1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24*81cd5c55SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 251a902ef8SHans Rosenfeld * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/zfs_context.h> 29ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 30fa9e4066Sahrens #include <sys/spa.h> 31fa9e4066Sahrens #include <sys/spa_impl.h> 32fa9e4066Sahrens #include <sys/dmu.h> 33fa9e4066Sahrens #include <sys/dmu_tx.h> 34fa9e4066Sahrens #include <sys/vdev_impl.h> 35fa9e4066Sahrens #include <sys/uberblock_impl.h> 36fa9e4066Sahrens #include <sys/metaslab.h> 37fa9e4066Sahrens #include <sys/metaslab_impl.h> 38fa9e4066Sahrens #include <sys/space_map.h> 390713e232SGeorge Wilson #include <sys/space_reftree.h> 40fa9e4066Sahrens #include <sys/zio.h> 41fa9e4066Sahrens #include <sys/zap.h> 42fa9e4066Sahrens #include <sys/fs/zfs.h> 43c5904d13Seschrock #include <sys/arc.h> 44e6ca193dSGeorge Wilson #include <sys/zil.h> 453f9d6ad7SLin Ling #include <sys/dsl_scan.h> 46fa9e4066Sahrens 47fa9e4066Sahrens /* 48fa9e4066Sahrens * Virtual device management. 49fa9e4066Sahrens */ 50fa9e4066Sahrens 51fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 52fa9e4066Sahrens &vdev_root_ops, 53fa9e4066Sahrens &vdev_raidz_ops, 54fa9e4066Sahrens &vdev_mirror_ops, 55fa9e4066Sahrens &vdev_replacing_ops, 5699653d4eSeschrock &vdev_spare_ops, 57fa9e4066Sahrens &vdev_disk_ops, 58fa9e4066Sahrens &vdev_file_ops, 59fa9e4066Sahrens &vdev_missing_ops, 6088ecc943SGeorge Wilson &vdev_hole_ops, 61fa9e4066Sahrens NULL 62fa9e4066Sahrens }; 63fa9e4066Sahrens 64088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 65088f3894Sahrens int zfs_scrub_limit = 10; 6605b2b3b8Smishra 67bf3e216cSMatthew Ahrens /* 68bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 69bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 70bf3e216cSMatthew Ahrens */ 71bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 72bf3e216cSMatthew Ahrens 73fa9e4066Sahrens /* 74fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 75fa9e4066Sahrens */ 76fa9e4066Sahrens static vdev_ops_t * 77fa9e4066Sahrens vdev_getops(const char *type) 78fa9e4066Sahrens { 79fa9e4066Sahrens vdev_ops_t *ops, **opspp; 80fa9e4066Sahrens 81fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 82fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 83fa9e4066Sahrens break; 84fa9e4066Sahrens 85fa9e4066Sahrens return (ops); 86fa9e4066Sahrens } 87fa9e4066Sahrens 88fa9e4066Sahrens /* 89fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 90fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 91fa9e4066Sahrens */ 92fa9e4066Sahrens uint64_t 93fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 94fa9e4066Sahrens { 95ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 96fa9e4066Sahrens uint64_t csize; 97fa9e4066Sahrens 98573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 99fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 100fa9e4066Sahrens asize = MAX(asize, csize); 101fa9e4066Sahrens } 102fa9e4066Sahrens 103fa9e4066Sahrens return (asize); 104fa9e4066Sahrens } 105fa9e4066Sahrens 1062a79c5feSlling /* 107573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 108573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 109573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 110573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1112a79c5feSlling */ 1122a79c5feSlling uint64_t 113573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1142a79c5feSlling { 115573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1162a79c5feSlling 117573ca77eSGeorge Wilson /* 1184263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 119573ca77eSGeorge Wilson * just return our own asize. 120573ca77eSGeorge Wilson */ 121573ca77eSGeorge Wilson if (pvd == NULL) 122573ca77eSGeorge Wilson return (vd->vdev_asize); 1232a79c5feSlling 1242a79c5feSlling /* 125573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 126573ca77eSGeorge Wilson * to the nearest metaslab. 1272a79c5feSlling */ 128573ca77eSGeorge Wilson if (vd == vd->vdev_top) 129573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1302a79c5feSlling 131573ca77eSGeorge Wilson /* 132573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 133573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 134573ca77eSGeorge Wilson */ 135573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 136573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1372a79c5feSlling 138573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 139573ca77eSGeorge Wilson } 1402a79c5feSlling 141573ca77eSGeorge Wilson void 142573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 143573ca77eSGeorge Wilson { 144573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 145573ca77eSGeorge Wilson 146573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 147573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1482a79c5feSlling } 1492a79c5feSlling 150fa9e4066Sahrens vdev_t * 151fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 152fa9e4066Sahrens { 153fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 154fa9e4066Sahrens 155e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 156e05725b1Sbonwick 157088f3894Sahrens if (vdev < rvd->vdev_children) { 158088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 159fa9e4066Sahrens return (rvd->vdev_child[vdev]); 160088f3894Sahrens } 161fa9e4066Sahrens 162fa9e4066Sahrens return (NULL); 163fa9e4066Sahrens } 164fa9e4066Sahrens 165fa9e4066Sahrens vdev_t * 166fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 167fa9e4066Sahrens { 168fa9e4066Sahrens vdev_t *mvd; 169fa9e4066Sahrens 1700e34b6a7Sbonwick if (vd->vdev_guid == guid) 171fa9e4066Sahrens return (vd); 172fa9e4066Sahrens 173573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 174fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 175fa9e4066Sahrens NULL) 176fa9e4066Sahrens return (mvd); 177fa9e4066Sahrens 178fa9e4066Sahrens return (NULL); 179fa9e4066Sahrens } 180fa9e4066Sahrens 18112380e1eSArne Jansen static int 18212380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 18312380e1eSArne Jansen { 18412380e1eSArne Jansen int n = 0; 18512380e1eSArne Jansen 18612380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 18712380e1eSArne Jansen return (1); 18812380e1eSArne Jansen 18912380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 19012380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 19112380e1eSArne Jansen 19212380e1eSArne Jansen return (n); 19312380e1eSArne Jansen } 19412380e1eSArne Jansen 19512380e1eSArne Jansen int 19612380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 19712380e1eSArne Jansen { 19812380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 19912380e1eSArne Jansen } 20012380e1eSArne Jansen 201fa9e4066Sahrens void 202fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 203fa9e4066Sahrens { 204fa9e4066Sahrens size_t oldsize, newsize; 205fa9e4066Sahrens uint64_t id = cvd->vdev_id; 206fa9e4066Sahrens vdev_t **newchild; 207*81cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 208fa9e4066Sahrens 209*81cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 210fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 211fa9e4066Sahrens 212fa9e4066Sahrens cvd->vdev_parent = pvd; 213fa9e4066Sahrens 214fa9e4066Sahrens if (pvd == NULL) 215fa9e4066Sahrens return; 216fa9e4066Sahrens 217fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 218fa9e4066Sahrens 219fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 220fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 221fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 222fa9e4066Sahrens 223fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 224fa9e4066Sahrens if (pvd->vdev_child != NULL) { 225fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 226fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 227fa9e4066Sahrens } 228fa9e4066Sahrens 229fa9e4066Sahrens pvd->vdev_child = newchild; 230fa9e4066Sahrens pvd->vdev_child[id] = cvd; 231fa9e4066Sahrens 232fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 233fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 234fa9e4066Sahrens 235fa9e4066Sahrens /* 236fa9e4066Sahrens * Walk up all ancestors to update guid sum. 237fa9e4066Sahrens */ 238fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 239fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 240fa9e4066Sahrens } 241fa9e4066Sahrens 242fa9e4066Sahrens void 243fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 244fa9e4066Sahrens { 245fa9e4066Sahrens int c; 246fa9e4066Sahrens uint_t id = cvd->vdev_id; 247fa9e4066Sahrens 248fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 249fa9e4066Sahrens 250fa9e4066Sahrens if (pvd == NULL) 251fa9e4066Sahrens return; 252fa9e4066Sahrens 253fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 254fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 255fa9e4066Sahrens 256fa9e4066Sahrens pvd->vdev_child[id] = NULL; 257fa9e4066Sahrens cvd->vdev_parent = NULL; 258fa9e4066Sahrens 259fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 260fa9e4066Sahrens if (pvd->vdev_child[c]) 261fa9e4066Sahrens break; 262fa9e4066Sahrens 263fa9e4066Sahrens if (c == pvd->vdev_children) { 264fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 265fa9e4066Sahrens pvd->vdev_child = NULL; 266fa9e4066Sahrens pvd->vdev_children = 0; 267fa9e4066Sahrens } 268fa9e4066Sahrens 269fa9e4066Sahrens /* 270fa9e4066Sahrens * Walk up all ancestors to update guid sum. 271fa9e4066Sahrens */ 272fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 273fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 274fa9e4066Sahrens } 275fa9e4066Sahrens 276fa9e4066Sahrens /* 277fa9e4066Sahrens * Remove any holes in the child array. 278fa9e4066Sahrens */ 279fa9e4066Sahrens void 280fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 281fa9e4066Sahrens { 282fa9e4066Sahrens vdev_t **newchild, *cvd; 283fa9e4066Sahrens int oldc = pvd->vdev_children; 284573ca77eSGeorge Wilson int newc; 285fa9e4066Sahrens 286e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 287fa9e4066Sahrens 288573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 289fa9e4066Sahrens if (pvd->vdev_child[c]) 290fa9e4066Sahrens newc++; 291fa9e4066Sahrens 292fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 293fa9e4066Sahrens 294573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 295fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 296fa9e4066Sahrens newchild[newc] = cvd; 297fa9e4066Sahrens cvd->vdev_id = newc++; 298fa9e4066Sahrens } 299fa9e4066Sahrens } 300fa9e4066Sahrens 301fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 302fa9e4066Sahrens pvd->vdev_child = newchild; 303fa9e4066Sahrens pvd->vdev_children = newc; 304fa9e4066Sahrens } 305fa9e4066Sahrens 306fa9e4066Sahrens /* 307fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 308fa9e4066Sahrens */ 30988ecc943SGeorge Wilson vdev_t * 310fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 311fa9e4066Sahrens { 312fa9e4066Sahrens vdev_t *vd; 313fa9e4066Sahrens 314fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 315fa9e4066Sahrens 3160e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 3170e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 3180e34b6a7Sbonwick spa->spa_root_vdev = vd; 319e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 3200e34b6a7Sbonwick } 3210e34b6a7Sbonwick 32288ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 3230e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 3240e34b6a7Sbonwick /* 3250e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3260e34b6a7Sbonwick * which must be unique among all pools. 3270e34b6a7Sbonwick */ 3281195e687SMark J Musante guid = spa_generate_guid(NULL); 3290e34b6a7Sbonwick } else { 3300e34b6a7Sbonwick /* 3310e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3320e34b6a7Sbonwick */ 3331195e687SMark J Musante guid = spa_generate_guid(spa); 3340e34b6a7Sbonwick } 3350e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3360e34b6a7Sbonwick } 3370e34b6a7Sbonwick 338fa9e4066Sahrens vd->vdev_spa = spa; 339fa9e4066Sahrens vd->vdev_id = id; 340fa9e4066Sahrens vd->vdev_guid = guid; 341fa9e4066Sahrens vd->vdev_guid_sum = guid; 342fa9e4066Sahrens vd->vdev_ops = ops; 343fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 34488ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 345fa9e4066Sahrens 346fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3475ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 348e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3498ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3500713e232SGeorge Wilson vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 3518ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3528ad4d6ddSJeff Bonwick } 353fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 354fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 355fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 356fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 357fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3583d7072f8Seschrock vdev_queue_init(vd); 3593d7072f8Seschrock vdev_cache_init(vd); 360fa9e4066Sahrens 361fa9e4066Sahrens return (vd); 362fa9e4066Sahrens } 363fa9e4066Sahrens 364fa9e4066Sahrens /* 365fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 366fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 367fa9e4066Sahrens * different for each case. 368fa9e4066Sahrens */ 36999653d4eSeschrock int 37099653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 37199653d4eSeschrock int alloctype) 372fa9e4066Sahrens { 373fa9e4066Sahrens vdev_ops_t *ops; 374fa9e4066Sahrens char *type; 3758654d025Sperrin uint64_t guid = 0, islog, nparity; 376fa9e4066Sahrens vdev_t *vd; 377fa9e4066Sahrens 378e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 379fa9e4066Sahrens 380fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 381be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 382fa9e4066Sahrens 383fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 384be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 385fa9e4066Sahrens 386fa9e4066Sahrens /* 387fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 388fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 389fa9e4066Sahrens */ 390fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 391fa9e4066Sahrens uint64_t label_id; 392fa9e4066Sahrens 393fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 394fa9e4066Sahrens label_id != id) 395be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 396fa9e4066Sahrens 397fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 398be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 39999653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 40099653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 401be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 402fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 403fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 404be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40521ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 40621ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 407be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 408fa9e4066Sahrens } 409fa9e4066Sahrens 41099653d4eSeschrock /* 41199653d4eSeschrock * The first allocated vdev must be of type 'root'. 41299653d4eSeschrock */ 41399653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 414be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 41599653d4eSeschrock 4168654d025Sperrin /* 4178654d025Sperrin * Determine whether we're a log vdev. 4188654d025Sperrin */ 4198654d025Sperrin islog = 0; 4208654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 421990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 422be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 423fa9e4066Sahrens 42488ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 425be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 42688ecc943SGeorge Wilson 42799653d4eSeschrock /* 4288654d025Sperrin * Set the nparity property for RAID-Z vdevs. 42999653d4eSeschrock */ 4308654d025Sperrin nparity = -1ULL; 43199653d4eSeschrock if (ops == &vdev_raidz_ops) { 43299653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4338654d025Sperrin &nparity) == 0) { 434b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 435be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 43699653d4eSeschrock /* 437f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 438f94275ceSAdam Leventhal * device. 43999653d4eSeschrock */ 440f94275ceSAdam Leventhal if (nparity > 1 && 441f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 442be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 443f94275ceSAdam Leventhal if (nparity > 2 && 444f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 445be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 44699653d4eSeschrock } else { 44799653d4eSeschrock /* 44899653d4eSeschrock * We require the parity to be specified for SPAs that 44999653d4eSeschrock * support multiple parity levels. 45099653d4eSeschrock */ 451f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 452be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 45399653d4eSeschrock /* 45499653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 45599653d4eSeschrock */ 4568654d025Sperrin nparity = 1; 45799653d4eSeschrock } 45899653d4eSeschrock } else { 4598654d025Sperrin nparity = 0; 46099653d4eSeschrock } 4618654d025Sperrin ASSERT(nparity != -1ULL); 4628654d025Sperrin 4638654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4648654d025Sperrin 4658654d025Sperrin vd->vdev_islog = islog; 4668654d025Sperrin vd->vdev_nparity = nparity; 4678654d025Sperrin 4688654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4698654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4708654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4718654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4728654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4738654d025Sperrin &vd->vdev_physpath) == 0) 4748654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4756809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4766809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 47799653d4eSeschrock 478afefbcddSeschrock /* 479afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 480afefbcddSeschrock * as -1. 481afefbcddSeschrock */ 482afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 483afefbcddSeschrock &vd->vdev_wholedisk) != 0) 484afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 485afefbcddSeschrock 486ea8dc4b6Seschrock /* 487ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 488ea8dc4b6Seschrock * was not present at the time of import. 489ea8dc4b6Seschrock */ 4906809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4916809eb4eSEric Schrock &vd->vdev_not_present); 492ea8dc4b6Seschrock 493ecc2d604Sbonwick /* 494ecc2d604Sbonwick * Get the alignment requirement. 495ecc2d604Sbonwick */ 496ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 497ecc2d604Sbonwick 49888ecc943SGeorge Wilson /* 49988ecc943SGeorge Wilson * Retrieve the vdev creation time. 50088ecc943SGeorge Wilson */ 50188ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 50288ecc943SGeorge Wilson &vd->vdev_crtxg); 50388ecc943SGeorge Wilson 504fa9e4066Sahrens /* 505fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 506fa9e4066Sahrens */ 5071195e687SMark J Musante if (parent && !parent->vdev_parent && 5081195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 509fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 510fa9e4066Sahrens &vd->vdev_ms_array); 511fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 512fa9e4066Sahrens &vd->vdev_ms_shift); 513fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 514fa9e4066Sahrens &vd->vdev_asize); 5153f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 5163f9d6ad7SLin Ling &vd->vdev_removing); 517fa9e4066Sahrens } 518fa9e4066Sahrens 519cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 520a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 5219f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 5221195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 5239f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 524a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 525a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 526a1521560SJeff Bonwick } 527a1521560SJeff Bonwick 528fa9e4066Sahrens /* 5293d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 530fa9e4066Sahrens */ 531c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 53221ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 53321ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 534c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 535c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5360713e232SGeorge Wilson &vd->vdev_dtl_object); 537c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 538c5904d13Seschrock &vd->vdev_unspare); 539c5904d13Seschrock } 54021ecdf64SLin Ling 54121ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 54221ecdf64SLin Ling uint64_t spare = 0; 54321ecdf64SLin Ling 54421ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 54521ecdf64SLin Ling &spare) == 0 && spare) 54621ecdf64SLin Ling spa_spare_add(vd); 54721ecdf64SLin Ling } 54821ecdf64SLin Ling 549ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 550ecc2d604Sbonwick &vd->vdev_offline); 551c5904d13Seschrock 552b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 553b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 554cb04b873SMark J Musante 5553d7072f8Seschrock /* 5563d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5573d7072f8Seschrock * state, as the diagnosis made on another system may not be 558069f55e2SEric Schrock * valid in the current context. Local vdevs will 559069f55e2SEric Schrock * remain in the faulted state. 5603d7072f8Seschrock */ 561b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5623d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5633d7072f8Seschrock &vd->vdev_faulted); 5643d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5653d7072f8Seschrock &vd->vdev_degraded); 5663d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5673d7072f8Seschrock &vd->vdev_removed); 568069f55e2SEric Schrock 569069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 570069f55e2SEric Schrock char *aux; 571069f55e2SEric Schrock 572069f55e2SEric Schrock vd->vdev_label_aux = 573069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 574069f55e2SEric Schrock if (nvlist_lookup_string(nv, 575069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 576069f55e2SEric Schrock strcmp(aux, "external") == 0) 577069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 578069f55e2SEric Schrock } 5793d7072f8Seschrock } 580fa9e4066Sahrens } 581fa9e4066Sahrens 582fa9e4066Sahrens /* 583fa9e4066Sahrens * Add ourselves to the parent's list of children. 584fa9e4066Sahrens */ 585fa9e4066Sahrens vdev_add_child(parent, vd); 586fa9e4066Sahrens 58799653d4eSeschrock *vdp = vd; 58899653d4eSeschrock 58999653d4eSeschrock return (0); 590fa9e4066Sahrens } 591fa9e4066Sahrens 592fa9e4066Sahrens void 593fa9e4066Sahrens vdev_free(vdev_t *vd) 594fa9e4066Sahrens { 5953d7072f8Seschrock spa_t *spa = vd->vdev_spa; 596fa9e4066Sahrens 597fa9e4066Sahrens /* 598fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 599fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 600fa9e4066Sahrens */ 601fa9e4066Sahrens vdev_close(vd); 602fa9e4066Sahrens 603e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 604b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 605fa9e4066Sahrens 606fa9e4066Sahrens /* 607fa9e4066Sahrens * Free all children. 608fa9e4066Sahrens */ 609573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 610fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 611fa9e4066Sahrens 612fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 613fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 614fa9e4066Sahrens 615fa9e4066Sahrens /* 616fa9e4066Sahrens * Discard allocation state. 617fa9e4066Sahrens */ 618a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 619fa9e4066Sahrens vdev_metaslab_fini(vd); 620a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 621a1521560SJeff Bonwick } 622fa9e4066Sahrens 623fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 624fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 625fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 626fa9e4066Sahrens 627fa9e4066Sahrens /* 628fa9e4066Sahrens * Remove this vdev from its parent's child list. 629fa9e4066Sahrens */ 630fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 631fa9e4066Sahrens 632fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 633fa9e4066Sahrens 6343d7072f8Seschrock /* 6353d7072f8Seschrock * Clean up vdev structure. 6363d7072f8Seschrock */ 6373d7072f8Seschrock vdev_queue_fini(vd); 6383d7072f8Seschrock vdev_cache_fini(vd); 6393d7072f8Seschrock 6403d7072f8Seschrock if (vd->vdev_path) 6413d7072f8Seschrock spa_strfree(vd->vdev_path); 6423d7072f8Seschrock if (vd->vdev_devid) 6433d7072f8Seschrock spa_strfree(vd->vdev_devid); 6443d7072f8Seschrock if (vd->vdev_physpath) 6453d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6466809eb4eSEric Schrock if (vd->vdev_fru) 6476809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6483d7072f8Seschrock 6493d7072f8Seschrock if (vd->vdev_isspare) 6503d7072f8Seschrock spa_spare_remove(vd); 651fa94a07fSbrendan if (vd->vdev_isl2cache) 652fa94a07fSbrendan spa_l2cache_remove(vd); 6533d7072f8Seschrock 6543d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6553d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6568ad4d6ddSJeff Bonwick 6573d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6580713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 6598ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6600713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 6610713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 6628ad4d6ddSJeff Bonwick } 6633d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6648ad4d6ddSJeff Bonwick 6653d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6663d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 667e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6683d7072f8Seschrock 6693d7072f8Seschrock if (vd == spa->spa_root_vdev) 6703d7072f8Seschrock spa->spa_root_vdev = NULL; 6713d7072f8Seschrock 6723d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 673fa9e4066Sahrens } 674fa9e4066Sahrens 675fa9e4066Sahrens /* 676fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 677fa9e4066Sahrens */ 678fa9e4066Sahrens static void 679fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 680fa9e4066Sahrens { 681fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 682fa9e4066Sahrens metaslab_t *msp; 683fa9e4066Sahrens vdev_t *vd; 684fa9e4066Sahrens int t; 685fa9e4066Sahrens 686fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 687fa9e4066Sahrens 688fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 689fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 690fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 691fa9e4066Sahrens 692fa9e4066Sahrens svd->vdev_ms_array = 0; 693fa9e4066Sahrens svd->vdev_ms_shift = 0; 694fa9e4066Sahrens svd->vdev_ms_count = 0; 695fa9e4066Sahrens 696cd0837ccSGeorge Wilson if (tvd->vdev_mg) 697cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 698fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 699fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 700fa9e4066Sahrens 701fa9e4066Sahrens svd->vdev_mg = NULL; 702fa9e4066Sahrens svd->vdev_ms = NULL; 703ecc2d604Sbonwick 704ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 705ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 706fa9e4066Sahrens 707fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 708fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 70999653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 710fa9e4066Sahrens 711fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 712fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 71399653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 714fa9e4066Sahrens 715fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 716fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 717fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 718fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 719fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 720fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 721fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 722fa9e4066Sahrens } 723fa9e4066Sahrens 724e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 725fa9e4066Sahrens vdev_config_clean(svd); 726fa9e4066Sahrens vdev_config_dirty(tvd); 727fa9e4066Sahrens } 728fa9e4066Sahrens 729e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 730e14bb325SJeff Bonwick vdev_state_clean(svd); 731e14bb325SJeff Bonwick vdev_state_dirty(tvd); 732e14bb325SJeff Bonwick } 733e14bb325SJeff Bonwick 73499653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 73599653d4eSeschrock svd->vdev_deflate_ratio = 0; 7368654d025Sperrin 7378654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7388654d025Sperrin svd->vdev_islog = 0; 739fa9e4066Sahrens } 740fa9e4066Sahrens 741fa9e4066Sahrens static void 742fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 743fa9e4066Sahrens { 744fa9e4066Sahrens if (vd == NULL) 745fa9e4066Sahrens return; 746fa9e4066Sahrens 747fa9e4066Sahrens vd->vdev_top = tvd; 748fa9e4066Sahrens 749573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 750fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 751fa9e4066Sahrens } 752fa9e4066Sahrens 753fa9e4066Sahrens /* 754fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 755fa9e4066Sahrens */ 756fa9e4066Sahrens vdev_t * 757fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 758fa9e4066Sahrens { 759fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 760fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 761fa9e4066Sahrens vdev_t *mvd; 762fa9e4066Sahrens 763e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 764fa9e4066Sahrens 765fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 766ecc2d604Sbonwick 767ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 768573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 7694263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 770ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 771ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 77288ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 773ecc2d604Sbonwick 774fa9e4066Sahrens vdev_remove_child(pvd, cvd); 775fa9e4066Sahrens vdev_add_child(pvd, mvd); 776fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 777fa9e4066Sahrens vdev_add_child(mvd, cvd); 778fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 779fa9e4066Sahrens 780fa9e4066Sahrens if (mvd == mvd->vdev_top) 781fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 782fa9e4066Sahrens 783fa9e4066Sahrens return (mvd); 784fa9e4066Sahrens } 785fa9e4066Sahrens 786fa9e4066Sahrens /* 787fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 788fa9e4066Sahrens */ 789fa9e4066Sahrens void 790fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 791fa9e4066Sahrens { 792fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 793fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 794fa9e4066Sahrens 795e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 796fa9e4066Sahrens 797fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 798fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 79999653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 80099653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 801ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 802fa9e4066Sahrens 803fa9e4066Sahrens vdev_remove_child(mvd, cvd); 804fa9e4066Sahrens vdev_remove_child(pvd, mvd); 8058ad4d6ddSJeff Bonwick 80699653d4eSeschrock /* 807e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 808e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 809e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 810e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 81199653d4eSeschrock */ 8128ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 8138ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 8141195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 8158ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 8168ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 8178ad4d6ddSJeff Bonwick } 818e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 819e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 820fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 821fa9e4066Sahrens 822fa9e4066Sahrens if (cvd == cvd->vdev_top) 823fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 824fa9e4066Sahrens 825fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 826fa9e4066Sahrens vdev_free(mvd); 827fa9e4066Sahrens } 828fa9e4066Sahrens 829ea8dc4b6Seschrock int 830fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 831fa9e4066Sahrens { 832fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 833ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 834ecc2d604Sbonwick uint64_t m; 835fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 836fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 837ecc2d604Sbonwick metaslab_t **mspp; 838ecc2d604Sbonwick int error; 839fa9e4066Sahrens 840a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 841a1521560SJeff Bonwick 84288ecc943SGeorge Wilson /* 84388ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 84488ecc943SGeorge Wilson */ 84588ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8460e34b6a7Sbonwick return (0); 8470e34b6a7Sbonwick 84888ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 84988ecc943SGeorge Wilson 850e6ca193dSGeorge Wilson /* 851e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 852b5152584SMatthew Ahrens * in 128k (1 << 17) because it is the "typical" blocksize. 853b5152584SMatthew Ahrens * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 854b5152584SMatthew Ahrens * otherwise it would inconsistently account for existing bp's. 855e6ca193dSGeorge Wilson */ 856e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 857e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 858e6ca193dSGeorge Wilson 859fa9e4066Sahrens ASSERT(oldc <= newc); 860fa9e4066Sahrens 861ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 862fa9e4066Sahrens 863ecc2d604Sbonwick if (oldc != 0) { 864ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 865ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 866ecc2d604Sbonwick } 867fa9e4066Sahrens 868ecc2d604Sbonwick vd->vdev_ms = mspp; 869ecc2d604Sbonwick vd->vdev_ms_count = newc; 870fa9e4066Sahrens 871ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 8720713e232SGeorge Wilson uint64_t object = 0; 8730713e232SGeorge Wilson 874ecc2d604Sbonwick if (txg == 0) { 875ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8767bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8777bfdf011SNeil Perrin DMU_READ_PREFETCH); 878ecc2d604Sbonwick if (error) 879ecc2d604Sbonwick return (error); 880fa9e4066Sahrens } 8811e9bd7ecSPrakash Surya 8821e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 8831e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 8841e9bd7ecSPrakash Surya if (error) 8851e9bd7ecSPrakash Surya return (error); 886fa9e4066Sahrens } 887fa9e4066Sahrens 888a1521560SJeff Bonwick if (txg == 0) 889a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 890a1521560SJeff Bonwick 8913f9d6ad7SLin Ling /* 8923f9d6ad7SLin Ling * If the vdev is being removed we don't activate 8933f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 8943f9d6ad7SLin Ling * allocations are performed on this device. 8953f9d6ad7SLin Ling */ 8963f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 897a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 898a1521560SJeff Bonwick 899a1521560SJeff Bonwick if (txg == 0) 900a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 901a1521560SJeff Bonwick 902ea8dc4b6Seschrock return (0); 903fa9e4066Sahrens } 904fa9e4066Sahrens 905fa9e4066Sahrens void 906fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 907fa9e4066Sahrens { 908fa9e4066Sahrens uint64_t m; 909fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 910fa9e4066Sahrens 911fa9e4066Sahrens if (vd->vdev_ms != NULL) { 912a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 9130713e232SGeorge Wilson for (m = 0; m < count; m++) { 9140713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 9150713e232SGeorge Wilson 9160713e232SGeorge Wilson if (msp != NULL) 9170713e232SGeorge Wilson metaslab_fini(msp); 9180713e232SGeorge Wilson } 919fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 920fa9e4066Sahrens vd->vdev_ms = NULL; 921fa9e4066Sahrens } 922fa9e4066Sahrens } 923fa9e4066Sahrens 924e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 925e14bb325SJeff Bonwick boolean_t vps_readable; 926e14bb325SJeff Bonwick boolean_t vps_writeable; 927e14bb325SJeff Bonwick int vps_flags; 928e14bb325SJeff Bonwick } vdev_probe_stats_t; 929e14bb325SJeff Bonwick 930e14bb325SJeff Bonwick static void 931e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9320a4e9518Sgw { 9338ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 934a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 935e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 936a3f829aeSBill Moore 937a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 938e14bb325SJeff Bonwick 939e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 940e14bb325SJeff Bonwick if (zio->io_error == 0) 941e14bb325SJeff Bonwick vps->vps_readable = 1; 9428ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 943a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 944e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 945e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 946e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 947e14bb325SJeff Bonwick } else { 948e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 949e14bb325SJeff Bonwick } 950e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 951e14bb325SJeff Bonwick if (zio->io_error == 0) 952e14bb325SJeff Bonwick vps->vps_writeable = 1; 953e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 954e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 955a3f829aeSBill Moore zio_t *pio; 956e14bb325SJeff Bonwick 957e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 958e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 959e14bb325SJeff Bonwick 960e14bb325SJeff Bonwick if (vdev_readable(vd) && 9618ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 962e14bb325SJeff Bonwick zio->io_error = 0; 963e14bb325SJeff Bonwick } else { 964e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 965e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9668ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 967be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 968e14bb325SJeff Bonwick } 969a3f829aeSBill Moore 970a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 971a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 972a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 973a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 974a3f829aeSBill Moore 975a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 976a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 977be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 978a3f829aeSBill Moore 979e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 980e14bb325SJeff Bonwick } 981e14bb325SJeff Bonwick } 9820a4e9518Sgw 983e14bb325SJeff Bonwick /* 984f7170741SWill Andrews * Determine whether this device is accessible. 985f7170741SWill Andrews * 986f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 987f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 988f7170741SWill Andrews * a VTOC. 989e14bb325SJeff Bonwick */ 990e14bb325SJeff Bonwick zio_t * 991a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 992e14bb325SJeff Bonwick { 993e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 994a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 995a3f829aeSBill Moore zio_t *pio; 996a3f829aeSBill Moore 997a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9980a4e9518Sgw 999a3f829aeSBill Moore /* 1000a3f829aeSBill Moore * Don't probe the probe. 1001a3f829aeSBill Moore */ 1002a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1003a3f829aeSBill Moore return (NULL); 1004e14bb325SJeff Bonwick 1005a3f829aeSBill Moore /* 1006a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1007a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1008a3f829aeSBill Moore * this vdev will become parents of the probe io. 1009a3f829aeSBill Moore */ 1010a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1011e14bb325SJeff Bonwick 1012a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1013a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1014a3f829aeSBill Moore 1015a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1016a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 10178956713aSEric Schrock ZIO_FLAG_TRYHARD; 1018a3f829aeSBill Moore 1019a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1020a3f829aeSBill Moore /* 1021a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1022a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1023a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1024a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1025a3f829aeSBill Moore * any zio looking at these values can assume that 1026a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1027a3f829aeSBill Moore * important because when a device has intermittent 1028a3f829aeSBill Moore * connectivity problems, we want to ensure that 1029a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1030a3f829aeSBill Moore * the zio (EIO). 1031a3f829aeSBill Moore * 1032a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1033a3f829aeSBill Moore * values so the probe can reevaluate from first 1034a3f829aeSBill Moore * principles. 1035a3f829aeSBill Moore */ 1036a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1037a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1038a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1039a3f829aeSBill Moore } 1040a3f829aeSBill Moore 1041a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1042a3f829aeSBill Moore vdev_probe_done, vps, 1043a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1044a3f829aeSBill Moore 104598d1cbfeSGeorge Wilson /* 104698d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 104798d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 104898d1cbfeSGeorge Wilson */ 1049a3f829aeSBill Moore if (zio != NULL) { 1050a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1051a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1052a3f829aeSBill Moore } 1053e14bb325SJeff Bonwick } 1054e14bb325SJeff Bonwick 1055a3f829aeSBill Moore if (zio != NULL) 1056a3f829aeSBill Moore zio_add_child(zio, pio); 1057e14bb325SJeff Bonwick 1058a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1059e14bb325SJeff Bonwick 1060a3f829aeSBill Moore if (vps == NULL) { 1061a3f829aeSBill Moore ASSERT(zio != NULL); 1062a3f829aeSBill Moore return (NULL); 1063a3f829aeSBill Moore } 1064e14bb325SJeff Bonwick 1065e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1066a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1067e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1068f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1069f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1070e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1071e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1072e14bb325SJeff Bonwick } 1073e14bb325SJeff Bonwick 1074a3f829aeSBill Moore if (zio == NULL) 1075a3f829aeSBill Moore return (pio); 1076a3f829aeSBill Moore 1077a3f829aeSBill Moore zio_nowait(pio); 1078a3f829aeSBill Moore return (NULL); 10790a4e9518Sgw } 10800a4e9518Sgw 1081f64c0e34SEric Taylor static void 1082f64c0e34SEric Taylor vdev_open_child(void *arg) 1083f64c0e34SEric Taylor { 1084f64c0e34SEric Taylor vdev_t *vd = arg; 1085f64c0e34SEric Taylor 1086f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1087f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1088f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1089f64c0e34SEric Taylor } 1090f64c0e34SEric Taylor 1091681d9761SEric Taylor boolean_t 1092681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1093681d9761SEric Taylor { 1094681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1095681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1096681d9761SEric Taylor return (B_TRUE); 1097681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1098681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1099681d9761SEric Taylor return (B_TRUE); 1100681d9761SEric Taylor return (B_FALSE); 1101681d9761SEric Taylor } 1102681d9761SEric Taylor 1103f64c0e34SEric Taylor void 1104f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1105f64c0e34SEric Taylor { 1106f64c0e34SEric Taylor taskq_t *tq; 1107f64c0e34SEric Taylor int children = vd->vdev_children; 1108f64c0e34SEric Taylor 1109681d9761SEric Taylor /* 1110681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1111681d9761SEric Taylor * in a single thread so that the same thread holds the 1112681d9761SEric Taylor * spa_namespace_lock 1113681d9761SEric Taylor */ 1114681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1115681d9761SEric Taylor for (int c = 0; c < children; c++) 1116681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1117681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1118681d9761SEric Taylor return; 1119681d9761SEric Taylor } 1120f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1121f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1122f64c0e34SEric Taylor 1123f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1124f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1125f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1126f64c0e34SEric Taylor 1127f64c0e34SEric Taylor taskq_destroy(tq); 1128f64c0e34SEric Taylor } 1129f64c0e34SEric Taylor 1130fa9e4066Sahrens /* 1131fa9e4066Sahrens * Prepare a virtual device for access. 1132fa9e4066Sahrens */ 1133fa9e4066Sahrens int 1134fa9e4066Sahrens vdev_open(vdev_t *vd) 1135fa9e4066Sahrens { 11368ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1137fa9e4066Sahrens int error; 1138fa9e4066Sahrens uint64_t osize = 0; 11394263d13fSGeorge Wilson uint64_t max_osize = 0; 11404263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1141ecc2d604Sbonwick uint64_t ashift = 0; 1142fa9e4066Sahrens 1143f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1144f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1145fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1146fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1147fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1148fa9e4066Sahrens 1149fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1150e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1151e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1152573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1153fa9e4066Sahrens 1154069f55e2SEric Schrock /* 1155069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1156069f55e2SEric Schrock * faulted, bail out of the open. 1157069f55e2SEric Schrock */ 11583d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11593d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1160069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1161069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11623d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1163069f55e2SEric Schrock vd->vdev_label_aux); 1164be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 11653d7072f8Seschrock } else if (vd->vdev_offline) { 1166fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1167ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1168be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1169fa9e4066Sahrens } 1170fa9e4066Sahrens 11714263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1172fa9e4066Sahrens 1173095bcd66SGeorge Wilson /* 1174095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1175095bcd66SGeorge Wilson * the vdev on error. 1176095bcd66SGeorge Wilson */ 1177095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1178ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11798956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1180ea8dc4b6Seschrock 1181fa9e4066Sahrens if (error) { 11823d7072f8Seschrock if (vd->vdev_removed && 11833d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11843d7072f8Seschrock vd->vdev_removed = B_FALSE; 11853d7072f8Seschrock 1186ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1187fa9e4066Sahrens vd->vdev_stat.vs_aux); 1188fa9e4066Sahrens return (error); 1189fa9e4066Sahrens } 1190fa9e4066Sahrens 11913d7072f8Seschrock vd->vdev_removed = B_FALSE; 11923d7072f8Seschrock 1193096d22d4SEric Schrock /* 1194096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1195096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1196096d22d4SEric Schrock */ 1197096d22d4SEric Schrock if (vd->vdev_faulted) { 1198096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1199096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1200096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1201096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1202096d22d4SEric Schrock vd->vdev_label_aux); 1203be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1204096d22d4SEric Schrock } 1205096d22d4SEric Schrock 12063d7072f8Seschrock if (vd->vdev_degraded) { 12073d7072f8Seschrock ASSERT(vd->vdev_children == 0); 12083d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 12093d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 12103d7072f8Seschrock } else { 1211069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 12123d7072f8Seschrock } 1213fa9e4066Sahrens 121488ecc943SGeorge Wilson /* 121588ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 121688ecc943SGeorge Wilson */ 121788ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 121888ecc943SGeorge Wilson return (0); 121988ecc943SGeorge Wilson 1220573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1221ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1222ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1223ea8dc4b6Seschrock VDEV_AUX_NONE); 1224ea8dc4b6Seschrock break; 1225ea8dc4b6Seschrock } 1226573ca77eSGeorge Wilson } 1227fa9e4066Sahrens 1228fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 12294263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1230fa9e4066Sahrens 1231fa9e4066Sahrens if (vd->vdev_children == 0) { 1232fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1233ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1234ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1235be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1236fa9e4066Sahrens } 1237fa9e4066Sahrens psize = osize; 1238fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 12394263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 12404263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1241fa9e4066Sahrens } else { 1242ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1243fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1244ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1245ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1246be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1247fa9e4066Sahrens } 1248fa9e4066Sahrens psize = 0; 1249fa9e4066Sahrens asize = osize; 12504263d13fSGeorge Wilson max_asize = max_osize; 1251fa9e4066Sahrens } 1252fa9e4066Sahrens 1253fa9e4066Sahrens vd->vdev_psize = psize; 1254fa9e4066Sahrens 1255573ca77eSGeorge Wilson /* 1256573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1257573ca77eSGeorge Wilson */ 1258573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1259573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1260573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1261be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1262573ca77eSGeorge Wilson } 1263573ca77eSGeorge Wilson 1264fa9e4066Sahrens if (vd->vdev_asize == 0) { 1265fa9e4066Sahrens /* 1266fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1267ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1268fa9e4066Sahrens */ 1269fa9e4066Sahrens vd->vdev_asize = asize; 12704263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1271ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1272fa9e4066Sahrens } else { 1273fa9e4066Sahrens /* 12742384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 12752384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 12762384d9f8SGeorge Wilson * issue a warning instead. 1277fa9e4066Sahrens */ 12782384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 12792384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 12802384d9f8SGeorge Wilson cmn_err(CE_WARN, 12812384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 12822384d9f8SGeorge Wilson "larger than the pool's alignment\n", 12832384d9f8SGeorge Wilson vd->vdev_path); 1284fa9e4066Sahrens } 12854263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1286573ca77eSGeorge Wilson } 1287fa9e4066Sahrens 1288573ca77eSGeorge Wilson /* 1289573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1290573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1291573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1292573ca77eSGeorge Wilson */ 1293573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1294573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1295573ca77eSGeorge Wilson vd->vdev_asize = asize; 1296fa9e4066Sahrens 1297573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1298fa9e4066Sahrens 12990a4e9518Sgw /* 13000a4e9518Sgw * Ensure we can issue some IO before declaring the 13010a4e9518Sgw * vdev open for business. 13020a4e9518Sgw */ 1303e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1304e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 130598d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 130698d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 13070a4e9518Sgw return (error); 13080a4e9518Sgw } 13090a4e9518Sgw 1310*81cd5c55SMatthew Ahrens /* 1311*81cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 1312*81cd5c55SMatthew Ahrens */ 1313*81cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1314*81cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 1315*81cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 1316*81cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 1317*81cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 1318*81cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 1319*81cd5c55SMatthew Ahrens } 1320*81cd5c55SMatthew Ahrens 1321088f3894Sahrens /* 1322088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 13238ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 13248ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1325088f3894Sahrens */ 13268ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 13278ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 13288ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1329088f3894Sahrens 1330fa9e4066Sahrens return (0); 1331fa9e4066Sahrens } 1332fa9e4066Sahrens 1333560e6e96Seschrock /* 1334560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1335560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 13363d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1337560e6e96Seschrock * 1338d7f601efSGeorge Wilson * If 'strict' is false ignore the spa guid check. This is necessary because 1339d7f601efSGeorge Wilson * if the machine crashed during a re-guid the new guid might have been written 1340d7f601efSGeorge Wilson * to all of the vdev labels, but not the cached config. The strict check 1341d7f601efSGeorge Wilson * will be performed when the pool is opened again using the mos config. 1342d7f601efSGeorge Wilson * 1343560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1344560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1345560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1346560e6e96Seschrock * will be updated but the function will return 0. 1347560e6e96Seschrock */ 1348560e6e96Seschrock int 1349d7f601efSGeorge Wilson vdev_validate(vdev_t *vd, boolean_t strict) 1350560e6e96Seschrock { 1351560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1352560e6e96Seschrock nvlist_t *label; 13531195e687SMark J Musante uint64_t guid = 0, top_guid; 1354560e6e96Seschrock uint64_t state; 1355560e6e96Seschrock 1356573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1357d7f601efSGeorge Wilson if (vdev_validate(vd->vdev_child[c], strict) != 0) 1358be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1359560e6e96Seschrock 1360b5989ec7Seschrock /* 1361b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1362b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1363b5989ec7Seschrock * overwrite the previous state. 1364b5989ec7Seschrock */ 1365e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13661195e687SMark J Musante uint64_t aux_guid = 0; 13671195e687SMark J Musante nvlist_t *nvl; 1368bda88194SGeorge Wilson uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1369bda88194SGeorge Wilson spa_last_synced_txg(spa) : -1ULL; 1370560e6e96Seschrock 1371dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1372560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1373560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1374560e6e96Seschrock return (0); 1375560e6e96Seschrock } 1376560e6e96Seschrock 13771195e687SMark J Musante /* 13781195e687SMark J Musante * Determine if this vdev has been split off into another 13791195e687SMark J Musante * pool. If so, then refuse to open it. 13801195e687SMark J Musante */ 13811195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 13821195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 13831195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 13841195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 13851195e687SMark J Musante nvlist_free(label); 13861195e687SMark J Musante return (0); 13871195e687SMark J Musante } 13881195e687SMark J Musante 1389d7f601efSGeorge Wilson if (strict && (nvlist_lookup_uint64(label, 1390d7f601efSGeorge Wilson ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1391d7f601efSGeorge Wilson guid != spa_guid(spa))) { 1392560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1393560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1394560e6e96Seschrock nvlist_free(label); 1395560e6e96Seschrock return (0); 1396560e6e96Seschrock } 1397560e6e96Seschrock 13981195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 13991195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 14001195e687SMark J Musante &aux_guid) != 0) 14011195e687SMark J Musante aux_guid = 0; 14021195e687SMark J Musante 1403e14bb325SJeff Bonwick /* 1404e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1405e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1406e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1407e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1408e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1409e14bb325SJeff Bonwick * safely compare to that instead. 14101195e687SMark J Musante * 14111195e687SMark J Musante * If we split this vdev off instead, then we also check the 14121195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 14131195e687SMark J Musante * corrupt if it is partway through a split operation. 1414e14bb325SJeff Bonwick */ 1415560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1416e14bb325SJeff Bonwick &guid) != 0 || 1417e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1418e14bb325SJeff Bonwick &top_guid) != 0 || 14191195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1420e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1421560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1422560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1423560e6e96Seschrock nvlist_free(label); 1424560e6e96Seschrock return (0); 1425560e6e96Seschrock } 1426560e6e96Seschrock 1427560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1428560e6e96Seschrock &state) != 0) { 1429560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1430560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1431560e6e96Seschrock nvlist_free(label); 1432560e6e96Seschrock return (0); 1433560e6e96Seschrock } 1434560e6e96Seschrock 1435560e6e96Seschrock nvlist_free(label); 1436560e6e96Seschrock 1437bc758434SLin Ling /* 14384b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1439bc758434SLin Ling * state of the pool. 1440bc758434SLin Ling */ 14414b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1442b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1443bc758434SLin Ling state != POOL_STATE_ACTIVE) 1444be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1445560e6e96Seschrock 144651ece835Seschrock /* 144751ece835Seschrock * If we were able to open and validate a vdev that was 144851ece835Seschrock * previously marked permanently unavailable, clear that state 144951ece835Seschrock * now. 145051ece835Seschrock */ 145151ece835Seschrock if (vd->vdev_not_present) 145251ece835Seschrock vd->vdev_not_present = 0; 145351ece835Seschrock } 1454560e6e96Seschrock 1455560e6e96Seschrock return (0); 1456560e6e96Seschrock } 1457560e6e96Seschrock 1458fa9e4066Sahrens /* 1459fa9e4066Sahrens * Close a virtual device. 1460fa9e4066Sahrens */ 1461fa9e4066Sahrens void 1462fa9e4066Sahrens vdev_close(vdev_t *vd) 1463fa9e4066Sahrens { 14648ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1465095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14668ad4d6ddSJeff Bonwick 14678ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14688ad4d6ddSJeff Bonwick 14691195e687SMark J Musante /* 14701195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14711195e687SMark J Musante * going offline. 14721195e687SMark J Musante */ 1473095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14741195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1475095bcd66SGeorge Wilson 1476fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1477fa9e4066Sahrens 14783d7072f8Seschrock vdev_cache_purge(vd); 1479fa9e4066Sahrens 1480560e6e96Seschrock /* 1481573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1482560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1483560e6e96Seschrock * it's still faulted. 1484560e6e96Seschrock */ 1485560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1486560e6e96Seschrock 1487fa9e4066Sahrens if (vd->vdev_offline) 1488fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1489fa9e4066Sahrens else 1490fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1491ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1492fa9e4066Sahrens } 1493fa9e4066Sahrens 1494dcba9f3fSGeorge Wilson void 1495dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1496dcba9f3fSGeorge Wilson { 1497dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1498dcba9f3fSGeorge Wilson 1499dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1500dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1501dcba9f3fSGeorge Wilson return; 1502dcba9f3fSGeorge Wilson 1503dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1504dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1505dcba9f3fSGeorge Wilson 1506dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1507dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1508dcba9f3fSGeorge Wilson } 1509dcba9f3fSGeorge Wilson 1510dcba9f3fSGeorge Wilson void 1511dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1512dcba9f3fSGeorge Wilson { 1513dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1514dcba9f3fSGeorge Wilson 1515dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1516dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1517dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1518dcba9f3fSGeorge Wilson 1519dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1520dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1521dcba9f3fSGeorge Wilson } 1522dcba9f3fSGeorge Wilson 1523095bcd66SGeorge Wilson /* 1524095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1525095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1526095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1527095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1528095bcd66SGeorge Wilson */ 1529fa9e4066Sahrens void 1530ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1531fa9e4066Sahrens { 1532ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1533fa9e4066Sahrens 1534e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1535ea8dc4b6Seschrock 15361195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 15371195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1538fa9e4066Sahrens vdev_close(vd); 1539fa9e4066Sahrens (void) vdev_open(vd); 1540fa9e4066Sahrens 154139c23413Seschrock /* 154239c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 154339c23413Seschrock * Otherwise, a device with an invalid label could be successfully 154439c23413Seschrock * opened in response to vdev_reopen(). 154539c23413Seschrock */ 1546c5904d13Seschrock if (vd->vdev_aux) { 1547c5904d13Seschrock (void) vdev_validate_aux(vd); 1548e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 15496809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1550573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1551573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1552c5904d13Seschrock } else { 1553bda88194SGeorge Wilson (void) vdev_validate(vd, B_TRUE); 1554c5904d13Seschrock } 155539c23413Seschrock 1556fa9e4066Sahrens /* 15573d7072f8Seschrock * Reassess parent vdev's health. 1558fa9e4066Sahrens */ 15593d7072f8Seschrock vdev_propagate_state(vd); 1560fa9e4066Sahrens } 1561fa9e4066Sahrens 1562fa9e4066Sahrens int 156399653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1564fa9e4066Sahrens { 1565fa9e4066Sahrens int error; 1566fa9e4066Sahrens 1567fa9e4066Sahrens /* 1568fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1569fa9e4066Sahrens * For a create, however, we want to fail the request if 1570fa9e4066Sahrens * there are any components we can't open. 1571fa9e4066Sahrens */ 1572fa9e4066Sahrens error = vdev_open(vd); 1573fa9e4066Sahrens 1574fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1575fa9e4066Sahrens vdev_close(vd); 1576fa9e4066Sahrens return (error ? error : ENXIO); 1577fa9e4066Sahrens } 1578fa9e4066Sahrens 1579fa9e4066Sahrens /* 15800713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1581fa9e4066Sahrens */ 15820713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 15830713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 158439c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1585fa9e4066Sahrens vdev_close(vd); 1586fa9e4066Sahrens return (error); 1587fa9e4066Sahrens } 1588fa9e4066Sahrens 1589fa9e4066Sahrens return (0); 1590fa9e4066Sahrens } 1591fa9e4066Sahrens 15920e34b6a7Sbonwick void 1593573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1594fa9e4066Sahrens { 1595fa9e4066Sahrens /* 1596bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1597fa9e4066Sahrens */ 1598bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1599fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1600fa9e4066Sahrens } 1601fa9e4066Sahrens 1602fa9e4066Sahrens void 1603ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1604fa9e4066Sahrens { 1605ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 160688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1607ecc2d604Sbonwick ASSERT(ISP2(flags)); 1608f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1609fa9e4066Sahrens 1610ecc2d604Sbonwick if (flags & VDD_METASLAB) 1611ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1612ecc2d604Sbonwick 1613ecc2d604Sbonwick if (flags & VDD_DTL) 1614ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1615ecc2d604Sbonwick 1616ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1617fa9e4066Sahrens } 1618fa9e4066Sahrens 16190713e232SGeorge Wilson void 16200713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 16210713e232SGeorge Wilson { 16220713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16230713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 16240713e232SGeorge Wilson 16250713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 16260713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 16270713e232SGeorge Wilson } 16280713e232SGeorge Wilson 16298ad4d6ddSJeff Bonwick /* 16308ad4d6ddSJeff Bonwick * DTLs. 16318ad4d6ddSJeff Bonwick * 16328ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 16339fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 16348ad4d6ddSJeff Bonwick * 16358ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 16368ad4d6ddSJeff Bonwick * 16378ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 16388ad4d6ddSJeff Bonwick * 16398ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 16408ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 16418ad4d6ddSJeff Bonwick * txgs that was scrubbed. 16428ad4d6ddSJeff Bonwick * 16438ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 16448ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 16458ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 16468ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 16478ad4d6ddSJeff Bonwick * determine whether a device can be detached. 16488ad4d6ddSJeff Bonwick * 16498ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 16508ad4d6ddSJeff Bonwick * either has the data or it doesn't. 16518ad4d6ddSJeff Bonwick * 16528ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 16538ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 16548ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 16558ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 16568ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 16578ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 16588ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 16598ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 16608ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 16618ad4d6ddSJeff Bonwick * 16628ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 16638ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 16648ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 16658ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 16668ad4d6ddSJeff Bonwick */ 1667fa9e4066Sahrens void 16688ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1669fa9e4066Sahrens { 16700713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16718ad4d6ddSJeff Bonwick 16728ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16738ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1674f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 16758ad4d6ddSJeff Bonwick 16760713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16770713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 16780713e232SGeorge Wilson range_tree_add(rt, txg, size); 16790713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1680fa9e4066Sahrens } 1681fa9e4066Sahrens 16828ad4d6ddSJeff Bonwick boolean_t 16838ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1684fa9e4066Sahrens { 16850713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16868ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1687fa9e4066Sahrens 16888ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16898ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1690fa9e4066Sahrens 16910713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16920713e232SGeorge Wilson if (range_tree_space(rt) != 0) 16930713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 16940713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1695fa9e4066Sahrens 1696fa9e4066Sahrens return (dirty); 1697fa9e4066Sahrens } 1698fa9e4066Sahrens 16998ad4d6ddSJeff Bonwick boolean_t 17008ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 17018ad4d6ddSJeff Bonwick { 17020713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17038ad4d6ddSJeff Bonwick boolean_t empty; 17048ad4d6ddSJeff Bonwick 17050713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17060713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 17070713e232SGeorge Wilson mutex_exit(rt->rt_lock); 17088ad4d6ddSJeff Bonwick 17098ad4d6ddSJeff Bonwick return (empty); 17108ad4d6ddSJeff Bonwick } 17118ad4d6ddSJeff Bonwick 1712b4952e17SGeorge Wilson /* 1713b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 1714b4952e17SGeorge Wilson */ 1715b4952e17SGeorge Wilson static uint64_t 1716b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 1717b4952e17SGeorge Wilson { 17180713e232SGeorge Wilson range_seg_t *rs; 1719b4952e17SGeorge Wilson 1720b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17210713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1722b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1723b4952e17SGeorge Wilson 17240713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17250713e232SGeorge Wilson return (rs->rs_start - 1); 1726b4952e17SGeorge Wilson } 1727b4952e17SGeorge Wilson 1728b4952e17SGeorge Wilson /* 1729b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 1730b4952e17SGeorge Wilson */ 1731b4952e17SGeorge Wilson static uint64_t 1732b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 1733b4952e17SGeorge Wilson { 17340713e232SGeorge Wilson range_seg_t *rs; 1735b4952e17SGeorge Wilson 1736b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17370713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1738b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1739b4952e17SGeorge Wilson 17400713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17410713e232SGeorge Wilson return (rs->rs_end); 1742b4952e17SGeorge Wilson } 1743b4952e17SGeorge Wilson 1744b4952e17SGeorge Wilson /* 1745b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 1746b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 1747b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 1748b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 1749b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 1750b4952e17SGeorge Wilson * excise the DTLs. 1751b4952e17SGeorge Wilson */ 1752b4952e17SGeorge Wilson static boolean_t 1753b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 1754b4952e17SGeorge Wilson { 1755b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 1756b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1757b4952e17SGeorge Wilson 1758b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 1759b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1760b4952e17SGeorge Wilson 1761b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 17620713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1763b4952e17SGeorge Wilson return (B_TRUE); 1764b4952e17SGeorge Wilson 1765b4952e17SGeorge Wilson /* 1766b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 1767b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 1768b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 1769b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 1770b4952e17SGeorge Wilson * for excision. 1771b4952e17SGeorge Wilson */ 1772b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1773b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1774b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1775b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1776b4952e17SGeorge Wilson return (B_TRUE); 1777b4952e17SGeorge Wilson } 1778b4952e17SGeorge Wilson return (B_FALSE); 1779b4952e17SGeorge Wilson } 1780b4952e17SGeorge Wilson 1781fa9e4066Sahrens /* 1782fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1783fa9e4066Sahrens */ 1784fa9e4066Sahrens void 1785fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1786fa9e4066Sahrens { 1787ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 17888ad4d6ddSJeff Bonwick avl_tree_t reftree; 17898ad4d6ddSJeff Bonwick int minref; 1790fa9e4066Sahrens 17918ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1792fa9e4066Sahrens 17938ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 17948ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 17958ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 17968ad4d6ddSJeff Bonwick 1797b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 17988ad4d6ddSJeff Bonwick return; 17998ad4d6ddSJeff Bonwick 18008ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 18013f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 18023f9d6ad7SLin Ling 1803fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1804b4952e17SGeorge Wilson 1805b4952e17SGeorge Wilson /* 1806b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 1807b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 1808b4952e17SGeorge Wilson * excise regions on vdevs that were available during 1809b4952e17SGeorge Wilson * the entire duration of this scan. 1810b4952e17SGeorge Wilson */ 1811088f3894Sahrens if (scrub_txg != 0 && 18123f9d6ad7SLin Ling (spa->spa_scrub_started || 1813b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1814b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 1815088f3894Sahrens /* 1816088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1817088f3894Sahrens * did it without rebooting, then the scrub dtl 1818088f3894Sahrens * will be valid, so excise the old region and 1819088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1820088f3894Sahrens * dtl as-is if there was an error. 18218ad4d6ddSJeff Bonwick * 18228ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 18238ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 18248ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 18258ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 18268ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 18278ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 18288ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 18298ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 18308ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1831088f3894Sahrens */ 18320713e232SGeorge Wilson space_reftree_create(&reftree); 18330713e232SGeorge Wilson space_reftree_add_map(&reftree, 18340713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18350713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 18360713e232SGeorge Wilson space_reftree_add_map(&reftree, 18370713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 18380713e232SGeorge Wilson space_reftree_generate_map(&reftree, 18390713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18400713e232SGeorge Wilson space_reftree_destroy(&reftree); 1841fa9e4066Sahrens } 18420713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 18430713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18440713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1845fa9e4066Sahrens if (scrub_done) 18460713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 18470713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 18488ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 18490713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 18508ad4d6ddSJeff Bonwick else 18510713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18520713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1853b4952e17SGeorge Wilson 1854b4952e17SGeorge Wilson /* 1855b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 1856b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 1857b4952e17SGeorge Wilson */ 1858b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 18590713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 18600713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 1861b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 1862b4952e17SGeorge Wilson 1863fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1864088f3894Sahrens 1865ecc2d604Sbonwick if (txg != 0) 1866ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1867fa9e4066Sahrens return; 1868fa9e4066Sahrens } 1869fa9e4066Sahrens 1870fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 18718ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 187299bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 187399bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 18748ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 18758ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 18768ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 18778ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 18788ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 18798ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 18808ad4d6ddSJeff Bonwick else 18818ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 18820713e232SGeorge Wilson space_reftree_create(&reftree); 18838ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 18848ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 18858ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 18860713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 18878ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 18888ad4d6ddSJeff Bonwick } 18890713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 18900713e232SGeorge Wilson space_reftree_destroy(&reftree); 1891fa9e4066Sahrens } 18928ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1893fa9e4066Sahrens } 1894fa9e4066Sahrens 18950713e232SGeorge Wilson int 1896fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1897fa9e4066Sahrens { 1898fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1899ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19000713e232SGeorge Wilson int error = 0; 1901fa9e4066Sahrens 19020713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 19030713e232SGeorge Wilson ASSERT(!vd->vdev_ishole); 1904fa9e4066Sahrens 19050713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 19060713e232SGeorge Wilson vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 19070713e232SGeorge Wilson if (error) 19080713e232SGeorge Wilson return (error); 19090713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1910fa9e4066Sahrens 19110713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 191288ecc943SGeorge Wilson 19130713e232SGeorge Wilson /* 19140713e232SGeorge Wilson * Now that we've opened the space_map we need to update 19150713e232SGeorge Wilson * the in-core DTL. 19160713e232SGeorge Wilson */ 19170713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 1918ecc2d604Sbonwick 19190713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 19200713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 19210713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1922fa9e4066Sahrens 19230713e232SGeorge Wilson return (error); 19240713e232SGeorge Wilson } 19250713e232SGeorge Wilson 19260713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 19270713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 19280713e232SGeorge Wilson if (error != 0) 19290713e232SGeorge Wilson break; 19300713e232SGeorge Wilson } 1931fa9e4066Sahrens 1932fa9e4066Sahrens return (error); 1933fa9e4066Sahrens } 1934fa9e4066Sahrens 1935fa9e4066Sahrens void 1936fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1937fa9e4066Sahrens { 1938fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 19390713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 1940ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19410713e232SGeorge Wilson range_tree_t *rtsync; 19420713e232SGeorge Wilson kmutex_t rtlock; 1943fa9e4066Sahrens dmu_tx_t *tx; 19440713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 1945fa9e4066Sahrens 194688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 19470713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 194888ecc943SGeorge Wilson 1949fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1950fa9e4066Sahrens 19510713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 19520713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 19530713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 19540713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 19550713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 19560713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1957fa9e4066Sahrens dmu_tx_commit(tx); 1958fa9e4066Sahrens return; 1959fa9e4066Sahrens } 1960fa9e4066Sahrens 19610713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 19620713e232SGeorge Wilson uint64_t new_object; 19630713e232SGeorge Wilson 19640713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 19650713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 19660713e232SGeorge Wilson 19670713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 19680713e232SGeorge Wilson 0, -1ULL, 0, &vd->vdev_dtl_lock)); 19690713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1970fa9e4066Sahrens } 1971fa9e4066Sahrens 19720713e232SGeorge Wilson mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 1973fa9e4066Sahrens 19740713e232SGeorge Wilson rtsync = range_tree_create(NULL, NULL, &rtlock); 1975fa9e4066Sahrens 19760713e232SGeorge Wilson mutex_enter(&rtlock); 1977fa9e4066Sahrens 1978fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 19790713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 1980fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1981fa9e4066Sahrens 19820713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 19830713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 19840713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 1985fa9e4066Sahrens 19860713e232SGeorge Wilson range_tree_destroy(rtsync); 1987fa9e4066Sahrens 19880713e232SGeorge Wilson mutex_exit(&rtlock); 19890713e232SGeorge Wilson mutex_destroy(&rtlock); 1990fa9e4066Sahrens 19910713e232SGeorge Wilson /* 19920713e232SGeorge Wilson * If the object for the space map has changed then dirty 19930713e232SGeorge Wilson * the top level so that we update the config. 19940713e232SGeorge Wilson */ 19950713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 19960713e232SGeorge Wilson zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 19970713e232SGeorge Wilson "new object %llu", txg, spa_name(spa), object, 19980713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 19990713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 20000713e232SGeorge Wilson } 2001fa9e4066Sahrens 2002fa9e4066Sahrens dmu_tx_commit(tx); 20030713e232SGeorge Wilson 20040713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20050713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 20060713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2007fa9e4066Sahrens } 2008fa9e4066Sahrens 20098ad4d6ddSJeff Bonwick /* 20108ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 20118ad4d6ddSJeff Bonwick * without losing data. 20128ad4d6ddSJeff Bonwick */ 20138ad4d6ddSJeff Bonwick boolean_t 20148ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 20158ad4d6ddSJeff Bonwick { 20168ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 20178ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 20188ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 20198ad4d6ddSJeff Bonwick boolean_t required; 20208ad4d6ddSJeff Bonwick 20218ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 20228ad4d6ddSJeff Bonwick 20238ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 20248ad4d6ddSJeff Bonwick return (B_TRUE); 20258ad4d6ddSJeff Bonwick 20268ad4d6ddSJeff Bonwick /* 20278ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 20288ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 20298ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 20308ad4d6ddSJeff Bonwick */ 20318ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 20328ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 20338ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 20348ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 20358ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 20368ad4d6ddSJeff Bonwick 2037cb04b873SMark J Musante if (!required && zio_injection_enabled) 2038cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2039cb04b873SMark J Musante 20408ad4d6ddSJeff Bonwick return (required); 20418ad4d6ddSJeff Bonwick } 20428ad4d6ddSJeff Bonwick 2043088f3894Sahrens /* 2044088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2045088f3894Sahrens */ 2046088f3894Sahrens boolean_t 2047088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2048088f3894Sahrens { 2049088f3894Sahrens boolean_t needed = B_FALSE; 2050088f3894Sahrens uint64_t thismin = UINT64_MAX; 2051088f3894Sahrens uint64_t thismax = 0; 2052088f3894Sahrens 2053088f3894Sahrens if (vd->vdev_children == 0) { 2054088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 20550713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 20568ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2057088f3894Sahrens 2058b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2059b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2060088f3894Sahrens needed = B_TRUE; 2061088f3894Sahrens } 2062088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2063088f3894Sahrens } else { 20648ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2065088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2066088f3894Sahrens uint64_t cmin, cmax; 2067088f3894Sahrens 2068088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2069088f3894Sahrens thismin = MIN(thismin, cmin); 2070088f3894Sahrens thismax = MAX(thismax, cmax); 2071088f3894Sahrens needed = B_TRUE; 2072088f3894Sahrens } 2073088f3894Sahrens } 2074088f3894Sahrens } 2075088f3894Sahrens 2076088f3894Sahrens if (needed && minp) { 2077088f3894Sahrens *minp = thismin; 2078088f3894Sahrens *maxp = thismax; 2079088f3894Sahrens } 2080088f3894Sahrens return (needed); 2081088f3894Sahrens } 2082088f3894Sahrens 2083560e6e96Seschrock void 2084ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2085fa9e4066Sahrens { 2086fa9e4066Sahrens /* 2087fa9e4066Sahrens * Recursively load all children. 2088fa9e4066Sahrens */ 20898ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2090560e6e96Seschrock vdev_load(vd->vdev_child[c]); 2091fa9e4066Sahrens 2092fa9e4066Sahrens /* 20930e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2094fa9e4066Sahrens */ 209588ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 2096560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2097560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 2098560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2099560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2100fa9e4066Sahrens 2101fa9e4066Sahrens /* 2102fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2103fa9e4066Sahrens */ 2104560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2105560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2106560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2107fa9e4066Sahrens } 2108fa9e4066Sahrens 210999653d4eSeschrock /* 2110fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2111fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2112fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2113fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2114fa94a07fSbrendan * repurposed to another pool. 211599653d4eSeschrock */ 211699653d4eSeschrock int 2117fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 211899653d4eSeschrock { 211999653d4eSeschrock nvlist_t *label; 212099653d4eSeschrock uint64_t guid, version; 212199653d4eSeschrock uint64_t state; 212299653d4eSeschrock 2123e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2124c5904d13Seschrock return (0); 2125c5904d13Seschrock 2126dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 212799653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 212899653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 212999653d4eSeschrock return (-1); 213099653d4eSeschrock } 213199653d4eSeschrock 213299653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2133ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 213499653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 213599653d4eSeschrock guid != vd->vdev_guid || 213699653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 213799653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 213899653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 213999653d4eSeschrock nvlist_free(label); 214099653d4eSeschrock return (-1); 214199653d4eSeschrock } 214299653d4eSeschrock 214399653d4eSeschrock /* 214499653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 214599653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 214699653d4eSeschrock */ 214799653d4eSeschrock nvlist_free(label); 214899653d4eSeschrock return (0); 214999653d4eSeschrock } 215099653d4eSeschrock 215188ecc943SGeorge Wilson void 215288ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 215388ecc943SGeorge Wilson { 215488ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 215588ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 215688ecc943SGeorge Wilson dmu_tx_t *tx; 215788ecc943SGeorge Wilson 215888ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 215988ecc943SGeorge Wilson 216088ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 21612e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 21622e4c9986SGeorge Wilson 21632e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21642e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21652e4c9986SGeorge Wilson 216688ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 216788ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 216888ecc943SGeorge Wilson 21690713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 217088ecc943SGeorge Wilson continue; 217188ecc943SGeorge Wilson 21720713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 21732e4c9986SGeorge Wilson /* 21742e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 21752e4c9986SGeorge Wilson * was removed then the histogram accounting may 21762e4c9986SGeorge Wilson * not be accurate. Update the histogram information 21772e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 21782e4c9986SGeorge Wilson * and metaslab class are up-to-date. 21792e4c9986SGeorge Wilson */ 21802e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 21812e4c9986SGeorge Wilson 21820713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 21830713e232SGeorge Wilson space_map_free(msp->ms_sm, tx); 21840713e232SGeorge Wilson space_map_close(msp->ms_sm); 21850713e232SGeorge Wilson msp->ms_sm = NULL; 21860713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 218788ecc943SGeorge Wilson } 21882e4c9986SGeorge Wilson 21892e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21902e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21912e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 21922e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 21932e4c9986SGeorge Wilson 219488ecc943SGeorge Wilson } 219588ecc943SGeorge Wilson 219688ecc943SGeorge Wilson if (vd->vdev_ms_array) { 219788ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 219888ecc943SGeorge Wilson vd->vdev_ms_array = 0; 219988ecc943SGeorge Wilson } 220088ecc943SGeorge Wilson dmu_tx_commit(tx); 220188ecc943SGeorge Wilson } 220288ecc943SGeorge Wilson 2203fa9e4066Sahrens void 2204fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2205fa9e4066Sahrens { 2206fa9e4066Sahrens metaslab_t *msp; 220780eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2208fa9e4066Sahrens 220988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 221088ecc943SGeorge Wilson 2211fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2212fa9e4066Sahrens metaslab_sync_done(msp, txg); 221380eb36f2SGeorge Wilson 221480eb36f2SGeorge Wilson if (reassess) 221580eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2216fa9e4066Sahrens } 2217fa9e4066Sahrens 2218fa9e4066Sahrens void 2219fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2220fa9e4066Sahrens { 2221fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2222fa9e4066Sahrens vdev_t *lvd; 2223fa9e4066Sahrens metaslab_t *msp; 2224ecc2d604Sbonwick dmu_tx_t *tx; 2225fa9e4066Sahrens 222688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 222788ecc943SGeorge Wilson 2228ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2229ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2230ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2231ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2232ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2233ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2234ecc2d604Sbonwick vdev_config_dirty(vd); 2235ecc2d604Sbonwick dmu_tx_commit(tx); 2236ecc2d604Sbonwick } 2237fa9e4066Sahrens 22383f9d6ad7SLin Ling /* 22393f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 22403f9d6ad7SLin Ling */ 22413f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 224288ecc943SGeorge Wilson vdev_remove(vd, txg); 224388ecc943SGeorge Wilson 2244ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2245fa9e4066Sahrens metaslab_sync(msp, txg); 2246ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2247ecc2d604Sbonwick } 2248fa9e4066Sahrens 2249fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2250fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2251fa9e4066Sahrens 2252fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2253fa9e4066Sahrens } 2254fa9e4066Sahrens 2255fa9e4066Sahrens uint64_t 2256fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2257fa9e4066Sahrens { 2258fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2259fa9e4066Sahrens } 2260fa9e4066Sahrens 22613d7072f8Seschrock /* 22623d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 22633d7072f8Seschrock * not be opened, and no I/O is attempted. 22643d7072f8Seschrock */ 2265fa9e4066Sahrens int 2266069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2267fa9e4066Sahrens { 22684b964adaSGeorge Wilson vdev_t *vd, *tvd; 2269fa9e4066Sahrens 22708f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2271fa9e4066Sahrens 2272c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2273e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2274e14bb325SJeff Bonwick 22753d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2276e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2277fa9e4066Sahrens 22784b964adaSGeorge Wilson tvd = vd->vdev_top; 22794b964adaSGeorge Wilson 2280069f55e2SEric Schrock /* 2281069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2282069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2283069f55e2SEric Schrock * were faulted. 2284069f55e2SEric Schrock */ 2285069f55e2SEric Schrock vd->vdev_label_aux = aux; 2286069f55e2SEric Schrock 22873d7072f8Seschrock /* 22883d7072f8Seschrock * Faulted state takes precedence over degraded. 22893d7072f8Seschrock */ 229098d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 22913d7072f8Seschrock vd->vdev_faulted = 1ULL; 22923d7072f8Seschrock vd->vdev_degraded = 0ULL; 2293069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 22943d7072f8Seschrock 22953d7072f8Seschrock /* 2296c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2297c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 22983d7072f8Seschrock */ 22994b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 23003d7072f8Seschrock vd->vdev_degraded = 1ULL; 23013d7072f8Seschrock vd->vdev_faulted = 0ULL; 23023d7072f8Seschrock 23033d7072f8Seschrock /* 23043d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 23053d7072f8Seschrock * mark it degraded. 23063d7072f8Seschrock */ 23074b964adaSGeorge Wilson vdev_reopen(tvd); 23083d7072f8Seschrock 2309069f55e2SEric Schrock if (vdev_readable(vd)) 2310069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 23113d7072f8Seschrock } 23123d7072f8Seschrock 2313e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23143d7072f8Seschrock } 23153d7072f8Seschrock 23163d7072f8Seschrock /* 23173d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 23183d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 23193d7072f8Seschrock * as I/O is concerned. 23203d7072f8Seschrock */ 23213d7072f8Seschrock int 2322069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 23233d7072f8Seschrock { 2324c5904d13Seschrock vdev_t *vd; 23250a4e9518Sgw 23268f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 23273d7072f8Seschrock 2328c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2329e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2330e14bb325SJeff Bonwick 23310e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2332e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 23330e34b6a7Sbonwick 23343d7072f8Seschrock /* 23353d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 23363d7072f8Seschrock */ 2337e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2338e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 23393d7072f8Seschrock 23403d7072f8Seschrock vd->vdev_degraded = 1ULL; 23413d7072f8Seschrock if (!vdev_is_dead(vd)) 23423d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2343069f55e2SEric Schrock aux); 23443d7072f8Seschrock 2345e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23463d7072f8Seschrock } 23473d7072f8Seschrock 23483d7072f8Seschrock /* 2349f7170741SWill Andrews * Online the given vdev. 2350f7170741SWill Andrews * 2351f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2352f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2353f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2354f7170741SWill Andrews * events are generated if the device fails to open. 23553d7072f8Seschrock */ 23563d7072f8Seschrock int 2357e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 23583d7072f8Seschrock { 2359573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 23603d7072f8Seschrock 23618f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 23623d7072f8Seschrock 2363c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2364e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 23653d7072f8Seschrock 23663d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2367e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2368fa9e4066Sahrens 2369573ca77eSGeorge Wilson tvd = vd->vdev_top; 2370fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2371441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2372e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2373e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2374573ca77eSGeorge Wilson 2375573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2376573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2377573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2378573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2379573ca77eSGeorge Wilson } 2380573ca77eSGeorge Wilson 2381573ca77eSGeorge Wilson vdev_reopen(tvd); 23823d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 23833d7072f8Seschrock 2384573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2385573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2386573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2387573ca77eSGeorge Wilson } 2388573ca77eSGeorge Wilson 23893d7072f8Seschrock if (newstate) 23903d7072f8Seschrock *newstate = vd->vdev_state; 23913d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 23923d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 23933d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 23943d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 23953d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2396fa9e4066Sahrens 2397573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2398573ca77eSGeorge Wilson 2399573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2400573ca77eSGeorge Wilson if (vd->vdev_aux) 2401573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2402573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2403573ca77eSGeorge Wilson } 24048ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2405fa9e4066Sahrens } 2406fa9e4066Sahrens 2407a1521560SJeff Bonwick static int 2408a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2409fa9e4066Sahrens { 2410e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 24118f18d1faSGeorge Wilson int error = 0; 24128f18d1faSGeorge Wilson uint64_t generation; 24138f18d1faSGeorge Wilson metaslab_group_t *mg; 24140a4e9518Sgw 24158f18d1faSGeorge Wilson top: 24168f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2417fa9e4066Sahrens 2418c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2419e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2420fa9e4066Sahrens 24210e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2422e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 24230e34b6a7Sbonwick 2424e6ca193dSGeorge Wilson tvd = vd->vdev_top; 24258f18d1faSGeorge Wilson mg = tvd->vdev_mg; 24268f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2427e6ca193dSGeorge Wilson 2428fa9e4066Sahrens /* 2429ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2430fa9e4066Sahrens */ 2431ecc2d604Sbonwick if (!vd->vdev_offline) { 2432ecc2d604Sbonwick /* 24338ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2434e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2435e6ca193dSGeorge Wilson * expendable. 2436ecc2d604Sbonwick */ 2437e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2438e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2439e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2440fa9e4066Sahrens 24418f18d1faSGeorge Wilson /* 2442b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2443b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2444b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2445b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 24468f18d1faSGeorge Wilson */ 24478f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 24488f18d1faSGeorge Wilson /* 24498f18d1faSGeorge Wilson * Prevent any future allocations. 24508f18d1faSGeorge Wilson */ 2451a1521560SJeff Bonwick metaslab_group_passivate(mg); 24528f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24538f18d1faSGeorge Wilson 24541195e687SMark J Musante error = spa_offline_log(spa); 24558f18d1faSGeorge Wilson 24568f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 24578f18d1faSGeorge Wilson 24588f18d1faSGeorge Wilson /* 24598f18d1faSGeorge Wilson * Check to see if the config has changed. 24608f18d1faSGeorge Wilson */ 24618f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2462a1521560SJeff Bonwick metaslab_group_activate(mg); 24638f18d1faSGeorge Wilson if (error) 24648f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 24658f18d1faSGeorge Wilson vd, error)); 24668f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24678f18d1faSGeorge Wilson goto top; 24688f18d1faSGeorge Wilson } 2469fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 24708f18d1faSGeorge Wilson } 24718f18d1faSGeorge Wilson 2472ecc2d604Sbonwick /* 2473ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2474e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2475e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2476e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2477ecc2d604Sbonwick */ 2478ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2479e6ca193dSGeorge Wilson vdev_reopen(tvd); 2480e6ca193dSGeorge Wilson 2481e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2482e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2483ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2484e6ca193dSGeorge Wilson vdev_reopen(tvd); 2485e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2486ecc2d604Sbonwick } 24878f18d1faSGeorge Wilson 24888f18d1faSGeorge Wilson /* 24898f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 24908f18d1faSGeorge Wilson * once we online the device it's open for business. 24918f18d1faSGeorge Wilson */ 24928f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2493a1521560SJeff Bonwick metaslab_group_activate(mg); 2494fa9e4066Sahrens } 2495fa9e4066Sahrens 2496e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2497ecc2d604Sbonwick 24988f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2499fa9e4066Sahrens } 2500fa9e4066Sahrens 2501a1521560SJeff Bonwick int 2502a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2503a1521560SJeff Bonwick { 2504a1521560SJeff Bonwick int error; 2505a1521560SJeff Bonwick 2506a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2507a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2508a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2509a1521560SJeff Bonwick 2510a1521560SJeff Bonwick return (error); 2511a1521560SJeff Bonwick } 2512a1521560SJeff Bonwick 2513ea8dc4b6Seschrock /* 2514ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2515ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2516ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2517ea8dc4b6Seschrock */ 2518ea8dc4b6Seschrock void 2519e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2520fa9e4066Sahrens { 2521e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2522e14bb325SJeff Bonwick 2523e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2524fa9e4066Sahrens 2525ea8dc4b6Seschrock if (vd == NULL) 2526e14bb325SJeff Bonwick vd = rvd; 2527fa9e4066Sahrens 2528ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2529ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2530ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2531fa9e4066Sahrens 2532e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2533e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 25343d7072f8Seschrock 25353d7072f8Seschrock /* 25368a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 25378a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 25388a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 25398a79c1b5Sek * written out to disk. 25403d7072f8Seschrock */ 2541e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2542e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 25438a79c1b5Sek 2544096d22d4SEric Schrock /* 2545096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2546096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2547096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2548096d22d4SEric Schrock */ 2549096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2550096d22d4SEric Schrock 25514b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2552e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2553e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2554e14bb325SJeff Bonwick 2555f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 25563d7072f8Seschrock 2557096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2558096d22d4SEric Schrock 2559f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2560e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2561e14bb325SJeff Bonwick 2562e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2563bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 25643d7072f8Seschrock 25653d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 25663d7072f8Seschrock } 2567096d22d4SEric Schrock 2568096d22d4SEric Schrock /* 2569096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2570096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2571096d22d4SEric Schrock * done in response to the FMA fault. 2572096d22d4SEric Schrock */ 2573096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2574096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2575096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2576096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2577fa9e4066Sahrens } 2578fa9e4066Sahrens 2579e14bb325SJeff Bonwick boolean_t 2580e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 25810a4e9518Sgw { 258288ecc943SGeorge Wilson /* 258388ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 258488ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 258588ecc943SGeorge Wilson * these types of devices in the various code paths. 258688ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 258788ecc943SGeorge Wilson * before issuing I/O to them. 258888ecc943SGeorge Wilson */ 258988ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 259088ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 25910a4e9518Sgw } 25920a4e9518Sgw 2593e14bb325SJeff Bonwick boolean_t 2594e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 25950a4e9518Sgw { 2596e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 25970a4e9518Sgw } 25980a4e9518Sgw 2599e14bb325SJeff Bonwick boolean_t 2600e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2601fa9e4066Sahrens { 2602e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2603fa9e4066Sahrens } 2604fa9e4066Sahrens 2605a31e6787SGeorge Wilson boolean_t 2606a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2607a31e6787SGeorge Wilson { 26088ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 26098ad4d6ddSJeff Bonwick 2610a31e6787SGeorge Wilson /* 26118ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2612a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2613a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 26148ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 26158ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 26168ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2617a31e6787SGeorge Wilson */ 26188ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 26193f9d6ad7SLin Ling !vd->vdev_cant_write && !vd->vdev_ishole); 2620a31e6787SGeorge Wilson } 2621a31e6787SGeorge Wilson 2622e14bb325SJeff Bonwick boolean_t 2623e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2624fa9e4066Sahrens { 2625e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2626fa9e4066Sahrens 2627e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2628e14bb325SJeff Bonwick return (B_FALSE); 2629fa9e4066Sahrens 2630e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2631e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2632fa9e4066Sahrens 2633e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2634e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2635fa9e4066Sahrens 2636e14bb325SJeff Bonwick return (B_TRUE); 2637fa9e4066Sahrens } 2638fa9e4066Sahrens 2639fa9e4066Sahrens /* 2640fa9e4066Sahrens * Get statistics for the given vdev. 2641fa9e4066Sahrens */ 2642fa9e4066Sahrens void 2643fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2644fa9e4066Sahrens { 26452e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 26462e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 26472e4c9986SGeorge Wilson 26482e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2649fa9e4066Sahrens 2650fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2651fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2652fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2653fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2654573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2655573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2656573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 26574263d13fSGeorge Wilson vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 26582986efa8SAlex Reece if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 26592e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 26602986efa8SAlex Reece } 2661fa9e4066Sahrens 2662fa9e4066Sahrens /* 2663fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2664fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2665fa9e4066Sahrens */ 2666fa9e4066Sahrens if (vd == rvd) { 2667e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2668fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2669fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2670fa9e4066Sahrens 2671e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2672fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2673fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2674fa9e4066Sahrens } 26753f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2676fa9e4066Sahrens } 2677fa9e4066Sahrens } 26782e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 2679fa9e4066Sahrens } 2680fa9e4066Sahrens 2681fa94a07fSbrendan void 2682fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2683fa94a07fSbrendan { 2684fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2685fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2686fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2687fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2688fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2689fa94a07fSbrendan } 2690fa94a07fSbrendan 26913f9d6ad7SLin Ling void 26923f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 26933f9d6ad7SLin Ling { 26943f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 26953f9d6ad7SLin Ling 26963f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 26973f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 26983f9d6ad7SLin Ling 26993f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 27003f9d6ad7SLin Ling vs->vs_scan_processed = 0; 27013f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 27023f9d6ad7SLin Ling } 27033f9d6ad7SLin Ling 2704fa9e4066Sahrens void 2705e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2706fa9e4066Sahrens { 27078ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 27088ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2709e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2710fa9e4066Sahrens vdev_t *pvd; 2711fa9e4066Sahrens uint64_t txg = zio->io_txg; 2712fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2713fa9e4066Sahrens zio_type_t type = zio->io_type; 2714fa9e4066Sahrens int flags = zio->io_flags; 2715fa9e4066Sahrens 2716e14bb325SJeff Bonwick /* 2717e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2718e14bb325SJeff Bonwick */ 2719e14bb325SJeff Bonwick if (zio->io_gang_tree) 2720e14bb325SJeff Bonwick return; 2721e14bb325SJeff Bonwick 2722fa9e4066Sahrens if (zio->io_error == 0) { 2723e14bb325SJeff Bonwick /* 2724e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2725e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2726e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2727e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2728e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2729e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2730e14bb325SJeff Bonwick * remain zero, which is what we want.) 2731e14bb325SJeff Bonwick * 2732e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2733e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2734e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2735e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2736e14bb325SJeff Bonwick */ 2737e14bb325SJeff Bonwick if (vd == rvd) 2738e14bb325SJeff Bonwick return; 2739e14bb325SJeff Bonwick 2740e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 27418ad4d6ddSJeff Bonwick 27428ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 27438ad4d6ddSJeff Bonwick return; 27448ad4d6ddSJeff Bonwick 27458ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 27468ad4d6ddSJeff Bonwick 2747e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 274844ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 27493f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 27503f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 27513f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 27523f9d6ad7SLin Ling 27533f9d6ad7SLin Ling /* XXX cleanup? */ 27543f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 27553f9d6ad7SLin Ling atomic_add_64(processed, psize); 27563f9d6ad7SLin Ling vs->vs_scan_processed += psize; 27573f9d6ad7SLin Ling } 27583f9d6ad7SLin Ling 27598ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2760e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2761fa9e4066Sahrens } 27628ad4d6ddSJeff Bonwick 27638ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 27648ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 27658ad4d6ddSJeff Bonwick 27668ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2767fa9e4066Sahrens return; 2768fa9e4066Sahrens } 2769fa9e4066Sahrens 2770fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2771fa9e4066Sahrens return; 2772fa9e4066Sahrens 27738956713aSEric Schrock /* 27748956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 27758956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 27768956713aSEric Schrock * hard errors, when in reality they can happen for any number of 27778956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 27788956713aSEric Schrock */ 27798956713aSEric Schrock if (zio->io_error == EIO && 27808956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 27818956713aSEric Schrock return; 27828956713aSEric Schrock 27838f18d1faSGeorge Wilson /* 27848f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 27858f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 27868f18d1faSGeorge Wilson * errors. 27878f18d1faSGeorge Wilson */ 27888f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 27898f18d1faSGeorge Wilson return; 27908f18d1faSGeorge Wilson 2791e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2792b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2793e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2794e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2795e14bb325SJeff Bonwick else 2796e14bb325SJeff Bonwick vs->vs_read_errors++; 2797fa9e4066Sahrens } 2798b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2799e14bb325SJeff Bonwick vs->vs_write_errors++; 2800e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2801fa9e4066Sahrens 28028ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 28038ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 280444ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2805b24ab676SJeff Bonwick spa->spa_claiming)) { 28068ad4d6ddSJeff Bonwick /* 2807b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2808b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2809b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2810b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2811b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2812b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2813b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2814b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 28158ad4d6ddSJeff Bonwick * 28168ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 28178ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 28188ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 28198ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 28208ad4d6ddSJeff Bonwick */ 28218ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 28228ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 282344ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 28248ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 28258ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 28268ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2827b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2828b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2829b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2830b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 28318ad4d6ddSJeff Bonwick } 2832b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 28338ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2834fa9e4066Sahrens return; 28358ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 28368ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 28378ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2838fa9e4066Sahrens } 28398ad4d6ddSJeff Bonwick if (vd != rvd) 28408ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2841fa9e4066Sahrens } 2842fa9e4066Sahrens } 2843fa9e4066Sahrens 2844fa9e4066Sahrens /* 2845b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2846b24ab676SJeff Bonwick * and the root vdev. 2847fa9e4066Sahrens */ 2848fa9e4066Sahrens void 2849b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2850b24ab676SJeff Bonwick int64_t space_delta) 2851fa9e4066Sahrens { 285299653d4eSeschrock int64_t dspace_delta = space_delta; 28538654d025Sperrin spa_t *spa = vd->vdev_spa; 28548654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2855b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2856b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2857fa9e4066Sahrens 28588654d025Sperrin ASSERT(vd == vd->vdev_top); 285999653d4eSeschrock 28608654d025Sperrin /* 28618654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 28628654d025Sperrin * factor. We must calculate this here and not at the root vdev 28638654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 28648654d025Sperrin * childrens', thus not accurate enough for us. 28658654d025Sperrin */ 28668654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2867e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 28688654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 28698654d025Sperrin vd->vdev_deflate_ratio; 28708654d025Sperrin 28718654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 28728654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2873b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 28748654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 28758654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 28768654d025Sperrin 2877b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2878fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2879fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2880b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2881fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2882fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2883fa94a07fSbrendan } 2884b24ab676SJeff Bonwick 2885b24ab676SJeff Bonwick if (mc != NULL) { 2886b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2887b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2888b24ab676SJeff Bonwick 2889b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2890b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2891b24ab676SJeff Bonwick } 2892fa9e4066Sahrens } 2893fa9e4066Sahrens 2894fa9e4066Sahrens /* 2895fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2896fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2897fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2898fa9e4066Sahrens */ 2899fa9e4066Sahrens void 2900fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2901fa9e4066Sahrens { 2902fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2903fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2904fa9e4066Sahrens int c; 2905fa9e4066Sahrens 2906f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2907f9af39baSGeorge Wilson 2908c5904d13Seschrock /* 29096809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 29106809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2911c5904d13Seschrock */ 2912c5904d13Seschrock if (vd->vdev_aux != NULL) { 2913c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2914c5904d13Seschrock nvlist_t **aux; 2915c5904d13Seschrock uint_t naux; 2916c5904d13Seschrock 2917c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2918c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2919c5904d13Seschrock break; 2920c5904d13Seschrock } 2921c5904d13Seschrock 2922e14bb325SJeff Bonwick if (c == sav->sav_count) { 2923e14bb325SJeff Bonwick /* 2924e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2925e14bb325SJeff Bonwick */ 2926e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2927e14bb325SJeff Bonwick return; 2928e14bb325SJeff Bonwick } 2929e14bb325SJeff Bonwick 2930c5904d13Seschrock sav->sav_sync = B_TRUE; 2931c5904d13Seschrock 29326809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 29336809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 29346809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 29356809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 29366809eb4eSEric Schrock } 2937c5904d13Seschrock 2938c5904d13Seschrock ASSERT(c < naux); 2939c5904d13Seschrock 2940c5904d13Seschrock /* 2941c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2942c5904d13Seschrock * sketchy, but it will work. 2943c5904d13Seschrock */ 2944c5904d13Seschrock nvlist_free(aux[c]); 29453f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2946c5904d13Seschrock 2947c5904d13Seschrock return; 2948c5904d13Seschrock } 2949c5904d13Seschrock 29505dabedeeSbonwick /* 2951e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2952e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2953e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 29545dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 29555dabedeeSbonwick */ 2956e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2957e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2958e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29595dabedeeSbonwick 2960fa9e4066Sahrens if (vd == rvd) { 2961fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2962fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2963fa9e4066Sahrens } else { 2964fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2965fa9e4066Sahrens 296688ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 296788ecc943SGeorge Wilson !vd->vdev_ishole) 2968e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2969fa9e4066Sahrens } 2970fa9e4066Sahrens } 2971fa9e4066Sahrens 2972fa9e4066Sahrens void 2973fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2974fa9e4066Sahrens { 29755dabedeeSbonwick spa_t *spa = vd->vdev_spa; 29765dabedeeSbonwick 2977e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2978e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2979e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29805dabedeeSbonwick 2981e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2982e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2983e14bb325SJeff Bonwick } 2984e14bb325SJeff Bonwick 2985e14bb325SJeff Bonwick /* 2986e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2987e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2988e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2989e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2990e14bb325SJeff Bonwick */ 2991e14bb325SJeff Bonwick void 2992e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2993e14bb325SJeff Bonwick { 2994e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2995e14bb325SJeff Bonwick 2996f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2997e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2998e14bb325SJeff Bonwick 2999e14bb325SJeff Bonwick /* 3000e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3001e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3002e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3003e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3004e14bb325SJeff Bonwick */ 3005e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3006e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3007e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3008e14bb325SJeff Bonwick 3009b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 3010e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3011e14bb325SJeff Bonwick } 3012e14bb325SJeff Bonwick 3013e14bb325SJeff Bonwick void 3014e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3015e14bb325SJeff Bonwick { 3016e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3017e14bb325SJeff Bonwick 3018e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3019e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3020e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3021e14bb325SJeff Bonwick 3022e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3023e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3024fa9e4066Sahrens } 3025fa9e4066Sahrens 302632b87932Sek /* 302732b87932Sek * Propagate vdev state up from children to parent. 302832b87932Sek */ 302944cd46caSbillm void 303044cd46caSbillm vdev_propagate_state(vdev_t *vd) 303144cd46caSbillm { 30328ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 30338ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 303444cd46caSbillm int degraded = 0, faulted = 0; 303544cd46caSbillm int corrupted = 0; 303644cd46caSbillm vdev_t *child; 303744cd46caSbillm 30383d7072f8Seschrock if (vd->vdev_children > 0) { 3039573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 30403d7072f8Seschrock child = vd->vdev_child[c]; 304151ece835Seschrock 304288ecc943SGeorge Wilson /* 304388ecc943SGeorge Wilson * Don't factor holes into the decision. 304488ecc943SGeorge Wilson */ 304588ecc943SGeorge Wilson if (child->vdev_ishole) 304688ecc943SGeorge Wilson continue; 304788ecc943SGeorge Wilson 3048e14bb325SJeff Bonwick if (!vdev_readable(child) || 30498ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 305051ece835Seschrock /* 305151ece835Seschrock * Root special: if there is a top-level log 305251ece835Seschrock * device, treat the root vdev as if it were 305351ece835Seschrock * degraded. 305451ece835Seschrock */ 305551ece835Seschrock if (child->vdev_islog && vd == rvd) 305651ece835Seschrock degraded++; 305751ece835Seschrock else 305851ece835Seschrock faulted++; 305951ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 30603d7072f8Seschrock degraded++; 306151ece835Seschrock } 306244cd46caSbillm 30633d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 30643d7072f8Seschrock corrupted++; 30653d7072f8Seschrock } 306644cd46caSbillm 30673d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 30683d7072f8Seschrock 30693d7072f8Seschrock /* 3070e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 30713d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 30723d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 30733d7072f8Seschrock * replicas'. 30743d7072f8Seschrock */ 30753d7072f8Seschrock if (corrupted && vd == rvd && 30763d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 30773d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 30783d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 30793d7072f8Seschrock } 30803d7072f8Seschrock 308151ece835Seschrock if (vd->vdev_parent) 30823d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 308344cd46caSbillm } 308444cd46caSbillm 3085fa9e4066Sahrens /* 3086ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3087ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3088ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3089ea8dc4b6Seschrock * 3090ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3091ea8dc4b6Seschrock * generated. 3092fa9e4066Sahrens */ 3093fa9e4066Sahrens void 3094ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3095fa9e4066Sahrens { 3096560e6e96Seschrock uint64_t save_state; 3097c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3098ea8dc4b6Seschrock 3099ea8dc4b6Seschrock if (state == vd->vdev_state) { 3100ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3101fa9e4066Sahrens return; 3102ea8dc4b6Seschrock } 3103ea8dc4b6Seschrock 3104560e6e96Seschrock save_state = vd->vdev_state; 3105fa9e4066Sahrens 3106fa9e4066Sahrens vd->vdev_state = state; 3107fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3108fa9e4066Sahrens 31093d7072f8Seschrock /* 31103d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 311198d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 311298d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 311398d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 311498d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 311598d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 311698d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 311798d1cbfeSGeorge Wilson * children. 311898d1cbfeSGeorge Wilson */ 311998d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 312098d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 31213d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 31223d7072f8Seschrock 3123069f55e2SEric Schrock /* 3124069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3125069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3126069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3127069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3128069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3129069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3130069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3131069f55e2SEric Schrock */ 3132069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3133069f55e2SEric Schrock vd->vdev_prevstate != state) 3134069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3135069f55e2SEric Schrock 31363d7072f8Seschrock if (vd->vdev_removed && 31373d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 31383d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 31393d7072f8Seschrock /* 31403d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 31413d7072f8Seschrock * device was previously marked removed and someone attempted to 31423d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 31433d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 31443d7072f8Seschrock * it is one of our special test online cases, which is only 31453d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 31463d7072f8Seschrock * fault. 31473d7072f8Seschrock */ 31483d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 31493d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 31503d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 31513d7072f8Seschrock vd->vdev_removed = B_TRUE; 31523d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3153ea8dc4b6Seschrock /* 3154cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3155cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3156cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3157cb04b873SMark J Musante * is not considered an error. 3158ea8dc4b6Seschrock */ 3159cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3160cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3161560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3162560e6e96Seschrock vd->vdev_not_present = 1; 3163560e6e96Seschrock 3164560e6e96Seschrock /* 3165560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3166560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3167560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3168560e6e96Seschrock * want to post the ereport if the device was already in the 3169560e6e96Seschrock * CANT_OPEN state beforehand. 31703d7072f8Seschrock * 31713d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 31723d7072f8Seschrock * online the device in response to an insertion event. If we 31733d7072f8Seschrock * hit this case, then we have detected an insertion event for a 31743d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 31753d7072f8Seschrock * In this scenario, we don't post an ereport because we are 31763d7072f8Seschrock * about to replace the device, or attempt an online with 31773d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3178560e6e96Seschrock */ 31793d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 31803d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3181c5904d13Seschrock vd != spa->spa_root_vdev) { 3182ea8dc4b6Seschrock const char *class; 3183ea8dc4b6Seschrock 3184ea8dc4b6Seschrock switch (aux) { 3185ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3186ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3187ea8dc4b6Seschrock break; 3188ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3189ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3190ea8dc4b6Seschrock break; 3191ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3192ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3193ea8dc4b6Seschrock break; 3194ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3195ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3196ea8dc4b6Seschrock break; 3197ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3198ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3199ea8dc4b6Seschrock break; 3200ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3201ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3202ea8dc4b6Seschrock break; 3203ea8dc4b6Seschrock default: 3204ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3205ea8dc4b6Seschrock } 3206ea8dc4b6Seschrock 3207c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3208ea8dc4b6Seschrock } 3209ea8dc4b6Seschrock 32103d7072f8Seschrock /* Erase any notion of persistent removed state */ 32113d7072f8Seschrock vd->vdev_removed = B_FALSE; 32123d7072f8Seschrock } else { 32133d7072f8Seschrock vd->vdev_removed = B_FALSE; 32143d7072f8Seschrock } 3215ea8dc4b6Seschrock 32168b33d774STim Haley if (!isopen && vd->vdev_parent) 32178b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3218fa9e4066Sahrens } 321915e6edf1Sgw 322015e6edf1Sgw /* 322115e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 322215e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 322315e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 322415e6edf1Sgw * can be wholedisks. 322515e6edf1Sgw */ 322615e6edf1Sgw boolean_t 322715e6edf1Sgw vdev_is_bootable(vdev_t *vd) 322815e6edf1Sgw { 322915e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 323015e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 323115e6edf1Sgw 323215e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 323315e6edf1Sgw vd->vdev_children > 1) { 323415e6edf1Sgw return (B_FALSE); 323515e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 323615e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 323715e6edf1Sgw return (B_FALSE); 323815e6edf1Sgw } 323915e6edf1Sgw } 324015e6edf1Sgw 3241573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 324215e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 324315e6edf1Sgw return (B_FALSE); 324415e6edf1Sgw } 324515e6edf1Sgw return (B_TRUE); 324615e6edf1Sgw } 3247e6ca193dSGeorge Wilson 324888ecc943SGeorge Wilson /* 324988ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 325088ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 32514b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 32524b964adaSGeorge Wilson * device in the current vdev tree (nvd). 325388ecc943SGeorge Wilson */ 3254e6ca193dSGeorge Wilson void 325588ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3256e6ca193dSGeorge Wilson { 325788ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3258e6ca193dSGeorge Wilson 32594b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 326088ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 326188ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3262e6ca193dSGeorge Wilson 326388ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 326488ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3265e6ca193dSGeorge Wilson 32664b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3267e6ca193dSGeorge Wilson /* 32684b964adaSGeorge Wilson * Restore the persistent vdev state 3269e6ca193dSGeorge Wilson */ 327088ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 32714b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 32724b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 32734b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3274e6ca193dSGeorge Wilson } 3275e6ca193dSGeorge Wilson } 3276573ca77eSGeorge Wilson 32774b964adaSGeorge Wilson /* 32784b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 32794b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 32804b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 32814b964adaSGeorge Wilson */ 32824b964adaSGeorge Wilson boolean_t 32834b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 32844b964adaSGeorge Wilson { 32854b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 32864b964adaSGeorge Wilson !vd->vdev_removed) 32874b964adaSGeorge Wilson return (B_TRUE); 32884b964adaSGeorge Wilson 32894b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 32904b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 32914b964adaSGeorge Wilson return (B_TRUE); 32924b964adaSGeorge Wilson 32934b964adaSGeorge Wilson return (B_FALSE); 32944b964adaSGeorge Wilson } 32954b964adaSGeorge Wilson 3296573ca77eSGeorge Wilson /* 3297573ca77eSGeorge Wilson * Expand a vdev if possible. 3298573ca77eSGeorge Wilson */ 3299573ca77eSGeorge Wilson void 3300573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3301573ca77eSGeorge Wilson { 3302573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3303573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3304573ca77eSGeorge Wilson 3305573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3306573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3307573ca77eSGeorge Wilson vdev_config_dirty(vd); 3308573ca77eSGeorge Wilson } 3309573ca77eSGeorge Wilson } 33101195e687SMark J Musante 33111195e687SMark J Musante /* 33121195e687SMark J Musante * Split a vdev. 33131195e687SMark J Musante */ 33141195e687SMark J Musante void 33151195e687SMark J Musante vdev_split(vdev_t *vd) 33161195e687SMark J Musante { 33171195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 33181195e687SMark J Musante 33191195e687SMark J Musante vdev_remove_child(pvd, vd); 33201195e687SMark J Musante vdev_compact_children(pvd); 33211195e687SMark J Musante 33221195e687SMark J Musante cvd = pvd->vdev_child[0]; 33231195e687SMark J Musante if (pvd->vdev_children == 1) { 33241195e687SMark J Musante vdev_remove_parent(cvd); 33251195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 33261195e687SMark J Musante } 33271195e687SMark J Musante vdev_propagate_state(cvd); 33281195e687SMark J Musante } 3329283b8460SGeorge.Wilson 3330283b8460SGeorge.Wilson void 3331283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3332283b8460SGeorge.Wilson { 3333283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3334283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3335283b8460SGeorge.Wilson 3336283b8460SGeorge.Wilson vdev_deadman(cvd); 3337283b8460SGeorge.Wilson } 3338283b8460SGeorge.Wilson 3339283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3340283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3341283b8460SGeorge.Wilson 3342283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 334369962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3344283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3345283b8460SGeorge.Wilson zio_t *fio; 3346283b8460SGeorge.Wilson uint64_t delta; 3347283b8460SGeorge.Wilson 3348283b8460SGeorge.Wilson /* 3349283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3350283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3351283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3352283b8460SGeorge.Wilson */ 335369962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3354c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3355c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 3356c55e05cbSMatthew Ahrens zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3357c55e05cbSMatthew Ahrens "delta %lluns, last io %lluns", 3358283b8460SGeorge.Wilson fio->io_timestamp, delta, 3359283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3360283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3361283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3362283b8460SGeorge.Wilson } 3363283b8460SGeorge.Wilson } 3364283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3365283b8460SGeorge.Wilson } 3366283b8460SGeorge.Wilson } 3367