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. 2481cd5c55SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 251a902ef8SHans Rosenfeld * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27*c8811bd3SToomas Soome * Copyright 2016 Toomas Soome <tsoome@me.com> 28fa9e4066Sahrens */ 29fa9e4066Sahrens 30fa9e4066Sahrens #include <sys/zfs_context.h> 31ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 32fa9e4066Sahrens #include <sys/spa.h> 33fa9e4066Sahrens #include <sys/spa_impl.h> 34fa9e4066Sahrens #include <sys/dmu.h> 35fa9e4066Sahrens #include <sys/dmu_tx.h> 36fa9e4066Sahrens #include <sys/vdev_impl.h> 37fa9e4066Sahrens #include <sys/uberblock_impl.h> 38fa9e4066Sahrens #include <sys/metaslab.h> 39fa9e4066Sahrens #include <sys/metaslab_impl.h> 40fa9e4066Sahrens #include <sys/space_map.h> 410713e232SGeorge Wilson #include <sys/space_reftree.h> 42fa9e4066Sahrens #include <sys/zio.h> 43fa9e4066Sahrens #include <sys/zap.h> 44fa9e4066Sahrens #include <sys/fs/zfs.h> 45c5904d13Seschrock #include <sys/arc.h> 46e6ca193dSGeorge Wilson #include <sys/zil.h> 473f9d6ad7SLin Ling #include <sys/dsl_scan.h> 48fa9e4066Sahrens 49fa9e4066Sahrens /* 50fa9e4066Sahrens * Virtual device management. 51fa9e4066Sahrens */ 52fa9e4066Sahrens 53fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 54fa9e4066Sahrens &vdev_root_ops, 55fa9e4066Sahrens &vdev_raidz_ops, 56fa9e4066Sahrens &vdev_mirror_ops, 57fa9e4066Sahrens &vdev_replacing_ops, 5899653d4eSeschrock &vdev_spare_ops, 59fa9e4066Sahrens &vdev_disk_ops, 60fa9e4066Sahrens &vdev_file_ops, 61fa9e4066Sahrens &vdev_missing_ops, 6288ecc943SGeorge Wilson &vdev_hole_ops, 63fa9e4066Sahrens NULL 64fa9e4066Sahrens }; 65fa9e4066Sahrens 66088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 67088f3894Sahrens int zfs_scrub_limit = 10; 6805b2b3b8Smishra 69bf3e216cSMatthew Ahrens /* 70bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 71bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 72bf3e216cSMatthew Ahrens */ 73bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 74bf3e216cSMatthew Ahrens 75fa9e4066Sahrens /* 76fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 77fa9e4066Sahrens */ 78fa9e4066Sahrens static vdev_ops_t * 79fa9e4066Sahrens vdev_getops(const char *type) 80fa9e4066Sahrens { 81fa9e4066Sahrens vdev_ops_t *ops, **opspp; 82fa9e4066Sahrens 83fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 84fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 85fa9e4066Sahrens break; 86fa9e4066Sahrens 87fa9e4066Sahrens return (ops); 88fa9e4066Sahrens } 89fa9e4066Sahrens 90fa9e4066Sahrens /* 91fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 92fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 93fa9e4066Sahrens */ 94fa9e4066Sahrens uint64_t 95fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 96fa9e4066Sahrens { 97ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 98fa9e4066Sahrens uint64_t csize; 99fa9e4066Sahrens 100573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 101fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 102fa9e4066Sahrens asize = MAX(asize, csize); 103fa9e4066Sahrens } 104fa9e4066Sahrens 105fa9e4066Sahrens return (asize); 106fa9e4066Sahrens } 107fa9e4066Sahrens 1082a79c5feSlling /* 109573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 110573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 111573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 112573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1132a79c5feSlling */ 1142a79c5feSlling uint64_t 115573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1162a79c5feSlling { 117573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1182a79c5feSlling 119573ca77eSGeorge Wilson /* 1204263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 121573ca77eSGeorge Wilson * just return our own asize. 122573ca77eSGeorge Wilson */ 123573ca77eSGeorge Wilson if (pvd == NULL) 124573ca77eSGeorge Wilson return (vd->vdev_asize); 1252a79c5feSlling 1262a79c5feSlling /* 127573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 128573ca77eSGeorge Wilson * to the nearest metaslab. 1292a79c5feSlling */ 130573ca77eSGeorge Wilson if (vd == vd->vdev_top) 131573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1322a79c5feSlling 133573ca77eSGeorge Wilson /* 134573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 135573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 136573ca77eSGeorge Wilson */ 137573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 138573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1392a79c5feSlling 140573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 141573ca77eSGeorge Wilson } 1422a79c5feSlling 143573ca77eSGeorge Wilson void 144573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 145573ca77eSGeorge Wilson { 146573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 147573ca77eSGeorge Wilson 148573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 149573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1502a79c5feSlling } 1512a79c5feSlling 152fa9e4066Sahrens vdev_t * 153fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 154fa9e4066Sahrens { 155fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 156fa9e4066Sahrens 157e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 158e05725b1Sbonwick 159088f3894Sahrens if (vdev < rvd->vdev_children) { 160088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 161fa9e4066Sahrens return (rvd->vdev_child[vdev]); 162088f3894Sahrens } 163fa9e4066Sahrens 164fa9e4066Sahrens return (NULL); 165fa9e4066Sahrens } 166fa9e4066Sahrens 167fa9e4066Sahrens vdev_t * 168fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 169fa9e4066Sahrens { 170fa9e4066Sahrens vdev_t *mvd; 171fa9e4066Sahrens 1720e34b6a7Sbonwick if (vd->vdev_guid == guid) 173fa9e4066Sahrens return (vd); 174fa9e4066Sahrens 175573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 176fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 177fa9e4066Sahrens NULL) 178fa9e4066Sahrens return (mvd); 179fa9e4066Sahrens 180fa9e4066Sahrens return (NULL); 181fa9e4066Sahrens } 182fa9e4066Sahrens 18312380e1eSArne Jansen static int 18412380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 18512380e1eSArne Jansen { 18612380e1eSArne Jansen int n = 0; 18712380e1eSArne Jansen 18812380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 18912380e1eSArne Jansen return (1); 19012380e1eSArne Jansen 19112380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 19212380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 19312380e1eSArne Jansen 19412380e1eSArne Jansen return (n); 19512380e1eSArne Jansen } 19612380e1eSArne Jansen 19712380e1eSArne Jansen int 19812380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 19912380e1eSArne Jansen { 20012380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 20112380e1eSArne Jansen } 20212380e1eSArne Jansen 203fa9e4066Sahrens void 204fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 205fa9e4066Sahrens { 206fa9e4066Sahrens size_t oldsize, newsize; 207fa9e4066Sahrens uint64_t id = cvd->vdev_id; 208fa9e4066Sahrens vdev_t **newchild; 20981cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 210fa9e4066Sahrens 21181cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 212fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 213fa9e4066Sahrens 214fa9e4066Sahrens cvd->vdev_parent = pvd; 215fa9e4066Sahrens 216fa9e4066Sahrens if (pvd == NULL) 217fa9e4066Sahrens return; 218fa9e4066Sahrens 219fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 220fa9e4066Sahrens 221fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 222fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 223fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 224fa9e4066Sahrens 225fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 226fa9e4066Sahrens if (pvd->vdev_child != NULL) { 227fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 228fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 229fa9e4066Sahrens } 230fa9e4066Sahrens 231fa9e4066Sahrens pvd->vdev_child = newchild; 232fa9e4066Sahrens pvd->vdev_child[id] = cvd; 233fa9e4066Sahrens 234fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 235fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 236fa9e4066Sahrens 237fa9e4066Sahrens /* 238fa9e4066Sahrens * Walk up all ancestors to update guid sum. 239fa9e4066Sahrens */ 240fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 241fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 242fa9e4066Sahrens } 243fa9e4066Sahrens 244fa9e4066Sahrens void 245fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 246fa9e4066Sahrens { 247fa9e4066Sahrens int c; 248fa9e4066Sahrens uint_t id = cvd->vdev_id; 249fa9e4066Sahrens 250fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 251fa9e4066Sahrens 252fa9e4066Sahrens if (pvd == NULL) 253fa9e4066Sahrens return; 254fa9e4066Sahrens 255fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 256fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 257fa9e4066Sahrens 258fa9e4066Sahrens pvd->vdev_child[id] = NULL; 259fa9e4066Sahrens cvd->vdev_parent = NULL; 260fa9e4066Sahrens 261fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 262fa9e4066Sahrens if (pvd->vdev_child[c]) 263fa9e4066Sahrens break; 264fa9e4066Sahrens 265fa9e4066Sahrens if (c == pvd->vdev_children) { 266fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 267fa9e4066Sahrens pvd->vdev_child = NULL; 268fa9e4066Sahrens pvd->vdev_children = 0; 269fa9e4066Sahrens } 270fa9e4066Sahrens 271fa9e4066Sahrens /* 272fa9e4066Sahrens * Walk up all ancestors to update guid sum. 273fa9e4066Sahrens */ 274fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 275fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 276fa9e4066Sahrens } 277fa9e4066Sahrens 278fa9e4066Sahrens /* 279fa9e4066Sahrens * Remove any holes in the child array. 280fa9e4066Sahrens */ 281fa9e4066Sahrens void 282fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 283fa9e4066Sahrens { 284fa9e4066Sahrens vdev_t **newchild, *cvd; 285fa9e4066Sahrens int oldc = pvd->vdev_children; 286573ca77eSGeorge Wilson int newc; 287fa9e4066Sahrens 288e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 289fa9e4066Sahrens 290573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 291fa9e4066Sahrens if (pvd->vdev_child[c]) 292fa9e4066Sahrens newc++; 293fa9e4066Sahrens 294fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 295fa9e4066Sahrens 296573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 297fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 298fa9e4066Sahrens newchild[newc] = cvd; 299fa9e4066Sahrens cvd->vdev_id = newc++; 300fa9e4066Sahrens } 301fa9e4066Sahrens } 302fa9e4066Sahrens 303fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 304fa9e4066Sahrens pvd->vdev_child = newchild; 305fa9e4066Sahrens pvd->vdev_children = newc; 306fa9e4066Sahrens } 307fa9e4066Sahrens 308fa9e4066Sahrens /* 309fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 310fa9e4066Sahrens */ 31188ecc943SGeorge Wilson vdev_t * 312fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 313fa9e4066Sahrens { 314fa9e4066Sahrens vdev_t *vd; 315fa9e4066Sahrens 316fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 317fa9e4066Sahrens 3180e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 3190e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 3200e34b6a7Sbonwick spa->spa_root_vdev = vd; 321e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 3220e34b6a7Sbonwick } 3230e34b6a7Sbonwick 32488ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 3250e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 3260e34b6a7Sbonwick /* 3270e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3280e34b6a7Sbonwick * which must be unique among all pools. 3290e34b6a7Sbonwick */ 3301195e687SMark J Musante guid = spa_generate_guid(NULL); 3310e34b6a7Sbonwick } else { 3320e34b6a7Sbonwick /* 3330e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3340e34b6a7Sbonwick */ 3351195e687SMark J Musante guid = spa_generate_guid(spa); 3360e34b6a7Sbonwick } 3370e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3380e34b6a7Sbonwick } 3390e34b6a7Sbonwick 340fa9e4066Sahrens vd->vdev_spa = spa; 341fa9e4066Sahrens vd->vdev_id = id; 342fa9e4066Sahrens vd->vdev_guid = guid; 343fa9e4066Sahrens vd->vdev_guid_sum = guid; 344fa9e4066Sahrens vd->vdev_ops = ops; 345fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 34688ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 347fa9e4066Sahrens 348fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3495ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 350e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3510f7643c7SGeorge Wilson mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); 3528ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3530713e232SGeorge Wilson vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 3548ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3558ad4d6ddSJeff Bonwick } 356fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 357fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 358fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 359fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 360fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3613d7072f8Seschrock vdev_queue_init(vd); 3623d7072f8Seschrock vdev_cache_init(vd); 363fa9e4066Sahrens 364fa9e4066Sahrens return (vd); 365fa9e4066Sahrens } 366fa9e4066Sahrens 367fa9e4066Sahrens /* 368fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 369fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 370fa9e4066Sahrens * different for each case. 371fa9e4066Sahrens */ 37299653d4eSeschrock int 37399653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 37499653d4eSeschrock int alloctype) 375fa9e4066Sahrens { 376fa9e4066Sahrens vdev_ops_t *ops; 377fa9e4066Sahrens char *type; 3788654d025Sperrin uint64_t guid = 0, islog, nparity; 379fa9e4066Sahrens vdev_t *vd; 380fa9e4066Sahrens 381e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 382fa9e4066Sahrens 383fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 384be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 385fa9e4066Sahrens 386fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 387be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 388fa9e4066Sahrens 389fa9e4066Sahrens /* 390fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 391fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 392fa9e4066Sahrens */ 393fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 394fa9e4066Sahrens uint64_t label_id; 395fa9e4066Sahrens 396fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 397fa9e4066Sahrens label_id != id) 398be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 399fa9e4066Sahrens 400fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 401be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40299653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 40399653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 404be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 405fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 406fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 407be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40821ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 40921ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 410be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 411fa9e4066Sahrens } 412fa9e4066Sahrens 41399653d4eSeschrock /* 41499653d4eSeschrock * The first allocated vdev must be of type 'root'. 41599653d4eSeschrock */ 41699653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 417be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 41899653d4eSeschrock 4198654d025Sperrin /* 4208654d025Sperrin * Determine whether we're a log vdev. 4218654d025Sperrin */ 4228654d025Sperrin islog = 0; 4238654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 424990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 425be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 426fa9e4066Sahrens 42788ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 428be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 42988ecc943SGeorge Wilson 43099653d4eSeschrock /* 4318654d025Sperrin * Set the nparity property for RAID-Z vdevs. 43299653d4eSeschrock */ 4338654d025Sperrin nparity = -1ULL; 43499653d4eSeschrock if (ops == &vdev_raidz_ops) { 43599653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4368654d025Sperrin &nparity) == 0) { 437b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 438be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 43999653d4eSeschrock /* 440f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 441f94275ceSAdam Leventhal * device. 44299653d4eSeschrock */ 443f94275ceSAdam Leventhal if (nparity > 1 && 444f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 445be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 446f94275ceSAdam Leventhal if (nparity > 2 && 447f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 448be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 44999653d4eSeschrock } else { 45099653d4eSeschrock /* 45199653d4eSeschrock * We require the parity to be specified for SPAs that 45299653d4eSeschrock * support multiple parity levels. 45399653d4eSeschrock */ 454f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 455be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 45699653d4eSeschrock /* 45799653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 45899653d4eSeschrock */ 4598654d025Sperrin nparity = 1; 46099653d4eSeschrock } 46199653d4eSeschrock } else { 4628654d025Sperrin nparity = 0; 46399653d4eSeschrock } 4648654d025Sperrin ASSERT(nparity != -1ULL); 4658654d025Sperrin 4668654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4678654d025Sperrin 4688654d025Sperrin vd->vdev_islog = islog; 4698654d025Sperrin vd->vdev_nparity = nparity; 4708654d025Sperrin 4718654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4728654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4738654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4748654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4758654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4768654d025Sperrin &vd->vdev_physpath) == 0) 4778654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4786809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4796809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 48099653d4eSeschrock 481afefbcddSeschrock /* 482afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 483afefbcddSeschrock * as -1. 484afefbcddSeschrock */ 485afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 486afefbcddSeschrock &vd->vdev_wholedisk) != 0) 487afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 488afefbcddSeschrock 489ea8dc4b6Seschrock /* 490ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 491ea8dc4b6Seschrock * was not present at the time of import. 492ea8dc4b6Seschrock */ 4936809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4946809eb4eSEric Schrock &vd->vdev_not_present); 495ea8dc4b6Seschrock 496ecc2d604Sbonwick /* 497ecc2d604Sbonwick * Get the alignment requirement. 498ecc2d604Sbonwick */ 499ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 500ecc2d604Sbonwick 50188ecc943SGeorge Wilson /* 50288ecc943SGeorge Wilson * Retrieve the vdev creation time. 50388ecc943SGeorge Wilson */ 50488ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 50588ecc943SGeorge Wilson &vd->vdev_crtxg); 50688ecc943SGeorge Wilson 507fa9e4066Sahrens /* 508fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 509fa9e4066Sahrens */ 5101195e687SMark J Musante if (parent && !parent->vdev_parent && 5111195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 512fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 513fa9e4066Sahrens &vd->vdev_ms_array); 514fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 515fa9e4066Sahrens &vd->vdev_ms_shift); 516fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 517fa9e4066Sahrens &vd->vdev_asize); 5183f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 5193f9d6ad7SLin Ling &vd->vdev_removing); 520215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 521215198a6SJoe Stein &vd->vdev_top_zap); 522215198a6SJoe Stein } else { 523215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 524fa9e4066Sahrens } 525fa9e4066Sahrens 526cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 527a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 5289f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 5291195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 5309f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 531a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 532a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 533a1521560SJeff Bonwick } 534a1521560SJeff Bonwick 535215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 536215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 537215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 538215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 539215198a6SJoe Stein } else { 540215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 541215198a6SJoe Stein } 542215198a6SJoe Stein 543fa9e4066Sahrens /* 5443d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 545fa9e4066Sahrens */ 546215198a6SJoe Stein 547c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 54821ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 54921ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 550c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 551c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5520713e232SGeorge Wilson &vd->vdev_dtl_object); 553c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 554c5904d13Seschrock &vd->vdev_unspare); 555c5904d13Seschrock } 55621ecdf64SLin Ling 55721ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 55821ecdf64SLin Ling uint64_t spare = 0; 55921ecdf64SLin Ling 56021ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 56121ecdf64SLin Ling &spare) == 0 && spare) 56221ecdf64SLin Ling spa_spare_add(vd); 56321ecdf64SLin Ling } 56421ecdf64SLin Ling 565ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 566ecc2d604Sbonwick &vd->vdev_offline); 567c5904d13Seschrock 568b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 569b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 570cb04b873SMark J Musante 5713d7072f8Seschrock /* 5723d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5733d7072f8Seschrock * state, as the diagnosis made on another system may not be 574069f55e2SEric Schrock * valid in the current context. Local vdevs will 575069f55e2SEric Schrock * remain in the faulted state. 5763d7072f8Seschrock */ 577b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5783d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5793d7072f8Seschrock &vd->vdev_faulted); 5803d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5813d7072f8Seschrock &vd->vdev_degraded); 5823d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5833d7072f8Seschrock &vd->vdev_removed); 584069f55e2SEric Schrock 585069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 586069f55e2SEric Schrock char *aux; 587069f55e2SEric Schrock 588069f55e2SEric Schrock vd->vdev_label_aux = 589069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 590069f55e2SEric Schrock if (nvlist_lookup_string(nv, 591069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 592069f55e2SEric Schrock strcmp(aux, "external") == 0) 593069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 594069f55e2SEric Schrock } 5953d7072f8Seschrock } 596fa9e4066Sahrens } 597fa9e4066Sahrens 598fa9e4066Sahrens /* 599fa9e4066Sahrens * Add ourselves to the parent's list of children. 600fa9e4066Sahrens */ 601fa9e4066Sahrens vdev_add_child(parent, vd); 602fa9e4066Sahrens 60399653d4eSeschrock *vdp = vd; 60499653d4eSeschrock 60599653d4eSeschrock return (0); 606fa9e4066Sahrens } 607fa9e4066Sahrens 608fa9e4066Sahrens void 609fa9e4066Sahrens vdev_free(vdev_t *vd) 610fa9e4066Sahrens { 6113d7072f8Seschrock spa_t *spa = vd->vdev_spa; 612fa9e4066Sahrens 613fa9e4066Sahrens /* 614fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 615fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 616fa9e4066Sahrens */ 617fa9e4066Sahrens vdev_close(vd); 618fa9e4066Sahrens 619e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 620b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 621fa9e4066Sahrens 622fa9e4066Sahrens /* 623fa9e4066Sahrens * Free all children. 624fa9e4066Sahrens */ 625573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 626fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 627fa9e4066Sahrens 628fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 629fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 630fa9e4066Sahrens 631fa9e4066Sahrens /* 632fa9e4066Sahrens * Discard allocation state. 633fa9e4066Sahrens */ 634a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 635fa9e4066Sahrens vdev_metaslab_fini(vd); 636a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 637a1521560SJeff Bonwick } 638fa9e4066Sahrens 639fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 640fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 641fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 642fa9e4066Sahrens 643fa9e4066Sahrens /* 644fa9e4066Sahrens * Remove this vdev from its parent's child list. 645fa9e4066Sahrens */ 646fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 647fa9e4066Sahrens 648fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 649fa9e4066Sahrens 6503d7072f8Seschrock /* 6513d7072f8Seschrock * Clean up vdev structure. 6523d7072f8Seschrock */ 6533d7072f8Seschrock vdev_queue_fini(vd); 6543d7072f8Seschrock vdev_cache_fini(vd); 6553d7072f8Seschrock 6563d7072f8Seschrock if (vd->vdev_path) 6573d7072f8Seschrock spa_strfree(vd->vdev_path); 6583d7072f8Seschrock if (vd->vdev_devid) 6593d7072f8Seschrock spa_strfree(vd->vdev_devid); 6603d7072f8Seschrock if (vd->vdev_physpath) 6613d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6626809eb4eSEric Schrock if (vd->vdev_fru) 6636809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6643d7072f8Seschrock 6653d7072f8Seschrock if (vd->vdev_isspare) 6663d7072f8Seschrock spa_spare_remove(vd); 667fa94a07fSbrendan if (vd->vdev_isl2cache) 668fa94a07fSbrendan spa_l2cache_remove(vd); 6693d7072f8Seschrock 6703d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6713d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6728ad4d6ddSJeff Bonwick 6733d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6740713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 6758ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6760713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 6770713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 6788ad4d6ddSJeff Bonwick } 6793d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6808ad4d6ddSJeff Bonwick 6810f7643c7SGeorge Wilson mutex_destroy(&vd->vdev_queue_lock); 6823d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6833d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 684e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6853d7072f8Seschrock 6863d7072f8Seschrock if (vd == spa->spa_root_vdev) 6873d7072f8Seschrock spa->spa_root_vdev = NULL; 6883d7072f8Seschrock 6893d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 690fa9e4066Sahrens } 691fa9e4066Sahrens 692fa9e4066Sahrens /* 693fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 694fa9e4066Sahrens */ 695fa9e4066Sahrens static void 696fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 697fa9e4066Sahrens { 698fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 699fa9e4066Sahrens metaslab_t *msp; 700fa9e4066Sahrens vdev_t *vd; 701fa9e4066Sahrens int t; 702fa9e4066Sahrens 703fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 704fa9e4066Sahrens 705fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 706fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 707fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 708215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 709fa9e4066Sahrens 710fa9e4066Sahrens svd->vdev_ms_array = 0; 711fa9e4066Sahrens svd->vdev_ms_shift = 0; 712fa9e4066Sahrens svd->vdev_ms_count = 0; 713215198a6SJoe Stein svd->vdev_top_zap = 0; 714fa9e4066Sahrens 715cd0837ccSGeorge Wilson if (tvd->vdev_mg) 716cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 717fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 718fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 719fa9e4066Sahrens 720fa9e4066Sahrens svd->vdev_mg = NULL; 721fa9e4066Sahrens svd->vdev_ms = NULL; 722ecc2d604Sbonwick 723ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 724ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 725fa9e4066Sahrens 726fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 727fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 72899653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 729fa9e4066Sahrens 730fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 731fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 73299653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 733fa9e4066Sahrens 734fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 735fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 736fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 737fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 738fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 739fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 740fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 741fa9e4066Sahrens } 742fa9e4066Sahrens 743e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 744fa9e4066Sahrens vdev_config_clean(svd); 745fa9e4066Sahrens vdev_config_dirty(tvd); 746fa9e4066Sahrens } 747fa9e4066Sahrens 748e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 749e14bb325SJeff Bonwick vdev_state_clean(svd); 750e14bb325SJeff Bonwick vdev_state_dirty(tvd); 751e14bb325SJeff Bonwick } 752e14bb325SJeff Bonwick 75399653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 75499653d4eSeschrock svd->vdev_deflate_ratio = 0; 7558654d025Sperrin 7568654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7578654d025Sperrin svd->vdev_islog = 0; 758fa9e4066Sahrens } 759fa9e4066Sahrens 760fa9e4066Sahrens static void 761fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 762fa9e4066Sahrens { 763fa9e4066Sahrens if (vd == NULL) 764fa9e4066Sahrens return; 765fa9e4066Sahrens 766fa9e4066Sahrens vd->vdev_top = tvd; 767fa9e4066Sahrens 768573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 769fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 770fa9e4066Sahrens } 771fa9e4066Sahrens 772fa9e4066Sahrens /* 773fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 774fa9e4066Sahrens */ 775fa9e4066Sahrens vdev_t * 776fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 777fa9e4066Sahrens { 778fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 779fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 780fa9e4066Sahrens vdev_t *mvd; 781fa9e4066Sahrens 782e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 783fa9e4066Sahrens 784fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 785ecc2d604Sbonwick 786ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 787573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 7884263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 789ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 790ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 79188ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 792ecc2d604Sbonwick 793fa9e4066Sahrens vdev_remove_child(pvd, cvd); 794fa9e4066Sahrens vdev_add_child(pvd, mvd); 795fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 796fa9e4066Sahrens vdev_add_child(mvd, cvd); 797fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 798fa9e4066Sahrens 799fa9e4066Sahrens if (mvd == mvd->vdev_top) 800fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 801fa9e4066Sahrens 802fa9e4066Sahrens return (mvd); 803fa9e4066Sahrens } 804fa9e4066Sahrens 805fa9e4066Sahrens /* 806fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 807fa9e4066Sahrens */ 808fa9e4066Sahrens void 809fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 810fa9e4066Sahrens { 811fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 812fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 813fa9e4066Sahrens 814e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 815fa9e4066Sahrens 816fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 817fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 81899653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 81999653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 820ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 821fa9e4066Sahrens 822fa9e4066Sahrens vdev_remove_child(mvd, cvd); 823fa9e4066Sahrens vdev_remove_child(pvd, mvd); 8248ad4d6ddSJeff Bonwick 82599653d4eSeschrock /* 826e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 827e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 828e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 829e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 83099653d4eSeschrock */ 8318ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 8328ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 8331195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 8348ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 8358ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 8368ad4d6ddSJeff Bonwick } 837e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 838e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 839fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 840fa9e4066Sahrens 841fa9e4066Sahrens if (cvd == cvd->vdev_top) 842fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 843fa9e4066Sahrens 844fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 845fa9e4066Sahrens vdev_free(mvd); 846fa9e4066Sahrens } 847fa9e4066Sahrens 848ea8dc4b6Seschrock int 849fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 850fa9e4066Sahrens { 851fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 852ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 853ecc2d604Sbonwick uint64_t m; 854fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 855fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 856ecc2d604Sbonwick metaslab_t **mspp; 857ecc2d604Sbonwick int error; 858fa9e4066Sahrens 859a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 860a1521560SJeff Bonwick 86188ecc943SGeorge Wilson /* 86288ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 86388ecc943SGeorge Wilson */ 86488ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8650e34b6a7Sbonwick return (0); 8660e34b6a7Sbonwick 86788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 86888ecc943SGeorge Wilson 869e6ca193dSGeorge Wilson /* 870e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 871b5152584SMatthew Ahrens * in 128k (1 << 17) because it is the "typical" blocksize. 872b5152584SMatthew Ahrens * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 873b5152584SMatthew Ahrens * otherwise it would inconsistently account for existing bp's. 874e6ca193dSGeorge Wilson */ 875e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 876e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 877e6ca193dSGeorge Wilson 878fa9e4066Sahrens ASSERT(oldc <= newc); 879fa9e4066Sahrens 880ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 881fa9e4066Sahrens 882ecc2d604Sbonwick if (oldc != 0) { 883ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 884ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 885ecc2d604Sbonwick } 886fa9e4066Sahrens 887ecc2d604Sbonwick vd->vdev_ms = mspp; 888ecc2d604Sbonwick vd->vdev_ms_count = newc; 889fa9e4066Sahrens 890ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 8910713e232SGeorge Wilson uint64_t object = 0; 8920713e232SGeorge Wilson 893ecc2d604Sbonwick if (txg == 0) { 894ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8957bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8967bfdf011SNeil Perrin DMU_READ_PREFETCH); 897ecc2d604Sbonwick if (error) 898ecc2d604Sbonwick return (error); 899fa9e4066Sahrens } 9001e9bd7ecSPrakash Surya 9011e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 9021e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 9031e9bd7ecSPrakash Surya if (error) 9041e9bd7ecSPrakash Surya return (error); 905fa9e4066Sahrens } 906fa9e4066Sahrens 907a1521560SJeff Bonwick if (txg == 0) 908a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 909a1521560SJeff Bonwick 9103f9d6ad7SLin Ling /* 9113f9d6ad7SLin Ling * If the vdev is being removed we don't activate 9123f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 9133f9d6ad7SLin Ling * allocations are performed on this device. 9143f9d6ad7SLin Ling */ 9153f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 916a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 917a1521560SJeff Bonwick 918a1521560SJeff Bonwick if (txg == 0) 919a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 920a1521560SJeff Bonwick 921ea8dc4b6Seschrock return (0); 922fa9e4066Sahrens } 923fa9e4066Sahrens 924fa9e4066Sahrens void 925fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 926fa9e4066Sahrens { 927fa9e4066Sahrens uint64_t m; 928fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 929fa9e4066Sahrens 930fa9e4066Sahrens if (vd->vdev_ms != NULL) { 931a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 9320713e232SGeorge Wilson for (m = 0; m < count; m++) { 9330713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 9340713e232SGeorge Wilson 9350713e232SGeorge Wilson if (msp != NULL) 9360713e232SGeorge Wilson metaslab_fini(msp); 9370713e232SGeorge Wilson } 938fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 939fa9e4066Sahrens vd->vdev_ms = NULL; 940fa9e4066Sahrens } 941fa9e4066Sahrens } 942fa9e4066Sahrens 943e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 944e14bb325SJeff Bonwick boolean_t vps_readable; 945e14bb325SJeff Bonwick boolean_t vps_writeable; 946e14bb325SJeff Bonwick int vps_flags; 947e14bb325SJeff Bonwick } vdev_probe_stats_t; 948e14bb325SJeff Bonwick 949e14bb325SJeff Bonwick static void 950e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9510a4e9518Sgw { 9528ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 953a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 954e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 955a3f829aeSBill Moore 956a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 957e14bb325SJeff Bonwick 958e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 959e14bb325SJeff Bonwick if (zio->io_error == 0) 960e14bb325SJeff Bonwick vps->vps_readable = 1; 9618ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 962a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 963e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 964e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 965e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 966e14bb325SJeff Bonwick } else { 967e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 968e14bb325SJeff Bonwick } 969e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 970e14bb325SJeff Bonwick if (zio->io_error == 0) 971e14bb325SJeff Bonwick vps->vps_writeable = 1; 972e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 973e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 974a3f829aeSBill Moore zio_t *pio; 975e14bb325SJeff Bonwick 976e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 977e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 978e14bb325SJeff Bonwick 979e14bb325SJeff Bonwick if (vdev_readable(vd) && 9808ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 981e14bb325SJeff Bonwick zio->io_error = 0; 982e14bb325SJeff Bonwick } else { 983e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 984e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9858ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 986be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 987e14bb325SJeff Bonwick } 988a3f829aeSBill Moore 989a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 990a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 991a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 992a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 993a3f829aeSBill Moore 9940f7643c7SGeorge Wilson zio_link_t *zl = NULL; 9950f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 996a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 997be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 998a3f829aeSBill Moore 999e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1000e14bb325SJeff Bonwick } 1001e14bb325SJeff Bonwick } 10020a4e9518Sgw 1003e14bb325SJeff Bonwick /* 1004f7170741SWill Andrews * Determine whether this device is accessible. 1005f7170741SWill Andrews * 1006f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1007f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1008f7170741SWill Andrews * a VTOC. 1009e14bb325SJeff Bonwick */ 1010e14bb325SJeff Bonwick zio_t * 1011a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1012e14bb325SJeff Bonwick { 1013e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1014a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1015a3f829aeSBill Moore zio_t *pio; 1016a3f829aeSBill Moore 1017a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 10180a4e9518Sgw 1019a3f829aeSBill Moore /* 1020a3f829aeSBill Moore * Don't probe the probe. 1021a3f829aeSBill Moore */ 1022a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1023a3f829aeSBill Moore return (NULL); 1024e14bb325SJeff Bonwick 1025a3f829aeSBill Moore /* 1026a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1027a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1028a3f829aeSBill Moore * this vdev will become parents of the probe io. 1029a3f829aeSBill Moore */ 1030a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1031e14bb325SJeff Bonwick 1032a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1033a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1034a3f829aeSBill Moore 1035a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1036a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 10378956713aSEric Schrock ZIO_FLAG_TRYHARD; 1038a3f829aeSBill Moore 1039a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1040a3f829aeSBill Moore /* 1041a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1042a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1043a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1044a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1045a3f829aeSBill Moore * any zio looking at these values can assume that 1046a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1047a3f829aeSBill Moore * important because when a device has intermittent 1048a3f829aeSBill Moore * connectivity problems, we want to ensure that 1049a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1050a3f829aeSBill Moore * the zio (EIO). 1051a3f829aeSBill Moore * 1052a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1053a3f829aeSBill Moore * values so the probe can reevaluate from first 1054a3f829aeSBill Moore * principles. 1055a3f829aeSBill Moore */ 1056a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1057a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1058a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1059a3f829aeSBill Moore } 1060a3f829aeSBill Moore 1061a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1062a3f829aeSBill Moore vdev_probe_done, vps, 1063a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1064a3f829aeSBill Moore 106598d1cbfeSGeorge Wilson /* 106698d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 106798d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 106898d1cbfeSGeorge Wilson */ 1069a3f829aeSBill Moore if (zio != NULL) { 1070a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1071a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1072a3f829aeSBill Moore } 1073e14bb325SJeff Bonwick } 1074e14bb325SJeff Bonwick 1075a3f829aeSBill Moore if (zio != NULL) 1076a3f829aeSBill Moore zio_add_child(zio, pio); 1077e14bb325SJeff Bonwick 1078a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1079e14bb325SJeff Bonwick 1080a3f829aeSBill Moore if (vps == NULL) { 1081a3f829aeSBill Moore ASSERT(zio != NULL); 1082a3f829aeSBill Moore return (NULL); 1083a3f829aeSBill Moore } 1084e14bb325SJeff Bonwick 1085e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1086a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1087e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1088f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1089f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1090e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1091e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1092e14bb325SJeff Bonwick } 1093e14bb325SJeff Bonwick 1094a3f829aeSBill Moore if (zio == NULL) 1095a3f829aeSBill Moore return (pio); 1096a3f829aeSBill Moore 1097a3f829aeSBill Moore zio_nowait(pio); 1098a3f829aeSBill Moore return (NULL); 10990a4e9518Sgw } 11000a4e9518Sgw 1101f64c0e34SEric Taylor static void 1102f64c0e34SEric Taylor vdev_open_child(void *arg) 1103f64c0e34SEric Taylor { 1104f64c0e34SEric Taylor vdev_t *vd = arg; 1105f64c0e34SEric Taylor 1106f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1107f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1108f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1109f64c0e34SEric Taylor } 1110f64c0e34SEric Taylor 1111681d9761SEric Taylor boolean_t 1112681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1113681d9761SEric Taylor { 1114681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1115681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1116681d9761SEric Taylor return (B_TRUE); 1117681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1118681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1119681d9761SEric Taylor return (B_TRUE); 1120681d9761SEric Taylor return (B_FALSE); 1121681d9761SEric Taylor } 1122681d9761SEric Taylor 1123f64c0e34SEric Taylor void 1124f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1125f64c0e34SEric Taylor { 1126f64c0e34SEric Taylor taskq_t *tq; 1127f64c0e34SEric Taylor int children = vd->vdev_children; 1128f64c0e34SEric Taylor 1129681d9761SEric Taylor /* 1130681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1131681d9761SEric Taylor * in a single thread so that the same thread holds the 1132681d9761SEric Taylor * spa_namespace_lock 1133681d9761SEric Taylor */ 1134681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1135681d9761SEric Taylor for (int c = 0; c < children; c++) 1136681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1137681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1138681d9761SEric Taylor return; 1139681d9761SEric Taylor } 1140f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1141f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1142f64c0e34SEric Taylor 1143f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1144f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1145f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1146f64c0e34SEric Taylor 1147f64c0e34SEric Taylor taskq_destroy(tq); 1148f64c0e34SEric Taylor } 1149f64c0e34SEric Taylor 1150fa9e4066Sahrens /* 1151fa9e4066Sahrens * Prepare a virtual device for access. 1152fa9e4066Sahrens */ 1153fa9e4066Sahrens int 1154fa9e4066Sahrens vdev_open(vdev_t *vd) 1155fa9e4066Sahrens { 11568ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1157fa9e4066Sahrens int error; 1158fa9e4066Sahrens uint64_t osize = 0; 11594263d13fSGeorge Wilson uint64_t max_osize = 0; 11604263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1161ecc2d604Sbonwick uint64_t ashift = 0; 1162fa9e4066Sahrens 1163f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1164f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1165fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1166fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1167fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1168fa9e4066Sahrens 1169fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1170e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1171e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1172573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1173fa9e4066Sahrens 1174069f55e2SEric Schrock /* 1175069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1176069f55e2SEric Schrock * faulted, bail out of the open. 1177069f55e2SEric Schrock */ 11783d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11793d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1180069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1181069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11823d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1183069f55e2SEric Schrock vd->vdev_label_aux); 1184be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 11853d7072f8Seschrock } else if (vd->vdev_offline) { 1186fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1187ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1188be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1189fa9e4066Sahrens } 1190fa9e4066Sahrens 11914263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1192fa9e4066Sahrens 1193095bcd66SGeorge Wilson /* 1194095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1195095bcd66SGeorge Wilson * the vdev on error. 1196095bcd66SGeorge Wilson */ 1197095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1198ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11998956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1200ea8dc4b6Seschrock 1201fa9e4066Sahrens if (error) { 12023d7072f8Seschrock if (vd->vdev_removed && 12033d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 12043d7072f8Seschrock vd->vdev_removed = B_FALSE; 12053d7072f8Seschrock 1206ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1207fa9e4066Sahrens vd->vdev_stat.vs_aux); 1208fa9e4066Sahrens return (error); 1209fa9e4066Sahrens } 1210fa9e4066Sahrens 12113d7072f8Seschrock vd->vdev_removed = B_FALSE; 12123d7072f8Seschrock 1213096d22d4SEric Schrock /* 1214096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1215096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1216096d22d4SEric Schrock */ 1217096d22d4SEric Schrock if (vd->vdev_faulted) { 1218096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1219096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1220096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1221096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1222096d22d4SEric Schrock vd->vdev_label_aux); 1223be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1224096d22d4SEric Schrock } 1225096d22d4SEric Schrock 12263d7072f8Seschrock if (vd->vdev_degraded) { 12273d7072f8Seschrock ASSERT(vd->vdev_children == 0); 12283d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 12293d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 12303d7072f8Seschrock } else { 1231069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 12323d7072f8Seschrock } 1233fa9e4066Sahrens 123488ecc943SGeorge Wilson /* 123588ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 123688ecc943SGeorge Wilson */ 123788ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 123888ecc943SGeorge Wilson return (0); 123988ecc943SGeorge Wilson 1240573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1241ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1242ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1243ea8dc4b6Seschrock VDEV_AUX_NONE); 1244ea8dc4b6Seschrock break; 1245ea8dc4b6Seschrock } 1246573ca77eSGeorge Wilson } 1247fa9e4066Sahrens 1248fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 12494263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1250fa9e4066Sahrens 1251fa9e4066Sahrens if (vd->vdev_children == 0) { 1252fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1253ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1254ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1255be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1256fa9e4066Sahrens } 1257fa9e4066Sahrens psize = osize; 1258fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 12594263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 12604263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1261fa9e4066Sahrens } else { 1262ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1263fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1264ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1265ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1266be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1267fa9e4066Sahrens } 1268fa9e4066Sahrens psize = 0; 1269fa9e4066Sahrens asize = osize; 12704263d13fSGeorge Wilson max_asize = max_osize; 1271fa9e4066Sahrens } 1272fa9e4066Sahrens 1273fa9e4066Sahrens vd->vdev_psize = psize; 1274fa9e4066Sahrens 1275573ca77eSGeorge Wilson /* 1276573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1277573ca77eSGeorge Wilson */ 1278573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1279573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1280573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1281be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1282573ca77eSGeorge Wilson } 1283573ca77eSGeorge Wilson 1284fa9e4066Sahrens if (vd->vdev_asize == 0) { 1285fa9e4066Sahrens /* 1286fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1287ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1288fa9e4066Sahrens */ 1289fa9e4066Sahrens vd->vdev_asize = asize; 12904263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1291ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1292fa9e4066Sahrens } else { 1293fa9e4066Sahrens /* 12942384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 12952384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 12962384d9f8SGeorge Wilson * issue a warning instead. 1297fa9e4066Sahrens */ 12982384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 12992384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 13002384d9f8SGeorge Wilson cmn_err(CE_WARN, 13012384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 13022384d9f8SGeorge Wilson "larger than the pool's alignment\n", 13032384d9f8SGeorge Wilson vd->vdev_path); 1304fa9e4066Sahrens } 13054263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1306573ca77eSGeorge Wilson } 1307fa9e4066Sahrens 1308573ca77eSGeorge Wilson /* 1309573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1310573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1311573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1312573ca77eSGeorge Wilson */ 1313573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1314573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1315573ca77eSGeorge Wilson vd->vdev_asize = asize; 1316fa9e4066Sahrens 1317573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1318fa9e4066Sahrens 13190a4e9518Sgw /* 13200a4e9518Sgw * Ensure we can issue some IO before declaring the 13210a4e9518Sgw * vdev open for business. 13220a4e9518Sgw */ 1323e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1324e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 132598d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 132698d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 13270a4e9518Sgw return (error); 13280a4e9518Sgw } 13290a4e9518Sgw 133081cd5c55SMatthew Ahrens /* 133181cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 133281cd5c55SMatthew Ahrens */ 133381cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 133481cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 133581cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 133681cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 133781cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 133881cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 133981cd5c55SMatthew Ahrens } 134081cd5c55SMatthew Ahrens 1341088f3894Sahrens /* 1342088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 13438ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 13448ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1345088f3894Sahrens */ 13468ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 13478ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 13488ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1349088f3894Sahrens 1350fa9e4066Sahrens return (0); 1351fa9e4066Sahrens } 1352fa9e4066Sahrens 1353560e6e96Seschrock /* 1354560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1355560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 13563d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1357560e6e96Seschrock * 1358d7f601efSGeorge Wilson * If 'strict' is false ignore the spa guid check. This is necessary because 1359d7f601efSGeorge Wilson * if the machine crashed during a re-guid the new guid might have been written 1360d7f601efSGeorge Wilson * to all of the vdev labels, but not the cached config. The strict check 1361d7f601efSGeorge Wilson * will be performed when the pool is opened again using the mos config. 1362d7f601efSGeorge Wilson * 1363560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1364560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1365560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1366560e6e96Seschrock * will be updated but the function will return 0. 1367560e6e96Seschrock */ 1368560e6e96Seschrock int 1369d7f601efSGeorge Wilson vdev_validate(vdev_t *vd, boolean_t strict) 1370560e6e96Seschrock { 1371560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1372560e6e96Seschrock nvlist_t *label; 13731195e687SMark J Musante uint64_t guid = 0, top_guid; 1374560e6e96Seschrock uint64_t state; 1375560e6e96Seschrock 1376573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1377d7f601efSGeorge Wilson if (vdev_validate(vd->vdev_child[c], strict) != 0) 1378be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1379560e6e96Seschrock 1380b5989ec7Seschrock /* 1381b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1382b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1383b5989ec7Seschrock * overwrite the previous state. 1384b5989ec7Seschrock */ 1385e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13861195e687SMark J Musante uint64_t aux_guid = 0; 13871195e687SMark J Musante nvlist_t *nvl; 1388bda88194SGeorge Wilson uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1389bda88194SGeorge Wilson spa_last_synced_txg(spa) : -1ULL; 1390560e6e96Seschrock 1391dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1392560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1393560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1394560e6e96Seschrock return (0); 1395560e6e96Seschrock } 1396560e6e96Seschrock 13971195e687SMark J Musante /* 13981195e687SMark J Musante * Determine if this vdev has been split off into another 13991195e687SMark J Musante * pool. If so, then refuse to open it. 14001195e687SMark J Musante */ 14011195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 14021195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 14031195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 14041195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 14051195e687SMark J Musante nvlist_free(label); 14061195e687SMark J Musante return (0); 14071195e687SMark J Musante } 14081195e687SMark J Musante 1409d7f601efSGeorge Wilson if (strict && (nvlist_lookup_uint64(label, 1410d7f601efSGeorge Wilson ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1411d7f601efSGeorge Wilson guid != spa_guid(spa))) { 1412560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1413560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1414560e6e96Seschrock nvlist_free(label); 1415560e6e96Seschrock return (0); 1416560e6e96Seschrock } 1417560e6e96Seschrock 14181195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 14191195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 14201195e687SMark J Musante &aux_guid) != 0) 14211195e687SMark J Musante aux_guid = 0; 14221195e687SMark J Musante 1423e14bb325SJeff Bonwick /* 1424e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1425e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1426e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1427e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1428e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1429e14bb325SJeff Bonwick * safely compare to that instead. 14301195e687SMark J Musante * 14311195e687SMark J Musante * If we split this vdev off instead, then we also check the 14321195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 14331195e687SMark J Musante * corrupt if it is partway through a split operation. 1434e14bb325SJeff Bonwick */ 1435560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1436e14bb325SJeff Bonwick &guid) != 0 || 1437e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1438e14bb325SJeff Bonwick &top_guid) != 0 || 14391195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1440e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1441560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1442560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1443560e6e96Seschrock nvlist_free(label); 1444560e6e96Seschrock return (0); 1445560e6e96Seschrock } 1446560e6e96Seschrock 1447560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1448560e6e96Seschrock &state) != 0) { 1449560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1450560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1451560e6e96Seschrock nvlist_free(label); 1452560e6e96Seschrock return (0); 1453560e6e96Seschrock } 1454560e6e96Seschrock 1455560e6e96Seschrock nvlist_free(label); 1456560e6e96Seschrock 1457bc758434SLin Ling /* 14584b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1459bc758434SLin Ling * state of the pool. 1460bc758434SLin Ling */ 14614b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1462b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1463bc758434SLin Ling state != POOL_STATE_ACTIVE) 1464be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1465560e6e96Seschrock 146651ece835Seschrock /* 146751ece835Seschrock * If we were able to open and validate a vdev that was 146851ece835Seschrock * previously marked permanently unavailable, clear that state 146951ece835Seschrock * now. 147051ece835Seschrock */ 147151ece835Seschrock if (vd->vdev_not_present) 147251ece835Seschrock vd->vdev_not_present = 0; 147351ece835Seschrock } 1474560e6e96Seschrock 1475560e6e96Seschrock return (0); 1476560e6e96Seschrock } 1477560e6e96Seschrock 1478fa9e4066Sahrens /* 1479fa9e4066Sahrens * Close a virtual device. 1480fa9e4066Sahrens */ 1481fa9e4066Sahrens void 1482fa9e4066Sahrens vdev_close(vdev_t *vd) 1483fa9e4066Sahrens { 14848ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1485095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14868ad4d6ddSJeff Bonwick 14878ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14888ad4d6ddSJeff Bonwick 14891195e687SMark J Musante /* 14901195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14911195e687SMark J Musante * going offline. 14921195e687SMark J Musante */ 1493095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14941195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1495095bcd66SGeorge Wilson 1496fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1497fa9e4066Sahrens 14983d7072f8Seschrock vdev_cache_purge(vd); 1499fa9e4066Sahrens 1500560e6e96Seschrock /* 1501573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1502560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1503560e6e96Seschrock * it's still faulted. 1504560e6e96Seschrock */ 1505560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1506560e6e96Seschrock 1507fa9e4066Sahrens if (vd->vdev_offline) 1508fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1509fa9e4066Sahrens else 1510fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1511ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1512fa9e4066Sahrens } 1513fa9e4066Sahrens 1514dcba9f3fSGeorge Wilson void 1515dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1516dcba9f3fSGeorge Wilson { 1517dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1518dcba9f3fSGeorge Wilson 1519dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1520dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1521dcba9f3fSGeorge Wilson return; 1522dcba9f3fSGeorge Wilson 1523dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1524dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1525dcba9f3fSGeorge Wilson 1526dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1527dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1528dcba9f3fSGeorge Wilson } 1529dcba9f3fSGeorge Wilson 1530dcba9f3fSGeorge Wilson void 1531dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1532dcba9f3fSGeorge Wilson { 1533dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1534dcba9f3fSGeorge Wilson 1535dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1536dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1537dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1538dcba9f3fSGeorge Wilson 1539dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1540dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1541dcba9f3fSGeorge Wilson } 1542dcba9f3fSGeorge Wilson 1543095bcd66SGeorge Wilson /* 1544095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1545095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1546095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1547095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1548095bcd66SGeorge Wilson */ 1549fa9e4066Sahrens void 1550ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1551fa9e4066Sahrens { 1552ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1553fa9e4066Sahrens 1554e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1555ea8dc4b6Seschrock 15561195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 15571195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1558fa9e4066Sahrens vdev_close(vd); 1559fa9e4066Sahrens (void) vdev_open(vd); 1560fa9e4066Sahrens 156139c23413Seschrock /* 156239c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 156339c23413Seschrock * Otherwise, a device with an invalid label could be successfully 156439c23413Seschrock * opened in response to vdev_reopen(). 156539c23413Seschrock */ 1566c5904d13Seschrock if (vd->vdev_aux) { 1567c5904d13Seschrock (void) vdev_validate_aux(vd); 1568e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 15696809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1570573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1571573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1572c5904d13Seschrock } else { 1573bda88194SGeorge Wilson (void) vdev_validate(vd, B_TRUE); 1574c5904d13Seschrock } 157539c23413Seschrock 1576fa9e4066Sahrens /* 15773d7072f8Seschrock * Reassess parent vdev's health. 1578fa9e4066Sahrens */ 15793d7072f8Seschrock vdev_propagate_state(vd); 1580fa9e4066Sahrens } 1581fa9e4066Sahrens 1582fa9e4066Sahrens int 158399653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1584fa9e4066Sahrens { 1585fa9e4066Sahrens int error; 1586fa9e4066Sahrens 1587fa9e4066Sahrens /* 1588fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1589fa9e4066Sahrens * For a create, however, we want to fail the request if 1590fa9e4066Sahrens * there are any components we can't open. 1591fa9e4066Sahrens */ 1592fa9e4066Sahrens error = vdev_open(vd); 1593fa9e4066Sahrens 1594fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1595fa9e4066Sahrens vdev_close(vd); 1596fa9e4066Sahrens return (error ? error : ENXIO); 1597fa9e4066Sahrens } 1598fa9e4066Sahrens 1599fa9e4066Sahrens /* 16000713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1601fa9e4066Sahrens */ 16020713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 16030713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 160439c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1605fa9e4066Sahrens vdev_close(vd); 1606fa9e4066Sahrens return (error); 1607fa9e4066Sahrens } 1608fa9e4066Sahrens 1609fa9e4066Sahrens return (0); 1610fa9e4066Sahrens } 1611fa9e4066Sahrens 16120e34b6a7Sbonwick void 1613573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1614fa9e4066Sahrens { 1615fa9e4066Sahrens /* 1616bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1617fa9e4066Sahrens */ 1618bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1619fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1620fa9e4066Sahrens } 1621fa9e4066Sahrens 1622fa9e4066Sahrens void 1623ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1624fa9e4066Sahrens { 1625ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 162688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1627ecc2d604Sbonwick ASSERT(ISP2(flags)); 1628f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1629fa9e4066Sahrens 1630ecc2d604Sbonwick if (flags & VDD_METASLAB) 1631ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1632ecc2d604Sbonwick 1633ecc2d604Sbonwick if (flags & VDD_DTL) 1634ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1635ecc2d604Sbonwick 1636ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1637fa9e4066Sahrens } 1638fa9e4066Sahrens 16390713e232SGeorge Wilson void 16400713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 16410713e232SGeorge Wilson { 16420713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16430713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 16440713e232SGeorge Wilson 16450713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 16460713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 16470713e232SGeorge Wilson } 16480713e232SGeorge Wilson 16498ad4d6ddSJeff Bonwick /* 16508ad4d6ddSJeff Bonwick * DTLs. 16518ad4d6ddSJeff Bonwick * 16528ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 16539fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 16548ad4d6ddSJeff Bonwick * 16558ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 16568ad4d6ddSJeff Bonwick * 16578ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 16588ad4d6ddSJeff Bonwick * 16598ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 16608ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 16618ad4d6ddSJeff Bonwick * txgs that was scrubbed. 16628ad4d6ddSJeff Bonwick * 16638ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 16648ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 16658ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 16668ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 16678ad4d6ddSJeff Bonwick * determine whether a device can be detached. 16688ad4d6ddSJeff Bonwick * 16698ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 16708ad4d6ddSJeff Bonwick * either has the data or it doesn't. 16718ad4d6ddSJeff Bonwick * 16728ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 16738ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 16748ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 16758ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 16768ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 16778ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 16788ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 16798ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 16808ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 16818ad4d6ddSJeff Bonwick * 16828ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 16838ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 16848ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 16858ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 16868ad4d6ddSJeff Bonwick */ 1687fa9e4066Sahrens void 16888ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1689fa9e4066Sahrens { 16900713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16918ad4d6ddSJeff Bonwick 16928ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16938ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1694f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 16958ad4d6ddSJeff Bonwick 16960713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16970713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 16980713e232SGeorge Wilson range_tree_add(rt, txg, size); 16990713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1700fa9e4066Sahrens } 1701fa9e4066Sahrens 17028ad4d6ddSJeff Bonwick boolean_t 17038ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1704fa9e4066Sahrens { 17050713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17068ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1707fa9e4066Sahrens 17088ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 17098ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1710fa9e4066Sahrens 17110713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17120713e232SGeorge Wilson if (range_tree_space(rt) != 0) 17130713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 17140713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1715fa9e4066Sahrens 1716fa9e4066Sahrens return (dirty); 1717fa9e4066Sahrens } 1718fa9e4066Sahrens 17198ad4d6ddSJeff Bonwick boolean_t 17208ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 17218ad4d6ddSJeff Bonwick { 17220713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17238ad4d6ddSJeff Bonwick boolean_t empty; 17248ad4d6ddSJeff Bonwick 17250713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17260713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 17270713e232SGeorge Wilson mutex_exit(rt->rt_lock); 17288ad4d6ddSJeff Bonwick 17298ad4d6ddSJeff Bonwick return (empty); 17308ad4d6ddSJeff Bonwick } 17318ad4d6ddSJeff Bonwick 1732b4952e17SGeorge Wilson /* 1733b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 1734b4952e17SGeorge Wilson */ 1735b4952e17SGeorge Wilson static uint64_t 1736b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 1737b4952e17SGeorge Wilson { 17380713e232SGeorge Wilson range_seg_t *rs; 1739b4952e17SGeorge Wilson 1740b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17410713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1742b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1743b4952e17SGeorge Wilson 17440713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17450713e232SGeorge Wilson return (rs->rs_start - 1); 1746b4952e17SGeorge Wilson } 1747b4952e17SGeorge Wilson 1748b4952e17SGeorge Wilson /* 1749b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 1750b4952e17SGeorge Wilson */ 1751b4952e17SGeorge Wilson static uint64_t 1752b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 1753b4952e17SGeorge Wilson { 17540713e232SGeorge Wilson range_seg_t *rs; 1755b4952e17SGeorge Wilson 1756b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17570713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1758b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1759b4952e17SGeorge Wilson 17600713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17610713e232SGeorge Wilson return (rs->rs_end); 1762b4952e17SGeorge Wilson } 1763b4952e17SGeorge Wilson 1764b4952e17SGeorge Wilson /* 1765b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 1766b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 1767b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 1768b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 1769b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 1770b4952e17SGeorge Wilson * excise the DTLs. 1771b4952e17SGeorge Wilson */ 1772b4952e17SGeorge Wilson static boolean_t 1773b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 1774b4952e17SGeorge Wilson { 1775b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 1776b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1777b4952e17SGeorge Wilson 1778b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 1779b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1780b4952e17SGeorge Wilson 1781b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 17820713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1783b4952e17SGeorge Wilson return (B_TRUE); 1784b4952e17SGeorge Wilson 1785b4952e17SGeorge Wilson /* 1786b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 1787b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 1788b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 1789b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 1790b4952e17SGeorge Wilson * for excision. 1791b4952e17SGeorge Wilson */ 1792b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1793b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1794b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1795b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1796b4952e17SGeorge Wilson return (B_TRUE); 1797b4952e17SGeorge Wilson } 1798b4952e17SGeorge Wilson return (B_FALSE); 1799b4952e17SGeorge Wilson } 1800b4952e17SGeorge Wilson 1801fa9e4066Sahrens /* 1802fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1803fa9e4066Sahrens */ 1804fa9e4066Sahrens void 1805fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1806fa9e4066Sahrens { 1807ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 18088ad4d6ddSJeff Bonwick avl_tree_t reftree; 18098ad4d6ddSJeff Bonwick int minref; 1810fa9e4066Sahrens 18118ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1812fa9e4066Sahrens 18138ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 18148ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 18158ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 18168ad4d6ddSJeff Bonwick 1817b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 18188ad4d6ddSJeff Bonwick return; 18198ad4d6ddSJeff Bonwick 18208ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 18213f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 18223f9d6ad7SLin Ling 1823fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1824b4952e17SGeorge Wilson 1825b4952e17SGeorge Wilson /* 1826b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 1827b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 1828b4952e17SGeorge Wilson * excise regions on vdevs that were available during 1829b4952e17SGeorge Wilson * the entire duration of this scan. 1830b4952e17SGeorge Wilson */ 1831088f3894Sahrens if (scrub_txg != 0 && 18323f9d6ad7SLin Ling (spa->spa_scrub_started || 1833b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1834b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 1835088f3894Sahrens /* 1836088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1837088f3894Sahrens * did it without rebooting, then the scrub dtl 1838088f3894Sahrens * will be valid, so excise the old region and 1839088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1840088f3894Sahrens * dtl as-is if there was an error. 18418ad4d6ddSJeff Bonwick * 18428ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 18438ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 18448ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 18458ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 18468ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 18478ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 18488ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 18498ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 18508ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1851088f3894Sahrens */ 18520713e232SGeorge Wilson space_reftree_create(&reftree); 18530713e232SGeorge Wilson space_reftree_add_map(&reftree, 18540713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18550713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 18560713e232SGeorge Wilson space_reftree_add_map(&reftree, 18570713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 18580713e232SGeorge Wilson space_reftree_generate_map(&reftree, 18590713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18600713e232SGeorge Wilson space_reftree_destroy(&reftree); 1861fa9e4066Sahrens } 18620713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 18630713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18640713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1865fa9e4066Sahrens if (scrub_done) 18660713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 18670713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 18688ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 18690713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 18708ad4d6ddSJeff Bonwick else 18710713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18720713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1873b4952e17SGeorge Wilson 1874b4952e17SGeorge Wilson /* 1875b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 1876b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 1877b4952e17SGeorge Wilson */ 1878b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 18790713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 18800713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 1881b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 1882b4952e17SGeorge Wilson 1883fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1884088f3894Sahrens 1885ecc2d604Sbonwick if (txg != 0) 1886ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1887fa9e4066Sahrens return; 1888fa9e4066Sahrens } 1889fa9e4066Sahrens 1890fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 18918ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 189299bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 189399bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 18948ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 18958ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 18968ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 18978ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 18988ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 18998ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 19008ad4d6ddSJeff Bonwick else 19018ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 19020713e232SGeorge Wilson space_reftree_create(&reftree); 19038ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 19048ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 19058ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 19060713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 19078ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 19088ad4d6ddSJeff Bonwick } 19090713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 19100713e232SGeorge Wilson space_reftree_destroy(&reftree); 1911fa9e4066Sahrens } 19128ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1913fa9e4066Sahrens } 1914fa9e4066Sahrens 19150713e232SGeorge Wilson int 1916fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1917fa9e4066Sahrens { 1918fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1919ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19200713e232SGeorge Wilson int error = 0; 1921fa9e4066Sahrens 19220713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 19230713e232SGeorge Wilson ASSERT(!vd->vdev_ishole); 1924fa9e4066Sahrens 19250713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 19260713e232SGeorge Wilson vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 19270713e232SGeorge Wilson if (error) 19280713e232SGeorge Wilson return (error); 19290713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1930fa9e4066Sahrens 19310713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 193288ecc943SGeorge Wilson 19330713e232SGeorge Wilson /* 19340713e232SGeorge Wilson * Now that we've opened the space_map we need to update 19350713e232SGeorge Wilson * the in-core DTL. 19360713e232SGeorge Wilson */ 19370713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 1938ecc2d604Sbonwick 19390713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 19400713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 19410713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1942fa9e4066Sahrens 19430713e232SGeorge Wilson return (error); 19440713e232SGeorge Wilson } 19450713e232SGeorge Wilson 19460713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 19470713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 19480713e232SGeorge Wilson if (error != 0) 19490713e232SGeorge Wilson break; 19500713e232SGeorge Wilson } 1951fa9e4066Sahrens 1952fa9e4066Sahrens return (error); 1953fa9e4066Sahrens } 1954fa9e4066Sahrens 1955215198a6SJoe Stein void 1956215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 1957215198a6SJoe Stein { 1958215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 1959215198a6SJoe Stein 1960215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 1961215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 1962215198a6SJoe Stein zapobj, tx)); 1963215198a6SJoe Stein } 1964215198a6SJoe Stein 1965215198a6SJoe Stein uint64_t 1966215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 1967215198a6SJoe Stein { 1968215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 1969215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 1970215198a6SJoe Stein DMU_OT_NONE, 0, tx); 1971215198a6SJoe Stein 1972215198a6SJoe Stein ASSERT(zap != 0); 1973215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 1974215198a6SJoe Stein zap, tx)); 1975215198a6SJoe Stein 1976215198a6SJoe Stein return (zap); 1977215198a6SJoe Stein } 1978215198a6SJoe Stein 1979215198a6SJoe Stein void 1980215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 1981215198a6SJoe Stein { 1982215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 1983215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 1984215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 1985215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 1986215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 1987215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 1988215198a6SJoe Stein } 1989215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 1990215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 1991215198a6SJoe Stein } 1992215198a6SJoe Stein } 1993215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 1994215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 1995215198a6SJoe Stein } 1996215198a6SJoe Stein } 1997215198a6SJoe Stein 1998fa9e4066Sahrens void 1999fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2000fa9e4066Sahrens { 2001fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 20020713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2003ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 20040713e232SGeorge Wilson range_tree_t *rtsync; 20050713e232SGeorge Wilson kmutex_t rtlock; 2006fa9e4066Sahrens dmu_tx_t *tx; 20070713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2008fa9e4066Sahrens 200988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 20100713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 201188ecc943SGeorge Wilson 2012fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2013fa9e4066Sahrens 20140713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 20150713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20160713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 20170713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 20180713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 20190713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2020215198a6SJoe Stein 2021215198a6SJoe Stein /* 2022215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2023215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2024215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2025215198a6SJoe Stein */ 2026215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2027215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2028215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2029215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2030215198a6SJoe Stein } 2031215198a6SJoe Stein 2032fa9e4066Sahrens dmu_tx_commit(tx); 2033fa9e4066Sahrens return; 2034fa9e4066Sahrens } 2035fa9e4066Sahrens 20360713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 20370713e232SGeorge Wilson uint64_t new_object; 20380713e232SGeorge Wilson 20390713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 20400713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 20410713e232SGeorge Wilson 20420713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 20430713e232SGeorge Wilson 0, -1ULL, 0, &vd->vdev_dtl_lock)); 20440713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2045fa9e4066Sahrens } 2046fa9e4066Sahrens 20470713e232SGeorge Wilson mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 2048fa9e4066Sahrens 20490713e232SGeorge Wilson rtsync = range_tree_create(NULL, NULL, &rtlock); 2050fa9e4066Sahrens 20510713e232SGeorge Wilson mutex_enter(&rtlock); 2052fa9e4066Sahrens 2053fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 20540713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2055fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2056fa9e4066Sahrens 20570713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 20580713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 20590713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2060fa9e4066Sahrens 20610713e232SGeorge Wilson range_tree_destroy(rtsync); 2062fa9e4066Sahrens 20630713e232SGeorge Wilson mutex_exit(&rtlock); 20640713e232SGeorge Wilson mutex_destroy(&rtlock); 2065fa9e4066Sahrens 20660713e232SGeorge Wilson /* 20670713e232SGeorge Wilson * If the object for the space map has changed then dirty 20680713e232SGeorge Wilson * the top level so that we update the config. 20690713e232SGeorge Wilson */ 20700713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 20710713e232SGeorge Wilson zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 20720713e232SGeorge Wilson "new object %llu", txg, spa_name(spa), object, 20730713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 20740713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 20750713e232SGeorge Wilson } 2076fa9e4066Sahrens 2077fa9e4066Sahrens dmu_tx_commit(tx); 20780713e232SGeorge Wilson 20790713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20800713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 20810713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2082fa9e4066Sahrens } 2083fa9e4066Sahrens 20848ad4d6ddSJeff Bonwick /* 20858ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 20868ad4d6ddSJeff Bonwick * without losing data. 20878ad4d6ddSJeff Bonwick */ 20888ad4d6ddSJeff Bonwick boolean_t 20898ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 20908ad4d6ddSJeff Bonwick { 20918ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 20928ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 20938ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 20948ad4d6ddSJeff Bonwick boolean_t required; 20958ad4d6ddSJeff Bonwick 20968ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 20978ad4d6ddSJeff Bonwick 20988ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 20998ad4d6ddSJeff Bonwick return (B_TRUE); 21008ad4d6ddSJeff Bonwick 21018ad4d6ddSJeff Bonwick /* 21028ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 21038ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 21048ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 21058ad4d6ddSJeff Bonwick */ 21068ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 21078ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 21088ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 21098ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 21108ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 21118ad4d6ddSJeff Bonwick 2112cb04b873SMark J Musante if (!required && zio_injection_enabled) 2113cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2114cb04b873SMark J Musante 21158ad4d6ddSJeff Bonwick return (required); 21168ad4d6ddSJeff Bonwick } 21178ad4d6ddSJeff Bonwick 2118088f3894Sahrens /* 2119088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2120088f3894Sahrens */ 2121088f3894Sahrens boolean_t 2122088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2123088f3894Sahrens { 2124088f3894Sahrens boolean_t needed = B_FALSE; 2125088f3894Sahrens uint64_t thismin = UINT64_MAX; 2126088f3894Sahrens uint64_t thismax = 0; 2127088f3894Sahrens 2128088f3894Sahrens if (vd->vdev_children == 0) { 2129088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 21300713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 21318ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2132088f3894Sahrens 2133b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2134b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2135088f3894Sahrens needed = B_TRUE; 2136088f3894Sahrens } 2137088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2138088f3894Sahrens } else { 21398ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2140088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2141088f3894Sahrens uint64_t cmin, cmax; 2142088f3894Sahrens 2143088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2144088f3894Sahrens thismin = MIN(thismin, cmin); 2145088f3894Sahrens thismax = MAX(thismax, cmax); 2146088f3894Sahrens needed = B_TRUE; 2147088f3894Sahrens } 2148088f3894Sahrens } 2149088f3894Sahrens } 2150088f3894Sahrens 2151088f3894Sahrens if (needed && minp) { 2152088f3894Sahrens *minp = thismin; 2153088f3894Sahrens *maxp = thismax; 2154088f3894Sahrens } 2155088f3894Sahrens return (needed); 2156088f3894Sahrens } 2157088f3894Sahrens 2158560e6e96Seschrock void 2159ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2160fa9e4066Sahrens { 2161fa9e4066Sahrens /* 2162fa9e4066Sahrens * Recursively load all children. 2163fa9e4066Sahrens */ 21648ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2165560e6e96Seschrock vdev_load(vd->vdev_child[c]); 2166fa9e4066Sahrens 2167fa9e4066Sahrens /* 21680e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2169fa9e4066Sahrens */ 217088ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 2171560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2172560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 2173560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2174560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2175fa9e4066Sahrens 2176fa9e4066Sahrens /* 2177fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2178fa9e4066Sahrens */ 2179560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2180560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2181560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2182fa9e4066Sahrens } 2183fa9e4066Sahrens 218499653d4eSeschrock /* 2185fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2186fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2187fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2188fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2189fa94a07fSbrendan * repurposed to another pool. 219099653d4eSeschrock */ 219199653d4eSeschrock int 2192fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 219399653d4eSeschrock { 219499653d4eSeschrock nvlist_t *label; 219599653d4eSeschrock uint64_t guid, version; 219699653d4eSeschrock uint64_t state; 219799653d4eSeschrock 2198e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2199c5904d13Seschrock return (0); 2200c5904d13Seschrock 2201dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 220299653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 220399653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 220499653d4eSeschrock return (-1); 220599653d4eSeschrock } 220699653d4eSeschrock 220799653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2208ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 220999653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 221099653d4eSeschrock guid != vd->vdev_guid || 221199653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 221299653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 221399653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 221499653d4eSeschrock nvlist_free(label); 221599653d4eSeschrock return (-1); 221699653d4eSeschrock } 221799653d4eSeschrock 221899653d4eSeschrock /* 221999653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 222099653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 222199653d4eSeschrock */ 222299653d4eSeschrock nvlist_free(label); 222399653d4eSeschrock return (0); 222499653d4eSeschrock } 222599653d4eSeschrock 222688ecc943SGeorge Wilson void 222788ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 222888ecc943SGeorge Wilson { 222988ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 223088ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 223188ecc943SGeorge Wilson dmu_tx_t *tx; 223288ecc943SGeorge Wilson 223388ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2234215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 2235215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 223688ecc943SGeorge Wilson 223788ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 22382e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 22392e4c9986SGeorge Wilson 22402e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 22412e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 22422e4c9986SGeorge Wilson 224388ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 224488ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 224588ecc943SGeorge Wilson 22460713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 224788ecc943SGeorge Wilson continue; 224888ecc943SGeorge Wilson 22490713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 22502e4c9986SGeorge Wilson /* 22512e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 22522e4c9986SGeorge Wilson * was removed then the histogram accounting may 22532e4c9986SGeorge Wilson * not be accurate. Update the histogram information 22542e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 22552e4c9986SGeorge Wilson * and metaslab class are up-to-date. 22562e4c9986SGeorge Wilson */ 22572e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 22582e4c9986SGeorge Wilson 22590713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 22600713e232SGeorge Wilson space_map_free(msp->ms_sm, tx); 22610713e232SGeorge Wilson space_map_close(msp->ms_sm); 22620713e232SGeorge Wilson msp->ms_sm = NULL; 22630713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 226488ecc943SGeorge Wilson } 22652e4c9986SGeorge Wilson 22662e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 22672e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 22682e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 22692e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 22702e4c9986SGeorge Wilson 227188ecc943SGeorge Wilson } 227288ecc943SGeorge Wilson 227388ecc943SGeorge Wilson if (vd->vdev_ms_array) { 227488ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 227588ecc943SGeorge Wilson vd->vdev_ms_array = 0; 227688ecc943SGeorge Wilson } 2277215198a6SJoe Stein 2278215198a6SJoe Stein if (vd->vdev_islog && vd->vdev_top_zap != 0) { 2279215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 2280215198a6SJoe Stein vd->vdev_top_zap = 0; 2281215198a6SJoe Stein } 228288ecc943SGeorge Wilson dmu_tx_commit(tx); 228388ecc943SGeorge Wilson } 228488ecc943SGeorge Wilson 2285fa9e4066Sahrens void 2286fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2287fa9e4066Sahrens { 2288fa9e4066Sahrens metaslab_t *msp; 228980eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2290fa9e4066Sahrens 229188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 229288ecc943SGeorge Wilson 2293fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2294fa9e4066Sahrens metaslab_sync_done(msp, txg); 229580eb36f2SGeorge Wilson 229680eb36f2SGeorge Wilson if (reassess) 229780eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2298fa9e4066Sahrens } 2299fa9e4066Sahrens 2300fa9e4066Sahrens void 2301fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2302fa9e4066Sahrens { 2303fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2304fa9e4066Sahrens vdev_t *lvd; 2305fa9e4066Sahrens metaslab_t *msp; 2306ecc2d604Sbonwick dmu_tx_t *tx; 2307fa9e4066Sahrens 230888ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 230988ecc943SGeorge Wilson 2310ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2311ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2312ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2313ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2314ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2315ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2316ecc2d604Sbonwick vdev_config_dirty(vd); 2317ecc2d604Sbonwick dmu_tx_commit(tx); 2318ecc2d604Sbonwick } 2319fa9e4066Sahrens 23203f9d6ad7SLin Ling /* 23213f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 23223f9d6ad7SLin Ling */ 23233f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 232488ecc943SGeorge Wilson vdev_remove(vd, txg); 232588ecc943SGeorge Wilson 2326ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2327fa9e4066Sahrens metaslab_sync(msp, txg); 2328ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2329ecc2d604Sbonwick } 2330fa9e4066Sahrens 2331fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2332fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2333fa9e4066Sahrens 2334fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2335fa9e4066Sahrens } 2336fa9e4066Sahrens 2337fa9e4066Sahrens uint64_t 2338fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2339fa9e4066Sahrens { 2340fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2341fa9e4066Sahrens } 2342fa9e4066Sahrens 23433d7072f8Seschrock /* 23443d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 23453d7072f8Seschrock * not be opened, and no I/O is attempted. 23463d7072f8Seschrock */ 2347fa9e4066Sahrens int 2348069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2349fa9e4066Sahrens { 23504b964adaSGeorge Wilson vdev_t *vd, *tvd; 2351fa9e4066Sahrens 23528f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2353fa9e4066Sahrens 2354c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2355e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2356e14bb325SJeff Bonwick 23573d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2358e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2359fa9e4066Sahrens 23604b964adaSGeorge Wilson tvd = vd->vdev_top; 23614b964adaSGeorge Wilson 2362069f55e2SEric Schrock /* 2363069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2364069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2365069f55e2SEric Schrock * were faulted. 2366069f55e2SEric Schrock */ 2367069f55e2SEric Schrock vd->vdev_label_aux = aux; 2368069f55e2SEric Schrock 23693d7072f8Seschrock /* 23703d7072f8Seschrock * Faulted state takes precedence over degraded. 23713d7072f8Seschrock */ 237298d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 23733d7072f8Seschrock vd->vdev_faulted = 1ULL; 23743d7072f8Seschrock vd->vdev_degraded = 0ULL; 2375069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 23763d7072f8Seschrock 23773d7072f8Seschrock /* 2378c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2379c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 23803d7072f8Seschrock */ 23814b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 23823d7072f8Seschrock vd->vdev_degraded = 1ULL; 23833d7072f8Seschrock vd->vdev_faulted = 0ULL; 23843d7072f8Seschrock 23853d7072f8Seschrock /* 23863d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 23873d7072f8Seschrock * mark it degraded. 23883d7072f8Seschrock */ 23894b964adaSGeorge Wilson vdev_reopen(tvd); 23903d7072f8Seschrock 2391069f55e2SEric Schrock if (vdev_readable(vd)) 2392069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 23933d7072f8Seschrock } 23943d7072f8Seschrock 2395e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23963d7072f8Seschrock } 23973d7072f8Seschrock 23983d7072f8Seschrock /* 23993d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 24003d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 24013d7072f8Seschrock * as I/O is concerned. 24023d7072f8Seschrock */ 24033d7072f8Seschrock int 2404069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 24053d7072f8Seschrock { 2406c5904d13Seschrock vdev_t *vd; 24070a4e9518Sgw 24088f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 24093d7072f8Seschrock 2410c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2411e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2412e14bb325SJeff Bonwick 24130e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2414e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 24150e34b6a7Sbonwick 24163d7072f8Seschrock /* 24173d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 24183d7072f8Seschrock */ 2419e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2420e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 24213d7072f8Seschrock 24223d7072f8Seschrock vd->vdev_degraded = 1ULL; 24233d7072f8Seschrock if (!vdev_is_dead(vd)) 24243d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2425069f55e2SEric Schrock aux); 24263d7072f8Seschrock 2427e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 24283d7072f8Seschrock } 24293d7072f8Seschrock 24303d7072f8Seschrock /* 2431f7170741SWill Andrews * Online the given vdev. 2432f7170741SWill Andrews * 2433f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2434f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2435f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2436f7170741SWill Andrews * events are generated if the device fails to open. 24373d7072f8Seschrock */ 24383d7072f8Seschrock int 2439e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 24403d7072f8Seschrock { 2441573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 244214372834SHans Rosenfeld boolean_t postevent = B_FALSE; 24433d7072f8Seschrock 24448f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 24453d7072f8Seschrock 2446c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2447e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 24483d7072f8Seschrock 24493d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2450e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2451fa9e4066Sahrens 245214372834SHans Rosenfeld postevent = 245314372834SHans Rosenfeld (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ? 245414372834SHans Rosenfeld B_TRUE : B_FALSE; 245514372834SHans Rosenfeld 2456573ca77eSGeorge Wilson tvd = vd->vdev_top; 2457fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2458441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2459e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2460e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2461573ca77eSGeorge Wilson 2462573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2463573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2464573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2465573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2466573ca77eSGeorge Wilson } 2467573ca77eSGeorge Wilson 2468573ca77eSGeorge Wilson vdev_reopen(tvd); 24693d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 24703d7072f8Seschrock 2471573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2472573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2473573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2474573ca77eSGeorge Wilson } 2475573ca77eSGeorge Wilson 24763d7072f8Seschrock if (newstate) 24773d7072f8Seschrock *newstate = vd->vdev_state; 24783d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 24793d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 24803d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 24813d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 24823d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2483fa9e4066Sahrens 2484573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2485573ca77eSGeorge Wilson 2486573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2487573ca77eSGeorge Wilson if (vd->vdev_aux) 2488573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2489573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2490573ca77eSGeorge Wilson } 249114372834SHans Rosenfeld 249214372834SHans Rosenfeld if (postevent) 249314372834SHans Rosenfeld spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE); 249414372834SHans Rosenfeld 24958ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2496fa9e4066Sahrens } 2497fa9e4066Sahrens 2498a1521560SJeff Bonwick static int 2499a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2500fa9e4066Sahrens { 2501e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 25028f18d1faSGeorge Wilson int error = 0; 25038f18d1faSGeorge Wilson uint64_t generation; 25048f18d1faSGeorge Wilson metaslab_group_t *mg; 25050a4e9518Sgw 25068f18d1faSGeorge Wilson top: 25078f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2508fa9e4066Sahrens 2509c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2510e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2511fa9e4066Sahrens 25120e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2513e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 25140e34b6a7Sbonwick 2515e6ca193dSGeorge Wilson tvd = vd->vdev_top; 25168f18d1faSGeorge Wilson mg = tvd->vdev_mg; 25178f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2518e6ca193dSGeorge Wilson 2519fa9e4066Sahrens /* 2520ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2521fa9e4066Sahrens */ 2522ecc2d604Sbonwick if (!vd->vdev_offline) { 2523ecc2d604Sbonwick /* 25248ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2525e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2526e6ca193dSGeorge Wilson * expendable. 2527ecc2d604Sbonwick */ 2528e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2529e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2530e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2531fa9e4066Sahrens 25328f18d1faSGeorge Wilson /* 2533b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2534b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2535b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2536b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 25378f18d1faSGeorge Wilson */ 25388f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 25398f18d1faSGeorge Wilson /* 25408f18d1faSGeorge Wilson * Prevent any future allocations. 25418f18d1faSGeorge Wilson */ 2542a1521560SJeff Bonwick metaslab_group_passivate(mg); 25438f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 25448f18d1faSGeorge Wilson 25451195e687SMark J Musante error = spa_offline_log(spa); 25468f18d1faSGeorge Wilson 25478f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 25488f18d1faSGeorge Wilson 25498f18d1faSGeorge Wilson /* 25508f18d1faSGeorge Wilson * Check to see if the config has changed. 25518f18d1faSGeorge Wilson */ 25528f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2553a1521560SJeff Bonwick metaslab_group_activate(mg); 25548f18d1faSGeorge Wilson if (error) 25558f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 25568f18d1faSGeorge Wilson vd, error)); 25578f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 25588f18d1faSGeorge Wilson goto top; 25598f18d1faSGeorge Wilson } 2560fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 25618f18d1faSGeorge Wilson } 25628f18d1faSGeorge Wilson 2563ecc2d604Sbonwick /* 2564ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2565e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2566e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2567e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2568ecc2d604Sbonwick */ 2569ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2570e6ca193dSGeorge Wilson vdev_reopen(tvd); 2571e6ca193dSGeorge Wilson 2572e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2573e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2574ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2575e6ca193dSGeorge Wilson vdev_reopen(tvd); 2576e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2577ecc2d604Sbonwick } 25788f18d1faSGeorge Wilson 25798f18d1faSGeorge Wilson /* 25808f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 25818f18d1faSGeorge Wilson * once we online the device it's open for business. 25828f18d1faSGeorge Wilson */ 25838f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2584a1521560SJeff Bonwick metaslab_group_activate(mg); 2585fa9e4066Sahrens } 2586fa9e4066Sahrens 2587e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2588ecc2d604Sbonwick 25898f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2590fa9e4066Sahrens } 2591fa9e4066Sahrens 2592a1521560SJeff Bonwick int 2593a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2594a1521560SJeff Bonwick { 2595a1521560SJeff Bonwick int error; 2596a1521560SJeff Bonwick 2597a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2598a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2599a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2600a1521560SJeff Bonwick 2601a1521560SJeff Bonwick return (error); 2602a1521560SJeff Bonwick } 2603a1521560SJeff Bonwick 2604ea8dc4b6Seschrock /* 2605ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2606ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2607ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2608ea8dc4b6Seschrock */ 2609ea8dc4b6Seschrock void 2610e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2611fa9e4066Sahrens { 2612e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2613e14bb325SJeff Bonwick 2614e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2615fa9e4066Sahrens 2616ea8dc4b6Seschrock if (vd == NULL) 2617e14bb325SJeff Bonwick vd = rvd; 2618fa9e4066Sahrens 2619ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2620ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2621ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2622fa9e4066Sahrens 2623e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2624e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 26253d7072f8Seschrock 26263d7072f8Seschrock /* 26278a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 26288a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 26298a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 26308a79c1b5Sek * written out to disk. 26313d7072f8Seschrock */ 2632e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2633e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 26348a79c1b5Sek 2635096d22d4SEric Schrock /* 2636096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2637096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2638096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2639096d22d4SEric Schrock */ 2640096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2641096d22d4SEric Schrock 26424b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2643e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2644e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2645e14bb325SJeff Bonwick 2646f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 26473d7072f8Seschrock 2648096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2649096d22d4SEric Schrock 2650f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2651e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2652e14bb325SJeff Bonwick 2653e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2654bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 26553d7072f8Seschrock 26563d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 26573d7072f8Seschrock } 2658096d22d4SEric Schrock 2659096d22d4SEric Schrock /* 2660096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2661096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2662096d22d4SEric Schrock * done in response to the FMA fault. 2663096d22d4SEric Schrock */ 2664096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2665096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2666096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2667096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2668fa9e4066Sahrens } 2669fa9e4066Sahrens 2670e14bb325SJeff Bonwick boolean_t 2671e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 26720a4e9518Sgw { 267388ecc943SGeorge Wilson /* 267488ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 267588ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 267688ecc943SGeorge Wilson * these types of devices in the various code paths. 267788ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 267888ecc943SGeorge Wilson * before issuing I/O to them. 267988ecc943SGeorge Wilson */ 268088ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 268188ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 26820a4e9518Sgw } 26830a4e9518Sgw 2684e14bb325SJeff Bonwick boolean_t 2685e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 26860a4e9518Sgw { 2687e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 26880a4e9518Sgw } 26890a4e9518Sgw 2690e14bb325SJeff Bonwick boolean_t 2691e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2692fa9e4066Sahrens { 2693e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2694fa9e4066Sahrens } 2695fa9e4066Sahrens 2696a31e6787SGeorge Wilson boolean_t 2697a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2698a31e6787SGeorge Wilson { 26998ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 27008ad4d6ddSJeff Bonwick 2701a31e6787SGeorge Wilson /* 27028ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2703a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2704a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 27058ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 27068ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 27078ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2708a31e6787SGeorge Wilson */ 27098ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 27100f7643c7SGeorge Wilson !vd->vdev_cant_write && !vd->vdev_ishole && 27110f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 2712a31e6787SGeorge Wilson } 2713a31e6787SGeorge Wilson 2714e14bb325SJeff Bonwick boolean_t 2715e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2716fa9e4066Sahrens { 2717e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2718fa9e4066Sahrens 2719e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2720e14bb325SJeff Bonwick return (B_FALSE); 2721fa9e4066Sahrens 2722e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2723e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2724fa9e4066Sahrens 2725e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2726e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2727fa9e4066Sahrens 2728e14bb325SJeff Bonwick return (B_TRUE); 2729fa9e4066Sahrens } 2730fa9e4066Sahrens 2731fa9e4066Sahrens /* 2732fa9e4066Sahrens * Get statistics for the given vdev. 2733fa9e4066Sahrens */ 2734fa9e4066Sahrens void 2735fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2736fa9e4066Sahrens { 27372e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 27382e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 2739c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 27402e4c9986SGeorge Wilson 27412e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2742fa9e4066Sahrens 2743fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2744fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2745fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2746fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2747573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2748573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2749573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2750c39a2aaeSGeorge Wilson /* 2751c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 2752c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 2753c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 2754c39a2aaeSGeorge Wilson */ 2755c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 2756c39a2aaeSGeorge Wilson vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize, 2757c39a2aaeSGeorge Wilson 1ULL << tvd->vdev_ms_shift); 2758c39a2aaeSGeorge Wilson } 27592986efa8SAlex Reece if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 27602e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 27612986efa8SAlex Reece } 2762fa9e4066Sahrens 2763fa9e4066Sahrens /* 2764fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2765fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2766fa9e4066Sahrens */ 2767fa9e4066Sahrens if (vd == rvd) { 2768e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2769fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2770fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2771fa9e4066Sahrens 2772e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2773fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2774fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2775fa9e4066Sahrens } 27763f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2777fa9e4066Sahrens } 2778fa9e4066Sahrens } 27792e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 2780fa9e4066Sahrens } 2781fa9e4066Sahrens 2782fa94a07fSbrendan void 2783fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2784fa94a07fSbrendan { 2785fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2786fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2787fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2788fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2789fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2790fa94a07fSbrendan } 2791fa94a07fSbrendan 27923f9d6ad7SLin Ling void 27933f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 27943f9d6ad7SLin Ling { 27953f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 27963f9d6ad7SLin Ling 27973f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 27983f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 27993f9d6ad7SLin Ling 28003f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 28013f9d6ad7SLin Ling vs->vs_scan_processed = 0; 28023f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 28033f9d6ad7SLin Ling } 28043f9d6ad7SLin Ling 2805fa9e4066Sahrens void 2806e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2807fa9e4066Sahrens { 28088ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 28098ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2810e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2811fa9e4066Sahrens vdev_t *pvd; 2812fa9e4066Sahrens uint64_t txg = zio->io_txg; 2813fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2814fa9e4066Sahrens zio_type_t type = zio->io_type; 2815fa9e4066Sahrens int flags = zio->io_flags; 2816fa9e4066Sahrens 2817e14bb325SJeff Bonwick /* 2818e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2819e14bb325SJeff Bonwick */ 2820e14bb325SJeff Bonwick if (zio->io_gang_tree) 2821e14bb325SJeff Bonwick return; 2822e14bb325SJeff Bonwick 2823fa9e4066Sahrens if (zio->io_error == 0) { 2824e14bb325SJeff Bonwick /* 2825e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2826e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2827e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2828e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2829e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2830e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2831e14bb325SJeff Bonwick * remain zero, which is what we want.) 2832e14bb325SJeff Bonwick * 2833e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2834e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2835e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2836e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2837e14bb325SJeff Bonwick */ 2838e14bb325SJeff Bonwick if (vd == rvd) 2839e14bb325SJeff Bonwick return; 2840e14bb325SJeff Bonwick 2841e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 28428ad4d6ddSJeff Bonwick 28438ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 28448ad4d6ddSJeff Bonwick return; 28458ad4d6ddSJeff Bonwick 28468ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 28478ad4d6ddSJeff Bonwick 2848e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 284944ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 28503f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 28513f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 28523f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 28533f9d6ad7SLin Ling 28543f9d6ad7SLin Ling /* XXX cleanup? */ 28553f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 28563f9d6ad7SLin Ling atomic_add_64(processed, psize); 28573f9d6ad7SLin Ling vs->vs_scan_processed += psize; 28583f9d6ad7SLin Ling } 28593f9d6ad7SLin Ling 28608ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2861e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2862fa9e4066Sahrens } 28638ad4d6ddSJeff Bonwick 28648ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 28658ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 28668ad4d6ddSJeff Bonwick 28678ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2868fa9e4066Sahrens return; 2869fa9e4066Sahrens } 2870fa9e4066Sahrens 2871fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2872fa9e4066Sahrens return; 2873fa9e4066Sahrens 28748956713aSEric Schrock /* 28758956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 28768956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 28778956713aSEric Schrock * hard errors, when in reality they can happen for any number of 28788956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 28798956713aSEric Schrock */ 28808956713aSEric Schrock if (zio->io_error == EIO && 28818956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 28828956713aSEric Schrock return; 28838956713aSEric Schrock 28848f18d1faSGeorge Wilson /* 28858f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 28868f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 28878f18d1faSGeorge Wilson * errors. 28888f18d1faSGeorge Wilson */ 28898f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 28908f18d1faSGeorge Wilson return; 28918f18d1faSGeorge Wilson 2892e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2893b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2894e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2895e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2896e14bb325SJeff Bonwick else 2897e14bb325SJeff Bonwick vs->vs_read_errors++; 2898fa9e4066Sahrens } 2899b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2900e14bb325SJeff Bonwick vs->vs_write_errors++; 2901e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2902fa9e4066Sahrens 29038ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 29048ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 290544ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2906b24ab676SJeff Bonwick spa->spa_claiming)) { 29078ad4d6ddSJeff Bonwick /* 2908b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2909b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2910b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2911b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2912b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2913b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2914b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2915b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 29168ad4d6ddSJeff Bonwick * 29178ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 29188ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 29198ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 29208ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 29218ad4d6ddSJeff Bonwick */ 29228ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 29238ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 292444ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 29258ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 29268ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 29278ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2928b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2929b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2930b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2931b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 29328ad4d6ddSJeff Bonwick } 2933b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 29348ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2935fa9e4066Sahrens return; 29368ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 29378ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 29388ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2939fa9e4066Sahrens } 29408ad4d6ddSJeff Bonwick if (vd != rvd) 29418ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2942fa9e4066Sahrens } 2943fa9e4066Sahrens } 2944fa9e4066Sahrens 2945fa9e4066Sahrens /* 2946b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2947b24ab676SJeff Bonwick * and the root vdev. 2948fa9e4066Sahrens */ 2949fa9e4066Sahrens void 2950b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2951b24ab676SJeff Bonwick int64_t space_delta) 2952fa9e4066Sahrens { 295399653d4eSeschrock int64_t dspace_delta = space_delta; 29548654d025Sperrin spa_t *spa = vd->vdev_spa; 29558654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2956b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2957b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2958fa9e4066Sahrens 29598654d025Sperrin ASSERT(vd == vd->vdev_top); 296099653d4eSeschrock 29618654d025Sperrin /* 29628654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 29638654d025Sperrin * factor. We must calculate this here and not at the root vdev 29648654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 29658654d025Sperrin * childrens', thus not accurate enough for us. 29668654d025Sperrin */ 29678654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2968e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 29698654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 29708654d025Sperrin vd->vdev_deflate_ratio; 29718654d025Sperrin 29728654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 29738654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2974b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 29758654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 29768654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 29778654d025Sperrin 2978b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2979fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2980fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2981b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2982fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2983fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2984fa94a07fSbrendan } 2985b24ab676SJeff Bonwick 2986b24ab676SJeff Bonwick if (mc != NULL) { 2987b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2988b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2989b24ab676SJeff Bonwick 2990b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2991b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2992b24ab676SJeff Bonwick } 2993fa9e4066Sahrens } 2994fa9e4066Sahrens 2995fa9e4066Sahrens /* 2996fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2997fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2998fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2999fa9e4066Sahrens */ 3000fa9e4066Sahrens void 3001fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3002fa9e4066Sahrens { 3003fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3004fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3005fa9e4066Sahrens int c; 3006fa9e4066Sahrens 3007f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3008f9af39baSGeorge Wilson 3009c5904d13Seschrock /* 30106809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 30116809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3012c5904d13Seschrock */ 3013c5904d13Seschrock if (vd->vdev_aux != NULL) { 3014c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3015c5904d13Seschrock nvlist_t **aux; 3016c5904d13Seschrock uint_t naux; 3017c5904d13Seschrock 3018c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3019c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3020c5904d13Seschrock break; 3021c5904d13Seschrock } 3022c5904d13Seschrock 3023e14bb325SJeff Bonwick if (c == sav->sav_count) { 3024e14bb325SJeff Bonwick /* 3025e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3026e14bb325SJeff Bonwick */ 3027e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3028e14bb325SJeff Bonwick return; 3029e14bb325SJeff Bonwick } 3030e14bb325SJeff Bonwick 3031c5904d13Seschrock sav->sav_sync = B_TRUE; 3032c5904d13Seschrock 30336809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 30346809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 30356809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 30366809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 30376809eb4eSEric Schrock } 3038c5904d13Seschrock 3039c5904d13Seschrock ASSERT(c < naux); 3040c5904d13Seschrock 3041c5904d13Seschrock /* 3042c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3043c5904d13Seschrock * sketchy, but it will work. 3044c5904d13Seschrock */ 3045c5904d13Seschrock nvlist_free(aux[c]); 30463f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3047c5904d13Seschrock 3048c5904d13Seschrock return; 3049c5904d13Seschrock } 3050c5904d13Seschrock 30515dabedeeSbonwick /* 3052e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3053e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3054e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 30555dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 30565dabedeeSbonwick */ 3057e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3058e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3059e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 30605dabedeeSbonwick 3061fa9e4066Sahrens if (vd == rvd) { 3062fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 3063fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 3064fa9e4066Sahrens } else { 3065fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 3066fa9e4066Sahrens 306788ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 306888ecc943SGeorge Wilson !vd->vdev_ishole) 3069e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 3070fa9e4066Sahrens } 3071fa9e4066Sahrens } 3072fa9e4066Sahrens 3073fa9e4066Sahrens void 3074fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 3075fa9e4066Sahrens { 30765dabedeeSbonwick spa_t *spa = vd->vdev_spa; 30775dabedeeSbonwick 3078e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3079e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3080e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 30815dabedeeSbonwick 3082e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3083e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 3084e14bb325SJeff Bonwick } 3085e14bb325SJeff Bonwick 3086e14bb325SJeff Bonwick /* 3087e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 3088e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3089e14bb325SJeff Bonwick * the state changes from larger config changes because they require 3090e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3091e14bb325SJeff Bonwick */ 3092e14bb325SJeff Bonwick void 3093e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3094e14bb325SJeff Bonwick { 3095e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3096e14bb325SJeff Bonwick 3097f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3098e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3099e14bb325SJeff Bonwick 3100e14bb325SJeff Bonwick /* 3101e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3102e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3103e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3104e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3105e14bb325SJeff Bonwick */ 3106e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3107e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3108e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3109e14bb325SJeff Bonwick 3110b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 3111e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3112e14bb325SJeff Bonwick } 3113e14bb325SJeff Bonwick 3114e14bb325SJeff Bonwick void 3115e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3116e14bb325SJeff Bonwick { 3117e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3118e14bb325SJeff Bonwick 3119e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3120e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3121e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3122e14bb325SJeff Bonwick 3123e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3124e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3125fa9e4066Sahrens } 3126fa9e4066Sahrens 312732b87932Sek /* 312832b87932Sek * Propagate vdev state up from children to parent. 312932b87932Sek */ 313044cd46caSbillm void 313144cd46caSbillm vdev_propagate_state(vdev_t *vd) 313244cd46caSbillm { 31338ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 31348ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 313544cd46caSbillm int degraded = 0, faulted = 0; 313644cd46caSbillm int corrupted = 0; 313744cd46caSbillm vdev_t *child; 313844cd46caSbillm 31393d7072f8Seschrock if (vd->vdev_children > 0) { 3140573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 31413d7072f8Seschrock child = vd->vdev_child[c]; 314251ece835Seschrock 314388ecc943SGeorge Wilson /* 314488ecc943SGeorge Wilson * Don't factor holes into the decision. 314588ecc943SGeorge Wilson */ 314688ecc943SGeorge Wilson if (child->vdev_ishole) 314788ecc943SGeorge Wilson continue; 314888ecc943SGeorge Wilson 3149e14bb325SJeff Bonwick if (!vdev_readable(child) || 31508ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 315151ece835Seschrock /* 315251ece835Seschrock * Root special: if there is a top-level log 315351ece835Seschrock * device, treat the root vdev as if it were 315451ece835Seschrock * degraded. 315551ece835Seschrock */ 315651ece835Seschrock if (child->vdev_islog && vd == rvd) 315751ece835Seschrock degraded++; 315851ece835Seschrock else 315951ece835Seschrock faulted++; 316051ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 31613d7072f8Seschrock degraded++; 316251ece835Seschrock } 316344cd46caSbillm 31643d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 31653d7072f8Seschrock corrupted++; 31663d7072f8Seschrock } 316744cd46caSbillm 31683d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 31693d7072f8Seschrock 31703d7072f8Seschrock /* 3171e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 31723d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 31733d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 31743d7072f8Seschrock * replicas'. 31753d7072f8Seschrock */ 31763d7072f8Seschrock if (corrupted && vd == rvd && 31773d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 31783d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 31793d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 31803d7072f8Seschrock } 31813d7072f8Seschrock 318251ece835Seschrock if (vd->vdev_parent) 31833d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 318444cd46caSbillm } 318544cd46caSbillm 3186fa9e4066Sahrens /* 3187ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3188ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3189ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3190ea8dc4b6Seschrock * 3191ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3192ea8dc4b6Seschrock * generated. 3193fa9e4066Sahrens */ 3194fa9e4066Sahrens void 3195ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3196fa9e4066Sahrens { 3197560e6e96Seschrock uint64_t save_state; 3198c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3199ea8dc4b6Seschrock 3200ea8dc4b6Seschrock if (state == vd->vdev_state) { 3201ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3202fa9e4066Sahrens return; 3203ea8dc4b6Seschrock } 3204ea8dc4b6Seschrock 3205560e6e96Seschrock save_state = vd->vdev_state; 3206fa9e4066Sahrens 3207fa9e4066Sahrens vd->vdev_state = state; 3208fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3209fa9e4066Sahrens 32103d7072f8Seschrock /* 32113d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 321298d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 321398d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 321498d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 321598d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 321698d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 321798d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 321898d1cbfeSGeorge Wilson * children. 321998d1cbfeSGeorge Wilson */ 322098d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 322198d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 32223d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 32233d7072f8Seschrock 3224069f55e2SEric Schrock /* 3225069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3226069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3227069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3228069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3229069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3230069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3231069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3232069f55e2SEric Schrock */ 3233069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3234069f55e2SEric Schrock vd->vdev_prevstate != state) 3235069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3236069f55e2SEric Schrock 32373d7072f8Seschrock if (vd->vdev_removed && 32383d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 32393d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 32403d7072f8Seschrock /* 32413d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 32423d7072f8Seschrock * device was previously marked removed and someone attempted to 32433d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 32443d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 32453d7072f8Seschrock * it is one of our special test online cases, which is only 32463d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 32473d7072f8Seschrock * fault. 32483d7072f8Seschrock */ 32493d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 32503d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 32513d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 32523d7072f8Seschrock vd->vdev_removed = B_TRUE; 32533d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3254ea8dc4b6Seschrock /* 3255cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3256cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3257cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3258cb04b873SMark J Musante * is not considered an error. 3259ea8dc4b6Seschrock */ 3260cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3261cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3262560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3263560e6e96Seschrock vd->vdev_not_present = 1; 3264560e6e96Seschrock 3265560e6e96Seschrock /* 3266560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3267560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3268560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3269560e6e96Seschrock * want to post the ereport if the device was already in the 3270560e6e96Seschrock * CANT_OPEN state beforehand. 32713d7072f8Seschrock * 32723d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 32733d7072f8Seschrock * online the device in response to an insertion event. If we 32743d7072f8Seschrock * hit this case, then we have detected an insertion event for a 32753d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 32763d7072f8Seschrock * In this scenario, we don't post an ereport because we are 32773d7072f8Seschrock * about to replace the device, or attempt an online with 32783d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3279560e6e96Seschrock */ 32803d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 32813d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3282c5904d13Seschrock vd != spa->spa_root_vdev) { 3283ea8dc4b6Seschrock const char *class; 3284ea8dc4b6Seschrock 3285ea8dc4b6Seschrock switch (aux) { 3286ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3287ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3288ea8dc4b6Seschrock break; 3289ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3290ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3291ea8dc4b6Seschrock break; 3292ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3293ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3294ea8dc4b6Seschrock break; 3295ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3296ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3297ea8dc4b6Seschrock break; 3298ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3299ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3300ea8dc4b6Seschrock break; 3301ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3302ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3303ea8dc4b6Seschrock break; 3304ea8dc4b6Seschrock default: 3305ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3306ea8dc4b6Seschrock } 3307ea8dc4b6Seschrock 3308c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3309ea8dc4b6Seschrock } 3310ea8dc4b6Seschrock 33113d7072f8Seschrock /* Erase any notion of persistent removed state */ 33123d7072f8Seschrock vd->vdev_removed = B_FALSE; 33133d7072f8Seschrock } else { 33143d7072f8Seschrock vd->vdev_removed = B_FALSE; 33153d7072f8Seschrock } 3316ea8dc4b6Seschrock 33178b33d774STim Haley if (!isopen && vd->vdev_parent) 33188b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3319fa9e4066Sahrens } 332015e6edf1Sgw 332115e6edf1Sgw /* 332215e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 3323*c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 3324*c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 332515e6edf1Sgw */ 332615e6edf1Sgw boolean_t 332715e6edf1Sgw vdev_is_bootable(vdev_t *vd) 332815e6edf1Sgw { 332915e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 333015e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 333115e6edf1Sgw 333215e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 333315e6edf1Sgw vd->vdev_children > 1) { 333415e6edf1Sgw return (B_FALSE); 3335*c8811bd3SToomas Soome } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 333615e6edf1Sgw return (B_FALSE); 333715e6edf1Sgw } 333815e6edf1Sgw } 333915e6edf1Sgw 3340573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 334115e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 334215e6edf1Sgw return (B_FALSE); 334315e6edf1Sgw } 334415e6edf1Sgw return (B_TRUE); 334515e6edf1Sgw } 3346e6ca193dSGeorge Wilson 334788ecc943SGeorge Wilson /* 334888ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 334988ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 33504b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 33514b964adaSGeorge Wilson * device in the current vdev tree (nvd). 335288ecc943SGeorge Wilson */ 3353e6ca193dSGeorge Wilson void 335488ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3355e6ca193dSGeorge Wilson { 335688ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3357e6ca193dSGeorge Wilson 33584b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 335988ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 336088ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3361e6ca193dSGeorge Wilson 336288ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 336388ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3364e6ca193dSGeorge Wilson 33654b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3366e6ca193dSGeorge Wilson /* 33674b964adaSGeorge Wilson * Restore the persistent vdev state 3368e6ca193dSGeorge Wilson */ 336988ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 33704b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 33714b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 33724b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3373e6ca193dSGeorge Wilson } 3374e6ca193dSGeorge Wilson } 3375573ca77eSGeorge Wilson 33764b964adaSGeorge Wilson /* 33774b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 33784b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 33794b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 33804b964adaSGeorge Wilson */ 33814b964adaSGeorge Wilson boolean_t 33824b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 33834b964adaSGeorge Wilson { 33844b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 33854b964adaSGeorge Wilson !vd->vdev_removed) 33864b964adaSGeorge Wilson return (B_TRUE); 33874b964adaSGeorge Wilson 33884b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 33894b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 33904b964adaSGeorge Wilson return (B_TRUE); 33914b964adaSGeorge Wilson 33924b964adaSGeorge Wilson return (B_FALSE); 33934b964adaSGeorge Wilson } 33944b964adaSGeorge Wilson 3395573ca77eSGeorge Wilson /* 3396573ca77eSGeorge Wilson * Expand a vdev if possible. 3397573ca77eSGeorge Wilson */ 3398573ca77eSGeorge Wilson void 3399573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3400573ca77eSGeorge Wilson { 3401573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3402573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3403573ca77eSGeorge Wilson 3404573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3405573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3406573ca77eSGeorge Wilson vdev_config_dirty(vd); 3407573ca77eSGeorge Wilson } 3408573ca77eSGeorge Wilson } 34091195e687SMark J Musante 34101195e687SMark J Musante /* 34111195e687SMark J Musante * Split a vdev. 34121195e687SMark J Musante */ 34131195e687SMark J Musante void 34141195e687SMark J Musante vdev_split(vdev_t *vd) 34151195e687SMark J Musante { 34161195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 34171195e687SMark J Musante 34181195e687SMark J Musante vdev_remove_child(pvd, vd); 34191195e687SMark J Musante vdev_compact_children(pvd); 34201195e687SMark J Musante 34211195e687SMark J Musante cvd = pvd->vdev_child[0]; 34221195e687SMark J Musante if (pvd->vdev_children == 1) { 34231195e687SMark J Musante vdev_remove_parent(cvd); 34241195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 34251195e687SMark J Musante } 34261195e687SMark J Musante vdev_propagate_state(cvd); 34271195e687SMark J Musante } 3428283b8460SGeorge.Wilson 3429283b8460SGeorge.Wilson void 3430283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3431283b8460SGeorge.Wilson { 3432283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3433283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3434283b8460SGeorge.Wilson 3435283b8460SGeorge.Wilson vdev_deadman(cvd); 3436283b8460SGeorge.Wilson } 3437283b8460SGeorge.Wilson 3438283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3439283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3440283b8460SGeorge.Wilson 3441283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 344269962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3443283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3444283b8460SGeorge.Wilson zio_t *fio; 3445283b8460SGeorge.Wilson uint64_t delta; 3446283b8460SGeorge.Wilson 3447283b8460SGeorge.Wilson /* 3448283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3449283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3450283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3451283b8460SGeorge.Wilson */ 345269962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3453c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3454c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 3455c55e05cbSMatthew Ahrens zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3456c55e05cbSMatthew Ahrens "delta %lluns, last io %lluns", 3457283b8460SGeorge.Wilson fio->io_timestamp, delta, 3458283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3459283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3460283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3461283b8460SGeorge.Wilson } 3462283b8460SGeorge.Wilson } 3463283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3464283b8460SGeorge.Wilson } 3465283b8460SGeorge.Wilson } 3466