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 /* 23a3f829aeSBill Moore * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24fa9e4066Sahrens * Use is subject to license terms. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/zfs_context.h> 28ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 29fa9e4066Sahrens #include <sys/spa.h> 30fa9e4066Sahrens #include <sys/spa_impl.h> 31fa9e4066Sahrens #include <sys/dmu.h> 32fa9e4066Sahrens #include <sys/dmu_tx.h> 33fa9e4066Sahrens #include <sys/vdev_impl.h> 34fa9e4066Sahrens #include <sys/uberblock_impl.h> 35fa9e4066Sahrens #include <sys/metaslab.h> 36fa9e4066Sahrens #include <sys/metaslab_impl.h> 37fa9e4066Sahrens #include <sys/space_map.h> 38fa9e4066Sahrens #include <sys/zio.h> 39fa9e4066Sahrens #include <sys/zap.h> 40fa9e4066Sahrens #include <sys/fs/zfs.h> 41c5904d13Seschrock #include <sys/arc.h> 42e6ca193dSGeorge Wilson #include <sys/zil.h> 43fa9e4066Sahrens 44fa9e4066Sahrens /* 45fa9e4066Sahrens * Virtual device management. 46fa9e4066Sahrens */ 47fa9e4066Sahrens 48fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 49fa9e4066Sahrens &vdev_root_ops, 50fa9e4066Sahrens &vdev_raidz_ops, 51fa9e4066Sahrens &vdev_mirror_ops, 52fa9e4066Sahrens &vdev_replacing_ops, 5399653d4eSeschrock &vdev_spare_ops, 54fa9e4066Sahrens &vdev_disk_ops, 55fa9e4066Sahrens &vdev_file_ops, 56fa9e4066Sahrens &vdev_missing_ops, 5788ecc943SGeorge Wilson &vdev_hole_ops, 58fa9e4066Sahrens NULL 59fa9e4066Sahrens }; 60fa9e4066Sahrens 61088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 62088f3894Sahrens int zfs_scrub_limit = 10; 6305b2b3b8Smishra 64fa9e4066Sahrens /* 65fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 66fa9e4066Sahrens */ 67fa9e4066Sahrens static vdev_ops_t * 68fa9e4066Sahrens vdev_getops(const char *type) 69fa9e4066Sahrens { 70fa9e4066Sahrens vdev_ops_t *ops, **opspp; 71fa9e4066Sahrens 72fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 73fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 74fa9e4066Sahrens break; 75fa9e4066Sahrens 76fa9e4066Sahrens return (ops); 77fa9e4066Sahrens } 78fa9e4066Sahrens 79fa9e4066Sahrens /* 80fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 81fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 82fa9e4066Sahrens */ 83fa9e4066Sahrens uint64_t 84fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 85fa9e4066Sahrens { 86ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 87fa9e4066Sahrens uint64_t csize; 88fa9e4066Sahrens 89573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 90fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 91fa9e4066Sahrens asize = MAX(asize, csize); 92fa9e4066Sahrens } 93fa9e4066Sahrens 94fa9e4066Sahrens return (asize); 95fa9e4066Sahrens } 96fa9e4066Sahrens 972a79c5feSlling /* 98573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 99573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 100573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 101573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1022a79c5feSlling */ 1032a79c5feSlling uint64_t 104573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1052a79c5feSlling { 106573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1072a79c5feSlling 108573ca77eSGeorge Wilson /* 109573ca77eSGeorge Wilson * The our parent is NULL (inactive spare or cache) or is the root, 110573ca77eSGeorge Wilson * just return our own asize. 111573ca77eSGeorge Wilson */ 112573ca77eSGeorge Wilson if (pvd == NULL) 113573ca77eSGeorge Wilson return (vd->vdev_asize); 1142a79c5feSlling 1152a79c5feSlling /* 116573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 117573ca77eSGeorge Wilson * to the nearest metaslab. 1182a79c5feSlling */ 119573ca77eSGeorge Wilson if (vd == vd->vdev_top) 120573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1212a79c5feSlling 122573ca77eSGeorge Wilson /* 123573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 124573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 125573ca77eSGeorge Wilson */ 126573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 127573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1282a79c5feSlling 129573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 130573ca77eSGeorge Wilson } 1312a79c5feSlling 132573ca77eSGeorge Wilson void 133573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 134573ca77eSGeorge Wilson { 135573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 136573ca77eSGeorge Wilson 137573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 138573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1392a79c5feSlling } 1402a79c5feSlling 141fa9e4066Sahrens vdev_t * 142fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 143fa9e4066Sahrens { 144fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 145fa9e4066Sahrens 146e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 147e05725b1Sbonwick 148088f3894Sahrens if (vdev < rvd->vdev_children) { 149088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 150fa9e4066Sahrens return (rvd->vdev_child[vdev]); 151088f3894Sahrens } 152fa9e4066Sahrens 153fa9e4066Sahrens return (NULL); 154fa9e4066Sahrens } 155fa9e4066Sahrens 156fa9e4066Sahrens vdev_t * 157fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 158fa9e4066Sahrens { 159fa9e4066Sahrens vdev_t *mvd; 160fa9e4066Sahrens 1610e34b6a7Sbonwick if (vd->vdev_guid == guid) 162fa9e4066Sahrens return (vd); 163fa9e4066Sahrens 164573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 165fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 166fa9e4066Sahrens NULL) 167fa9e4066Sahrens return (mvd); 168fa9e4066Sahrens 169fa9e4066Sahrens return (NULL); 170fa9e4066Sahrens } 171fa9e4066Sahrens 172fa9e4066Sahrens void 173fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 174fa9e4066Sahrens { 175fa9e4066Sahrens size_t oldsize, newsize; 176fa9e4066Sahrens uint64_t id = cvd->vdev_id; 177fa9e4066Sahrens vdev_t **newchild; 178fa9e4066Sahrens 179e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 180fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 181fa9e4066Sahrens 182fa9e4066Sahrens cvd->vdev_parent = pvd; 183fa9e4066Sahrens 184fa9e4066Sahrens if (pvd == NULL) 185fa9e4066Sahrens return; 186fa9e4066Sahrens 187fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 188fa9e4066Sahrens 189fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 190fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 191fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 192fa9e4066Sahrens 193fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 194fa9e4066Sahrens if (pvd->vdev_child != NULL) { 195fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 196fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 197fa9e4066Sahrens } 198fa9e4066Sahrens 199fa9e4066Sahrens pvd->vdev_child = newchild; 200fa9e4066Sahrens pvd->vdev_child[id] = cvd; 201fa9e4066Sahrens 202fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 203fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 204fa9e4066Sahrens 205fa9e4066Sahrens /* 206fa9e4066Sahrens * Walk up all ancestors to update guid sum. 207fa9e4066Sahrens */ 208fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 209fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 21005b2b3b8Smishra 21105b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 21205b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; 213fa9e4066Sahrens } 214fa9e4066Sahrens 215fa9e4066Sahrens void 216fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 217fa9e4066Sahrens { 218fa9e4066Sahrens int c; 219fa9e4066Sahrens uint_t id = cvd->vdev_id; 220fa9e4066Sahrens 221fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 222fa9e4066Sahrens 223fa9e4066Sahrens if (pvd == NULL) 224fa9e4066Sahrens return; 225fa9e4066Sahrens 226fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 227fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 228fa9e4066Sahrens 229fa9e4066Sahrens pvd->vdev_child[id] = NULL; 230fa9e4066Sahrens cvd->vdev_parent = NULL; 231fa9e4066Sahrens 232fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 233fa9e4066Sahrens if (pvd->vdev_child[c]) 234fa9e4066Sahrens break; 235fa9e4066Sahrens 236fa9e4066Sahrens if (c == pvd->vdev_children) { 237fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 238fa9e4066Sahrens pvd->vdev_child = NULL; 239fa9e4066Sahrens pvd->vdev_children = 0; 240fa9e4066Sahrens } 241fa9e4066Sahrens 242fa9e4066Sahrens /* 243fa9e4066Sahrens * Walk up all ancestors to update guid sum. 244fa9e4066Sahrens */ 245fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 246fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 24705b2b3b8Smishra 24805b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 24905b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; 250fa9e4066Sahrens } 251fa9e4066Sahrens 252fa9e4066Sahrens /* 253fa9e4066Sahrens * Remove any holes in the child array. 254fa9e4066Sahrens */ 255fa9e4066Sahrens void 256fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 257fa9e4066Sahrens { 258fa9e4066Sahrens vdev_t **newchild, *cvd; 259fa9e4066Sahrens int oldc = pvd->vdev_children; 260573ca77eSGeorge Wilson int newc; 261fa9e4066Sahrens 262e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 263fa9e4066Sahrens 264573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 265fa9e4066Sahrens if (pvd->vdev_child[c]) 266fa9e4066Sahrens newc++; 267fa9e4066Sahrens 268fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 269fa9e4066Sahrens 270573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 271fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 272fa9e4066Sahrens newchild[newc] = cvd; 273fa9e4066Sahrens cvd->vdev_id = newc++; 274fa9e4066Sahrens } 275fa9e4066Sahrens } 276fa9e4066Sahrens 277fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 278fa9e4066Sahrens pvd->vdev_child = newchild; 279fa9e4066Sahrens pvd->vdev_children = newc; 280fa9e4066Sahrens } 281fa9e4066Sahrens 282fa9e4066Sahrens /* 283fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 284fa9e4066Sahrens */ 28588ecc943SGeorge Wilson vdev_t * 286fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 287fa9e4066Sahrens { 288fa9e4066Sahrens vdev_t *vd; 289fa9e4066Sahrens 290fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 291fa9e4066Sahrens 2920e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2930e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2940e34b6a7Sbonwick spa->spa_root_vdev = vd; 2950e34b6a7Sbonwick } 2960e34b6a7Sbonwick 29788ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 2980e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 2990e34b6a7Sbonwick /* 3000e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3010e34b6a7Sbonwick * which must be unique among all pools. 3020e34b6a7Sbonwick */ 3030e34b6a7Sbonwick while (guid == 0 || spa_guid_exists(guid, 0)) 3040e34b6a7Sbonwick guid = spa_get_random(-1ULL); 3050e34b6a7Sbonwick } else { 3060e34b6a7Sbonwick /* 3070e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3080e34b6a7Sbonwick */ 3090e34b6a7Sbonwick while (guid == 0 || 3100e34b6a7Sbonwick spa_guid_exists(spa_guid(spa), guid)) 3110e34b6a7Sbonwick guid = spa_get_random(-1ULL); 3120e34b6a7Sbonwick } 3130e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3140e34b6a7Sbonwick } 3150e34b6a7Sbonwick 316fa9e4066Sahrens vd->vdev_spa = spa; 317fa9e4066Sahrens vd->vdev_id = id; 318fa9e4066Sahrens vd->vdev_guid = guid; 319fa9e4066Sahrens vd->vdev_guid_sum = guid; 320fa9e4066Sahrens vd->vdev_ops = ops; 321fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 32288ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 323fa9e4066Sahrens 324fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3255ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 326e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3278ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3288ad4d6ddSJeff Bonwick space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 3298ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3308ad4d6ddSJeff Bonwick } 331fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 332fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 333fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 334fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 335fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3363d7072f8Seschrock vdev_queue_init(vd); 3373d7072f8Seschrock vdev_cache_init(vd); 338fa9e4066Sahrens 339fa9e4066Sahrens return (vd); 340fa9e4066Sahrens } 341fa9e4066Sahrens 342fa9e4066Sahrens /* 343fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 344fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 345fa9e4066Sahrens * different for each case. 346fa9e4066Sahrens */ 34799653d4eSeschrock int 34899653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 34999653d4eSeschrock int alloctype) 350fa9e4066Sahrens { 351fa9e4066Sahrens vdev_ops_t *ops; 352fa9e4066Sahrens char *type; 3538654d025Sperrin uint64_t guid = 0, islog, nparity; 354fa9e4066Sahrens vdev_t *vd; 355fa9e4066Sahrens 356e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 357fa9e4066Sahrens 358fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 35999653d4eSeschrock return (EINVAL); 360fa9e4066Sahrens 361fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 36299653d4eSeschrock return (EINVAL); 363fa9e4066Sahrens 364fa9e4066Sahrens /* 365fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 366fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 367fa9e4066Sahrens */ 368fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 369fa9e4066Sahrens uint64_t label_id; 370fa9e4066Sahrens 371fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 372fa9e4066Sahrens label_id != id) 37399653d4eSeschrock return (EINVAL); 374fa9e4066Sahrens 375fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37699653d4eSeschrock return (EINVAL); 37799653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 37899653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37999653d4eSeschrock return (EINVAL); 380fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 381fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 382fa94a07fSbrendan return (EINVAL); 38321ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 38421ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 38521ecdf64SLin Ling return (EINVAL); 386fa9e4066Sahrens } 387fa9e4066Sahrens 38899653d4eSeschrock /* 38999653d4eSeschrock * The first allocated vdev must be of type 'root'. 39099653d4eSeschrock */ 39199653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 39299653d4eSeschrock return (EINVAL); 39399653d4eSeschrock 3948654d025Sperrin /* 3958654d025Sperrin * Determine whether we're a log vdev. 3968654d025Sperrin */ 3978654d025Sperrin islog = 0; 3988654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 399990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 4008654d025Sperrin return (ENOTSUP); 401fa9e4066Sahrens 40288ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 40388ecc943SGeorge Wilson return (ENOTSUP); 40488ecc943SGeorge Wilson 40599653d4eSeschrock /* 4068654d025Sperrin * Set the nparity property for RAID-Z vdevs. 40799653d4eSeschrock */ 4088654d025Sperrin nparity = -1ULL; 40999653d4eSeschrock if (ops == &vdev_raidz_ops) { 41099653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4118654d025Sperrin &nparity) == 0) { 412*b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 41399653d4eSeschrock return (EINVAL); 41499653d4eSeschrock /* 415f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 416f94275ceSAdam Leventhal * device. 41799653d4eSeschrock */ 418f94275ceSAdam Leventhal if (nparity > 1 && 419f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 420f94275ceSAdam Leventhal return (ENOTSUP); 421f94275ceSAdam Leventhal if (nparity > 2 && 422f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 42399653d4eSeschrock return (ENOTSUP); 42499653d4eSeschrock } else { 42599653d4eSeschrock /* 42699653d4eSeschrock * We require the parity to be specified for SPAs that 42799653d4eSeschrock * support multiple parity levels. 42899653d4eSeschrock */ 429f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 43099653d4eSeschrock return (EINVAL); 43199653d4eSeschrock /* 43299653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 43399653d4eSeschrock */ 4348654d025Sperrin nparity = 1; 43599653d4eSeschrock } 43699653d4eSeschrock } else { 4378654d025Sperrin nparity = 0; 43899653d4eSeschrock } 4398654d025Sperrin ASSERT(nparity != -1ULL); 4408654d025Sperrin 4418654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4428654d025Sperrin 4438654d025Sperrin vd->vdev_islog = islog; 4448654d025Sperrin vd->vdev_nparity = nparity; 4458654d025Sperrin 4468654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4478654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4488654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4498654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4508654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4518654d025Sperrin &vd->vdev_physpath) == 0) 4528654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4536809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4546809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 45599653d4eSeschrock 456afefbcddSeschrock /* 457afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 458afefbcddSeschrock * as -1. 459afefbcddSeschrock */ 460afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 461afefbcddSeschrock &vd->vdev_wholedisk) != 0) 462afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 463afefbcddSeschrock 464ea8dc4b6Seschrock /* 465ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 466ea8dc4b6Seschrock * was not present at the time of import. 467ea8dc4b6Seschrock */ 4686809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4696809eb4eSEric Schrock &vd->vdev_not_present); 470ea8dc4b6Seschrock 471ecc2d604Sbonwick /* 472ecc2d604Sbonwick * Get the alignment requirement. 473ecc2d604Sbonwick */ 474ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 475ecc2d604Sbonwick 47688ecc943SGeorge Wilson /* 47788ecc943SGeorge Wilson * Retrieve the vdev creation time. 47888ecc943SGeorge Wilson */ 47988ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 48088ecc943SGeorge Wilson &vd->vdev_crtxg); 48188ecc943SGeorge Wilson 482fa9e4066Sahrens /* 483fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 484fa9e4066Sahrens */ 485fa9e4066Sahrens if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) { 486fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 487fa9e4066Sahrens &vd->vdev_ms_array); 488fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 489fa9e4066Sahrens &vd->vdev_ms_shift); 490fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 491fa9e4066Sahrens &vd->vdev_asize); 492fa9e4066Sahrens } 493fa9e4066Sahrens 494fa9e4066Sahrens /* 4953d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 496fa9e4066Sahrens */ 497c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 49821ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 49921ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 500c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 501c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5028ad4d6ddSJeff Bonwick &vd->vdev_dtl_smo.smo_object); 503c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 504c5904d13Seschrock &vd->vdev_unspare); 505c5904d13Seschrock } 50621ecdf64SLin Ling 50721ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 50821ecdf64SLin Ling uint64_t spare = 0; 50921ecdf64SLin Ling 51021ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 51121ecdf64SLin Ling &spare) == 0 && spare) 51221ecdf64SLin Ling spa_spare_add(vd); 51321ecdf64SLin Ling } 51421ecdf64SLin Ling 515ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 516ecc2d604Sbonwick &vd->vdev_offline); 517c5904d13Seschrock 5183d7072f8Seschrock /* 5193d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5203d7072f8Seschrock * state, as the diagnosis made on another system may not be 521069f55e2SEric Schrock * valid in the current context. Local vdevs will 522069f55e2SEric Schrock * remain in the faulted state. 5233d7072f8Seschrock */ 5243d7072f8Seschrock if (spa->spa_load_state == SPA_LOAD_OPEN) { 5253d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5263d7072f8Seschrock &vd->vdev_faulted); 5273d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5283d7072f8Seschrock &vd->vdev_degraded); 5293d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5303d7072f8Seschrock &vd->vdev_removed); 531069f55e2SEric Schrock 532069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 533069f55e2SEric Schrock char *aux; 534069f55e2SEric Schrock 535069f55e2SEric Schrock vd->vdev_label_aux = 536069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 537069f55e2SEric Schrock if (nvlist_lookup_string(nv, 538069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 539069f55e2SEric Schrock strcmp(aux, "external") == 0) 540069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 541069f55e2SEric Schrock } 5423d7072f8Seschrock } 543fa9e4066Sahrens } 544fa9e4066Sahrens 545fa9e4066Sahrens /* 546fa9e4066Sahrens * Add ourselves to the parent's list of children. 547fa9e4066Sahrens */ 548fa9e4066Sahrens vdev_add_child(parent, vd); 549fa9e4066Sahrens 55099653d4eSeschrock *vdp = vd; 55199653d4eSeschrock 55299653d4eSeschrock return (0); 553fa9e4066Sahrens } 554fa9e4066Sahrens 555fa9e4066Sahrens void 556fa9e4066Sahrens vdev_free(vdev_t *vd) 557fa9e4066Sahrens { 5583d7072f8Seschrock spa_t *spa = vd->vdev_spa; 559fa9e4066Sahrens 560fa9e4066Sahrens /* 561fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 562fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 563fa9e4066Sahrens */ 564fa9e4066Sahrens vdev_close(vd); 565fa9e4066Sahrens 566e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 567*b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 568fa9e4066Sahrens 569fa9e4066Sahrens /* 570fa9e4066Sahrens * Free all children. 571fa9e4066Sahrens */ 572573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 573fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 574fa9e4066Sahrens 575fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 576fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 577fa9e4066Sahrens 578fa9e4066Sahrens /* 579fa9e4066Sahrens * Discard allocation state. 580fa9e4066Sahrens */ 581fa9e4066Sahrens if (vd == vd->vdev_top) 582fa9e4066Sahrens vdev_metaslab_fini(vd); 583fa9e4066Sahrens 584fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 58599653d4eSeschrock ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 586fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 587fa9e4066Sahrens 588fa9e4066Sahrens /* 589fa9e4066Sahrens * Remove this vdev from its parent's child list. 590fa9e4066Sahrens */ 591fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 592fa9e4066Sahrens 593fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 594fa9e4066Sahrens 5953d7072f8Seschrock /* 5963d7072f8Seschrock * Clean up vdev structure. 5973d7072f8Seschrock */ 5983d7072f8Seschrock vdev_queue_fini(vd); 5993d7072f8Seschrock vdev_cache_fini(vd); 6003d7072f8Seschrock 6013d7072f8Seschrock if (vd->vdev_path) 6023d7072f8Seschrock spa_strfree(vd->vdev_path); 6033d7072f8Seschrock if (vd->vdev_devid) 6043d7072f8Seschrock spa_strfree(vd->vdev_devid); 6053d7072f8Seschrock if (vd->vdev_physpath) 6063d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6076809eb4eSEric Schrock if (vd->vdev_fru) 6086809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6093d7072f8Seschrock 6103d7072f8Seschrock if (vd->vdev_isspare) 6113d7072f8Seschrock spa_spare_remove(vd); 612fa94a07fSbrendan if (vd->vdev_isl2cache) 613fa94a07fSbrendan spa_l2cache_remove(vd); 6143d7072f8Seschrock 6153d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6163d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6178ad4d6ddSJeff Bonwick 6183d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6198ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6208ad4d6ddSJeff Bonwick space_map_unload(&vd->vdev_dtl[t]); 6218ad4d6ddSJeff Bonwick space_map_destroy(&vd->vdev_dtl[t]); 6228ad4d6ddSJeff Bonwick } 6233d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6248ad4d6ddSJeff Bonwick 6253d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6263d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 627e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6283d7072f8Seschrock 6293d7072f8Seschrock if (vd == spa->spa_root_vdev) 6303d7072f8Seschrock spa->spa_root_vdev = NULL; 6313d7072f8Seschrock 6323d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 633fa9e4066Sahrens } 634fa9e4066Sahrens 635fa9e4066Sahrens /* 636fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 637fa9e4066Sahrens */ 638fa9e4066Sahrens static void 639fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 640fa9e4066Sahrens { 641fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 642fa9e4066Sahrens metaslab_t *msp; 643fa9e4066Sahrens vdev_t *vd; 644fa9e4066Sahrens int t; 645fa9e4066Sahrens 646fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 647fa9e4066Sahrens 648fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 649fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 650fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 651fa9e4066Sahrens 652fa9e4066Sahrens svd->vdev_ms_array = 0; 653fa9e4066Sahrens svd->vdev_ms_shift = 0; 654fa9e4066Sahrens svd->vdev_ms_count = 0; 655fa9e4066Sahrens 656fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 657fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 658fa9e4066Sahrens 659fa9e4066Sahrens svd->vdev_mg = NULL; 660fa9e4066Sahrens svd->vdev_ms = NULL; 661ecc2d604Sbonwick 662ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 663ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 664fa9e4066Sahrens 665fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 666fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 66799653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 668fa9e4066Sahrens 669fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 670fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 67199653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 672fa9e4066Sahrens 673fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 674fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 675fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 676fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 677fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 678fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 679fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 680fa9e4066Sahrens } 681fa9e4066Sahrens 682e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 683fa9e4066Sahrens vdev_config_clean(svd); 684fa9e4066Sahrens vdev_config_dirty(tvd); 685fa9e4066Sahrens } 686fa9e4066Sahrens 687e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 688e14bb325SJeff Bonwick vdev_state_clean(svd); 689e14bb325SJeff Bonwick vdev_state_dirty(tvd); 690e14bb325SJeff Bonwick } 691e14bb325SJeff Bonwick 69299653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 69399653d4eSeschrock svd->vdev_deflate_ratio = 0; 6948654d025Sperrin 6958654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 6968654d025Sperrin svd->vdev_islog = 0; 697fa9e4066Sahrens } 698fa9e4066Sahrens 699fa9e4066Sahrens static void 700fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 701fa9e4066Sahrens { 702fa9e4066Sahrens if (vd == NULL) 703fa9e4066Sahrens return; 704fa9e4066Sahrens 705fa9e4066Sahrens vd->vdev_top = tvd; 706fa9e4066Sahrens 707573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 708fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 709fa9e4066Sahrens } 710fa9e4066Sahrens 711fa9e4066Sahrens /* 712fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 713fa9e4066Sahrens */ 714fa9e4066Sahrens vdev_t * 715fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 716fa9e4066Sahrens { 717fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 718fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 719fa9e4066Sahrens vdev_t *mvd; 720fa9e4066Sahrens 721e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 722fa9e4066Sahrens 723fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 724ecc2d604Sbonwick 725ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 726573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 727ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 728ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 72988ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 730ecc2d604Sbonwick 731fa9e4066Sahrens vdev_remove_child(pvd, cvd); 732fa9e4066Sahrens vdev_add_child(pvd, mvd); 733fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 734fa9e4066Sahrens vdev_add_child(mvd, cvd); 735fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 736fa9e4066Sahrens 737fa9e4066Sahrens if (mvd == mvd->vdev_top) 738fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 739fa9e4066Sahrens 740fa9e4066Sahrens return (mvd); 741fa9e4066Sahrens } 742fa9e4066Sahrens 743fa9e4066Sahrens /* 744fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 745fa9e4066Sahrens */ 746fa9e4066Sahrens void 747fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 748fa9e4066Sahrens { 749fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 750fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 751fa9e4066Sahrens 752e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 753fa9e4066Sahrens 754fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 755fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 75699653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 75799653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 758ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 759fa9e4066Sahrens 760fa9e4066Sahrens vdev_remove_child(mvd, cvd); 761fa9e4066Sahrens vdev_remove_child(pvd, mvd); 7628ad4d6ddSJeff Bonwick 76399653d4eSeschrock /* 764e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 765e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 766e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 767e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 76899653d4eSeschrock */ 7698ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 7708ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 7718ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 7728ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 7738ad4d6ddSJeff Bonwick } 774e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 775e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 776fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 777fa9e4066Sahrens 778fa9e4066Sahrens if (cvd == cvd->vdev_top) 779fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 780fa9e4066Sahrens 781fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 782fa9e4066Sahrens vdev_free(mvd); 783fa9e4066Sahrens } 784fa9e4066Sahrens 785ea8dc4b6Seschrock int 786fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 787fa9e4066Sahrens { 788fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 789ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 7908654d025Sperrin metaslab_class_t *mc; 791ecc2d604Sbonwick uint64_t m; 792fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 793fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 794ecc2d604Sbonwick metaslab_t **mspp; 795ecc2d604Sbonwick int error; 796fa9e4066Sahrens 79788ecc943SGeorge Wilson /* 79888ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 79988ecc943SGeorge Wilson */ 80088ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8010e34b6a7Sbonwick return (0); 8020e34b6a7Sbonwick 80388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 80488ecc943SGeorge Wilson 805e6ca193dSGeorge Wilson /* 806e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 807e6ca193dSGeorge Wilson * in 128k (1 << 17) because it is the current "typical" blocksize. 808e6ca193dSGeorge Wilson * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 809e6ca193dSGeorge Wilson * or we will inconsistently account for existing bp's. 810e6ca193dSGeorge Wilson */ 811e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 812e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 813e6ca193dSGeorge Wilson 814fa9e4066Sahrens ASSERT(oldc <= newc); 815fa9e4066Sahrens 8168654d025Sperrin if (vd->vdev_islog) 817*b24ab676SJeff Bonwick mc = spa_log_class(spa); 8188654d025Sperrin else 819*b24ab676SJeff Bonwick mc = spa_normal_class(spa); 8208654d025Sperrin 821ecc2d604Sbonwick if (vd->vdev_mg == NULL) 822ecc2d604Sbonwick vd->vdev_mg = metaslab_group_create(mc, vd); 823fa9e4066Sahrens 824ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 825fa9e4066Sahrens 826ecc2d604Sbonwick if (oldc != 0) { 827ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 828ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 829ecc2d604Sbonwick } 830fa9e4066Sahrens 831ecc2d604Sbonwick vd->vdev_ms = mspp; 832ecc2d604Sbonwick vd->vdev_ms_count = newc; 833fa9e4066Sahrens 834ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 835ecc2d604Sbonwick space_map_obj_t smo = { 0, 0, 0 }; 836ecc2d604Sbonwick if (txg == 0) { 837ecc2d604Sbonwick uint64_t object = 0; 838ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8397bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8407bfdf011SNeil Perrin DMU_READ_PREFETCH); 841ecc2d604Sbonwick if (error) 842ecc2d604Sbonwick return (error); 843ecc2d604Sbonwick if (object != 0) { 844ecc2d604Sbonwick dmu_buf_t *db; 845ecc2d604Sbonwick error = dmu_bonus_hold(mos, object, FTAG, &db); 846ecc2d604Sbonwick if (error) 847ecc2d604Sbonwick return (error); 8481934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (smo)); 8491934e92fSmaybee bcopy(db->db_data, &smo, sizeof (smo)); 850ecc2d604Sbonwick ASSERT3U(smo.smo_object, ==, object); 851ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 852fa9e4066Sahrens } 853fa9e4066Sahrens } 854ecc2d604Sbonwick vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 855ecc2d604Sbonwick m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 856fa9e4066Sahrens } 857fa9e4066Sahrens 858ea8dc4b6Seschrock return (0); 859fa9e4066Sahrens } 860fa9e4066Sahrens 861fa9e4066Sahrens void 862fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 863fa9e4066Sahrens { 864fa9e4066Sahrens uint64_t m; 865fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 866fa9e4066Sahrens 867fa9e4066Sahrens if (vd->vdev_ms != NULL) { 868fa9e4066Sahrens for (m = 0; m < count; m++) 869ecc2d604Sbonwick if (vd->vdev_ms[m] != NULL) 870ecc2d604Sbonwick metaslab_fini(vd->vdev_ms[m]); 871fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 872fa9e4066Sahrens vd->vdev_ms = NULL; 873fa9e4066Sahrens } 874fa9e4066Sahrens } 875fa9e4066Sahrens 876e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 877e14bb325SJeff Bonwick boolean_t vps_readable; 878e14bb325SJeff Bonwick boolean_t vps_writeable; 879e14bb325SJeff Bonwick int vps_flags; 880e14bb325SJeff Bonwick } vdev_probe_stats_t; 881e14bb325SJeff Bonwick 882e14bb325SJeff Bonwick static void 883e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 8840a4e9518Sgw { 8858ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 886a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 887e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 888a3f829aeSBill Moore 889a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 890e14bb325SJeff Bonwick 891e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 892e14bb325SJeff Bonwick if (zio->io_error == 0) 893e14bb325SJeff Bonwick vps->vps_readable = 1; 8948ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 895a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 896e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 897e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 898e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 899e14bb325SJeff Bonwick } else { 900e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 901e14bb325SJeff Bonwick } 902e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 903e14bb325SJeff Bonwick if (zio->io_error == 0) 904e14bb325SJeff Bonwick vps->vps_writeable = 1; 905e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 906e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 907a3f829aeSBill Moore zio_t *pio; 908e14bb325SJeff Bonwick 909e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 910e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 911e14bb325SJeff Bonwick 912e14bb325SJeff Bonwick if (vdev_readable(vd) && 9138ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 914e14bb325SJeff Bonwick zio->io_error = 0; 915e14bb325SJeff Bonwick } else { 916e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 917e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9188ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 919e14bb325SJeff Bonwick zio->io_error = ENXIO; 920e14bb325SJeff Bonwick } 921a3f829aeSBill Moore 922a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 923a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 924a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 925a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 926a3f829aeSBill Moore 927a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 928a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 929a3f829aeSBill Moore pio->io_error = ENXIO; 930a3f829aeSBill Moore 931e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 932e14bb325SJeff Bonwick } 933e14bb325SJeff Bonwick } 9340a4e9518Sgw 935e14bb325SJeff Bonwick /* 936e14bb325SJeff Bonwick * Determine whether this device is accessible by reading and writing 937e14bb325SJeff Bonwick * to several known locations: the pad regions of each vdev label 938e14bb325SJeff Bonwick * but the first (which we leave alone in case it contains a VTOC). 939e14bb325SJeff Bonwick */ 940e14bb325SJeff Bonwick zio_t * 941a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 942e14bb325SJeff Bonwick { 943e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 944a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 945a3f829aeSBill Moore zio_t *pio; 946a3f829aeSBill Moore 947a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9480a4e9518Sgw 949a3f829aeSBill Moore /* 950a3f829aeSBill Moore * Don't probe the probe. 951a3f829aeSBill Moore */ 952a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 953a3f829aeSBill Moore return (NULL); 954e14bb325SJeff Bonwick 955a3f829aeSBill Moore /* 956a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 957a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 958a3f829aeSBill Moore * this vdev will become parents of the probe io. 959a3f829aeSBill Moore */ 960a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 961e14bb325SJeff Bonwick 962a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 963a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 964a3f829aeSBill Moore 965a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 966a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 9678956713aSEric Schrock ZIO_FLAG_TRYHARD; 968a3f829aeSBill Moore 969a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 970a3f829aeSBill Moore /* 971a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 972a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 973a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 974a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 975a3f829aeSBill Moore * any zio looking at these values can assume that 976a3f829aeSBill Moore * failures persist for the life of the I/O. That's 977a3f829aeSBill Moore * important because when a device has intermittent 978a3f829aeSBill Moore * connectivity problems, we want to ensure that 979a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 980a3f829aeSBill Moore * the zio (EIO). 981a3f829aeSBill Moore * 982a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 983a3f829aeSBill Moore * values so the probe can reevaluate from first 984a3f829aeSBill Moore * principles. 985a3f829aeSBill Moore */ 986a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 987a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 988a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 989a3f829aeSBill Moore } 990a3f829aeSBill Moore 991a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 992a3f829aeSBill Moore vdev_probe_done, vps, 993a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 994a3f829aeSBill Moore 995a3f829aeSBill Moore if (zio != NULL) { 996a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 997a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 998a3f829aeSBill Moore } 999e14bb325SJeff Bonwick } 1000e14bb325SJeff Bonwick 1001a3f829aeSBill Moore if (zio != NULL) 1002a3f829aeSBill Moore zio_add_child(zio, pio); 1003e14bb325SJeff Bonwick 1004a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1005e14bb325SJeff Bonwick 1006a3f829aeSBill Moore if (vps == NULL) { 1007a3f829aeSBill Moore ASSERT(zio != NULL); 1008a3f829aeSBill Moore return (NULL); 1009a3f829aeSBill Moore } 1010e14bb325SJeff Bonwick 1011e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1012a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1013e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1014f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1015f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1016e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1017e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1018e14bb325SJeff Bonwick } 1019e14bb325SJeff Bonwick 1020a3f829aeSBill Moore if (zio == NULL) 1021a3f829aeSBill Moore return (pio); 1022a3f829aeSBill Moore 1023a3f829aeSBill Moore zio_nowait(pio); 1024a3f829aeSBill Moore return (NULL); 10250a4e9518Sgw } 10260a4e9518Sgw 1027f64c0e34SEric Taylor static void 1028f64c0e34SEric Taylor vdev_open_child(void *arg) 1029f64c0e34SEric Taylor { 1030f64c0e34SEric Taylor vdev_t *vd = arg; 1031f64c0e34SEric Taylor 1032f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1033f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1034f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1035f64c0e34SEric Taylor } 1036f64c0e34SEric Taylor 1037681d9761SEric Taylor boolean_t 1038681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1039681d9761SEric Taylor { 1040681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1041681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1042681d9761SEric Taylor return (B_TRUE); 1043681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1044681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1045681d9761SEric Taylor return (B_TRUE); 1046681d9761SEric Taylor return (B_FALSE); 1047681d9761SEric Taylor } 1048681d9761SEric Taylor 1049f64c0e34SEric Taylor void 1050f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1051f64c0e34SEric Taylor { 1052f64c0e34SEric Taylor taskq_t *tq; 1053f64c0e34SEric Taylor int children = vd->vdev_children; 1054f64c0e34SEric Taylor 1055681d9761SEric Taylor /* 1056681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1057681d9761SEric Taylor * in a single thread so that the same thread holds the 1058681d9761SEric Taylor * spa_namespace_lock 1059681d9761SEric Taylor */ 1060681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1061681d9761SEric Taylor for (int c = 0; c < children; c++) 1062681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1063681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1064681d9761SEric Taylor return; 1065681d9761SEric Taylor } 1066f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1067f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1068f64c0e34SEric Taylor 1069f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1070f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1071f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1072f64c0e34SEric Taylor 1073f64c0e34SEric Taylor taskq_destroy(tq); 1074f64c0e34SEric Taylor } 1075f64c0e34SEric Taylor 1076fa9e4066Sahrens /* 1077fa9e4066Sahrens * Prepare a virtual device for access. 1078fa9e4066Sahrens */ 1079fa9e4066Sahrens int 1080fa9e4066Sahrens vdev_open(vdev_t *vd) 1081fa9e4066Sahrens { 10828ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1083fa9e4066Sahrens int error; 1084fa9e4066Sahrens uint64_t osize = 0; 1085fa9e4066Sahrens uint64_t asize, psize; 1086ecc2d604Sbonwick uint64_t ashift = 0; 1087fa9e4066Sahrens 1088f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1089f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1090fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1091fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1092fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1093fa9e4066Sahrens 1094fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1095e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1096e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1097573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1098fa9e4066Sahrens 1099069f55e2SEric Schrock /* 1100069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1101069f55e2SEric Schrock * faulted, bail out of the open. 1102069f55e2SEric Schrock */ 11033d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11043d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1105069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1106069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11073d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1108069f55e2SEric Schrock vd->vdev_label_aux); 11093d7072f8Seschrock return (ENXIO); 11103d7072f8Seschrock } else if (vd->vdev_offline) { 1111fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1112ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1113fa9e4066Sahrens return (ENXIO); 1114fa9e4066Sahrens } 1115fa9e4066Sahrens 1116fa9e4066Sahrens error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 1117fa9e4066Sahrens 1118095bcd66SGeorge Wilson /* 1119095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1120095bcd66SGeorge Wilson * the vdev on error. 1121095bcd66SGeorge Wilson */ 1122095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1123ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11248956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1125ea8dc4b6Seschrock 1126fa9e4066Sahrens if (error) { 11273d7072f8Seschrock if (vd->vdev_removed && 11283d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11293d7072f8Seschrock vd->vdev_removed = B_FALSE; 11303d7072f8Seschrock 1131ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1132fa9e4066Sahrens vd->vdev_stat.vs_aux); 1133fa9e4066Sahrens return (error); 1134fa9e4066Sahrens } 1135fa9e4066Sahrens 11363d7072f8Seschrock vd->vdev_removed = B_FALSE; 11373d7072f8Seschrock 1138096d22d4SEric Schrock /* 1139096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1140096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1141096d22d4SEric Schrock */ 1142096d22d4SEric Schrock if (vd->vdev_faulted) { 1143096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1144096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1145096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1146096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1147096d22d4SEric Schrock vd->vdev_label_aux); 1148096d22d4SEric Schrock return (ENXIO); 1149096d22d4SEric Schrock } 1150096d22d4SEric Schrock 11513d7072f8Seschrock if (vd->vdev_degraded) { 11523d7072f8Seschrock ASSERT(vd->vdev_children == 0); 11533d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 11543d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 11553d7072f8Seschrock } else { 1156069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 11573d7072f8Seschrock } 1158fa9e4066Sahrens 115988ecc943SGeorge Wilson /* 116088ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 116188ecc943SGeorge Wilson */ 116288ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 116388ecc943SGeorge Wilson return (0); 116488ecc943SGeorge Wilson 1165573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1166ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1167ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1168ea8dc4b6Seschrock VDEV_AUX_NONE); 1169ea8dc4b6Seschrock break; 1170ea8dc4b6Seschrock } 1171573ca77eSGeorge Wilson } 1172fa9e4066Sahrens 1173fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1174fa9e4066Sahrens 1175fa9e4066Sahrens if (vd->vdev_children == 0) { 1176fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1177ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1178ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1179fa9e4066Sahrens return (EOVERFLOW); 1180fa9e4066Sahrens } 1181fa9e4066Sahrens psize = osize; 1182fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1183fa9e4066Sahrens } else { 1184ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1185fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1186ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1187ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1188fa9e4066Sahrens return (EOVERFLOW); 1189fa9e4066Sahrens } 1190fa9e4066Sahrens psize = 0; 1191fa9e4066Sahrens asize = osize; 1192fa9e4066Sahrens } 1193fa9e4066Sahrens 1194fa9e4066Sahrens vd->vdev_psize = psize; 1195fa9e4066Sahrens 1196573ca77eSGeorge Wilson /* 1197573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1198573ca77eSGeorge Wilson */ 1199573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1200573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1201573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1202573ca77eSGeorge Wilson return (EINVAL); 1203573ca77eSGeorge Wilson } 1204573ca77eSGeorge Wilson 1205fa9e4066Sahrens if (vd->vdev_asize == 0) { 1206fa9e4066Sahrens /* 1207fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1208ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1209fa9e4066Sahrens */ 1210fa9e4066Sahrens vd->vdev_asize = asize; 1211ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1212fa9e4066Sahrens } else { 1213fa9e4066Sahrens /* 1214fa9e4066Sahrens * Make sure the alignment requirement hasn't increased. 1215fa9e4066Sahrens */ 1216ecc2d604Sbonwick if (ashift > vd->vdev_top->vdev_ashift) { 1217ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1218ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 1219fa9e4066Sahrens return (EINVAL); 1220fa9e4066Sahrens } 1221573ca77eSGeorge Wilson } 1222fa9e4066Sahrens 1223573ca77eSGeorge Wilson /* 1224573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1225573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1226573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1227573ca77eSGeorge Wilson */ 1228573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1229573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1230573ca77eSGeorge Wilson vd->vdev_asize = asize; 1231fa9e4066Sahrens 1232573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1233fa9e4066Sahrens 12340a4e9518Sgw /* 12350a4e9518Sgw * Ensure we can issue some IO before declaring the 12360a4e9518Sgw * vdev open for business. 12370a4e9518Sgw */ 1238e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1239e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 12400a4e9518Sgw vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1241e14bb325SJeff Bonwick VDEV_AUX_IO_FAILURE); 12420a4e9518Sgw return (error); 12430a4e9518Sgw } 12440a4e9518Sgw 1245088f3894Sahrens /* 1246088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 12478ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 12488ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1249088f3894Sahrens */ 12508ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 12518ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 12528ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1253088f3894Sahrens 1254fa9e4066Sahrens return (0); 1255fa9e4066Sahrens } 1256fa9e4066Sahrens 1257560e6e96Seschrock /* 1258560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1259560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 12603d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1261560e6e96Seschrock * 1262560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1263560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1264560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1265560e6e96Seschrock * will be updated but the function will return 0. 1266560e6e96Seschrock */ 1267560e6e96Seschrock int 1268560e6e96Seschrock vdev_validate(vdev_t *vd) 1269560e6e96Seschrock { 1270560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1271560e6e96Seschrock nvlist_t *label; 1272e14bb325SJeff Bonwick uint64_t guid, top_guid; 1273560e6e96Seschrock uint64_t state; 1274560e6e96Seschrock 1275573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1276560e6e96Seschrock if (vdev_validate(vd->vdev_child[c]) != 0) 12770bf246f5Smc return (EBADF); 1278560e6e96Seschrock 1279b5989ec7Seschrock /* 1280b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1281b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1282b5989ec7Seschrock * overwrite the previous state. 1283b5989ec7Seschrock */ 1284e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1285560e6e96Seschrock 1286560e6e96Seschrock if ((label = vdev_label_read_config(vd)) == NULL) { 1287560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1288560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1289560e6e96Seschrock return (0); 1290560e6e96Seschrock } 1291560e6e96Seschrock 1292560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1293560e6e96Seschrock &guid) != 0 || guid != spa_guid(spa)) { 1294560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1295560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1296560e6e96Seschrock nvlist_free(label); 1297560e6e96Seschrock return (0); 1298560e6e96Seschrock } 1299560e6e96Seschrock 1300e14bb325SJeff Bonwick /* 1301e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1302e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1303e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1304e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1305e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1306e14bb325SJeff Bonwick * safely compare to that instead. 1307e14bb325SJeff Bonwick */ 1308560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1309e14bb325SJeff Bonwick &guid) != 0 || 1310e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1311e14bb325SJeff Bonwick &top_guid) != 0 || 1312e14bb325SJeff Bonwick (vd->vdev_guid != guid && 1313e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1314560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1315560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1316560e6e96Seschrock nvlist_free(label); 1317560e6e96Seschrock return (0); 1318560e6e96Seschrock } 1319560e6e96Seschrock 1320560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1321560e6e96Seschrock &state) != 0) { 1322560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1323560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1324560e6e96Seschrock nvlist_free(label); 1325560e6e96Seschrock return (0); 1326560e6e96Seschrock } 1327560e6e96Seschrock 1328560e6e96Seschrock nvlist_free(label); 1329560e6e96Seschrock 1330bc758434SLin Ling /* 1331bc758434SLin Ling * If spa->spa_load_verbatim is true, no need to check the 1332bc758434SLin Ling * state of the pool. 1333bc758434SLin Ling */ 1334bc758434SLin Ling if (!spa->spa_load_verbatim && 1335bc758434SLin Ling spa->spa_load_state == SPA_LOAD_OPEN && 1336bc758434SLin Ling state != POOL_STATE_ACTIVE) 13370bf246f5Smc return (EBADF); 1338560e6e96Seschrock 133951ece835Seschrock /* 134051ece835Seschrock * If we were able to open and validate a vdev that was 134151ece835Seschrock * previously marked permanently unavailable, clear that state 134251ece835Seschrock * now. 134351ece835Seschrock */ 134451ece835Seschrock if (vd->vdev_not_present) 134551ece835Seschrock vd->vdev_not_present = 0; 134651ece835Seschrock } 1347560e6e96Seschrock 1348560e6e96Seschrock return (0); 1349560e6e96Seschrock } 1350560e6e96Seschrock 1351fa9e4066Sahrens /* 1352fa9e4066Sahrens * Close a virtual device. 1353fa9e4066Sahrens */ 1354fa9e4066Sahrens void 1355fa9e4066Sahrens vdev_close(vdev_t *vd) 1356fa9e4066Sahrens { 13578ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1358095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 13598ad4d6ddSJeff Bonwick 13608ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 13618ad4d6ddSJeff Bonwick 1362095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 1363095bcd66SGeorge Wilson vd->vdev_reopening = pvd->vdev_reopening; 1364095bcd66SGeorge Wilson 1365fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1366fa9e4066Sahrens 13673d7072f8Seschrock vdev_cache_purge(vd); 1368fa9e4066Sahrens 1369560e6e96Seschrock /* 1370573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1371560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1372560e6e96Seschrock * it's still faulted. 1373560e6e96Seschrock */ 1374560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1375560e6e96Seschrock 1376fa9e4066Sahrens if (vd->vdev_offline) 1377fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1378fa9e4066Sahrens else 1379fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1380ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1381fa9e4066Sahrens } 1382fa9e4066Sahrens 1383095bcd66SGeorge Wilson /* 1384095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1385095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1386095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1387095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1388095bcd66SGeorge Wilson */ 1389fa9e4066Sahrens void 1390ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1391fa9e4066Sahrens { 1392ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1393fa9e4066Sahrens 1394e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1395ea8dc4b6Seschrock 1396095bcd66SGeorge Wilson vd->vdev_reopening = B_TRUE; 1397fa9e4066Sahrens vdev_close(vd); 1398fa9e4066Sahrens (void) vdev_open(vd); 1399fa9e4066Sahrens 140039c23413Seschrock /* 140139c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 140239c23413Seschrock * Otherwise, a device with an invalid label could be successfully 140339c23413Seschrock * opened in response to vdev_reopen(). 140439c23413Seschrock */ 1405c5904d13Seschrock if (vd->vdev_aux) { 1406c5904d13Seschrock (void) vdev_validate_aux(vd); 1407e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 14086809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1409573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1410573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1411c5904d13Seschrock } else { 1412c5904d13Seschrock (void) vdev_validate(vd); 1413c5904d13Seschrock } 141439c23413Seschrock 1415fa9e4066Sahrens /* 14163d7072f8Seschrock * Reassess parent vdev's health. 1417fa9e4066Sahrens */ 14183d7072f8Seschrock vdev_propagate_state(vd); 1419fa9e4066Sahrens } 1420fa9e4066Sahrens 1421fa9e4066Sahrens int 142299653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1423fa9e4066Sahrens { 1424fa9e4066Sahrens int error; 1425fa9e4066Sahrens 1426fa9e4066Sahrens /* 1427fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1428fa9e4066Sahrens * For a create, however, we want to fail the request if 1429fa9e4066Sahrens * there are any components we can't open. 1430fa9e4066Sahrens */ 1431fa9e4066Sahrens error = vdev_open(vd); 1432fa9e4066Sahrens 1433fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1434fa9e4066Sahrens vdev_close(vd); 1435fa9e4066Sahrens return (error ? error : ENXIO); 1436fa9e4066Sahrens } 1437fa9e4066Sahrens 1438fa9e4066Sahrens /* 1439fa9e4066Sahrens * Recursively initialize all labels. 1440fa9e4066Sahrens */ 144139c23413Seschrock if ((error = vdev_label_init(vd, txg, isreplacing ? 144239c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1443fa9e4066Sahrens vdev_close(vd); 1444fa9e4066Sahrens return (error); 1445fa9e4066Sahrens } 1446fa9e4066Sahrens 1447fa9e4066Sahrens return (0); 1448fa9e4066Sahrens } 1449fa9e4066Sahrens 14500e34b6a7Sbonwick void 1451573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1452fa9e4066Sahrens { 1453fa9e4066Sahrens /* 1454fa9e4066Sahrens * Aim for roughly 200 metaslabs per vdev. 1455fa9e4066Sahrens */ 1456fa9e4066Sahrens vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1457fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1458fa9e4066Sahrens } 1459fa9e4066Sahrens 1460fa9e4066Sahrens void 1461ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1462fa9e4066Sahrens { 1463ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 146488ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1465ecc2d604Sbonwick ASSERT(ISP2(flags)); 1466fa9e4066Sahrens 1467ecc2d604Sbonwick if (flags & VDD_METASLAB) 1468ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1469ecc2d604Sbonwick 1470ecc2d604Sbonwick if (flags & VDD_DTL) 1471ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1472ecc2d604Sbonwick 1473ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1474fa9e4066Sahrens } 1475fa9e4066Sahrens 14768ad4d6ddSJeff Bonwick /* 14778ad4d6ddSJeff Bonwick * DTLs. 14788ad4d6ddSJeff Bonwick * 14798ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 14808ad4d6ddSJeff Bonwick * the vdev has less than perfect replication. There are three kinds of DTL: 14818ad4d6ddSJeff Bonwick * 14828ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 14838ad4d6ddSJeff Bonwick * 14848ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 14858ad4d6ddSJeff Bonwick * 14868ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 14878ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 14888ad4d6ddSJeff Bonwick * txgs that was scrubbed. 14898ad4d6ddSJeff Bonwick * 14908ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 14918ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 14928ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 14938ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 14948ad4d6ddSJeff Bonwick * determine whether a device can be detached. 14958ad4d6ddSJeff Bonwick * 14968ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 14978ad4d6ddSJeff Bonwick * either has the data or it doesn't. 14988ad4d6ddSJeff Bonwick * 14998ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 15008ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 15018ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 15028ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 15038ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 15048ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 15058ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 15068ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 15078ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 15088ad4d6ddSJeff Bonwick * 15098ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 15108ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 15118ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 15128ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 15138ad4d6ddSJeff Bonwick */ 1514fa9e4066Sahrens void 15158ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1516fa9e4066Sahrens { 15178ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 15188ad4d6ddSJeff Bonwick 15198ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 15208ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 15218ad4d6ddSJeff Bonwick 1522fa9e4066Sahrens mutex_enter(sm->sm_lock); 1523fa9e4066Sahrens if (!space_map_contains(sm, txg, size)) 1524fa9e4066Sahrens space_map_add(sm, txg, size); 1525fa9e4066Sahrens mutex_exit(sm->sm_lock); 1526fa9e4066Sahrens } 1527fa9e4066Sahrens 15288ad4d6ddSJeff Bonwick boolean_t 15298ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1530fa9e4066Sahrens { 15318ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 15328ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1533fa9e4066Sahrens 15348ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 15358ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1536fa9e4066Sahrens 1537fa9e4066Sahrens mutex_enter(sm->sm_lock); 15388ad4d6ddSJeff Bonwick if (sm->sm_space != 0) 15398ad4d6ddSJeff Bonwick dirty = space_map_contains(sm, txg, size); 1540fa9e4066Sahrens mutex_exit(sm->sm_lock); 1541fa9e4066Sahrens 1542fa9e4066Sahrens return (dirty); 1543fa9e4066Sahrens } 1544fa9e4066Sahrens 15458ad4d6ddSJeff Bonwick boolean_t 15468ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 15478ad4d6ddSJeff Bonwick { 15488ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 15498ad4d6ddSJeff Bonwick boolean_t empty; 15508ad4d6ddSJeff Bonwick 15518ad4d6ddSJeff Bonwick mutex_enter(sm->sm_lock); 15528ad4d6ddSJeff Bonwick empty = (sm->sm_space == 0); 15538ad4d6ddSJeff Bonwick mutex_exit(sm->sm_lock); 15548ad4d6ddSJeff Bonwick 15558ad4d6ddSJeff Bonwick return (empty); 15568ad4d6ddSJeff Bonwick } 15578ad4d6ddSJeff Bonwick 1558fa9e4066Sahrens /* 1559fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1560fa9e4066Sahrens */ 1561fa9e4066Sahrens void 1562fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1563fa9e4066Sahrens { 1564ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 15658ad4d6ddSJeff Bonwick avl_tree_t reftree; 15668ad4d6ddSJeff Bonwick int minref; 1567fa9e4066Sahrens 15688ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1569fa9e4066Sahrens 15708ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 15718ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 15728ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 15738ad4d6ddSJeff Bonwick 1574*b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 15758ad4d6ddSJeff Bonwick return; 15768ad4d6ddSJeff Bonwick 15778ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 1578fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1579088f3894Sahrens if (scrub_txg != 0 && 1580088f3894Sahrens (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { 1581088f3894Sahrens /* XXX should check scrub_done? */ 1582088f3894Sahrens /* 1583088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1584088f3894Sahrens * did it without rebooting, then the scrub dtl 1585088f3894Sahrens * will be valid, so excise the old region and 1586088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1587088f3894Sahrens * dtl as-is if there was an error. 15888ad4d6ddSJeff Bonwick * 15898ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 15908ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 15918ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 15928ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 15938ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 15948ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 15958ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 15968ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 15978ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1598088f3894Sahrens */ 15998ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 16008ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16018ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16028ad4d6ddSJeff Bonwick space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 16038ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16048ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_SCRUB], 2); 16058ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, 16068ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16078ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1608fa9e4066Sahrens } 16098ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 16108ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 16118ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1612fa9e4066Sahrens if (scrub_done) 16138ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 16148ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 16158ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 16168ad4d6ddSJeff Bonwick space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 16178ad4d6ddSJeff Bonwick else 16188ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 16198ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1620fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1621088f3894Sahrens 1622ecc2d604Sbonwick if (txg != 0) 1623ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1624fa9e4066Sahrens return; 1625fa9e4066Sahrens } 1626fa9e4066Sahrens 1627fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 16288ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 162999bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 163099bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 16318ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 16328ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 16338ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 16348ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 16358ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 16368ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 16378ad4d6ddSJeff Bonwick else 16388ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 16398ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 16408ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 16418ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 16428ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 164399bb17e2SEric Taylor space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 16448ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 16458ad4d6ddSJeff Bonwick } 16468ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 16478ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1648fa9e4066Sahrens } 16498ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1650fa9e4066Sahrens } 1651fa9e4066Sahrens 1652fa9e4066Sahrens static int 1653fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1654fa9e4066Sahrens { 1655fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 16568ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 1657ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1658fa9e4066Sahrens dmu_buf_t *db; 1659fa9e4066Sahrens int error; 1660fa9e4066Sahrens 1661fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1662fa9e4066Sahrens 1663fa9e4066Sahrens if (smo->smo_object == 0) 1664fa9e4066Sahrens return (0); 1665fa9e4066Sahrens 166688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 166788ecc943SGeorge Wilson 1668ecc2d604Sbonwick if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1669ea8dc4b6Seschrock return (error); 1670ecc2d604Sbonwick 16711934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 16721934e92fSmaybee bcopy(db->db_data, smo, sizeof (*smo)); 1673ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1674fa9e4066Sahrens 1675fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 16768ad4d6ddSJeff Bonwick error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 16778ad4d6ddSJeff Bonwick NULL, SM_ALLOC, smo, mos); 1678fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1679fa9e4066Sahrens 1680fa9e4066Sahrens return (error); 1681fa9e4066Sahrens } 1682fa9e4066Sahrens 1683fa9e4066Sahrens void 1684fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1685fa9e4066Sahrens { 1686fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 16878ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 16888ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1689ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1690fa9e4066Sahrens space_map_t smsync; 1691fa9e4066Sahrens kmutex_t smlock; 1692fa9e4066Sahrens dmu_buf_t *db; 1693fa9e4066Sahrens dmu_tx_t *tx; 1694fa9e4066Sahrens 169588ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 169688ecc943SGeorge Wilson 1697fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1698fa9e4066Sahrens 1699fa9e4066Sahrens if (vd->vdev_detached) { 1700fa9e4066Sahrens if (smo->smo_object != 0) { 1701ecc2d604Sbonwick int err = dmu_object_free(mos, smo->smo_object, tx); 1702fa9e4066Sahrens ASSERT3U(err, ==, 0); 1703fa9e4066Sahrens smo->smo_object = 0; 1704fa9e4066Sahrens } 1705fa9e4066Sahrens dmu_tx_commit(tx); 1706fa9e4066Sahrens return; 1707fa9e4066Sahrens } 1708fa9e4066Sahrens 1709fa9e4066Sahrens if (smo->smo_object == 0) { 1710fa9e4066Sahrens ASSERT(smo->smo_objsize == 0); 1711fa9e4066Sahrens ASSERT(smo->smo_alloc == 0); 1712ecc2d604Sbonwick smo->smo_object = dmu_object_alloc(mos, 1713fa9e4066Sahrens DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1714fa9e4066Sahrens DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1715fa9e4066Sahrens ASSERT(smo->smo_object != 0); 1716fa9e4066Sahrens vdev_config_dirty(vd->vdev_top); 1717fa9e4066Sahrens } 1718fa9e4066Sahrens 1719fa9e4066Sahrens mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1720fa9e4066Sahrens 1721fa9e4066Sahrens space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1722fa9e4066Sahrens &smlock); 1723fa9e4066Sahrens 1724fa9e4066Sahrens mutex_enter(&smlock); 1725fa9e4066Sahrens 1726fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1727ecc2d604Sbonwick space_map_walk(sm, space_map_add, &smsync); 1728fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1729fa9e4066Sahrens 1730ecc2d604Sbonwick space_map_truncate(smo, mos, tx); 1731ecc2d604Sbonwick space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1732fa9e4066Sahrens 1733fa9e4066Sahrens space_map_destroy(&smsync); 1734fa9e4066Sahrens 1735fa9e4066Sahrens mutex_exit(&smlock); 1736fa9e4066Sahrens mutex_destroy(&smlock); 1737fa9e4066Sahrens 1738ecc2d604Sbonwick VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1739fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 17401934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 17411934e92fSmaybee bcopy(smo, db->db_data, sizeof (*smo)); 1742ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1743fa9e4066Sahrens 1744fa9e4066Sahrens dmu_tx_commit(tx); 1745fa9e4066Sahrens } 1746fa9e4066Sahrens 17478ad4d6ddSJeff Bonwick /* 17488ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 17498ad4d6ddSJeff Bonwick * without losing data. 17508ad4d6ddSJeff Bonwick */ 17518ad4d6ddSJeff Bonwick boolean_t 17528ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 17538ad4d6ddSJeff Bonwick { 17548ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 17558ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 17568ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 17578ad4d6ddSJeff Bonwick boolean_t required; 17588ad4d6ddSJeff Bonwick 17598ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 17608ad4d6ddSJeff Bonwick 17618ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 17628ad4d6ddSJeff Bonwick return (B_TRUE); 17638ad4d6ddSJeff Bonwick 17648ad4d6ddSJeff Bonwick /* 17658ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 17668ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 17678ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 17688ad4d6ddSJeff Bonwick */ 17698ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 17708ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 17718ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 17728ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 17738ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 17748ad4d6ddSJeff Bonwick 17758ad4d6ddSJeff Bonwick return (required); 17768ad4d6ddSJeff Bonwick } 17778ad4d6ddSJeff Bonwick 1778088f3894Sahrens /* 1779088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 1780088f3894Sahrens */ 1781088f3894Sahrens boolean_t 1782088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1783088f3894Sahrens { 1784088f3894Sahrens boolean_t needed = B_FALSE; 1785088f3894Sahrens uint64_t thismin = UINT64_MAX; 1786088f3894Sahrens uint64_t thismax = 0; 1787088f3894Sahrens 1788088f3894Sahrens if (vd->vdev_children == 0) { 1789088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 17908ad4d6ddSJeff Bonwick if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 17918ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 1792088f3894Sahrens space_seg_t *ss; 1793088f3894Sahrens 17948ad4d6ddSJeff Bonwick ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1795088f3894Sahrens thismin = ss->ss_start - 1; 17968ad4d6ddSJeff Bonwick ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1797088f3894Sahrens thismax = ss->ss_end; 1798088f3894Sahrens needed = B_TRUE; 1799088f3894Sahrens } 1800088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 1801088f3894Sahrens } else { 18028ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 1803088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 1804088f3894Sahrens uint64_t cmin, cmax; 1805088f3894Sahrens 1806088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1807088f3894Sahrens thismin = MIN(thismin, cmin); 1808088f3894Sahrens thismax = MAX(thismax, cmax); 1809088f3894Sahrens needed = B_TRUE; 1810088f3894Sahrens } 1811088f3894Sahrens } 1812088f3894Sahrens } 1813088f3894Sahrens 1814088f3894Sahrens if (needed && minp) { 1815088f3894Sahrens *minp = thismin; 1816088f3894Sahrens *maxp = thismax; 1817088f3894Sahrens } 1818088f3894Sahrens return (needed); 1819088f3894Sahrens } 1820088f3894Sahrens 1821560e6e96Seschrock void 1822ea8dc4b6Seschrock vdev_load(vdev_t *vd) 1823fa9e4066Sahrens { 1824fa9e4066Sahrens /* 1825fa9e4066Sahrens * Recursively load all children. 1826fa9e4066Sahrens */ 18278ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1828560e6e96Seschrock vdev_load(vd->vdev_child[c]); 1829fa9e4066Sahrens 1830fa9e4066Sahrens /* 18310e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 1832fa9e4066Sahrens */ 183388ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 1834560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1835560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 1836560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1837560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1838fa9e4066Sahrens 1839fa9e4066Sahrens /* 1840fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 1841fa9e4066Sahrens */ 1842560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1843560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1844560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1845fa9e4066Sahrens } 1846fa9e4066Sahrens 184799653d4eSeschrock /* 1848fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 1849fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 1850fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 1851fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 1852fa94a07fSbrendan * repurposed to another pool. 185399653d4eSeschrock */ 185499653d4eSeschrock int 1855fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 185699653d4eSeschrock { 185799653d4eSeschrock nvlist_t *label; 185899653d4eSeschrock uint64_t guid, version; 185999653d4eSeschrock uint64_t state; 186099653d4eSeschrock 1861e14bb325SJeff Bonwick if (!vdev_readable(vd)) 1862c5904d13Seschrock return (0); 1863c5904d13Seschrock 186499653d4eSeschrock if ((label = vdev_label_read_config(vd)) == NULL) { 186599653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 186699653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 186799653d4eSeschrock return (-1); 186899653d4eSeschrock } 186999653d4eSeschrock 187099653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1871e7437265Sahrens version > SPA_VERSION || 187299653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 187399653d4eSeschrock guid != vd->vdev_guid || 187499653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 187599653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 187699653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 187799653d4eSeschrock nvlist_free(label); 187899653d4eSeschrock return (-1); 187999653d4eSeschrock } 188099653d4eSeschrock 188199653d4eSeschrock /* 188299653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 188399653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 188499653d4eSeschrock */ 188599653d4eSeschrock nvlist_free(label); 188699653d4eSeschrock return (0); 188799653d4eSeschrock } 188899653d4eSeschrock 188988ecc943SGeorge Wilson void 189088ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 189188ecc943SGeorge Wilson { 189288ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 189388ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 189488ecc943SGeorge Wilson dmu_tx_t *tx; 189588ecc943SGeorge Wilson 189688ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 189788ecc943SGeorge Wilson 189888ecc943SGeorge Wilson if (vd->vdev_dtl_smo.smo_object) { 189988ecc943SGeorge Wilson ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); 190088ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 190188ecc943SGeorge Wilson vd->vdev_dtl_smo.smo_object = 0; 190288ecc943SGeorge Wilson } 190388ecc943SGeorge Wilson 190488ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 190588ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 190688ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 190788ecc943SGeorge Wilson 190888ecc943SGeorge Wilson if (msp == NULL || msp->ms_smo.smo_object == 0) 190988ecc943SGeorge Wilson continue; 191088ecc943SGeorge Wilson 191188ecc943SGeorge Wilson ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); 191288ecc943SGeorge Wilson (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 191388ecc943SGeorge Wilson msp->ms_smo.smo_object = 0; 191488ecc943SGeorge Wilson } 191588ecc943SGeorge Wilson } 191688ecc943SGeorge Wilson 191788ecc943SGeorge Wilson if (vd->vdev_ms_array) { 191888ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 191988ecc943SGeorge Wilson vd->vdev_ms_array = 0; 192088ecc943SGeorge Wilson vd->vdev_ms_shift = 0; 192188ecc943SGeorge Wilson } 192288ecc943SGeorge Wilson dmu_tx_commit(tx); 192388ecc943SGeorge Wilson } 192488ecc943SGeorge Wilson 1925fa9e4066Sahrens void 1926fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 1927fa9e4066Sahrens { 1928fa9e4066Sahrens metaslab_t *msp; 1929fa9e4066Sahrens 193088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 193188ecc943SGeorge Wilson 1932fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 1933fa9e4066Sahrens metaslab_sync_done(msp, txg); 1934fa9e4066Sahrens } 1935fa9e4066Sahrens 1936fa9e4066Sahrens void 1937fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 1938fa9e4066Sahrens { 1939fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1940fa9e4066Sahrens vdev_t *lvd; 1941fa9e4066Sahrens metaslab_t *msp; 1942ecc2d604Sbonwick dmu_tx_t *tx; 1943fa9e4066Sahrens 194488ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 194588ecc943SGeorge Wilson 1946ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 1947ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 1948ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1949ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 1950ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 1951ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 1952ecc2d604Sbonwick vdev_config_dirty(vd); 1953ecc2d604Sbonwick dmu_tx_commit(tx); 1954ecc2d604Sbonwick } 1955fa9e4066Sahrens 195688ecc943SGeorge Wilson if (vd->vdev_removing) 195788ecc943SGeorge Wilson vdev_remove(vd, txg); 195888ecc943SGeorge Wilson 1959ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 1960fa9e4066Sahrens metaslab_sync(msp, txg); 1961ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 1962ecc2d604Sbonwick } 1963fa9e4066Sahrens 1964fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 1965fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 1966fa9e4066Sahrens 1967fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 1968fa9e4066Sahrens } 1969fa9e4066Sahrens 1970fa9e4066Sahrens uint64_t 1971fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 1972fa9e4066Sahrens { 1973fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 1974fa9e4066Sahrens } 1975fa9e4066Sahrens 19763d7072f8Seschrock /* 19773d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 19783d7072f8Seschrock * not be opened, and no I/O is attempted. 19793d7072f8Seschrock */ 1980fa9e4066Sahrens int 1981069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 1982fa9e4066Sahrens { 1983c5904d13Seschrock vdev_t *vd; 1984fa9e4066Sahrens 19858f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 1986fa9e4066Sahrens 1987c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1988e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1989e14bb325SJeff Bonwick 19903d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 1991e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1992fa9e4066Sahrens 1993069f55e2SEric Schrock /* 1994069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 1995069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 1996069f55e2SEric Schrock * were faulted. 1997069f55e2SEric Schrock */ 1998069f55e2SEric Schrock vd->vdev_label_aux = aux; 1999069f55e2SEric Schrock 20003d7072f8Seschrock /* 20013d7072f8Seschrock * Faulted state takes precedence over degraded. 20023d7072f8Seschrock */ 20033d7072f8Seschrock vd->vdev_faulted = 1ULL; 20043d7072f8Seschrock vd->vdev_degraded = 0ULL; 2005069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 20063d7072f8Seschrock 20073d7072f8Seschrock /* 20086988b9faSDavid Marker * If marking the vdev as faulted cause the top-level vdev to become 20093d7072f8Seschrock * unavailable, then back off and simply mark the vdev as degraded 20103d7072f8Seschrock * instead. 20113d7072f8Seschrock */ 20128f18d1faSGeorge Wilson if (vdev_is_dead(vd->vdev_top) && !vd->vdev_islog && 20138f18d1faSGeorge Wilson vd->vdev_aux == NULL) { 20143d7072f8Seschrock vd->vdev_degraded = 1ULL; 20153d7072f8Seschrock vd->vdev_faulted = 0ULL; 20163d7072f8Seschrock 20173d7072f8Seschrock /* 20183d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 20193d7072f8Seschrock * mark it degraded. 20203d7072f8Seschrock */ 20213d7072f8Seschrock vdev_reopen(vd); 20223d7072f8Seschrock 2023069f55e2SEric Schrock if (vdev_readable(vd)) 2024069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 20253d7072f8Seschrock } 20263d7072f8Seschrock 2027e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 20283d7072f8Seschrock } 20293d7072f8Seschrock 20303d7072f8Seschrock /* 20313d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 20323d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 20333d7072f8Seschrock * as I/O is concerned. 20343d7072f8Seschrock */ 20353d7072f8Seschrock int 2036069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 20373d7072f8Seschrock { 2038c5904d13Seschrock vdev_t *vd; 20390a4e9518Sgw 20408f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 20413d7072f8Seschrock 2042c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2043e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2044e14bb325SJeff Bonwick 20450e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2046e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 20470e34b6a7Sbonwick 20483d7072f8Seschrock /* 20493d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 20503d7072f8Seschrock */ 2051e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2052e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 20533d7072f8Seschrock 20543d7072f8Seschrock vd->vdev_degraded = 1ULL; 20553d7072f8Seschrock if (!vdev_is_dead(vd)) 20563d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2057069f55e2SEric Schrock aux); 20583d7072f8Seschrock 2059e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 20603d7072f8Seschrock } 20613d7072f8Seschrock 20623d7072f8Seschrock /* 20633d7072f8Seschrock * Online the given vdev. If 'unspare' is set, it implies two things. First, 20643d7072f8Seschrock * any attached spare device should be detached when the device finishes 20653d7072f8Seschrock * resilvering. Second, the online should be treated like a 'test' online case, 20663d7072f8Seschrock * so no FMA events are generated if the device fails to open. 20673d7072f8Seschrock */ 20683d7072f8Seschrock int 2069e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 20703d7072f8Seschrock { 2071573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 20723d7072f8Seschrock 20738f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 20743d7072f8Seschrock 2075c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2076e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 20773d7072f8Seschrock 20783d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2079e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2080fa9e4066Sahrens 2081573ca77eSGeorge Wilson tvd = vd->vdev_top; 2082fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2083441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2084e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2085e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2086573ca77eSGeorge Wilson 2087573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2088573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2089573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2090573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2091573ca77eSGeorge Wilson } 2092573ca77eSGeorge Wilson 2093573ca77eSGeorge Wilson vdev_reopen(tvd); 20943d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 20953d7072f8Seschrock 2096573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2097573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2098573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2099573ca77eSGeorge Wilson } 2100573ca77eSGeorge Wilson 21013d7072f8Seschrock if (newstate) 21023d7072f8Seschrock *newstate = vd->vdev_state; 21033d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 21043d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 21053d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 21063d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 21073d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2108fa9e4066Sahrens 2109573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2110573ca77eSGeorge Wilson 2111573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2112573ca77eSGeorge Wilson if (vd->vdev_aux) 2113573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2114573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2115573ca77eSGeorge Wilson } 21168ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2117fa9e4066Sahrens } 2118fa9e4066Sahrens 21198f18d1faSGeorge Wilson int 21208f18d1faSGeorge Wilson vdev_offline_log(spa_t *spa) 21218f18d1faSGeorge Wilson { 21228f18d1faSGeorge Wilson int error = 0; 21238f18d1faSGeorge Wilson 21248f18d1faSGeorge Wilson if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 21258f18d1faSGeorge Wilson NULL, DS_FIND_CHILDREN)) == 0) { 21268f18d1faSGeorge Wilson 21278f18d1faSGeorge Wilson /* 21288f18d1faSGeorge Wilson * We successfully offlined the log device, sync out the 21298f18d1faSGeorge Wilson * current txg so that the "stubby" block can be removed 21308f18d1faSGeorge Wilson * by zil_sync(). 21318f18d1faSGeorge Wilson */ 21328f18d1faSGeorge Wilson txg_wait_synced(spa->spa_dsl_pool, 0); 21338f18d1faSGeorge Wilson } 21348f18d1faSGeorge Wilson return (error); 21358f18d1faSGeorge Wilson } 21368f18d1faSGeorge Wilson 2137fa9e4066Sahrens int 21383d7072f8Seschrock vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2139fa9e4066Sahrens { 2140e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 21418f18d1faSGeorge Wilson int error = 0; 21428f18d1faSGeorge Wilson uint64_t generation; 21438f18d1faSGeorge Wilson metaslab_group_t *mg; 21440a4e9518Sgw 21458f18d1faSGeorge Wilson top: 21468f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2147fa9e4066Sahrens 2148c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2149e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2150fa9e4066Sahrens 21510e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2152e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 21530e34b6a7Sbonwick 2154e6ca193dSGeorge Wilson tvd = vd->vdev_top; 21558f18d1faSGeorge Wilson mg = tvd->vdev_mg; 21568f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2157e6ca193dSGeorge Wilson 2158fa9e4066Sahrens /* 2159ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2160fa9e4066Sahrens */ 2161ecc2d604Sbonwick if (!vd->vdev_offline) { 2162ecc2d604Sbonwick /* 21638ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2164e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2165e6ca193dSGeorge Wilson * expendable. 2166ecc2d604Sbonwick */ 2167e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2168e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2169e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2170fa9e4066Sahrens 21718f18d1faSGeorge Wilson /* 2172*b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2173*b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2174*b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2175*b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 21768f18d1faSGeorge Wilson */ 21778f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 21788f18d1faSGeorge Wilson /* 21798f18d1faSGeorge Wilson * Prevent any future allocations. 21808f18d1faSGeorge Wilson */ 21818f18d1faSGeorge Wilson metaslab_class_remove(spa->spa_log_class, mg); 21828f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 21838f18d1faSGeorge Wilson 21848f18d1faSGeorge Wilson error = vdev_offline_log(spa); 21858f18d1faSGeorge Wilson 21868f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 21878f18d1faSGeorge Wilson 21888f18d1faSGeorge Wilson /* 21898f18d1faSGeorge Wilson * Check to see if the config has changed. 21908f18d1faSGeorge Wilson */ 21918f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 21928f18d1faSGeorge Wilson metaslab_class_add(spa->spa_log_class, mg); 21938f18d1faSGeorge Wilson if (error) 21948f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 21958f18d1faSGeorge Wilson vd, error)); 21968f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 21978f18d1faSGeorge Wilson goto top; 21988f18d1faSGeorge Wilson } 21998f18d1faSGeorge Wilson ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); 22008f18d1faSGeorge Wilson } 22018f18d1faSGeorge Wilson 2202ecc2d604Sbonwick /* 2203ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2204e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2205e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2206e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2207ecc2d604Sbonwick */ 2208ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2209e6ca193dSGeorge Wilson vdev_reopen(tvd); 2210e6ca193dSGeorge Wilson 2211e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2212e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2213ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2214e6ca193dSGeorge Wilson vdev_reopen(tvd); 2215e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2216ecc2d604Sbonwick } 22178f18d1faSGeorge Wilson 22188f18d1faSGeorge Wilson /* 22198f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 22208f18d1faSGeorge Wilson * once we online the device it's open for business. 22218f18d1faSGeorge Wilson */ 22228f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 22238f18d1faSGeorge Wilson metaslab_class_add(spa->spa_log_class, mg); 2224fa9e4066Sahrens } 2225fa9e4066Sahrens 2226e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2227ecc2d604Sbonwick 22288f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2229fa9e4066Sahrens } 2230fa9e4066Sahrens 2231ea8dc4b6Seschrock /* 2232ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2233ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2234ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2235ea8dc4b6Seschrock */ 2236ea8dc4b6Seschrock void 2237e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2238fa9e4066Sahrens { 2239e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2240e14bb325SJeff Bonwick 2241e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2242fa9e4066Sahrens 2243ea8dc4b6Seschrock if (vd == NULL) 2244e14bb325SJeff Bonwick vd = rvd; 2245fa9e4066Sahrens 2246ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2247ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2248ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2249fa9e4066Sahrens 2250e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2251e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 22523d7072f8Seschrock 22533d7072f8Seschrock /* 22548a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 22558a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 22568a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 22578a79c1b5Sek * written out to disk. 22583d7072f8Seschrock */ 2259e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2260e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 22618a79c1b5Sek 2262096d22d4SEric Schrock /* 2263096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2264096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2265096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2266096d22d4SEric Schrock */ 2267096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2268096d22d4SEric Schrock 22693d7072f8Seschrock vd->vdev_faulted = vd->vdev_degraded = 0; 2270e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2271e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2272e14bb325SJeff Bonwick 22733d7072f8Seschrock vdev_reopen(vd); 22743d7072f8Seschrock 2275096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2276096d22d4SEric Schrock 2277e14bb325SJeff Bonwick if (vd != rvd) 2278e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2279e14bb325SJeff Bonwick 2280e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2281bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 22823d7072f8Seschrock 22833d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 22843d7072f8Seschrock } 2285096d22d4SEric Schrock 2286096d22d4SEric Schrock /* 2287096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2288096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2289096d22d4SEric Schrock * done in response to the FMA fault. 2290096d22d4SEric Schrock */ 2291096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2292096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2293096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2294096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2295fa9e4066Sahrens } 2296fa9e4066Sahrens 2297e14bb325SJeff Bonwick boolean_t 2298e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 22990a4e9518Sgw { 230088ecc943SGeorge Wilson /* 230188ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 230288ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 230388ecc943SGeorge Wilson * these types of devices in the various code paths. 230488ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 230588ecc943SGeorge Wilson * before issuing I/O to them. 230688ecc943SGeorge Wilson */ 230788ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 230888ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 23090a4e9518Sgw } 23100a4e9518Sgw 2311e14bb325SJeff Bonwick boolean_t 2312e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 23130a4e9518Sgw { 2314e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 23150a4e9518Sgw } 23160a4e9518Sgw 2317e14bb325SJeff Bonwick boolean_t 2318e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2319fa9e4066Sahrens { 2320e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2321fa9e4066Sahrens } 2322fa9e4066Sahrens 2323a31e6787SGeorge Wilson boolean_t 2324a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2325a31e6787SGeorge Wilson { 23268ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 23278ad4d6ddSJeff Bonwick 2328a31e6787SGeorge Wilson /* 23298ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2330a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2331a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 23328ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 23338ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 23348ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2335a31e6787SGeorge Wilson */ 23368ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 233788ecc943SGeorge Wilson !vd->vdev_cant_write && !vd->vdev_ishole && !vd->vdev_removing); 2338a31e6787SGeorge Wilson } 2339a31e6787SGeorge Wilson 2340e14bb325SJeff Bonwick boolean_t 2341e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2342fa9e4066Sahrens { 2343e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2344fa9e4066Sahrens 2345e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2346e14bb325SJeff Bonwick return (B_FALSE); 2347fa9e4066Sahrens 2348e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2349e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2350fa9e4066Sahrens 2351e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2352e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2353fa9e4066Sahrens 2354e14bb325SJeff Bonwick return (B_TRUE); 2355fa9e4066Sahrens } 2356fa9e4066Sahrens 2357fa9e4066Sahrens /* 2358fa9e4066Sahrens * Get statistics for the given vdev. 2359fa9e4066Sahrens */ 2360fa9e4066Sahrens void 2361fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2362fa9e4066Sahrens { 2363fa9e4066Sahrens vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2364fa9e4066Sahrens 2365fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2366fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2367088f3894Sahrens vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; 2368fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2369fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2370573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2371573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2372573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2373fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2374fa9e4066Sahrens 2375fa9e4066Sahrens /* 2376fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2377fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2378fa9e4066Sahrens */ 2379fa9e4066Sahrens if (vd == rvd) { 2380e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2381fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2382fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2383fa9e4066Sahrens 2384fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2385e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2386fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2387fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2388fa9e4066Sahrens } 2389fa9e4066Sahrens vs->vs_scrub_examined += cvs->vs_scrub_examined; 2390fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2391fa9e4066Sahrens } 2392fa9e4066Sahrens } 2393fa9e4066Sahrens } 2394fa9e4066Sahrens 2395fa94a07fSbrendan void 2396fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2397fa94a07fSbrendan { 2398fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2399fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2400fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2401fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2402fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2403fa94a07fSbrendan } 2404fa94a07fSbrendan 2405fa9e4066Sahrens void 2406e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2407fa9e4066Sahrens { 24088ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 24098ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2410e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2411fa9e4066Sahrens vdev_t *pvd; 2412fa9e4066Sahrens uint64_t txg = zio->io_txg; 2413fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2414fa9e4066Sahrens zio_type_t type = zio->io_type; 2415fa9e4066Sahrens int flags = zio->io_flags; 2416fa9e4066Sahrens 2417e14bb325SJeff Bonwick /* 2418e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2419e14bb325SJeff Bonwick */ 2420e14bb325SJeff Bonwick if (zio->io_gang_tree) 2421e14bb325SJeff Bonwick return; 2422e14bb325SJeff Bonwick 2423fa9e4066Sahrens if (zio->io_error == 0) { 2424e14bb325SJeff Bonwick /* 2425e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2426e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2427e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2428e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2429e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2430e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2431e14bb325SJeff Bonwick * remain zero, which is what we want.) 2432e14bb325SJeff Bonwick * 2433e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2434e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2435e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2436e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2437e14bb325SJeff Bonwick */ 2438e14bb325SJeff Bonwick if (vd == rvd) 2439e14bb325SJeff Bonwick return; 2440e14bb325SJeff Bonwick 2441e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 24428ad4d6ddSJeff Bonwick 24438ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 24448ad4d6ddSJeff Bonwick return; 24458ad4d6ddSJeff Bonwick 24468ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 24478ad4d6ddSJeff Bonwick 2448e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 2449d80c45e0Sbonwick if (flags & ZIO_FLAG_SCRUB_THREAD) 2450e14bb325SJeff Bonwick vs->vs_scrub_repaired += psize; 24518ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2452e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2453fa9e4066Sahrens } 24548ad4d6ddSJeff Bonwick 24558ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 24568ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 24578ad4d6ddSJeff Bonwick 24588ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2459fa9e4066Sahrens return; 2460fa9e4066Sahrens } 2461fa9e4066Sahrens 2462fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2463fa9e4066Sahrens return; 2464fa9e4066Sahrens 24658956713aSEric Schrock /* 24668956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 24678956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 24688956713aSEric Schrock * hard errors, when in reality they can happen for any number of 24698956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 24708956713aSEric Schrock */ 24718956713aSEric Schrock if (zio->io_error == EIO && 24728956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 24738956713aSEric Schrock return; 24748956713aSEric Schrock 24758f18d1faSGeorge Wilson /* 24768f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 24778f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 24788f18d1faSGeorge Wilson * errors. 24798f18d1faSGeorge Wilson */ 24808f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 24818f18d1faSGeorge Wilson return; 24828f18d1faSGeorge Wilson 2483e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2484b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2485e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2486e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2487e14bb325SJeff Bonwick else 2488e14bb325SJeff Bonwick vs->vs_read_errors++; 2489fa9e4066Sahrens } 2490b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2491e14bb325SJeff Bonwick vs->vs_write_errors++; 2492e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2493fa9e4066Sahrens 24948ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 24958ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 2496*b24ab676SJeff Bonwick (flags & ZIO_FLAG_SCRUB_THREAD) || 2497*b24ab676SJeff Bonwick spa->spa_claiming)) { 24988ad4d6ddSJeff Bonwick /* 2499*b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2500*b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2501*b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2502*b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2503*b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2504*b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2505*b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2506*b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 25078ad4d6ddSJeff Bonwick * 25088ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 25098ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 25108ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 25118ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 25128ad4d6ddSJeff Bonwick */ 25138ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 25148ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 25158ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SCRUB_THREAD) { 25168ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 25178ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 25188ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2519*b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2520*b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2521*b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2522*b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 25238ad4d6ddSJeff Bonwick } 2524*b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 25258ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2526fa9e4066Sahrens return; 25278ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 25288ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 25298ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2530fa9e4066Sahrens } 25318ad4d6ddSJeff Bonwick if (vd != rvd) 25328ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2533fa9e4066Sahrens } 2534fa9e4066Sahrens } 2535fa9e4066Sahrens 2536fa9e4066Sahrens void 2537fa9e4066Sahrens vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) 2538fa9e4066Sahrens { 2539fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2540fa9e4066Sahrens 2541573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2542fa9e4066Sahrens vdev_scrub_stat_update(vd->vdev_child[c], type, complete); 2543fa9e4066Sahrens 2544fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2545fa9e4066Sahrens 2546fa9e4066Sahrens if (type == POOL_SCRUB_NONE) { 2547fa9e4066Sahrens /* 2548fa9e4066Sahrens * Update completion and end time. Leave everything else alone 2549fa9e4066Sahrens * so we can report what happened during the previous scrub. 2550fa9e4066Sahrens */ 2551fa9e4066Sahrens vs->vs_scrub_complete = complete; 2552fa9e4066Sahrens vs->vs_scrub_end = gethrestime_sec(); 2553fa9e4066Sahrens } else { 2554fa9e4066Sahrens vs->vs_scrub_type = type; 2555fa9e4066Sahrens vs->vs_scrub_complete = 0; 2556fa9e4066Sahrens vs->vs_scrub_examined = 0; 2557fa9e4066Sahrens vs->vs_scrub_repaired = 0; 2558fa9e4066Sahrens vs->vs_scrub_start = gethrestime_sec(); 2559fa9e4066Sahrens vs->vs_scrub_end = 0; 2560fa9e4066Sahrens } 2561fa9e4066Sahrens 2562fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2563fa9e4066Sahrens } 2564fa9e4066Sahrens 2565fa9e4066Sahrens /* 2566*b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2567*b24ab676SJeff Bonwick * and the root vdev. 2568fa9e4066Sahrens */ 2569fa9e4066Sahrens void 2570*b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2571*b24ab676SJeff Bonwick int64_t space_delta) 2572fa9e4066Sahrens { 257399653d4eSeschrock int64_t dspace_delta = space_delta; 25748654d025Sperrin spa_t *spa = vd->vdev_spa; 25758654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2576*b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2577*b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2578fa9e4066Sahrens 25798654d025Sperrin ASSERT(vd == vd->vdev_top); 258099653d4eSeschrock 25818654d025Sperrin /* 25828654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 25838654d025Sperrin * factor. We must calculate this here and not at the root vdev 25848654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 25858654d025Sperrin * childrens', thus not accurate enough for us. 25868654d025Sperrin */ 25878654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2588e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 25898654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 25908654d025Sperrin vd->vdev_deflate_ratio; 25918654d025Sperrin 25928654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 25938654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2594*b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 25958654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 25968654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 25978654d025Sperrin 2598*b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2599fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2600fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2601*b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2602fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2603fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2604fa94a07fSbrendan } 2605*b24ab676SJeff Bonwick 2606*b24ab676SJeff Bonwick if (mc != NULL) { 2607*b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2608*b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2609*b24ab676SJeff Bonwick 2610*b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2611*b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2612*b24ab676SJeff Bonwick } 2613fa9e4066Sahrens } 2614fa9e4066Sahrens 2615fa9e4066Sahrens /* 2616fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2617fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2618fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2619fa9e4066Sahrens */ 2620fa9e4066Sahrens void 2621fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2622fa9e4066Sahrens { 2623fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2624fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2625fa9e4066Sahrens int c; 2626fa9e4066Sahrens 2627c5904d13Seschrock /* 26286809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 26296809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2630c5904d13Seschrock */ 2631c5904d13Seschrock if (vd->vdev_aux != NULL) { 2632c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2633c5904d13Seschrock nvlist_t **aux; 2634c5904d13Seschrock uint_t naux; 2635c5904d13Seschrock 2636c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2637c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2638c5904d13Seschrock break; 2639c5904d13Seschrock } 2640c5904d13Seschrock 2641e14bb325SJeff Bonwick if (c == sav->sav_count) { 2642e14bb325SJeff Bonwick /* 2643e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2644e14bb325SJeff Bonwick */ 2645e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2646e14bb325SJeff Bonwick return; 2647e14bb325SJeff Bonwick } 2648e14bb325SJeff Bonwick 2649c5904d13Seschrock sav->sav_sync = B_TRUE; 2650c5904d13Seschrock 26516809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 26526809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 26536809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 26546809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 26556809eb4eSEric Schrock } 2656c5904d13Seschrock 2657c5904d13Seschrock ASSERT(c < naux); 2658c5904d13Seschrock 2659c5904d13Seschrock /* 2660c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2661c5904d13Seschrock * sketchy, but it will work. 2662c5904d13Seschrock */ 2663c5904d13Seschrock nvlist_free(aux[c]); 2664c5904d13Seschrock aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); 2665c5904d13Seschrock 2666c5904d13Seschrock return; 2667c5904d13Seschrock } 2668c5904d13Seschrock 26695dabedeeSbonwick /* 2670e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2671e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2672e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 26735dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 26745dabedeeSbonwick */ 2675e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2676e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2677e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 26785dabedeeSbonwick 2679fa9e4066Sahrens if (vd == rvd) { 2680fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2681fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2682fa9e4066Sahrens } else { 2683fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2684fa9e4066Sahrens 268588ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 268688ecc943SGeorge Wilson !vd->vdev_ishole) 2687e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2688fa9e4066Sahrens } 2689fa9e4066Sahrens } 2690fa9e4066Sahrens 2691fa9e4066Sahrens void 2692fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2693fa9e4066Sahrens { 26945dabedeeSbonwick spa_t *spa = vd->vdev_spa; 26955dabedeeSbonwick 2696e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2697e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2698e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 26995dabedeeSbonwick 2700e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2701e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2702e14bb325SJeff Bonwick } 2703e14bb325SJeff Bonwick 2704e14bb325SJeff Bonwick /* 2705e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2706e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2707e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2708e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2709e14bb325SJeff Bonwick */ 2710e14bb325SJeff Bonwick void 2711e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2712e14bb325SJeff Bonwick { 2713e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2714e14bb325SJeff Bonwick 2715e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2716e14bb325SJeff Bonwick 2717e14bb325SJeff Bonwick /* 2718e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 2719e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 2720e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 2721e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 2722e14bb325SJeff Bonwick */ 2723e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2724e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2725e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2726e14bb325SJeff Bonwick 2727*b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2728e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 2729e14bb325SJeff Bonwick } 2730e14bb325SJeff Bonwick 2731e14bb325SJeff Bonwick void 2732e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 2733e14bb325SJeff Bonwick { 2734e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2735e14bb325SJeff Bonwick 2736e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2737e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2738e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2739e14bb325SJeff Bonwick 2740e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2741e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 2742fa9e4066Sahrens } 2743fa9e4066Sahrens 274432b87932Sek /* 274532b87932Sek * Propagate vdev state up from children to parent. 274632b87932Sek */ 274744cd46caSbillm void 274844cd46caSbillm vdev_propagate_state(vdev_t *vd) 274944cd46caSbillm { 27508ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 27518ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 275244cd46caSbillm int degraded = 0, faulted = 0; 275344cd46caSbillm int corrupted = 0; 275444cd46caSbillm vdev_t *child; 275544cd46caSbillm 27563d7072f8Seschrock if (vd->vdev_children > 0) { 2757573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 27583d7072f8Seschrock child = vd->vdev_child[c]; 275951ece835Seschrock 276088ecc943SGeorge Wilson /* 276188ecc943SGeorge Wilson * Don't factor holes into the decision. 276288ecc943SGeorge Wilson */ 276388ecc943SGeorge Wilson if (child->vdev_ishole) 276488ecc943SGeorge Wilson continue; 276588ecc943SGeorge Wilson 2766e14bb325SJeff Bonwick if (!vdev_readable(child) || 27678ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 276851ece835Seschrock /* 276951ece835Seschrock * Root special: if there is a top-level log 277051ece835Seschrock * device, treat the root vdev as if it were 277151ece835Seschrock * degraded. 277251ece835Seschrock */ 277351ece835Seschrock if (child->vdev_islog && vd == rvd) 277451ece835Seschrock degraded++; 277551ece835Seschrock else 277651ece835Seschrock faulted++; 277751ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 27783d7072f8Seschrock degraded++; 277951ece835Seschrock } 278044cd46caSbillm 27813d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 27823d7072f8Seschrock corrupted++; 27833d7072f8Seschrock } 278444cd46caSbillm 27853d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 27863d7072f8Seschrock 27873d7072f8Seschrock /* 2788e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 27893d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 27903d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 27913d7072f8Seschrock * replicas'. 27923d7072f8Seschrock */ 27933d7072f8Seschrock if (corrupted && vd == rvd && 27943d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 27953d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 27963d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 27973d7072f8Seschrock } 27983d7072f8Seschrock 279951ece835Seschrock if (vd->vdev_parent) 28003d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 280144cd46caSbillm } 280244cd46caSbillm 2803fa9e4066Sahrens /* 2804ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 2805ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 2806ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 2807ea8dc4b6Seschrock * 2808ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 2809ea8dc4b6Seschrock * generated. 2810fa9e4066Sahrens */ 2811fa9e4066Sahrens void 2812ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2813fa9e4066Sahrens { 2814560e6e96Seschrock uint64_t save_state; 2815c5904d13Seschrock spa_t *spa = vd->vdev_spa; 2816ea8dc4b6Seschrock 2817ea8dc4b6Seschrock if (state == vd->vdev_state) { 2818ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 2819fa9e4066Sahrens return; 2820ea8dc4b6Seschrock } 2821ea8dc4b6Seschrock 2822560e6e96Seschrock save_state = vd->vdev_state; 2823fa9e4066Sahrens 2824fa9e4066Sahrens vd->vdev_state = state; 2825fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 2826fa9e4066Sahrens 28273d7072f8Seschrock /* 28283d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 28293d7072f8Seschrock * always close the underlying device. Otherwise, we keep accessible 28303d7072f8Seschrock * but invalid devices open forever. We don't call vdev_close() itself, 28313d7072f8Seschrock * because that implies some extra checks (offline, etc) that we don't 28323d7072f8Seschrock * want here. This is limited to leaf devices, because otherwise 28333d7072f8Seschrock * closing the device will affect other children. 28343d7072f8Seschrock */ 2835cbd2b15eSJeff Bonwick if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) 28363d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 28373d7072f8Seschrock 2838069f55e2SEric Schrock /* 2839069f55e2SEric Schrock * If we have brought this vdev back into service, we need 2840069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 2841069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 2842069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 2843069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 2844069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 2845069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 2846069f55e2SEric Schrock */ 2847069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2848069f55e2SEric Schrock vd->vdev_prevstate != state) 2849069f55e2SEric Schrock zfs_post_state_change(spa, vd); 2850069f55e2SEric Schrock 28513d7072f8Seschrock if (vd->vdev_removed && 28523d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 28533d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 28543d7072f8Seschrock /* 28553d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 28563d7072f8Seschrock * device was previously marked removed and someone attempted to 28573d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 28583d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 28593d7072f8Seschrock * it is one of our special test online cases, which is only 28603d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 28613d7072f8Seschrock * fault. 28623d7072f8Seschrock */ 28633d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 28643d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 28653d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 28663d7072f8Seschrock vd->vdev_removed = B_TRUE; 28673d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 2868ea8dc4b6Seschrock /* 2869ea8dc4b6Seschrock * If we fail to open a vdev during an import, we mark it as 2870ea8dc4b6Seschrock * "not available", which signifies that it was never there to 2871ea8dc4b6Seschrock * begin with. Failure to open such a device is not considered 2872ea8dc4b6Seschrock * an error. 2873ea8dc4b6Seschrock */ 2874c5904d13Seschrock if (spa->spa_load_state == SPA_LOAD_IMPORT && 2875560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 2876560e6e96Seschrock vd->vdev_not_present = 1; 2877560e6e96Seschrock 2878560e6e96Seschrock /* 2879560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 2880560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 2881560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 2882560e6e96Seschrock * want to post the ereport if the device was already in the 2883560e6e96Seschrock * CANT_OPEN state beforehand. 28843d7072f8Seschrock * 28853d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 28863d7072f8Seschrock * online the device in response to an insertion event. If we 28873d7072f8Seschrock * hit this case, then we have detected an insertion event for a 28883d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 28893d7072f8Seschrock * In this scenario, we don't post an ereport because we are 28903d7072f8Seschrock * about to replace the device, or attempt an online with 28913d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 2892560e6e96Seschrock */ 28933d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 28943d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 2895c5904d13Seschrock vd != spa->spa_root_vdev) { 2896ea8dc4b6Seschrock const char *class; 2897ea8dc4b6Seschrock 2898ea8dc4b6Seschrock switch (aux) { 2899ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 2900ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2901ea8dc4b6Seschrock break; 2902ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 2903ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2904ea8dc4b6Seschrock break; 2905ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 2906ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2907ea8dc4b6Seschrock break; 2908ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 2909ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2910ea8dc4b6Seschrock break; 2911ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 2912ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 2913ea8dc4b6Seschrock break; 2914ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 2915ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 2916ea8dc4b6Seschrock break; 2917e14bb325SJeff Bonwick case VDEV_AUX_IO_FAILURE: 2918e14bb325SJeff Bonwick class = FM_EREPORT_ZFS_IO_FAILURE; 2919e14bb325SJeff Bonwick break; 2920ea8dc4b6Seschrock default: 2921ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 2922ea8dc4b6Seschrock } 2923ea8dc4b6Seschrock 2924c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 2925ea8dc4b6Seschrock } 2926ea8dc4b6Seschrock 29273d7072f8Seschrock /* Erase any notion of persistent removed state */ 29283d7072f8Seschrock vd->vdev_removed = B_FALSE; 29293d7072f8Seschrock } else { 29303d7072f8Seschrock vd->vdev_removed = B_FALSE; 29313d7072f8Seschrock } 2932ea8dc4b6Seschrock 29338b33d774STim Haley if (!isopen && vd->vdev_parent) 29348b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 2935fa9e4066Sahrens } 293615e6edf1Sgw 293715e6edf1Sgw /* 293815e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 293915e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 294015e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 294115e6edf1Sgw * can be wholedisks. 294215e6edf1Sgw */ 294315e6edf1Sgw boolean_t 294415e6edf1Sgw vdev_is_bootable(vdev_t *vd) 294515e6edf1Sgw { 294615e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 294715e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 294815e6edf1Sgw 294915e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 295015e6edf1Sgw vd->vdev_children > 1) { 295115e6edf1Sgw return (B_FALSE); 295215e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 295315e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 295415e6edf1Sgw return (B_FALSE); 295515e6edf1Sgw } 295615e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 295715e6edf1Sgw return (B_FALSE); 295815e6edf1Sgw } 295915e6edf1Sgw 2960573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 296115e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 296215e6edf1Sgw return (B_FALSE); 296315e6edf1Sgw } 296415e6edf1Sgw return (B_TRUE); 296515e6edf1Sgw } 2966e6ca193dSGeorge Wilson 296788ecc943SGeorge Wilson /* 296888ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 296988ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 297088ecc943SGeorge Wilson * vdev was offline then we transfer that state to the device 297188ecc943SGeorge Wilson * in the current vdev tree (nvd). 297288ecc943SGeorge Wilson */ 2973e6ca193dSGeorge Wilson void 297488ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 2975e6ca193dSGeorge Wilson { 297688ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 2977e6ca193dSGeorge Wilson 297888ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 297988ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 2980e6ca193dSGeorge Wilson 298188ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 298288ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 2983e6ca193dSGeorge Wilson 298488ecc943SGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf && ovd->vdev_offline) { 2985e6ca193dSGeorge Wilson /* 2986e6ca193dSGeorge Wilson * It would be nice to call vdev_offline() 2987e6ca193dSGeorge Wilson * directly but the pool isn't fully loaded and 2988e6ca193dSGeorge Wilson * the txg threads have not been started yet. 2989e6ca193dSGeorge Wilson */ 299088ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 299188ecc943SGeorge Wilson vdev_reopen(nvd->vdev_top); 2992e6ca193dSGeorge Wilson } 2993e6ca193dSGeorge Wilson } 2994573ca77eSGeorge Wilson 2995573ca77eSGeorge Wilson /* 2996573ca77eSGeorge Wilson * Expand a vdev if possible. 2997573ca77eSGeorge Wilson */ 2998573ca77eSGeorge Wilson void 2999573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3000573ca77eSGeorge Wilson { 3001573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3002573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3003573ca77eSGeorge Wilson 3004573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3005573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3006573ca77eSGeorge Wilson vdev_config_dirty(vd); 3007573ca77eSGeorge Wilson } 3008573ca77eSGeorge Wilson } 3009