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. 25*5f368aefSYuri Pankov * Copyright 2017 Nexenta Systems, Inc. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27c8811bd3SToomas 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> 4866094686SDan Kimmel #include <sys/abd.h> 49fa9e4066Sahrens 50fa9e4066Sahrens /* 51fa9e4066Sahrens * Virtual device management. 52fa9e4066Sahrens */ 53fa9e4066Sahrens 54fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 55fa9e4066Sahrens &vdev_root_ops, 56fa9e4066Sahrens &vdev_raidz_ops, 57fa9e4066Sahrens &vdev_mirror_ops, 58fa9e4066Sahrens &vdev_replacing_ops, 5999653d4eSeschrock &vdev_spare_ops, 60fa9e4066Sahrens &vdev_disk_ops, 61fa9e4066Sahrens &vdev_file_ops, 62fa9e4066Sahrens &vdev_missing_ops, 6388ecc943SGeorge Wilson &vdev_hole_ops, 64fa9e4066Sahrens NULL 65fa9e4066Sahrens }; 66fa9e4066Sahrens 67088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 68088f3894Sahrens int zfs_scrub_limit = 10; 6905b2b3b8Smishra 70bf3e216cSMatthew Ahrens /* 71bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 72bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 73bf3e216cSMatthew Ahrens */ 74bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 75bf3e216cSMatthew Ahrens 76fa9e4066Sahrens /* 77fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 78fa9e4066Sahrens */ 79fa9e4066Sahrens static vdev_ops_t * 80fa9e4066Sahrens vdev_getops(const char *type) 81fa9e4066Sahrens { 82fa9e4066Sahrens vdev_ops_t *ops, **opspp; 83fa9e4066Sahrens 84fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 85fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 86fa9e4066Sahrens break; 87fa9e4066Sahrens 88fa9e4066Sahrens return (ops); 89fa9e4066Sahrens } 90fa9e4066Sahrens 91fa9e4066Sahrens /* 92fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 93fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 94fa9e4066Sahrens */ 95fa9e4066Sahrens uint64_t 96fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 97fa9e4066Sahrens { 98ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 99fa9e4066Sahrens uint64_t csize; 100fa9e4066Sahrens 101573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 102fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 103fa9e4066Sahrens asize = MAX(asize, csize); 104fa9e4066Sahrens } 105fa9e4066Sahrens 106fa9e4066Sahrens return (asize); 107fa9e4066Sahrens } 108fa9e4066Sahrens 1092a79c5feSlling /* 110573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 111573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 112573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 113573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1142a79c5feSlling */ 1152a79c5feSlling uint64_t 116573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1172a79c5feSlling { 118573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1192a79c5feSlling 120573ca77eSGeorge Wilson /* 1214263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 122573ca77eSGeorge Wilson * just return our own asize. 123573ca77eSGeorge Wilson */ 124573ca77eSGeorge Wilson if (pvd == NULL) 125573ca77eSGeorge Wilson return (vd->vdev_asize); 1262a79c5feSlling 1272a79c5feSlling /* 128573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 129573ca77eSGeorge Wilson * to the nearest metaslab. 1302a79c5feSlling */ 131573ca77eSGeorge Wilson if (vd == vd->vdev_top) 132573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1332a79c5feSlling 134573ca77eSGeorge Wilson /* 135573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 136573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 137573ca77eSGeorge Wilson */ 138573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 139c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 140c040c10cSSteven Hartland pvd->vdev_children); 1412a79c5feSlling 142573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 143573ca77eSGeorge Wilson } 1442a79c5feSlling 145573ca77eSGeorge Wilson void 146573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 147573ca77eSGeorge Wilson { 148573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 149573ca77eSGeorge Wilson 150573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 151573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1522a79c5feSlling } 1532a79c5feSlling 154fa9e4066Sahrens vdev_t * 155fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 156fa9e4066Sahrens { 157fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 158fa9e4066Sahrens 159e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 160e05725b1Sbonwick 161088f3894Sahrens if (vdev < rvd->vdev_children) { 162088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 163fa9e4066Sahrens return (rvd->vdev_child[vdev]); 164088f3894Sahrens } 165fa9e4066Sahrens 166fa9e4066Sahrens return (NULL); 167fa9e4066Sahrens } 168fa9e4066Sahrens 169fa9e4066Sahrens vdev_t * 170fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 171fa9e4066Sahrens { 172fa9e4066Sahrens vdev_t *mvd; 173fa9e4066Sahrens 1740e34b6a7Sbonwick if (vd->vdev_guid == guid) 175fa9e4066Sahrens return (vd); 176fa9e4066Sahrens 177573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 178fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 179fa9e4066Sahrens NULL) 180fa9e4066Sahrens return (mvd); 181fa9e4066Sahrens 182fa9e4066Sahrens return (NULL); 183fa9e4066Sahrens } 184fa9e4066Sahrens 18512380e1eSArne Jansen static int 18612380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 18712380e1eSArne Jansen { 18812380e1eSArne Jansen int n = 0; 18912380e1eSArne Jansen 19012380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 19112380e1eSArne Jansen return (1); 19212380e1eSArne Jansen 19312380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 19412380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 19512380e1eSArne Jansen 19612380e1eSArne Jansen return (n); 19712380e1eSArne Jansen } 19812380e1eSArne Jansen 19912380e1eSArne Jansen int 20012380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 20112380e1eSArne Jansen { 20212380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 20312380e1eSArne Jansen } 20412380e1eSArne Jansen 205fa9e4066Sahrens void 206fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 207fa9e4066Sahrens { 208fa9e4066Sahrens size_t oldsize, newsize; 209fa9e4066Sahrens uint64_t id = cvd->vdev_id; 210fa9e4066Sahrens vdev_t **newchild; 21181cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 212fa9e4066Sahrens 21381cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 214fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 215fa9e4066Sahrens 216fa9e4066Sahrens cvd->vdev_parent = pvd; 217fa9e4066Sahrens 218fa9e4066Sahrens if (pvd == NULL) 219fa9e4066Sahrens return; 220fa9e4066Sahrens 221fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 222fa9e4066Sahrens 223fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 224fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 225fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 226fa9e4066Sahrens 227fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 228fa9e4066Sahrens if (pvd->vdev_child != NULL) { 229fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 230fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 231fa9e4066Sahrens } 232fa9e4066Sahrens 233fa9e4066Sahrens pvd->vdev_child = newchild; 234fa9e4066Sahrens pvd->vdev_child[id] = cvd; 235fa9e4066Sahrens 236fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 237fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 238fa9e4066Sahrens 239fa9e4066Sahrens /* 240fa9e4066Sahrens * Walk up all ancestors to update guid sum. 241fa9e4066Sahrens */ 242fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 243fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 244fa9e4066Sahrens } 245fa9e4066Sahrens 246fa9e4066Sahrens void 247fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 248fa9e4066Sahrens { 249fa9e4066Sahrens int c; 250fa9e4066Sahrens uint_t id = cvd->vdev_id; 251fa9e4066Sahrens 252fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 253fa9e4066Sahrens 254fa9e4066Sahrens if (pvd == NULL) 255fa9e4066Sahrens return; 256fa9e4066Sahrens 257fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 258fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 259fa9e4066Sahrens 260fa9e4066Sahrens pvd->vdev_child[id] = NULL; 261fa9e4066Sahrens cvd->vdev_parent = NULL; 262fa9e4066Sahrens 263fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 264fa9e4066Sahrens if (pvd->vdev_child[c]) 265fa9e4066Sahrens break; 266fa9e4066Sahrens 267fa9e4066Sahrens if (c == pvd->vdev_children) { 268fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 269fa9e4066Sahrens pvd->vdev_child = NULL; 270fa9e4066Sahrens pvd->vdev_children = 0; 271fa9e4066Sahrens } 272fa9e4066Sahrens 273fa9e4066Sahrens /* 274fa9e4066Sahrens * Walk up all ancestors to update guid sum. 275fa9e4066Sahrens */ 276fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 277fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 278fa9e4066Sahrens } 279fa9e4066Sahrens 280fa9e4066Sahrens /* 281fa9e4066Sahrens * Remove any holes in the child array. 282fa9e4066Sahrens */ 283fa9e4066Sahrens void 284fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 285fa9e4066Sahrens { 286fa9e4066Sahrens vdev_t **newchild, *cvd; 287fa9e4066Sahrens int oldc = pvd->vdev_children; 288573ca77eSGeorge Wilson int newc; 289fa9e4066Sahrens 290e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 291fa9e4066Sahrens 292573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 293fa9e4066Sahrens if (pvd->vdev_child[c]) 294fa9e4066Sahrens newc++; 295fa9e4066Sahrens 296fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 297fa9e4066Sahrens 298573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 299fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 300fa9e4066Sahrens newchild[newc] = cvd; 301fa9e4066Sahrens cvd->vdev_id = newc++; 302fa9e4066Sahrens } 303fa9e4066Sahrens } 304fa9e4066Sahrens 305fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 306fa9e4066Sahrens pvd->vdev_child = newchild; 307fa9e4066Sahrens pvd->vdev_children = newc; 308fa9e4066Sahrens } 309fa9e4066Sahrens 310fa9e4066Sahrens /* 311fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 312fa9e4066Sahrens */ 31388ecc943SGeorge Wilson vdev_t * 314fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 315fa9e4066Sahrens { 316fa9e4066Sahrens vdev_t *vd; 317fa9e4066Sahrens 318fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 319fa9e4066Sahrens 3200e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 3210e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 3220e34b6a7Sbonwick spa->spa_root_vdev = vd; 323e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 3240e34b6a7Sbonwick } 3250e34b6a7Sbonwick 32688ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 3270e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 3280e34b6a7Sbonwick /* 3290e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3300e34b6a7Sbonwick * which must be unique among all pools. 3310e34b6a7Sbonwick */ 3321195e687SMark J Musante guid = spa_generate_guid(NULL); 3330e34b6a7Sbonwick } else { 3340e34b6a7Sbonwick /* 3350e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3360e34b6a7Sbonwick */ 3371195e687SMark J Musante guid = spa_generate_guid(spa); 3380e34b6a7Sbonwick } 3390e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3400e34b6a7Sbonwick } 3410e34b6a7Sbonwick 342fa9e4066Sahrens vd->vdev_spa = spa; 343fa9e4066Sahrens vd->vdev_id = id; 344fa9e4066Sahrens vd->vdev_guid = guid; 345fa9e4066Sahrens vd->vdev_guid_sum = guid; 346fa9e4066Sahrens vd->vdev_ops = ops; 347fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 34888ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 349fa9e4066Sahrens 350fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3515ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 352e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3530f7643c7SGeorge Wilson mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); 3548ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3550713e232SGeorge Wilson vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 3568ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3578ad4d6ddSJeff Bonwick } 358fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 359fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 360fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 361fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 362fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3633d7072f8Seschrock vdev_queue_init(vd); 3643d7072f8Seschrock vdev_cache_init(vd); 365fa9e4066Sahrens 366fa9e4066Sahrens return (vd); 367fa9e4066Sahrens } 368fa9e4066Sahrens 369fa9e4066Sahrens /* 370fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 371fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 372fa9e4066Sahrens * different for each case. 373fa9e4066Sahrens */ 37499653d4eSeschrock int 37599653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 37699653d4eSeschrock int alloctype) 377fa9e4066Sahrens { 378fa9e4066Sahrens vdev_ops_t *ops; 379fa9e4066Sahrens char *type; 3808654d025Sperrin uint64_t guid = 0, islog, nparity; 381fa9e4066Sahrens vdev_t *vd; 382fa9e4066Sahrens 383e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 384fa9e4066Sahrens 385fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 386be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 387fa9e4066Sahrens 388fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 389be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 390fa9e4066Sahrens 391fa9e4066Sahrens /* 392fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 393fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 394fa9e4066Sahrens */ 395fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 396fa9e4066Sahrens uint64_t label_id; 397fa9e4066Sahrens 398fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 399fa9e4066Sahrens label_id != id) 400be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 401fa9e4066Sahrens 402fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 403be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40499653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 40599653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 406be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 407fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 408fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 409be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 41021ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 41121ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 412be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 413fa9e4066Sahrens } 414fa9e4066Sahrens 41599653d4eSeschrock /* 41699653d4eSeschrock * The first allocated vdev must be of type 'root'. 41799653d4eSeschrock */ 41899653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 419be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 42099653d4eSeschrock 4218654d025Sperrin /* 4228654d025Sperrin * Determine whether we're a log vdev. 4238654d025Sperrin */ 4248654d025Sperrin islog = 0; 4258654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 426990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 427be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 428fa9e4066Sahrens 42988ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 430be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 43188ecc943SGeorge Wilson 43299653d4eSeschrock /* 4338654d025Sperrin * Set the nparity property for RAID-Z vdevs. 43499653d4eSeschrock */ 4358654d025Sperrin nparity = -1ULL; 43699653d4eSeschrock if (ops == &vdev_raidz_ops) { 43799653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4388654d025Sperrin &nparity) == 0) { 439b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 440be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 44199653d4eSeschrock /* 442f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 443f94275ceSAdam Leventhal * device. 44499653d4eSeschrock */ 445f94275ceSAdam Leventhal if (nparity > 1 && 446f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 447be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 448f94275ceSAdam Leventhal if (nparity > 2 && 449f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 450be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 45199653d4eSeschrock } else { 45299653d4eSeschrock /* 45399653d4eSeschrock * We require the parity to be specified for SPAs that 45499653d4eSeschrock * support multiple parity levels. 45599653d4eSeschrock */ 456f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 457be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 45899653d4eSeschrock /* 45999653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 46099653d4eSeschrock */ 4618654d025Sperrin nparity = 1; 46299653d4eSeschrock } 46399653d4eSeschrock } else { 4648654d025Sperrin nparity = 0; 46599653d4eSeschrock } 4668654d025Sperrin ASSERT(nparity != -1ULL); 4678654d025Sperrin 4688654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4698654d025Sperrin 4708654d025Sperrin vd->vdev_islog = islog; 4718654d025Sperrin vd->vdev_nparity = nparity; 4728654d025Sperrin 4738654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4748654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4758654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4768654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4778654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4788654d025Sperrin &vd->vdev_physpath) == 0) 4798654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4806809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4816809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 48299653d4eSeschrock 483afefbcddSeschrock /* 484afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 485afefbcddSeschrock * as -1. 486afefbcddSeschrock */ 487afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 488afefbcddSeschrock &vd->vdev_wholedisk) != 0) 489afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 490afefbcddSeschrock 491ea8dc4b6Seschrock /* 492ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 493ea8dc4b6Seschrock * was not present at the time of import. 494ea8dc4b6Seschrock */ 4956809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4966809eb4eSEric Schrock &vd->vdev_not_present); 497ea8dc4b6Seschrock 498ecc2d604Sbonwick /* 499ecc2d604Sbonwick * Get the alignment requirement. 500ecc2d604Sbonwick */ 501ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 502ecc2d604Sbonwick 50388ecc943SGeorge Wilson /* 50488ecc943SGeorge Wilson * Retrieve the vdev creation time. 50588ecc943SGeorge Wilson */ 50688ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 50788ecc943SGeorge Wilson &vd->vdev_crtxg); 50888ecc943SGeorge Wilson 509fa9e4066Sahrens /* 510fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 511fa9e4066Sahrens */ 5121195e687SMark J Musante if (parent && !parent->vdev_parent && 5131195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 514fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 515fa9e4066Sahrens &vd->vdev_ms_array); 516fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 517fa9e4066Sahrens &vd->vdev_ms_shift); 518fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 519fa9e4066Sahrens &vd->vdev_asize); 5203f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 5213f9d6ad7SLin Ling &vd->vdev_removing); 522215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 523215198a6SJoe Stein &vd->vdev_top_zap); 524215198a6SJoe Stein } else { 525215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 526fa9e4066Sahrens } 527fa9e4066Sahrens 528cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 529a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 5309f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 5311195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 5329f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 533a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 534a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 535a1521560SJeff Bonwick } 536a1521560SJeff Bonwick 537215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 538215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 539215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 540215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 541215198a6SJoe Stein } else { 542215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 543215198a6SJoe Stein } 544215198a6SJoe Stein 545fa9e4066Sahrens /* 5463d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 547fa9e4066Sahrens */ 548215198a6SJoe Stein 549c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 55021ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 55121ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 552c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 553c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5540713e232SGeorge Wilson &vd->vdev_dtl_object); 555c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 556c5904d13Seschrock &vd->vdev_unspare); 557c5904d13Seschrock } 55821ecdf64SLin Ling 55921ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 56021ecdf64SLin Ling uint64_t spare = 0; 56121ecdf64SLin Ling 56221ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 56321ecdf64SLin Ling &spare) == 0 && spare) 56421ecdf64SLin Ling spa_spare_add(vd); 56521ecdf64SLin Ling } 56621ecdf64SLin Ling 567ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 568ecc2d604Sbonwick &vd->vdev_offline); 569c5904d13Seschrock 570b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 571b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 572cb04b873SMark J Musante 5733d7072f8Seschrock /* 5743d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5753d7072f8Seschrock * state, as the diagnosis made on another system may not be 576069f55e2SEric Schrock * valid in the current context. Local vdevs will 577069f55e2SEric Schrock * remain in the faulted state. 5783d7072f8Seschrock */ 579b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5803d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5813d7072f8Seschrock &vd->vdev_faulted); 5823d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5833d7072f8Seschrock &vd->vdev_degraded); 5843d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5853d7072f8Seschrock &vd->vdev_removed); 586069f55e2SEric Schrock 587069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 588069f55e2SEric Schrock char *aux; 589069f55e2SEric Schrock 590069f55e2SEric Schrock vd->vdev_label_aux = 591069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 592069f55e2SEric Schrock if (nvlist_lookup_string(nv, 593069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 594069f55e2SEric Schrock strcmp(aux, "external") == 0) 595069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 596069f55e2SEric Schrock } 5973d7072f8Seschrock } 598fa9e4066Sahrens } 599fa9e4066Sahrens 600fa9e4066Sahrens /* 601fa9e4066Sahrens * Add ourselves to the parent's list of children. 602fa9e4066Sahrens */ 603fa9e4066Sahrens vdev_add_child(parent, vd); 604fa9e4066Sahrens 60599653d4eSeschrock *vdp = vd; 60699653d4eSeschrock 60799653d4eSeschrock return (0); 608fa9e4066Sahrens } 609fa9e4066Sahrens 610fa9e4066Sahrens void 611fa9e4066Sahrens vdev_free(vdev_t *vd) 612fa9e4066Sahrens { 6133d7072f8Seschrock spa_t *spa = vd->vdev_spa; 614fa9e4066Sahrens 615fa9e4066Sahrens /* 616fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 617fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 618fa9e4066Sahrens */ 619fa9e4066Sahrens vdev_close(vd); 620fa9e4066Sahrens 621e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 622b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 623fa9e4066Sahrens 624fa9e4066Sahrens /* 625fa9e4066Sahrens * Free all children. 626fa9e4066Sahrens */ 627573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 628fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 629fa9e4066Sahrens 630fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 631fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 632fa9e4066Sahrens 633fa9e4066Sahrens /* 634fa9e4066Sahrens * Discard allocation state. 635fa9e4066Sahrens */ 636a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 637fa9e4066Sahrens vdev_metaslab_fini(vd); 638a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 639a1521560SJeff Bonwick } 640fa9e4066Sahrens 641fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 642fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 643fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 644fa9e4066Sahrens 645fa9e4066Sahrens /* 646fa9e4066Sahrens * Remove this vdev from its parent's child list. 647fa9e4066Sahrens */ 648fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 649fa9e4066Sahrens 650fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 651fa9e4066Sahrens 6523d7072f8Seschrock /* 6533d7072f8Seschrock * Clean up vdev structure. 6543d7072f8Seschrock */ 6553d7072f8Seschrock vdev_queue_fini(vd); 6563d7072f8Seschrock vdev_cache_fini(vd); 6573d7072f8Seschrock 6583d7072f8Seschrock if (vd->vdev_path) 6593d7072f8Seschrock spa_strfree(vd->vdev_path); 6603d7072f8Seschrock if (vd->vdev_devid) 6613d7072f8Seschrock spa_strfree(vd->vdev_devid); 6623d7072f8Seschrock if (vd->vdev_physpath) 6633d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6646809eb4eSEric Schrock if (vd->vdev_fru) 6656809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6663d7072f8Seschrock 6673d7072f8Seschrock if (vd->vdev_isspare) 6683d7072f8Seschrock spa_spare_remove(vd); 669fa94a07fSbrendan if (vd->vdev_isl2cache) 670fa94a07fSbrendan spa_l2cache_remove(vd); 6713d7072f8Seschrock 6723d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6733d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6748ad4d6ddSJeff Bonwick 6753d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6760713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 6778ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6780713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 6790713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 6808ad4d6ddSJeff Bonwick } 6813d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6828ad4d6ddSJeff Bonwick 6830f7643c7SGeorge Wilson mutex_destroy(&vd->vdev_queue_lock); 6843d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6853d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 686e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6873d7072f8Seschrock 6883d7072f8Seschrock if (vd == spa->spa_root_vdev) 6893d7072f8Seschrock spa->spa_root_vdev = NULL; 6903d7072f8Seschrock 6913d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 692fa9e4066Sahrens } 693fa9e4066Sahrens 694fa9e4066Sahrens /* 695fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 696fa9e4066Sahrens */ 697fa9e4066Sahrens static void 698fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 699fa9e4066Sahrens { 700fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 701fa9e4066Sahrens metaslab_t *msp; 702fa9e4066Sahrens vdev_t *vd; 703fa9e4066Sahrens int t; 704fa9e4066Sahrens 705fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 706fa9e4066Sahrens 707fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 708fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 709fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 710215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 711fa9e4066Sahrens 712fa9e4066Sahrens svd->vdev_ms_array = 0; 713fa9e4066Sahrens svd->vdev_ms_shift = 0; 714fa9e4066Sahrens svd->vdev_ms_count = 0; 715215198a6SJoe Stein svd->vdev_top_zap = 0; 716fa9e4066Sahrens 717cd0837ccSGeorge Wilson if (tvd->vdev_mg) 718cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 719fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 720fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 721fa9e4066Sahrens 722fa9e4066Sahrens svd->vdev_mg = NULL; 723fa9e4066Sahrens svd->vdev_ms = NULL; 724ecc2d604Sbonwick 725ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 726ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 727fa9e4066Sahrens 728fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 729fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 73099653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 731fa9e4066Sahrens 732fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 733fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 73499653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 735fa9e4066Sahrens 736fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 737fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 738fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 739fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 740fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 741fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 742fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 743fa9e4066Sahrens } 744fa9e4066Sahrens 745e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 746fa9e4066Sahrens vdev_config_clean(svd); 747fa9e4066Sahrens vdev_config_dirty(tvd); 748fa9e4066Sahrens } 749fa9e4066Sahrens 750e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 751e14bb325SJeff Bonwick vdev_state_clean(svd); 752e14bb325SJeff Bonwick vdev_state_dirty(tvd); 753e14bb325SJeff Bonwick } 754e14bb325SJeff Bonwick 75599653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 75699653d4eSeschrock svd->vdev_deflate_ratio = 0; 7578654d025Sperrin 7588654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7598654d025Sperrin svd->vdev_islog = 0; 760fa9e4066Sahrens } 761fa9e4066Sahrens 762fa9e4066Sahrens static void 763fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 764fa9e4066Sahrens { 765fa9e4066Sahrens if (vd == NULL) 766fa9e4066Sahrens return; 767fa9e4066Sahrens 768fa9e4066Sahrens vd->vdev_top = tvd; 769fa9e4066Sahrens 770573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 771fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 772fa9e4066Sahrens } 773fa9e4066Sahrens 774fa9e4066Sahrens /* 775fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 776fa9e4066Sahrens */ 777fa9e4066Sahrens vdev_t * 778fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 779fa9e4066Sahrens { 780fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 781fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 782fa9e4066Sahrens vdev_t *mvd; 783fa9e4066Sahrens 784e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 785fa9e4066Sahrens 786fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 787ecc2d604Sbonwick 788ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 789573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 7904263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 791ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 792ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 79388ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 794ecc2d604Sbonwick 795fa9e4066Sahrens vdev_remove_child(pvd, cvd); 796fa9e4066Sahrens vdev_add_child(pvd, mvd); 797fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 798fa9e4066Sahrens vdev_add_child(mvd, cvd); 799fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 800fa9e4066Sahrens 801fa9e4066Sahrens if (mvd == mvd->vdev_top) 802fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 803fa9e4066Sahrens 804fa9e4066Sahrens return (mvd); 805fa9e4066Sahrens } 806fa9e4066Sahrens 807fa9e4066Sahrens /* 808fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 809fa9e4066Sahrens */ 810fa9e4066Sahrens void 811fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 812fa9e4066Sahrens { 813fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 814fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 815fa9e4066Sahrens 816e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 817fa9e4066Sahrens 818fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 819fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 82099653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 82199653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 822ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 823fa9e4066Sahrens 824fa9e4066Sahrens vdev_remove_child(mvd, cvd); 825fa9e4066Sahrens vdev_remove_child(pvd, mvd); 8268ad4d6ddSJeff Bonwick 82799653d4eSeschrock /* 828e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 829e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 830e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 831e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 83299653d4eSeschrock */ 8338ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 8348ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 8351195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 8368ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 8378ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 8388ad4d6ddSJeff Bonwick } 839e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 840e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 841fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 842fa9e4066Sahrens 843fa9e4066Sahrens if (cvd == cvd->vdev_top) 844fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 845fa9e4066Sahrens 846fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 847fa9e4066Sahrens vdev_free(mvd); 848fa9e4066Sahrens } 849fa9e4066Sahrens 850ea8dc4b6Seschrock int 851fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 852fa9e4066Sahrens { 853fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 854ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 855ecc2d604Sbonwick uint64_t m; 856fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 857fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 858ecc2d604Sbonwick metaslab_t **mspp; 859ecc2d604Sbonwick int error; 860fa9e4066Sahrens 861a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 862a1521560SJeff Bonwick 86388ecc943SGeorge Wilson /* 86488ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 86588ecc943SGeorge Wilson */ 86688ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8670e34b6a7Sbonwick return (0); 8680e34b6a7Sbonwick 86988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 87088ecc943SGeorge Wilson 871e6ca193dSGeorge Wilson /* 872e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 873b5152584SMatthew Ahrens * in 128k (1 << 17) because it is the "typical" blocksize. 874b5152584SMatthew Ahrens * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 875b5152584SMatthew Ahrens * otherwise it would inconsistently account for existing bp's. 876e6ca193dSGeorge Wilson */ 877e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 878e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 879e6ca193dSGeorge Wilson 880fa9e4066Sahrens ASSERT(oldc <= newc); 881fa9e4066Sahrens 882ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 883fa9e4066Sahrens 884ecc2d604Sbonwick if (oldc != 0) { 885ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 886ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 887ecc2d604Sbonwick } 888fa9e4066Sahrens 889ecc2d604Sbonwick vd->vdev_ms = mspp; 890ecc2d604Sbonwick vd->vdev_ms_count = newc; 891fa9e4066Sahrens 892ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 8930713e232SGeorge Wilson uint64_t object = 0; 8940713e232SGeorge Wilson 895ecc2d604Sbonwick if (txg == 0) { 896ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8977bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8987bfdf011SNeil Perrin DMU_READ_PREFETCH); 899ecc2d604Sbonwick if (error) 900ecc2d604Sbonwick return (error); 901fa9e4066Sahrens } 9021e9bd7ecSPrakash Surya 9031e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 9041e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 9051e9bd7ecSPrakash Surya if (error) 9061e9bd7ecSPrakash Surya return (error); 907fa9e4066Sahrens } 908fa9e4066Sahrens 909a1521560SJeff Bonwick if (txg == 0) 910a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 911a1521560SJeff Bonwick 9123f9d6ad7SLin Ling /* 9133f9d6ad7SLin Ling * If the vdev is being removed we don't activate 9143f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 9153f9d6ad7SLin Ling * allocations are performed on this device. 9163f9d6ad7SLin Ling */ 9173f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 918a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 919a1521560SJeff Bonwick 920a1521560SJeff Bonwick if (txg == 0) 921a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 922a1521560SJeff Bonwick 923ea8dc4b6Seschrock return (0); 924fa9e4066Sahrens } 925fa9e4066Sahrens 926fa9e4066Sahrens void 927fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 928fa9e4066Sahrens { 929fa9e4066Sahrens uint64_t m; 930fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 931fa9e4066Sahrens 932fa9e4066Sahrens if (vd->vdev_ms != NULL) { 933a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 9340713e232SGeorge Wilson for (m = 0; m < count; m++) { 9350713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 9360713e232SGeorge Wilson 9370713e232SGeorge Wilson if (msp != NULL) 9380713e232SGeorge Wilson metaslab_fini(msp); 9390713e232SGeorge Wilson } 940fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 941fa9e4066Sahrens vd->vdev_ms = NULL; 942fa9e4066Sahrens } 943fa9e4066Sahrens } 944fa9e4066Sahrens 945e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 946e14bb325SJeff Bonwick boolean_t vps_readable; 947e14bb325SJeff Bonwick boolean_t vps_writeable; 948e14bb325SJeff Bonwick int vps_flags; 949e14bb325SJeff Bonwick } vdev_probe_stats_t; 950e14bb325SJeff Bonwick 951e14bb325SJeff Bonwick static void 952e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9530a4e9518Sgw { 9548ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 955a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 956e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 957a3f829aeSBill Moore 958a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 959e14bb325SJeff Bonwick 960e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 961e14bb325SJeff Bonwick if (zio->io_error == 0) 962e14bb325SJeff Bonwick vps->vps_readable = 1; 9638ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 964a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 96566094686SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 966e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 967e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 968e14bb325SJeff Bonwick } else { 96966094686SDan Kimmel abd_free(zio->io_abd); 970e14bb325SJeff Bonwick } 971e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 972e14bb325SJeff Bonwick if (zio->io_error == 0) 973e14bb325SJeff Bonwick vps->vps_writeable = 1; 97466094686SDan Kimmel abd_free(zio->io_abd); 975e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 976a3f829aeSBill Moore zio_t *pio; 977e14bb325SJeff Bonwick 978e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 979e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 980e14bb325SJeff Bonwick 981e14bb325SJeff Bonwick if (vdev_readable(vd) && 9828ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 983e14bb325SJeff Bonwick zio->io_error = 0; 984e14bb325SJeff Bonwick } else { 985e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 986e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9878ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 988be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 989e14bb325SJeff Bonwick } 990a3f829aeSBill Moore 991a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 992a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 993a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 994a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 995a3f829aeSBill Moore 9960f7643c7SGeorge Wilson zio_link_t *zl = NULL; 9970f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 998a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 999be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1000a3f829aeSBill Moore 1001e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1002e14bb325SJeff Bonwick } 1003e14bb325SJeff Bonwick } 10040a4e9518Sgw 1005e14bb325SJeff Bonwick /* 1006f7170741SWill Andrews * Determine whether this device is accessible. 1007f7170741SWill Andrews * 1008f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1009f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1010f7170741SWill Andrews * a VTOC. 1011e14bb325SJeff Bonwick */ 1012e14bb325SJeff Bonwick zio_t * 1013a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1014e14bb325SJeff Bonwick { 1015e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1016a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1017a3f829aeSBill Moore zio_t *pio; 1018a3f829aeSBill Moore 1019a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 10200a4e9518Sgw 1021a3f829aeSBill Moore /* 1022a3f829aeSBill Moore * Don't probe the probe. 1023a3f829aeSBill Moore */ 1024a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1025a3f829aeSBill Moore return (NULL); 1026e14bb325SJeff Bonwick 1027a3f829aeSBill Moore /* 1028a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1029a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1030a3f829aeSBill Moore * this vdev will become parents of the probe io. 1031a3f829aeSBill Moore */ 1032a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1033e14bb325SJeff Bonwick 1034a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1035a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1036a3f829aeSBill Moore 1037a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1038a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 10398956713aSEric Schrock ZIO_FLAG_TRYHARD; 1040a3f829aeSBill Moore 1041a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1042a3f829aeSBill Moore /* 1043a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1044a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1045a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1046a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1047a3f829aeSBill Moore * any zio looking at these values can assume that 1048a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1049a3f829aeSBill Moore * important because when a device has intermittent 1050a3f829aeSBill Moore * connectivity problems, we want to ensure that 1051a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1052a3f829aeSBill Moore * the zio (EIO). 1053a3f829aeSBill Moore * 1054a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1055a3f829aeSBill Moore * values so the probe can reevaluate from first 1056a3f829aeSBill Moore * principles. 1057a3f829aeSBill Moore */ 1058a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1059a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1060a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1061a3f829aeSBill Moore } 1062a3f829aeSBill Moore 1063a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1064a3f829aeSBill Moore vdev_probe_done, vps, 1065a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1066a3f829aeSBill Moore 106798d1cbfeSGeorge Wilson /* 106898d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 106998d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 107098d1cbfeSGeorge Wilson */ 1071a3f829aeSBill Moore if (zio != NULL) { 1072a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1073a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1074a3f829aeSBill Moore } 1075e14bb325SJeff Bonwick } 1076e14bb325SJeff Bonwick 1077a3f829aeSBill Moore if (zio != NULL) 1078a3f829aeSBill Moore zio_add_child(zio, pio); 1079e14bb325SJeff Bonwick 1080a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1081e14bb325SJeff Bonwick 1082a3f829aeSBill Moore if (vps == NULL) { 1083a3f829aeSBill Moore ASSERT(zio != NULL); 1084a3f829aeSBill Moore return (NULL); 1085a3f829aeSBill Moore } 1086e14bb325SJeff Bonwick 1087e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1088a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1089e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 109066094686SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 109166094686SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1092e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1093e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1094e14bb325SJeff Bonwick } 1095e14bb325SJeff Bonwick 1096a3f829aeSBill Moore if (zio == NULL) 1097a3f829aeSBill Moore return (pio); 1098a3f829aeSBill Moore 1099a3f829aeSBill Moore zio_nowait(pio); 1100a3f829aeSBill Moore return (NULL); 11010a4e9518Sgw } 11020a4e9518Sgw 1103f64c0e34SEric Taylor static void 1104f64c0e34SEric Taylor vdev_open_child(void *arg) 1105f64c0e34SEric Taylor { 1106f64c0e34SEric Taylor vdev_t *vd = arg; 1107f64c0e34SEric Taylor 1108f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1109f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1110f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1111f64c0e34SEric Taylor } 1112f64c0e34SEric Taylor 1113681d9761SEric Taylor boolean_t 1114681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1115681d9761SEric Taylor { 1116681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1117681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1118681d9761SEric Taylor return (B_TRUE); 1119681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1120681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1121681d9761SEric Taylor return (B_TRUE); 1122681d9761SEric Taylor return (B_FALSE); 1123681d9761SEric Taylor } 1124681d9761SEric Taylor 1125f64c0e34SEric Taylor void 1126f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1127f64c0e34SEric Taylor { 1128f64c0e34SEric Taylor taskq_t *tq; 1129f64c0e34SEric Taylor int children = vd->vdev_children; 1130f64c0e34SEric Taylor 1131681d9761SEric Taylor /* 1132681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1133681d9761SEric Taylor * in a single thread so that the same thread holds the 1134681d9761SEric Taylor * spa_namespace_lock 1135681d9761SEric Taylor */ 1136681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1137681d9761SEric Taylor for (int c = 0; c < children; c++) 1138681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1139681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1140681d9761SEric Taylor return; 1141681d9761SEric Taylor } 1142f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1143f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1144f64c0e34SEric Taylor 1145f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1146f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1147f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1148f64c0e34SEric Taylor 1149f64c0e34SEric Taylor taskq_destroy(tq); 1150f64c0e34SEric Taylor } 1151f64c0e34SEric Taylor 1152fa9e4066Sahrens /* 1153fa9e4066Sahrens * Prepare a virtual device for access. 1154fa9e4066Sahrens */ 1155fa9e4066Sahrens int 1156fa9e4066Sahrens vdev_open(vdev_t *vd) 1157fa9e4066Sahrens { 11588ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1159fa9e4066Sahrens int error; 1160fa9e4066Sahrens uint64_t osize = 0; 11614263d13fSGeorge Wilson uint64_t max_osize = 0; 11624263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1163ecc2d604Sbonwick uint64_t ashift = 0; 1164fa9e4066Sahrens 1165f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1166f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1167fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1168fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1169fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1170fa9e4066Sahrens 1171fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1172e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1173e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1174573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1175fa9e4066Sahrens 1176069f55e2SEric Schrock /* 1177069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1178069f55e2SEric Schrock * faulted, bail out of the open. 1179069f55e2SEric Schrock */ 11803d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11813d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1182069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1183069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11843d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1185069f55e2SEric Schrock vd->vdev_label_aux); 1186be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 11873d7072f8Seschrock } else if (vd->vdev_offline) { 1188fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1189ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1190be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1191fa9e4066Sahrens } 1192fa9e4066Sahrens 11934263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1194fa9e4066Sahrens 1195095bcd66SGeorge Wilson /* 1196095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1197095bcd66SGeorge Wilson * the vdev on error. 1198095bcd66SGeorge Wilson */ 1199095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1200ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 12018956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1202ea8dc4b6Seschrock 1203fa9e4066Sahrens if (error) { 12043d7072f8Seschrock if (vd->vdev_removed && 12053d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 12063d7072f8Seschrock vd->vdev_removed = B_FALSE; 12073d7072f8Seschrock 1208ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1209fa9e4066Sahrens vd->vdev_stat.vs_aux); 1210fa9e4066Sahrens return (error); 1211fa9e4066Sahrens } 1212fa9e4066Sahrens 12133d7072f8Seschrock vd->vdev_removed = B_FALSE; 12143d7072f8Seschrock 1215096d22d4SEric Schrock /* 1216096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1217096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1218096d22d4SEric Schrock */ 1219096d22d4SEric Schrock if (vd->vdev_faulted) { 1220096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1221096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1222096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1223096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1224096d22d4SEric Schrock vd->vdev_label_aux); 1225be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1226096d22d4SEric Schrock } 1227096d22d4SEric Schrock 12283d7072f8Seschrock if (vd->vdev_degraded) { 12293d7072f8Seschrock ASSERT(vd->vdev_children == 0); 12303d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 12313d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 12323d7072f8Seschrock } else { 1233069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 12343d7072f8Seschrock } 1235fa9e4066Sahrens 123688ecc943SGeorge Wilson /* 123788ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 123888ecc943SGeorge Wilson */ 123988ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 124088ecc943SGeorge Wilson return (0); 124188ecc943SGeorge Wilson 1242573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1243ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1244ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1245ea8dc4b6Seschrock VDEV_AUX_NONE); 1246ea8dc4b6Seschrock break; 1247ea8dc4b6Seschrock } 1248573ca77eSGeorge Wilson } 1249fa9e4066Sahrens 1250fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 12514263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1252fa9e4066Sahrens 1253fa9e4066Sahrens if (vd->vdev_children == 0) { 1254fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1255ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1256ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1257be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1258fa9e4066Sahrens } 1259fa9e4066Sahrens psize = osize; 1260fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 12614263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 12624263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1263fa9e4066Sahrens } else { 1264ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1265fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1266ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1267ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1268be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1269fa9e4066Sahrens } 1270fa9e4066Sahrens psize = 0; 1271fa9e4066Sahrens asize = osize; 12724263d13fSGeorge Wilson max_asize = max_osize; 1273fa9e4066Sahrens } 1274fa9e4066Sahrens 1275fa9e4066Sahrens vd->vdev_psize = psize; 1276fa9e4066Sahrens 1277573ca77eSGeorge Wilson /* 1278c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1279573ca77eSGeorge Wilson */ 1280573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1281573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1282573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1283be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1284573ca77eSGeorge Wilson } 1285573ca77eSGeorge Wilson 1286fa9e4066Sahrens if (vd->vdev_asize == 0) { 1287fa9e4066Sahrens /* 1288fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1289ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1290fa9e4066Sahrens */ 1291fa9e4066Sahrens vd->vdev_asize = asize; 12924263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1293ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1294fa9e4066Sahrens } else { 1295fa9e4066Sahrens /* 12962384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 12972384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 12982384d9f8SGeorge Wilson * issue a warning instead. 1299fa9e4066Sahrens */ 13002384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 13012384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 13022384d9f8SGeorge Wilson cmn_err(CE_WARN, 13032384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 13042384d9f8SGeorge Wilson "larger than the pool's alignment\n", 13052384d9f8SGeorge Wilson vd->vdev_path); 1306fa9e4066Sahrens } 13074263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1308573ca77eSGeorge Wilson } 1309fa9e4066Sahrens 1310573ca77eSGeorge Wilson /* 1311c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1312c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1313c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1314c040c10cSSteven Hartland * making the additional space available. 1315c040c10cSSteven Hartland * 1316c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1317c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1318c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1319c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1320c040c10cSSteven Hartland * vdev_min_asize. 1321573ca77eSGeorge Wilson */ 1322c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1323c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1324c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1325c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1326573ca77eSGeorge Wilson vd->vdev_asize = asize; 1327fa9e4066Sahrens 1328573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1329fa9e4066Sahrens 13300a4e9518Sgw /* 13310a4e9518Sgw * Ensure we can issue some IO before declaring the 13320a4e9518Sgw * vdev open for business. 13330a4e9518Sgw */ 1334e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1335e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 133698d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 133798d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 13380a4e9518Sgw return (error); 13390a4e9518Sgw } 13400a4e9518Sgw 134181cd5c55SMatthew Ahrens /* 134281cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 134381cd5c55SMatthew Ahrens */ 134481cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 134581cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 134681cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 134781cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 134881cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 134981cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 135081cd5c55SMatthew Ahrens } 135181cd5c55SMatthew Ahrens 1352088f3894Sahrens /* 1353088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 13548ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 13558ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1356088f3894Sahrens */ 13578ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 13588ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 13598ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1360088f3894Sahrens 1361fa9e4066Sahrens return (0); 1362fa9e4066Sahrens } 1363fa9e4066Sahrens 1364560e6e96Seschrock /* 1365560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1366560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 13673d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1368560e6e96Seschrock * 1369d7f601efSGeorge Wilson * If 'strict' is false ignore the spa guid check. This is necessary because 1370d7f601efSGeorge Wilson * if the machine crashed during a re-guid the new guid might have been written 1371d7f601efSGeorge Wilson * to all of the vdev labels, but not the cached config. The strict check 1372d7f601efSGeorge Wilson * will be performed when the pool is opened again using the mos config. 1373d7f601efSGeorge Wilson * 1374560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1375560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1376560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1377560e6e96Seschrock * will be updated but the function will return 0. 1378560e6e96Seschrock */ 1379560e6e96Seschrock int 1380d7f601efSGeorge Wilson vdev_validate(vdev_t *vd, boolean_t strict) 1381560e6e96Seschrock { 1382560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1383560e6e96Seschrock nvlist_t *label; 13841195e687SMark J Musante uint64_t guid = 0, top_guid; 1385560e6e96Seschrock uint64_t state; 1386560e6e96Seschrock 1387573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1388d7f601efSGeorge Wilson if (vdev_validate(vd->vdev_child[c], strict) != 0) 1389be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1390560e6e96Seschrock 1391b5989ec7Seschrock /* 1392b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1393b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1394b5989ec7Seschrock * overwrite the previous state. 1395b5989ec7Seschrock */ 1396e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13971195e687SMark J Musante uint64_t aux_guid = 0; 13981195e687SMark J Musante nvlist_t *nvl; 1399bda88194SGeorge Wilson uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1400bda88194SGeorge Wilson spa_last_synced_txg(spa) : -1ULL; 1401560e6e96Seschrock 1402dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1403560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1404560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1405560e6e96Seschrock return (0); 1406560e6e96Seschrock } 1407560e6e96Seschrock 14081195e687SMark J Musante /* 14091195e687SMark J Musante * Determine if this vdev has been split off into another 14101195e687SMark J Musante * pool. If so, then refuse to open it. 14111195e687SMark J Musante */ 14121195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 14131195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 14141195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 14151195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 14161195e687SMark J Musante nvlist_free(label); 14171195e687SMark J Musante return (0); 14181195e687SMark J Musante } 14191195e687SMark J Musante 1420d7f601efSGeorge Wilson if (strict && (nvlist_lookup_uint64(label, 1421d7f601efSGeorge Wilson ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1422d7f601efSGeorge Wilson guid != spa_guid(spa))) { 1423560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1424560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1425560e6e96Seschrock nvlist_free(label); 1426560e6e96Seschrock return (0); 1427560e6e96Seschrock } 1428560e6e96Seschrock 14291195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 14301195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 14311195e687SMark J Musante &aux_guid) != 0) 14321195e687SMark J Musante aux_guid = 0; 14331195e687SMark J Musante 1434e14bb325SJeff Bonwick /* 1435e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1436e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1437e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1438e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1439e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1440e14bb325SJeff Bonwick * safely compare to that instead. 14411195e687SMark J Musante * 14421195e687SMark J Musante * If we split this vdev off instead, then we also check the 14431195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 14441195e687SMark J Musante * corrupt if it is partway through a split operation. 1445e14bb325SJeff Bonwick */ 1446560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1447e14bb325SJeff Bonwick &guid) != 0 || 1448e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1449e14bb325SJeff Bonwick &top_guid) != 0 || 14501195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1451e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1452560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1453560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1454560e6e96Seschrock nvlist_free(label); 1455560e6e96Seschrock return (0); 1456560e6e96Seschrock } 1457560e6e96Seschrock 1458560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1459560e6e96Seschrock &state) != 0) { 1460560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1461560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1462560e6e96Seschrock nvlist_free(label); 1463560e6e96Seschrock return (0); 1464560e6e96Seschrock } 1465560e6e96Seschrock 1466560e6e96Seschrock nvlist_free(label); 1467560e6e96Seschrock 1468bc758434SLin Ling /* 14694b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1470bc758434SLin Ling * state of the pool. 1471bc758434SLin Ling */ 14724b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1473b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1474bc758434SLin Ling state != POOL_STATE_ACTIVE) 1475be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1476560e6e96Seschrock 147751ece835Seschrock /* 147851ece835Seschrock * If we were able to open and validate a vdev that was 147951ece835Seschrock * previously marked permanently unavailable, clear that state 148051ece835Seschrock * now. 148151ece835Seschrock */ 148251ece835Seschrock if (vd->vdev_not_present) 148351ece835Seschrock vd->vdev_not_present = 0; 148451ece835Seschrock } 1485560e6e96Seschrock 1486560e6e96Seschrock return (0); 1487560e6e96Seschrock } 1488560e6e96Seschrock 1489fa9e4066Sahrens /* 1490fa9e4066Sahrens * Close a virtual device. 1491fa9e4066Sahrens */ 1492fa9e4066Sahrens void 1493fa9e4066Sahrens vdev_close(vdev_t *vd) 1494fa9e4066Sahrens { 14958ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1496095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14978ad4d6ddSJeff Bonwick 14988ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14998ad4d6ddSJeff Bonwick 15001195e687SMark J Musante /* 15011195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 15021195e687SMark J Musante * going offline. 15031195e687SMark J Musante */ 1504095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 15051195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1506095bcd66SGeorge Wilson 1507fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1508fa9e4066Sahrens 15093d7072f8Seschrock vdev_cache_purge(vd); 1510fa9e4066Sahrens 1511560e6e96Seschrock /* 1512573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1513560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1514560e6e96Seschrock * it's still faulted. 1515560e6e96Seschrock */ 1516560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1517560e6e96Seschrock 1518fa9e4066Sahrens if (vd->vdev_offline) 1519fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1520fa9e4066Sahrens else 1521fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1522ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1523fa9e4066Sahrens } 1524fa9e4066Sahrens 1525dcba9f3fSGeorge Wilson void 1526dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1527dcba9f3fSGeorge Wilson { 1528dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1529dcba9f3fSGeorge Wilson 1530dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1531dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1532dcba9f3fSGeorge Wilson return; 1533dcba9f3fSGeorge Wilson 1534dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1535dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1536dcba9f3fSGeorge Wilson 1537dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1538dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1539dcba9f3fSGeorge Wilson } 1540dcba9f3fSGeorge Wilson 1541dcba9f3fSGeorge Wilson void 1542dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1543dcba9f3fSGeorge Wilson { 1544dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1545dcba9f3fSGeorge Wilson 1546dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1547dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1548dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1549dcba9f3fSGeorge Wilson 1550dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1551dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1552dcba9f3fSGeorge Wilson } 1553dcba9f3fSGeorge Wilson 1554095bcd66SGeorge Wilson /* 1555095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1556095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1557095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1558095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1559095bcd66SGeorge Wilson */ 1560fa9e4066Sahrens void 1561ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1562fa9e4066Sahrens { 1563ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1564fa9e4066Sahrens 1565e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1566ea8dc4b6Seschrock 15671195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 15681195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1569fa9e4066Sahrens vdev_close(vd); 1570fa9e4066Sahrens (void) vdev_open(vd); 1571fa9e4066Sahrens 157239c23413Seschrock /* 157339c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 157439c23413Seschrock * Otherwise, a device with an invalid label could be successfully 157539c23413Seschrock * opened in response to vdev_reopen(). 157639c23413Seschrock */ 1577c5904d13Seschrock if (vd->vdev_aux) { 1578c5904d13Seschrock (void) vdev_validate_aux(vd); 1579e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 15806809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1581573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1582573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1583c5904d13Seschrock } else { 1584bda88194SGeorge Wilson (void) vdev_validate(vd, B_TRUE); 1585c5904d13Seschrock } 158639c23413Seschrock 1587fa9e4066Sahrens /* 15883d7072f8Seschrock * Reassess parent vdev's health. 1589fa9e4066Sahrens */ 15903d7072f8Seschrock vdev_propagate_state(vd); 1591fa9e4066Sahrens } 1592fa9e4066Sahrens 1593fa9e4066Sahrens int 159499653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1595fa9e4066Sahrens { 1596fa9e4066Sahrens int error; 1597fa9e4066Sahrens 1598fa9e4066Sahrens /* 1599fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1600fa9e4066Sahrens * For a create, however, we want to fail the request if 1601fa9e4066Sahrens * there are any components we can't open. 1602fa9e4066Sahrens */ 1603fa9e4066Sahrens error = vdev_open(vd); 1604fa9e4066Sahrens 1605fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1606fa9e4066Sahrens vdev_close(vd); 1607fa9e4066Sahrens return (error ? error : ENXIO); 1608fa9e4066Sahrens } 1609fa9e4066Sahrens 1610fa9e4066Sahrens /* 16110713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1612fa9e4066Sahrens */ 16130713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 16140713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 161539c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1616fa9e4066Sahrens vdev_close(vd); 1617fa9e4066Sahrens return (error); 1618fa9e4066Sahrens } 1619fa9e4066Sahrens 1620fa9e4066Sahrens return (0); 1621fa9e4066Sahrens } 1622fa9e4066Sahrens 16230e34b6a7Sbonwick void 1624573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1625fa9e4066Sahrens { 1626fa9e4066Sahrens /* 1627bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1628fa9e4066Sahrens */ 1629bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1630fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1631fa9e4066Sahrens } 1632fa9e4066Sahrens 1633fa9e4066Sahrens void 1634ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1635fa9e4066Sahrens { 1636ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 163788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1638ecc2d604Sbonwick ASSERT(ISP2(flags)); 1639f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1640fa9e4066Sahrens 1641ecc2d604Sbonwick if (flags & VDD_METASLAB) 1642ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1643ecc2d604Sbonwick 1644ecc2d604Sbonwick if (flags & VDD_DTL) 1645ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1646ecc2d604Sbonwick 1647ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1648fa9e4066Sahrens } 1649fa9e4066Sahrens 16500713e232SGeorge Wilson void 16510713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 16520713e232SGeorge Wilson { 16530713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16540713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 16550713e232SGeorge Wilson 16560713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 16570713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 16580713e232SGeorge Wilson } 16590713e232SGeorge Wilson 16608ad4d6ddSJeff Bonwick /* 16618ad4d6ddSJeff Bonwick * DTLs. 16628ad4d6ddSJeff Bonwick * 16638ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 16649fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 16658ad4d6ddSJeff Bonwick * 16668ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 16678ad4d6ddSJeff Bonwick * 16688ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 16698ad4d6ddSJeff Bonwick * 16708ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 16718ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 16728ad4d6ddSJeff Bonwick * txgs that was scrubbed. 16738ad4d6ddSJeff Bonwick * 16748ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 16758ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 16768ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 16778ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 16788ad4d6ddSJeff Bonwick * determine whether a device can be detached. 16798ad4d6ddSJeff Bonwick * 16808ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 16818ad4d6ddSJeff Bonwick * either has the data or it doesn't. 16828ad4d6ddSJeff Bonwick * 16838ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 16848ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 16858ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 16868ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 16878ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 16888ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 16898ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 16908ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 16918ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 16928ad4d6ddSJeff Bonwick * 16938ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 16948ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 16958ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 16968ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 16978ad4d6ddSJeff Bonwick */ 1698fa9e4066Sahrens void 16998ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1700fa9e4066Sahrens { 17010713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17028ad4d6ddSJeff Bonwick 17038ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 17048ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1705f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 17068ad4d6ddSJeff Bonwick 17070713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17080713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 17090713e232SGeorge Wilson range_tree_add(rt, txg, size); 17100713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1711fa9e4066Sahrens } 1712fa9e4066Sahrens 17138ad4d6ddSJeff Bonwick boolean_t 17148ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1715fa9e4066Sahrens { 17160713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17178ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1718fa9e4066Sahrens 17198ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 17208ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1721fa9e4066Sahrens 17220713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17230713e232SGeorge Wilson if (range_tree_space(rt) != 0) 17240713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 17250713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1726fa9e4066Sahrens 1727fa9e4066Sahrens return (dirty); 1728fa9e4066Sahrens } 1729fa9e4066Sahrens 17308ad4d6ddSJeff Bonwick boolean_t 17318ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 17328ad4d6ddSJeff Bonwick { 17330713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17348ad4d6ddSJeff Bonwick boolean_t empty; 17358ad4d6ddSJeff Bonwick 17360713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17370713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 17380713e232SGeorge Wilson mutex_exit(rt->rt_lock); 17398ad4d6ddSJeff Bonwick 17408ad4d6ddSJeff Bonwick return (empty); 17418ad4d6ddSJeff Bonwick } 17428ad4d6ddSJeff Bonwick 1743b4952e17SGeorge Wilson /* 1744b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 1745b4952e17SGeorge Wilson */ 1746b4952e17SGeorge Wilson static uint64_t 1747b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 1748b4952e17SGeorge Wilson { 17490713e232SGeorge Wilson range_seg_t *rs; 1750b4952e17SGeorge Wilson 1751b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17520713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1753b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1754b4952e17SGeorge Wilson 17550713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17560713e232SGeorge Wilson return (rs->rs_start - 1); 1757b4952e17SGeorge Wilson } 1758b4952e17SGeorge Wilson 1759b4952e17SGeorge Wilson /* 1760b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 1761b4952e17SGeorge Wilson */ 1762b4952e17SGeorge Wilson static uint64_t 1763b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 1764b4952e17SGeorge Wilson { 17650713e232SGeorge Wilson range_seg_t *rs; 1766b4952e17SGeorge Wilson 1767b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17680713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1769b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1770b4952e17SGeorge Wilson 17710713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17720713e232SGeorge Wilson return (rs->rs_end); 1773b4952e17SGeorge Wilson } 1774b4952e17SGeorge Wilson 1775b4952e17SGeorge Wilson /* 1776b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 1777b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 1778b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 1779b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 1780b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 1781b4952e17SGeorge Wilson * excise the DTLs. 1782b4952e17SGeorge Wilson */ 1783b4952e17SGeorge Wilson static boolean_t 1784b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 1785b4952e17SGeorge Wilson { 1786b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 1787b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1788b4952e17SGeorge Wilson 1789b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 1790b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1791b4952e17SGeorge Wilson 1792b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 17930713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1794b4952e17SGeorge Wilson return (B_TRUE); 1795b4952e17SGeorge Wilson 1796b4952e17SGeorge Wilson /* 1797b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 1798b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 1799b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 1800b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 1801b4952e17SGeorge Wilson * for excision. 1802b4952e17SGeorge Wilson */ 1803b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1804b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1805b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1806b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1807b4952e17SGeorge Wilson return (B_TRUE); 1808b4952e17SGeorge Wilson } 1809b4952e17SGeorge Wilson return (B_FALSE); 1810b4952e17SGeorge Wilson } 1811b4952e17SGeorge Wilson 1812fa9e4066Sahrens /* 1813fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1814fa9e4066Sahrens */ 1815fa9e4066Sahrens void 1816fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1817fa9e4066Sahrens { 1818ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 18198ad4d6ddSJeff Bonwick avl_tree_t reftree; 18208ad4d6ddSJeff Bonwick int minref; 1821fa9e4066Sahrens 18228ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1823fa9e4066Sahrens 18248ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 18258ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 18268ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 18278ad4d6ddSJeff Bonwick 1828b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 18298ad4d6ddSJeff Bonwick return; 18308ad4d6ddSJeff Bonwick 18318ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 18323f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 18333f9d6ad7SLin Ling 1834fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1835b4952e17SGeorge Wilson 1836b4952e17SGeorge Wilson /* 1837b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 1838b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 1839b4952e17SGeorge Wilson * excise regions on vdevs that were available during 1840b4952e17SGeorge Wilson * the entire duration of this scan. 1841b4952e17SGeorge Wilson */ 1842088f3894Sahrens if (scrub_txg != 0 && 18433f9d6ad7SLin Ling (spa->spa_scrub_started || 1844b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1845b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 1846088f3894Sahrens /* 1847088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1848088f3894Sahrens * did it without rebooting, then the scrub dtl 1849088f3894Sahrens * will be valid, so excise the old region and 1850088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1851088f3894Sahrens * dtl as-is if there was an error. 18528ad4d6ddSJeff Bonwick * 18538ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 18548ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 18558ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 18568ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 18578ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 18588ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 18598ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 18608ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 18618ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1862088f3894Sahrens */ 18630713e232SGeorge Wilson space_reftree_create(&reftree); 18640713e232SGeorge Wilson space_reftree_add_map(&reftree, 18650713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18660713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 18670713e232SGeorge Wilson space_reftree_add_map(&reftree, 18680713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 18690713e232SGeorge Wilson space_reftree_generate_map(&reftree, 18700713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18710713e232SGeorge Wilson space_reftree_destroy(&reftree); 1872fa9e4066Sahrens } 18730713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 18740713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18750713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1876fa9e4066Sahrens if (scrub_done) 18770713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 18780713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 18798ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 18800713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 18818ad4d6ddSJeff Bonwick else 18820713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18830713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1884b4952e17SGeorge Wilson 1885b4952e17SGeorge Wilson /* 1886b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 1887b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 1888b4952e17SGeorge Wilson */ 1889b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 18900713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 18910713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 1892b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 1893b4952e17SGeorge Wilson 1894fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1895088f3894Sahrens 1896ecc2d604Sbonwick if (txg != 0) 1897ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1898fa9e4066Sahrens return; 1899fa9e4066Sahrens } 1900fa9e4066Sahrens 1901fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 19028ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 190399bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 190499bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 19058ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 19068ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 19078ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 19088ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 19098ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 19108ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 19118ad4d6ddSJeff Bonwick else 19128ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 19130713e232SGeorge Wilson space_reftree_create(&reftree); 19148ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 19158ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 19168ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 19170713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 19188ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 19198ad4d6ddSJeff Bonwick } 19200713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 19210713e232SGeorge Wilson space_reftree_destroy(&reftree); 1922fa9e4066Sahrens } 19238ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1924fa9e4066Sahrens } 1925fa9e4066Sahrens 19260713e232SGeorge Wilson int 1927fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1928fa9e4066Sahrens { 1929fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1930ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19310713e232SGeorge Wilson int error = 0; 1932fa9e4066Sahrens 19330713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 19340713e232SGeorge Wilson ASSERT(!vd->vdev_ishole); 1935fa9e4066Sahrens 19360713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 19370713e232SGeorge Wilson vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 19380713e232SGeorge Wilson if (error) 19390713e232SGeorge Wilson return (error); 19400713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1941fa9e4066Sahrens 19420713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 194388ecc943SGeorge Wilson 19440713e232SGeorge Wilson /* 19450713e232SGeorge Wilson * Now that we've opened the space_map we need to update 19460713e232SGeorge Wilson * the in-core DTL. 19470713e232SGeorge Wilson */ 19480713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 1949ecc2d604Sbonwick 19500713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 19510713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 19520713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1953fa9e4066Sahrens 19540713e232SGeorge Wilson return (error); 19550713e232SGeorge Wilson } 19560713e232SGeorge Wilson 19570713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 19580713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 19590713e232SGeorge Wilson if (error != 0) 19600713e232SGeorge Wilson break; 19610713e232SGeorge Wilson } 1962fa9e4066Sahrens 1963fa9e4066Sahrens return (error); 1964fa9e4066Sahrens } 1965fa9e4066Sahrens 1966215198a6SJoe Stein void 1967215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 1968215198a6SJoe Stein { 1969215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 1970215198a6SJoe Stein 1971215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 1972215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 1973215198a6SJoe Stein zapobj, tx)); 1974215198a6SJoe Stein } 1975215198a6SJoe Stein 1976215198a6SJoe Stein uint64_t 1977215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 1978215198a6SJoe Stein { 1979215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 1980215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 1981215198a6SJoe Stein DMU_OT_NONE, 0, tx); 1982215198a6SJoe Stein 1983215198a6SJoe Stein ASSERT(zap != 0); 1984215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 1985215198a6SJoe Stein zap, tx)); 1986215198a6SJoe Stein 1987215198a6SJoe Stein return (zap); 1988215198a6SJoe Stein } 1989215198a6SJoe Stein 1990215198a6SJoe Stein void 1991215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 1992215198a6SJoe Stein { 1993215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 1994215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 1995215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 1996215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 1997215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 1998215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 1999215198a6SJoe Stein } 2000215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2001215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2002215198a6SJoe Stein } 2003215198a6SJoe Stein } 2004215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2005215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2006215198a6SJoe Stein } 2007215198a6SJoe Stein } 2008215198a6SJoe Stein 2009fa9e4066Sahrens void 2010fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2011fa9e4066Sahrens { 2012fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 20130713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2014ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 20150713e232SGeorge Wilson range_tree_t *rtsync; 20160713e232SGeorge Wilson kmutex_t rtlock; 2017fa9e4066Sahrens dmu_tx_t *tx; 20180713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2019fa9e4066Sahrens 202088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 20210713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 202288ecc943SGeorge Wilson 2023fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2024fa9e4066Sahrens 20250713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 20260713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20270713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 20280713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 20290713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 20300713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2031215198a6SJoe Stein 2032215198a6SJoe Stein /* 2033215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2034215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2035215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2036215198a6SJoe Stein */ 2037215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2038215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2039215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2040215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2041215198a6SJoe Stein } 2042215198a6SJoe Stein 2043fa9e4066Sahrens dmu_tx_commit(tx); 2044fa9e4066Sahrens return; 2045fa9e4066Sahrens } 2046fa9e4066Sahrens 20470713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 20480713e232SGeorge Wilson uint64_t new_object; 20490713e232SGeorge Wilson 20500713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 20510713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 20520713e232SGeorge Wilson 20530713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 20540713e232SGeorge Wilson 0, -1ULL, 0, &vd->vdev_dtl_lock)); 20550713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2056fa9e4066Sahrens } 2057fa9e4066Sahrens 20580713e232SGeorge Wilson mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 2059fa9e4066Sahrens 20600713e232SGeorge Wilson rtsync = range_tree_create(NULL, NULL, &rtlock); 2061fa9e4066Sahrens 20620713e232SGeorge Wilson mutex_enter(&rtlock); 2063fa9e4066Sahrens 2064fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 20650713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2066fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2067fa9e4066Sahrens 20680713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 20690713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 20700713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2071fa9e4066Sahrens 20720713e232SGeorge Wilson range_tree_destroy(rtsync); 2073fa9e4066Sahrens 20740713e232SGeorge Wilson mutex_exit(&rtlock); 20750713e232SGeorge Wilson mutex_destroy(&rtlock); 2076fa9e4066Sahrens 20770713e232SGeorge Wilson /* 20780713e232SGeorge Wilson * If the object for the space map has changed then dirty 20790713e232SGeorge Wilson * the top level so that we update the config. 20800713e232SGeorge Wilson */ 20810713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 20820713e232SGeorge Wilson zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 20830713e232SGeorge Wilson "new object %llu", txg, spa_name(spa), object, 20840713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 20850713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 20860713e232SGeorge Wilson } 2087fa9e4066Sahrens 2088fa9e4066Sahrens dmu_tx_commit(tx); 20890713e232SGeorge Wilson 20900713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20910713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 20920713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2093fa9e4066Sahrens } 2094fa9e4066Sahrens 20958ad4d6ddSJeff Bonwick /* 20968ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 20978ad4d6ddSJeff Bonwick * without losing data. 20988ad4d6ddSJeff Bonwick */ 20998ad4d6ddSJeff Bonwick boolean_t 21008ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 21018ad4d6ddSJeff Bonwick { 21028ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 21038ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 21048ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 21058ad4d6ddSJeff Bonwick boolean_t required; 21068ad4d6ddSJeff Bonwick 21078ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 21088ad4d6ddSJeff Bonwick 21098ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 21108ad4d6ddSJeff Bonwick return (B_TRUE); 21118ad4d6ddSJeff Bonwick 21128ad4d6ddSJeff Bonwick /* 21138ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 21148ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 21158ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 21168ad4d6ddSJeff Bonwick */ 21178ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 21188ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 21198ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 21208ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 21218ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 21228ad4d6ddSJeff Bonwick 2123cb04b873SMark J Musante if (!required && zio_injection_enabled) 2124cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2125cb04b873SMark J Musante 21268ad4d6ddSJeff Bonwick return (required); 21278ad4d6ddSJeff Bonwick } 21288ad4d6ddSJeff Bonwick 2129088f3894Sahrens /* 2130088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2131088f3894Sahrens */ 2132088f3894Sahrens boolean_t 2133088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2134088f3894Sahrens { 2135088f3894Sahrens boolean_t needed = B_FALSE; 2136088f3894Sahrens uint64_t thismin = UINT64_MAX; 2137088f3894Sahrens uint64_t thismax = 0; 2138088f3894Sahrens 2139088f3894Sahrens if (vd->vdev_children == 0) { 2140088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 21410713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 21428ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2143088f3894Sahrens 2144b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2145b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2146088f3894Sahrens needed = B_TRUE; 2147088f3894Sahrens } 2148088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2149088f3894Sahrens } else { 21508ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2151088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2152088f3894Sahrens uint64_t cmin, cmax; 2153088f3894Sahrens 2154088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2155088f3894Sahrens thismin = MIN(thismin, cmin); 2156088f3894Sahrens thismax = MAX(thismax, cmax); 2157088f3894Sahrens needed = B_TRUE; 2158088f3894Sahrens } 2159088f3894Sahrens } 2160088f3894Sahrens } 2161088f3894Sahrens 2162088f3894Sahrens if (needed && minp) { 2163088f3894Sahrens *minp = thismin; 2164088f3894Sahrens *maxp = thismax; 2165088f3894Sahrens } 2166088f3894Sahrens return (needed); 2167088f3894Sahrens } 2168088f3894Sahrens 2169560e6e96Seschrock void 2170ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2171fa9e4066Sahrens { 2172fa9e4066Sahrens /* 2173fa9e4066Sahrens * Recursively load all children. 2174fa9e4066Sahrens */ 21758ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2176560e6e96Seschrock vdev_load(vd->vdev_child[c]); 2177fa9e4066Sahrens 2178fa9e4066Sahrens /* 21790e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2180fa9e4066Sahrens */ 218188ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 2182560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2183560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 2184560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2185560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2186fa9e4066Sahrens 2187fa9e4066Sahrens /* 2188fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2189fa9e4066Sahrens */ 2190560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2191560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2192560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2193fa9e4066Sahrens } 2194fa9e4066Sahrens 219599653d4eSeschrock /* 2196fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2197fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2198fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2199fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2200fa94a07fSbrendan * repurposed to another pool. 220199653d4eSeschrock */ 220299653d4eSeschrock int 2203fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 220499653d4eSeschrock { 220599653d4eSeschrock nvlist_t *label; 220699653d4eSeschrock uint64_t guid, version; 220799653d4eSeschrock uint64_t state; 220899653d4eSeschrock 2209e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2210c5904d13Seschrock return (0); 2211c5904d13Seschrock 2212dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 221399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 221499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 221599653d4eSeschrock return (-1); 221699653d4eSeschrock } 221799653d4eSeschrock 221899653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2219ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 222099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 222199653d4eSeschrock guid != vd->vdev_guid || 222299653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 222399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 222499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 222599653d4eSeschrock nvlist_free(label); 222699653d4eSeschrock return (-1); 222799653d4eSeschrock } 222899653d4eSeschrock 222999653d4eSeschrock /* 223099653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 223199653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 223299653d4eSeschrock */ 223399653d4eSeschrock nvlist_free(label); 223499653d4eSeschrock return (0); 223599653d4eSeschrock } 223699653d4eSeschrock 223788ecc943SGeorge Wilson void 223888ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 223988ecc943SGeorge Wilson { 224088ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 224188ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 224288ecc943SGeorge Wilson dmu_tx_t *tx; 224388ecc943SGeorge Wilson 224488ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2245215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 2246215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 224788ecc943SGeorge Wilson 224888ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 22492e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 22502e4c9986SGeorge Wilson 22512e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 22522e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 22532e4c9986SGeorge Wilson 225488ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 225588ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 225688ecc943SGeorge Wilson 22570713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 225888ecc943SGeorge Wilson continue; 225988ecc943SGeorge Wilson 22600713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 22612e4c9986SGeorge Wilson /* 22622e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 22632e4c9986SGeorge Wilson * was removed then the histogram accounting may 22642e4c9986SGeorge Wilson * not be accurate. Update the histogram information 22652e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 22662e4c9986SGeorge Wilson * and metaslab class are up-to-date. 22672e4c9986SGeorge Wilson */ 22682e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 22692e4c9986SGeorge Wilson 22700713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 22710713e232SGeorge Wilson space_map_free(msp->ms_sm, tx); 22720713e232SGeorge Wilson space_map_close(msp->ms_sm); 22730713e232SGeorge Wilson msp->ms_sm = NULL; 22740713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 227588ecc943SGeorge Wilson } 22762e4c9986SGeorge Wilson 22772e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 22782e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 22792e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 22802e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 22812e4c9986SGeorge Wilson 228288ecc943SGeorge Wilson } 228388ecc943SGeorge Wilson 228488ecc943SGeorge Wilson if (vd->vdev_ms_array) { 228588ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 228688ecc943SGeorge Wilson vd->vdev_ms_array = 0; 228788ecc943SGeorge Wilson } 2288215198a6SJoe Stein 2289215198a6SJoe Stein if (vd->vdev_islog && vd->vdev_top_zap != 0) { 2290215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 2291215198a6SJoe Stein vd->vdev_top_zap = 0; 2292215198a6SJoe Stein } 229388ecc943SGeorge Wilson dmu_tx_commit(tx); 229488ecc943SGeorge Wilson } 229588ecc943SGeorge Wilson 2296fa9e4066Sahrens void 2297fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2298fa9e4066Sahrens { 2299fa9e4066Sahrens metaslab_t *msp; 230080eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2301fa9e4066Sahrens 230288ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 230388ecc943SGeorge Wilson 2304fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2305fa9e4066Sahrens metaslab_sync_done(msp, txg); 230680eb36f2SGeorge Wilson 230780eb36f2SGeorge Wilson if (reassess) 230880eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2309fa9e4066Sahrens } 2310fa9e4066Sahrens 2311fa9e4066Sahrens void 2312fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2313fa9e4066Sahrens { 2314fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2315fa9e4066Sahrens vdev_t *lvd; 2316fa9e4066Sahrens metaslab_t *msp; 2317ecc2d604Sbonwick dmu_tx_t *tx; 2318fa9e4066Sahrens 231988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 232088ecc943SGeorge Wilson 2321ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2322ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2323ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2324ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2325ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2326ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2327ecc2d604Sbonwick vdev_config_dirty(vd); 2328ecc2d604Sbonwick dmu_tx_commit(tx); 2329ecc2d604Sbonwick } 2330fa9e4066Sahrens 23313f9d6ad7SLin Ling /* 23323f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 23333f9d6ad7SLin Ling */ 23343f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 233588ecc943SGeorge Wilson vdev_remove(vd, txg); 233688ecc943SGeorge Wilson 2337ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2338fa9e4066Sahrens metaslab_sync(msp, txg); 2339ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2340ecc2d604Sbonwick } 2341fa9e4066Sahrens 2342fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2343fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2344fa9e4066Sahrens 2345fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2346fa9e4066Sahrens } 2347fa9e4066Sahrens 2348fa9e4066Sahrens uint64_t 2349fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2350fa9e4066Sahrens { 2351fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2352fa9e4066Sahrens } 2353fa9e4066Sahrens 23543d7072f8Seschrock /* 23553d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 23563d7072f8Seschrock * not be opened, and no I/O is attempted. 23573d7072f8Seschrock */ 2358fa9e4066Sahrens int 2359069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2360fa9e4066Sahrens { 23614b964adaSGeorge Wilson vdev_t *vd, *tvd; 2362fa9e4066Sahrens 23638f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2364fa9e4066Sahrens 2365c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2366e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2367e14bb325SJeff Bonwick 23683d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2369e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2370fa9e4066Sahrens 23714b964adaSGeorge Wilson tvd = vd->vdev_top; 23724b964adaSGeorge Wilson 2373069f55e2SEric Schrock /* 2374069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2375069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2376069f55e2SEric Schrock * were faulted. 2377069f55e2SEric Schrock */ 2378069f55e2SEric Schrock vd->vdev_label_aux = aux; 2379069f55e2SEric Schrock 23803d7072f8Seschrock /* 23813d7072f8Seschrock * Faulted state takes precedence over degraded. 23823d7072f8Seschrock */ 238398d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 23843d7072f8Seschrock vd->vdev_faulted = 1ULL; 23853d7072f8Seschrock vd->vdev_degraded = 0ULL; 2386069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 23873d7072f8Seschrock 23883d7072f8Seschrock /* 2389c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2390c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 23913d7072f8Seschrock */ 23924b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 23933d7072f8Seschrock vd->vdev_degraded = 1ULL; 23943d7072f8Seschrock vd->vdev_faulted = 0ULL; 23953d7072f8Seschrock 23963d7072f8Seschrock /* 23973d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 23983d7072f8Seschrock * mark it degraded. 23993d7072f8Seschrock */ 24004b964adaSGeorge Wilson vdev_reopen(tvd); 24013d7072f8Seschrock 2402069f55e2SEric Schrock if (vdev_readable(vd)) 2403069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 24043d7072f8Seschrock } 24053d7072f8Seschrock 2406e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 24073d7072f8Seschrock } 24083d7072f8Seschrock 24093d7072f8Seschrock /* 24103d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 24113d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 24123d7072f8Seschrock * as I/O is concerned. 24133d7072f8Seschrock */ 24143d7072f8Seschrock int 2415069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 24163d7072f8Seschrock { 2417c5904d13Seschrock vdev_t *vd; 24180a4e9518Sgw 24198f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 24203d7072f8Seschrock 2421c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2422e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2423e14bb325SJeff Bonwick 24240e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2425e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 24260e34b6a7Sbonwick 24273d7072f8Seschrock /* 24283d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 24293d7072f8Seschrock */ 2430e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2431e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 24323d7072f8Seschrock 24333d7072f8Seschrock vd->vdev_degraded = 1ULL; 24343d7072f8Seschrock if (!vdev_is_dead(vd)) 24353d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2436069f55e2SEric Schrock aux); 24373d7072f8Seschrock 2438e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 24393d7072f8Seschrock } 24403d7072f8Seschrock 24413d7072f8Seschrock /* 2442f7170741SWill Andrews * Online the given vdev. 2443f7170741SWill Andrews * 2444f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2445f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2446f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2447f7170741SWill Andrews * events are generated if the device fails to open. 24483d7072f8Seschrock */ 24493d7072f8Seschrock int 2450e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 24513d7072f8Seschrock { 2452573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2453*5f368aefSYuri Pankov boolean_t wasoffline; 2454*5f368aefSYuri Pankov vdev_state_t oldstate; 24553d7072f8Seschrock 24568f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 24573d7072f8Seschrock 2458c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2459e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 24603d7072f8Seschrock 24613d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2462e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2463fa9e4066Sahrens 2464*5f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 2465*5f368aefSYuri Pankov oldstate = vd->vdev_state; 246614372834SHans Rosenfeld 2467573ca77eSGeorge Wilson tvd = vd->vdev_top; 2468fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2469441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2470e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2471e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2472573ca77eSGeorge Wilson 2473573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2474573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2475573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2476573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2477573ca77eSGeorge Wilson } 2478573ca77eSGeorge Wilson 2479573ca77eSGeorge Wilson vdev_reopen(tvd); 24803d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 24813d7072f8Seschrock 2482573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2483573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2484573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2485573ca77eSGeorge Wilson } 2486573ca77eSGeorge Wilson 24873d7072f8Seschrock if (newstate) 24883d7072f8Seschrock *newstate = vd->vdev_state; 24893d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 24903d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 24913d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 24923d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 24933d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2494fa9e4066Sahrens 2495573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2496573ca77eSGeorge Wilson 2497573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2498573ca77eSGeorge Wilson if (vd->vdev_aux) 2499573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2500573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2501573ca77eSGeorge Wilson } 250214372834SHans Rosenfeld 2503*5f368aefSYuri Pankov if (wasoffline || 2504*5f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 2505*5f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 250614372834SHans Rosenfeld spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE); 250714372834SHans Rosenfeld 25088ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2509fa9e4066Sahrens } 2510fa9e4066Sahrens 2511a1521560SJeff Bonwick static int 2512a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2513fa9e4066Sahrens { 2514e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 25158f18d1faSGeorge Wilson int error = 0; 25168f18d1faSGeorge Wilson uint64_t generation; 25178f18d1faSGeorge Wilson metaslab_group_t *mg; 25180a4e9518Sgw 25198f18d1faSGeorge Wilson top: 25208f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2521fa9e4066Sahrens 2522c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2523e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2524fa9e4066Sahrens 25250e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2526e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 25270e34b6a7Sbonwick 2528e6ca193dSGeorge Wilson tvd = vd->vdev_top; 25298f18d1faSGeorge Wilson mg = tvd->vdev_mg; 25308f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2531e6ca193dSGeorge Wilson 2532fa9e4066Sahrens /* 2533ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2534fa9e4066Sahrens */ 2535ecc2d604Sbonwick if (!vd->vdev_offline) { 2536ecc2d604Sbonwick /* 25378ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2538e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2539e6ca193dSGeorge Wilson * expendable. 2540ecc2d604Sbonwick */ 2541e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2542e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2543e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2544fa9e4066Sahrens 25458f18d1faSGeorge Wilson /* 2546b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2547b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2548b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2549b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 25508f18d1faSGeorge Wilson */ 25518f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 25528f18d1faSGeorge Wilson /* 25538f18d1faSGeorge Wilson * Prevent any future allocations. 25548f18d1faSGeorge Wilson */ 2555a1521560SJeff Bonwick metaslab_group_passivate(mg); 25568f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 25578f18d1faSGeorge Wilson 25581195e687SMark J Musante error = spa_offline_log(spa); 25598f18d1faSGeorge Wilson 25608f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 25618f18d1faSGeorge Wilson 25628f18d1faSGeorge Wilson /* 25638f18d1faSGeorge Wilson * Check to see if the config has changed. 25648f18d1faSGeorge Wilson */ 25658f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2566a1521560SJeff Bonwick metaslab_group_activate(mg); 25678f18d1faSGeorge Wilson if (error) 25688f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 25698f18d1faSGeorge Wilson vd, error)); 25708f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 25718f18d1faSGeorge Wilson goto top; 25728f18d1faSGeorge Wilson } 2573fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 25748f18d1faSGeorge Wilson } 25758f18d1faSGeorge Wilson 2576ecc2d604Sbonwick /* 2577ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2578e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2579e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2580e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2581ecc2d604Sbonwick */ 2582ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2583e6ca193dSGeorge Wilson vdev_reopen(tvd); 2584e6ca193dSGeorge Wilson 2585e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2586e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2587ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2588e6ca193dSGeorge Wilson vdev_reopen(tvd); 2589e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2590ecc2d604Sbonwick } 25918f18d1faSGeorge Wilson 25928f18d1faSGeorge Wilson /* 25938f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 25948f18d1faSGeorge Wilson * once we online the device it's open for business. 25958f18d1faSGeorge Wilson */ 25968f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2597a1521560SJeff Bonwick metaslab_group_activate(mg); 2598fa9e4066Sahrens } 2599fa9e4066Sahrens 2600e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2601ecc2d604Sbonwick 26028f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2603fa9e4066Sahrens } 2604fa9e4066Sahrens 2605a1521560SJeff Bonwick int 2606a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2607a1521560SJeff Bonwick { 2608a1521560SJeff Bonwick int error; 2609a1521560SJeff Bonwick 2610a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2611a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2612a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2613a1521560SJeff Bonwick 2614a1521560SJeff Bonwick return (error); 2615a1521560SJeff Bonwick } 2616a1521560SJeff Bonwick 2617ea8dc4b6Seschrock /* 2618ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2619ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2620ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2621ea8dc4b6Seschrock */ 2622ea8dc4b6Seschrock void 2623e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2624fa9e4066Sahrens { 2625e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2626e14bb325SJeff Bonwick 2627e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2628fa9e4066Sahrens 2629ea8dc4b6Seschrock if (vd == NULL) 2630e14bb325SJeff Bonwick vd = rvd; 2631fa9e4066Sahrens 2632ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2633ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2634ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2635fa9e4066Sahrens 2636e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2637e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 26383d7072f8Seschrock 26393d7072f8Seschrock /* 26408a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 26418a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 26428a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 26438a79c1b5Sek * written out to disk. 26443d7072f8Seschrock */ 2645e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2646e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 26478a79c1b5Sek 2648096d22d4SEric Schrock /* 2649096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2650096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2651096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2652096d22d4SEric Schrock */ 2653096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2654096d22d4SEric Schrock 26554b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2656e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2657e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2658e14bb325SJeff Bonwick 2659f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 26603d7072f8Seschrock 2661096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2662096d22d4SEric Schrock 2663f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2664e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2665e14bb325SJeff Bonwick 2666e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2667bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 26683d7072f8Seschrock 26693d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 26703d7072f8Seschrock } 2671096d22d4SEric Schrock 2672096d22d4SEric Schrock /* 2673096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2674096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2675096d22d4SEric Schrock * done in response to the FMA fault. 2676096d22d4SEric Schrock */ 2677096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2678096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2679096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2680096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2681fa9e4066Sahrens } 2682fa9e4066Sahrens 2683e14bb325SJeff Bonwick boolean_t 2684e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 26850a4e9518Sgw { 268688ecc943SGeorge Wilson /* 268788ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 268888ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 268988ecc943SGeorge Wilson * these types of devices in the various code paths. 269088ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 269188ecc943SGeorge Wilson * before issuing I/O to them. 269288ecc943SGeorge Wilson */ 269388ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 269488ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 26950a4e9518Sgw } 26960a4e9518Sgw 2697e14bb325SJeff Bonwick boolean_t 2698e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 26990a4e9518Sgw { 2700e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 27010a4e9518Sgw } 27020a4e9518Sgw 2703e14bb325SJeff Bonwick boolean_t 2704e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2705fa9e4066Sahrens { 2706e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2707fa9e4066Sahrens } 2708fa9e4066Sahrens 2709a31e6787SGeorge Wilson boolean_t 2710a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2711a31e6787SGeorge Wilson { 27128ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 27138ad4d6ddSJeff Bonwick 2714a31e6787SGeorge Wilson /* 27158ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2716a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2717a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 27188ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 27198ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 27208ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2721a31e6787SGeorge Wilson */ 27228ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 27230f7643c7SGeorge Wilson !vd->vdev_cant_write && !vd->vdev_ishole && 27240f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 2725a31e6787SGeorge Wilson } 2726a31e6787SGeorge Wilson 2727e14bb325SJeff Bonwick boolean_t 2728e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2729fa9e4066Sahrens { 2730e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2731fa9e4066Sahrens 2732e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2733e14bb325SJeff Bonwick return (B_FALSE); 2734fa9e4066Sahrens 2735e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2736e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2737fa9e4066Sahrens 2738e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2739e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2740fa9e4066Sahrens 2741e14bb325SJeff Bonwick return (B_TRUE); 2742fa9e4066Sahrens } 2743fa9e4066Sahrens 2744fa9e4066Sahrens /* 2745fa9e4066Sahrens * Get statistics for the given vdev. 2746fa9e4066Sahrens */ 2747fa9e4066Sahrens void 2748fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2749fa9e4066Sahrens { 27502e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 27512e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 2752c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 27532e4c9986SGeorge Wilson 27542e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2755fa9e4066Sahrens 2756fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2757fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2758fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2759fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2760573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2761573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2762573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2763c39a2aaeSGeorge Wilson /* 2764c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 2765c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 2766c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 2767c39a2aaeSGeorge Wilson */ 2768c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 2769c39a2aaeSGeorge Wilson vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize, 2770c39a2aaeSGeorge Wilson 1ULL << tvd->vdev_ms_shift); 2771c39a2aaeSGeorge Wilson } 27722986efa8SAlex Reece if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 27732e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 27742986efa8SAlex Reece } 2775fa9e4066Sahrens 2776fa9e4066Sahrens /* 2777fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2778fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2779fa9e4066Sahrens */ 2780fa9e4066Sahrens if (vd == rvd) { 2781e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2782fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2783fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2784fa9e4066Sahrens 2785e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2786fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2787fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2788fa9e4066Sahrens } 27893f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2790fa9e4066Sahrens } 2791fa9e4066Sahrens } 27922e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 2793fa9e4066Sahrens } 2794fa9e4066Sahrens 2795fa94a07fSbrendan void 2796fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2797fa94a07fSbrendan { 2798fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2799fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2800fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2801fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2802fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2803fa94a07fSbrendan } 2804fa94a07fSbrendan 28053f9d6ad7SLin Ling void 28063f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 28073f9d6ad7SLin Ling { 28083f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 28093f9d6ad7SLin Ling 28103f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 28113f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 28123f9d6ad7SLin Ling 28133f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 28143f9d6ad7SLin Ling vs->vs_scan_processed = 0; 28153f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 28163f9d6ad7SLin Ling } 28173f9d6ad7SLin Ling 2818fa9e4066Sahrens void 2819e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2820fa9e4066Sahrens { 28218ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 28228ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2823e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2824fa9e4066Sahrens vdev_t *pvd; 2825fa9e4066Sahrens uint64_t txg = zio->io_txg; 2826fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2827fa9e4066Sahrens zio_type_t type = zio->io_type; 2828fa9e4066Sahrens int flags = zio->io_flags; 2829fa9e4066Sahrens 2830e14bb325SJeff Bonwick /* 2831e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2832e14bb325SJeff Bonwick */ 2833e14bb325SJeff Bonwick if (zio->io_gang_tree) 2834e14bb325SJeff Bonwick return; 2835e14bb325SJeff Bonwick 2836fa9e4066Sahrens if (zio->io_error == 0) { 2837e14bb325SJeff Bonwick /* 2838e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2839e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2840e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2841e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2842e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2843e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2844e14bb325SJeff Bonwick * remain zero, which is what we want.) 2845e14bb325SJeff Bonwick * 2846e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2847e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2848e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2849e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2850e14bb325SJeff Bonwick */ 2851e14bb325SJeff Bonwick if (vd == rvd) 2852e14bb325SJeff Bonwick return; 2853e14bb325SJeff Bonwick 2854e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 28558ad4d6ddSJeff Bonwick 28568ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 28578ad4d6ddSJeff Bonwick return; 28588ad4d6ddSJeff Bonwick 28598ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 28608ad4d6ddSJeff Bonwick 2861e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 286244ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 28633f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 28643f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 28653f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 28663f9d6ad7SLin Ling 28673f9d6ad7SLin Ling /* XXX cleanup? */ 28683f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 28693f9d6ad7SLin Ling atomic_add_64(processed, psize); 28703f9d6ad7SLin Ling vs->vs_scan_processed += psize; 28713f9d6ad7SLin Ling } 28723f9d6ad7SLin Ling 28738ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2874e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2875fa9e4066Sahrens } 28768ad4d6ddSJeff Bonwick 28778ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 28788ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 28798ad4d6ddSJeff Bonwick 28808ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2881fa9e4066Sahrens return; 2882fa9e4066Sahrens } 2883fa9e4066Sahrens 2884fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2885fa9e4066Sahrens return; 2886fa9e4066Sahrens 28878956713aSEric Schrock /* 28888956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 28898956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 28908956713aSEric Schrock * hard errors, when in reality they can happen for any number of 28918956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 28928956713aSEric Schrock */ 28938956713aSEric Schrock if (zio->io_error == EIO && 28948956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 28958956713aSEric Schrock return; 28968956713aSEric Schrock 28978f18d1faSGeorge Wilson /* 28988f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 28998f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 29008f18d1faSGeorge Wilson * errors. 29018f18d1faSGeorge Wilson */ 29028f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 29038f18d1faSGeorge Wilson return; 29048f18d1faSGeorge Wilson 2905e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2906b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2907e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2908e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2909e14bb325SJeff Bonwick else 2910e14bb325SJeff Bonwick vs->vs_read_errors++; 2911fa9e4066Sahrens } 2912b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2913e14bb325SJeff Bonwick vs->vs_write_errors++; 2914e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2915fa9e4066Sahrens 29168ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 29178ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 291844ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2919b24ab676SJeff Bonwick spa->spa_claiming)) { 29208ad4d6ddSJeff Bonwick /* 2921b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2922b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2923b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2924b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2925b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2926b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2927b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2928b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 29298ad4d6ddSJeff Bonwick * 29308ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 29318ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 29328ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 29338ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 29348ad4d6ddSJeff Bonwick */ 29358ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 29368ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 293744ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 29388ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 29398ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 29408ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2941b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2942b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2943b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2944b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 29458ad4d6ddSJeff Bonwick } 2946b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 29478ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2948fa9e4066Sahrens return; 29498ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 29508ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 29518ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2952fa9e4066Sahrens } 29538ad4d6ddSJeff Bonwick if (vd != rvd) 29548ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2955fa9e4066Sahrens } 2956fa9e4066Sahrens } 2957fa9e4066Sahrens 2958fa9e4066Sahrens /* 2959b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2960b24ab676SJeff Bonwick * and the root vdev. 2961fa9e4066Sahrens */ 2962fa9e4066Sahrens void 2963b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2964b24ab676SJeff Bonwick int64_t space_delta) 2965fa9e4066Sahrens { 296699653d4eSeschrock int64_t dspace_delta = space_delta; 29678654d025Sperrin spa_t *spa = vd->vdev_spa; 29688654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2969b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2970b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2971fa9e4066Sahrens 29728654d025Sperrin ASSERT(vd == vd->vdev_top); 297399653d4eSeschrock 29748654d025Sperrin /* 29758654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 29768654d025Sperrin * factor. We must calculate this here and not at the root vdev 29778654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 29788654d025Sperrin * childrens', thus not accurate enough for us. 29798654d025Sperrin */ 29808654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2981e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 29828654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 29838654d025Sperrin vd->vdev_deflate_ratio; 29848654d025Sperrin 29858654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 29868654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2987b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 29888654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 29898654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 29908654d025Sperrin 2991b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2992fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2993fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2994b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2995fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2996fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2997fa94a07fSbrendan } 2998b24ab676SJeff Bonwick 2999b24ab676SJeff Bonwick if (mc != NULL) { 3000b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 3001b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 3002b24ab676SJeff Bonwick 3003b24ab676SJeff Bonwick metaslab_class_space_update(mc, 3004b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 3005b24ab676SJeff Bonwick } 3006fa9e4066Sahrens } 3007fa9e4066Sahrens 3008fa9e4066Sahrens /* 3009fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 3010fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 3011fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3012fa9e4066Sahrens */ 3013fa9e4066Sahrens void 3014fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3015fa9e4066Sahrens { 3016fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3017fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3018fa9e4066Sahrens int c; 3019fa9e4066Sahrens 3020f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3021f9af39baSGeorge Wilson 3022c5904d13Seschrock /* 30236809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 30246809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3025c5904d13Seschrock */ 3026c5904d13Seschrock if (vd->vdev_aux != NULL) { 3027c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3028c5904d13Seschrock nvlist_t **aux; 3029c5904d13Seschrock uint_t naux; 3030c5904d13Seschrock 3031c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3032c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3033c5904d13Seschrock break; 3034c5904d13Seschrock } 3035c5904d13Seschrock 3036e14bb325SJeff Bonwick if (c == sav->sav_count) { 3037e14bb325SJeff Bonwick /* 3038e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3039e14bb325SJeff Bonwick */ 3040e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3041e14bb325SJeff Bonwick return; 3042e14bb325SJeff Bonwick } 3043e14bb325SJeff Bonwick 3044c5904d13Seschrock sav->sav_sync = B_TRUE; 3045c5904d13Seschrock 30466809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 30476809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 30486809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 30496809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 30506809eb4eSEric Schrock } 3051c5904d13Seschrock 3052c5904d13Seschrock ASSERT(c < naux); 3053c5904d13Seschrock 3054c5904d13Seschrock /* 3055c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3056c5904d13Seschrock * sketchy, but it will work. 3057c5904d13Seschrock */ 3058c5904d13Seschrock nvlist_free(aux[c]); 30593f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3060c5904d13Seschrock 3061c5904d13Seschrock return; 3062c5904d13Seschrock } 3063c5904d13Seschrock 30645dabedeeSbonwick /* 3065e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3066e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3067e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 30685dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 30695dabedeeSbonwick */ 3070e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3071e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3072e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 30735dabedeeSbonwick 3074fa9e4066Sahrens if (vd == rvd) { 3075fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 3076fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 3077fa9e4066Sahrens } else { 3078fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 3079fa9e4066Sahrens 308088ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 308188ecc943SGeorge Wilson !vd->vdev_ishole) 3082e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 3083fa9e4066Sahrens } 3084fa9e4066Sahrens } 3085fa9e4066Sahrens 3086fa9e4066Sahrens void 3087fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 3088fa9e4066Sahrens { 30895dabedeeSbonwick spa_t *spa = vd->vdev_spa; 30905dabedeeSbonwick 3091e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3092e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3093e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 30945dabedeeSbonwick 3095e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3096e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 3097e14bb325SJeff Bonwick } 3098e14bb325SJeff Bonwick 3099e14bb325SJeff Bonwick /* 3100e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 3101e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3102e14bb325SJeff Bonwick * the state changes from larger config changes because they require 3103e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3104e14bb325SJeff Bonwick */ 3105e14bb325SJeff Bonwick void 3106e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3107e14bb325SJeff Bonwick { 3108e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3109e14bb325SJeff Bonwick 3110f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3111e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3112e14bb325SJeff Bonwick 3113e14bb325SJeff Bonwick /* 3114e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3115e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3116e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3117e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 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 3123b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 3124e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3125e14bb325SJeff Bonwick } 3126e14bb325SJeff Bonwick 3127e14bb325SJeff Bonwick void 3128e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3129e14bb325SJeff Bonwick { 3130e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3131e14bb325SJeff Bonwick 3132e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3133e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3134e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3135e14bb325SJeff Bonwick 3136e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3137e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3138fa9e4066Sahrens } 3139fa9e4066Sahrens 314032b87932Sek /* 314132b87932Sek * Propagate vdev state up from children to parent. 314232b87932Sek */ 314344cd46caSbillm void 314444cd46caSbillm vdev_propagate_state(vdev_t *vd) 314544cd46caSbillm { 31468ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 31478ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 314844cd46caSbillm int degraded = 0, faulted = 0; 314944cd46caSbillm int corrupted = 0; 315044cd46caSbillm vdev_t *child; 315144cd46caSbillm 31523d7072f8Seschrock if (vd->vdev_children > 0) { 3153573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 31543d7072f8Seschrock child = vd->vdev_child[c]; 315551ece835Seschrock 315688ecc943SGeorge Wilson /* 315788ecc943SGeorge Wilson * Don't factor holes into the decision. 315888ecc943SGeorge Wilson */ 315988ecc943SGeorge Wilson if (child->vdev_ishole) 316088ecc943SGeorge Wilson continue; 316188ecc943SGeorge Wilson 3162e14bb325SJeff Bonwick if (!vdev_readable(child) || 31638ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 316451ece835Seschrock /* 316551ece835Seschrock * Root special: if there is a top-level log 316651ece835Seschrock * device, treat the root vdev as if it were 316751ece835Seschrock * degraded. 316851ece835Seschrock */ 316951ece835Seschrock if (child->vdev_islog && vd == rvd) 317051ece835Seschrock degraded++; 317151ece835Seschrock else 317251ece835Seschrock faulted++; 317351ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 31743d7072f8Seschrock degraded++; 317551ece835Seschrock } 317644cd46caSbillm 31773d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 31783d7072f8Seschrock corrupted++; 31793d7072f8Seschrock } 318044cd46caSbillm 31813d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 31823d7072f8Seschrock 31833d7072f8Seschrock /* 3184e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 31853d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 31863d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 31873d7072f8Seschrock * replicas'. 31883d7072f8Seschrock */ 31893d7072f8Seschrock if (corrupted && vd == rvd && 31903d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 31913d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 31923d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 31933d7072f8Seschrock } 31943d7072f8Seschrock 319551ece835Seschrock if (vd->vdev_parent) 31963d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 319744cd46caSbillm } 319844cd46caSbillm 3199fa9e4066Sahrens /* 3200ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3201ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3202ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3203ea8dc4b6Seschrock * 3204ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3205ea8dc4b6Seschrock * generated. 3206fa9e4066Sahrens */ 3207fa9e4066Sahrens void 3208ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3209fa9e4066Sahrens { 3210560e6e96Seschrock uint64_t save_state; 3211c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3212ea8dc4b6Seschrock 3213ea8dc4b6Seschrock if (state == vd->vdev_state) { 3214ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3215fa9e4066Sahrens return; 3216ea8dc4b6Seschrock } 3217ea8dc4b6Seschrock 3218560e6e96Seschrock save_state = vd->vdev_state; 3219fa9e4066Sahrens 3220fa9e4066Sahrens vd->vdev_state = state; 3221fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3222fa9e4066Sahrens 32233d7072f8Seschrock /* 32243d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 322598d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 322698d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 322798d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 322898d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 322998d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 323098d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 323198d1cbfeSGeorge Wilson * children. 323298d1cbfeSGeorge Wilson */ 323398d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 323498d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 32353d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 32363d7072f8Seschrock 3237069f55e2SEric Schrock /* 3238069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3239069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3240069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3241069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3242069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3243069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3244069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3245069f55e2SEric Schrock */ 3246069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3247069f55e2SEric Schrock vd->vdev_prevstate != state) 3248069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3249069f55e2SEric Schrock 32503d7072f8Seschrock if (vd->vdev_removed && 32513d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 32523d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 32533d7072f8Seschrock /* 32543d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 32553d7072f8Seschrock * device was previously marked removed and someone attempted to 32563d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 32573d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 32583d7072f8Seschrock * it is one of our special test online cases, which is only 32593d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 32603d7072f8Seschrock * fault. 32613d7072f8Seschrock */ 32623d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 32633d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 32643d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 32653d7072f8Seschrock vd->vdev_removed = B_TRUE; 32663d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3267ea8dc4b6Seschrock /* 3268cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3269cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3270cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3271cb04b873SMark J Musante * is not considered an error. 3272ea8dc4b6Seschrock */ 3273cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3274cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3275560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3276560e6e96Seschrock vd->vdev_not_present = 1; 3277560e6e96Seschrock 3278560e6e96Seschrock /* 3279560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3280560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3281560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3282560e6e96Seschrock * want to post the ereport if the device was already in the 3283560e6e96Seschrock * CANT_OPEN state beforehand. 32843d7072f8Seschrock * 32853d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 32863d7072f8Seschrock * online the device in response to an insertion event. If we 32873d7072f8Seschrock * hit this case, then we have detected an insertion event for a 32883d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 32893d7072f8Seschrock * In this scenario, we don't post an ereport because we are 32903d7072f8Seschrock * about to replace the device, or attempt an online with 32913d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3292560e6e96Seschrock */ 32933d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 32943d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3295c5904d13Seschrock vd != spa->spa_root_vdev) { 3296ea8dc4b6Seschrock const char *class; 3297ea8dc4b6Seschrock 3298ea8dc4b6Seschrock switch (aux) { 3299ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3300ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3301ea8dc4b6Seschrock break; 3302ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3303ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3304ea8dc4b6Seschrock break; 3305ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3306ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3307ea8dc4b6Seschrock break; 3308ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3309ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3310ea8dc4b6Seschrock break; 3311ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3312ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3313ea8dc4b6Seschrock break; 3314ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3315ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3316ea8dc4b6Seschrock break; 3317ea8dc4b6Seschrock default: 3318ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3319ea8dc4b6Seschrock } 3320ea8dc4b6Seschrock 3321c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3322ea8dc4b6Seschrock } 3323ea8dc4b6Seschrock 33243d7072f8Seschrock /* Erase any notion of persistent removed state */ 33253d7072f8Seschrock vd->vdev_removed = B_FALSE; 33263d7072f8Seschrock } else { 33273d7072f8Seschrock vd->vdev_removed = B_FALSE; 33283d7072f8Seschrock } 3329ea8dc4b6Seschrock 33308b33d774STim Haley if (!isopen && vd->vdev_parent) 33318b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3332fa9e4066Sahrens } 333315e6edf1Sgw 333415e6edf1Sgw /* 333515e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 3336c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 3337c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 333815e6edf1Sgw */ 333915e6edf1Sgw boolean_t 334015e6edf1Sgw vdev_is_bootable(vdev_t *vd) 334115e6edf1Sgw { 334215e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 334315e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 334415e6edf1Sgw 334515e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 334615e6edf1Sgw vd->vdev_children > 1) { 334715e6edf1Sgw return (B_FALSE); 3348c8811bd3SToomas Soome } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 334915e6edf1Sgw return (B_FALSE); 335015e6edf1Sgw } 335115e6edf1Sgw } 335215e6edf1Sgw 3353573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 335415e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 335515e6edf1Sgw return (B_FALSE); 335615e6edf1Sgw } 335715e6edf1Sgw return (B_TRUE); 335815e6edf1Sgw } 3359e6ca193dSGeorge Wilson 336088ecc943SGeorge Wilson /* 336188ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 336288ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 33634b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 33644b964adaSGeorge Wilson * device in the current vdev tree (nvd). 336588ecc943SGeorge Wilson */ 3366e6ca193dSGeorge Wilson void 336788ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3368e6ca193dSGeorge Wilson { 336988ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3370e6ca193dSGeorge Wilson 33714b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 337288ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 337388ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3374e6ca193dSGeorge Wilson 337588ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 337688ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3377e6ca193dSGeorge Wilson 33784b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3379e6ca193dSGeorge Wilson /* 33804b964adaSGeorge Wilson * Restore the persistent vdev state 3381e6ca193dSGeorge Wilson */ 338288ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 33834b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 33844b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 33854b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3386e6ca193dSGeorge Wilson } 3387e6ca193dSGeorge Wilson } 3388573ca77eSGeorge Wilson 33894b964adaSGeorge Wilson /* 33904b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 33914b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 33924b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 33934b964adaSGeorge Wilson */ 33944b964adaSGeorge Wilson boolean_t 33954b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 33964b964adaSGeorge Wilson { 33974b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 33984b964adaSGeorge Wilson !vd->vdev_removed) 33994b964adaSGeorge Wilson return (B_TRUE); 34004b964adaSGeorge Wilson 34014b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 34024b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 34034b964adaSGeorge Wilson return (B_TRUE); 34044b964adaSGeorge Wilson 34054b964adaSGeorge Wilson return (B_FALSE); 34064b964adaSGeorge Wilson } 34074b964adaSGeorge Wilson 3408573ca77eSGeorge Wilson /* 3409573ca77eSGeorge Wilson * Expand a vdev if possible. 3410573ca77eSGeorge Wilson */ 3411573ca77eSGeorge Wilson void 3412573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3413573ca77eSGeorge Wilson { 3414573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3415573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3416573ca77eSGeorge Wilson 3417573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3418573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3419573ca77eSGeorge Wilson vdev_config_dirty(vd); 3420573ca77eSGeorge Wilson } 3421573ca77eSGeorge Wilson } 34221195e687SMark J Musante 34231195e687SMark J Musante /* 34241195e687SMark J Musante * Split a vdev. 34251195e687SMark J Musante */ 34261195e687SMark J Musante void 34271195e687SMark J Musante vdev_split(vdev_t *vd) 34281195e687SMark J Musante { 34291195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 34301195e687SMark J Musante 34311195e687SMark J Musante vdev_remove_child(pvd, vd); 34321195e687SMark J Musante vdev_compact_children(pvd); 34331195e687SMark J Musante 34341195e687SMark J Musante cvd = pvd->vdev_child[0]; 34351195e687SMark J Musante if (pvd->vdev_children == 1) { 34361195e687SMark J Musante vdev_remove_parent(cvd); 34371195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 34381195e687SMark J Musante } 34391195e687SMark J Musante vdev_propagate_state(cvd); 34401195e687SMark J Musante } 3441283b8460SGeorge.Wilson 3442283b8460SGeorge.Wilson void 3443283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3444283b8460SGeorge.Wilson { 3445283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3446283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3447283b8460SGeorge.Wilson 3448283b8460SGeorge.Wilson vdev_deadman(cvd); 3449283b8460SGeorge.Wilson } 3450283b8460SGeorge.Wilson 3451283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3452283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3453283b8460SGeorge.Wilson 3454283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 345569962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3456283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3457283b8460SGeorge.Wilson zio_t *fio; 3458283b8460SGeorge.Wilson uint64_t delta; 3459283b8460SGeorge.Wilson 3460283b8460SGeorge.Wilson /* 3461283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3462283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3463283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3464283b8460SGeorge.Wilson */ 346569962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3466c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3467c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 3468c55e05cbSMatthew Ahrens zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3469c55e05cbSMatthew Ahrens "delta %lluns, last io %lluns", 3470283b8460SGeorge.Wilson fio->io_timestamp, delta, 3471283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3472283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3473283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3474283b8460SGeorge.Wilson } 3475283b8460SGeorge.Wilson } 3476283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3477283b8460SGeorge.Wilson } 3478283b8460SGeorge.Wilson } 3479