1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 2481cd5c55SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 251a902ef8SHans Rosenfeld * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 26*c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27fa9e4066Sahrens */ 28fa9e4066Sahrens 29fa9e4066Sahrens #include <sys/zfs_context.h> 30ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 31fa9e4066Sahrens #include <sys/spa.h> 32fa9e4066Sahrens #include <sys/spa_impl.h> 33fa9e4066Sahrens #include <sys/dmu.h> 34fa9e4066Sahrens #include <sys/dmu_tx.h> 35fa9e4066Sahrens #include <sys/vdev_impl.h> 36fa9e4066Sahrens #include <sys/uberblock_impl.h> 37fa9e4066Sahrens #include <sys/metaslab.h> 38fa9e4066Sahrens #include <sys/metaslab_impl.h> 39fa9e4066Sahrens #include <sys/space_map.h> 400713e232SGeorge Wilson #include <sys/space_reftree.h> 41fa9e4066Sahrens #include <sys/zio.h> 42fa9e4066Sahrens #include <sys/zap.h> 43fa9e4066Sahrens #include <sys/fs/zfs.h> 44c5904d13Seschrock #include <sys/arc.h> 45e6ca193dSGeorge Wilson #include <sys/zil.h> 463f9d6ad7SLin Ling #include <sys/dsl_scan.h> 47fa9e4066Sahrens 48fa9e4066Sahrens /* 49fa9e4066Sahrens * Virtual device management. 50fa9e4066Sahrens */ 51fa9e4066Sahrens 52fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 53fa9e4066Sahrens &vdev_root_ops, 54fa9e4066Sahrens &vdev_raidz_ops, 55fa9e4066Sahrens &vdev_mirror_ops, 56fa9e4066Sahrens &vdev_replacing_ops, 5799653d4eSeschrock &vdev_spare_ops, 58fa9e4066Sahrens &vdev_disk_ops, 59fa9e4066Sahrens &vdev_file_ops, 60fa9e4066Sahrens &vdev_missing_ops, 6188ecc943SGeorge Wilson &vdev_hole_ops, 62fa9e4066Sahrens NULL 63fa9e4066Sahrens }; 64fa9e4066Sahrens 65088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 66088f3894Sahrens int zfs_scrub_limit = 10; 6705b2b3b8Smishra 68bf3e216cSMatthew Ahrens /* 69bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 70bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 71bf3e216cSMatthew Ahrens */ 72bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 73bf3e216cSMatthew Ahrens 74fa9e4066Sahrens /* 75fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 76fa9e4066Sahrens */ 77fa9e4066Sahrens static vdev_ops_t * 78fa9e4066Sahrens vdev_getops(const char *type) 79fa9e4066Sahrens { 80fa9e4066Sahrens vdev_ops_t *ops, **opspp; 81fa9e4066Sahrens 82fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 83fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 84fa9e4066Sahrens break; 85fa9e4066Sahrens 86fa9e4066Sahrens return (ops); 87fa9e4066Sahrens } 88fa9e4066Sahrens 89fa9e4066Sahrens /* 90fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 91fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 92fa9e4066Sahrens */ 93fa9e4066Sahrens uint64_t 94fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 95fa9e4066Sahrens { 96ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 97fa9e4066Sahrens uint64_t csize; 98fa9e4066Sahrens 99573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 100fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 101fa9e4066Sahrens asize = MAX(asize, csize); 102fa9e4066Sahrens } 103fa9e4066Sahrens 104fa9e4066Sahrens return (asize); 105fa9e4066Sahrens } 106fa9e4066Sahrens 1072a79c5feSlling /* 108573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 109573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 110573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 111573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1122a79c5feSlling */ 1132a79c5feSlling uint64_t 114573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1152a79c5feSlling { 116573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1172a79c5feSlling 118573ca77eSGeorge Wilson /* 1194263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 120573ca77eSGeorge Wilson * just return our own asize. 121573ca77eSGeorge Wilson */ 122573ca77eSGeorge Wilson if (pvd == NULL) 123573ca77eSGeorge Wilson return (vd->vdev_asize); 1242a79c5feSlling 1252a79c5feSlling /* 126573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 127573ca77eSGeorge Wilson * to the nearest metaslab. 1282a79c5feSlling */ 129573ca77eSGeorge Wilson if (vd == vd->vdev_top) 130573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1312a79c5feSlling 132573ca77eSGeorge Wilson /* 133573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 134573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 135573ca77eSGeorge Wilson */ 136573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 137573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1382a79c5feSlling 139573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 140573ca77eSGeorge Wilson } 1412a79c5feSlling 142573ca77eSGeorge Wilson void 143573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 144573ca77eSGeorge Wilson { 145573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 146573ca77eSGeorge Wilson 147573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 148573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1492a79c5feSlling } 1502a79c5feSlling 151fa9e4066Sahrens vdev_t * 152fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 153fa9e4066Sahrens { 154fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 155fa9e4066Sahrens 156e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 157e05725b1Sbonwick 158088f3894Sahrens if (vdev < rvd->vdev_children) { 159088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 160fa9e4066Sahrens return (rvd->vdev_child[vdev]); 161088f3894Sahrens } 162fa9e4066Sahrens 163fa9e4066Sahrens return (NULL); 164fa9e4066Sahrens } 165fa9e4066Sahrens 166fa9e4066Sahrens vdev_t * 167fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 168fa9e4066Sahrens { 169fa9e4066Sahrens vdev_t *mvd; 170fa9e4066Sahrens 1710e34b6a7Sbonwick if (vd->vdev_guid == guid) 172fa9e4066Sahrens return (vd); 173fa9e4066Sahrens 174573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 175fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 176fa9e4066Sahrens NULL) 177fa9e4066Sahrens return (mvd); 178fa9e4066Sahrens 179fa9e4066Sahrens return (NULL); 180fa9e4066Sahrens } 181fa9e4066Sahrens 18212380e1eSArne Jansen static int 18312380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 18412380e1eSArne Jansen { 18512380e1eSArne Jansen int n = 0; 18612380e1eSArne Jansen 18712380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 18812380e1eSArne Jansen return (1); 18912380e1eSArne Jansen 19012380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 19112380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 19212380e1eSArne Jansen 19312380e1eSArne Jansen return (n); 19412380e1eSArne Jansen } 19512380e1eSArne Jansen 19612380e1eSArne Jansen int 19712380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 19812380e1eSArne Jansen { 19912380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 20012380e1eSArne Jansen } 20112380e1eSArne Jansen 202fa9e4066Sahrens void 203fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 204fa9e4066Sahrens { 205fa9e4066Sahrens size_t oldsize, newsize; 206fa9e4066Sahrens uint64_t id = cvd->vdev_id; 207fa9e4066Sahrens vdev_t **newchild; 20881cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 209fa9e4066Sahrens 21081cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 211fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 212fa9e4066Sahrens 213fa9e4066Sahrens cvd->vdev_parent = pvd; 214fa9e4066Sahrens 215fa9e4066Sahrens if (pvd == NULL) 216fa9e4066Sahrens return; 217fa9e4066Sahrens 218fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 219fa9e4066Sahrens 220fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 221fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 222fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 223fa9e4066Sahrens 224fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 225fa9e4066Sahrens if (pvd->vdev_child != NULL) { 226fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 227fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 228fa9e4066Sahrens } 229fa9e4066Sahrens 230fa9e4066Sahrens pvd->vdev_child = newchild; 231fa9e4066Sahrens pvd->vdev_child[id] = cvd; 232fa9e4066Sahrens 233fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 234fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 235fa9e4066Sahrens 236fa9e4066Sahrens /* 237fa9e4066Sahrens * Walk up all ancestors to update guid sum. 238fa9e4066Sahrens */ 239fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 240fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 241fa9e4066Sahrens } 242fa9e4066Sahrens 243fa9e4066Sahrens void 244fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 245fa9e4066Sahrens { 246fa9e4066Sahrens int c; 247fa9e4066Sahrens uint_t id = cvd->vdev_id; 248fa9e4066Sahrens 249fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 250fa9e4066Sahrens 251fa9e4066Sahrens if (pvd == NULL) 252fa9e4066Sahrens return; 253fa9e4066Sahrens 254fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 255fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 256fa9e4066Sahrens 257fa9e4066Sahrens pvd->vdev_child[id] = NULL; 258fa9e4066Sahrens cvd->vdev_parent = NULL; 259fa9e4066Sahrens 260fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 261fa9e4066Sahrens if (pvd->vdev_child[c]) 262fa9e4066Sahrens break; 263fa9e4066Sahrens 264fa9e4066Sahrens if (c == pvd->vdev_children) { 265fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 266fa9e4066Sahrens pvd->vdev_child = NULL; 267fa9e4066Sahrens pvd->vdev_children = 0; 268fa9e4066Sahrens } 269fa9e4066Sahrens 270fa9e4066Sahrens /* 271fa9e4066Sahrens * Walk up all ancestors to update guid sum. 272fa9e4066Sahrens */ 273fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 274fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 275fa9e4066Sahrens } 276fa9e4066Sahrens 277fa9e4066Sahrens /* 278fa9e4066Sahrens * Remove any holes in the child array. 279fa9e4066Sahrens */ 280fa9e4066Sahrens void 281fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 282fa9e4066Sahrens { 283fa9e4066Sahrens vdev_t **newchild, *cvd; 284fa9e4066Sahrens int oldc = pvd->vdev_children; 285573ca77eSGeorge Wilson int newc; 286fa9e4066Sahrens 287e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 288fa9e4066Sahrens 289573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 290fa9e4066Sahrens if (pvd->vdev_child[c]) 291fa9e4066Sahrens newc++; 292fa9e4066Sahrens 293fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 294fa9e4066Sahrens 295573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 296fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 297fa9e4066Sahrens newchild[newc] = cvd; 298fa9e4066Sahrens cvd->vdev_id = newc++; 299fa9e4066Sahrens } 300fa9e4066Sahrens } 301fa9e4066Sahrens 302fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 303fa9e4066Sahrens pvd->vdev_child = newchild; 304fa9e4066Sahrens pvd->vdev_children = newc; 305fa9e4066Sahrens } 306fa9e4066Sahrens 307fa9e4066Sahrens /* 308fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 309fa9e4066Sahrens */ 31088ecc943SGeorge Wilson vdev_t * 311fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 312fa9e4066Sahrens { 313fa9e4066Sahrens vdev_t *vd; 314fa9e4066Sahrens 315fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 316fa9e4066Sahrens 3170e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 3180e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 3190e34b6a7Sbonwick spa->spa_root_vdev = vd; 320e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 3210e34b6a7Sbonwick } 3220e34b6a7Sbonwick 32388ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 3240e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 3250e34b6a7Sbonwick /* 3260e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 3270e34b6a7Sbonwick * which must be unique among all pools. 3280e34b6a7Sbonwick */ 3291195e687SMark J Musante guid = spa_generate_guid(NULL); 3300e34b6a7Sbonwick } else { 3310e34b6a7Sbonwick /* 3320e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3330e34b6a7Sbonwick */ 3341195e687SMark J Musante guid = spa_generate_guid(spa); 3350e34b6a7Sbonwick } 3360e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3370e34b6a7Sbonwick } 3380e34b6a7Sbonwick 339fa9e4066Sahrens vd->vdev_spa = spa; 340fa9e4066Sahrens vd->vdev_id = id; 341fa9e4066Sahrens vd->vdev_guid = guid; 342fa9e4066Sahrens vd->vdev_guid_sum = guid; 343fa9e4066Sahrens vd->vdev_ops = ops; 344fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 34588ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 346fa9e4066Sahrens 347fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3485ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 349e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3508ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3510713e232SGeorge Wilson vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 3528ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3538ad4d6ddSJeff Bonwick } 354fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 355fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 356fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 357fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 358fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3593d7072f8Seschrock vdev_queue_init(vd); 3603d7072f8Seschrock vdev_cache_init(vd); 361fa9e4066Sahrens 362fa9e4066Sahrens return (vd); 363fa9e4066Sahrens } 364fa9e4066Sahrens 365fa9e4066Sahrens /* 366fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 367fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 368fa9e4066Sahrens * different for each case. 369fa9e4066Sahrens */ 37099653d4eSeschrock int 37199653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 37299653d4eSeschrock int alloctype) 373fa9e4066Sahrens { 374fa9e4066Sahrens vdev_ops_t *ops; 375fa9e4066Sahrens char *type; 3768654d025Sperrin uint64_t guid = 0, islog, nparity; 377fa9e4066Sahrens vdev_t *vd; 378fa9e4066Sahrens 379e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 380fa9e4066Sahrens 381fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 382be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 383fa9e4066Sahrens 384fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 385be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 386fa9e4066Sahrens 387fa9e4066Sahrens /* 388fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 389fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 390fa9e4066Sahrens */ 391fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 392fa9e4066Sahrens uint64_t label_id; 393fa9e4066Sahrens 394fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 395fa9e4066Sahrens label_id != id) 396be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 397fa9e4066Sahrens 398fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 399be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40099653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 40199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 402be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 403fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 404fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 405be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 40621ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 40721ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 408be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 409fa9e4066Sahrens } 410fa9e4066Sahrens 41199653d4eSeschrock /* 41299653d4eSeschrock * The first allocated vdev must be of type 'root'. 41399653d4eSeschrock */ 41499653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 415be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 41699653d4eSeschrock 4178654d025Sperrin /* 4188654d025Sperrin * Determine whether we're a log vdev. 4198654d025Sperrin */ 4208654d025Sperrin islog = 0; 4218654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 422990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 423be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 424fa9e4066Sahrens 42588ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 426be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 42788ecc943SGeorge Wilson 42899653d4eSeschrock /* 4298654d025Sperrin * Set the nparity property for RAID-Z vdevs. 43099653d4eSeschrock */ 4318654d025Sperrin nparity = -1ULL; 43299653d4eSeschrock if (ops == &vdev_raidz_ops) { 43399653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4348654d025Sperrin &nparity) == 0) { 435b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 436be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 43799653d4eSeschrock /* 438f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 439f94275ceSAdam Leventhal * device. 44099653d4eSeschrock */ 441f94275ceSAdam Leventhal if (nparity > 1 && 442f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 443be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 444f94275ceSAdam Leventhal if (nparity > 2 && 445f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 446be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 44799653d4eSeschrock } else { 44899653d4eSeschrock /* 44999653d4eSeschrock * We require the parity to be specified for SPAs that 45099653d4eSeschrock * support multiple parity levels. 45199653d4eSeschrock */ 452f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 453be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 45499653d4eSeschrock /* 45599653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 45699653d4eSeschrock */ 4578654d025Sperrin nparity = 1; 45899653d4eSeschrock } 45999653d4eSeschrock } else { 4608654d025Sperrin nparity = 0; 46199653d4eSeschrock } 4628654d025Sperrin ASSERT(nparity != -1ULL); 4638654d025Sperrin 4648654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4658654d025Sperrin 4668654d025Sperrin vd->vdev_islog = islog; 4678654d025Sperrin vd->vdev_nparity = nparity; 4688654d025Sperrin 4698654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4708654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4718654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4728654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4738654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4748654d025Sperrin &vd->vdev_physpath) == 0) 4758654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4766809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4776809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 47899653d4eSeschrock 479afefbcddSeschrock /* 480afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 481afefbcddSeschrock * as -1. 482afefbcddSeschrock */ 483afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 484afefbcddSeschrock &vd->vdev_wholedisk) != 0) 485afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 486afefbcddSeschrock 487ea8dc4b6Seschrock /* 488ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 489ea8dc4b6Seschrock * was not present at the time of import. 490ea8dc4b6Seschrock */ 4916809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4926809eb4eSEric Schrock &vd->vdev_not_present); 493ea8dc4b6Seschrock 494ecc2d604Sbonwick /* 495ecc2d604Sbonwick * Get the alignment requirement. 496ecc2d604Sbonwick */ 497ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 498ecc2d604Sbonwick 49988ecc943SGeorge Wilson /* 50088ecc943SGeorge Wilson * Retrieve the vdev creation time. 50188ecc943SGeorge Wilson */ 50288ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 50388ecc943SGeorge Wilson &vd->vdev_crtxg); 50488ecc943SGeorge Wilson 505fa9e4066Sahrens /* 506fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 507fa9e4066Sahrens */ 5081195e687SMark J Musante if (parent && !parent->vdev_parent && 5091195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 510fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 511fa9e4066Sahrens &vd->vdev_ms_array); 512fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 513fa9e4066Sahrens &vd->vdev_ms_shift); 514fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 515fa9e4066Sahrens &vd->vdev_asize); 5163f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 5173f9d6ad7SLin Ling &vd->vdev_removing); 518fa9e4066Sahrens } 519fa9e4066Sahrens 520cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 521a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 5229f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 5231195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 5249f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 525a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 526a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 527a1521560SJeff Bonwick } 528a1521560SJeff Bonwick 529fa9e4066Sahrens /* 5303d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 531fa9e4066Sahrens */ 532c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 53321ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 53421ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 535c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 536c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5370713e232SGeorge Wilson &vd->vdev_dtl_object); 538c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 539c5904d13Seschrock &vd->vdev_unspare); 540c5904d13Seschrock } 54121ecdf64SLin Ling 54221ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 54321ecdf64SLin Ling uint64_t spare = 0; 54421ecdf64SLin Ling 54521ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 54621ecdf64SLin Ling &spare) == 0 && spare) 54721ecdf64SLin Ling spa_spare_add(vd); 54821ecdf64SLin Ling } 54921ecdf64SLin Ling 550ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 551ecc2d604Sbonwick &vd->vdev_offline); 552c5904d13Seschrock 553b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 554b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 555cb04b873SMark J Musante 5563d7072f8Seschrock /* 5573d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5583d7072f8Seschrock * state, as the diagnosis made on another system may not be 559069f55e2SEric Schrock * valid in the current context. Local vdevs will 560069f55e2SEric Schrock * remain in the faulted state. 5613d7072f8Seschrock */ 562b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5633d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5643d7072f8Seschrock &vd->vdev_faulted); 5653d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5663d7072f8Seschrock &vd->vdev_degraded); 5673d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5683d7072f8Seschrock &vd->vdev_removed); 569069f55e2SEric Schrock 570069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 571069f55e2SEric Schrock char *aux; 572069f55e2SEric Schrock 573069f55e2SEric Schrock vd->vdev_label_aux = 574069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 575069f55e2SEric Schrock if (nvlist_lookup_string(nv, 576069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 577069f55e2SEric Schrock strcmp(aux, "external") == 0) 578069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 579069f55e2SEric Schrock } 5803d7072f8Seschrock } 581fa9e4066Sahrens } 582fa9e4066Sahrens 583fa9e4066Sahrens /* 584fa9e4066Sahrens * Add ourselves to the parent's list of children. 585fa9e4066Sahrens */ 586fa9e4066Sahrens vdev_add_child(parent, vd); 587fa9e4066Sahrens 58899653d4eSeschrock *vdp = vd; 58999653d4eSeschrock 59099653d4eSeschrock return (0); 591fa9e4066Sahrens } 592fa9e4066Sahrens 593fa9e4066Sahrens void 594fa9e4066Sahrens vdev_free(vdev_t *vd) 595fa9e4066Sahrens { 5963d7072f8Seschrock spa_t *spa = vd->vdev_spa; 597fa9e4066Sahrens 598fa9e4066Sahrens /* 599fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 600fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 601fa9e4066Sahrens */ 602fa9e4066Sahrens vdev_close(vd); 603fa9e4066Sahrens 604e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 605b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 606fa9e4066Sahrens 607fa9e4066Sahrens /* 608fa9e4066Sahrens * Free all children. 609fa9e4066Sahrens */ 610573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 611fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 612fa9e4066Sahrens 613fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 614fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 615fa9e4066Sahrens 616fa9e4066Sahrens /* 617fa9e4066Sahrens * Discard allocation state. 618fa9e4066Sahrens */ 619a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 620fa9e4066Sahrens vdev_metaslab_fini(vd); 621a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 622a1521560SJeff Bonwick } 623fa9e4066Sahrens 624fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 625fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 626fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 627fa9e4066Sahrens 628fa9e4066Sahrens /* 629fa9e4066Sahrens * Remove this vdev from its parent's child list. 630fa9e4066Sahrens */ 631fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 632fa9e4066Sahrens 633fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 634fa9e4066Sahrens 6353d7072f8Seschrock /* 6363d7072f8Seschrock * Clean up vdev structure. 6373d7072f8Seschrock */ 6383d7072f8Seschrock vdev_queue_fini(vd); 6393d7072f8Seschrock vdev_cache_fini(vd); 6403d7072f8Seschrock 6413d7072f8Seschrock if (vd->vdev_path) 6423d7072f8Seschrock spa_strfree(vd->vdev_path); 6433d7072f8Seschrock if (vd->vdev_devid) 6443d7072f8Seschrock spa_strfree(vd->vdev_devid); 6453d7072f8Seschrock if (vd->vdev_physpath) 6463d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6476809eb4eSEric Schrock if (vd->vdev_fru) 6486809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6493d7072f8Seschrock 6503d7072f8Seschrock if (vd->vdev_isspare) 6513d7072f8Seschrock spa_spare_remove(vd); 652fa94a07fSbrendan if (vd->vdev_isl2cache) 653fa94a07fSbrendan spa_l2cache_remove(vd); 6543d7072f8Seschrock 6553d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6563d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6578ad4d6ddSJeff Bonwick 6583d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6590713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 6608ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6610713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 6620713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 6638ad4d6ddSJeff Bonwick } 6643d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6658ad4d6ddSJeff Bonwick 6663d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6673d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 668e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6693d7072f8Seschrock 6703d7072f8Seschrock if (vd == spa->spa_root_vdev) 6713d7072f8Seschrock spa->spa_root_vdev = NULL; 6723d7072f8Seschrock 6733d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 674fa9e4066Sahrens } 675fa9e4066Sahrens 676fa9e4066Sahrens /* 677fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 678fa9e4066Sahrens */ 679fa9e4066Sahrens static void 680fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 681fa9e4066Sahrens { 682fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 683fa9e4066Sahrens metaslab_t *msp; 684fa9e4066Sahrens vdev_t *vd; 685fa9e4066Sahrens int t; 686fa9e4066Sahrens 687fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 688fa9e4066Sahrens 689fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 690fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 691fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 692fa9e4066Sahrens 693fa9e4066Sahrens svd->vdev_ms_array = 0; 694fa9e4066Sahrens svd->vdev_ms_shift = 0; 695fa9e4066Sahrens svd->vdev_ms_count = 0; 696fa9e4066Sahrens 697cd0837ccSGeorge Wilson if (tvd->vdev_mg) 698cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 699fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 700fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 701fa9e4066Sahrens 702fa9e4066Sahrens svd->vdev_mg = NULL; 703fa9e4066Sahrens svd->vdev_ms = NULL; 704ecc2d604Sbonwick 705ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 706ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 707fa9e4066Sahrens 708fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 709fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 71099653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 711fa9e4066Sahrens 712fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 713fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 71499653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 715fa9e4066Sahrens 716fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 717fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 718fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 719fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 720fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 721fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 722fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 723fa9e4066Sahrens } 724fa9e4066Sahrens 725e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 726fa9e4066Sahrens vdev_config_clean(svd); 727fa9e4066Sahrens vdev_config_dirty(tvd); 728fa9e4066Sahrens } 729fa9e4066Sahrens 730e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 731e14bb325SJeff Bonwick vdev_state_clean(svd); 732e14bb325SJeff Bonwick vdev_state_dirty(tvd); 733e14bb325SJeff Bonwick } 734e14bb325SJeff Bonwick 73599653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 73699653d4eSeschrock svd->vdev_deflate_ratio = 0; 7378654d025Sperrin 7388654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7398654d025Sperrin svd->vdev_islog = 0; 740fa9e4066Sahrens } 741fa9e4066Sahrens 742fa9e4066Sahrens static void 743fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 744fa9e4066Sahrens { 745fa9e4066Sahrens if (vd == NULL) 746fa9e4066Sahrens return; 747fa9e4066Sahrens 748fa9e4066Sahrens vd->vdev_top = tvd; 749fa9e4066Sahrens 750573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 751fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 752fa9e4066Sahrens } 753fa9e4066Sahrens 754fa9e4066Sahrens /* 755fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 756fa9e4066Sahrens */ 757fa9e4066Sahrens vdev_t * 758fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 759fa9e4066Sahrens { 760fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 761fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 762fa9e4066Sahrens vdev_t *mvd; 763fa9e4066Sahrens 764e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 765fa9e4066Sahrens 766fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 767ecc2d604Sbonwick 768ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 769573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 7704263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 771ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 772ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 77388ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 774ecc2d604Sbonwick 775fa9e4066Sahrens vdev_remove_child(pvd, cvd); 776fa9e4066Sahrens vdev_add_child(pvd, mvd); 777fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 778fa9e4066Sahrens vdev_add_child(mvd, cvd); 779fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 780fa9e4066Sahrens 781fa9e4066Sahrens if (mvd == mvd->vdev_top) 782fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 783fa9e4066Sahrens 784fa9e4066Sahrens return (mvd); 785fa9e4066Sahrens } 786fa9e4066Sahrens 787fa9e4066Sahrens /* 788fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 789fa9e4066Sahrens */ 790fa9e4066Sahrens void 791fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 792fa9e4066Sahrens { 793fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 794fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 795fa9e4066Sahrens 796e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 797fa9e4066Sahrens 798fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 799fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 80099653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 80199653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 802ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 803fa9e4066Sahrens 804fa9e4066Sahrens vdev_remove_child(mvd, cvd); 805fa9e4066Sahrens vdev_remove_child(pvd, mvd); 8068ad4d6ddSJeff Bonwick 80799653d4eSeschrock /* 808e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 809e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 810e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 811e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 81299653d4eSeschrock */ 8138ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 8148ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 8151195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 8168ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 8178ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 8188ad4d6ddSJeff Bonwick } 819e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 820e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 821fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 822fa9e4066Sahrens 823fa9e4066Sahrens if (cvd == cvd->vdev_top) 824fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 825fa9e4066Sahrens 826fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 827fa9e4066Sahrens vdev_free(mvd); 828fa9e4066Sahrens } 829fa9e4066Sahrens 830ea8dc4b6Seschrock int 831fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 832fa9e4066Sahrens { 833fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 834ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 835ecc2d604Sbonwick uint64_t m; 836fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 837fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 838ecc2d604Sbonwick metaslab_t **mspp; 839ecc2d604Sbonwick int error; 840fa9e4066Sahrens 841a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 842a1521560SJeff Bonwick 84388ecc943SGeorge Wilson /* 84488ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 84588ecc943SGeorge Wilson */ 84688ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8470e34b6a7Sbonwick return (0); 8480e34b6a7Sbonwick 84988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 85088ecc943SGeorge Wilson 851e6ca193dSGeorge Wilson /* 852e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 853b5152584SMatthew Ahrens * in 128k (1 << 17) because it is the "typical" blocksize. 854b5152584SMatthew Ahrens * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 855b5152584SMatthew Ahrens * otherwise it would inconsistently account for existing bp's. 856e6ca193dSGeorge Wilson */ 857e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 858e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 859e6ca193dSGeorge Wilson 860fa9e4066Sahrens ASSERT(oldc <= newc); 861fa9e4066Sahrens 862ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 863fa9e4066Sahrens 864ecc2d604Sbonwick if (oldc != 0) { 865ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 866ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 867ecc2d604Sbonwick } 868fa9e4066Sahrens 869ecc2d604Sbonwick vd->vdev_ms = mspp; 870ecc2d604Sbonwick vd->vdev_ms_count = newc; 871fa9e4066Sahrens 872ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 8730713e232SGeorge Wilson uint64_t object = 0; 8740713e232SGeorge Wilson 875ecc2d604Sbonwick if (txg == 0) { 876ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8777bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8787bfdf011SNeil Perrin DMU_READ_PREFETCH); 879ecc2d604Sbonwick if (error) 880ecc2d604Sbonwick return (error); 881fa9e4066Sahrens } 8821e9bd7ecSPrakash Surya 8831e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 8841e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 8851e9bd7ecSPrakash Surya if (error) 8861e9bd7ecSPrakash Surya return (error); 887fa9e4066Sahrens } 888fa9e4066Sahrens 889a1521560SJeff Bonwick if (txg == 0) 890a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 891a1521560SJeff Bonwick 8923f9d6ad7SLin Ling /* 8933f9d6ad7SLin Ling * If the vdev is being removed we don't activate 8943f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 8953f9d6ad7SLin Ling * allocations are performed on this device. 8963f9d6ad7SLin Ling */ 8973f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 898a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 899a1521560SJeff Bonwick 900a1521560SJeff Bonwick if (txg == 0) 901a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 902a1521560SJeff Bonwick 903ea8dc4b6Seschrock return (0); 904fa9e4066Sahrens } 905fa9e4066Sahrens 906fa9e4066Sahrens void 907fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 908fa9e4066Sahrens { 909fa9e4066Sahrens uint64_t m; 910fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 911fa9e4066Sahrens 912fa9e4066Sahrens if (vd->vdev_ms != NULL) { 913a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 9140713e232SGeorge Wilson for (m = 0; m < count; m++) { 9150713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 9160713e232SGeorge Wilson 9170713e232SGeorge Wilson if (msp != NULL) 9180713e232SGeorge Wilson metaslab_fini(msp); 9190713e232SGeorge Wilson } 920fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 921fa9e4066Sahrens vd->vdev_ms = NULL; 922fa9e4066Sahrens } 923fa9e4066Sahrens } 924fa9e4066Sahrens 925e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 926e14bb325SJeff Bonwick boolean_t vps_readable; 927e14bb325SJeff Bonwick boolean_t vps_writeable; 928e14bb325SJeff Bonwick int vps_flags; 929e14bb325SJeff Bonwick } vdev_probe_stats_t; 930e14bb325SJeff Bonwick 931e14bb325SJeff Bonwick static void 932e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9330a4e9518Sgw { 9348ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 935a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 936e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 937a3f829aeSBill Moore 938a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 939e14bb325SJeff Bonwick 940e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 941e14bb325SJeff Bonwick if (zio->io_error == 0) 942e14bb325SJeff Bonwick vps->vps_readable = 1; 9438ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 944a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 945e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 946e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 947e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 948e14bb325SJeff Bonwick } else { 949e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 950e14bb325SJeff Bonwick } 951e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 952e14bb325SJeff Bonwick if (zio->io_error == 0) 953e14bb325SJeff Bonwick vps->vps_writeable = 1; 954e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 955e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 956a3f829aeSBill Moore zio_t *pio; 957e14bb325SJeff Bonwick 958e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 959e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 960e14bb325SJeff Bonwick 961e14bb325SJeff Bonwick if (vdev_readable(vd) && 9628ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 963e14bb325SJeff Bonwick zio->io_error = 0; 964e14bb325SJeff Bonwick } else { 965e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 966e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9678ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 968be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 969e14bb325SJeff Bonwick } 970a3f829aeSBill Moore 971a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 972a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 973a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 974a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 975a3f829aeSBill Moore 976a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 977a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 978be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 979a3f829aeSBill Moore 980e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 981e14bb325SJeff Bonwick } 982e14bb325SJeff Bonwick } 9830a4e9518Sgw 984e14bb325SJeff Bonwick /* 985f7170741SWill Andrews * Determine whether this device is accessible. 986f7170741SWill Andrews * 987f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 988f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 989f7170741SWill Andrews * a VTOC. 990e14bb325SJeff Bonwick */ 991e14bb325SJeff Bonwick zio_t * 992a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 993e14bb325SJeff Bonwick { 994e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 995a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 996a3f829aeSBill Moore zio_t *pio; 997a3f829aeSBill Moore 998a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9990a4e9518Sgw 1000a3f829aeSBill Moore /* 1001a3f829aeSBill Moore * Don't probe the probe. 1002a3f829aeSBill Moore */ 1003a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1004a3f829aeSBill Moore return (NULL); 1005e14bb325SJeff Bonwick 1006a3f829aeSBill Moore /* 1007a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1008a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1009a3f829aeSBill Moore * this vdev will become parents of the probe io. 1010a3f829aeSBill Moore */ 1011a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1012e14bb325SJeff Bonwick 1013a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1014a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1015a3f829aeSBill Moore 1016a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1017a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 10188956713aSEric Schrock ZIO_FLAG_TRYHARD; 1019a3f829aeSBill Moore 1020a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1021a3f829aeSBill Moore /* 1022a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1023a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1024a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1025a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1026a3f829aeSBill Moore * any zio looking at these values can assume that 1027a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1028a3f829aeSBill Moore * important because when a device has intermittent 1029a3f829aeSBill Moore * connectivity problems, we want to ensure that 1030a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1031a3f829aeSBill Moore * the zio (EIO). 1032a3f829aeSBill Moore * 1033a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1034a3f829aeSBill Moore * values so the probe can reevaluate from first 1035a3f829aeSBill Moore * principles. 1036a3f829aeSBill Moore */ 1037a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1038a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1039a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1040a3f829aeSBill Moore } 1041a3f829aeSBill Moore 1042a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1043a3f829aeSBill Moore vdev_probe_done, vps, 1044a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1045a3f829aeSBill Moore 104698d1cbfeSGeorge Wilson /* 104798d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 104898d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 104998d1cbfeSGeorge Wilson */ 1050a3f829aeSBill Moore if (zio != NULL) { 1051a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1052a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1053a3f829aeSBill Moore } 1054e14bb325SJeff Bonwick } 1055e14bb325SJeff Bonwick 1056a3f829aeSBill Moore if (zio != NULL) 1057a3f829aeSBill Moore zio_add_child(zio, pio); 1058e14bb325SJeff Bonwick 1059a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1060e14bb325SJeff Bonwick 1061a3f829aeSBill Moore if (vps == NULL) { 1062a3f829aeSBill Moore ASSERT(zio != NULL); 1063a3f829aeSBill Moore return (NULL); 1064a3f829aeSBill Moore } 1065e14bb325SJeff Bonwick 1066e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1067a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1068e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1069f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1070f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1071e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1072e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1073e14bb325SJeff Bonwick } 1074e14bb325SJeff Bonwick 1075a3f829aeSBill Moore if (zio == NULL) 1076a3f829aeSBill Moore return (pio); 1077a3f829aeSBill Moore 1078a3f829aeSBill Moore zio_nowait(pio); 1079a3f829aeSBill Moore return (NULL); 10800a4e9518Sgw } 10810a4e9518Sgw 1082f64c0e34SEric Taylor static void 1083f64c0e34SEric Taylor vdev_open_child(void *arg) 1084f64c0e34SEric Taylor { 1085f64c0e34SEric Taylor vdev_t *vd = arg; 1086f64c0e34SEric Taylor 1087f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1088f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1089f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1090f64c0e34SEric Taylor } 1091f64c0e34SEric Taylor 1092681d9761SEric Taylor boolean_t 1093681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1094681d9761SEric Taylor { 1095681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1096681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1097681d9761SEric Taylor return (B_TRUE); 1098681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1099681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1100681d9761SEric Taylor return (B_TRUE); 1101681d9761SEric Taylor return (B_FALSE); 1102681d9761SEric Taylor } 1103681d9761SEric Taylor 1104f64c0e34SEric Taylor void 1105f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1106f64c0e34SEric Taylor { 1107f64c0e34SEric Taylor taskq_t *tq; 1108f64c0e34SEric Taylor int children = vd->vdev_children; 1109f64c0e34SEric Taylor 1110681d9761SEric Taylor /* 1111681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1112681d9761SEric Taylor * in a single thread so that the same thread holds the 1113681d9761SEric Taylor * spa_namespace_lock 1114681d9761SEric Taylor */ 1115681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1116681d9761SEric Taylor for (int c = 0; c < children; c++) 1117681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1118681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1119681d9761SEric Taylor return; 1120681d9761SEric Taylor } 1121f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1122f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1123f64c0e34SEric Taylor 1124f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1125f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1126f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1127f64c0e34SEric Taylor 1128f64c0e34SEric Taylor taskq_destroy(tq); 1129f64c0e34SEric Taylor } 1130f64c0e34SEric Taylor 1131fa9e4066Sahrens /* 1132fa9e4066Sahrens * Prepare a virtual device for access. 1133fa9e4066Sahrens */ 1134fa9e4066Sahrens int 1135fa9e4066Sahrens vdev_open(vdev_t *vd) 1136fa9e4066Sahrens { 11378ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1138fa9e4066Sahrens int error; 1139fa9e4066Sahrens uint64_t osize = 0; 11404263d13fSGeorge Wilson uint64_t max_osize = 0; 11414263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1142ecc2d604Sbonwick uint64_t ashift = 0; 1143fa9e4066Sahrens 1144f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1145f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1146fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1147fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1148fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1149fa9e4066Sahrens 1150fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1151e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1152e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1153573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1154fa9e4066Sahrens 1155069f55e2SEric Schrock /* 1156069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1157069f55e2SEric Schrock * faulted, bail out of the open. 1158069f55e2SEric Schrock */ 11593d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11603d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1161069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1162069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11633d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1164069f55e2SEric Schrock vd->vdev_label_aux); 1165be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 11663d7072f8Seschrock } else if (vd->vdev_offline) { 1167fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1168ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1169be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1170fa9e4066Sahrens } 1171fa9e4066Sahrens 11724263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1173fa9e4066Sahrens 1174095bcd66SGeorge Wilson /* 1175095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1176095bcd66SGeorge Wilson * the vdev on error. 1177095bcd66SGeorge Wilson */ 1178095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1179ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11808956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1181ea8dc4b6Seschrock 1182fa9e4066Sahrens if (error) { 11833d7072f8Seschrock if (vd->vdev_removed && 11843d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11853d7072f8Seschrock vd->vdev_removed = B_FALSE; 11863d7072f8Seschrock 1187ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1188fa9e4066Sahrens vd->vdev_stat.vs_aux); 1189fa9e4066Sahrens return (error); 1190fa9e4066Sahrens } 1191fa9e4066Sahrens 11923d7072f8Seschrock vd->vdev_removed = B_FALSE; 11933d7072f8Seschrock 1194096d22d4SEric Schrock /* 1195096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1196096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1197096d22d4SEric Schrock */ 1198096d22d4SEric Schrock if (vd->vdev_faulted) { 1199096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1200096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1201096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1202096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1203096d22d4SEric Schrock vd->vdev_label_aux); 1204be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1205096d22d4SEric Schrock } 1206096d22d4SEric Schrock 12073d7072f8Seschrock if (vd->vdev_degraded) { 12083d7072f8Seschrock ASSERT(vd->vdev_children == 0); 12093d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 12103d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 12113d7072f8Seschrock } else { 1212069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 12133d7072f8Seschrock } 1214fa9e4066Sahrens 121588ecc943SGeorge Wilson /* 121688ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 121788ecc943SGeorge Wilson */ 121888ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 121988ecc943SGeorge Wilson return (0); 122088ecc943SGeorge Wilson 1221573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1222ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1223ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1224ea8dc4b6Seschrock VDEV_AUX_NONE); 1225ea8dc4b6Seschrock break; 1226ea8dc4b6Seschrock } 1227573ca77eSGeorge Wilson } 1228fa9e4066Sahrens 1229fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 12304263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1231fa9e4066Sahrens 1232fa9e4066Sahrens if (vd->vdev_children == 0) { 1233fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1234ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1235ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1236be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1237fa9e4066Sahrens } 1238fa9e4066Sahrens psize = osize; 1239fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 12404263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 12414263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1242fa9e4066Sahrens } else { 1243ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1244fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1245ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1246ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1247be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1248fa9e4066Sahrens } 1249fa9e4066Sahrens psize = 0; 1250fa9e4066Sahrens asize = osize; 12514263d13fSGeorge Wilson max_asize = max_osize; 1252fa9e4066Sahrens } 1253fa9e4066Sahrens 1254fa9e4066Sahrens vd->vdev_psize = psize; 1255fa9e4066Sahrens 1256573ca77eSGeorge Wilson /* 1257573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1258573ca77eSGeorge Wilson */ 1259573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1260573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1261573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1262be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1263573ca77eSGeorge Wilson } 1264573ca77eSGeorge Wilson 1265fa9e4066Sahrens if (vd->vdev_asize == 0) { 1266fa9e4066Sahrens /* 1267fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1268ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1269fa9e4066Sahrens */ 1270fa9e4066Sahrens vd->vdev_asize = asize; 12714263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1272ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1273fa9e4066Sahrens } else { 1274fa9e4066Sahrens /* 12752384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 12762384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 12772384d9f8SGeorge Wilson * issue a warning instead. 1278fa9e4066Sahrens */ 12792384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 12802384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 12812384d9f8SGeorge Wilson cmn_err(CE_WARN, 12822384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 12832384d9f8SGeorge Wilson "larger than the pool's alignment\n", 12842384d9f8SGeorge Wilson vd->vdev_path); 1285fa9e4066Sahrens } 12864263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1287573ca77eSGeorge Wilson } 1288fa9e4066Sahrens 1289573ca77eSGeorge Wilson /* 1290573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1291573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1292573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1293573ca77eSGeorge Wilson */ 1294573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1295573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1296573ca77eSGeorge Wilson vd->vdev_asize = asize; 1297fa9e4066Sahrens 1298573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1299fa9e4066Sahrens 13000a4e9518Sgw /* 13010a4e9518Sgw * Ensure we can issue some IO before declaring the 13020a4e9518Sgw * vdev open for business. 13030a4e9518Sgw */ 1304e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1305e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 130698d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 130798d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 13080a4e9518Sgw return (error); 13090a4e9518Sgw } 13100a4e9518Sgw 131181cd5c55SMatthew Ahrens /* 131281cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 131381cd5c55SMatthew Ahrens */ 131481cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 131581cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 131681cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 131781cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 131881cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 131981cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 132081cd5c55SMatthew Ahrens } 132181cd5c55SMatthew Ahrens 1322088f3894Sahrens /* 1323088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 13248ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 13258ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1326088f3894Sahrens */ 13278ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 13288ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 13298ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1330088f3894Sahrens 1331fa9e4066Sahrens return (0); 1332fa9e4066Sahrens } 1333fa9e4066Sahrens 1334560e6e96Seschrock /* 1335560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1336560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 13373d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1338560e6e96Seschrock * 1339d7f601efSGeorge Wilson * If 'strict' is false ignore the spa guid check. This is necessary because 1340d7f601efSGeorge Wilson * if the machine crashed during a re-guid the new guid might have been written 1341d7f601efSGeorge Wilson * to all of the vdev labels, but not the cached config. The strict check 1342d7f601efSGeorge Wilson * will be performed when the pool is opened again using the mos config. 1343d7f601efSGeorge Wilson * 1344560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1345560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1346560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1347560e6e96Seschrock * will be updated but the function will return 0. 1348560e6e96Seschrock */ 1349560e6e96Seschrock int 1350d7f601efSGeorge Wilson vdev_validate(vdev_t *vd, boolean_t strict) 1351560e6e96Seschrock { 1352560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1353560e6e96Seschrock nvlist_t *label; 13541195e687SMark J Musante uint64_t guid = 0, top_guid; 1355560e6e96Seschrock uint64_t state; 1356560e6e96Seschrock 1357573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1358d7f601efSGeorge Wilson if (vdev_validate(vd->vdev_child[c], strict) != 0) 1359be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1360560e6e96Seschrock 1361b5989ec7Seschrock /* 1362b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1363b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1364b5989ec7Seschrock * overwrite the previous state. 1365b5989ec7Seschrock */ 1366e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13671195e687SMark J Musante uint64_t aux_guid = 0; 13681195e687SMark J Musante nvlist_t *nvl; 1369bda88194SGeorge Wilson uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1370bda88194SGeorge Wilson spa_last_synced_txg(spa) : -1ULL; 1371560e6e96Seschrock 1372dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1373560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1374560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1375560e6e96Seschrock return (0); 1376560e6e96Seschrock } 1377560e6e96Seschrock 13781195e687SMark J Musante /* 13791195e687SMark J Musante * Determine if this vdev has been split off into another 13801195e687SMark J Musante * pool. If so, then refuse to open it. 13811195e687SMark J Musante */ 13821195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 13831195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 13841195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 13851195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 13861195e687SMark J Musante nvlist_free(label); 13871195e687SMark J Musante return (0); 13881195e687SMark J Musante } 13891195e687SMark J Musante 1390d7f601efSGeorge Wilson if (strict && (nvlist_lookup_uint64(label, 1391d7f601efSGeorge Wilson ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1392d7f601efSGeorge Wilson guid != spa_guid(spa))) { 1393560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1394560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1395560e6e96Seschrock nvlist_free(label); 1396560e6e96Seschrock return (0); 1397560e6e96Seschrock } 1398560e6e96Seschrock 13991195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 14001195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 14011195e687SMark J Musante &aux_guid) != 0) 14021195e687SMark J Musante aux_guid = 0; 14031195e687SMark J Musante 1404e14bb325SJeff Bonwick /* 1405e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1406e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1407e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1408e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1409e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1410e14bb325SJeff Bonwick * safely compare to that instead. 14111195e687SMark J Musante * 14121195e687SMark J Musante * If we split this vdev off instead, then we also check the 14131195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 14141195e687SMark J Musante * corrupt if it is partway through a split operation. 1415e14bb325SJeff Bonwick */ 1416560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1417e14bb325SJeff Bonwick &guid) != 0 || 1418e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1419e14bb325SJeff Bonwick &top_guid) != 0 || 14201195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1421e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1422560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1423560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1424560e6e96Seschrock nvlist_free(label); 1425560e6e96Seschrock return (0); 1426560e6e96Seschrock } 1427560e6e96Seschrock 1428560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1429560e6e96Seschrock &state) != 0) { 1430560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1431560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1432560e6e96Seschrock nvlist_free(label); 1433560e6e96Seschrock return (0); 1434560e6e96Seschrock } 1435560e6e96Seschrock 1436560e6e96Seschrock nvlist_free(label); 1437560e6e96Seschrock 1438bc758434SLin Ling /* 14394b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1440bc758434SLin Ling * state of the pool. 1441bc758434SLin Ling */ 14424b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1443b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1444bc758434SLin Ling state != POOL_STATE_ACTIVE) 1445be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1446560e6e96Seschrock 144751ece835Seschrock /* 144851ece835Seschrock * If we were able to open and validate a vdev that was 144951ece835Seschrock * previously marked permanently unavailable, clear that state 145051ece835Seschrock * now. 145151ece835Seschrock */ 145251ece835Seschrock if (vd->vdev_not_present) 145351ece835Seschrock vd->vdev_not_present = 0; 145451ece835Seschrock } 1455560e6e96Seschrock 1456560e6e96Seschrock return (0); 1457560e6e96Seschrock } 1458560e6e96Seschrock 1459fa9e4066Sahrens /* 1460fa9e4066Sahrens * Close a virtual device. 1461fa9e4066Sahrens */ 1462fa9e4066Sahrens void 1463fa9e4066Sahrens vdev_close(vdev_t *vd) 1464fa9e4066Sahrens { 14658ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1466095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14678ad4d6ddSJeff Bonwick 14688ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14698ad4d6ddSJeff Bonwick 14701195e687SMark J Musante /* 14711195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14721195e687SMark J Musante * going offline. 14731195e687SMark J Musante */ 1474095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14751195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1476095bcd66SGeorge Wilson 1477fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1478fa9e4066Sahrens 14793d7072f8Seschrock vdev_cache_purge(vd); 1480fa9e4066Sahrens 1481560e6e96Seschrock /* 1482573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1483560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1484560e6e96Seschrock * it's still faulted. 1485560e6e96Seschrock */ 1486560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1487560e6e96Seschrock 1488fa9e4066Sahrens if (vd->vdev_offline) 1489fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1490fa9e4066Sahrens else 1491fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1492ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1493fa9e4066Sahrens } 1494fa9e4066Sahrens 1495dcba9f3fSGeorge Wilson void 1496dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1497dcba9f3fSGeorge Wilson { 1498dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1499dcba9f3fSGeorge Wilson 1500dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1501dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1502dcba9f3fSGeorge Wilson return; 1503dcba9f3fSGeorge Wilson 1504dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1505dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1506dcba9f3fSGeorge Wilson 1507dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1508dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1509dcba9f3fSGeorge Wilson } 1510dcba9f3fSGeorge Wilson 1511dcba9f3fSGeorge Wilson void 1512dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1513dcba9f3fSGeorge Wilson { 1514dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1515dcba9f3fSGeorge Wilson 1516dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1517dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1518dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1519dcba9f3fSGeorge Wilson 1520dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1521dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1522dcba9f3fSGeorge Wilson } 1523dcba9f3fSGeorge Wilson 1524095bcd66SGeorge Wilson /* 1525095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1526095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1527095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1528095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1529095bcd66SGeorge Wilson */ 1530fa9e4066Sahrens void 1531ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1532fa9e4066Sahrens { 1533ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1534fa9e4066Sahrens 1535e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1536ea8dc4b6Seschrock 15371195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 15381195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1539fa9e4066Sahrens vdev_close(vd); 1540fa9e4066Sahrens (void) vdev_open(vd); 1541fa9e4066Sahrens 154239c23413Seschrock /* 154339c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 154439c23413Seschrock * Otherwise, a device with an invalid label could be successfully 154539c23413Seschrock * opened in response to vdev_reopen(). 154639c23413Seschrock */ 1547c5904d13Seschrock if (vd->vdev_aux) { 1548c5904d13Seschrock (void) vdev_validate_aux(vd); 1549e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 15506809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1551573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1552573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1553c5904d13Seschrock } else { 1554bda88194SGeorge Wilson (void) vdev_validate(vd, B_TRUE); 1555c5904d13Seschrock } 155639c23413Seschrock 1557fa9e4066Sahrens /* 15583d7072f8Seschrock * Reassess parent vdev's health. 1559fa9e4066Sahrens */ 15603d7072f8Seschrock vdev_propagate_state(vd); 1561fa9e4066Sahrens } 1562fa9e4066Sahrens 1563fa9e4066Sahrens int 156499653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1565fa9e4066Sahrens { 1566fa9e4066Sahrens int error; 1567fa9e4066Sahrens 1568fa9e4066Sahrens /* 1569fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1570fa9e4066Sahrens * For a create, however, we want to fail the request if 1571fa9e4066Sahrens * there are any components we can't open. 1572fa9e4066Sahrens */ 1573fa9e4066Sahrens error = vdev_open(vd); 1574fa9e4066Sahrens 1575fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1576fa9e4066Sahrens vdev_close(vd); 1577fa9e4066Sahrens return (error ? error : ENXIO); 1578fa9e4066Sahrens } 1579fa9e4066Sahrens 1580fa9e4066Sahrens /* 15810713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1582fa9e4066Sahrens */ 15830713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 15840713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 158539c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1586fa9e4066Sahrens vdev_close(vd); 1587fa9e4066Sahrens return (error); 1588fa9e4066Sahrens } 1589fa9e4066Sahrens 1590fa9e4066Sahrens return (0); 1591fa9e4066Sahrens } 1592fa9e4066Sahrens 15930e34b6a7Sbonwick void 1594573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1595fa9e4066Sahrens { 1596fa9e4066Sahrens /* 1597bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1598fa9e4066Sahrens */ 1599bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1600fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1601fa9e4066Sahrens } 1602fa9e4066Sahrens 1603fa9e4066Sahrens void 1604ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1605fa9e4066Sahrens { 1606ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 160788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1608ecc2d604Sbonwick ASSERT(ISP2(flags)); 1609f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1610fa9e4066Sahrens 1611ecc2d604Sbonwick if (flags & VDD_METASLAB) 1612ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1613ecc2d604Sbonwick 1614ecc2d604Sbonwick if (flags & VDD_DTL) 1615ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1616ecc2d604Sbonwick 1617ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1618fa9e4066Sahrens } 1619fa9e4066Sahrens 16200713e232SGeorge Wilson void 16210713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 16220713e232SGeorge Wilson { 16230713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 16240713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 16250713e232SGeorge Wilson 16260713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 16270713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 16280713e232SGeorge Wilson } 16290713e232SGeorge Wilson 16308ad4d6ddSJeff Bonwick /* 16318ad4d6ddSJeff Bonwick * DTLs. 16328ad4d6ddSJeff Bonwick * 16338ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 16349fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 16358ad4d6ddSJeff Bonwick * 16368ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 16378ad4d6ddSJeff Bonwick * 16388ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 16398ad4d6ddSJeff Bonwick * 16408ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 16418ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 16428ad4d6ddSJeff Bonwick * txgs that was scrubbed. 16438ad4d6ddSJeff Bonwick * 16448ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 16458ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 16468ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 16478ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 16488ad4d6ddSJeff Bonwick * determine whether a device can be detached. 16498ad4d6ddSJeff Bonwick * 16508ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 16518ad4d6ddSJeff Bonwick * either has the data or it doesn't. 16528ad4d6ddSJeff Bonwick * 16538ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 16548ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 16558ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 16568ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 16578ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 16588ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 16598ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 16608ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 16618ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 16628ad4d6ddSJeff Bonwick * 16638ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 16648ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 16658ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 16668ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 16678ad4d6ddSJeff Bonwick */ 1668fa9e4066Sahrens void 16698ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1670fa9e4066Sahrens { 16710713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16728ad4d6ddSJeff Bonwick 16738ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16748ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1675f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 16768ad4d6ddSJeff Bonwick 16770713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16780713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 16790713e232SGeorge Wilson range_tree_add(rt, txg, size); 16800713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1681fa9e4066Sahrens } 1682fa9e4066Sahrens 16838ad4d6ddSJeff Bonwick boolean_t 16848ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1685fa9e4066Sahrens { 16860713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 16878ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1688fa9e4066Sahrens 16898ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16908ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1691fa9e4066Sahrens 16920713e232SGeorge Wilson mutex_enter(rt->rt_lock); 16930713e232SGeorge Wilson if (range_tree_space(rt) != 0) 16940713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 16950713e232SGeorge Wilson mutex_exit(rt->rt_lock); 1696fa9e4066Sahrens 1697fa9e4066Sahrens return (dirty); 1698fa9e4066Sahrens } 1699fa9e4066Sahrens 17008ad4d6ddSJeff Bonwick boolean_t 17018ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 17028ad4d6ddSJeff Bonwick { 17030713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 17048ad4d6ddSJeff Bonwick boolean_t empty; 17058ad4d6ddSJeff Bonwick 17060713e232SGeorge Wilson mutex_enter(rt->rt_lock); 17070713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 17080713e232SGeorge Wilson mutex_exit(rt->rt_lock); 17098ad4d6ddSJeff Bonwick 17108ad4d6ddSJeff Bonwick return (empty); 17118ad4d6ddSJeff Bonwick } 17128ad4d6ddSJeff Bonwick 1713b4952e17SGeorge Wilson /* 1714b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 1715b4952e17SGeorge Wilson */ 1716b4952e17SGeorge Wilson static uint64_t 1717b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 1718b4952e17SGeorge Wilson { 17190713e232SGeorge Wilson range_seg_t *rs; 1720b4952e17SGeorge Wilson 1721b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17220713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1723b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1724b4952e17SGeorge Wilson 17250713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17260713e232SGeorge Wilson return (rs->rs_start - 1); 1727b4952e17SGeorge Wilson } 1728b4952e17SGeorge Wilson 1729b4952e17SGeorge Wilson /* 1730b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 1731b4952e17SGeorge Wilson */ 1732b4952e17SGeorge Wilson static uint64_t 1733b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 1734b4952e17SGeorge Wilson { 17350713e232SGeorge Wilson range_seg_t *rs; 1736b4952e17SGeorge Wilson 1737b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 17380713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1739b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1740b4952e17SGeorge Wilson 17410713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 17420713e232SGeorge Wilson return (rs->rs_end); 1743b4952e17SGeorge Wilson } 1744b4952e17SGeorge Wilson 1745b4952e17SGeorge Wilson /* 1746b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 1747b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 1748b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 1749b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 1750b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 1751b4952e17SGeorge Wilson * excise the DTLs. 1752b4952e17SGeorge Wilson */ 1753b4952e17SGeorge Wilson static boolean_t 1754b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 1755b4952e17SGeorge Wilson { 1756b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 1757b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1758b4952e17SGeorge Wilson 1759b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 1760b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 1761b4952e17SGeorge Wilson 1762b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 17630713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1764b4952e17SGeorge Wilson return (B_TRUE); 1765b4952e17SGeorge Wilson 1766b4952e17SGeorge Wilson /* 1767b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 1768b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 1769b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 1770b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 1771b4952e17SGeorge Wilson * for excision. 1772b4952e17SGeorge Wilson */ 1773b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1774b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1775b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1776b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1777b4952e17SGeorge Wilson return (B_TRUE); 1778b4952e17SGeorge Wilson } 1779b4952e17SGeorge Wilson return (B_FALSE); 1780b4952e17SGeorge Wilson } 1781b4952e17SGeorge Wilson 1782fa9e4066Sahrens /* 1783fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1784fa9e4066Sahrens */ 1785fa9e4066Sahrens void 1786fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1787fa9e4066Sahrens { 1788ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 17898ad4d6ddSJeff Bonwick avl_tree_t reftree; 17908ad4d6ddSJeff Bonwick int minref; 1791fa9e4066Sahrens 17928ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1793fa9e4066Sahrens 17948ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 17958ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 17968ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 17978ad4d6ddSJeff Bonwick 1798b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 17998ad4d6ddSJeff Bonwick return; 18008ad4d6ddSJeff Bonwick 18018ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 18023f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 18033f9d6ad7SLin Ling 1804fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1805b4952e17SGeorge Wilson 1806b4952e17SGeorge Wilson /* 1807b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 1808b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 1809b4952e17SGeorge Wilson * excise regions on vdevs that were available during 1810b4952e17SGeorge Wilson * the entire duration of this scan. 1811b4952e17SGeorge Wilson */ 1812088f3894Sahrens if (scrub_txg != 0 && 18133f9d6ad7SLin Ling (spa->spa_scrub_started || 1814b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1815b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 1816088f3894Sahrens /* 1817088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1818088f3894Sahrens * did it without rebooting, then the scrub dtl 1819088f3894Sahrens * will be valid, so excise the old region and 1820088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1821088f3894Sahrens * dtl as-is if there was an error. 18228ad4d6ddSJeff Bonwick * 18238ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 18248ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 18258ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 18268ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 18278ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 18288ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 18298ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 18308ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 18318ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1832088f3894Sahrens */ 18330713e232SGeorge Wilson space_reftree_create(&reftree); 18340713e232SGeorge Wilson space_reftree_add_map(&reftree, 18350713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18360713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 18370713e232SGeorge Wilson space_reftree_add_map(&reftree, 18380713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 18390713e232SGeorge Wilson space_reftree_generate_map(&reftree, 18400713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 18410713e232SGeorge Wilson space_reftree_destroy(&reftree); 1842fa9e4066Sahrens } 18430713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 18440713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18450713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1846fa9e4066Sahrens if (scrub_done) 18470713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 18480713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 18498ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 18500713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 18518ad4d6ddSJeff Bonwick else 18520713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 18530713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1854b4952e17SGeorge Wilson 1855b4952e17SGeorge Wilson /* 1856b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 1857b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 1858b4952e17SGeorge Wilson */ 1859b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 18600713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 18610713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 1862b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 1863b4952e17SGeorge Wilson 1864fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1865088f3894Sahrens 1866ecc2d604Sbonwick if (txg != 0) 1867ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1868fa9e4066Sahrens return; 1869fa9e4066Sahrens } 1870fa9e4066Sahrens 1871fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 18728ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 187399bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 187499bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 18758ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 18768ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 18778ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 18788ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 18798ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 18808ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 18818ad4d6ddSJeff Bonwick else 18828ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 18830713e232SGeorge Wilson space_reftree_create(&reftree); 18848ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 18858ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 18868ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 18870713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 18888ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 18898ad4d6ddSJeff Bonwick } 18900713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 18910713e232SGeorge Wilson space_reftree_destroy(&reftree); 1892fa9e4066Sahrens } 18938ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1894fa9e4066Sahrens } 1895fa9e4066Sahrens 18960713e232SGeorge Wilson int 1897fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1898fa9e4066Sahrens { 1899fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1900ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19010713e232SGeorge Wilson int error = 0; 1902fa9e4066Sahrens 19030713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 19040713e232SGeorge Wilson ASSERT(!vd->vdev_ishole); 1905fa9e4066Sahrens 19060713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 19070713e232SGeorge Wilson vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 19080713e232SGeorge Wilson if (error) 19090713e232SGeorge Wilson return (error); 19100713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1911fa9e4066Sahrens 19120713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 191388ecc943SGeorge Wilson 19140713e232SGeorge Wilson /* 19150713e232SGeorge Wilson * Now that we've opened the space_map we need to update 19160713e232SGeorge Wilson * the in-core DTL. 19170713e232SGeorge Wilson */ 19180713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 1919ecc2d604Sbonwick 19200713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 19210713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 19220713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1923fa9e4066Sahrens 19240713e232SGeorge Wilson return (error); 19250713e232SGeorge Wilson } 19260713e232SGeorge Wilson 19270713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 19280713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 19290713e232SGeorge Wilson if (error != 0) 19300713e232SGeorge Wilson break; 19310713e232SGeorge Wilson } 1932fa9e4066Sahrens 1933fa9e4066Sahrens return (error); 1934fa9e4066Sahrens } 1935fa9e4066Sahrens 1936fa9e4066Sahrens void 1937fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1938fa9e4066Sahrens { 1939fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 19400713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 1941ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 19420713e232SGeorge Wilson range_tree_t *rtsync; 19430713e232SGeorge Wilson kmutex_t rtlock; 1944fa9e4066Sahrens dmu_tx_t *tx; 19450713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 1946fa9e4066Sahrens 194788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 19480713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 194988ecc943SGeorge Wilson 1950fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1951fa9e4066Sahrens 19520713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 19530713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 19540713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 19550713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 19560713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 19570713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 1958fa9e4066Sahrens dmu_tx_commit(tx); 1959fa9e4066Sahrens return; 1960fa9e4066Sahrens } 1961fa9e4066Sahrens 19620713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 19630713e232SGeorge Wilson uint64_t new_object; 19640713e232SGeorge Wilson 19650713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 19660713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 19670713e232SGeorge Wilson 19680713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 19690713e232SGeorge Wilson 0, -1ULL, 0, &vd->vdev_dtl_lock)); 19700713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 1971fa9e4066Sahrens } 1972fa9e4066Sahrens 19730713e232SGeorge Wilson mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 1974fa9e4066Sahrens 19750713e232SGeorge Wilson rtsync = range_tree_create(NULL, NULL, &rtlock); 1976fa9e4066Sahrens 19770713e232SGeorge Wilson mutex_enter(&rtlock); 1978fa9e4066Sahrens 1979fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 19800713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 1981fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1982fa9e4066Sahrens 19830713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 19840713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 19850713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 1986fa9e4066Sahrens 19870713e232SGeorge Wilson range_tree_destroy(rtsync); 1988fa9e4066Sahrens 19890713e232SGeorge Wilson mutex_exit(&rtlock); 19900713e232SGeorge Wilson mutex_destroy(&rtlock); 1991fa9e4066Sahrens 19920713e232SGeorge Wilson /* 19930713e232SGeorge Wilson * If the object for the space map has changed then dirty 19940713e232SGeorge Wilson * the top level so that we update the config. 19950713e232SGeorge Wilson */ 19960713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 19970713e232SGeorge Wilson zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 19980713e232SGeorge Wilson "new object %llu", txg, spa_name(spa), object, 19990713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 20000713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 20010713e232SGeorge Wilson } 2002fa9e4066Sahrens 2003fa9e4066Sahrens dmu_tx_commit(tx); 20040713e232SGeorge Wilson 20050713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 20060713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 20070713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2008fa9e4066Sahrens } 2009fa9e4066Sahrens 20108ad4d6ddSJeff Bonwick /* 20118ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 20128ad4d6ddSJeff Bonwick * without losing data. 20138ad4d6ddSJeff Bonwick */ 20148ad4d6ddSJeff Bonwick boolean_t 20158ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 20168ad4d6ddSJeff Bonwick { 20178ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 20188ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 20198ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 20208ad4d6ddSJeff Bonwick boolean_t required; 20218ad4d6ddSJeff Bonwick 20228ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 20238ad4d6ddSJeff Bonwick 20248ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 20258ad4d6ddSJeff Bonwick return (B_TRUE); 20268ad4d6ddSJeff Bonwick 20278ad4d6ddSJeff Bonwick /* 20288ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 20298ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 20308ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 20318ad4d6ddSJeff Bonwick */ 20328ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 20338ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 20348ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 20358ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 20368ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 20378ad4d6ddSJeff Bonwick 2038cb04b873SMark J Musante if (!required && zio_injection_enabled) 2039cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2040cb04b873SMark J Musante 20418ad4d6ddSJeff Bonwick return (required); 20428ad4d6ddSJeff Bonwick } 20438ad4d6ddSJeff Bonwick 2044088f3894Sahrens /* 2045088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2046088f3894Sahrens */ 2047088f3894Sahrens boolean_t 2048088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2049088f3894Sahrens { 2050088f3894Sahrens boolean_t needed = B_FALSE; 2051088f3894Sahrens uint64_t thismin = UINT64_MAX; 2052088f3894Sahrens uint64_t thismax = 0; 2053088f3894Sahrens 2054088f3894Sahrens if (vd->vdev_children == 0) { 2055088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 20560713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 20578ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2058088f3894Sahrens 2059b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2060b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2061088f3894Sahrens needed = B_TRUE; 2062088f3894Sahrens } 2063088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2064088f3894Sahrens } else { 20658ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2066088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2067088f3894Sahrens uint64_t cmin, cmax; 2068088f3894Sahrens 2069088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2070088f3894Sahrens thismin = MIN(thismin, cmin); 2071088f3894Sahrens thismax = MAX(thismax, cmax); 2072088f3894Sahrens needed = B_TRUE; 2073088f3894Sahrens } 2074088f3894Sahrens } 2075088f3894Sahrens } 2076088f3894Sahrens 2077088f3894Sahrens if (needed && minp) { 2078088f3894Sahrens *minp = thismin; 2079088f3894Sahrens *maxp = thismax; 2080088f3894Sahrens } 2081088f3894Sahrens return (needed); 2082088f3894Sahrens } 2083088f3894Sahrens 2084560e6e96Seschrock void 2085ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2086fa9e4066Sahrens { 2087fa9e4066Sahrens /* 2088fa9e4066Sahrens * Recursively load all children. 2089fa9e4066Sahrens */ 20908ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2091560e6e96Seschrock vdev_load(vd->vdev_child[c]); 2092fa9e4066Sahrens 2093fa9e4066Sahrens /* 20940e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2095fa9e4066Sahrens */ 209688ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 2097560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2098560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 2099560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2100560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2101fa9e4066Sahrens 2102fa9e4066Sahrens /* 2103fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2104fa9e4066Sahrens */ 2105560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2106560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2107560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 2108fa9e4066Sahrens } 2109fa9e4066Sahrens 211099653d4eSeschrock /* 2111fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2112fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2113fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2114fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2115fa94a07fSbrendan * repurposed to another pool. 211699653d4eSeschrock */ 211799653d4eSeschrock int 2118fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 211999653d4eSeschrock { 212099653d4eSeschrock nvlist_t *label; 212199653d4eSeschrock uint64_t guid, version; 212299653d4eSeschrock uint64_t state; 212399653d4eSeschrock 2124e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2125c5904d13Seschrock return (0); 2126c5904d13Seschrock 2127dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 212899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 212999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 213099653d4eSeschrock return (-1); 213199653d4eSeschrock } 213299653d4eSeschrock 213399653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2134ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 213599653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 213699653d4eSeschrock guid != vd->vdev_guid || 213799653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 213899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 213999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 214099653d4eSeschrock nvlist_free(label); 214199653d4eSeschrock return (-1); 214299653d4eSeschrock } 214399653d4eSeschrock 214499653d4eSeschrock /* 214599653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 214699653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 214799653d4eSeschrock */ 214899653d4eSeschrock nvlist_free(label); 214999653d4eSeschrock return (0); 215099653d4eSeschrock } 215199653d4eSeschrock 215288ecc943SGeorge Wilson void 215388ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 215488ecc943SGeorge Wilson { 215588ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 215688ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 215788ecc943SGeorge Wilson dmu_tx_t *tx; 215888ecc943SGeorge Wilson 215988ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 216088ecc943SGeorge Wilson 216188ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 21622e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 21632e4c9986SGeorge Wilson 21642e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21652e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21662e4c9986SGeorge Wilson 216788ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 216888ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 216988ecc943SGeorge Wilson 21700713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 217188ecc943SGeorge Wilson continue; 217288ecc943SGeorge Wilson 21730713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 21742e4c9986SGeorge Wilson /* 21752e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 21762e4c9986SGeorge Wilson * was removed then the histogram accounting may 21772e4c9986SGeorge Wilson * not be accurate. Update the histogram information 21782e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 21792e4c9986SGeorge Wilson * and metaslab class are up-to-date. 21802e4c9986SGeorge Wilson */ 21812e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 21822e4c9986SGeorge Wilson 21830713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 21840713e232SGeorge Wilson space_map_free(msp->ms_sm, tx); 21850713e232SGeorge Wilson space_map_close(msp->ms_sm); 21860713e232SGeorge Wilson msp->ms_sm = NULL; 21870713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 218888ecc943SGeorge Wilson } 21892e4c9986SGeorge Wilson 21902e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 21912e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 21922e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 21932e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 21942e4c9986SGeorge Wilson 219588ecc943SGeorge Wilson } 219688ecc943SGeorge Wilson 219788ecc943SGeorge Wilson if (vd->vdev_ms_array) { 219888ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 219988ecc943SGeorge Wilson vd->vdev_ms_array = 0; 220088ecc943SGeorge Wilson } 220188ecc943SGeorge Wilson dmu_tx_commit(tx); 220288ecc943SGeorge Wilson } 220388ecc943SGeorge Wilson 2204fa9e4066Sahrens void 2205fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2206fa9e4066Sahrens { 2207fa9e4066Sahrens metaslab_t *msp; 220880eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2209fa9e4066Sahrens 221088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 221188ecc943SGeorge Wilson 2212fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2213fa9e4066Sahrens metaslab_sync_done(msp, txg); 221480eb36f2SGeorge Wilson 221580eb36f2SGeorge Wilson if (reassess) 221680eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2217fa9e4066Sahrens } 2218fa9e4066Sahrens 2219fa9e4066Sahrens void 2220fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2221fa9e4066Sahrens { 2222fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2223fa9e4066Sahrens vdev_t *lvd; 2224fa9e4066Sahrens metaslab_t *msp; 2225ecc2d604Sbonwick dmu_tx_t *tx; 2226fa9e4066Sahrens 222788ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 222888ecc943SGeorge Wilson 2229ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2230ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2231ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2232ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2233ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2234ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2235ecc2d604Sbonwick vdev_config_dirty(vd); 2236ecc2d604Sbonwick dmu_tx_commit(tx); 2237ecc2d604Sbonwick } 2238fa9e4066Sahrens 22393f9d6ad7SLin Ling /* 22403f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 22413f9d6ad7SLin Ling */ 22423f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 224388ecc943SGeorge Wilson vdev_remove(vd, txg); 224488ecc943SGeorge Wilson 2245ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2246fa9e4066Sahrens metaslab_sync(msp, txg); 2247ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2248ecc2d604Sbonwick } 2249fa9e4066Sahrens 2250fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2251fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2252fa9e4066Sahrens 2253fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2254fa9e4066Sahrens } 2255fa9e4066Sahrens 2256fa9e4066Sahrens uint64_t 2257fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2258fa9e4066Sahrens { 2259fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2260fa9e4066Sahrens } 2261fa9e4066Sahrens 22623d7072f8Seschrock /* 22633d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 22643d7072f8Seschrock * not be opened, and no I/O is attempted. 22653d7072f8Seschrock */ 2266fa9e4066Sahrens int 2267069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2268fa9e4066Sahrens { 22694b964adaSGeorge Wilson vdev_t *vd, *tvd; 2270fa9e4066Sahrens 22718f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2272fa9e4066Sahrens 2273c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2274e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2275e14bb325SJeff Bonwick 22763d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2277e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2278fa9e4066Sahrens 22794b964adaSGeorge Wilson tvd = vd->vdev_top; 22804b964adaSGeorge Wilson 2281069f55e2SEric Schrock /* 2282069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2283069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2284069f55e2SEric Schrock * were faulted. 2285069f55e2SEric Schrock */ 2286069f55e2SEric Schrock vd->vdev_label_aux = aux; 2287069f55e2SEric Schrock 22883d7072f8Seschrock /* 22893d7072f8Seschrock * Faulted state takes precedence over degraded. 22903d7072f8Seschrock */ 229198d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 22923d7072f8Seschrock vd->vdev_faulted = 1ULL; 22933d7072f8Seschrock vd->vdev_degraded = 0ULL; 2294069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 22953d7072f8Seschrock 22963d7072f8Seschrock /* 2297c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2298c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 22993d7072f8Seschrock */ 23004b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 23013d7072f8Seschrock vd->vdev_degraded = 1ULL; 23023d7072f8Seschrock vd->vdev_faulted = 0ULL; 23033d7072f8Seschrock 23043d7072f8Seschrock /* 23053d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 23063d7072f8Seschrock * mark it degraded. 23073d7072f8Seschrock */ 23084b964adaSGeorge Wilson vdev_reopen(tvd); 23093d7072f8Seschrock 2310069f55e2SEric Schrock if (vdev_readable(vd)) 2311069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 23123d7072f8Seschrock } 23133d7072f8Seschrock 2314e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23153d7072f8Seschrock } 23163d7072f8Seschrock 23173d7072f8Seschrock /* 23183d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 23193d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 23203d7072f8Seschrock * as I/O is concerned. 23213d7072f8Seschrock */ 23223d7072f8Seschrock int 2323069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 23243d7072f8Seschrock { 2325c5904d13Seschrock vdev_t *vd; 23260a4e9518Sgw 23278f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 23283d7072f8Seschrock 2329c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2330e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2331e14bb325SJeff Bonwick 23320e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2333e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 23340e34b6a7Sbonwick 23353d7072f8Seschrock /* 23363d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 23373d7072f8Seschrock */ 2338e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2339e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 23403d7072f8Seschrock 23413d7072f8Seschrock vd->vdev_degraded = 1ULL; 23423d7072f8Seschrock if (!vdev_is_dead(vd)) 23433d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2344069f55e2SEric Schrock aux); 23453d7072f8Seschrock 2346e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 23473d7072f8Seschrock } 23483d7072f8Seschrock 23493d7072f8Seschrock /* 2350f7170741SWill Andrews * Online the given vdev. 2351f7170741SWill Andrews * 2352f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2353f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2354f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2355f7170741SWill Andrews * events are generated if the device fails to open. 23563d7072f8Seschrock */ 23573d7072f8Seschrock int 2358e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 23593d7072f8Seschrock { 2360573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 236114372834SHans Rosenfeld boolean_t postevent = B_FALSE; 23623d7072f8Seschrock 23638f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 23643d7072f8Seschrock 2365c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2366e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 23673d7072f8Seschrock 23683d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2369e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2370fa9e4066Sahrens 237114372834SHans Rosenfeld postevent = 237214372834SHans Rosenfeld (vd->vdev_offline == B_TRUE || vd->vdev_tmpoffline == B_TRUE) ? 237314372834SHans Rosenfeld B_TRUE : B_FALSE; 237414372834SHans Rosenfeld 2375573ca77eSGeorge Wilson tvd = vd->vdev_top; 2376fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2377441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2378e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2379e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2380573ca77eSGeorge Wilson 2381573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2382573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2383573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2384573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2385573ca77eSGeorge Wilson } 2386573ca77eSGeorge Wilson 2387573ca77eSGeorge Wilson vdev_reopen(tvd); 23883d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 23893d7072f8Seschrock 2390573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2391573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2392573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2393573ca77eSGeorge Wilson } 2394573ca77eSGeorge Wilson 23953d7072f8Seschrock if (newstate) 23963d7072f8Seschrock *newstate = vd->vdev_state; 23973d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 23983d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 23993d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 24003d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 24013d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2402fa9e4066Sahrens 2403573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2404573ca77eSGeorge Wilson 2405573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2406573ca77eSGeorge Wilson if (vd->vdev_aux) 2407573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2408573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2409573ca77eSGeorge Wilson } 241014372834SHans Rosenfeld 241114372834SHans Rosenfeld if (postevent) 241214372834SHans Rosenfeld spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE); 241314372834SHans Rosenfeld 24148ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2415fa9e4066Sahrens } 2416fa9e4066Sahrens 2417a1521560SJeff Bonwick static int 2418a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2419fa9e4066Sahrens { 2420e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 24218f18d1faSGeorge Wilson int error = 0; 24228f18d1faSGeorge Wilson uint64_t generation; 24238f18d1faSGeorge Wilson metaslab_group_t *mg; 24240a4e9518Sgw 24258f18d1faSGeorge Wilson top: 24268f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2427fa9e4066Sahrens 2428c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2429e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2430fa9e4066Sahrens 24310e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2432e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 24330e34b6a7Sbonwick 2434e6ca193dSGeorge Wilson tvd = vd->vdev_top; 24358f18d1faSGeorge Wilson mg = tvd->vdev_mg; 24368f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2437e6ca193dSGeorge Wilson 2438fa9e4066Sahrens /* 2439ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2440fa9e4066Sahrens */ 2441ecc2d604Sbonwick if (!vd->vdev_offline) { 2442ecc2d604Sbonwick /* 24438ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2444e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2445e6ca193dSGeorge Wilson * expendable. 2446ecc2d604Sbonwick */ 2447e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2448e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2449e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2450fa9e4066Sahrens 24518f18d1faSGeorge Wilson /* 2452b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2453b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2454b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2455b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 24568f18d1faSGeorge Wilson */ 24578f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 24588f18d1faSGeorge Wilson /* 24598f18d1faSGeorge Wilson * Prevent any future allocations. 24608f18d1faSGeorge Wilson */ 2461a1521560SJeff Bonwick metaslab_group_passivate(mg); 24628f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24638f18d1faSGeorge Wilson 24641195e687SMark J Musante error = spa_offline_log(spa); 24658f18d1faSGeorge Wilson 24668f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 24678f18d1faSGeorge Wilson 24688f18d1faSGeorge Wilson /* 24698f18d1faSGeorge Wilson * Check to see if the config has changed. 24708f18d1faSGeorge Wilson */ 24718f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2472a1521560SJeff Bonwick metaslab_group_activate(mg); 24738f18d1faSGeorge Wilson if (error) 24748f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 24758f18d1faSGeorge Wilson vd, error)); 24768f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 24778f18d1faSGeorge Wilson goto top; 24788f18d1faSGeorge Wilson } 2479fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 24808f18d1faSGeorge Wilson } 24818f18d1faSGeorge Wilson 2482ecc2d604Sbonwick /* 2483ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2484e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2485e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2486e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2487ecc2d604Sbonwick */ 2488ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2489e6ca193dSGeorge Wilson vdev_reopen(tvd); 2490e6ca193dSGeorge Wilson 2491e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2492e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2493ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2494e6ca193dSGeorge Wilson vdev_reopen(tvd); 2495e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2496ecc2d604Sbonwick } 24978f18d1faSGeorge Wilson 24988f18d1faSGeorge Wilson /* 24998f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 25008f18d1faSGeorge Wilson * once we online the device it's open for business. 25018f18d1faSGeorge Wilson */ 25028f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2503a1521560SJeff Bonwick metaslab_group_activate(mg); 2504fa9e4066Sahrens } 2505fa9e4066Sahrens 2506e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2507ecc2d604Sbonwick 25088f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2509fa9e4066Sahrens } 2510fa9e4066Sahrens 2511a1521560SJeff Bonwick int 2512a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2513a1521560SJeff Bonwick { 2514a1521560SJeff Bonwick int error; 2515a1521560SJeff Bonwick 2516a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2517a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2518a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2519a1521560SJeff Bonwick 2520a1521560SJeff Bonwick return (error); 2521a1521560SJeff Bonwick } 2522a1521560SJeff Bonwick 2523ea8dc4b6Seschrock /* 2524ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2525ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2526ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2527ea8dc4b6Seschrock */ 2528ea8dc4b6Seschrock void 2529e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2530fa9e4066Sahrens { 2531e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2532e14bb325SJeff Bonwick 2533e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2534fa9e4066Sahrens 2535ea8dc4b6Seschrock if (vd == NULL) 2536e14bb325SJeff Bonwick vd = rvd; 2537fa9e4066Sahrens 2538ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2539ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2540ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2541fa9e4066Sahrens 2542e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2543e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 25443d7072f8Seschrock 25453d7072f8Seschrock /* 25468a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 25478a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 25488a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 25498a79c1b5Sek * written out to disk. 25503d7072f8Seschrock */ 2551e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2552e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 25538a79c1b5Sek 2554096d22d4SEric Schrock /* 2555096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2556096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2557096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2558096d22d4SEric Schrock */ 2559096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2560096d22d4SEric Schrock 25614b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2562e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2563e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2564e14bb325SJeff Bonwick 2565f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 25663d7072f8Seschrock 2567096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2568096d22d4SEric Schrock 2569f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2570e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2571e14bb325SJeff Bonwick 2572e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2573bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 25743d7072f8Seschrock 25753d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 25763d7072f8Seschrock } 2577096d22d4SEric Schrock 2578096d22d4SEric Schrock /* 2579096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2580096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2581096d22d4SEric Schrock * done in response to the FMA fault. 2582096d22d4SEric Schrock */ 2583096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2584096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2585096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2586096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2587fa9e4066Sahrens } 2588fa9e4066Sahrens 2589e14bb325SJeff Bonwick boolean_t 2590e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 25910a4e9518Sgw { 259288ecc943SGeorge Wilson /* 259388ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 259488ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 259588ecc943SGeorge Wilson * these types of devices in the various code paths. 259688ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 259788ecc943SGeorge Wilson * before issuing I/O to them. 259888ecc943SGeorge Wilson */ 259988ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 260088ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 26010a4e9518Sgw } 26020a4e9518Sgw 2603e14bb325SJeff Bonwick boolean_t 2604e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 26050a4e9518Sgw { 2606e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 26070a4e9518Sgw } 26080a4e9518Sgw 2609e14bb325SJeff Bonwick boolean_t 2610e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2611fa9e4066Sahrens { 2612e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2613fa9e4066Sahrens } 2614fa9e4066Sahrens 2615a31e6787SGeorge Wilson boolean_t 2616a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2617a31e6787SGeorge Wilson { 26188ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 26198ad4d6ddSJeff Bonwick 2620a31e6787SGeorge Wilson /* 26218ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2622a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2623a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 26248ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 26258ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 26268ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2627a31e6787SGeorge Wilson */ 26288ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 26293f9d6ad7SLin Ling !vd->vdev_cant_write && !vd->vdev_ishole); 2630a31e6787SGeorge Wilson } 2631a31e6787SGeorge Wilson 2632e14bb325SJeff Bonwick boolean_t 2633e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2634fa9e4066Sahrens { 2635e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2636fa9e4066Sahrens 2637e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2638e14bb325SJeff Bonwick return (B_FALSE); 2639fa9e4066Sahrens 2640e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2641e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2642fa9e4066Sahrens 2643e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2644e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2645fa9e4066Sahrens 2646e14bb325SJeff Bonwick return (B_TRUE); 2647fa9e4066Sahrens } 2648fa9e4066Sahrens 2649fa9e4066Sahrens /* 2650fa9e4066Sahrens * Get statistics for the given vdev. 2651fa9e4066Sahrens */ 2652fa9e4066Sahrens void 2653fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2654fa9e4066Sahrens { 26552e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 26562e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 26572e4c9986SGeorge Wilson 26582e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2659fa9e4066Sahrens 2660fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2661fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2662fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2663fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2664573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2665573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2666573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 26674263d13fSGeorge Wilson vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 26682986efa8SAlex Reece if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 26692e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 26702986efa8SAlex Reece } 2671fa9e4066Sahrens 2672fa9e4066Sahrens /* 2673fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2674fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2675fa9e4066Sahrens */ 2676fa9e4066Sahrens if (vd == rvd) { 2677e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2678fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2679fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2680fa9e4066Sahrens 2681e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2682fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2683fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2684fa9e4066Sahrens } 26853f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2686fa9e4066Sahrens } 2687fa9e4066Sahrens } 26882e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 2689fa9e4066Sahrens } 2690fa9e4066Sahrens 2691fa94a07fSbrendan void 2692fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2693fa94a07fSbrendan { 2694fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2695fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2696fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2697fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2698fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2699fa94a07fSbrendan } 2700fa94a07fSbrendan 27013f9d6ad7SLin Ling void 27023f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 27033f9d6ad7SLin Ling { 27043f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 27053f9d6ad7SLin Ling 27063f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 27073f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 27083f9d6ad7SLin Ling 27093f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 27103f9d6ad7SLin Ling vs->vs_scan_processed = 0; 27113f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 27123f9d6ad7SLin Ling } 27133f9d6ad7SLin Ling 2714fa9e4066Sahrens void 2715e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2716fa9e4066Sahrens { 27178ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 27188ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2719e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2720fa9e4066Sahrens vdev_t *pvd; 2721fa9e4066Sahrens uint64_t txg = zio->io_txg; 2722fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2723fa9e4066Sahrens zio_type_t type = zio->io_type; 2724fa9e4066Sahrens int flags = zio->io_flags; 2725fa9e4066Sahrens 2726e14bb325SJeff Bonwick /* 2727e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2728e14bb325SJeff Bonwick */ 2729e14bb325SJeff Bonwick if (zio->io_gang_tree) 2730e14bb325SJeff Bonwick return; 2731e14bb325SJeff Bonwick 2732fa9e4066Sahrens if (zio->io_error == 0) { 2733e14bb325SJeff Bonwick /* 2734e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2735e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2736e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2737e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2738e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2739e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2740e14bb325SJeff Bonwick * remain zero, which is what we want.) 2741e14bb325SJeff Bonwick * 2742e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2743e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2744e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2745e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2746e14bb325SJeff Bonwick */ 2747e14bb325SJeff Bonwick if (vd == rvd) 2748e14bb325SJeff Bonwick return; 2749e14bb325SJeff Bonwick 2750e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 27518ad4d6ddSJeff Bonwick 27528ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 27538ad4d6ddSJeff Bonwick return; 27548ad4d6ddSJeff Bonwick 27558ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 27568ad4d6ddSJeff Bonwick 2757e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 275844ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 27593f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 27603f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 27613f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 27623f9d6ad7SLin Ling 27633f9d6ad7SLin Ling /* XXX cleanup? */ 27643f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 27653f9d6ad7SLin Ling atomic_add_64(processed, psize); 27663f9d6ad7SLin Ling vs->vs_scan_processed += psize; 27673f9d6ad7SLin Ling } 27683f9d6ad7SLin Ling 27698ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2770e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2771fa9e4066Sahrens } 27728ad4d6ddSJeff Bonwick 27738ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 27748ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 27758ad4d6ddSJeff Bonwick 27768ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2777fa9e4066Sahrens return; 2778fa9e4066Sahrens } 2779fa9e4066Sahrens 2780fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2781fa9e4066Sahrens return; 2782fa9e4066Sahrens 27838956713aSEric Schrock /* 27848956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 27858956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 27868956713aSEric Schrock * hard errors, when in reality they can happen for any number of 27878956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 27888956713aSEric Schrock */ 27898956713aSEric Schrock if (zio->io_error == EIO && 27908956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 27918956713aSEric Schrock return; 27928956713aSEric Schrock 27938f18d1faSGeorge Wilson /* 27948f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 27958f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 27968f18d1faSGeorge Wilson * errors. 27978f18d1faSGeorge Wilson */ 27988f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 27998f18d1faSGeorge Wilson return; 28008f18d1faSGeorge Wilson 2801e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2802b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2803e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2804e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2805e14bb325SJeff Bonwick else 2806e14bb325SJeff Bonwick vs->vs_read_errors++; 2807fa9e4066Sahrens } 2808b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2809e14bb325SJeff Bonwick vs->vs_write_errors++; 2810e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2811fa9e4066Sahrens 28128ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 28138ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 281444ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2815b24ab676SJeff Bonwick spa->spa_claiming)) { 28168ad4d6ddSJeff Bonwick /* 2817b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2818b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2819b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2820b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2821b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2822b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2823b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2824b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 28258ad4d6ddSJeff Bonwick * 28268ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 28278ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 28288ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 28298ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 28308ad4d6ddSJeff Bonwick */ 28318ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 28328ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 283344ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 28348ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 28358ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 28368ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2837b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2838b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2839b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2840b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 28418ad4d6ddSJeff Bonwick } 2842b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 28438ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2844fa9e4066Sahrens return; 28458ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 28468ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 28478ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2848fa9e4066Sahrens } 28498ad4d6ddSJeff Bonwick if (vd != rvd) 28508ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2851fa9e4066Sahrens } 2852fa9e4066Sahrens } 2853fa9e4066Sahrens 2854fa9e4066Sahrens /* 2855b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2856b24ab676SJeff Bonwick * and the root vdev. 2857fa9e4066Sahrens */ 2858fa9e4066Sahrens void 2859b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2860b24ab676SJeff Bonwick int64_t space_delta) 2861fa9e4066Sahrens { 286299653d4eSeschrock int64_t dspace_delta = space_delta; 28638654d025Sperrin spa_t *spa = vd->vdev_spa; 28648654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2865b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2866b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2867fa9e4066Sahrens 28688654d025Sperrin ASSERT(vd == vd->vdev_top); 286999653d4eSeschrock 28708654d025Sperrin /* 28718654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 28728654d025Sperrin * factor. We must calculate this here and not at the root vdev 28738654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 28748654d025Sperrin * childrens', thus not accurate enough for us. 28758654d025Sperrin */ 28768654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2877e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 28788654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 28798654d025Sperrin vd->vdev_deflate_ratio; 28808654d025Sperrin 28818654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 28828654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2883b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 28848654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 28858654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 28868654d025Sperrin 2887b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2888fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2889fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2890b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2891fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2892fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2893fa94a07fSbrendan } 2894b24ab676SJeff Bonwick 2895b24ab676SJeff Bonwick if (mc != NULL) { 2896b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2897b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2898b24ab676SJeff Bonwick 2899b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2900b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2901b24ab676SJeff Bonwick } 2902fa9e4066Sahrens } 2903fa9e4066Sahrens 2904fa9e4066Sahrens /* 2905fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2906fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2907fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2908fa9e4066Sahrens */ 2909fa9e4066Sahrens void 2910fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2911fa9e4066Sahrens { 2912fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2913fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2914fa9e4066Sahrens int c; 2915fa9e4066Sahrens 2916f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2917f9af39baSGeorge Wilson 2918c5904d13Seschrock /* 29196809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 29206809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2921c5904d13Seschrock */ 2922c5904d13Seschrock if (vd->vdev_aux != NULL) { 2923c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2924c5904d13Seschrock nvlist_t **aux; 2925c5904d13Seschrock uint_t naux; 2926c5904d13Seschrock 2927c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2928c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2929c5904d13Seschrock break; 2930c5904d13Seschrock } 2931c5904d13Seschrock 2932e14bb325SJeff Bonwick if (c == sav->sav_count) { 2933e14bb325SJeff Bonwick /* 2934e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2935e14bb325SJeff Bonwick */ 2936e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2937e14bb325SJeff Bonwick return; 2938e14bb325SJeff Bonwick } 2939e14bb325SJeff Bonwick 2940c5904d13Seschrock sav->sav_sync = B_TRUE; 2941c5904d13Seschrock 29426809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 29436809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 29446809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 29456809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 29466809eb4eSEric Schrock } 2947c5904d13Seschrock 2948c5904d13Seschrock ASSERT(c < naux); 2949c5904d13Seschrock 2950c5904d13Seschrock /* 2951c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2952c5904d13Seschrock * sketchy, but it will work. 2953c5904d13Seschrock */ 2954c5904d13Seschrock nvlist_free(aux[c]); 29553f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2956c5904d13Seschrock 2957c5904d13Seschrock return; 2958c5904d13Seschrock } 2959c5904d13Seschrock 29605dabedeeSbonwick /* 2961e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2962e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2963e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 29645dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 29655dabedeeSbonwick */ 2966e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2967e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2968e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29695dabedeeSbonwick 2970fa9e4066Sahrens if (vd == rvd) { 2971fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2972fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2973fa9e4066Sahrens } else { 2974fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2975fa9e4066Sahrens 297688ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 297788ecc943SGeorge Wilson !vd->vdev_ishole) 2978e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2979fa9e4066Sahrens } 2980fa9e4066Sahrens } 2981fa9e4066Sahrens 2982fa9e4066Sahrens void 2983fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2984fa9e4066Sahrens { 29855dabedeeSbonwick spa_t *spa = vd->vdev_spa; 29865dabedeeSbonwick 2987e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2988e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2989e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 29905dabedeeSbonwick 2991e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2992e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2993e14bb325SJeff Bonwick } 2994e14bb325SJeff Bonwick 2995e14bb325SJeff Bonwick /* 2996e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2997e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2998e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2999e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3000e14bb325SJeff Bonwick */ 3001e14bb325SJeff Bonwick void 3002e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3003e14bb325SJeff Bonwick { 3004e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3005e14bb325SJeff Bonwick 3006f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3007e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3008e14bb325SJeff Bonwick 3009e14bb325SJeff Bonwick /* 3010e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3011e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3012e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3013e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3014e14bb325SJeff Bonwick */ 3015e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3016e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3017e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3018e14bb325SJeff Bonwick 3019b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 3020e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3021e14bb325SJeff Bonwick } 3022e14bb325SJeff Bonwick 3023e14bb325SJeff Bonwick void 3024e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3025e14bb325SJeff Bonwick { 3026e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3027e14bb325SJeff Bonwick 3028e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3029e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3030e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3031e14bb325SJeff Bonwick 3032e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3033e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3034fa9e4066Sahrens } 3035fa9e4066Sahrens 303632b87932Sek /* 303732b87932Sek * Propagate vdev state up from children to parent. 303832b87932Sek */ 303944cd46caSbillm void 304044cd46caSbillm vdev_propagate_state(vdev_t *vd) 304144cd46caSbillm { 30428ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 30438ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 304444cd46caSbillm int degraded = 0, faulted = 0; 304544cd46caSbillm int corrupted = 0; 304644cd46caSbillm vdev_t *child; 304744cd46caSbillm 30483d7072f8Seschrock if (vd->vdev_children > 0) { 3049573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 30503d7072f8Seschrock child = vd->vdev_child[c]; 305151ece835Seschrock 305288ecc943SGeorge Wilson /* 305388ecc943SGeorge Wilson * Don't factor holes into the decision. 305488ecc943SGeorge Wilson */ 305588ecc943SGeorge Wilson if (child->vdev_ishole) 305688ecc943SGeorge Wilson continue; 305788ecc943SGeorge Wilson 3058e14bb325SJeff Bonwick if (!vdev_readable(child) || 30598ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 306051ece835Seschrock /* 306151ece835Seschrock * Root special: if there is a top-level log 306251ece835Seschrock * device, treat the root vdev as if it were 306351ece835Seschrock * degraded. 306451ece835Seschrock */ 306551ece835Seschrock if (child->vdev_islog && vd == rvd) 306651ece835Seschrock degraded++; 306751ece835Seschrock else 306851ece835Seschrock faulted++; 306951ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 30703d7072f8Seschrock degraded++; 307151ece835Seschrock } 307244cd46caSbillm 30733d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 30743d7072f8Seschrock corrupted++; 30753d7072f8Seschrock } 307644cd46caSbillm 30773d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 30783d7072f8Seschrock 30793d7072f8Seschrock /* 3080e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 30813d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 30823d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 30833d7072f8Seschrock * replicas'. 30843d7072f8Seschrock */ 30853d7072f8Seschrock if (corrupted && vd == rvd && 30863d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 30873d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 30883d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 30893d7072f8Seschrock } 30903d7072f8Seschrock 309151ece835Seschrock if (vd->vdev_parent) 30923d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 309344cd46caSbillm } 309444cd46caSbillm 3095fa9e4066Sahrens /* 3096ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3097ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3098ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3099ea8dc4b6Seschrock * 3100ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3101ea8dc4b6Seschrock * generated. 3102fa9e4066Sahrens */ 3103fa9e4066Sahrens void 3104ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3105fa9e4066Sahrens { 3106560e6e96Seschrock uint64_t save_state; 3107c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3108ea8dc4b6Seschrock 3109ea8dc4b6Seschrock if (state == vd->vdev_state) { 3110ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3111fa9e4066Sahrens return; 3112ea8dc4b6Seschrock } 3113ea8dc4b6Seschrock 3114560e6e96Seschrock save_state = vd->vdev_state; 3115fa9e4066Sahrens 3116fa9e4066Sahrens vd->vdev_state = state; 3117fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3118fa9e4066Sahrens 31193d7072f8Seschrock /* 31203d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 312198d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 312298d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 312398d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 312498d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 312598d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 312698d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 312798d1cbfeSGeorge Wilson * children. 312898d1cbfeSGeorge Wilson */ 312998d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 313098d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 31313d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 31323d7072f8Seschrock 3133069f55e2SEric Schrock /* 3134069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3135069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3136069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3137069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3138069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3139069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3140069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3141069f55e2SEric Schrock */ 3142069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3143069f55e2SEric Schrock vd->vdev_prevstate != state) 3144069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3145069f55e2SEric Schrock 31463d7072f8Seschrock if (vd->vdev_removed && 31473d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 31483d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 31493d7072f8Seschrock /* 31503d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 31513d7072f8Seschrock * device was previously marked removed and someone attempted to 31523d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 31533d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 31543d7072f8Seschrock * it is one of our special test online cases, which is only 31553d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 31563d7072f8Seschrock * fault. 31573d7072f8Seschrock */ 31583d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 31593d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 31603d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 31613d7072f8Seschrock vd->vdev_removed = B_TRUE; 31623d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3163ea8dc4b6Seschrock /* 3164cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3165cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3166cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3167cb04b873SMark J Musante * is not considered an error. 3168ea8dc4b6Seschrock */ 3169cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3170cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3171560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3172560e6e96Seschrock vd->vdev_not_present = 1; 3173560e6e96Seschrock 3174560e6e96Seschrock /* 3175560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3176560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3177560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3178560e6e96Seschrock * want to post the ereport if the device was already in the 3179560e6e96Seschrock * CANT_OPEN state beforehand. 31803d7072f8Seschrock * 31813d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 31823d7072f8Seschrock * online the device in response to an insertion event. If we 31833d7072f8Seschrock * hit this case, then we have detected an insertion event for a 31843d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 31853d7072f8Seschrock * In this scenario, we don't post an ereport because we are 31863d7072f8Seschrock * about to replace the device, or attempt an online with 31873d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3188560e6e96Seschrock */ 31893d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 31903d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3191c5904d13Seschrock vd != spa->spa_root_vdev) { 3192ea8dc4b6Seschrock const char *class; 3193ea8dc4b6Seschrock 3194ea8dc4b6Seschrock switch (aux) { 3195ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3196ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3197ea8dc4b6Seschrock break; 3198ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3199ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3200ea8dc4b6Seschrock break; 3201ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3202ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3203ea8dc4b6Seschrock break; 3204ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3205ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3206ea8dc4b6Seschrock break; 3207ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3208ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3209ea8dc4b6Seschrock break; 3210ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3211ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3212ea8dc4b6Seschrock break; 3213ea8dc4b6Seschrock default: 3214ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3215ea8dc4b6Seschrock } 3216ea8dc4b6Seschrock 3217c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3218ea8dc4b6Seschrock } 3219ea8dc4b6Seschrock 32203d7072f8Seschrock /* Erase any notion of persistent removed state */ 32213d7072f8Seschrock vd->vdev_removed = B_FALSE; 32223d7072f8Seschrock } else { 32233d7072f8Seschrock vd->vdev_removed = B_FALSE; 32243d7072f8Seschrock } 3225ea8dc4b6Seschrock 32268b33d774STim Haley if (!isopen && vd->vdev_parent) 32278b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3228fa9e4066Sahrens } 322915e6edf1Sgw 323015e6edf1Sgw /* 323115e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 323215e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 323315e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 323415e6edf1Sgw * can be wholedisks. 323515e6edf1Sgw */ 323615e6edf1Sgw boolean_t 323715e6edf1Sgw vdev_is_bootable(vdev_t *vd) 323815e6edf1Sgw { 323915e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 324015e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 324115e6edf1Sgw 324215e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 324315e6edf1Sgw vd->vdev_children > 1) { 324415e6edf1Sgw return (B_FALSE); 324515e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 324615e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 324715e6edf1Sgw return (B_FALSE); 324815e6edf1Sgw } 324915e6edf1Sgw } 325015e6edf1Sgw 3251573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 325215e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 325315e6edf1Sgw return (B_FALSE); 325415e6edf1Sgw } 325515e6edf1Sgw return (B_TRUE); 325615e6edf1Sgw } 3257e6ca193dSGeorge Wilson 325888ecc943SGeorge Wilson /* 325988ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 326088ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 32614b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 32624b964adaSGeorge Wilson * device in the current vdev tree (nvd). 326388ecc943SGeorge Wilson */ 3264e6ca193dSGeorge Wilson void 326588ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3266e6ca193dSGeorge Wilson { 326788ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3268e6ca193dSGeorge Wilson 32694b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 327088ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 327188ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3272e6ca193dSGeorge Wilson 327388ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 327488ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3275e6ca193dSGeorge Wilson 32764b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3277e6ca193dSGeorge Wilson /* 32784b964adaSGeorge Wilson * Restore the persistent vdev state 3279e6ca193dSGeorge Wilson */ 328088ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 32814b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 32824b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 32834b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3284e6ca193dSGeorge Wilson } 3285e6ca193dSGeorge Wilson } 3286573ca77eSGeorge Wilson 32874b964adaSGeorge Wilson /* 32884b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 32894b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 32904b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 32914b964adaSGeorge Wilson */ 32924b964adaSGeorge Wilson boolean_t 32934b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 32944b964adaSGeorge Wilson { 32954b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 32964b964adaSGeorge Wilson !vd->vdev_removed) 32974b964adaSGeorge Wilson return (B_TRUE); 32984b964adaSGeorge Wilson 32994b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 33004b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 33014b964adaSGeorge Wilson return (B_TRUE); 33024b964adaSGeorge Wilson 33034b964adaSGeorge Wilson return (B_FALSE); 33044b964adaSGeorge Wilson } 33054b964adaSGeorge Wilson 3306573ca77eSGeorge Wilson /* 3307573ca77eSGeorge Wilson * Expand a vdev if possible. 3308573ca77eSGeorge Wilson */ 3309573ca77eSGeorge Wilson void 3310573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3311573ca77eSGeorge Wilson { 3312573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3313573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3314573ca77eSGeorge Wilson 3315573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3316573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3317573ca77eSGeorge Wilson vdev_config_dirty(vd); 3318573ca77eSGeorge Wilson } 3319573ca77eSGeorge Wilson } 33201195e687SMark J Musante 33211195e687SMark J Musante /* 33221195e687SMark J Musante * Split a vdev. 33231195e687SMark J Musante */ 33241195e687SMark J Musante void 33251195e687SMark J Musante vdev_split(vdev_t *vd) 33261195e687SMark J Musante { 33271195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 33281195e687SMark J Musante 33291195e687SMark J Musante vdev_remove_child(pvd, vd); 33301195e687SMark J Musante vdev_compact_children(pvd); 33311195e687SMark J Musante 33321195e687SMark J Musante cvd = pvd->vdev_child[0]; 33331195e687SMark J Musante if (pvd->vdev_children == 1) { 33341195e687SMark J Musante vdev_remove_parent(cvd); 33351195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 33361195e687SMark J Musante } 33371195e687SMark J Musante vdev_propagate_state(cvd); 33381195e687SMark J Musante } 3339283b8460SGeorge.Wilson 3340283b8460SGeorge.Wilson void 3341283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3342283b8460SGeorge.Wilson { 3343283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3344283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3345283b8460SGeorge.Wilson 3346283b8460SGeorge.Wilson vdev_deadman(cvd); 3347283b8460SGeorge.Wilson } 3348283b8460SGeorge.Wilson 3349283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3350283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3351283b8460SGeorge.Wilson 3352283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 335369962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3354283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3355283b8460SGeorge.Wilson zio_t *fio; 3356283b8460SGeorge.Wilson uint64_t delta; 3357283b8460SGeorge.Wilson 3358283b8460SGeorge.Wilson /* 3359283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3360283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3361283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3362283b8460SGeorge.Wilson */ 336369962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3364c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3365c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 3366c55e05cbSMatthew Ahrens zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3367c55e05cbSMatthew Ahrens "delta %lluns, last io %lluns", 3368283b8460SGeorge.Wilson fio->io_timestamp, delta, 3369283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3370283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3371283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3372283b8460SGeorge.Wilson } 3373283b8460SGeorge.Wilson } 3374283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3375283b8460SGeorge.Wilson } 3376283b8460SGeorge.Wilson } 3377