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 /* 23*98d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/zfs_context.h> 27ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 28fa9e4066Sahrens #include <sys/spa.h> 29fa9e4066Sahrens #include <sys/spa_impl.h> 30fa9e4066Sahrens #include <sys/dmu.h> 31fa9e4066Sahrens #include <sys/dmu_tx.h> 32fa9e4066Sahrens #include <sys/vdev_impl.h> 33fa9e4066Sahrens #include <sys/uberblock_impl.h> 34fa9e4066Sahrens #include <sys/metaslab.h> 35fa9e4066Sahrens #include <sys/metaslab_impl.h> 36fa9e4066Sahrens #include <sys/space_map.h> 37fa9e4066Sahrens #include <sys/zio.h> 38fa9e4066Sahrens #include <sys/zap.h> 39fa9e4066Sahrens #include <sys/fs/zfs.h> 40c5904d13Seschrock #include <sys/arc.h> 41e6ca193dSGeorge Wilson #include <sys/zil.h> 42fa9e4066Sahrens 43fa9e4066Sahrens /* 44fa9e4066Sahrens * Virtual device management. 45fa9e4066Sahrens */ 46fa9e4066Sahrens 47fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 48fa9e4066Sahrens &vdev_root_ops, 49fa9e4066Sahrens &vdev_raidz_ops, 50fa9e4066Sahrens &vdev_mirror_ops, 51fa9e4066Sahrens &vdev_replacing_ops, 5299653d4eSeschrock &vdev_spare_ops, 53fa9e4066Sahrens &vdev_disk_ops, 54fa9e4066Sahrens &vdev_file_ops, 55fa9e4066Sahrens &vdev_missing_ops, 5688ecc943SGeorge Wilson &vdev_hole_ops, 57fa9e4066Sahrens NULL 58fa9e4066Sahrens }; 59fa9e4066Sahrens 60088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 61088f3894Sahrens int zfs_scrub_limit = 10; 6205b2b3b8Smishra 63fa9e4066Sahrens /* 64fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 65fa9e4066Sahrens */ 66fa9e4066Sahrens static vdev_ops_t * 67fa9e4066Sahrens vdev_getops(const char *type) 68fa9e4066Sahrens { 69fa9e4066Sahrens vdev_ops_t *ops, **opspp; 70fa9e4066Sahrens 71fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 72fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 73fa9e4066Sahrens break; 74fa9e4066Sahrens 75fa9e4066Sahrens return (ops); 76fa9e4066Sahrens } 77fa9e4066Sahrens 78fa9e4066Sahrens /* 79fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 80fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 81fa9e4066Sahrens */ 82fa9e4066Sahrens uint64_t 83fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 84fa9e4066Sahrens { 85ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 86fa9e4066Sahrens uint64_t csize; 87fa9e4066Sahrens 88573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 89fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 90fa9e4066Sahrens asize = MAX(asize, csize); 91fa9e4066Sahrens } 92fa9e4066Sahrens 93fa9e4066Sahrens return (asize); 94fa9e4066Sahrens } 95fa9e4066Sahrens 962a79c5feSlling /* 97573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 98573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 99573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 100573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1012a79c5feSlling */ 1022a79c5feSlling uint64_t 103573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1042a79c5feSlling { 105573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1062a79c5feSlling 107573ca77eSGeorge Wilson /* 108573ca77eSGeorge Wilson * The our parent is NULL (inactive spare or cache) or is the root, 109573ca77eSGeorge Wilson * just return our own asize. 110573ca77eSGeorge Wilson */ 111573ca77eSGeorge Wilson if (pvd == NULL) 112573ca77eSGeorge Wilson return (vd->vdev_asize); 1132a79c5feSlling 1142a79c5feSlling /* 115573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 116573ca77eSGeorge Wilson * to the nearest metaslab. 1172a79c5feSlling */ 118573ca77eSGeorge Wilson if (vd == vd->vdev_top) 119573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1202a79c5feSlling 121573ca77eSGeorge Wilson /* 122573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 123573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 124573ca77eSGeorge Wilson */ 125573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 126573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1272a79c5feSlling 128573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 129573ca77eSGeorge Wilson } 1302a79c5feSlling 131573ca77eSGeorge Wilson void 132573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 133573ca77eSGeorge Wilson { 134573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 135573ca77eSGeorge Wilson 136573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 137573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1382a79c5feSlling } 1392a79c5feSlling 140fa9e4066Sahrens vdev_t * 141fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 142fa9e4066Sahrens { 143fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 144fa9e4066Sahrens 145e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 146e05725b1Sbonwick 147088f3894Sahrens if (vdev < rvd->vdev_children) { 148088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 149fa9e4066Sahrens return (rvd->vdev_child[vdev]); 150088f3894Sahrens } 151fa9e4066Sahrens 152fa9e4066Sahrens return (NULL); 153fa9e4066Sahrens } 154fa9e4066Sahrens 155fa9e4066Sahrens vdev_t * 156fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 157fa9e4066Sahrens { 158fa9e4066Sahrens vdev_t *mvd; 159fa9e4066Sahrens 1600e34b6a7Sbonwick if (vd->vdev_guid == guid) 161fa9e4066Sahrens return (vd); 162fa9e4066Sahrens 163573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 164fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 165fa9e4066Sahrens NULL) 166fa9e4066Sahrens return (mvd); 167fa9e4066Sahrens 168fa9e4066Sahrens return (NULL); 169fa9e4066Sahrens } 170fa9e4066Sahrens 171fa9e4066Sahrens void 172fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 173fa9e4066Sahrens { 174fa9e4066Sahrens size_t oldsize, newsize; 175fa9e4066Sahrens uint64_t id = cvd->vdev_id; 176fa9e4066Sahrens vdev_t **newchild; 177fa9e4066Sahrens 178e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 179fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 180fa9e4066Sahrens 181fa9e4066Sahrens cvd->vdev_parent = pvd; 182fa9e4066Sahrens 183fa9e4066Sahrens if (pvd == NULL) 184fa9e4066Sahrens return; 185fa9e4066Sahrens 186fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 187fa9e4066Sahrens 188fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 189fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 190fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 191fa9e4066Sahrens 192fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 193fa9e4066Sahrens if (pvd->vdev_child != NULL) { 194fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 195fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 196fa9e4066Sahrens } 197fa9e4066Sahrens 198fa9e4066Sahrens pvd->vdev_child = newchild; 199fa9e4066Sahrens pvd->vdev_child[id] = cvd; 200fa9e4066Sahrens 201fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 202fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 203fa9e4066Sahrens 204fa9e4066Sahrens /* 205fa9e4066Sahrens * Walk up all ancestors to update guid sum. 206fa9e4066Sahrens */ 207fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 208fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 20905b2b3b8Smishra 21005b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 21105b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; 212fa9e4066Sahrens } 213fa9e4066Sahrens 214fa9e4066Sahrens void 215fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 216fa9e4066Sahrens { 217fa9e4066Sahrens int c; 218fa9e4066Sahrens uint_t id = cvd->vdev_id; 219fa9e4066Sahrens 220fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 221fa9e4066Sahrens 222fa9e4066Sahrens if (pvd == NULL) 223fa9e4066Sahrens return; 224fa9e4066Sahrens 225fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 226fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 227fa9e4066Sahrens 228fa9e4066Sahrens pvd->vdev_child[id] = NULL; 229fa9e4066Sahrens cvd->vdev_parent = NULL; 230fa9e4066Sahrens 231fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 232fa9e4066Sahrens if (pvd->vdev_child[c]) 233fa9e4066Sahrens break; 234fa9e4066Sahrens 235fa9e4066Sahrens if (c == pvd->vdev_children) { 236fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 237fa9e4066Sahrens pvd->vdev_child = NULL; 238fa9e4066Sahrens pvd->vdev_children = 0; 239fa9e4066Sahrens } 240fa9e4066Sahrens 241fa9e4066Sahrens /* 242fa9e4066Sahrens * Walk up all ancestors to update guid sum. 243fa9e4066Sahrens */ 244fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 245fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 24605b2b3b8Smishra 24705b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 24805b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; 249fa9e4066Sahrens } 250fa9e4066Sahrens 251fa9e4066Sahrens /* 252fa9e4066Sahrens * Remove any holes in the child array. 253fa9e4066Sahrens */ 254fa9e4066Sahrens void 255fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 256fa9e4066Sahrens { 257fa9e4066Sahrens vdev_t **newchild, *cvd; 258fa9e4066Sahrens int oldc = pvd->vdev_children; 259573ca77eSGeorge Wilson int newc; 260fa9e4066Sahrens 261e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 262fa9e4066Sahrens 263573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 264fa9e4066Sahrens if (pvd->vdev_child[c]) 265fa9e4066Sahrens newc++; 266fa9e4066Sahrens 267fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 268fa9e4066Sahrens 269573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 270fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 271fa9e4066Sahrens newchild[newc] = cvd; 272fa9e4066Sahrens cvd->vdev_id = newc++; 273fa9e4066Sahrens } 274fa9e4066Sahrens } 275fa9e4066Sahrens 276fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 277fa9e4066Sahrens pvd->vdev_child = newchild; 278fa9e4066Sahrens pvd->vdev_children = newc; 279fa9e4066Sahrens } 280fa9e4066Sahrens 281fa9e4066Sahrens /* 282fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 283fa9e4066Sahrens */ 28488ecc943SGeorge Wilson vdev_t * 285fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 286fa9e4066Sahrens { 287fa9e4066Sahrens vdev_t *vd; 288fa9e4066Sahrens 289fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 290fa9e4066Sahrens 2910e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2920e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2930e34b6a7Sbonwick spa->spa_root_vdev = vd; 2940e34b6a7Sbonwick } 2950e34b6a7Sbonwick 29688ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 2970e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 2980e34b6a7Sbonwick /* 2990e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3000e34b6a7Sbonwick * which must be unique among all pools. 3010e34b6a7Sbonwick */ 3021195e687SMark J Musante guid = spa_generate_guid(NULL); 3030e34b6a7Sbonwick } else { 3040e34b6a7Sbonwick /* 3050e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3060e34b6a7Sbonwick */ 3071195e687SMark J Musante guid = spa_generate_guid(spa); 3080e34b6a7Sbonwick } 3090e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3100e34b6a7Sbonwick } 3110e34b6a7Sbonwick 312fa9e4066Sahrens vd->vdev_spa = spa; 313fa9e4066Sahrens vd->vdev_id = id; 314fa9e4066Sahrens vd->vdev_guid = guid; 315fa9e4066Sahrens vd->vdev_guid_sum = guid; 316fa9e4066Sahrens vd->vdev_ops = ops; 317fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 31888ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 319fa9e4066Sahrens 320fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3215ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 322e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3238ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3248ad4d6ddSJeff Bonwick space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 3258ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3268ad4d6ddSJeff Bonwick } 327fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 328fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 329fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 330fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 331fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3323d7072f8Seschrock vdev_queue_init(vd); 3333d7072f8Seschrock vdev_cache_init(vd); 334fa9e4066Sahrens 335fa9e4066Sahrens return (vd); 336fa9e4066Sahrens } 337fa9e4066Sahrens 338fa9e4066Sahrens /* 339fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 340fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 341fa9e4066Sahrens * different for each case. 342fa9e4066Sahrens */ 34399653d4eSeschrock int 34499653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 34599653d4eSeschrock int alloctype) 346fa9e4066Sahrens { 347fa9e4066Sahrens vdev_ops_t *ops; 348fa9e4066Sahrens char *type; 3498654d025Sperrin uint64_t guid = 0, islog, nparity; 350fa9e4066Sahrens vdev_t *vd; 351fa9e4066Sahrens 352e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 353fa9e4066Sahrens 354fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 35599653d4eSeschrock return (EINVAL); 356fa9e4066Sahrens 357fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 35899653d4eSeschrock return (EINVAL); 359fa9e4066Sahrens 360fa9e4066Sahrens /* 361fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 362fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 363fa9e4066Sahrens */ 364fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 365fa9e4066Sahrens uint64_t label_id; 366fa9e4066Sahrens 367fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 368fa9e4066Sahrens label_id != id) 36999653d4eSeschrock return (EINVAL); 370fa9e4066Sahrens 371fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37299653d4eSeschrock return (EINVAL); 37399653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 37499653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37599653d4eSeschrock return (EINVAL); 376fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 377fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 378fa94a07fSbrendan return (EINVAL); 37921ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 38021ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 38121ecdf64SLin Ling return (EINVAL); 382fa9e4066Sahrens } 383fa9e4066Sahrens 38499653d4eSeschrock /* 38599653d4eSeschrock * The first allocated vdev must be of type 'root'. 38699653d4eSeschrock */ 38799653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 38899653d4eSeschrock return (EINVAL); 38999653d4eSeschrock 3908654d025Sperrin /* 3918654d025Sperrin * Determine whether we're a log vdev. 3928654d025Sperrin */ 3938654d025Sperrin islog = 0; 3948654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 395990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 3968654d025Sperrin return (ENOTSUP); 397fa9e4066Sahrens 39888ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 39988ecc943SGeorge Wilson return (ENOTSUP); 40088ecc943SGeorge Wilson 40199653d4eSeschrock /* 4028654d025Sperrin * Set the nparity property for RAID-Z vdevs. 40399653d4eSeschrock */ 4048654d025Sperrin nparity = -1ULL; 40599653d4eSeschrock if (ops == &vdev_raidz_ops) { 40699653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4078654d025Sperrin &nparity) == 0) { 408b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 40999653d4eSeschrock return (EINVAL); 41099653d4eSeschrock /* 411f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 412f94275ceSAdam Leventhal * device. 41399653d4eSeschrock */ 414f94275ceSAdam Leventhal if (nparity > 1 && 415f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 416f94275ceSAdam Leventhal return (ENOTSUP); 417f94275ceSAdam Leventhal if (nparity > 2 && 418f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 41999653d4eSeschrock return (ENOTSUP); 42099653d4eSeschrock } else { 42199653d4eSeschrock /* 42299653d4eSeschrock * We require the parity to be specified for SPAs that 42399653d4eSeschrock * support multiple parity levels. 42499653d4eSeschrock */ 425f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 42699653d4eSeschrock return (EINVAL); 42799653d4eSeschrock /* 42899653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 42999653d4eSeschrock */ 4308654d025Sperrin nparity = 1; 43199653d4eSeschrock } 43299653d4eSeschrock } else { 4338654d025Sperrin nparity = 0; 43499653d4eSeschrock } 4358654d025Sperrin ASSERT(nparity != -1ULL); 4368654d025Sperrin 4378654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4388654d025Sperrin 4398654d025Sperrin vd->vdev_islog = islog; 4408654d025Sperrin vd->vdev_nparity = nparity; 4418654d025Sperrin 4428654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4438654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4448654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4458654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4468654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4478654d025Sperrin &vd->vdev_physpath) == 0) 4488654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4496809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4506809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 45199653d4eSeschrock 452afefbcddSeschrock /* 453afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 454afefbcddSeschrock * as -1. 455afefbcddSeschrock */ 456afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 457afefbcddSeschrock &vd->vdev_wholedisk) != 0) 458afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 459afefbcddSeschrock 460ea8dc4b6Seschrock /* 461ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 462ea8dc4b6Seschrock * was not present at the time of import. 463ea8dc4b6Seschrock */ 4646809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4656809eb4eSEric Schrock &vd->vdev_not_present); 466ea8dc4b6Seschrock 467ecc2d604Sbonwick /* 468ecc2d604Sbonwick * Get the alignment requirement. 469ecc2d604Sbonwick */ 470ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 471ecc2d604Sbonwick 47288ecc943SGeorge Wilson /* 47388ecc943SGeorge Wilson * Retrieve the vdev creation time. 47488ecc943SGeorge Wilson */ 47588ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 47688ecc943SGeorge Wilson &vd->vdev_crtxg); 47788ecc943SGeorge Wilson 478fa9e4066Sahrens /* 479fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 480fa9e4066Sahrens */ 4811195e687SMark J Musante if (parent && !parent->vdev_parent && 4821195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 483fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 484fa9e4066Sahrens &vd->vdev_ms_array); 485fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 486fa9e4066Sahrens &vd->vdev_ms_shift); 487fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 488fa9e4066Sahrens &vd->vdev_asize); 489fa9e4066Sahrens } 490fa9e4066Sahrens 491a1521560SJeff Bonwick if (parent && !parent->vdev_parent) { 492a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 4939f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 4941195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 4959f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 496a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 497a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 498a1521560SJeff Bonwick } 499a1521560SJeff Bonwick 500fa9e4066Sahrens /* 5013d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 502fa9e4066Sahrens */ 503c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 50421ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 50521ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 506c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 507c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5088ad4d6ddSJeff Bonwick &vd->vdev_dtl_smo.smo_object); 509c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 510c5904d13Seschrock &vd->vdev_unspare); 511c5904d13Seschrock } 51221ecdf64SLin Ling 51321ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 51421ecdf64SLin Ling uint64_t spare = 0; 51521ecdf64SLin Ling 51621ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 51721ecdf64SLin Ling &spare) == 0 && spare) 51821ecdf64SLin Ling spa_spare_add(vd); 51921ecdf64SLin Ling } 52021ecdf64SLin Ling 521ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 522ecc2d604Sbonwick &vd->vdev_offline); 523c5904d13Seschrock 5243d7072f8Seschrock /* 5253d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5263d7072f8Seschrock * state, as the diagnosis made on another system may not be 527069f55e2SEric Schrock * valid in the current context. Local vdevs will 528069f55e2SEric Schrock * remain in the faulted state. 5293d7072f8Seschrock */ 530b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5313d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5323d7072f8Seschrock &vd->vdev_faulted); 5333d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5343d7072f8Seschrock &vd->vdev_degraded); 5353d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5363d7072f8Seschrock &vd->vdev_removed); 537069f55e2SEric Schrock 538069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 539069f55e2SEric Schrock char *aux; 540069f55e2SEric Schrock 541069f55e2SEric Schrock vd->vdev_label_aux = 542069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 543069f55e2SEric Schrock if (nvlist_lookup_string(nv, 544069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 545069f55e2SEric Schrock strcmp(aux, "external") == 0) 546069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 547069f55e2SEric Schrock } 5483d7072f8Seschrock } 549fa9e4066Sahrens } 550fa9e4066Sahrens 551fa9e4066Sahrens /* 552fa9e4066Sahrens * Add ourselves to the parent's list of children. 553fa9e4066Sahrens */ 554fa9e4066Sahrens vdev_add_child(parent, vd); 555fa9e4066Sahrens 55699653d4eSeschrock *vdp = vd; 55799653d4eSeschrock 55899653d4eSeschrock return (0); 559fa9e4066Sahrens } 560fa9e4066Sahrens 561fa9e4066Sahrens void 562fa9e4066Sahrens vdev_free(vdev_t *vd) 563fa9e4066Sahrens { 5643d7072f8Seschrock spa_t *spa = vd->vdev_spa; 565fa9e4066Sahrens 566fa9e4066Sahrens /* 567fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 568fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 569fa9e4066Sahrens */ 570fa9e4066Sahrens vdev_close(vd); 571fa9e4066Sahrens 572e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 573b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 574fa9e4066Sahrens 575fa9e4066Sahrens /* 576fa9e4066Sahrens * Free all children. 577fa9e4066Sahrens */ 578573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 579fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 580fa9e4066Sahrens 581fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 582fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 583fa9e4066Sahrens 584fa9e4066Sahrens /* 585fa9e4066Sahrens * Discard allocation state. 586fa9e4066Sahrens */ 587a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 588fa9e4066Sahrens vdev_metaslab_fini(vd); 589a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 590a1521560SJeff Bonwick } 591fa9e4066Sahrens 592fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 59399653d4eSeschrock ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 594fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 595fa9e4066Sahrens 596fa9e4066Sahrens /* 597fa9e4066Sahrens * Remove this vdev from its parent's child list. 598fa9e4066Sahrens */ 599fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 600fa9e4066Sahrens 601fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 602fa9e4066Sahrens 6033d7072f8Seschrock /* 6043d7072f8Seschrock * Clean up vdev structure. 6053d7072f8Seschrock */ 6063d7072f8Seschrock vdev_queue_fini(vd); 6073d7072f8Seschrock vdev_cache_fini(vd); 6083d7072f8Seschrock 6093d7072f8Seschrock if (vd->vdev_path) 6103d7072f8Seschrock spa_strfree(vd->vdev_path); 6113d7072f8Seschrock if (vd->vdev_devid) 6123d7072f8Seschrock spa_strfree(vd->vdev_devid); 6133d7072f8Seschrock if (vd->vdev_physpath) 6143d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6156809eb4eSEric Schrock if (vd->vdev_fru) 6166809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6173d7072f8Seschrock 6183d7072f8Seschrock if (vd->vdev_isspare) 6193d7072f8Seschrock spa_spare_remove(vd); 620fa94a07fSbrendan if (vd->vdev_isl2cache) 621fa94a07fSbrendan spa_l2cache_remove(vd); 6223d7072f8Seschrock 6233d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6243d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6258ad4d6ddSJeff Bonwick 6263d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6278ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6288ad4d6ddSJeff Bonwick space_map_unload(&vd->vdev_dtl[t]); 6298ad4d6ddSJeff Bonwick space_map_destroy(&vd->vdev_dtl[t]); 6308ad4d6ddSJeff Bonwick } 6313d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6328ad4d6ddSJeff Bonwick 6333d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6343d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 635e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6363d7072f8Seschrock 6373d7072f8Seschrock if (vd == spa->spa_root_vdev) 6383d7072f8Seschrock spa->spa_root_vdev = NULL; 6393d7072f8Seschrock 6403d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 641fa9e4066Sahrens } 642fa9e4066Sahrens 643fa9e4066Sahrens /* 644fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 645fa9e4066Sahrens */ 646fa9e4066Sahrens static void 647fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 648fa9e4066Sahrens { 649fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 650fa9e4066Sahrens metaslab_t *msp; 651fa9e4066Sahrens vdev_t *vd; 652fa9e4066Sahrens int t; 653fa9e4066Sahrens 654fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 655fa9e4066Sahrens 656fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 657fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 658fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 659fa9e4066Sahrens 660fa9e4066Sahrens svd->vdev_ms_array = 0; 661fa9e4066Sahrens svd->vdev_ms_shift = 0; 662fa9e4066Sahrens svd->vdev_ms_count = 0; 663fa9e4066Sahrens 664fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 665fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 666fa9e4066Sahrens 667fa9e4066Sahrens svd->vdev_mg = NULL; 668fa9e4066Sahrens svd->vdev_ms = NULL; 669ecc2d604Sbonwick 670ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 671ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 672fa9e4066Sahrens 673fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 674fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 67599653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 676fa9e4066Sahrens 677fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 678fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 67999653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 680fa9e4066Sahrens 681fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 682fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 683fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 684fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 685fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 686fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 687fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 688fa9e4066Sahrens } 689fa9e4066Sahrens 690e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 691fa9e4066Sahrens vdev_config_clean(svd); 692fa9e4066Sahrens vdev_config_dirty(tvd); 693fa9e4066Sahrens } 694fa9e4066Sahrens 695e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 696e14bb325SJeff Bonwick vdev_state_clean(svd); 697e14bb325SJeff Bonwick vdev_state_dirty(tvd); 698e14bb325SJeff Bonwick } 699e14bb325SJeff Bonwick 70099653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 70199653d4eSeschrock svd->vdev_deflate_ratio = 0; 7028654d025Sperrin 7038654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7048654d025Sperrin svd->vdev_islog = 0; 705fa9e4066Sahrens } 706fa9e4066Sahrens 707fa9e4066Sahrens static void 708fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 709fa9e4066Sahrens { 710fa9e4066Sahrens if (vd == NULL) 711fa9e4066Sahrens return; 712fa9e4066Sahrens 713fa9e4066Sahrens vd->vdev_top = tvd; 714fa9e4066Sahrens 715573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 716fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 717fa9e4066Sahrens } 718fa9e4066Sahrens 719fa9e4066Sahrens /* 720fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 721fa9e4066Sahrens */ 722fa9e4066Sahrens vdev_t * 723fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 724fa9e4066Sahrens { 725fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 726fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 727fa9e4066Sahrens vdev_t *mvd; 728fa9e4066Sahrens 729e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 730fa9e4066Sahrens 731fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 732ecc2d604Sbonwick 733ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 734573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 735ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 736ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 73788ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 738ecc2d604Sbonwick 739fa9e4066Sahrens vdev_remove_child(pvd, cvd); 740fa9e4066Sahrens vdev_add_child(pvd, mvd); 741fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 742fa9e4066Sahrens vdev_add_child(mvd, cvd); 743fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 744fa9e4066Sahrens 745fa9e4066Sahrens if (mvd == mvd->vdev_top) 746fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 747fa9e4066Sahrens 748fa9e4066Sahrens return (mvd); 749fa9e4066Sahrens } 750fa9e4066Sahrens 751fa9e4066Sahrens /* 752fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 753fa9e4066Sahrens */ 754fa9e4066Sahrens void 755fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 756fa9e4066Sahrens { 757fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 758fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 759fa9e4066Sahrens 760e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 761fa9e4066Sahrens 762fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 763fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 76499653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 76599653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 766ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 767fa9e4066Sahrens 768fa9e4066Sahrens vdev_remove_child(mvd, cvd); 769fa9e4066Sahrens vdev_remove_child(pvd, mvd); 7708ad4d6ddSJeff Bonwick 77199653d4eSeschrock /* 772e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 773e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 774e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 775e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 77699653d4eSeschrock */ 7778ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 7788ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 7791195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 7808ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 7818ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 7828ad4d6ddSJeff Bonwick } 783e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 784e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 785fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 786fa9e4066Sahrens 787fa9e4066Sahrens if (cvd == cvd->vdev_top) 788fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 789fa9e4066Sahrens 790fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 791fa9e4066Sahrens vdev_free(mvd); 792fa9e4066Sahrens } 793fa9e4066Sahrens 794ea8dc4b6Seschrock int 795fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 796fa9e4066Sahrens { 797fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 798ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 799ecc2d604Sbonwick uint64_t m; 800fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 801fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 802ecc2d604Sbonwick metaslab_t **mspp; 803ecc2d604Sbonwick int error; 804fa9e4066Sahrens 805a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 806a1521560SJeff Bonwick 80788ecc943SGeorge Wilson /* 80888ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 80988ecc943SGeorge Wilson */ 81088ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8110e34b6a7Sbonwick return (0); 8120e34b6a7Sbonwick 81388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 81488ecc943SGeorge Wilson 815e6ca193dSGeorge Wilson /* 816e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 817e6ca193dSGeorge Wilson * in 128k (1 << 17) because it is the current "typical" blocksize. 818e6ca193dSGeorge Wilson * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 819e6ca193dSGeorge Wilson * or we will inconsistently account for existing bp's. 820e6ca193dSGeorge Wilson */ 821e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 822e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 823e6ca193dSGeorge Wilson 824fa9e4066Sahrens ASSERT(oldc <= newc); 825fa9e4066Sahrens 826ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 827fa9e4066Sahrens 828ecc2d604Sbonwick if (oldc != 0) { 829ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 830ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 831ecc2d604Sbonwick } 832fa9e4066Sahrens 833ecc2d604Sbonwick vd->vdev_ms = mspp; 834ecc2d604Sbonwick vd->vdev_ms_count = newc; 835fa9e4066Sahrens 836ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 837ecc2d604Sbonwick space_map_obj_t smo = { 0, 0, 0 }; 838ecc2d604Sbonwick if (txg == 0) { 839ecc2d604Sbonwick uint64_t object = 0; 840ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8417bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8427bfdf011SNeil Perrin DMU_READ_PREFETCH); 843ecc2d604Sbonwick if (error) 844ecc2d604Sbonwick return (error); 845ecc2d604Sbonwick if (object != 0) { 846ecc2d604Sbonwick dmu_buf_t *db; 847ecc2d604Sbonwick error = dmu_bonus_hold(mos, object, FTAG, &db); 848ecc2d604Sbonwick if (error) 849ecc2d604Sbonwick return (error); 8501934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (smo)); 8511934e92fSmaybee bcopy(db->db_data, &smo, sizeof (smo)); 852ecc2d604Sbonwick ASSERT3U(smo.smo_object, ==, object); 853ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 854fa9e4066Sahrens } 855fa9e4066Sahrens } 856ecc2d604Sbonwick vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 857ecc2d604Sbonwick m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 858fa9e4066Sahrens } 859fa9e4066Sahrens 860a1521560SJeff Bonwick if (txg == 0) 861a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 862a1521560SJeff Bonwick 863a1521560SJeff Bonwick if (oldc == 0) 864a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 865a1521560SJeff Bonwick 866a1521560SJeff Bonwick if (txg == 0) 867a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 868a1521560SJeff Bonwick 869ea8dc4b6Seschrock return (0); 870fa9e4066Sahrens } 871fa9e4066Sahrens 872fa9e4066Sahrens void 873fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 874fa9e4066Sahrens { 875fa9e4066Sahrens uint64_t m; 876fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 877fa9e4066Sahrens 878fa9e4066Sahrens if (vd->vdev_ms != NULL) { 879a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 880fa9e4066Sahrens for (m = 0; m < count; m++) 881ecc2d604Sbonwick if (vd->vdev_ms[m] != NULL) 882ecc2d604Sbonwick metaslab_fini(vd->vdev_ms[m]); 883fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 884fa9e4066Sahrens vd->vdev_ms = NULL; 885fa9e4066Sahrens } 886fa9e4066Sahrens } 887fa9e4066Sahrens 888e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 889e14bb325SJeff Bonwick boolean_t vps_readable; 890e14bb325SJeff Bonwick boolean_t vps_writeable; 891e14bb325SJeff Bonwick int vps_flags; 892e14bb325SJeff Bonwick } vdev_probe_stats_t; 893e14bb325SJeff Bonwick 894e14bb325SJeff Bonwick static void 895e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 8960a4e9518Sgw { 8978ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 898a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 899e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 900a3f829aeSBill Moore 901a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 902e14bb325SJeff Bonwick 903e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 904e14bb325SJeff Bonwick if (zio->io_error == 0) 905e14bb325SJeff Bonwick vps->vps_readable = 1; 9068ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 907a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 908e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 909e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 910e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 911e14bb325SJeff Bonwick } else { 912e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 913e14bb325SJeff Bonwick } 914e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 915e14bb325SJeff Bonwick if (zio->io_error == 0) 916e14bb325SJeff Bonwick vps->vps_writeable = 1; 917e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 918e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 919a3f829aeSBill Moore zio_t *pio; 920e14bb325SJeff Bonwick 921e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 922e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 923e14bb325SJeff Bonwick 924e14bb325SJeff Bonwick if (vdev_readable(vd) && 9258ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 926e14bb325SJeff Bonwick zio->io_error = 0; 927e14bb325SJeff Bonwick } else { 928e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 929e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9308ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 931e14bb325SJeff Bonwick zio->io_error = ENXIO; 932e14bb325SJeff Bonwick } 933a3f829aeSBill Moore 934a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 935a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 936a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 937a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 938a3f829aeSBill Moore 939a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 940a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 941a3f829aeSBill Moore pio->io_error = ENXIO; 942a3f829aeSBill Moore 943e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 944e14bb325SJeff Bonwick } 945e14bb325SJeff Bonwick } 9460a4e9518Sgw 947e14bb325SJeff Bonwick /* 948e14bb325SJeff Bonwick * Determine whether this device is accessible by reading and writing 949e14bb325SJeff Bonwick * to several known locations: the pad regions of each vdev label 950e14bb325SJeff Bonwick * but the first (which we leave alone in case it contains a VTOC). 951e14bb325SJeff Bonwick */ 952e14bb325SJeff Bonwick zio_t * 953a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 954e14bb325SJeff Bonwick { 955e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 956a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 957a3f829aeSBill Moore zio_t *pio; 958a3f829aeSBill Moore 959a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9600a4e9518Sgw 961a3f829aeSBill Moore /* 962a3f829aeSBill Moore * Don't probe the probe. 963a3f829aeSBill Moore */ 964a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 965a3f829aeSBill Moore return (NULL); 966e14bb325SJeff Bonwick 967a3f829aeSBill Moore /* 968a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 969a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 970a3f829aeSBill Moore * this vdev will become parents of the probe io. 971a3f829aeSBill Moore */ 972a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 973e14bb325SJeff Bonwick 974a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 975a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 976a3f829aeSBill Moore 977a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 978a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 9798956713aSEric Schrock ZIO_FLAG_TRYHARD; 980a3f829aeSBill Moore 981a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 982a3f829aeSBill Moore /* 983a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 984a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 985a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 986a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 987a3f829aeSBill Moore * any zio looking at these values can assume that 988a3f829aeSBill Moore * failures persist for the life of the I/O. That's 989a3f829aeSBill Moore * important because when a device has intermittent 990a3f829aeSBill Moore * connectivity problems, we want to ensure that 991a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 992a3f829aeSBill Moore * the zio (EIO). 993a3f829aeSBill Moore * 994a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 995a3f829aeSBill Moore * values so the probe can reevaluate from first 996a3f829aeSBill Moore * principles. 997a3f829aeSBill Moore */ 998a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 999a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1000a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1001a3f829aeSBill Moore } 1002a3f829aeSBill Moore 1003a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1004a3f829aeSBill Moore vdev_probe_done, vps, 1005a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1006a3f829aeSBill Moore 1007*98d1cbfeSGeorge Wilson /* 1008*98d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 1009*98d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 1010*98d1cbfeSGeorge Wilson */ 1011a3f829aeSBill Moore if (zio != NULL) { 1012a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1013a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1014a3f829aeSBill Moore } 1015e14bb325SJeff Bonwick } 1016e14bb325SJeff Bonwick 1017a3f829aeSBill Moore if (zio != NULL) 1018a3f829aeSBill Moore zio_add_child(zio, pio); 1019e14bb325SJeff Bonwick 1020a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1021e14bb325SJeff Bonwick 1022a3f829aeSBill Moore if (vps == NULL) { 1023a3f829aeSBill Moore ASSERT(zio != NULL); 1024a3f829aeSBill Moore return (NULL); 1025a3f829aeSBill Moore } 1026e14bb325SJeff Bonwick 1027e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1028a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1029e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1030f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1031f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1032e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1033e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1034e14bb325SJeff Bonwick } 1035e14bb325SJeff Bonwick 1036a3f829aeSBill Moore if (zio == NULL) 1037a3f829aeSBill Moore return (pio); 1038a3f829aeSBill Moore 1039a3f829aeSBill Moore zio_nowait(pio); 1040a3f829aeSBill Moore return (NULL); 10410a4e9518Sgw } 10420a4e9518Sgw 1043f64c0e34SEric Taylor static void 1044f64c0e34SEric Taylor vdev_open_child(void *arg) 1045f64c0e34SEric Taylor { 1046f64c0e34SEric Taylor vdev_t *vd = arg; 1047f64c0e34SEric Taylor 1048f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1049f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1050f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1051f64c0e34SEric Taylor } 1052f64c0e34SEric Taylor 1053681d9761SEric Taylor boolean_t 1054681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1055681d9761SEric Taylor { 1056681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1057681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1058681d9761SEric Taylor return (B_TRUE); 1059681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1060681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1061681d9761SEric Taylor return (B_TRUE); 1062681d9761SEric Taylor return (B_FALSE); 1063681d9761SEric Taylor } 1064681d9761SEric Taylor 1065f64c0e34SEric Taylor void 1066f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1067f64c0e34SEric Taylor { 1068f64c0e34SEric Taylor taskq_t *tq; 1069f64c0e34SEric Taylor int children = vd->vdev_children; 1070f64c0e34SEric Taylor 1071681d9761SEric Taylor /* 1072681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1073681d9761SEric Taylor * in a single thread so that the same thread holds the 1074681d9761SEric Taylor * spa_namespace_lock 1075681d9761SEric Taylor */ 1076681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1077681d9761SEric Taylor for (int c = 0; c < children; c++) 1078681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1079681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1080681d9761SEric Taylor return; 1081681d9761SEric Taylor } 1082f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1083f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1084f64c0e34SEric Taylor 1085f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1086f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1087f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1088f64c0e34SEric Taylor 1089f64c0e34SEric Taylor taskq_destroy(tq); 1090f64c0e34SEric Taylor } 1091f64c0e34SEric Taylor 1092fa9e4066Sahrens /* 1093fa9e4066Sahrens * Prepare a virtual device for access. 1094fa9e4066Sahrens */ 1095fa9e4066Sahrens int 1096fa9e4066Sahrens vdev_open(vdev_t *vd) 1097fa9e4066Sahrens { 10988ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1099fa9e4066Sahrens int error; 1100fa9e4066Sahrens uint64_t osize = 0; 1101fa9e4066Sahrens uint64_t asize, psize; 1102ecc2d604Sbonwick uint64_t ashift = 0; 1103fa9e4066Sahrens 1104f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1105f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1106fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1107fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1108fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1109fa9e4066Sahrens 1110fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1111e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1112e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1113573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1114fa9e4066Sahrens 1115069f55e2SEric Schrock /* 1116069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1117069f55e2SEric Schrock * faulted, bail out of the open. 1118069f55e2SEric Schrock */ 11193d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11203d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1121069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1122069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11233d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1124069f55e2SEric Schrock vd->vdev_label_aux); 11253d7072f8Seschrock return (ENXIO); 11263d7072f8Seschrock } else if (vd->vdev_offline) { 1127fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1128ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1129fa9e4066Sahrens return (ENXIO); 1130fa9e4066Sahrens } 1131fa9e4066Sahrens 1132fa9e4066Sahrens error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 1133fa9e4066Sahrens 1134095bcd66SGeorge Wilson /* 1135095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1136095bcd66SGeorge Wilson * the vdev on error. 1137095bcd66SGeorge Wilson */ 1138095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1139ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11408956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1141ea8dc4b6Seschrock 1142fa9e4066Sahrens if (error) { 11433d7072f8Seschrock if (vd->vdev_removed && 11443d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11453d7072f8Seschrock vd->vdev_removed = B_FALSE; 11463d7072f8Seschrock 1147ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1148fa9e4066Sahrens vd->vdev_stat.vs_aux); 1149fa9e4066Sahrens return (error); 1150fa9e4066Sahrens } 1151fa9e4066Sahrens 11523d7072f8Seschrock vd->vdev_removed = B_FALSE; 11533d7072f8Seschrock 1154096d22d4SEric Schrock /* 1155096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1156096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1157096d22d4SEric Schrock */ 1158096d22d4SEric Schrock if (vd->vdev_faulted) { 1159096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1160096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1161096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1162096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1163096d22d4SEric Schrock vd->vdev_label_aux); 1164096d22d4SEric Schrock return (ENXIO); 1165096d22d4SEric Schrock } 1166096d22d4SEric Schrock 11673d7072f8Seschrock if (vd->vdev_degraded) { 11683d7072f8Seschrock ASSERT(vd->vdev_children == 0); 11693d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 11703d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 11713d7072f8Seschrock } else { 1172069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 11733d7072f8Seschrock } 1174fa9e4066Sahrens 117588ecc943SGeorge Wilson /* 117688ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 117788ecc943SGeorge Wilson */ 117888ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 117988ecc943SGeorge Wilson return (0); 118088ecc943SGeorge Wilson 1181573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1182ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1183ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1184ea8dc4b6Seschrock VDEV_AUX_NONE); 1185ea8dc4b6Seschrock break; 1186ea8dc4b6Seschrock } 1187573ca77eSGeorge Wilson } 1188fa9e4066Sahrens 1189fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1190fa9e4066Sahrens 1191fa9e4066Sahrens if (vd->vdev_children == 0) { 1192fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1193ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1194ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1195fa9e4066Sahrens return (EOVERFLOW); 1196fa9e4066Sahrens } 1197fa9e4066Sahrens psize = osize; 1198fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1199fa9e4066Sahrens } else { 1200ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1201fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1202ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1203ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1204fa9e4066Sahrens return (EOVERFLOW); 1205fa9e4066Sahrens } 1206fa9e4066Sahrens psize = 0; 1207fa9e4066Sahrens asize = osize; 1208fa9e4066Sahrens } 1209fa9e4066Sahrens 1210fa9e4066Sahrens vd->vdev_psize = psize; 1211fa9e4066Sahrens 1212573ca77eSGeorge Wilson /* 1213573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1214573ca77eSGeorge Wilson */ 1215573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1216573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1217573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1218573ca77eSGeorge Wilson return (EINVAL); 1219573ca77eSGeorge Wilson } 1220573ca77eSGeorge Wilson 1221fa9e4066Sahrens if (vd->vdev_asize == 0) { 1222fa9e4066Sahrens /* 1223fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1224ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1225fa9e4066Sahrens */ 1226fa9e4066Sahrens vd->vdev_asize = asize; 1227ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1228fa9e4066Sahrens } else { 1229fa9e4066Sahrens /* 1230fa9e4066Sahrens * Make sure the alignment requirement hasn't increased. 1231fa9e4066Sahrens */ 1232ecc2d604Sbonwick if (ashift > vd->vdev_top->vdev_ashift) { 1233ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1234ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 1235fa9e4066Sahrens return (EINVAL); 1236fa9e4066Sahrens } 1237573ca77eSGeorge Wilson } 1238fa9e4066Sahrens 1239573ca77eSGeorge Wilson /* 1240573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1241573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1242573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1243573ca77eSGeorge Wilson */ 1244573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1245573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1246573ca77eSGeorge Wilson vd->vdev_asize = asize; 1247fa9e4066Sahrens 1248573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1249fa9e4066Sahrens 12500a4e9518Sgw /* 12510a4e9518Sgw * Ensure we can issue some IO before declaring the 12520a4e9518Sgw * vdev open for business. 12530a4e9518Sgw */ 1254e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1255e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1256*98d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1257*98d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 12580a4e9518Sgw return (error); 12590a4e9518Sgw } 12600a4e9518Sgw 1261088f3894Sahrens /* 1262088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 12638ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 12648ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1265088f3894Sahrens */ 12668ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 12678ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 12688ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1269088f3894Sahrens 1270fa9e4066Sahrens return (0); 1271fa9e4066Sahrens } 1272fa9e4066Sahrens 1273560e6e96Seschrock /* 1274560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1275560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 12763d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1277560e6e96Seschrock * 1278560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1279560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1280560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1281560e6e96Seschrock * will be updated but the function will return 0. 1282560e6e96Seschrock */ 1283560e6e96Seschrock int 1284560e6e96Seschrock vdev_validate(vdev_t *vd) 1285560e6e96Seschrock { 1286560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1287560e6e96Seschrock nvlist_t *label; 12881195e687SMark J Musante uint64_t guid = 0, top_guid; 1289560e6e96Seschrock uint64_t state; 1290560e6e96Seschrock 1291573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1292560e6e96Seschrock if (vdev_validate(vd->vdev_child[c]) != 0) 12930bf246f5Smc return (EBADF); 1294560e6e96Seschrock 1295b5989ec7Seschrock /* 1296b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1297b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1298b5989ec7Seschrock * overwrite the previous state. 1299b5989ec7Seschrock */ 1300e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13011195e687SMark J Musante uint64_t aux_guid = 0; 13021195e687SMark J Musante nvlist_t *nvl; 1303560e6e96Seschrock 1304560e6e96Seschrock if ((label = vdev_label_read_config(vd)) == NULL) { 1305560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1306560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1307560e6e96Seschrock return (0); 1308560e6e96Seschrock } 1309560e6e96Seschrock 13101195e687SMark J Musante /* 13111195e687SMark J Musante * Determine if this vdev has been split off into another 13121195e687SMark J Musante * pool. If so, then refuse to open it. 13131195e687SMark J Musante */ 13141195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 13151195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 13161195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 13171195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 13181195e687SMark J Musante nvlist_free(label); 13191195e687SMark J Musante return (0); 13201195e687SMark J Musante } 13211195e687SMark J Musante 1322560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1323560e6e96Seschrock &guid) != 0 || guid != spa_guid(spa)) { 1324560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1325560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1326560e6e96Seschrock nvlist_free(label); 1327560e6e96Seschrock return (0); 1328560e6e96Seschrock } 1329560e6e96Seschrock 13301195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 13311195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 13321195e687SMark J Musante &aux_guid) != 0) 13331195e687SMark J Musante aux_guid = 0; 13341195e687SMark J Musante 1335e14bb325SJeff Bonwick /* 1336e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1337e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1338e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1339e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1340e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1341e14bb325SJeff Bonwick * safely compare to that instead. 13421195e687SMark J Musante * 13431195e687SMark J Musante * If we split this vdev off instead, then we also check the 13441195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 13451195e687SMark J Musante * corrupt if it is partway through a split operation. 1346e14bb325SJeff Bonwick */ 1347560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1348e14bb325SJeff Bonwick &guid) != 0 || 1349e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1350e14bb325SJeff Bonwick &top_guid) != 0 || 13511195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1352e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1353560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1354560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1355560e6e96Seschrock nvlist_free(label); 1356560e6e96Seschrock return (0); 1357560e6e96Seschrock } 1358560e6e96Seschrock 1359560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1360560e6e96Seschrock &state) != 0) { 1361560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1362560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1363560e6e96Seschrock nvlist_free(label); 1364560e6e96Seschrock return (0); 1365560e6e96Seschrock } 1366560e6e96Seschrock 1367560e6e96Seschrock nvlist_free(label); 1368560e6e96Seschrock 1369bc758434SLin Ling /* 1370bc758434SLin Ling * If spa->spa_load_verbatim is true, no need to check the 1371bc758434SLin Ling * state of the pool. 1372bc758434SLin Ling */ 1373bc758434SLin Ling if (!spa->spa_load_verbatim && 1374b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1375bc758434SLin Ling state != POOL_STATE_ACTIVE) 13760bf246f5Smc return (EBADF); 1377560e6e96Seschrock 137851ece835Seschrock /* 137951ece835Seschrock * If we were able to open and validate a vdev that was 138051ece835Seschrock * previously marked permanently unavailable, clear that state 138151ece835Seschrock * now. 138251ece835Seschrock */ 138351ece835Seschrock if (vd->vdev_not_present) 138451ece835Seschrock vd->vdev_not_present = 0; 138551ece835Seschrock } 1386560e6e96Seschrock 1387560e6e96Seschrock return (0); 1388560e6e96Seschrock } 1389560e6e96Seschrock 1390fa9e4066Sahrens /* 1391fa9e4066Sahrens * Close a virtual device. 1392fa9e4066Sahrens */ 1393fa9e4066Sahrens void 1394fa9e4066Sahrens vdev_close(vdev_t *vd) 1395fa9e4066Sahrens { 13968ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1397095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 13988ad4d6ddSJeff Bonwick 13998ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14008ad4d6ddSJeff Bonwick 14011195e687SMark J Musante /* 14021195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14031195e687SMark J Musante * going offline. 14041195e687SMark J Musante */ 1405095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14061195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1407095bcd66SGeorge Wilson 1408fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1409fa9e4066Sahrens 14103d7072f8Seschrock vdev_cache_purge(vd); 1411fa9e4066Sahrens 1412560e6e96Seschrock /* 1413573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1414560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1415560e6e96Seschrock * it's still faulted. 1416560e6e96Seschrock */ 1417560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1418560e6e96Seschrock 1419fa9e4066Sahrens if (vd->vdev_offline) 1420fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1421fa9e4066Sahrens else 1422fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1423ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1424fa9e4066Sahrens } 1425fa9e4066Sahrens 1426dcba9f3fSGeorge Wilson void 1427dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1428dcba9f3fSGeorge Wilson { 1429dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1430dcba9f3fSGeorge Wilson 1431dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1432dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1433dcba9f3fSGeorge Wilson return; 1434dcba9f3fSGeorge Wilson 1435dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1436dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1437dcba9f3fSGeorge Wilson 1438dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1439dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1440dcba9f3fSGeorge Wilson } 1441dcba9f3fSGeorge Wilson 1442dcba9f3fSGeorge Wilson void 1443dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1444dcba9f3fSGeorge Wilson { 1445dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1446dcba9f3fSGeorge Wilson 1447dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1448dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1449dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1450dcba9f3fSGeorge Wilson 1451dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1452dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1453dcba9f3fSGeorge Wilson } 1454dcba9f3fSGeorge Wilson 1455095bcd66SGeorge Wilson /* 1456095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1457095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1458095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1459095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1460095bcd66SGeorge Wilson */ 1461fa9e4066Sahrens void 1462ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1463fa9e4066Sahrens { 1464ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1465fa9e4066Sahrens 1466e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1467ea8dc4b6Seschrock 14681195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 14691195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1470fa9e4066Sahrens vdev_close(vd); 1471fa9e4066Sahrens (void) vdev_open(vd); 1472fa9e4066Sahrens 147339c23413Seschrock /* 147439c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 147539c23413Seschrock * Otherwise, a device with an invalid label could be successfully 147639c23413Seschrock * opened in response to vdev_reopen(). 147739c23413Seschrock */ 1478c5904d13Seschrock if (vd->vdev_aux) { 1479c5904d13Seschrock (void) vdev_validate_aux(vd); 1480e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 14816809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1482573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1483573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1484c5904d13Seschrock } else { 1485c5904d13Seschrock (void) vdev_validate(vd); 1486c5904d13Seschrock } 148739c23413Seschrock 1488fa9e4066Sahrens /* 14893d7072f8Seschrock * Reassess parent vdev's health. 1490fa9e4066Sahrens */ 14913d7072f8Seschrock vdev_propagate_state(vd); 1492fa9e4066Sahrens } 1493fa9e4066Sahrens 1494fa9e4066Sahrens int 149599653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1496fa9e4066Sahrens { 1497fa9e4066Sahrens int error; 1498fa9e4066Sahrens 1499fa9e4066Sahrens /* 1500fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1501fa9e4066Sahrens * For a create, however, we want to fail the request if 1502fa9e4066Sahrens * there are any components we can't open. 1503fa9e4066Sahrens */ 1504fa9e4066Sahrens error = vdev_open(vd); 1505fa9e4066Sahrens 1506fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1507fa9e4066Sahrens vdev_close(vd); 1508fa9e4066Sahrens return (error ? error : ENXIO); 1509fa9e4066Sahrens } 1510fa9e4066Sahrens 1511fa9e4066Sahrens /* 1512fa9e4066Sahrens * Recursively initialize all labels. 1513fa9e4066Sahrens */ 151439c23413Seschrock if ((error = vdev_label_init(vd, txg, isreplacing ? 151539c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1516fa9e4066Sahrens vdev_close(vd); 1517fa9e4066Sahrens return (error); 1518fa9e4066Sahrens } 1519fa9e4066Sahrens 1520fa9e4066Sahrens return (0); 1521fa9e4066Sahrens } 1522fa9e4066Sahrens 15230e34b6a7Sbonwick void 1524573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1525fa9e4066Sahrens { 1526fa9e4066Sahrens /* 1527fa9e4066Sahrens * Aim for roughly 200 metaslabs per vdev. 1528fa9e4066Sahrens */ 1529fa9e4066Sahrens vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1530fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1531fa9e4066Sahrens } 1532fa9e4066Sahrens 1533fa9e4066Sahrens void 1534ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1535fa9e4066Sahrens { 1536ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 153788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1538ecc2d604Sbonwick ASSERT(ISP2(flags)); 1539fa9e4066Sahrens 1540ecc2d604Sbonwick if (flags & VDD_METASLAB) 1541ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1542ecc2d604Sbonwick 1543ecc2d604Sbonwick if (flags & VDD_DTL) 1544ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1545ecc2d604Sbonwick 1546ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1547fa9e4066Sahrens } 1548fa9e4066Sahrens 15498ad4d6ddSJeff Bonwick /* 15508ad4d6ddSJeff Bonwick * DTLs. 15518ad4d6ddSJeff Bonwick * 15528ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 15539fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 15548ad4d6ddSJeff Bonwick * 15558ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 15568ad4d6ddSJeff Bonwick * 15578ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 15588ad4d6ddSJeff Bonwick * 15598ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 15608ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 15618ad4d6ddSJeff Bonwick * txgs that was scrubbed. 15628ad4d6ddSJeff Bonwick * 15638ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 15648ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 15658ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 15668ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 15678ad4d6ddSJeff Bonwick * determine whether a device can be detached. 15688ad4d6ddSJeff Bonwick * 15698ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 15708ad4d6ddSJeff Bonwick * either has the data or it doesn't. 15718ad4d6ddSJeff Bonwick * 15728ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 15738ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 15748ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 15758ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 15768ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 15778ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 15788ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 15798ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 15808ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 15818ad4d6ddSJeff Bonwick * 15828ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 15838ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 15848ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 15858ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 15868ad4d6ddSJeff Bonwick */ 1587fa9e4066Sahrens void 15888ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1589fa9e4066Sahrens { 15908ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 15918ad4d6ddSJeff Bonwick 15928ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 15938ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 15948ad4d6ddSJeff Bonwick 1595fa9e4066Sahrens mutex_enter(sm->sm_lock); 1596fa9e4066Sahrens if (!space_map_contains(sm, txg, size)) 1597fa9e4066Sahrens space_map_add(sm, txg, size); 1598fa9e4066Sahrens mutex_exit(sm->sm_lock); 1599fa9e4066Sahrens } 1600fa9e4066Sahrens 16018ad4d6ddSJeff Bonwick boolean_t 16028ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1603fa9e4066Sahrens { 16048ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 16058ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1606fa9e4066Sahrens 16078ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16088ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1609fa9e4066Sahrens 1610fa9e4066Sahrens mutex_enter(sm->sm_lock); 16118ad4d6ddSJeff Bonwick if (sm->sm_space != 0) 16128ad4d6ddSJeff Bonwick dirty = space_map_contains(sm, txg, size); 1613fa9e4066Sahrens mutex_exit(sm->sm_lock); 1614fa9e4066Sahrens 1615fa9e4066Sahrens return (dirty); 1616fa9e4066Sahrens } 1617fa9e4066Sahrens 16188ad4d6ddSJeff Bonwick boolean_t 16198ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 16208ad4d6ddSJeff Bonwick { 16218ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 16228ad4d6ddSJeff Bonwick boolean_t empty; 16238ad4d6ddSJeff Bonwick 16248ad4d6ddSJeff Bonwick mutex_enter(sm->sm_lock); 16258ad4d6ddSJeff Bonwick empty = (sm->sm_space == 0); 16268ad4d6ddSJeff Bonwick mutex_exit(sm->sm_lock); 16278ad4d6ddSJeff Bonwick 16288ad4d6ddSJeff Bonwick return (empty); 16298ad4d6ddSJeff Bonwick } 16308ad4d6ddSJeff Bonwick 1631fa9e4066Sahrens /* 1632fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1633fa9e4066Sahrens */ 1634fa9e4066Sahrens void 1635fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1636fa9e4066Sahrens { 1637ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 16388ad4d6ddSJeff Bonwick avl_tree_t reftree; 16398ad4d6ddSJeff Bonwick int minref; 1640fa9e4066Sahrens 16418ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1642fa9e4066Sahrens 16438ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 16448ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 16458ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 16468ad4d6ddSJeff Bonwick 1647b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 16488ad4d6ddSJeff Bonwick return; 16498ad4d6ddSJeff Bonwick 16508ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 1651fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1652088f3894Sahrens if (scrub_txg != 0 && 1653088f3894Sahrens (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { 1654088f3894Sahrens /* 1655088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1656088f3894Sahrens * did it without rebooting, then the scrub dtl 1657088f3894Sahrens * will be valid, so excise the old region and 1658088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1659088f3894Sahrens * dtl as-is if there was an error. 16608ad4d6ddSJeff Bonwick * 16618ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 16628ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 16638ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 16648ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 16658ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 16668ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 16678ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 16688ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 16698ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1670088f3894Sahrens */ 16718ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 16728ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16738ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16748ad4d6ddSJeff Bonwick space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 16758ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16768ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_SCRUB], 2); 16778ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, 16788ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16798ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1680fa9e4066Sahrens } 16818ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 16828ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 16838ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1684fa9e4066Sahrens if (scrub_done) 16858ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 16868ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 16878ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 16888ad4d6ddSJeff Bonwick space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 16898ad4d6ddSJeff Bonwick else 16908ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 16918ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1692fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1693088f3894Sahrens 1694ecc2d604Sbonwick if (txg != 0) 1695ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1696fa9e4066Sahrens return; 1697fa9e4066Sahrens } 1698fa9e4066Sahrens 1699fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 17008ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 170199bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 170299bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 17038ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 17048ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 17058ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 17068ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 17078ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 17088ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 17098ad4d6ddSJeff Bonwick else 17108ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 17118ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 17128ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 17138ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 17148ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 171599bb17e2SEric Taylor space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 17168ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 17178ad4d6ddSJeff Bonwick } 17188ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 17198ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1720fa9e4066Sahrens } 17218ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1722fa9e4066Sahrens } 1723fa9e4066Sahrens 1724fa9e4066Sahrens static int 1725fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1726fa9e4066Sahrens { 1727fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 17288ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 1729ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1730fa9e4066Sahrens dmu_buf_t *db; 1731fa9e4066Sahrens int error; 1732fa9e4066Sahrens 1733fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1734fa9e4066Sahrens 1735fa9e4066Sahrens if (smo->smo_object == 0) 1736fa9e4066Sahrens return (0); 1737fa9e4066Sahrens 173888ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 173988ecc943SGeorge Wilson 1740ecc2d604Sbonwick if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1741ea8dc4b6Seschrock return (error); 1742ecc2d604Sbonwick 17431934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 17441934e92fSmaybee bcopy(db->db_data, smo, sizeof (*smo)); 1745ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1746fa9e4066Sahrens 1747fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 17488ad4d6ddSJeff Bonwick error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 17498ad4d6ddSJeff Bonwick NULL, SM_ALLOC, smo, mos); 1750fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1751fa9e4066Sahrens 1752fa9e4066Sahrens return (error); 1753fa9e4066Sahrens } 1754fa9e4066Sahrens 1755fa9e4066Sahrens void 1756fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1757fa9e4066Sahrens { 1758fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 17598ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 17608ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1761ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1762fa9e4066Sahrens space_map_t smsync; 1763fa9e4066Sahrens kmutex_t smlock; 1764fa9e4066Sahrens dmu_buf_t *db; 1765fa9e4066Sahrens dmu_tx_t *tx; 1766fa9e4066Sahrens 176788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 176888ecc943SGeorge Wilson 1769fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1770fa9e4066Sahrens 1771fa9e4066Sahrens if (vd->vdev_detached) { 1772fa9e4066Sahrens if (smo->smo_object != 0) { 1773ecc2d604Sbonwick int err = dmu_object_free(mos, smo->smo_object, tx); 1774fa9e4066Sahrens ASSERT3U(err, ==, 0); 1775fa9e4066Sahrens smo->smo_object = 0; 1776fa9e4066Sahrens } 1777fa9e4066Sahrens dmu_tx_commit(tx); 1778fa9e4066Sahrens return; 1779fa9e4066Sahrens } 1780fa9e4066Sahrens 1781fa9e4066Sahrens if (smo->smo_object == 0) { 1782fa9e4066Sahrens ASSERT(smo->smo_objsize == 0); 1783fa9e4066Sahrens ASSERT(smo->smo_alloc == 0); 1784ecc2d604Sbonwick smo->smo_object = dmu_object_alloc(mos, 1785fa9e4066Sahrens DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1786fa9e4066Sahrens DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1787fa9e4066Sahrens ASSERT(smo->smo_object != 0); 1788fa9e4066Sahrens vdev_config_dirty(vd->vdev_top); 1789fa9e4066Sahrens } 1790fa9e4066Sahrens 1791fa9e4066Sahrens mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1792fa9e4066Sahrens 1793fa9e4066Sahrens space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1794fa9e4066Sahrens &smlock); 1795fa9e4066Sahrens 1796fa9e4066Sahrens mutex_enter(&smlock); 1797fa9e4066Sahrens 1798fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1799ecc2d604Sbonwick space_map_walk(sm, space_map_add, &smsync); 1800fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1801fa9e4066Sahrens 1802ecc2d604Sbonwick space_map_truncate(smo, mos, tx); 1803ecc2d604Sbonwick space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1804fa9e4066Sahrens 1805fa9e4066Sahrens space_map_destroy(&smsync); 1806fa9e4066Sahrens 1807fa9e4066Sahrens mutex_exit(&smlock); 1808fa9e4066Sahrens mutex_destroy(&smlock); 1809fa9e4066Sahrens 1810ecc2d604Sbonwick VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1811fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 18121934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 18131934e92fSmaybee bcopy(smo, db->db_data, sizeof (*smo)); 1814ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1815fa9e4066Sahrens 1816fa9e4066Sahrens dmu_tx_commit(tx); 1817fa9e4066Sahrens } 1818fa9e4066Sahrens 18198ad4d6ddSJeff Bonwick /* 18208ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 18218ad4d6ddSJeff Bonwick * without losing data. 18228ad4d6ddSJeff Bonwick */ 18238ad4d6ddSJeff Bonwick boolean_t 18248ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 18258ad4d6ddSJeff Bonwick { 18268ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 18278ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 18288ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 18298ad4d6ddSJeff Bonwick boolean_t required; 18308ad4d6ddSJeff Bonwick 18318ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 18328ad4d6ddSJeff Bonwick 18338ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 18348ad4d6ddSJeff Bonwick return (B_TRUE); 18358ad4d6ddSJeff Bonwick 18368ad4d6ddSJeff Bonwick /* 18378ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 18388ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 18398ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 18408ad4d6ddSJeff Bonwick */ 18418ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 18428ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 18438ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 18448ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 18458ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 18468ad4d6ddSJeff Bonwick 18478ad4d6ddSJeff Bonwick return (required); 18488ad4d6ddSJeff Bonwick } 18498ad4d6ddSJeff Bonwick 1850088f3894Sahrens /* 1851088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 1852088f3894Sahrens */ 1853088f3894Sahrens boolean_t 1854088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1855088f3894Sahrens { 1856088f3894Sahrens boolean_t needed = B_FALSE; 1857088f3894Sahrens uint64_t thismin = UINT64_MAX; 1858088f3894Sahrens uint64_t thismax = 0; 1859088f3894Sahrens 1860088f3894Sahrens if (vd->vdev_children == 0) { 1861088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 18628ad4d6ddSJeff Bonwick if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 18638ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 1864088f3894Sahrens space_seg_t *ss; 1865088f3894Sahrens 18668ad4d6ddSJeff Bonwick ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1867088f3894Sahrens thismin = ss->ss_start - 1; 18688ad4d6ddSJeff Bonwick ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1869088f3894Sahrens thismax = ss->ss_end; 1870088f3894Sahrens needed = B_TRUE; 1871088f3894Sahrens } 1872088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 1873088f3894Sahrens } else { 18748ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 1875088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 1876088f3894Sahrens uint64_t cmin, cmax; 1877088f3894Sahrens 1878088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1879088f3894Sahrens thismin = MIN(thismin, cmin); 1880088f3894Sahrens thismax = MAX(thismax, cmax); 1881088f3894Sahrens needed = B_TRUE; 1882088f3894Sahrens } 1883088f3894Sahrens } 1884088f3894Sahrens } 1885088f3894Sahrens 1886088f3894Sahrens if (needed && minp) { 1887088f3894Sahrens *minp = thismin; 1888088f3894Sahrens *maxp = thismax; 1889088f3894Sahrens } 1890088f3894Sahrens return (needed); 1891088f3894Sahrens } 1892088f3894Sahrens 1893560e6e96Seschrock void 1894ea8dc4b6Seschrock vdev_load(vdev_t *vd) 1895fa9e4066Sahrens { 1896fa9e4066Sahrens /* 1897fa9e4066Sahrens * Recursively load all children. 1898fa9e4066Sahrens */ 18998ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1900560e6e96Seschrock vdev_load(vd->vdev_child[c]); 1901fa9e4066Sahrens 1902fa9e4066Sahrens /* 19030e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 1904fa9e4066Sahrens */ 190588ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 1906560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1907560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 1908560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1909560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1910fa9e4066Sahrens 1911fa9e4066Sahrens /* 1912fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 1913fa9e4066Sahrens */ 1914560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1915560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1916560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1917fa9e4066Sahrens } 1918fa9e4066Sahrens 191999653d4eSeschrock /* 1920fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 1921fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 1922fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 1923fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 1924fa94a07fSbrendan * repurposed to another pool. 192599653d4eSeschrock */ 192699653d4eSeschrock int 1927fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 192899653d4eSeschrock { 192999653d4eSeschrock nvlist_t *label; 193099653d4eSeschrock uint64_t guid, version; 193199653d4eSeschrock uint64_t state; 193299653d4eSeschrock 1933e14bb325SJeff Bonwick if (!vdev_readable(vd)) 1934c5904d13Seschrock return (0); 1935c5904d13Seschrock 193699653d4eSeschrock if ((label = vdev_label_read_config(vd)) == NULL) { 193799653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 193899653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 193999653d4eSeschrock return (-1); 194099653d4eSeschrock } 194199653d4eSeschrock 194299653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1943e7437265Sahrens version > SPA_VERSION || 194499653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 194599653d4eSeschrock guid != vd->vdev_guid || 194699653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 194799653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 194899653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 194999653d4eSeschrock nvlist_free(label); 195099653d4eSeschrock return (-1); 195199653d4eSeschrock } 195299653d4eSeschrock 195399653d4eSeschrock /* 195499653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 195599653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 195699653d4eSeschrock */ 195799653d4eSeschrock nvlist_free(label); 195899653d4eSeschrock return (0); 195999653d4eSeschrock } 196099653d4eSeschrock 196188ecc943SGeorge Wilson void 196288ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 196388ecc943SGeorge Wilson { 196488ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 196588ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 196688ecc943SGeorge Wilson dmu_tx_t *tx; 196788ecc943SGeorge Wilson 196888ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 196988ecc943SGeorge Wilson 197088ecc943SGeorge Wilson if (vd->vdev_dtl_smo.smo_object) { 197188ecc943SGeorge Wilson ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); 197288ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 197388ecc943SGeorge Wilson vd->vdev_dtl_smo.smo_object = 0; 197488ecc943SGeorge Wilson } 197588ecc943SGeorge Wilson 197688ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 197788ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 197888ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 197988ecc943SGeorge Wilson 198088ecc943SGeorge Wilson if (msp == NULL || msp->ms_smo.smo_object == 0) 198188ecc943SGeorge Wilson continue; 198288ecc943SGeorge Wilson 198388ecc943SGeorge Wilson ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); 198488ecc943SGeorge Wilson (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 198588ecc943SGeorge Wilson msp->ms_smo.smo_object = 0; 198688ecc943SGeorge Wilson } 198788ecc943SGeorge Wilson } 198888ecc943SGeorge Wilson 198988ecc943SGeorge Wilson if (vd->vdev_ms_array) { 199088ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 199188ecc943SGeorge Wilson vd->vdev_ms_array = 0; 199288ecc943SGeorge Wilson vd->vdev_ms_shift = 0; 199388ecc943SGeorge Wilson } 199488ecc943SGeorge Wilson dmu_tx_commit(tx); 199588ecc943SGeorge Wilson } 199688ecc943SGeorge Wilson 1997fa9e4066Sahrens void 1998fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 1999fa9e4066Sahrens { 2000fa9e4066Sahrens metaslab_t *msp; 200180eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2002fa9e4066Sahrens 200388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 200488ecc943SGeorge Wilson 2005fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2006fa9e4066Sahrens metaslab_sync_done(msp, txg); 200780eb36f2SGeorge Wilson 200880eb36f2SGeorge Wilson if (reassess) 200980eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2010fa9e4066Sahrens } 2011fa9e4066Sahrens 2012fa9e4066Sahrens void 2013fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2014fa9e4066Sahrens { 2015fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2016fa9e4066Sahrens vdev_t *lvd; 2017fa9e4066Sahrens metaslab_t *msp; 2018ecc2d604Sbonwick dmu_tx_t *tx; 2019fa9e4066Sahrens 202088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 202188ecc943SGeorge Wilson 2022ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2023ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2024ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2025ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2026ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2027ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2028ecc2d604Sbonwick vdev_config_dirty(vd); 2029ecc2d604Sbonwick dmu_tx_commit(tx); 2030ecc2d604Sbonwick } 2031fa9e4066Sahrens 203288ecc943SGeorge Wilson if (vd->vdev_removing) 203388ecc943SGeorge Wilson vdev_remove(vd, txg); 203488ecc943SGeorge Wilson 2035ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2036fa9e4066Sahrens metaslab_sync(msp, txg); 2037ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2038ecc2d604Sbonwick } 2039fa9e4066Sahrens 2040fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2041fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2042fa9e4066Sahrens 2043fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2044fa9e4066Sahrens } 2045fa9e4066Sahrens 2046fa9e4066Sahrens uint64_t 2047fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2048fa9e4066Sahrens { 2049fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2050fa9e4066Sahrens } 2051fa9e4066Sahrens 20523d7072f8Seschrock /* 20533d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 20543d7072f8Seschrock * not be opened, and no I/O is attempted. 20553d7072f8Seschrock */ 2056fa9e4066Sahrens int 2057069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2058fa9e4066Sahrens { 2059c5904d13Seschrock vdev_t *vd; 2060fa9e4066Sahrens 20618f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2062fa9e4066Sahrens 2063c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2064e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2065e14bb325SJeff Bonwick 20663d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2067e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2068fa9e4066Sahrens 2069069f55e2SEric Schrock /* 2070069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2071069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2072069f55e2SEric Schrock * were faulted. 2073069f55e2SEric Schrock */ 2074069f55e2SEric Schrock vd->vdev_label_aux = aux; 2075069f55e2SEric Schrock 20763d7072f8Seschrock /* 20773d7072f8Seschrock * Faulted state takes precedence over degraded. 20783d7072f8Seschrock */ 2079*98d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 20803d7072f8Seschrock vd->vdev_faulted = 1ULL; 20813d7072f8Seschrock vd->vdev_degraded = 0ULL; 2082069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 20833d7072f8Seschrock 20843d7072f8Seschrock /* 2085c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2086c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 20873d7072f8Seschrock */ 2088c79790bcSGeorge Wilson if (!vd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 20893d7072f8Seschrock vd->vdev_degraded = 1ULL; 20903d7072f8Seschrock vd->vdev_faulted = 0ULL; 20913d7072f8Seschrock 20923d7072f8Seschrock /* 20933d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 20943d7072f8Seschrock * mark it degraded. 20953d7072f8Seschrock */ 20963d7072f8Seschrock vdev_reopen(vd); 20973d7072f8Seschrock 2098069f55e2SEric Schrock if (vdev_readable(vd)) 2099069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 21003d7072f8Seschrock } 21013d7072f8Seschrock 2102e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 21033d7072f8Seschrock } 21043d7072f8Seschrock 21053d7072f8Seschrock /* 21063d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 21073d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 21083d7072f8Seschrock * as I/O is concerned. 21093d7072f8Seschrock */ 21103d7072f8Seschrock int 2111069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 21123d7072f8Seschrock { 2113c5904d13Seschrock vdev_t *vd; 21140a4e9518Sgw 21158f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 21163d7072f8Seschrock 2117c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2118e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2119e14bb325SJeff Bonwick 21200e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2121e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 21220e34b6a7Sbonwick 21233d7072f8Seschrock /* 21243d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 21253d7072f8Seschrock */ 2126e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2127e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 21283d7072f8Seschrock 21293d7072f8Seschrock vd->vdev_degraded = 1ULL; 21303d7072f8Seschrock if (!vdev_is_dead(vd)) 21313d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2132069f55e2SEric Schrock aux); 21333d7072f8Seschrock 2134e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 21353d7072f8Seschrock } 21363d7072f8Seschrock 21373d7072f8Seschrock /* 21383d7072f8Seschrock * Online the given vdev. If 'unspare' is set, it implies two things. First, 21393d7072f8Seschrock * any attached spare device should be detached when the device finishes 21403d7072f8Seschrock * resilvering. Second, the online should be treated like a 'test' online case, 21413d7072f8Seschrock * so no FMA events are generated if the device fails to open. 21423d7072f8Seschrock */ 21433d7072f8Seschrock int 2144e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 21453d7072f8Seschrock { 2146573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 21473d7072f8Seschrock 21488f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 21493d7072f8Seschrock 2150c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2151e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 21523d7072f8Seschrock 21533d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2154e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2155fa9e4066Sahrens 2156573ca77eSGeorge Wilson tvd = vd->vdev_top; 2157fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2158441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2159e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2160e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2161573ca77eSGeorge Wilson 2162573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2163573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2164573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2165573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2166573ca77eSGeorge Wilson } 2167573ca77eSGeorge Wilson 2168573ca77eSGeorge Wilson vdev_reopen(tvd); 21693d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 21703d7072f8Seschrock 2171573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2172573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2173573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2174573ca77eSGeorge Wilson } 2175573ca77eSGeorge Wilson 21763d7072f8Seschrock if (newstate) 21773d7072f8Seschrock *newstate = vd->vdev_state; 21783d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 21793d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 21803d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 21813d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 21823d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2183fa9e4066Sahrens 2184573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2185573ca77eSGeorge Wilson 2186573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2187573ca77eSGeorge Wilson if (vd->vdev_aux) 2188573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2189573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2190573ca77eSGeorge Wilson } 21918ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2192fa9e4066Sahrens } 2193fa9e4066Sahrens 2194a1521560SJeff Bonwick static int 2195a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2196fa9e4066Sahrens { 2197e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 21988f18d1faSGeorge Wilson int error = 0; 21998f18d1faSGeorge Wilson uint64_t generation; 22008f18d1faSGeorge Wilson metaslab_group_t *mg; 22010a4e9518Sgw 22028f18d1faSGeorge Wilson top: 22038f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2204fa9e4066Sahrens 2205c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2206e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2207fa9e4066Sahrens 22080e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2209e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 22100e34b6a7Sbonwick 2211e6ca193dSGeorge Wilson tvd = vd->vdev_top; 22128f18d1faSGeorge Wilson mg = tvd->vdev_mg; 22138f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2214e6ca193dSGeorge Wilson 2215fa9e4066Sahrens /* 2216ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2217fa9e4066Sahrens */ 2218ecc2d604Sbonwick if (!vd->vdev_offline) { 2219ecc2d604Sbonwick /* 22208ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2221e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2222e6ca193dSGeorge Wilson * expendable. 2223ecc2d604Sbonwick */ 2224e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2225e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2226e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2227fa9e4066Sahrens 22288f18d1faSGeorge Wilson /* 2229b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2230b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2231b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2232b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 22338f18d1faSGeorge Wilson */ 22348f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 22358f18d1faSGeorge Wilson /* 22368f18d1faSGeorge Wilson * Prevent any future allocations. 22378f18d1faSGeorge Wilson */ 2238a1521560SJeff Bonwick metaslab_group_passivate(mg); 22398f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 22408f18d1faSGeorge Wilson 22411195e687SMark J Musante error = spa_offline_log(spa); 22428f18d1faSGeorge Wilson 22438f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 22448f18d1faSGeorge Wilson 22458f18d1faSGeorge Wilson /* 22468f18d1faSGeorge Wilson * Check to see if the config has changed. 22478f18d1faSGeorge Wilson */ 22488f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2249a1521560SJeff Bonwick metaslab_group_activate(mg); 22508f18d1faSGeorge Wilson if (error) 22518f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 22528f18d1faSGeorge Wilson vd, error)); 22538f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 22548f18d1faSGeorge Wilson goto top; 22558f18d1faSGeorge Wilson } 22568f18d1faSGeorge Wilson ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); 22578f18d1faSGeorge Wilson } 22588f18d1faSGeorge Wilson 2259ecc2d604Sbonwick /* 2260ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2261e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2262e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2263e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2264ecc2d604Sbonwick */ 2265ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2266e6ca193dSGeorge Wilson vdev_reopen(tvd); 2267e6ca193dSGeorge Wilson 2268e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2269e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2270ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2271e6ca193dSGeorge Wilson vdev_reopen(tvd); 2272e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2273ecc2d604Sbonwick } 22748f18d1faSGeorge Wilson 22758f18d1faSGeorge Wilson /* 22768f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 22778f18d1faSGeorge Wilson * once we online the device it's open for business. 22788f18d1faSGeorge Wilson */ 22798f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2280a1521560SJeff Bonwick metaslab_group_activate(mg); 2281fa9e4066Sahrens } 2282fa9e4066Sahrens 2283e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2284ecc2d604Sbonwick 22858f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2286fa9e4066Sahrens } 2287fa9e4066Sahrens 2288a1521560SJeff Bonwick int 2289a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2290a1521560SJeff Bonwick { 2291a1521560SJeff Bonwick int error; 2292a1521560SJeff Bonwick 2293a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2294a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2295a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2296a1521560SJeff Bonwick 2297a1521560SJeff Bonwick return (error); 2298a1521560SJeff Bonwick } 2299a1521560SJeff Bonwick 2300ea8dc4b6Seschrock /* 2301ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2302ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2303ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2304ea8dc4b6Seschrock */ 2305ea8dc4b6Seschrock void 2306e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2307fa9e4066Sahrens { 2308e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2309e14bb325SJeff Bonwick 2310e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2311fa9e4066Sahrens 2312ea8dc4b6Seschrock if (vd == NULL) 2313e14bb325SJeff Bonwick vd = rvd; 2314fa9e4066Sahrens 2315ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2316ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2317ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2318fa9e4066Sahrens 2319e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2320e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 23213d7072f8Seschrock 23223d7072f8Seschrock /* 23238a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 23248a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 23258a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 23268a79c1b5Sek * written out to disk. 23273d7072f8Seschrock */ 2328e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2329e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 23308a79c1b5Sek 2331096d22d4SEric Schrock /* 2332096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2333096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2334096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2335096d22d4SEric Schrock */ 2336096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2337096d22d4SEric Schrock 23383d7072f8Seschrock vd->vdev_faulted = vd->vdev_degraded = 0; 2339e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2340e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2341e14bb325SJeff Bonwick 23423d7072f8Seschrock vdev_reopen(vd); 23433d7072f8Seschrock 2344096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2345096d22d4SEric Schrock 2346e14bb325SJeff Bonwick if (vd != rvd) 2347e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2348e14bb325SJeff Bonwick 2349e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2350bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 23513d7072f8Seschrock 23523d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 23533d7072f8Seschrock } 2354096d22d4SEric Schrock 2355096d22d4SEric Schrock /* 2356096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2357096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2358096d22d4SEric Schrock * done in response to the FMA fault. 2359096d22d4SEric Schrock */ 2360096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2361096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2362096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2363096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2364fa9e4066Sahrens } 2365fa9e4066Sahrens 2366e14bb325SJeff Bonwick boolean_t 2367e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 23680a4e9518Sgw { 236988ecc943SGeorge Wilson /* 237088ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 237188ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 237288ecc943SGeorge Wilson * these types of devices in the various code paths. 237388ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 237488ecc943SGeorge Wilson * before issuing I/O to them. 237588ecc943SGeorge Wilson */ 237688ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 237788ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 23780a4e9518Sgw } 23790a4e9518Sgw 2380e14bb325SJeff Bonwick boolean_t 2381e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 23820a4e9518Sgw { 2383e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 23840a4e9518Sgw } 23850a4e9518Sgw 2386e14bb325SJeff Bonwick boolean_t 2387e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2388fa9e4066Sahrens { 2389e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2390fa9e4066Sahrens } 2391fa9e4066Sahrens 2392a31e6787SGeorge Wilson boolean_t 2393a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2394a31e6787SGeorge Wilson { 23958ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 23968ad4d6ddSJeff Bonwick 2397a31e6787SGeorge Wilson /* 23988ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2399a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2400a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 24018ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 24028ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 24038ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2404a31e6787SGeorge Wilson */ 24058ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 240688ecc943SGeorge Wilson !vd->vdev_cant_write && !vd->vdev_ishole && !vd->vdev_removing); 2407a31e6787SGeorge Wilson } 2408a31e6787SGeorge Wilson 2409e14bb325SJeff Bonwick boolean_t 2410e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2411fa9e4066Sahrens { 2412e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2413fa9e4066Sahrens 2414e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2415e14bb325SJeff Bonwick return (B_FALSE); 2416fa9e4066Sahrens 2417e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2418e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2419fa9e4066Sahrens 2420e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2421e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2422fa9e4066Sahrens 2423e14bb325SJeff Bonwick return (B_TRUE); 2424fa9e4066Sahrens } 2425fa9e4066Sahrens 2426fa9e4066Sahrens /* 2427fa9e4066Sahrens * Get statistics for the given vdev. 2428fa9e4066Sahrens */ 2429fa9e4066Sahrens void 2430fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2431fa9e4066Sahrens { 2432fa9e4066Sahrens vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2433fa9e4066Sahrens 2434fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2435fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2436088f3894Sahrens vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; 2437fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2438fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2439573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2440573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2441573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2442fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2443fa9e4066Sahrens 2444fa9e4066Sahrens /* 2445fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2446fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2447fa9e4066Sahrens */ 2448fa9e4066Sahrens if (vd == rvd) { 2449e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2450fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2451fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2452fa9e4066Sahrens 2453fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2454e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2455fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2456fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2457fa9e4066Sahrens } 2458fa9e4066Sahrens vs->vs_scrub_examined += cvs->vs_scrub_examined; 2459fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2460fa9e4066Sahrens } 2461fa9e4066Sahrens } 2462fa9e4066Sahrens } 2463fa9e4066Sahrens 2464fa94a07fSbrendan void 2465fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2466fa94a07fSbrendan { 2467fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2468fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2469fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2470fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2471fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2472fa94a07fSbrendan } 2473fa94a07fSbrendan 2474fa9e4066Sahrens void 2475e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2476fa9e4066Sahrens { 24778ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 24788ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2479e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2480fa9e4066Sahrens vdev_t *pvd; 2481fa9e4066Sahrens uint64_t txg = zio->io_txg; 2482fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2483fa9e4066Sahrens zio_type_t type = zio->io_type; 2484fa9e4066Sahrens int flags = zio->io_flags; 2485fa9e4066Sahrens 2486e14bb325SJeff Bonwick /* 2487e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2488e14bb325SJeff Bonwick */ 2489e14bb325SJeff Bonwick if (zio->io_gang_tree) 2490e14bb325SJeff Bonwick return; 2491e14bb325SJeff Bonwick 2492fa9e4066Sahrens if (zio->io_error == 0) { 2493e14bb325SJeff Bonwick /* 2494e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2495e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2496e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2497e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2498e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2499e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2500e14bb325SJeff Bonwick * remain zero, which is what we want.) 2501e14bb325SJeff Bonwick * 2502e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2503e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2504e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2505e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2506e14bb325SJeff Bonwick */ 2507e14bb325SJeff Bonwick if (vd == rvd) 2508e14bb325SJeff Bonwick return; 2509e14bb325SJeff Bonwick 2510e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 25118ad4d6ddSJeff Bonwick 25128ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 25138ad4d6ddSJeff Bonwick return; 25148ad4d6ddSJeff Bonwick 25158ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 25168ad4d6ddSJeff Bonwick 2517e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 2518d80c45e0Sbonwick if (flags & ZIO_FLAG_SCRUB_THREAD) 2519e14bb325SJeff Bonwick vs->vs_scrub_repaired += psize; 25208ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2521e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2522fa9e4066Sahrens } 25238ad4d6ddSJeff Bonwick 25248ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 25258ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 25268ad4d6ddSJeff Bonwick 25278ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2528fa9e4066Sahrens return; 2529fa9e4066Sahrens } 2530fa9e4066Sahrens 2531fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2532fa9e4066Sahrens return; 2533fa9e4066Sahrens 25348956713aSEric Schrock /* 25358956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 25368956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 25378956713aSEric Schrock * hard errors, when in reality they can happen for any number of 25388956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 25398956713aSEric Schrock */ 25408956713aSEric Schrock if (zio->io_error == EIO && 25418956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 25428956713aSEric Schrock return; 25438956713aSEric Schrock 25448f18d1faSGeorge Wilson /* 25458f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 25468f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 25478f18d1faSGeorge Wilson * errors. 25488f18d1faSGeorge Wilson */ 25498f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 25508f18d1faSGeorge Wilson return; 25518f18d1faSGeorge Wilson 2552e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2553b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2554e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2555e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2556e14bb325SJeff Bonwick else 2557e14bb325SJeff Bonwick vs->vs_read_errors++; 2558fa9e4066Sahrens } 2559b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2560e14bb325SJeff Bonwick vs->vs_write_errors++; 2561e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2562fa9e4066Sahrens 25638ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 25648ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 2565b24ab676SJeff Bonwick (flags & ZIO_FLAG_SCRUB_THREAD) || 2566b24ab676SJeff Bonwick spa->spa_claiming)) { 25678ad4d6ddSJeff Bonwick /* 2568b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2569b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2570b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2571b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2572b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2573b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2574b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2575b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 25768ad4d6ddSJeff Bonwick * 25778ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 25788ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 25798ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 25808ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 25818ad4d6ddSJeff Bonwick */ 25828ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 25838ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 25848ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SCRUB_THREAD) { 25858ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 25868ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 25878ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2588b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2589b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2590b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2591b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 25928ad4d6ddSJeff Bonwick } 2593b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 25948ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2595fa9e4066Sahrens return; 25968ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 25978ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 25988ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2599fa9e4066Sahrens } 26008ad4d6ddSJeff Bonwick if (vd != rvd) 26018ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2602fa9e4066Sahrens } 2603fa9e4066Sahrens } 2604fa9e4066Sahrens 2605fa9e4066Sahrens void 2606fa9e4066Sahrens vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) 2607fa9e4066Sahrens { 2608fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2609fa9e4066Sahrens 2610573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2611fa9e4066Sahrens vdev_scrub_stat_update(vd->vdev_child[c], type, complete); 2612fa9e4066Sahrens 2613fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2614fa9e4066Sahrens 2615fa9e4066Sahrens if (type == POOL_SCRUB_NONE) { 2616fa9e4066Sahrens /* 2617fa9e4066Sahrens * Update completion and end time. Leave everything else alone 2618fa9e4066Sahrens * so we can report what happened during the previous scrub. 2619fa9e4066Sahrens */ 2620fa9e4066Sahrens vs->vs_scrub_complete = complete; 2621fa9e4066Sahrens vs->vs_scrub_end = gethrestime_sec(); 2622fa9e4066Sahrens } else { 2623fa9e4066Sahrens vs->vs_scrub_type = type; 2624fa9e4066Sahrens vs->vs_scrub_complete = 0; 2625fa9e4066Sahrens vs->vs_scrub_examined = 0; 2626fa9e4066Sahrens vs->vs_scrub_repaired = 0; 2627fa9e4066Sahrens vs->vs_scrub_start = gethrestime_sec(); 2628fa9e4066Sahrens vs->vs_scrub_end = 0; 2629fa9e4066Sahrens } 2630fa9e4066Sahrens 2631fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2632fa9e4066Sahrens } 2633fa9e4066Sahrens 2634fa9e4066Sahrens /* 2635b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2636b24ab676SJeff Bonwick * and the root vdev. 2637fa9e4066Sahrens */ 2638fa9e4066Sahrens void 2639b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2640b24ab676SJeff Bonwick int64_t space_delta) 2641fa9e4066Sahrens { 264299653d4eSeschrock int64_t dspace_delta = space_delta; 26438654d025Sperrin spa_t *spa = vd->vdev_spa; 26448654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2645b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2646b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2647fa9e4066Sahrens 26488654d025Sperrin ASSERT(vd == vd->vdev_top); 264999653d4eSeschrock 26508654d025Sperrin /* 26518654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 26528654d025Sperrin * factor. We must calculate this here and not at the root vdev 26538654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 26548654d025Sperrin * childrens', thus not accurate enough for us. 26558654d025Sperrin */ 26568654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2657e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 26588654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 26598654d025Sperrin vd->vdev_deflate_ratio; 26608654d025Sperrin 26618654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 26628654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2663b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 26648654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 26658654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 26668654d025Sperrin 2667b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2668fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2669fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2670b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2671fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2672fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2673fa94a07fSbrendan } 2674b24ab676SJeff Bonwick 2675b24ab676SJeff Bonwick if (mc != NULL) { 2676b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2677b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2678b24ab676SJeff Bonwick 2679b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2680b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2681b24ab676SJeff Bonwick } 2682fa9e4066Sahrens } 2683fa9e4066Sahrens 2684fa9e4066Sahrens /* 2685fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2686fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2687fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2688fa9e4066Sahrens */ 2689fa9e4066Sahrens void 2690fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2691fa9e4066Sahrens { 2692fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2693fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2694fa9e4066Sahrens int c; 2695fa9e4066Sahrens 2696c5904d13Seschrock /* 26976809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 26986809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2699c5904d13Seschrock */ 2700c5904d13Seschrock if (vd->vdev_aux != NULL) { 2701c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2702c5904d13Seschrock nvlist_t **aux; 2703c5904d13Seschrock uint_t naux; 2704c5904d13Seschrock 2705c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2706c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2707c5904d13Seschrock break; 2708c5904d13Seschrock } 2709c5904d13Seschrock 2710e14bb325SJeff Bonwick if (c == sav->sav_count) { 2711e14bb325SJeff Bonwick /* 2712e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2713e14bb325SJeff Bonwick */ 2714e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2715e14bb325SJeff Bonwick return; 2716e14bb325SJeff Bonwick } 2717e14bb325SJeff Bonwick 2718c5904d13Seschrock sav->sav_sync = B_TRUE; 2719c5904d13Seschrock 27206809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 27216809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 27226809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 27236809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 27246809eb4eSEric Schrock } 2725c5904d13Seschrock 2726c5904d13Seschrock ASSERT(c < naux); 2727c5904d13Seschrock 2728c5904d13Seschrock /* 2729c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2730c5904d13Seschrock * sketchy, but it will work. 2731c5904d13Seschrock */ 2732c5904d13Seschrock nvlist_free(aux[c]); 2733c5904d13Seschrock aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); 2734c5904d13Seschrock 2735c5904d13Seschrock return; 2736c5904d13Seschrock } 2737c5904d13Seschrock 27385dabedeeSbonwick /* 2739e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2740e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2741e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 27425dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 27435dabedeeSbonwick */ 2744e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2745e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2746e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 27475dabedeeSbonwick 2748fa9e4066Sahrens if (vd == rvd) { 2749fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2750fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2751fa9e4066Sahrens } else { 2752fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2753fa9e4066Sahrens 275488ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 275588ecc943SGeorge Wilson !vd->vdev_ishole) 2756e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2757fa9e4066Sahrens } 2758fa9e4066Sahrens } 2759fa9e4066Sahrens 2760fa9e4066Sahrens void 2761fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2762fa9e4066Sahrens { 27635dabedeeSbonwick spa_t *spa = vd->vdev_spa; 27645dabedeeSbonwick 2765e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2766e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2767e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 27685dabedeeSbonwick 2769e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2770e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2771e14bb325SJeff Bonwick } 2772e14bb325SJeff Bonwick 2773e14bb325SJeff Bonwick /* 2774e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2775e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2776e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2777e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2778e14bb325SJeff Bonwick */ 2779e14bb325SJeff Bonwick void 2780e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2781e14bb325SJeff Bonwick { 2782e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2783e14bb325SJeff Bonwick 2784e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2785e14bb325SJeff Bonwick 2786e14bb325SJeff Bonwick /* 2787e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 2788e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 2789e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 2790e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 2791e14bb325SJeff Bonwick */ 2792e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2793e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2794e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2795e14bb325SJeff Bonwick 2796b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2797e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 2798e14bb325SJeff Bonwick } 2799e14bb325SJeff Bonwick 2800e14bb325SJeff Bonwick void 2801e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 2802e14bb325SJeff Bonwick { 2803e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2804e14bb325SJeff Bonwick 2805e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2806e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2807e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2808e14bb325SJeff Bonwick 2809e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2810e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 2811fa9e4066Sahrens } 2812fa9e4066Sahrens 281332b87932Sek /* 281432b87932Sek * Propagate vdev state up from children to parent. 281532b87932Sek */ 281644cd46caSbillm void 281744cd46caSbillm vdev_propagate_state(vdev_t *vd) 281844cd46caSbillm { 28198ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 28208ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 282144cd46caSbillm int degraded = 0, faulted = 0; 282244cd46caSbillm int corrupted = 0; 282344cd46caSbillm vdev_t *child; 282444cd46caSbillm 28253d7072f8Seschrock if (vd->vdev_children > 0) { 2826573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 28273d7072f8Seschrock child = vd->vdev_child[c]; 282851ece835Seschrock 282988ecc943SGeorge Wilson /* 283088ecc943SGeorge Wilson * Don't factor holes into the decision. 283188ecc943SGeorge Wilson */ 283288ecc943SGeorge Wilson if (child->vdev_ishole) 283388ecc943SGeorge Wilson continue; 283488ecc943SGeorge Wilson 2835e14bb325SJeff Bonwick if (!vdev_readable(child) || 28368ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 283751ece835Seschrock /* 283851ece835Seschrock * Root special: if there is a top-level log 283951ece835Seschrock * device, treat the root vdev as if it were 284051ece835Seschrock * degraded. 284151ece835Seschrock */ 284251ece835Seschrock if (child->vdev_islog && vd == rvd) 284351ece835Seschrock degraded++; 284451ece835Seschrock else 284551ece835Seschrock faulted++; 284651ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 28473d7072f8Seschrock degraded++; 284851ece835Seschrock } 284944cd46caSbillm 28503d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 28513d7072f8Seschrock corrupted++; 28523d7072f8Seschrock } 285344cd46caSbillm 28543d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 28553d7072f8Seschrock 28563d7072f8Seschrock /* 2857e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 28583d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 28593d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 28603d7072f8Seschrock * replicas'. 28613d7072f8Seschrock */ 28623d7072f8Seschrock if (corrupted && vd == rvd && 28633d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 28643d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 28653d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 28663d7072f8Seschrock } 28673d7072f8Seschrock 286851ece835Seschrock if (vd->vdev_parent) 28693d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 287044cd46caSbillm } 287144cd46caSbillm 2872fa9e4066Sahrens /* 2873ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 2874ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 2875ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 2876ea8dc4b6Seschrock * 2877ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 2878ea8dc4b6Seschrock * generated. 2879fa9e4066Sahrens */ 2880fa9e4066Sahrens void 2881ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2882fa9e4066Sahrens { 2883560e6e96Seschrock uint64_t save_state; 2884c5904d13Seschrock spa_t *spa = vd->vdev_spa; 2885ea8dc4b6Seschrock 2886ea8dc4b6Seschrock if (state == vd->vdev_state) { 2887ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 2888fa9e4066Sahrens return; 2889ea8dc4b6Seschrock } 2890ea8dc4b6Seschrock 2891560e6e96Seschrock save_state = vd->vdev_state; 2892fa9e4066Sahrens 2893fa9e4066Sahrens vd->vdev_state = state; 2894fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 2895fa9e4066Sahrens 28963d7072f8Seschrock /* 28973d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 2898*98d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 2899*98d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 2900*98d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 2901*98d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 2902*98d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 2903*98d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 2904*98d1cbfeSGeorge Wilson * children. 2905*98d1cbfeSGeorge Wilson */ 2906*98d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 2907*98d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 29083d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 29093d7072f8Seschrock 2910069f55e2SEric Schrock /* 2911069f55e2SEric Schrock * If we have brought this vdev back into service, we need 2912069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 2913069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 2914069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 2915069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 2916069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 2917069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 2918069f55e2SEric Schrock */ 2919069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2920069f55e2SEric Schrock vd->vdev_prevstate != state) 2921069f55e2SEric Schrock zfs_post_state_change(spa, vd); 2922069f55e2SEric Schrock 29233d7072f8Seschrock if (vd->vdev_removed && 29243d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 29253d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 29263d7072f8Seschrock /* 29273d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 29283d7072f8Seschrock * device was previously marked removed and someone attempted to 29293d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 29303d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 29313d7072f8Seschrock * it is one of our special test online cases, which is only 29323d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 29333d7072f8Seschrock * fault. 29343d7072f8Seschrock */ 29353d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 29363d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 29373d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 29383d7072f8Seschrock vd->vdev_removed = B_TRUE; 29393d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 2940ea8dc4b6Seschrock /* 2941ea8dc4b6Seschrock * If we fail to open a vdev during an import, we mark it as 2942ea8dc4b6Seschrock * "not available", which signifies that it was never there to 2943ea8dc4b6Seschrock * begin with. Failure to open such a device is not considered 2944ea8dc4b6Seschrock * an error. 2945ea8dc4b6Seschrock */ 2946b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_IMPORT && 2947560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 2948560e6e96Seschrock vd->vdev_not_present = 1; 2949560e6e96Seschrock 2950560e6e96Seschrock /* 2951560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 2952560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 2953560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 2954560e6e96Seschrock * want to post the ereport if the device was already in the 2955560e6e96Seschrock * CANT_OPEN state beforehand. 29563d7072f8Seschrock * 29573d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 29583d7072f8Seschrock * online the device in response to an insertion event. If we 29593d7072f8Seschrock * hit this case, then we have detected an insertion event for a 29603d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 29613d7072f8Seschrock * In this scenario, we don't post an ereport because we are 29623d7072f8Seschrock * about to replace the device, or attempt an online with 29633d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 2964560e6e96Seschrock */ 29653d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 29663d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 2967c5904d13Seschrock vd != spa->spa_root_vdev) { 2968ea8dc4b6Seschrock const char *class; 2969ea8dc4b6Seschrock 2970ea8dc4b6Seschrock switch (aux) { 2971ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 2972ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2973ea8dc4b6Seschrock break; 2974ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 2975ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2976ea8dc4b6Seschrock break; 2977ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 2978ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2979ea8dc4b6Seschrock break; 2980ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 2981ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2982ea8dc4b6Seschrock break; 2983ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 2984ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 2985ea8dc4b6Seschrock break; 2986ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 2987ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 2988ea8dc4b6Seschrock break; 2989ea8dc4b6Seschrock default: 2990ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 2991ea8dc4b6Seschrock } 2992ea8dc4b6Seschrock 2993c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 2994ea8dc4b6Seschrock } 2995ea8dc4b6Seschrock 29963d7072f8Seschrock /* Erase any notion of persistent removed state */ 29973d7072f8Seschrock vd->vdev_removed = B_FALSE; 29983d7072f8Seschrock } else { 29993d7072f8Seschrock vd->vdev_removed = B_FALSE; 30003d7072f8Seschrock } 3001ea8dc4b6Seschrock 30028b33d774STim Haley if (!isopen && vd->vdev_parent) 30038b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3004fa9e4066Sahrens } 300515e6edf1Sgw 300615e6edf1Sgw /* 300715e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 300815e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 300915e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 301015e6edf1Sgw * can be wholedisks. 301115e6edf1Sgw */ 301215e6edf1Sgw boolean_t 301315e6edf1Sgw vdev_is_bootable(vdev_t *vd) 301415e6edf1Sgw { 301515e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 301615e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 301715e6edf1Sgw 301815e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 301915e6edf1Sgw vd->vdev_children > 1) { 302015e6edf1Sgw return (B_FALSE); 302115e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 302215e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 302315e6edf1Sgw return (B_FALSE); 302415e6edf1Sgw } 302515e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 302615e6edf1Sgw return (B_FALSE); 302715e6edf1Sgw } 302815e6edf1Sgw 3029573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 303015e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 303115e6edf1Sgw return (B_FALSE); 303215e6edf1Sgw } 303315e6edf1Sgw return (B_TRUE); 303415e6edf1Sgw } 3035e6ca193dSGeorge Wilson 303688ecc943SGeorge Wilson /* 303788ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 303888ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 303988ecc943SGeorge Wilson * vdev was offline then we transfer that state to the device 304088ecc943SGeorge Wilson * in the current vdev tree (nvd). 304188ecc943SGeorge Wilson */ 3042e6ca193dSGeorge Wilson void 304388ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3044e6ca193dSGeorge Wilson { 304588ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3046e6ca193dSGeorge Wilson 304788ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 304888ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3049e6ca193dSGeorge Wilson 305088ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 305188ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3052e6ca193dSGeorge Wilson 305388ecc943SGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf && ovd->vdev_offline) { 3054e6ca193dSGeorge Wilson /* 3055e6ca193dSGeorge Wilson * It would be nice to call vdev_offline() 3056e6ca193dSGeorge Wilson * directly but the pool isn't fully loaded and 3057e6ca193dSGeorge Wilson * the txg threads have not been started yet. 3058e6ca193dSGeorge Wilson */ 305988ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 306088ecc943SGeorge Wilson vdev_reopen(nvd->vdev_top); 3061e6ca193dSGeorge Wilson } 3062e6ca193dSGeorge Wilson } 3063573ca77eSGeorge Wilson 3064573ca77eSGeorge Wilson /* 3065573ca77eSGeorge Wilson * Expand a vdev if possible. 3066573ca77eSGeorge Wilson */ 3067573ca77eSGeorge Wilson void 3068573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3069573ca77eSGeorge Wilson { 3070573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3071573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3072573ca77eSGeorge Wilson 3073573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3074573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3075573ca77eSGeorge Wilson vdev_config_dirty(vd); 3076573ca77eSGeorge Wilson } 3077573ca77eSGeorge Wilson } 30781195e687SMark J Musante 30791195e687SMark J Musante /* 30801195e687SMark J Musante * Split a vdev. 30811195e687SMark J Musante */ 30821195e687SMark J Musante void 30831195e687SMark J Musante vdev_split(vdev_t *vd) 30841195e687SMark J Musante { 30851195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 30861195e687SMark J Musante 30871195e687SMark J Musante vdev_remove_child(pvd, vd); 30881195e687SMark J Musante vdev_compact_children(pvd); 30891195e687SMark J Musante 30901195e687SMark J Musante cvd = pvd->vdev_child[0]; 30911195e687SMark J Musante if (pvd->vdev_children == 1) { 30921195e687SMark J Musante vdev_remove_parent(cvd); 30931195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 30941195e687SMark J Musante } 30951195e687SMark J Musante vdev_propagate_state(cvd); 30961195e687SMark J Musante } 3097