (EIO). 1006a3f829aeSBill Moore * 1007a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1008a3f829aeSBill Moore * values so the probe can reevaluate from first 1009a3f829aeSBill Moore * principles. 1010a3f829aeSBill Moore */ 1011a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1012a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1013a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1014a3f829aeSBill Moore } 1015a3f829aeSBill Moore 1016a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1017a3f829aeSBill Moore vdev_probe_done, vps, 1018a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1019a3f829aeSBill Moore 102098d1cbfeSGeorge Wilson /* 102198d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 102298d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 102398d1cbfeSGeorge Wilson */ 1024a3f829aeSBill Moore if (zio != NULL) { 1025a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1026a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1027a3f829aeSBill Moore } 1028e14bb325SJeff Bonwick } 1029e14bb325SJeff Bonwick 1030a3f829aeSBill Moore if (zio != NULL) 1031a3f829aeSBill Moore zio_add_child(zio, pio); 1032e14bb325SJeff Bonwick 1033a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1034e14bb325SJeff Bonwick 1035a3f829aeSBill Moore if (vps == NULL) { 1036a3f829aeSBill Moore ASSERT(zio != NULL); 1037a3f829aeSBill Moore return (NULL); 1038a3f829aeSBill Moore } 1039e14bb325SJeff Bonwick 1040e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1041a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1042e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1043f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1044f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1045e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1046e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1047e14bb325SJeff Bonwick } 1048e14bb325SJeff Bonwick 1049a3f829aeSBill Moore if (zio == NULL) 1050a3f829aeSBill Moore return (pio); 1051a3f829aeSBill Moore 1052a3f829aeSBill Moore zio_nowait(pio); 1053a3f829aeSBill Moore return (NULL); 10540a4e9518Sgw } 10550a4e9518Sgw 1056f64c0e34SEric Taylor static void 1057f64c0e34SEric Taylor vdev_open_child(void *arg) 1058f64c0e34SEric Taylor { 1059f64c0e34SEric Taylor vdev_t *vd = arg; 1060f64c0e34SEric Taylor 1061f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1062f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1063f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1064f64c0e34SEric Taylor } 1065f64c0e34SEric Taylor 1066681d9761SEric Taylor boolean_t 1067681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1068681d9761SEric Taylor { 1069681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1070681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1071681d9761SEric Taylor return (B_TRUE); 1072681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1073681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1074681d9761SEric Taylor return (B_TRUE); 1075681d9761SEric Taylor return (B_FALSE); 1076681d9761SEric Taylor } 1077681d9761SEric Taylor 1078f64c0e34SEric Taylor void 1079f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1080f64c0e34SEric Taylor { 1081f64c0e34SEric Taylor taskq_t *tq; 1082f64c0e34SEric Taylor int children = vd->vdev_children; 1083f64c0e34SEric Taylor 1084681d9761SEric Taylor /* 1085681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1086681d9761SEric Taylor * in a single thread so that the same thread holds the 1087681d9761SEric Taylor * spa_namespace_lock 1088681d9761SEric Taylor */ 1089681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1090681d9761SEric Taylor for (int c = 0; c < children; c++) 1091681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1092681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1093681d9761SEric Taylor return; 1094681d9761SEric Taylor } 1095f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1096f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1097f64c0e34SEric Taylor 1098f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1099f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1100f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1101f64c0e34SEric Taylor 1102f64c0e34SEric Taylor taskq_destroy(tq); 1103f64c0e34SEric Taylor } 1104f64c0e34SEric Taylor 1105fa9e4066Sahrens /* 1106fa9e4066Sahrens * Prepare a virtual device for access. 1107fa9e4066Sahrens */ 1108fa9e4066Sahrens int 1109fa9e4066Sahrens vdev_open(vdev_t *vd) 1110fa9e4066Sahrens { 11118ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1112fa9e4066Sahrens int error; 1113fa9e4066Sahrens uint64_t osize = 0; 11144263d13fSGeorge Wilson uint64_t max_osize = 0; 11154263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1116ecc2d604Sbonwick uint64_t ashift = 0; 1117fa9e4066Sahrens 1118f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1119f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1120fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1121fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1122fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1123fa9e4066Sahrens 1124fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1125e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1126e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1127573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1128fa9e4066Sahrens 1129069f55e2SEric Schrock /* 1130069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1131069f55e2SEric Schrock * faulted, bail out of the open. 1132069f55e2SEric Schrock */ 11333d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11343d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1135069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1136069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11373d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1138069f55e2SEric Schrock vd->vdev_label_aux); 1139be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 11403d7072f8Seschrock } else if (vd->vdev_offline) { 1141fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1142ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1143be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1144fa9e4066Sahrens } 1145fa9e4066Sahrens 11464263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1147fa9e4066Sahrens 1148095bcd66SGeorge Wilson /* 1149095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1150095bcd66SGeorge Wilson * the vdev on error. 1151095bcd66SGeorge Wilson */ 1152095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1153ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11548956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1155ea8dc4b6Seschrock 1156fa9e4066Sahrens if (error) { 11573d7072f8Seschrock if (vd->vdev_removed && 11583d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11593d7072f8Seschrock vd->vdev_removed = B_FALSE; 11603d7072f8Seschrock 1161ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1162fa9e4066Sahrens vd->vdev_stat.vs_aux); 1163fa9e4066Sahrens return (error); 1164fa9e4066Sahrens } 1165fa9e4066Sahrens 11663d7072f8Seschrock vd->vdev_removed = B_FALSE; 11673d7072f8Seschrock 1168096d22d4SEric Schrock /* 1169096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1170096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1171096d22d4SEric Schrock */ 1172096d22d4SEric Schrock if (vd->vdev_faulted) { 1173096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1174096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1175096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1176096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1177096d22d4SEric Schrock vd->vdev_label_aux); 1178be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1179096d22d4SEric Schrock } 1180096d22d4SEric Schrock 11813d7072f8Seschrock if (vd->vdev_degraded) { 11823d7072f8Seschrock ASSERT(vd->vdev_children == 0); 11833d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 11843d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 11853d7072f8Seschrock } else { 1186069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 11873d7072f8Seschrock } 1188fa9e4066Sahrens 118988ecc943SGeorge Wilson /* 119088ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 119188ecc943SGeorge Wilson */ 119288ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 119388ecc943SGeorge Wilson return (0); 119488ecc943SGeorge Wilson 1195573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1196ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1197ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1198ea8dc4b6Seschrock VDEV_AUX_NONE); 1199ea8dc4b6Seschrock break; 1200ea8dc4b6Seschrock } 1201573ca77eSGeorge Wilson } 1202fa9e4066Sahrens 1203fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 12044263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1205fa9e4066Sahrens 1206fa9e4066Sahrens if (vd->vdev_children == 0) { 1207fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1208ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1209ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1210be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1211fa9e4066Sahrens } 1212fa9e4066Sahrens psize = osize; 1213fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 12144263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 12154263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1216fa9e4066Sahrens } else { 1217ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1218fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1219ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1220ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1221be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1222fa9e4066Sahrens } 1223fa9e4066Sahrens psize = 0; 1224fa9e4066Sahrens asize = osize; 12254263d13fSGeorge Wilson max_asize = max_osize; 1226fa9e4066Sahrens } 1227fa9e4066Sahrens 1228fa9e4066Sahrens vd->vdev_psize = psize; 1229fa9e4066Sahrens 1230573ca77eSGeorge Wilson /* 1231573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1232573ca77eSGeorge Wilson */ 1233573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1234573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1235573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1236be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1237573ca77eSGeorge Wilson } 1238573ca77eSGeorge Wilson 1239fa9e4066Sahrens if (vd->vdev_asize == 0) { 1240fa9e4066Sahrens /* 1241fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1242ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1243fa9e4066Sahrens */ 1244fa9e4066Sahrens vd->vdev_asize = asize; 12454263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1246ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1247fa9e4066Sahrens } else { 1248fa9e4066Sahrens /* 12492384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 12502384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 12512384d9f8SGeorge Wilson * issue a warning instead. 1252fa9e4066Sahrens */ 12532384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 12542384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 12552384d9f8SGeorge Wilson cmn_err(CE_WARN, 12562384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 12572384d9f8SGeorge Wilson "larger than the pool's alignment\n", 12582384d9f8SGeorge Wilson vd->vdev_path); 1259fa9e4066Sahrens } 12604263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1261573ca77eSGeorge Wilson } 1262fa9e4066Sahrens 1263573ca77eSGeorge Wilson /* 1264573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1265573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1266573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1267573ca77eSGeorge Wilson */ 1268573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1269573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1270573ca77eSGeorge Wilson vd->vdev_asize = asize; 1271fa9e4066Sahrens 1272573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1273fa9e4066Sahrens 12740a4e9518Sgw /* 12750a4e9518Sgw * Ensure we can issue some IO before declaring the 12760a4e9518Sgw * vdev open for business. 12770a4e9518Sgw */ 1278e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1279e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 128098d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 128198d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 12820a4e9518Sgw return (error); 12830a4e9518Sgw } 12840a4e9518Sgw 1285088f3894Sahrens /* 1286088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 12878ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 12888ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1289088f3894Sahrens */ 12908ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 12918ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 12928ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1293088f3894Sahrens 1294fa9e4066Sahrens return (0); 1295fa9e4066Sahrens } 1296fa9e4066Sahrens 1297560e6e96Seschrock /* 1298560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1299560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 13003d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1301560e6e96Seschrock * 1302d7f601efSGeorge Wilson * If 'strict' is false ignore the spa guid check. This is necessary because 1303d7f601efSGeorge Wilson * if the machine crashed during a re-guid the new guid might have been written 1304d7f601efSGeorge Wilson * to all of the vdev labels, but not the cached config. The strict check 1305d7f601efSGeorge Wilson * will be performed when the pool is opened again using the mos config. 1306d7f601efSGeorge Wilson * 1307560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1308560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1309560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1310560e6e96Seschrock * will be updated but the function will return 0. 1311560e6e96Seschrock */ 1312560e6e96Seschrock int 1313d7f601efSGeorge Wilson vdev_validate(vdev_t *vd, boolean_t strict) 1314560e6e96Seschrock { 1315560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1316560e6e96Seschrock nvlist_t *label; 13171195e687SMark J Musante uint64_t guid = 0, top_guid; 1318560e6e96Seschrock uint64_t state; 1319560e6e96Seschrock 1320573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1321d7f601efSGeorge Wilson if (vdev_validate(vd->vdev_child[c], strict) != 0) 1322be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1323560e6e96Seschrock 1324b5989ec7Seschrock /* 1325b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1326b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1327b5989ec7Seschrock * overwrite the previous state. 1328b5989ec7Seschrock */ 1329e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13301195e687SMark J Musante uint64_t aux_guid = 0; 13311195e687SMark J Musante nvlist_t *nvl; 1332bda88194SGeorge Wilson uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1333bda88194SGeorge Wilson spa_last_synced_txg(spa) : -1ULL; 1334560e6e96Seschrock 1335dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1336560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1337560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1338560e6e96Seschrock return (0); 1339560e6e96Seschrock } 1340560e6e96Seschrock 13411195e687SMark J Musante /* 13421195e687SMark J Musante * Determine if this vdev has been split off into another 13431195e687SMark J Musante * pool. If so, then refuse to open it. 13441195e687SMark J Musante */ 13451195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 13461195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 13471195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 13481195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 13491195e687SMark J Musante nvlist_free(label); 13501195e687SMark J Musante return (0); 13511195e687SMark J Musante } 13521195e687SMark J Musante 1353d7f601efSGeorge Wilson if (strict && (nvlist_lookup_uint64(label, 1354d7f601efSGeorge Wilson ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1355d7f601efSGeorge Wilson guid != spa_guid(spa))) { 1356560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1357560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1358560e6e96Seschrock nvlist_free(label); 1359560e6e96Seschrock return (0); 1360560e6e96Seschrock } 1361560e6e96Seschrock 13621195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 13631195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 13641195e687SMark J Musante &aux_guid) != 0) 13651195e687SMark J Musante aux_guid = 0; 13661195e687SMark J Musante 1367e14bb325SJeff Bonwick /* 1368e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1369e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1370e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1371e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1372e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1373e14bb325SJeff Bonwick * safely compare to that instead. 13741195e687SMark J Musante * 13751195e687SMark J Musante * If we split this vdev off instead, then we also check the 13761195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 13771195e687SMark J Musante * corrupt if it is partway through a split operation. 1378e14bb325SJeff Bonwick */ 1379560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1380e14bb325SJeff Bonwick &guid) != 0 || 1381e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1382e14bb325SJeff Bonwick &top_guid) != 0 || 13831195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1384e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1385560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1386560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1387560e6e96Seschrock nvlist_free(label); 1388560e6e96Seschrock return (0); 1389560e6e96Seschrock } 1390560e6e96Seschrock 1391560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1392560e6e96Seschrock &state) != 0) { 1393560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1394560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1395560e6e96Seschrock nvlist_free(label); 1396560e6e96Seschrock return (0); 1397560e6e96Seschrock } 1398560e6e96Seschrock 1399560e6e96Seschrock nvlist_free(label); 1400560e6e96Seschrock 1401bc758434SLin Ling /* 14024b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1403bc758434SLin Ling * state of the pool. 1404bc758434SLin Ling */ 14054b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1406b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1407bc758434SLin Ling state != POOL_STATE_ACTIVE) 1408be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1409560e6e96Seschrock 141051ece835Seschrock /* 141151ece835Seschrock * If we were able to open and validate a vdev that was 141251ece835Seschrock * previously marked permanently unavailable, clear that state 141351ece835Seschrock * now. 141451ece835Seschrock */ 141551ece835Seschrock if (vd->vdev_not_present) 141651ece835Seschrock vd->vdev_not_present = 0; 141751ece835Seschrock } 1418560e6e96Seschrock 1419560e6e96Seschrock return (0); 1420560e6e96Seschrock } 1421560e6e96Seschrock 1422fa9e4066Sahrens /* 1423fa9e4066Sahrens * Close a virtual device. 1424fa9e4066Sahrens */ 1425fa9e4066Sahrens void 1426fa9e4066Sahrens vdev_close(vdev_t *vd) 1427fa9e4066Sahrens { 14288ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1429095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14308ad4d6ddSJeff Bonwick 14318ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14328ad4d6ddSJeff Bonwick 14331195e687SMark J Musante /* 14341195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14351195e687SMark J Musante * going offline. 14361195e687SMark J Musante */ 1437095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14381195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1439095bcd66SGeorge Wilson 1440fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1441fa9e4066Sahrens 14423d7072f8Seschrock vdev_cache_purge(vd); 1443fa9e4066Sahrens 1444560e6e96Seschrock /* 1445573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1446560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1447560e6e96Seschrock * it's still faulted. 1448560e6e96Seschrock */ 1449560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1450560e6e96Seschrock 1451fa9e4066Sahrens if (vd->vdev_offline) 1452fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1453fa9e4066Sahrens else 1454fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1455ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1456fa9e4066Sahrens } 1457fa9e4066Sahrens 1458dcba9f3fSGeorge Wilson void 1459dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1460dcba9f3fSGeorge Wilson { 1461dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1462dcba9f3fSGeorge Wilson 1463dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1464dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1465dcba9f3fSGeorge Wilson return; 1466dcba9f3fSGeorge Wilson 1467dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1468dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1469dcba9f3fSGeorge Wilson 1470dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1471dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1472dcba9f3fSGeorge Wilson } 1473dcba9f3fSGeorge Wilson 1474dcba9f3fSGeorge Wilson void 1475dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1476dcba9f3fSGeorge Wilson { 1477dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1478dcba9f3fSGeorge Wilson 1479dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1480dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1481dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1482dcba9f3fSGeorge Wilson 1483dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1484dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1485dcba9f3fSGeorge Wilson } 1486dcba9f3fSGeorge Wilson 1487095bcd66SGeorge Wilson /* 1488095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1489095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1490095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1491095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1492095bcd66SGeorge Wilson */ 1493fa9e4066Sahrens void 1494ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1495fa9e4066Sahrens { 1496ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1497fa9e4066Sahrens 1498e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1499ea8dc4b6Seschrock 15001195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 15011195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1502fa9e4066Sahrens vdev_close(vd); 1503fa9e4066Sahrens (void) vdev_open(vd); 1504fa9e4066Sahrens 150539c23413Seschrock /* 150639c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 150739c23413Seschrock * Otherwise, a device with an invalid label could be successfully 150839c23413Seschrock * opened in response to vdev_reopen(). 150939c23413Seschrock */ 1510c5904d13Seschrock if (vd->vdev_aux) { 1511c5904d13Seschrock (void) vdev_validate_aux(vd); 1512e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 15136809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1514573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1515573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1516c5904d13Seschrock } else { 1517bda88194SGeorge Wilson (void) vdev_validate(vd, B_TRUE); 1518c5904d13Seschrock } 151939c23413Seschrock 1520fa9e4066Sahrens /* 15213d7072f8Seschrock * Reassess parent vdev's health. 1522fa9e4066Sahrens */ 15233d7072f8Seschrock vdev_propagate_state(vd); 1524fa9e4066Sahrens } 1525fa9e4066Sahrens 1526fa9e4066Sahrens int 152799653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1528fa9e4066Sahrens { 1529fa9e4066Sahrens int error; 1530fa9e4066Sahrens 1531fa9e4066Sahrens /* 1532fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1533fa9e4066Sahrens * For a create, however, we want to fail the request if 1534fa9e4066Sahrens * there are any components we can't open. 1535fa9e4066Sahrens */ 1536fa9e4066Sahrens error = vdev_open(vd); 1537fa9e4066Sahrens 1538fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1539fa9e4066Sahrens vdev_close(vd); 1540fa9e4066Sahrens return (error ? error : ENXIO); 1541fa9e4066Sahrens } 1542fa9e4066Sahrens 1543fa9e4066Sahrens /* 15440713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1545fa9e4066Sahrens */ 15460713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 15470713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 154839c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1549fa9e4066Sahrens vdev_close(vd); 1550fa9e4066Sahrens return (error); 1551fa9e4066Sahrens } 1552fa9e4066Sahrens 1553fa9e4066Sahrens return (0); 1554fa9e4066Sahrens } 1555fa9e4066Sahrens 15560e34b6a7Sbonwick void 1557573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1558fa9e4066Sahrens { 1559fa9e4066Sahrens /* 1560*bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1561fa9e4066Sahrens */ 1562*bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1563fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1564fa9e4066Sahrens } 1565fa9e4066Sahrens 1566fa9e4066Sahrens void 1567ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1568fa9e4066Sahrens { 1569ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 157088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1571ecc2d604Sbonwick ASSERT(ISP2(flags)); 1572f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1573fa9e4066Sahrens 1574ecc2d604Sbonwick if (flags & VDD_METASLAB) 1575ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1576ecc2d604Sbonwick 1577ecc2d604Sbonwick if (flags & VDD_DTL) 1578ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1579ecc2d604Sbonwick 1580ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1581fa9e4066Sahrens } 1582fa9e4066Sahrens 15830713e232SGeorge Wilson void 15840713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 15850713e232SGeorge Wilson { 15860713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 15870713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 15880713e232SGeorge Wilson 15890713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 15900713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 15910713e232SGeorge Wilson } 15920713e232SGeorge Wilson 15938ad4d6ddSJeff Bonwick /* 15948ad4d6ddSJeff Bonwick * DTLs. 15958ad4d6ddSJeff Bonwick * 15968ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 15979fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 15988ad4d6ddSJeff Bonwick * 15998ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 16008ad4d6ddSJeff Bonwick * 16018ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 16028ad4d6ddSJeff Bonwick * 16038ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 16048ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 16058ad4d6ddSJeff Bonwick * txgs that was scrubbed. 16068ad4d6ddSJeff Bonwick * 16078ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 16088ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 16098ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 16108ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 16118ad4d6ddSJeff Bonwick * determine whether a device can be detached. 16128ad4d6ddSJeff Bonwick * 16138ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 16148ad4d6ddSJeff Bonwick * either has the data or it doesn't. 16158ad4d6ddSJeff Bonwick * 16168ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 16178ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 16188ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 16198ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 16208ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 16218ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 16228ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 16238ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 16248ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 16258ad4d6ddSJeff Bonwick * 16268ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 16278ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 16288ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 16298ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 16308ad4d6ddSJeff Bonwick */ 1631fa9e4066Sahrens void 16328ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1633fa9e4066Sahrens { 16340713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16358ad4d6ddSJeff Bonwick 16368ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16378ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1638f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 16398ad4d6ddSJeff Bonwick 16400713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16410713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 16420713e232SGeorge Wilson range_tree_add(rt, txg, size); 16430713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1644fa9e4066Sahrens } 1645fa9e4066Sahrens 16468ad4d6ddSJeff Bonwick boolean_t 16478ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1648fa9e4066Sahrens { 16490713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16508ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1651fa9e4066Sahrens 16528ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16538ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1654fa9e4066Sahrens 16550713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16560713e232SGeorge Wilson if (range_tree_space(rt) != 0) 16570713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 16580713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1659fa9e4066Sahrens 1660fa9e4066Sahrens return (dirty); 1661fa9e4066Sahrens } 1662fa9e4066Sahrens 16638ad4d6ddSJeff Bonwick boolean_t 16648ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 16658ad4d6ddSJeff Bonwick { 16660713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16678ad4d6ddSJeff Bonwick boolean_t empty; 16688ad4d6ddSJeff Bonwick 16690713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16700713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 16710713e232SGeorge Wilson mutex_exit(rt->rt_lock); 16728ad4d6ddSJeff Bonwick 16738ad4d6ddSJeff Bonwick return (empty); 16748ad4d6ddSJeff Bonwick } 16758ad4d6ddSJeff Bonwick 1676b4952e17SGeorge Wilson /* 1677b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 1678b4952e17SGeorge Wilson */ 1679b4952e17SGeorge Wilson static uint64_t 1680b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 1681b4952e17SGeorge Wilson { 16820713e232SGeorge Wilson range_seg_t *rs; 1683b4952e17SGeorge Wilson 1684b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 16850713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1686b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1687b4952e17SGeorge Wilson 16880713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 16890713e232SGeorge Wilson return (rs->rs_start - 1); 1690b4952e17SGeorge Wilson } 1691b4952e17SGeorge Wilson 1692b4952e17SGeorge Wilson /* 1693b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 1694b4952e17SGeorge Wilson */ 1695b4952e17SGeorge Wilson static uint64_t 1696b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 1697b4952e17SGeorge Wilson { 16980713e232SGeorge Wilson range_seg_t *rs; 1699b4952e17SGeorge Wilson 1700b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17010713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1702b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1703b4952e17SGeorge Wilson 17040713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17050713e232SGeorge Wilson return (rs->rs_end); 1706b4952e17SGeorge Wilson } 1707b4952e17SGeorge Wilson 1708b4952e17SGeorge Wilson /* 1709b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 1710b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 1711b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 1712b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 1713b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 1714b4952e17SGeorge Wilson * excise the DTLs. 1715b4952e17SGeorge Wilson */ 1716b4952e17SGeorge Wilson static boolean_t 1717b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 1718b4952e17SGeorge Wilson { 1719b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 1720b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1721b4952e17SGeorge Wilson 1722b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 1723b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1724b4952e17SGeorge Wilson 1725b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 17260713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1727b4952e17SGeorge Wilson return (B_TRUE); 1728b4952e17SGeorge Wilson 1729b4952e17SGeorge Wilson /* 1730b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 1731b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 1732b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 1733b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 1734b4952e17SGeorge Wilson * for excision. 1735b4952e17SGeorge Wilson */ 1736b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1737b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1738b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1739b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1740b4952e17SGeorge Wilson return (B_TRUE); 1741b4952e17SGeorge Wilson } 1742b4952e17SGeorge Wilson return (B_FALSE); 1743b4952e17SGeorge Wilson } 1744b4952e17SGeorge Wilson 1745fa9e4066Sahrens /* 1746fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1747fa9e4066Sahrens */ 1748fa9e4066Sahrens void 1749fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1750fa9e4066Sahrens { 1751ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 17528ad4d6ddSJeff Bonwick avl_tree_t reftree; 17538ad4d6ddSJeff Bonwick int minref; 1754fa9e4066Sahrens 17558ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1756fa9e4066Sahrens 17578ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 17588ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 17598ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 17608ad4d6ddSJeff Bonwick 1761b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 17628ad4d6ddSJeff Bonwick return; 17638ad4d6ddSJeff Bonwick 17648ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 17653f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 17663f9d6ad7SLin Ling 1767fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1768b4952e17SGeorge Wilson 1769b4952e17SGeorge Wilson /* 1770b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 1771b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 1772b4952e17SGeorge Wilson * excise regions on vdevs that were available during 1773b4952e17SGeorge Wilson * the entire duration of this scan. 1774b4952e17SGeorge Wilson */ 1775088f3894Sahrens if (scrub_txg != 0 && 17763f9d6ad7SLin Ling (spa->spa_scrub_started || 1777b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1778b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 1779088f3894Sahrens /* 1780088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1781088f3894Sahrens * did it without rebooting, then the scrub dtl 1782088f3894Sahrens * will be valid, so excise the old region and 1783088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1784088f3894Sahrens * dtl as-is if there was an error. 17858ad4d6ddSJeff Bonwick * 17868ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 17878ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 17888ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 17898ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 17908ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 17918ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 17928ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 17938ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 17948ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1795088f3894Sahrens */ 17960713e232SGeorge Wilson space_reftree_create(&reftree); 17970713e232SGeorge Wilson space_reftree_add_map(&reftree, 17980713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 17990713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 18000713e232SGeorge Wilson space_reftree_add_map(&reftree, 18010713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 18020713e232SGeorge Wilson space_reftree_generate_map(&reftree, 18030713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18040713e232SGeorge Wilson space_reftree_destroy(&reftree); 1805fa9e4066Sahrens } 18060713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 18070713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18080713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1809fa9e4066Sahrens if (scrub_done) 18100713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 18110713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 18128ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 18130713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 18148ad4d6ddSJeff Bonwick else 18150713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18160713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1817b4952e17SGeorge Wilson 1818b4952e17SGeorge Wilson /* 1819b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 1820b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 1821b4952e17SGeorge Wilson */ 1822b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 18230713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 18240713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 1825b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 1826b4952e17SGeorge Wilson 1827fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1828088f3894Sahrens 1829ecc2d604Sbonwick if (txg != 0) 1830ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1831fa9e4066Sahrens return; 1832fa9e4066Sahrens } 1833fa9e4066Sahrens 1834fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 18358ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 183699bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 183799bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 18388ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 18398ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 18408ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 18418ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 18428ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 18438ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 18448ad4d6ddSJeff Bonwick else 18458ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 18460713e232SGeorge Wilson space_reftree_create(&reftree); 18478ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 18488ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 18498ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 18500713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 18518ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 18528ad4d6ddSJeff Bonwick } 18530713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 18540713e232SGeorge Wilson space_reftree_destroy(&reftree); 1855fa9e4066Sahrens } 18568ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1857fa9e4066Sahrens } 1858fa9e4066Sahrens 18590713e232SGeorge Wilson int 1860fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1861fa9e4066Sahrens { 1862fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1863ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 18640713e232SGeorge Wilson int error = 0; 1865fa9e4066Sahrens 18660713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 18670713e232SGeorge Wilson ASSERT(!vd->vdev_ishole); 1868fa9e4066Sahrens 18690713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 18700713e232SGeorge Wilson vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 18710713e232SGeorge Wilson if (error) 18720713e232SGeorge Wilson return (error); 18730713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1874fa9e4066Sahrens 18750713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 187688ecc943SGeorge Wilson 18770713e232SGeorge Wilson /* 18780713e232SGeorge Wilson * Now that we've opened the space_map we need to update 18790713e232SGeorge Wilson * the in-core DTL. 18800713e232SGeorge Wilson */ 18810713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 1882ecc2d604Sbonwick 18830713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 18840713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 18850713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1886fa9e4066Sahrens 18870713e232SGeorge Wilson return (error); 18880713e232SGeorge Wilson } 18890713e232SGeorge Wilson 18900713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 18910713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 18920713e232SGeorge Wilson if (error != 0) 18930713e232SGeorge Wilson break; 18940713e232SGeorge Wilson } 1895fa9e4066Sahrens 1896fa9e4066Sahrens return (error); 1897fa9e4066Sahrens } 1898fa9e4066Sahrens 1899fa9e4066Sahrens void 1900fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1901fa9e4066Sahrens { 1902fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 19030713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 1904ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19050713e232SGeorge Wilson range_tree_t *rtsync; 19060713e232SGeorge Wilson kmutex_t rtlock; 1907fa9e4066Sahrens dmu_tx_t *tx; 19080713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 1909fa9e4066Sahrens 191088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 19110713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 191288ecc943SGeorge Wilson 1913fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1914fa9e4066Sahrens 19150713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 19160713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 19170713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 19180713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 19190713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 19200713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1921fa9e4066Sahrens dmu_tx_commit(tx); 1922fa9e4066Sahrens return; 1923fa9e4066Sahrens } 1924fa9e4066Sahrens 19250713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 19260713e232SGeorge Wilson uint64_t new_object; 19270713e232SGeorge Wilson 19280713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 19290713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 19300713e232SGeorge Wilson 19310713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 19320713e232SGeorge Wilson 0, -1ULL, 0, &vd->vdev_dtl_lock)); 19330713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1934fa9e4066Sahrens } 1935fa9e4066Sahrens 19360713e232SGeorge Wilson mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 1937fa9e4066Sahrens 19380713e232SGeorge Wilson rtsync = range_tree_create(NULL, NULL, &rtlock); 1939fa9e4066Sahrens 19400713e232SGeorge Wilson mutex_enter(&rtlock); 1941fa9e4066Sahrens 1942fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 19430713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 1944fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1945fa9e4066Sahrens 19460713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 19470713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 19480713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 1949fa9e4066Sahrens 19500713e232SGeorge Wilson range_tree_destroy(rtsync); 1951fa9e4066Sahrens 19520713e232SGeorge Wilson mutex_exit(&rtlock); 19530713e232SGeorge Wilson mutex_destroy(&rtlock); 1954fa9e4066Sahrens 19550713e232SGeorge Wilson /* 19560713e232SGeorge Wilson * If the object for the space map has changed then dirty 19570713e232SGeorge Wilson * the top level so that we update the config. 19580713e232SGeorge Wilson */ 19590713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 19600713e232SGeorge Wilson zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 19610713e232SGeorge Wilson "new object %llu", txg, spa_name(spa), object, 19620713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 19630713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 19640713e232SGeorge Wilson } 1965fa9e4066Sahrens 1966fa9e4066Sahrens dmu_tx_commit(tx); 19670713e232SGeorge Wilson 19680713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 19690713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 19700713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1971fa9e4066Sahrens } 1972fa9e4066Sahrens 19738ad4d6ddSJeff Bonwick /* 19748ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 19758ad4d6ddSJeff Bonwick * without losing data. 19768ad4d6ddSJeff Bonwick */ 19778ad4d6ddSJeff Bonwick boolean_t 19788ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 19798ad4d6ddSJeff Bonwick { 19808ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 19818ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 19828ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 19838ad4d6ddSJeff Bonwick boolean_t required; 19848ad4d6ddSJeff Bonwick 19858ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 19868ad4d6ddSJeff Bonwick 19878ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 19888ad4d6ddSJeff Bonwick return (B_TRUE); 19898ad4d6ddSJeff Bonwick 19908ad4d6ddSJeff Bonwick /* 19918ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 19928ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 19938ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 19948ad4d6ddSJeff Bonwick */ 19958ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 19968ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 19978ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 19988ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 19998ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 20008ad4d6ddSJeff Bonwick 2001cb04b873SMark J Musante if (!required && zio_injection_enabled) 2002cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2003cb04b873SMark J Musante 20048ad4d6ddSJeff Bonwick return (required); 20058ad4d6ddSJeff Bonwick } 20068ad4d6ddSJeff Bonwick 2007088f3894Sahrens /* 2008088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2009088f3894Sahrens */ 2010088f3894Sahrens boolean_t 2011088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2012088f3894Sahrens { 2013088f3894Sahrens boolean_t needed = B_FALSE; 2014088f3894Sahrens uint64_t thismin = UINT64_MAX; 2015088f3894Sahrens uint64_t thismax = 0; 2016088f3894Sahrens 2017088f3894Sahrens if (vd->vdev_children == 0) { 2018088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 20190713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 20208ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2021088f3894Sahrens 2022b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2023b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2024088f3894Sahrens needed = B_TRUE; 2025088f3894Sahrens } 2026088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2027088f3894Sahrens } else { 20288ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2029088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2030088f3894Sahrens uint64_t cmin, cmax; 2031088f3894Sahrens 2032088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2033088f3894Sahrens thismin = MIN(thismin, cmin); 2034088f3894Sahrens thismax = MAX(thismax, cmax); 2035088f3894Sahrens needed = B_TRUE; 2036088f3894Sahrens } 2037088f3894Sahrens } 2038088f3894Sahrens } 2039088f3894Sahrens 2040088f3894Sahrens if (needed && minp) { 2041088f3894Sahrens *minp = thismin; 2042088f3894Sahrens *maxp = thismax; 2043088f3894Sahrens } 2044088f3894Sahrens return (needed); 2045088f3894Sahrens } 2046088f3894Sahrens 2047560e6e96Seschrock void 2048ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2049fa9e4066Sahrens { 2050fa9e4066Sahrens /* 2051fa9e4066Sahrens * Recursively load all children. 2052fa9e4066Sahrens */ 20538ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2054560e6e96Seschrock vdev_load(vd->vdev_child[c]); 2055fa9e4066Sahrens 2056fa9e4066Sahrens /* 20570e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2058fa9e4066Sahrens */ 205988ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 2060560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2061560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 2062560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2063560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2064fa9e4066Sahrens 2065fa9e4066Sahrens /* 2066fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2067fa9e4066Sahrens */ 2068560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2069560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2070560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2071fa9e4066Sahrens } 2072fa9e4066Sahrens 207399653d4eSeschrock /* 2074fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2075fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2076fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2077fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2078fa94a07fSbrendan * repurposed to another pool. 207999653d4eSeschrock */ 208099653d4eSeschrock int 2081fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 208299653d4eSeschrock { 208399653d4eSeschrock nvlist_t *label; 208499653d4eSeschrock uint64_t guid, version; 208599653d4eSeschrock uint64_t state; 208699653d4eSeschrock 2087e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2088c5904d13Seschrock return (0); 2089c5904d13Seschrock 2090dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 209199653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 209299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 209399653d4eSeschrock return (-1); 209499653d4eSeschrock } 209599653d4eSeschrock 209699653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2097ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 209899653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 209999653d4eSeschrock guid != vd->vdev_guid || 210099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 210199653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 210299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 210399653d4eSeschrock nvlist_free(label); 210499653d4eSeschrock return (-1); 210599653d4eSeschrock } 210699653d4eSeschrock 210799653d4eSeschrock /* 210899653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 210999653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 211099653d4eSeschrock */ 211199653d4eSeschrock nvlist_free(label); 211299653d4eSeschrock return (0); 211399653d4eSeschrock } 211499653d4eSeschrock 211588ecc943SGeorge Wilson void 211688ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 211788ecc943SGeorge Wilson { 211888ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 211988ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 212088ecc943SGeorge Wilson dmu_tx_t *tx; 212188ecc943SGeorge Wilson 212288ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 212388ecc943SGeorge Wilson 212488ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 21252e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 21262e4c9986SGeorge Wilson 21272e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21282e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21292e4c9986SGeorge Wilson 213088ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 213188ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 213288ecc943SGeorge Wilson 21330713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 213488ecc943SGeorge Wilson continue; 213588ecc943SGeorge Wilson 21360713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 21372e4c9986SGeorge Wilson /* 21382e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 21392e4c9986SGeorge Wilson * was removed then the histogram accounting may 21402e4c9986SGeorge Wilson * not be accurate. Update the histogram information 21412e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 21422e4c9986SGeorge Wilson * and metaslab class are up-to-date. 21432e4c9986SGeorge Wilson */ 21442e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 21452e4c9986SGeorge Wilson 21460713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 21470713e232SGeorge Wilson space_map_free(msp->ms_sm, tx); 21480713e232SGeorge Wilson space_map_close(msp->ms_sm); 21490713e232SGeorge Wilson msp->ms_sm = NULL; 21500713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 215188ecc943SGeorge Wilson } 21522e4c9986SGeorge Wilson 21532e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21542e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21552e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 21562e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 21572e4c9986SGeorge Wilson 215888ecc943SGeorge Wilson } 215988ecc943SGeorge Wilson 216088ecc943SGeorge Wilson if (vd->vdev_ms_array) { 216188ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 216288ecc943SGeorge Wilson vd->vdev_ms_array = 0; 216388ecc943SGeorge Wilson } 216488ecc943SGeorge Wilson dmu_tx_commit(tx); 216588ecc943SGeorge Wilson } 216688ecc943SGeorge Wilson 2167fa9e4066Sahrens void 2168fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2169fa9e4066Sahrens { 2170fa9e4066Sahrens metaslab_t *msp; 217180eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2172fa9e4066Sahrens 217388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 217488ecc943SGeorge Wilson 2175fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2176fa9e4066Sahrens metaslab_sync_done(msp, txg); 217780eb36f2SGeorge Wilson 217880eb36f2SGeorge Wilson if (reassess) 217980eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2180fa9e4066Sahrens } 2181fa9e4066Sahrens 2182fa9e4066Sahrens void 2183fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2184fa9e4066Sahrens { 2185fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2186fa9e4066Sahrens vdev_t *lvd; 2187fa9e4066Sahrens metaslab_t *msp; 2188ecc2d604Sbonwick dmu_tx_t *tx; 2189fa9e4066Sahrens 219088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 219188ecc943SGeorge Wilson 2192ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2193ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2194ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2195ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2196ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2197ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2198ecc2d604Sbonwick vdev_config_dirty(vd); 2199ecc2d604Sbonwick dmu_tx_commit(tx); 2200ecc2d604Sbonwick } 2201fa9e4066Sahrens 22023f9d6ad7SLin Ling /* 22033f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 22043f9d6ad7SLin Ling */ 22053f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 220688ecc943SGeorge Wilson vdev_remove(vd, txg); 220788ecc943SGeorge Wilson 2208ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2209fa9e4066Sahrens metaslab_sync(msp, txg); 2210ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2211ecc2d604Sbonwick } 2212fa9e4066Sahrens 2213fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2214fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2215fa9e4066Sahrens 2216fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2217fa9e4066Sahrens } 2218fa9e4066Sahrens 2219fa9e4066Sahrens uint64_t 2220fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2221fa9e4066Sahrens { 2222fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2223fa9e4066Sahrens } 2224fa9e4066Sahrens 22253d7072f8Seschrock /* 22263d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 22273d7072f8Seschrock * not be opened, and no I/O is attempted. 22283d7072f8Seschrock */ 2229fa9e4066Sahrens int 2230069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2231fa9e4066Sahrens { 22324b964adaSGeorge Wilson vdev_t *vd, *tvd; 2233fa9e4066Sahrens 22348f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2235fa9e4066Sahrens 2236c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2237e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2238e14bb325SJeff Bonwick 22393d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2240e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2241fa9e4066Sahrens 22424b964adaSGeorge Wilson tvd = vd->vdev_top; 22434b964adaSGeorge Wilson 2244069f55e2SEric Schrock /* 2245069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2246069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2247069f55e2SEric Schrock * were faulted. 2248069f55e2SEric Schrock */ 2249069f55e2SEric Schrock vd->vdev_label_aux = aux; 2250069f55e2SEric Schrock 22513d7072f8Seschrock /* 22523d7072f8Seschrock * Faulted state takes precedence over degraded. 22533d7072f8Seschrock */ 225498d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 22553d7072f8Seschrock vd->vdev_faulted = 1ULL; 22563d7072f8Seschrock vd->vdev_degraded = 0ULL; 2257069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 22583d7072f8Seschrock 22593d7072f8Seschrock /* 2260c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2261c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 22623d7072f8Seschrock */ 22634b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 22643d7072f8Seschrock vd->vdev_degraded = 1ULL; 22653d7072f8Seschrock vd->vdev_faulted = 0ULL; 22663d7072f8Seschrock 22673d7072f8Seschrock /* 22683d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 22693d7072f8Seschrock * mark it degraded. 22703d7072f8Seschrock */ 22714b964adaSGeorge Wilson vdev_reopen(tvd); 22723d7072f8Seschrock 2273069f55e2SEric Schrock if (vdev_readable(vd)) 2274069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 22753d7072f8Seschrock } 22763d7072f8Seschrock 2277e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 22783d7072f8Seschrock } 22793d7072f8Seschrock 22803d7072f8Seschrock /* 22813d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 22823d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 22833d7072f8Seschrock * as I/O is concerned. 22843d7072f8Seschrock */ 22853d7072f8Seschrock int 2286069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 22873d7072f8Seschrock { 2288c5904d13Seschrock vdev_t *vd; 22890a4e9518Sgw 22908f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 22913d7072f8Seschrock 2292c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2293e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2294e14bb325SJeff Bonwick 22950e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2296e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 22970e34b6a7Sbonwick 22983d7072f8Seschrock /* 22993d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 23003d7072f8Seschrock */ 2301e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2302e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 23033d7072f8Seschrock 23043d7072f8Seschrock vd->vdev_degraded = 1ULL; 23053d7072f8Seschrock if (!vdev_is_dead(vd)) 23063d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2307069f55e2SEric Schrock aux); 23083d7072f8Seschrock 2309e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23103d7072f8Seschrock } 23113d7072f8Seschrock 23123d7072f8Seschrock /* 2313f7170741SWill Andrews * Online the given vdev. 2314f7170741SWill Andrews * 2315f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2316f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2317f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2318f7170741SWill Andrews * events are generated if the device fails to open. 23193d7072f8Seschrock */ 23203d7072f8Seschrock int 2321e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 23223d7072f8Seschrock { 2323573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 23243d7072f8Seschrock 23258f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 23263d7072f8Seschrock 2327c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2328e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 23293d7072f8Seschrock 23303d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2331e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2332fa9e4066Sahrens 2333573ca77eSGeorge Wilson tvd = vd->vdev_top; 2334fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2335441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2336e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2337e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2338573ca77eSGeorge Wilson 2339573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2340573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2341573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2342573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2343573ca77eSGeorge Wilson } 2344573ca77eSGeorge Wilson 2345573ca77eSGeorge Wilson vdev_reopen(tvd); 23463d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 23473d7072f8Seschrock 2348573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2349573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2350573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2351573ca77eSGeorge Wilson } 2352573ca77eSGeorge Wilson 23533d7072f8Seschrock if (newstate) 23543d7072f8Seschrock *newstate = vd->vdev_state; 23553d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 23563d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 23573d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 23583d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 23593d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2360fa9e4066Sahrens 2361573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2362573ca77eSGeorge Wilson 2363573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2364573ca77eSGeorge Wilson if (vd->vdev_aux) 2365573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2366573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2367573ca77eSGeorge Wilson } 23688ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2369fa9e4066Sahrens } 2370fa9e4066Sahrens 2371a1521560SJeff Bonwick static int 2372a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2373fa9e4066Sahrens { 2374e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 23758f18d1faSGeorge Wilson int error = 0; 23768f18d1faSGeorge Wilson uint64_t generation; 23778f18d1faSGeorge Wilson metaslab_group_t *mg; 23780a4e9518Sgw 23798f18d1faSGeorge Wilson top: 23808f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2381fa9e4066Sahrens 2382c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2383e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2384fa9e4066Sahrens 23850e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2386e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 23870e34b6a7Sbonwick 2388e6ca193dSGeorge Wilson tvd = vd->vdev_top; 23898f18d1faSGeorge Wilson mg = tvd->vdev_mg; 23908f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2391e6ca193dSGeorge Wilson 2392fa9e4066Sahrens /* 2393ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2394fa9e4066Sahrens */ 2395ecc2d604Sbonwick if (!vd->vdev_offline) { 2396ecc2d604Sbonwick /* 23978ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2398e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2399e6ca193dSGeorge Wilson * expendable. 2400ecc2d604Sbonwick */ 2401e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2402e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2403e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2404fa9e4066Sahrens 24058f18d1faSGeorge Wilson /* 2406b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2407b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2408b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2409b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 24108f18d1faSGeorge Wilson */ 24118f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 24128f18d1faSGeorge Wilson /* 24138f18d1faSGeorge Wilson * Prevent any future allocations. 24148f18d1faSGeorge Wilson */ 2415a1521560SJeff Bonwick metaslab_group_passivate(mg); 24168f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24178f18d1faSGeorge Wilson 24181195e687SMark J Musante error = spa_offline_log(spa); 24198f18d1faSGeorge Wilson 24208f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 24218f18d1faSGeorge Wilson 24228f18d1faSGeorge Wilson /* 24238f18d1faSGeorge Wilson * Check to see if the config has changed. 24248f18d1faSGeorge Wilson */ 24258f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2426a1521560SJeff Bonwick metaslab_group_activate(mg); 24278f18d1faSGeorge Wilson if (error) 24288f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 24298f18d1faSGeorge Wilson vd, error)); 24308f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24318f18d1faSGeorge Wilson goto top; 24328f18d1faSGeorge Wilson } 2433fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 24348f18d1faSGeorge Wilson } 24358f18d1faSGeorge Wilson 2436ecc2d604Sbonwick /* 2437ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2438e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2439e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2440e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2441ecc2d604Sbonwick */ 2442ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2443e6ca193dSGeorge Wilson vdev_reopen(tvd); 2444e6ca193dSGeorge Wilson 2445e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2446e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2447ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2448e6ca193dSGeorge Wilson vdev_reopen(tvd); 2449e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2450ecc2d604Sbonwick } 24518f18d1faSGeorge Wilson 24528f18d1faSGeorge Wilson /* 24538f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 24548f18d1faSGeorge Wilson * once we online the device it's open for business. 24558f18d1faSGeorge Wilson */ 24568f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2457a1521560SJeff Bonwick metaslab_group_activate(mg); 2458fa9e4066Sahrens } 2459fa9e4066Sahrens 2460e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2461ecc2d604Sbonwick 24628f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2463fa9e4066Sahrens } 2464fa9e4066Sahrens 2465a1521560SJeff Bonwick int 2466a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2467a1521560SJeff Bonwick { 2468a1521560SJeff Bonwick int error; 2469a1521560SJeff Bonwick 2470a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2471a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2472a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2473a1521560SJeff Bonwick 2474a1521560SJeff Bonwick return (error); 2475a1521560SJeff Bonwick } 2476a1521560SJeff Bonwick 2477ea8dc4b6Seschrock /* 2478ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2479ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2480ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2481ea8dc4b6Seschrock */ 2482ea8dc4b6Seschrock void 2483e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2484fa9e4066Sahrens { 2485e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2486e14bb325SJeff Bonwick 2487e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2488fa9e4066Sahrens 2489ea8dc4b6Seschrock if (vd == NULL) 2490e14bb325SJeff Bonwick vd = rvd; 2491fa9e4066Sahrens 2492ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2493ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2494ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2495fa9e4066Sahrens 2496e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2497e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 24983d7072f8Seschrock 24993d7072f8Seschrock /* 25008a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 25018a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 25028a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 25038a79c1b5Sek * written out to disk. 25043d7072f8Seschrock */ 2505e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2506e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 25078a79c1b5Sek 2508096d22d4SEric Schrock /* 2509096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2510096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2511096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2512096d22d4SEric Schrock */ 2513096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2514096d22d4SEric Schrock 25154b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2516e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2517e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2518e14bb325SJeff Bonwick 2519f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 25203d7072f8Seschrock 2521096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2522096d22d4SEric Schrock 2523f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2524e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2525e14bb325SJeff Bonwick 2526e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2527bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 25283d7072f8Seschrock 25293d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 25303d7072f8Seschrock } 2531096d22d4SEric Schrock 2532096d22d4SEric Schrock /* 2533096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2534096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2535096d22d4SEric Schrock * done in response to the FMA fault. 2536096d22d4SEric Schrock */ 2537096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2538096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2539096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2540096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2541fa9e4066Sahrens } 2542fa9e4066Sahrens 2543e14bb325SJeff Bonwick boolean_t 2544e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 25450a4e9518Sgw { 254688ecc943SGeorge Wilson /* 254788ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 254888ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 254988ecc943SGeorge Wilson * these types of devices in the various code paths. 255088ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 255188ecc943SGeorge Wilson * before issuing I/O to them. 255288ecc943SGeorge Wilson */ 255388ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 255488ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 25550a4e9518Sgw } 25560a4e9518Sgw 2557e14bb325SJeff Bonwick boolean_t 2558e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 25590a4e9518Sgw { 2560e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 25610a4e9518Sgw } 25620a4e9518Sgw 2563e14bb325SJeff Bonwick boolean_t 2564e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2565fa9e4066Sahrens { 2566e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2567fa9e4066Sahrens } 2568fa9e4066Sahrens 2569a31e6787SGeorge Wilson boolean_t 2570a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2571a31e6787SGeorge Wilson { 25728ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 25738ad4d6ddSJeff Bonwick 2574a31e6787SGeorge Wilson /* 25758ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2576a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2577a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 25788ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 25798ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 25808ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2581a31e6787SGeorge Wilson */ 25828ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 25833f9d6ad7SLin Ling !vd->vdev_cant_write && !vd->vdev_ishole); 2584a31e6787SGeorge Wilson } 2585a31e6787SGeorge Wilson 2586e14bb325SJeff Bonwick boolean_t 2587e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2588fa9e4066Sahrens { 2589e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2590fa9e4066Sahrens 2591e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2592e14bb325SJeff Bonwick return (B_FALSE); 2593fa9e4066Sahrens 2594e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2595e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2596fa9e4066Sahrens 2597e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2598e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2599fa9e4066Sahrens 2600e14bb325SJeff Bonwick return (B_TRUE); 2601fa9e4066Sahrens } 2602fa9e4066Sahrens 2603fa9e4066Sahrens /* 2604fa9e4066Sahrens * Get statistics for the given vdev. 2605fa9e4066Sahrens */ 2606fa9e4066Sahrens void 2607fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2608fa9e4066Sahrens { 26092e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 26102e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 26112e4c9986SGeorge Wilson 26122e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2613fa9e4066Sahrens 2614fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2615fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2616fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2617fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2618573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2619573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2620573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 26214263d13fSGeorge Wilson vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 26222986efa8SAlex Reece if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 26232e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 26242986efa8SAlex Reece } 2625fa9e4066Sahrens 2626fa9e4066Sahrens /* 2627fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2628fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2629fa9e4066Sahrens */ 2630fa9e4066Sahrens if (vd == rvd) { 2631e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2632fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2633fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2634fa9e4066Sahrens 2635e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2636fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2637fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2638fa9e4066Sahrens } 26393f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2640fa9e4066Sahrens } 2641fa9e4066Sahrens } 26422e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 2643fa9e4066Sahrens } 2644fa9e4066Sahrens 2645fa94a07fSbrendan void 2646fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2647fa94a07fSbrendan { 2648fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2649fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2650fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2651fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2652fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2653fa94a07fSbrendan } 2654fa94a07fSbrendan 26553f9d6ad7SLin Ling void 26563f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 26573f9d6ad7SLin Ling { 26583f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 26593f9d6ad7SLin Ling 26603f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 26613f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 26623f9d6ad7SLin Ling 26633f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 26643f9d6ad7SLin Ling vs->vs_scan_processed = 0; 26653f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 26663f9d6ad7SLin Ling } 26673f9d6ad7SLin Ling 2668fa9e4066Sahrens void 2669e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2670fa9e4066Sahrens { 26718ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 26728ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2673e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2674fa9e4066Sahrens vdev_t *pvd; 2675fa9e4066Sahrens uint64_t txg = zio->io_txg; 2676fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2677fa9e4066Sahrens zio_type_t type = zio->io_type; 2678fa9e4066Sahrens int flags = zio->io_flags; 2679fa9e4066Sahrens 2680e14bb325SJeff Bonwick /* 2681e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2682e14bb325SJeff Bonwick */ 2683e14bb325SJeff Bonwick if (zio->io_gang_tree) 2684e14bb325SJeff Bonwick return; 2685e14bb325SJeff Bonwick 2686fa9e4066Sahrens if (zio->io_error == 0) { 2687e14bb325SJeff Bonwick /* 2688e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2689e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2690e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2691e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2692e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2693e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2694e14bb325SJeff Bonwick * remain zero, which is what we want.) 2695e14bb325SJeff Bonwick * 2696e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2697e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2698e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2699e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2700e14bb325SJeff Bonwick */ 2701e14bb325SJeff Bonwick if (vd == rvd) 2702e14bb325SJeff Bonwick return; 2703e14bb325SJeff Bonwick 2704e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 27058ad4d6ddSJeff Bonwick 27068ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 27078ad4d6ddSJeff Bonwick return; 27088ad4d6ddSJeff Bonwick 27098ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 27108ad4d6ddSJeff Bonwick 2711e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 271244ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 27133f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 27143f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 27153f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 27163f9d6ad7SLin Ling 27173f9d6ad7SLin Ling /* XXX cleanup? */ 27183f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 27193f9d6ad7SLin Ling atomic_add_64(processed, psize); 27203f9d6ad7SLin Ling vs->vs_scan_processed += psize; 27213f9d6ad7SLin Ling } 27223f9d6ad7SLin Ling 27238ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2724e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2725fa9e4066Sahrens } 27268ad4d6ddSJeff Bonwick 27278ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 27288ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 27298ad4d6ddSJeff Bonwick 27308ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2731fa9e4066Sahrens return; 2732fa9e4066Sahrens } 2733fa9e4066Sahrens 2734fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2735fa9e4066Sahrens return; 2736fa9e4066Sahrens 27378956713aSEric Schrock /* 27388956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 27398956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 27408956713aSEric Schrock * hard errors, when in reality they can happen for any number of 27418956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 27428956713aSEric Schrock */ 27438956713aSEric Schrock if (zio->io_error == EIO && 27448956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 27458956713aSEric Schrock return; 27468956713aSEric Schrock 27478f18d1faSGeorge Wilson /* 27488f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 27498f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 27508f18d1faSGeorge Wilson * errors. 27518f18d1faSGeorge Wilson */ 27528f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 27538f18d1faSGeorge Wilson return; 27548f18d1faSGeorge Wilson 2755e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2756b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2757e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2758e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2759e14bb325SJeff Bonwick else 2760e14bb325SJeff Bonwick vs->vs_read_errors++; 2761fa9e4066Sahrens } 2762b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2763e14bb325SJeff Bonwick vs->vs_write_errors++; 2764e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2765fa9e4066Sahrens 27668ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 27678ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 276844ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2769b24ab676SJeff Bonwick spa->spa_claiming)) { 27708ad4d6ddSJeff Bonwick /* 2771b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2772b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2773b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2774b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2775b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2776b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2777b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2778b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 27798ad4d6ddSJeff Bonwick * 27808ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 27818ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 27828ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 27838ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 27848ad4d6ddSJeff Bonwick */ 27858ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 27868ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 278744ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 27888ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 27898ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 27908ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2791b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2792b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2793b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2794b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 27958ad4d6ddSJeff Bonwick } 2796b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 27978ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2798fa9e4066Sahrens return; 27998ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 28008ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 28018ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2802fa9e4066Sahrens } 28038ad4d6ddSJeff Bonwick if (vd != rvd) 28048ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2805fa9e4066Sahrens } 2806fa9e4066Sahrens } 2807fa9e4066Sahrens 2808fa9e4066Sahrens /* 2809b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2810b24ab676SJeff Bonwick * and the root vdev. 2811fa9e4066Sahrens */ 2812fa9e4066Sahrens void 2813b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2814b24ab676SJeff Bonwick int64_t space_delta) 2815fa9e4066Sahrens { 281699653d4eSeschrock int64_t dspace_delta = space_delta; 28178654d025Sperrin spa_t *spa = vd->vdev_spa; 28188654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2819b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2820b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2821fa9e4066Sahrens 28228654d025Sperrin ASSERT(vd == vd->vdev_top); 282399653d4eSeschrock 28248654d025Sperrin /* 28258654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 28268654d025Sperrin * factor. We must calculate this here and not at the root vdev 28278654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 28288654d025Sperrin * childrens', thus not accurate enough for us. 28298654d025Sperrin */ 28308654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2831e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 28328654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 28338654d025Sperrin vd->vdev_deflate_ratio; 28348654d025Sperrin 28358654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 28368654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2837b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 28388654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 28398654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 28408654d025Sperrin 2841b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2842fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2843fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2844b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2845fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2846fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2847fa94a07fSbrendan } 2848b24ab676SJeff Bonwick 2849b24ab676SJeff Bonwick if (mc != NULL) { 2850b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2851b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2852b24ab676SJeff Bonwick 2853b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2854b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2855b24ab676SJeff Bonwick } 2856fa9e4066Sahrens } 2857fa9e4066Sahrens 2858fa9e4066Sahrens /* 2859fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2860fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2861fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2862fa9e4066Sahrens */ 2863fa9e4066Sahrens void 2864fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2865fa9e4066Sahrens { 2866fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2867fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2868fa9e4066Sahrens int c; 2869fa9e4066Sahrens 2870f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2871f9af39baSGeorge Wilson 2872c5904d13Seschrock /* 28736809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 28746809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2875c5904d13Seschrock */ 2876c5904d13Seschrock if (vd->vdev_aux != NULL) { 2877c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2878c5904d13Seschrock nvlist_t **aux; 2879c5904d13Seschrock uint_t naux; 2880c5904d13Seschrock 2881c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2882c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2883c5904d13Seschrock break; 2884c5904d13Seschrock } 2885c5904d13Seschrock 2886e14bb325SJeff Bonwick if (c == sav->sav_count) { 2887e14bb325SJeff Bonwick /* 2888e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2889e14bb325SJeff Bonwick */ 2890e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2891e14bb325SJeff Bonwick return; 2892e14bb325SJeff Bonwick } 2893e14bb325SJeff Bonwick 2894c5904d13Seschrock sav->sav_sync = B_TRUE; 2895c5904d13Seschrock 28966809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 28976809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 28986809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 28996809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 29006809eb4eSEric Schrock } 2901c5904d13Seschrock 2902c5904d13Seschrock ASSERT(c < naux); 2903c5904d13Seschrock 2904c5904d13Seschrock /* 2905c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2906c5904d13Seschrock * sketchy, but it will work. 2907c5904d13Seschrock */ 2908c5904d13Seschrock nvlist_free(aux[c]); 29093f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2910c5904d13Seschrock 2911c5904d13Seschrock return; 2912c5904d13Seschrock } 2913c5904d13Seschrock 29145dabedeeSbonwick /* 2915e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2916e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2917e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 29185dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 29195dabedeeSbonwick */ 2920e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2921e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2922e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29235dabedeeSbonwick 2924fa9e4066Sahrens if (vd == rvd) { 2925fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2926fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2927fa9e4066Sahrens } else { 2928fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2929fa9e4066Sahrens 293088ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 293188ecc943SGeorge Wilson !vd->vdev_ishole) 2932e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2933fa9e4066Sahrens } 2934fa9e4066Sahrens } 2935fa9e4066Sahrens 2936fa9e4066Sahrens void 2937fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2938fa9e4066Sahrens { 29395dabedeeSbonwick spa_t *spa = vd->vdev_spa; 29405dabedeeSbonwick 2941e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2942e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2943e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29445dabedeeSbonwick 2945e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2946e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2947e14bb325SJeff Bonwick } 2948e14bb325SJeff Bonwick 2949e14bb325SJeff Bonwick /* 2950e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2951e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2952e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2953e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2954e14bb325SJeff Bonwick */ 2955e14bb325SJeff Bonwick void 2956e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2957e14bb325SJeff Bonwick { 2958e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2959e14bb325SJeff Bonwick 2960f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2961e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2962e14bb325SJeff Bonwick 2963e14bb325SJeff Bonwick /* 2964e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 2965e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 2966e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 2967e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 2968e14bb325SJeff Bonwick */ 2969e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2970e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2971e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2972e14bb325SJeff Bonwick 2973b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2974e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 2975e14bb325SJeff Bonwick } 2976e14bb325SJeff Bonwick 2977e14bb325SJeff Bonwick void 2978e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 2979e14bb325SJeff Bonwick { 2980e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2981e14bb325SJeff Bonwick 2982e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2983e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2984e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2985e14bb325SJeff Bonwick 2986e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2987e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 2988fa9e4066Sahrens } 2989fa9e4066Sahrens 299032b87932Sek /* 299132b87932Sek * Propagate vdev state up from children to parent. 299232b87932Sek */ 299344cd46caSbillm void 299444cd46caSbillm vdev_propagate_state(vdev_t *vd) 299544cd46caSbillm { 29968ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 29978ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 299844cd46caSbillm int degraded = 0, faulted = 0; 299944cd46caSbillm int corrupted = 0; 300044cd46caSbillm vdev_t *child; 300144cd46caSbillm 30023d7072f8Seschrock if (vd->vdev_children > 0) { 3003573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 30043d7072f8Seschrock child = vd->vdev_child[c]; 300551ece835Seschrock 300688ecc943SGeorge Wilson /* 300788ecc943SGeorge Wilson * Don't factor holes into the decision. 300888ecc943SGeorge Wilson */ 300988ecc943SGeorge Wilson if (child->vdev_ishole) 301088ecc943SGeorge Wilson continue; 301188ecc943SGeorge Wilson 3012e14bb325SJeff Bonwick if (!vdev_readable(child) || 30138ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 301451ece835Seschrock /* 301551ece835Seschrock * Root special: if there is a top-level log 301651ece835Seschrock * device, treat the root vdev as if it were 301751ece835Seschrock * degraded. 301851ece835Seschrock */ 301951ece835Seschrock if (child->vdev_islog && vd == rvd) 302051ece835Seschrock degraded++; 302151ece835Seschrock else 302251ece835Seschrock faulted++; 302351ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 30243d7072f8Seschrock degraded++; 302551ece835Seschrock } 302644cd46caSbillm 30273d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 30283d7072f8Seschrock corrupted++; 30293d7072f8Seschrock } 303044cd46caSbillm 30313d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 30323d7072f8Seschrock 30333d7072f8Seschrock /* 3034e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 30353d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 30363d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 30373d7072f8Seschrock * replicas'. 30383d7072f8Seschrock */ 30393d7072f8Seschrock if (corrupted && vd == rvd && 30403d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 30413d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 30423d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 30433d7072f8Seschrock } 30443d7072f8Seschrock 304551ece835Seschrock if (vd->vdev_parent) 30463d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 304744cd46caSbillm } 304844cd46caSbillm 3049fa9e4066Sahrens /* 3050ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3051ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3052ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3053ea8dc4b6Seschrock * 3054ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3055ea8dc4b6Seschrock * generated. 3056fa9e4066Sahrens */ 3057fa9e4066Sahrens void 3058ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3059fa9e4066Sahrens { 3060560e6e96Seschrock uint64_t save_state; 3061c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3062ea8dc4b6Seschrock 3063ea8dc4b6Seschrock if (state == vd->vdev_state) { 3064ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3065fa9e4066Sahrens return; 3066ea8dc4b6Seschrock } 3067ea8dc4b6Seschrock 3068560e6e96Seschrock save_state = vd->vdev_state; 3069fa9e4066Sahrens 3070fa9e4066Sahrens vd->vdev_state = state; 3071fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3072fa9e4066Sahrens 30733d7072f8Seschrock /* 30743d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 307598d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 307698d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 307798d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 307898d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 307998d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 308098d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 308198d1cbfeSGeorge Wilson * children. 308298d1cbfeSGeorge Wilson */ 308398d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 308498d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 30853d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 30863d7072f8Seschrock 3087069f55e2SEric Schrock /* 3088069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3089069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3090069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3091069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3092069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3093069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3094069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3095069f55e2SEric Schrock */ 3096069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3097069f55e2SEric Schrock vd->vdev_prevstate != state) 3098069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3099069f55e2SEric Schrock 31003d7072f8Seschrock if (vd->vdev_removed && 31013d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 31023d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 31033d7072f8Seschrock /* 31043d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 31053d7072f8Seschrock * device was previously marked removed and someone attempted to 31063d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 31073d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 31083d7072f8Seschrock * it is one of our special test online cases, which is only 31093d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 31103d7072f8Seschrock * fault. 31113d7072f8Seschrock */ 31123d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 31133d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 31143d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 31153d7072f8Seschrock vd->vdev_removed = B_TRUE; 31163d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3117ea8dc4b6Seschrock /* 3118cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3119cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3120cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3121cb04b873SMark J Musante * is not considered an error. 3122ea8dc4b6Seschrock */ 3123cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3124cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3125560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3126560e6e96Seschrock vd->vdev_not_present = 1; 3127560e6e96Seschrock 3128560e6e96Seschrock /* 3129560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3130560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3131560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3132560e6e96Seschrock * want to post the ereport if the device was already in the 3133560e6e96Seschrock * CANT_OPEN state beforehand. 31343d7072f8Seschrock * 31353d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 31363d7072f8Seschrock * online the device in response to an insertion event. If we 31373d7072f8Seschrock * hit this case, then we have detected an insertion event for a 31383d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 31393d7072f8Seschrock * In this scenario, we don't post an ereport because we are 31403d7072f8Seschrock * about to replace the device, or attempt an online with 31413d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3142560e6e96Seschrock */ 31433d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 31443d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3145c5904d13Seschrock vd != spa->spa_root_vdev) { 3146ea8dc4b6Seschrock const char *class; 3147ea8dc4b6Seschrock 3148ea8dc4b6Seschrock switch (aux) { 3149ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3150ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3151ea8dc4b6Seschrock break; 3152ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3153ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3154ea8dc4b6Seschrock break; 3155ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3156ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3157ea8dc4b6Seschrock break; 3158ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3159ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3160ea8dc4b6Seschrock break; 3161ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3162ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3163ea8dc4b6Seschrock break; 3164ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3165ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3166ea8dc4b6Seschrock break; 3167ea8dc4b6Seschrock default: 3168ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3169ea8dc4b6Seschrock } 3170ea8dc4b6Seschrock 3171c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3172ea8dc4b6Seschrock } 3173ea8dc4b6Seschrock 31743d7072f8Seschrock /* Erase any notion of persistent removed state */ 31753d7072f8Seschrock vd->vdev_removed = B_FALSE; 31763d7072f8Seschrock } else { 31773d7072f8Seschrock vd->vdev_removed = B_FALSE; 31783d7072f8Seschrock } 3179ea8dc4b6Seschrock 31808b33d774STim Haley if (!isopen && vd->vdev_parent) 31818b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3182fa9e4066Sahrens } 318315e6edf1Sgw 318415e6edf1Sgw /* 318515e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 318615e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 318715e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 318815e6edf1Sgw * can be wholedisks. 318915e6edf1Sgw */ 319015e6edf1Sgw boolean_t 319115e6edf1Sgw vdev_is_bootable(vdev_t *vd) 319215e6edf1Sgw { 319315e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 319415e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 319515e6edf1Sgw 319615e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 319715e6edf1Sgw vd->vdev_children > 1) { 319815e6edf1Sgw return (B_FALSE); 319915e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 320015e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 320115e6edf1Sgw return (B_FALSE); 320215e6edf1Sgw } 320315e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 320415e6edf1Sgw return (B_FALSE); 320515e6edf1Sgw } 320615e6edf1Sgw 3207573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 320815e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 320915e6edf1Sgw return (B_FALSE); 321015e6edf1Sgw } 321115e6edf1Sgw return (B_TRUE); 321215e6edf1Sgw } 3213e6ca193dSGeorge Wilson 321488ecc943SGeorge Wilson /* 321588ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 321688ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 32174b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 32184b964adaSGeorge Wilson * device in the current vdev tree (nvd). 321988ecc943SGeorge Wilson */ 3220e6ca193dSGeorge Wilson void 322188ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3222e6ca193dSGeorge Wilson { 322388ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3224e6ca193dSGeorge Wilson 32254b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 322688ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 322788ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3228e6ca193dSGeorge Wilson 322988ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 323088ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3231e6ca193dSGeorge Wilson 32324b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3233e6ca193dSGeorge Wilson /* 32344b964adaSGeorge Wilson * Restore the persistent vdev state 3235e6ca193dSGeorge Wilson */ 323688ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 32374b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 32384b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 32394b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3240e6ca193dSGeorge Wilson } 3241e6ca193dSGeorge Wilson } 3242573ca77eSGeorge Wilson 32434b964adaSGeorge Wilson /* 32444b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 32454b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 32464b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 32474b964adaSGeorge Wilson */ 32484b964adaSGeorge Wilson boolean_t 32494b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 32504b964adaSGeorge Wilson { 32514b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 32524b964adaSGeorge Wilson !vd->vdev_removed) 32534b964adaSGeorge Wilson return (B_TRUE); 32544b964adaSGeorge Wilson 32554b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 32564b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 32574b964adaSGeorge Wilson return (B_TRUE); 32584b964adaSGeorge Wilson 32594b964adaSGeorge Wilson return (B_FALSE); 32604b964adaSGeorge Wilson } 32614b964adaSGeorge Wilson 3262573ca77eSGeorge Wilson /* 3263573ca77eSGeorge Wilson * Expand a vdev if possible. 3264573ca77eSGeorge Wilson */ 3265573ca77eSGeorge Wilson void 3266573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3267573ca77eSGeorge Wilson { 3268573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3269573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3270573ca77eSGeorge Wilson 3271573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3272573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3273573ca77eSGeorge Wilson vdev_config_dirty(vd); 3274573ca77eSGeorge Wilson } 3275573ca77eSGeorge Wilson } 32761195e687SMark J Musante 32771195e687SMark J Musante /* 32781195e687SMark J Musante * Split a vdev. 32791195e687SMark J Musante */ 32801195e687SMark J Musante void 32811195e687SMark J Musante vdev_split(vdev_t *vd) 32821195e687SMark J Musante { 32831195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 32841195e687SMark J Musante 32851195e687SMark J Musante vdev_remove_child(pvd, vd); 32861195e687SMark J Musante vdev_compact_children(pvd); 32871195e687SMark J Musante 32881195e687SMark J Musante cvd = pvd->vdev_child[0]; 32891195e687SMark J Musante if (pvd->vdev_children == 1) { 32901195e687SMark J Musante vdev_remove_parent(cvd); 32911195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 32921195e687SMark J Musante } 32931195e687SMark J Musante vdev_propagate_state(cvd); 32941195e687SMark J Musante } 3295283b8460SGeorge.Wilson 3296283b8460SGeorge.Wilson void 3297283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3298283b8460SGeorge.Wilson { 3299283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3300283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3301283b8460SGeorge.Wilson 3302283b8460SGeorge.Wilson vdev_deadman(cvd); 3303283b8460SGeorge.Wilson } 3304283b8460SGeorge.Wilson 3305283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3306283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3307283b8460SGeorge.Wilson 3308283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 330969962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3310283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3311283b8460SGeorge.Wilson zio_t *fio; 3312283b8460SGeorge.Wilson uint64_t delta; 3313283b8460SGeorge.Wilson 3314283b8460SGeorge.Wilson /* 3315283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3316283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3317283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3318283b8460SGeorge.Wilson */ 331969962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3320c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3321c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 3322c55e05cbSMatthew Ahrens zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3323c55e05cbSMatthew Ahrens "delta %lluns, last io %lluns", 3324283b8460SGeorge.Wilson fio->io_timestamp, delta, 3325283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3326283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3327283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3328283b8460SGeorge.Wilson } 3329283b8460SGeorge.Wilson } 3330283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3331283b8460SGeorge.Wilson } 3332283b8460SGeorge.Wilson } 3333