1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24*e9103aaeSGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25*e9103aaeSGarrett D'Amore * Copyright (c) 2011 by Delphix. All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/zfs_context.h> 29ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 30fa9e4066Sahrens #include <sys/spa.h> 31fa9e4066Sahrens #include <sys/spa_impl.h> 32fa9e4066Sahrens #include <sys/dmu.h> 33fa9e4066Sahrens #include <sys/dmu_tx.h> 34fa9e4066Sahrens #include <sys/vdev_impl.h> 35fa9e4066Sahrens #include <sys/uberblock_impl.h> 36fa9e4066Sahrens #include <sys/metaslab.h> 37fa9e4066Sahrens #include <sys/metaslab_impl.h> 38fa9e4066Sahrens #include <sys/space_map.h> 39fa9e4066Sahrens #include <sys/zio.h> 40fa9e4066Sahrens #include <sys/zap.h> 41fa9e4066Sahrens #include <sys/fs/zfs.h> 42c5904d13Seschrock #include <sys/arc.h> 43e6ca193dSGeorge Wilson #include <sys/zil.h> 443f9d6ad7SLin Ling #include <sys/dsl_scan.h> 45fa9e4066Sahrens 46fa9e4066Sahrens /* 47fa9e4066Sahrens * Virtual device management. 48fa9e4066Sahrens */ 49fa9e4066Sahrens 50fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 51fa9e4066Sahrens &vdev_root_ops, 52fa9e4066Sahrens &vdev_raidz_ops, 53fa9e4066Sahrens &vdev_mirror_ops, 54fa9e4066Sahrens &vdev_replacing_ops, 5599653d4eSeschrock &vdev_spare_ops, 56fa9e4066Sahrens &vdev_disk_ops, 57fa9e4066Sahrens &vdev_file_ops, 58fa9e4066Sahrens &vdev_missing_ops, 5988ecc943SGeorge Wilson &vdev_hole_ops, 60fa9e4066Sahrens NULL 61fa9e4066Sahrens }; 62fa9e4066Sahrens 63088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 64088f3894Sahrens int zfs_scrub_limit = 10; 6505b2b3b8Smishra 66fa9e4066Sahrens /* 67fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 68fa9e4066Sahrens */ 69fa9e4066Sahrens static vdev_ops_t * 70fa9e4066Sahrens vdev_getops(const char *type) 71fa9e4066Sahrens { 72fa9e4066Sahrens vdev_ops_t *ops, **opspp; 73fa9e4066Sahrens 74fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 75fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 76fa9e4066Sahrens break; 77fa9e4066Sahrens 78fa9e4066Sahrens return (ops); 79fa9e4066Sahrens } 80fa9e4066Sahrens 81fa9e4066Sahrens /* 82fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 83fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 84fa9e4066Sahrens */ 85fa9e4066Sahrens uint64_t 86fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 87fa9e4066Sahrens { 88ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 89fa9e4066Sahrens uint64_t csize; 90fa9e4066Sahrens 91573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 92fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 93fa9e4066Sahrens asize = MAX(asize, csize); 94fa9e4066Sahrens } 95fa9e4066Sahrens 96fa9e4066Sahrens return (asize); 97fa9e4066Sahrens } 98fa9e4066Sahrens 992a79c5feSlling /* 100573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 101573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 102573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 103573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1042a79c5feSlling */ 1052a79c5feSlling uint64_t 106573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1072a79c5feSlling { 108573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1092a79c5feSlling 110573ca77eSGeorge Wilson /* 111573ca77eSGeorge Wilson * The our parent is NULL (inactive spare or cache) or is the root, 112573ca77eSGeorge Wilson * just return our own asize. 113573ca77eSGeorge Wilson */ 114573ca77eSGeorge Wilson if (pvd == NULL) 115573ca77eSGeorge Wilson return (vd->vdev_asize); 1162a79c5feSlling 1172a79c5feSlling /* 118573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 119573ca77eSGeorge Wilson * to the nearest metaslab. 1202a79c5feSlling */ 121573ca77eSGeorge Wilson if (vd == vd->vdev_top) 122573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 1232a79c5feSlling 124573ca77eSGeorge Wilson /* 125573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 126573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 127573ca77eSGeorge Wilson */ 128573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 129573ca77eSGeorge Wilson return (pvd->vdev_min_asize / pvd->vdev_children); 1302a79c5feSlling 131573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 132573ca77eSGeorge Wilson } 1332a79c5feSlling 134573ca77eSGeorge Wilson void 135573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 136573ca77eSGeorge Wilson { 137573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 138573ca77eSGeorge Wilson 139573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 140573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 1412a79c5feSlling } 1422a79c5feSlling 143fa9e4066Sahrens vdev_t * 144fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 145fa9e4066Sahrens { 146fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 147fa9e4066Sahrens 148e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 149e05725b1Sbonwick 150088f3894Sahrens if (vdev < rvd->vdev_children) { 151088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 152fa9e4066Sahrens return (rvd->vdev_child[vdev]); 153088f3894Sahrens } 154fa9e4066Sahrens 155fa9e4066Sahrens return (NULL); 156fa9e4066Sahrens } 157fa9e4066Sahrens 158fa9e4066Sahrens vdev_t * 159fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 160fa9e4066Sahrens { 161fa9e4066Sahrens vdev_t *mvd; 162fa9e4066Sahrens 1630e34b6a7Sbonwick if (vd->vdev_guid == guid) 164fa9e4066Sahrens return (vd); 165fa9e4066Sahrens 166573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 167fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 168fa9e4066Sahrens NULL) 169fa9e4066Sahrens return (mvd); 170fa9e4066Sahrens 171fa9e4066Sahrens return (NULL); 172fa9e4066Sahrens } 173fa9e4066Sahrens 174fa9e4066Sahrens void 175fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 176fa9e4066Sahrens { 177fa9e4066Sahrens size_t oldsize, newsize; 178fa9e4066Sahrens uint64_t id = cvd->vdev_id; 179fa9e4066Sahrens vdev_t **newchild; 180fa9e4066Sahrens 181e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 182fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 183fa9e4066Sahrens 184fa9e4066Sahrens cvd->vdev_parent = pvd; 185fa9e4066Sahrens 186fa9e4066Sahrens if (pvd == NULL) 187fa9e4066Sahrens return; 188fa9e4066Sahrens 189fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 190fa9e4066Sahrens 191fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 192fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 193fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 194fa9e4066Sahrens 195fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 196fa9e4066Sahrens if (pvd->vdev_child != NULL) { 197fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 198fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 199fa9e4066Sahrens } 200fa9e4066Sahrens 201fa9e4066Sahrens pvd->vdev_child = newchild; 202fa9e4066Sahrens pvd->vdev_child[id] = cvd; 203fa9e4066Sahrens 204fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 205fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 206fa9e4066Sahrens 207fa9e4066Sahrens /* 208fa9e4066Sahrens * Walk up all ancestors to update guid sum. 209fa9e4066Sahrens */ 210fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 211fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 212fa9e4066Sahrens } 213fa9e4066Sahrens 214fa9e4066Sahrens void 215fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 216fa9e4066Sahrens { 217fa9e4066Sahrens int c; 218fa9e4066Sahrens uint_t id = cvd->vdev_id; 219fa9e4066Sahrens 220fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 221fa9e4066Sahrens 222fa9e4066Sahrens if (pvd == NULL) 223fa9e4066Sahrens return; 224fa9e4066Sahrens 225fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 226fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 227fa9e4066Sahrens 228fa9e4066Sahrens pvd->vdev_child[id] = NULL; 229fa9e4066Sahrens cvd->vdev_parent = NULL; 230fa9e4066Sahrens 231fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 232fa9e4066Sahrens if (pvd->vdev_child[c]) 233fa9e4066Sahrens break; 234fa9e4066Sahrens 235fa9e4066Sahrens if (c == pvd->vdev_children) { 236fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 237fa9e4066Sahrens pvd->vdev_child = NULL; 238fa9e4066Sahrens pvd->vdev_children = 0; 239fa9e4066Sahrens } 240fa9e4066Sahrens 241fa9e4066Sahrens /* 242fa9e4066Sahrens * Walk up all ancestors to update guid sum. 243fa9e4066Sahrens */ 244fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 245fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 246fa9e4066Sahrens } 247fa9e4066Sahrens 248fa9e4066Sahrens /* 249fa9e4066Sahrens * Remove any holes in the child array. 250fa9e4066Sahrens */ 251fa9e4066Sahrens void 252fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 253fa9e4066Sahrens { 254fa9e4066Sahrens vdev_t **newchild, *cvd; 255fa9e4066Sahrens int oldc = pvd->vdev_children; 256573ca77eSGeorge Wilson int newc; 257fa9e4066Sahrens 258e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 259fa9e4066Sahrens 260573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 261fa9e4066Sahrens if (pvd->vdev_child[c]) 262fa9e4066Sahrens newc++; 263fa9e4066Sahrens 264fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 265fa9e4066Sahrens 266573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 267fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 268fa9e4066Sahrens newchild[newc] = cvd; 269fa9e4066Sahrens cvd->vdev_id = newc++; 270fa9e4066Sahrens } 271fa9e4066Sahrens } 272fa9e4066Sahrens 273fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 274fa9e4066Sahrens pvd->vdev_child = newchild; 275fa9e4066Sahrens pvd->vdev_children = newc; 276fa9e4066Sahrens } 277fa9e4066Sahrens 278fa9e4066Sahrens /* 279fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 280fa9e4066Sahrens */ 28188ecc943SGeorge Wilson vdev_t * 282fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 283fa9e4066Sahrens { 284fa9e4066Sahrens vdev_t *vd; 285fa9e4066Sahrens 286fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 287fa9e4066Sahrens 2880e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2890e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2900e34b6a7Sbonwick spa->spa_root_vdev = vd; 291*e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 2920e34b6a7Sbonwick } 2930e34b6a7Sbonwick 29488ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 2950e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 2960e34b6a7Sbonwick /* 2970e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 2980e34b6a7Sbonwick * which must be unique among all pools. 2990e34b6a7Sbonwick */ 3001195e687SMark J Musante guid = spa_generate_guid(NULL); 3010e34b6a7Sbonwick } else { 3020e34b6a7Sbonwick /* 3030e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3040e34b6a7Sbonwick */ 3051195e687SMark J Musante guid = spa_generate_guid(spa); 3060e34b6a7Sbonwick } 3070e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3080e34b6a7Sbonwick } 3090e34b6a7Sbonwick 310fa9e4066Sahrens vd->vdev_spa = spa; 311fa9e4066Sahrens vd->vdev_id = id; 312fa9e4066Sahrens vd->vdev_guid = guid; 313fa9e4066Sahrens vd->vdev_guid_sum = guid; 314fa9e4066Sahrens vd->vdev_ops = ops; 315fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 31688ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 317fa9e4066Sahrens 318fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3195ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 320e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3218ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3228ad4d6ddSJeff Bonwick space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 3238ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3248ad4d6ddSJeff Bonwick } 325fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 326fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 327fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 328fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 329fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3303d7072f8Seschrock vdev_queue_init(vd); 3313d7072f8Seschrock vdev_cache_init(vd); 332fa9e4066Sahrens 333fa9e4066Sahrens return (vd); 334fa9e4066Sahrens } 335fa9e4066Sahrens 336fa9e4066Sahrens /* 337fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 338fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 339fa9e4066Sahrens * different for each case. 340fa9e4066Sahrens */ 34199653d4eSeschrock int 34299653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 34399653d4eSeschrock int alloctype) 344fa9e4066Sahrens { 345fa9e4066Sahrens vdev_ops_t *ops; 346fa9e4066Sahrens char *type; 3478654d025Sperrin uint64_t guid = 0, islog, nparity; 348fa9e4066Sahrens vdev_t *vd; 349fa9e4066Sahrens 350e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 351fa9e4066Sahrens 352fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 35399653d4eSeschrock return (EINVAL); 354fa9e4066Sahrens 355fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 35699653d4eSeschrock return (EINVAL); 357fa9e4066Sahrens 358fa9e4066Sahrens /* 359fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 360fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 361fa9e4066Sahrens */ 362fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 363fa9e4066Sahrens uint64_t label_id; 364fa9e4066Sahrens 365fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 366fa9e4066Sahrens label_id != id) 36799653d4eSeschrock return (EINVAL); 368fa9e4066Sahrens 369fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37099653d4eSeschrock return (EINVAL); 37199653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 37299653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37399653d4eSeschrock return (EINVAL); 374fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 375fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 376fa94a07fSbrendan return (EINVAL); 37721ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 37821ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37921ecdf64SLin Ling return (EINVAL); 380fa9e4066Sahrens } 381fa9e4066Sahrens 38299653d4eSeschrock /* 38399653d4eSeschrock * The first allocated vdev must be of type 'root'. 38499653d4eSeschrock */ 38599653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 38699653d4eSeschrock return (EINVAL); 38799653d4eSeschrock 3888654d025Sperrin /* 3898654d025Sperrin * Determine whether we're a log vdev. 3908654d025Sperrin */ 3918654d025Sperrin islog = 0; 3928654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 393990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 3948654d025Sperrin return (ENOTSUP); 395fa9e4066Sahrens 39688ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 39788ecc943SGeorge Wilson return (ENOTSUP); 39888ecc943SGeorge Wilson 39999653d4eSeschrock /* 4008654d025Sperrin * Set the nparity property for RAID-Z vdevs. 40199653d4eSeschrock */ 4028654d025Sperrin nparity = -1ULL; 40399653d4eSeschrock if (ops == &vdev_raidz_ops) { 40499653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 4058654d025Sperrin &nparity) == 0) { 406b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 40799653d4eSeschrock return (EINVAL); 40899653d4eSeschrock /* 409f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 410f94275ceSAdam Leventhal * device. 41199653d4eSeschrock */ 412f94275ceSAdam Leventhal if (nparity > 1 && 413f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 414f94275ceSAdam Leventhal return (ENOTSUP); 415f94275ceSAdam Leventhal if (nparity > 2 && 416f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 41799653d4eSeschrock return (ENOTSUP); 41899653d4eSeschrock } else { 41999653d4eSeschrock /* 42099653d4eSeschrock * We require the parity to be specified for SPAs that 42199653d4eSeschrock * support multiple parity levels. 42299653d4eSeschrock */ 423f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 42499653d4eSeschrock return (EINVAL); 42599653d4eSeschrock /* 42699653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 42799653d4eSeschrock */ 4288654d025Sperrin nparity = 1; 42999653d4eSeschrock } 43099653d4eSeschrock } else { 4318654d025Sperrin nparity = 0; 43299653d4eSeschrock } 4338654d025Sperrin ASSERT(nparity != -1ULL); 4348654d025Sperrin 4358654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4368654d025Sperrin 4378654d025Sperrin vd->vdev_islog = islog; 4388654d025Sperrin vd->vdev_nparity = nparity; 4398654d025Sperrin 4408654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4418654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4428654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4438654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4448654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4458654d025Sperrin &vd->vdev_physpath) == 0) 4468654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4476809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4486809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 44999653d4eSeschrock 450afefbcddSeschrock /* 451afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 452afefbcddSeschrock * as -1. 453afefbcddSeschrock */ 454afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 455afefbcddSeschrock &vd->vdev_wholedisk) != 0) 456afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 457afefbcddSeschrock 458ea8dc4b6Seschrock /* 459ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 460ea8dc4b6Seschrock * was not present at the time of import. 461ea8dc4b6Seschrock */ 4626809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4636809eb4eSEric Schrock &vd->vdev_not_present); 464ea8dc4b6Seschrock 465ecc2d604Sbonwick /* 466ecc2d604Sbonwick * Get the alignment requirement. 467ecc2d604Sbonwick */ 468ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 469ecc2d604Sbonwick 47088ecc943SGeorge Wilson /* 47188ecc943SGeorge Wilson * Retrieve the vdev creation time. 47288ecc943SGeorge Wilson */ 47388ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 47488ecc943SGeorge Wilson &vd->vdev_crtxg); 47588ecc943SGeorge Wilson 476fa9e4066Sahrens /* 477fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 478fa9e4066Sahrens */ 4791195e687SMark J Musante if (parent && !parent->vdev_parent && 4801195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 481fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 482fa9e4066Sahrens &vd->vdev_ms_array); 483fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 484fa9e4066Sahrens &vd->vdev_ms_shift); 485fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 486fa9e4066Sahrens &vd->vdev_asize); 4873f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 4883f9d6ad7SLin Ling &vd->vdev_removing); 489fa9e4066Sahrens } 490fa9e4066Sahrens 491a1521560SJeff Bonwick if (parent && !parent->vdev_parent) { 492a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 4939f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 4941195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 4959f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 496a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 497a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 498a1521560SJeff Bonwick } 499a1521560SJeff Bonwick 500fa9e4066Sahrens /* 5013d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 502fa9e4066Sahrens */ 503c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 50421ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 50521ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 506c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 507c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 5088ad4d6ddSJeff Bonwick &vd->vdev_dtl_smo.smo_object); 509c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 510c5904d13Seschrock &vd->vdev_unspare); 511c5904d13Seschrock } 51221ecdf64SLin Ling 51321ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 51421ecdf64SLin Ling uint64_t spare = 0; 51521ecdf64SLin Ling 51621ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 51721ecdf64SLin Ling &spare) == 0 && spare) 51821ecdf64SLin Ling spa_spare_add(vd); 51921ecdf64SLin Ling } 52021ecdf64SLin Ling 521ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 522ecc2d604Sbonwick &vd->vdev_offline); 523c5904d13Seschrock 524cb04b873SMark J Musante (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, 525cb04b873SMark J Musante &vd->vdev_resilvering); 526cb04b873SMark J Musante 5273d7072f8Seschrock /* 5283d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 5293d7072f8Seschrock * state, as the diagnosis made on another system may not be 530069f55e2SEric Schrock * valid in the current context. Local vdevs will 531069f55e2SEric Schrock * remain in the faulted state. 5323d7072f8Seschrock */ 533b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 5343d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 5353d7072f8Seschrock &vd->vdev_faulted); 5363d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 5373d7072f8Seschrock &vd->vdev_degraded); 5383d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 5393d7072f8Seschrock &vd->vdev_removed); 540069f55e2SEric Schrock 541069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 542069f55e2SEric Schrock char *aux; 543069f55e2SEric Schrock 544069f55e2SEric Schrock vd->vdev_label_aux = 545069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 546069f55e2SEric Schrock if (nvlist_lookup_string(nv, 547069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 548069f55e2SEric Schrock strcmp(aux, "external") == 0) 549069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 550069f55e2SEric Schrock } 5513d7072f8Seschrock } 552fa9e4066Sahrens } 553fa9e4066Sahrens 554fa9e4066Sahrens /* 555fa9e4066Sahrens * Add ourselves to the parent's list of children. 556fa9e4066Sahrens */ 557fa9e4066Sahrens vdev_add_child(parent, vd); 558fa9e4066Sahrens 55999653d4eSeschrock *vdp = vd; 56099653d4eSeschrock 56199653d4eSeschrock return (0); 562fa9e4066Sahrens } 563fa9e4066Sahrens 564fa9e4066Sahrens void 565fa9e4066Sahrens vdev_free(vdev_t *vd) 566fa9e4066Sahrens { 5673d7072f8Seschrock spa_t *spa = vd->vdev_spa; 568fa9e4066Sahrens 569fa9e4066Sahrens /* 570fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 571fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 572fa9e4066Sahrens */ 573fa9e4066Sahrens vdev_close(vd); 574fa9e4066Sahrens 575e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 576b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 577fa9e4066Sahrens 578fa9e4066Sahrens /* 579fa9e4066Sahrens * Free all children. 580fa9e4066Sahrens */ 581573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 582fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 583fa9e4066Sahrens 584fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 585fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 586fa9e4066Sahrens 587fa9e4066Sahrens /* 588fa9e4066Sahrens * Discard allocation state. 589fa9e4066Sahrens */ 590a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 591fa9e4066Sahrens vdev_metaslab_fini(vd); 592a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 593a1521560SJeff Bonwick } 594fa9e4066Sahrens 595fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 59699653d4eSeschrock ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 597fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 598fa9e4066Sahrens 599fa9e4066Sahrens /* 600fa9e4066Sahrens * Remove this vdev from its parent's child list. 601fa9e4066Sahrens */ 602fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 603fa9e4066Sahrens 604fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 605fa9e4066Sahrens 6063d7072f8Seschrock /* 6073d7072f8Seschrock * Clean up vdev structure. 6083d7072f8Seschrock */ 6093d7072f8Seschrock vdev_queue_fini(vd); 6103d7072f8Seschrock vdev_cache_fini(vd); 6113d7072f8Seschrock 6123d7072f8Seschrock if (vd->vdev_path) 6133d7072f8Seschrock spa_strfree(vd->vdev_path); 6143d7072f8Seschrock if (vd->vdev_devid) 6153d7072f8Seschrock spa_strfree(vd->vdev_devid); 6163d7072f8Seschrock if (vd->vdev_physpath) 6173d7072f8Seschrock spa_strfree(vd->vdev_physpath); 6186809eb4eSEric Schrock if (vd->vdev_fru) 6196809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 6203d7072f8Seschrock 6213d7072f8Seschrock if (vd->vdev_isspare) 6223d7072f8Seschrock spa_spare_remove(vd); 623fa94a07fSbrendan if (vd->vdev_isl2cache) 624fa94a07fSbrendan spa_l2cache_remove(vd); 6253d7072f8Seschrock 6263d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 6273d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 6288ad4d6ddSJeff Bonwick 6293d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 6308ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 6318ad4d6ddSJeff Bonwick space_map_unload(&vd->vdev_dtl[t]); 6328ad4d6ddSJeff Bonwick space_map_destroy(&vd->vdev_dtl[t]); 6338ad4d6ddSJeff Bonwick } 6343d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 6358ad4d6ddSJeff Bonwick 6363d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 6373d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 638e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 6393d7072f8Seschrock 6403d7072f8Seschrock if (vd == spa->spa_root_vdev) 6413d7072f8Seschrock spa->spa_root_vdev = NULL; 6423d7072f8Seschrock 6433d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 644fa9e4066Sahrens } 645fa9e4066Sahrens 646fa9e4066Sahrens /* 647fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 648fa9e4066Sahrens */ 649fa9e4066Sahrens static void 650fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 651fa9e4066Sahrens { 652fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 653fa9e4066Sahrens metaslab_t *msp; 654fa9e4066Sahrens vdev_t *vd; 655fa9e4066Sahrens int t; 656fa9e4066Sahrens 657fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 658fa9e4066Sahrens 659fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 660fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 661fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 662fa9e4066Sahrens 663fa9e4066Sahrens svd->vdev_ms_array = 0; 664fa9e4066Sahrens svd->vdev_ms_shift = 0; 665fa9e4066Sahrens svd->vdev_ms_count = 0; 666fa9e4066Sahrens 667fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 668fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 669fa9e4066Sahrens 670fa9e4066Sahrens svd->vdev_mg = NULL; 671fa9e4066Sahrens svd->vdev_ms = NULL; 672ecc2d604Sbonwick 673ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 674ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 675fa9e4066Sahrens 676fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 677fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 67899653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 679fa9e4066Sahrens 680fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 681fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 68299653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 683fa9e4066Sahrens 684fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 685fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 686fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 687fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 688fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 689fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 690fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 691fa9e4066Sahrens } 692fa9e4066Sahrens 693e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 694fa9e4066Sahrens vdev_config_clean(svd); 695fa9e4066Sahrens vdev_config_dirty(tvd); 696fa9e4066Sahrens } 697fa9e4066Sahrens 698e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 699e14bb325SJeff Bonwick vdev_state_clean(svd); 700e14bb325SJeff Bonwick vdev_state_dirty(tvd); 701e14bb325SJeff Bonwick } 702e14bb325SJeff Bonwick 70399653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 70499653d4eSeschrock svd->vdev_deflate_ratio = 0; 7058654d025Sperrin 7068654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 7078654d025Sperrin svd->vdev_islog = 0; 708fa9e4066Sahrens } 709fa9e4066Sahrens 710fa9e4066Sahrens static void 711fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 712fa9e4066Sahrens { 713fa9e4066Sahrens if (vd == NULL) 714fa9e4066Sahrens return; 715fa9e4066Sahrens 716fa9e4066Sahrens vd->vdev_top = tvd; 717fa9e4066Sahrens 718573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 719fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 720fa9e4066Sahrens } 721fa9e4066Sahrens 722fa9e4066Sahrens /* 723fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 724fa9e4066Sahrens */ 725fa9e4066Sahrens vdev_t * 726fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 727fa9e4066Sahrens { 728fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 729fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 730fa9e4066Sahrens vdev_t *mvd; 731fa9e4066Sahrens 732e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 733fa9e4066Sahrens 734fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 735ecc2d604Sbonwick 736ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 737573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 738ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 739ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 74088ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 741ecc2d604Sbonwick 742fa9e4066Sahrens vdev_remove_child(pvd, cvd); 743fa9e4066Sahrens vdev_add_child(pvd, mvd); 744fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 745fa9e4066Sahrens vdev_add_child(mvd, cvd); 746fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 747fa9e4066Sahrens 748fa9e4066Sahrens if (mvd == mvd->vdev_top) 749fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 750fa9e4066Sahrens 751fa9e4066Sahrens return (mvd); 752fa9e4066Sahrens } 753fa9e4066Sahrens 754fa9e4066Sahrens /* 755fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 756fa9e4066Sahrens */ 757fa9e4066Sahrens void 758fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 759fa9e4066Sahrens { 760fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 761fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 762fa9e4066Sahrens 763e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 764fa9e4066Sahrens 765fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 766fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 76799653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 76899653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 769ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 770fa9e4066Sahrens 771fa9e4066Sahrens vdev_remove_child(mvd, cvd); 772fa9e4066Sahrens vdev_remove_child(pvd, mvd); 7738ad4d6ddSJeff Bonwick 77499653d4eSeschrock /* 775e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 776e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 777e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 778e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 77999653d4eSeschrock */ 7808ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 7818ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 7821195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 7838ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 7848ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 7858ad4d6ddSJeff Bonwick } 786e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 787e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 788fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 789fa9e4066Sahrens 790fa9e4066Sahrens if (cvd == cvd->vdev_top) 791fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 792fa9e4066Sahrens 793fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 794fa9e4066Sahrens vdev_free(mvd); 795fa9e4066Sahrens } 796fa9e4066Sahrens 797ea8dc4b6Seschrock int 798fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 799fa9e4066Sahrens { 800fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 801ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 802ecc2d604Sbonwick uint64_t m; 803fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 804fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 805ecc2d604Sbonwick metaslab_t **mspp; 806ecc2d604Sbonwick int error; 807fa9e4066Sahrens 808a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 809a1521560SJeff Bonwick 81088ecc943SGeorge Wilson /* 81188ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 81288ecc943SGeorge Wilson */ 81388ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 8140e34b6a7Sbonwick return (0); 8150e34b6a7Sbonwick 81688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 81788ecc943SGeorge Wilson 818e6ca193dSGeorge Wilson /* 819e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 820e6ca193dSGeorge Wilson * in 128k (1 << 17) because it is the current "typical" blocksize. 821e6ca193dSGeorge Wilson * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 822e6ca193dSGeorge Wilson * or we will inconsistently account for existing bp's. 823e6ca193dSGeorge Wilson */ 824e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 825e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 826e6ca193dSGeorge Wilson 827fa9e4066Sahrens ASSERT(oldc <= newc); 828fa9e4066Sahrens 829ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 830fa9e4066Sahrens 831ecc2d604Sbonwick if (oldc != 0) { 832ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 833ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 834ecc2d604Sbonwick } 835fa9e4066Sahrens 836ecc2d604Sbonwick vd->vdev_ms = mspp; 837ecc2d604Sbonwick vd->vdev_ms_count = newc; 838fa9e4066Sahrens 839ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 840ecc2d604Sbonwick space_map_obj_t smo = { 0, 0, 0 }; 841ecc2d604Sbonwick if (txg == 0) { 842ecc2d604Sbonwick uint64_t object = 0; 843ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 8447bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 8457bfdf011SNeil Perrin DMU_READ_PREFETCH); 846ecc2d604Sbonwick if (error) 847ecc2d604Sbonwick return (error); 848ecc2d604Sbonwick if (object != 0) { 849ecc2d604Sbonwick dmu_buf_t *db; 850ecc2d604Sbonwick error = dmu_bonus_hold(mos, object, FTAG, &db); 851ecc2d604Sbonwick if (error) 852ecc2d604Sbonwick return (error); 8531934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (smo)); 8541934e92fSmaybee bcopy(db->db_data, &smo, sizeof (smo)); 855ecc2d604Sbonwick ASSERT3U(smo.smo_object, ==, object); 856ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 857fa9e4066Sahrens } 858fa9e4066Sahrens } 859ecc2d604Sbonwick vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 860ecc2d604Sbonwick m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 861fa9e4066Sahrens } 862fa9e4066Sahrens 863a1521560SJeff Bonwick if (txg == 0) 864a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 865a1521560SJeff Bonwick 8663f9d6ad7SLin Ling /* 8673f9d6ad7SLin Ling * If the vdev is being removed we don't activate 8683f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 8693f9d6ad7SLin Ling * allocations are performed on this device. 8703f9d6ad7SLin Ling */ 8713f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 872a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 873a1521560SJeff Bonwick 874a1521560SJeff Bonwick if (txg == 0) 875a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 876a1521560SJeff Bonwick 877ea8dc4b6Seschrock return (0); 878fa9e4066Sahrens } 879fa9e4066Sahrens 880fa9e4066Sahrens void 881fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 882fa9e4066Sahrens { 883fa9e4066Sahrens uint64_t m; 884fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 885fa9e4066Sahrens 886fa9e4066Sahrens if (vd->vdev_ms != NULL) { 887a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 888fa9e4066Sahrens for (m = 0; m < count; m++) 889ecc2d604Sbonwick if (vd->vdev_ms[m] != NULL) 890ecc2d604Sbonwick metaslab_fini(vd->vdev_ms[m]); 891fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 892fa9e4066Sahrens vd->vdev_ms = NULL; 893fa9e4066Sahrens } 894fa9e4066Sahrens } 895fa9e4066Sahrens 896e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 897e14bb325SJeff Bonwick boolean_t vps_readable; 898e14bb325SJeff Bonwick boolean_t vps_writeable; 899e14bb325SJeff Bonwick int vps_flags; 900e14bb325SJeff Bonwick } vdev_probe_stats_t; 901e14bb325SJeff Bonwick 902e14bb325SJeff Bonwick static void 903e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 9040a4e9518Sgw { 9058ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 906a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 907e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 908a3f829aeSBill Moore 909a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 910e14bb325SJeff Bonwick 911e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 912e14bb325SJeff Bonwick if (zio->io_error == 0) 913e14bb325SJeff Bonwick vps->vps_readable = 1; 9148ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 915a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 916e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 917e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 918e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 919e14bb325SJeff Bonwick } else { 920e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 921e14bb325SJeff Bonwick } 922e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 923e14bb325SJeff Bonwick if (zio->io_error == 0) 924e14bb325SJeff Bonwick vps->vps_writeable = 1; 925e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 926e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 927a3f829aeSBill Moore zio_t *pio; 928e14bb325SJeff Bonwick 929e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 930e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 931e14bb325SJeff Bonwick 932e14bb325SJeff Bonwick if (vdev_readable(vd) && 9338ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 934e14bb325SJeff Bonwick zio->io_error = 0; 935e14bb325SJeff Bonwick } else { 936e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 937e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 9388ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 939e14bb325SJeff Bonwick zio->io_error = ENXIO; 940e14bb325SJeff Bonwick } 941a3f829aeSBill Moore 942a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 943a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 944a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 945a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 946a3f829aeSBill Moore 947a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 948a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 949a3f829aeSBill Moore pio->io_error = ENXIO; 950a3f829aeSBill Moore 951e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 952e14bb325SJeff Bonwick } 953e14bb325SJeff Bonwick } 9540a4e9518Sgw 955e14bb325SJeff Bonwick /* 956e14bb325SJeff Bonwick * Determine whether this device is accessible by reading and writing 957e14bb325SJeff Bonwick * to several known locations: the pad regions of each vdev label 958e14bb325SJeff Bonwick * but the first (which we leave alone in case it contains a VTOC). 959e14bb325SJeff Bonwick */ 960e14bb325SJeff Bonwick zio_t * 961a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 962e14bb325SJeff Bonwick { 963e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 964a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 965a3f829aeSBill Moore zio_t *pio; 966a3f829aeSBill Moore 967a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9680a4e9518Sgw 969a3f829aeSBill Moore /* 970a3f829aeSBill Moore * Don't probe the probe. 971a3f829aeSBill Moore */ 972a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 973a3f829aeSBill Moore return (NULL); 974e14bb325SJeff Bonwick 975a3f829aeSBill Moore /* 976a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 977a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 978a3f829aeSBill Moore * this vdev will become parents of the probe io. 979a3f829aeSBill Moore */ 980a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 981e14bb325SJeff Bonwick 982a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 983a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 984a3f829aeSBill Moore 985a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 986a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 9878956713aSEric Schrock ZIO_FLAG_TRYHARD; 988a3f829aeSBill Moore 989a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 990a3f829aeSBill Moore /* 991a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 992a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 993a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 994a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 995a3f829aeSBill Moore * any zio looking at these values can assume that 996a3f829aeSBill Moore * failures persist for the life of the I/O. That's 997a3f829aeSBill Moore * important because when a device has intermittent 998a3f829aeSBill Moore * connectivity problems, we want to ensure that 999a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1000a3f829aeSBill Moore * the zio (EIO). 1001a3f829aeSBill Moore * 1002a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1003a3f829aeSBill Moore * values so the probe can reevaluate from first 1004a3f829aeSBill Moore * principles. 1005a3f829aeSBill Moore */ 1006a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1007a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1008a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1009a3f829aeSBill Moore } 1010a3f829aeSBill Moore 1011a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1012a3f829aeSBill Moore vdev_probe_done, vps, 1013a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1014a3f829aeSBill Moore 101598d1cbfeSGeorge Wilson /* 101698d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 101798d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 101898d1cbfeSGeorge Wilson */ 1019a3f829aeSBill Moore if (zio != NULL) { 1020a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1021a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1022a3f829aeSBill Moore } 1023e14bb325SJeff Bonwick } 1024e14bb325SJeff Bonwick 1025a3f829aeSBill Moore if (zio != NULL) 1026a3f829aeSBill Moore zio_add_child(zio, pio); 1027e14bb325SJeff Bonwick 1028a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1029e14bb325SJeff Bonwick 1030a3f829aeSBill Moore if (vps == NULL) { 1031a3f829aeSBill Moore ASSERT(zio != NULL); 1032a3f829aeSBill Moore return (NULL); 1033a3f829aeSBill Moore } 1034e14bb325SJeff Bonwick 1035e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1036a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1037e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1038f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 1039f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1040e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1041e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1042e14bb325SJeff Bonwick } 1043e14bb325SJeff Bonwick 1044a3f829aeSBill Moore if (zio == NULL) 1045a3f829aeSBill Moore return (pio); 1046a3f829aeSBill Moore 1047a3f829aeSBill Moore zio_nowait(pio); 1048a3f829aeSBill Moore return (NULL); 10490a4e9518Sgw } 10500a4e9518Sgw 1051f64c0e34SEric Taylor static void 1052f64c0e34SEric Taylor vdev_open_child(void *arg) 1053f64c0e34SEric Taylor { 1054f64c0e34SEric Taylor vdev_t *vd = arg; 1055f64c0e34SEric Taylor 1056f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1057f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1058f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1059f64c0e34SEric Taylor } 1060f64c0e34SEric Taylor 1061681d9761SEric Taylor boolean_t 1062681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1063681d9761SEric Taylor { 1064681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1065681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1066681d9761SEric Taylor return (B_TRUE); 1067681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1068681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1069681d9761SEric Taylor return (B_TRUE); 1070681d9761SEric Taylor return (B_FALSE); 1071681d9761SEric Taylor } 1072681d9761SEric Taylor 1073f64c0e34SEric Taylor void 1074f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1075f64c0e34SEric Taylor { 1076f64c0e34SEric Taylor taskq_t *tq; 1077f64c0e34SEric Taylor int children = vd->vdev_children; 1078f64c0e34SEric Taylor 1079681d9761SEric Taylor /* 1080681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1081681d9761SEric Taylor * in a single thread so that the same thread holds the 1082681d9761SEric Taylor * spa_namespace_lock 1083681d9761SEric Taylor */ 1084681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1085681d9761SEric Taylor for (int c = 0; c < children; c++) 1086681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1087681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1088681d9761SEric Taylor return; 1089681d9761SEric Taylor } 1090f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1091f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1092f64c0e34SEric Taylor 1093f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1094f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1095f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1096f64c0e34SEric Taylor 1097f64c0e34SEric Taylor taskq_destroy(tq); 1098f64c0e34SEric Taylor } 1099f64c0e34SEric Taylor 1100fa9e4066Sahrens /* 1101fa9e4066Sahrens * Prepare a virtual device for access. 1102fa9e4066Sahrens */ 1103fa9e4066Sahrens int 1104fa9e4066Sahrens vdev_open(vdev_t *vd) 1105fa9e4066Sahrens { 11068ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1107fa9e4066Sahrens int error; 1108fa9e4066Sahrens uint64_t osize = 0; 1109fa9e4066Sahrens uint64_t asize, psize; 1110ecc2d604Sbonwick uint64_t ashift = 0; 1111fa9e4066Sahrens 1112f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1113f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1114fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1115fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1116fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1117fa9e4066Sahrens 1118fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1119e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1120e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1121573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1122fa9e4066Sahrens 1123069f55e2SEric Schrock /* 1124069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1125069f55e2SEric Schrock * faulted, bail out of the open. 1126069f55e2SEric Schrock */ 11273d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 11283d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1129069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1130069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 11313d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1132069f55e2SEric Schrock vd->vdev_label_aux); 11333d7072f8Seschrock return (ENXIO); 11343d7072f8Seschrock } else if (vd->vdev_offline) { 1135fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1136ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1137fa9e4066Sahrens return (ENXIO); 1138fa9e4066Sahrens } 1139fa9e4066Sahrens 1140fa9e4066Sahrens error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 1141fa9e4066Sahrens 1142095bcd66SGeorge Wilson /* 1143095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1144095bcd66SGeorge Wilson * the vdev on error. 1145095bcd66SGeorge Wilson */ 1146095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1147ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 11488956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1149ea8dc4b6Seschrock 1150fa9e4066Sahrens if (error) { 11513d7072f8Seschrock if (vd->vdev_removed && 11523d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 11533d7072f8Seschrock vd->vdev_removed = B_FALSE; 11543d7072f8Seschrock 1155ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1156fa9e4066Sahrens vd->vdev_stat.vs_aux); 1157fa9e4066Sahrens return (error); 1158fa9e4066Sahrens } 1159fa9e4066Sahrens 11603d7072f8Seschrock vd->vdev_removed = B_FALSE; 11613d7072f8Seschrock 1162096d22d4SEric Schrock /* 1163096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1164096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1165096d22d4SEric Schrock */ 1166096d22d4SEric Schrock if (vd->vdev_faulted) { 1167096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1168096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1169096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1170096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1171096d22d4SEric Schrock vd->vdev_label_aux); 1172096d22d4SEric Schrock return (ENXIO); 1173096d22d4SEric Schrock } 1174096d22d4SEric Schrock 11753d7072f8Seschrock if (vd->vdev_degraded) { 11763d7072f8Seschrock ASSERT(vd->vdev_children == 0); 11773d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 11783d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 11793d7072f8Seschrock } else { 1180069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 11813d7072f8Seschrock } 1182fa9e4066Sahrens 118388ecc943SGeorge Wilson /* 118488ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 118588ecc943SGeorge Wilson */ 118688ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 118788ecc943SGeorge Wilson return (0); 118888ecc943SGeorge Wilson 1189573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1190ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1191ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1192ea8dc4b6Seschrock VDEV_AUX_NONE); 1193ea8dc4b6Seschrock break; 1194ea8dc4b6Seschrock } 1195573ca77eSGeorge Wilson } 1196fa9e4066Sahrens 1197fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1198fa9e4066Sahrens 1199fa9e4066Sahrens if (vd->vdev_children == 0) { 1200fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1201ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1202ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1203fa9e4066Sahrens return (EOVERFLOW); 1204fa9e4066Sahrens } 1205fa9e4066Sahrens psize = osize; 1206fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1207fa9e4066Sahrens } else { 1208ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1209fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1210ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1211ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1212fa9e4066Sahrens return (EOVERFLOW); 1213fa9e4066Sahrens } 1214fa9e4066Sahrens psize = 0; 1215fa9e4066Sahrens asize = osize; 1216fa9e4066Sahrens } 1217fa9e4066Sahrens 1218fa9e4066Sahrens vd->vdev_psize = psize; 1219fa9e4066Sahrens 1220573ca77eSGeorge Wilson /* 1221573ca77eSGeorge Wilson * Make sure the allocatable size hasn't shrunk. 1222573ca77eSGeorge Wilson */ 1223573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1224573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1225573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1226573ca77eSGeorge Wilson return (EINVAL); 1227573ca77eSGeorge Wilson } 1228573ca77eSGeorge Wilson 1229fa9e4066Sahrens if (vd->vdev_asize == 0) { 1230fa9e4066Sahrens /* 1231fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1232ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1233fa9e4066Sahrens */ 1234fa9e4066Sahrens vd->vdev_asize = asize; 1235ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1236fa9e4066Sahrens } else { 1237fa9e4066Sahrens /* 1238fa9e4066Sahrens * Make sure the alignment requirement hasn't increased. 1239fa9e4066Sahrens */ 1240ecc2d604Sbonwick if (ashift > vd->vdev_top->vdev_ashift) { 1241ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1242ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 1243fa9e4066Sahrens return (EINVAL); 1244fa9e4066Sahrens } 1245573ca77eSGeorge Wilson } 1246fa9e4066Sahrens 1247573ca77eSGeorge Wilson /* 1248573ca77eSGeorge Wilson * If all children are healthy and the asize has increased, 1249573ca77eSGeorge Wilson * then we've experienced dynamic LUN growth. If automatic 1250573ca77eSGeorge Wilson * expansion is enabled then use the additional space. 1251573ca77eSGeorge Wilson */ 1252573ca77eSGeorge Wilson if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1253573ca77eSGeorge Wilson (vd->vdev_expanding || spa->spa_autoexpand)) 1254573ca77eSGeorge Wilson vd->vdev_asize = asize; 1255fa9e4066Sahrens 1256573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1257fa9e4066Sahrens 12580a4e9518Sgw /* 12590a4e9518Sgw * Ensure we can issue some IO before declaring the 12600a4e9518Sgw * vdev open for business. 12610a4e9518Sgw */ 1262e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1263e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 126498d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 126598d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 12660a4e9518Sgw return (error); 12670a4e9518Sgw } 12680a4e9518Sgw 1269088f3894Sahrens /* 1270088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 12718ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 12728ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1273088f3894Sahrens */ 12748ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 12758ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 12768ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1277088f3894Sahrens 1278fa9e4066Sahrens return (0); 1279fa9e4066Sahrens } 1280fa9e4066Sahrens 1281560e6e96Seschrock /* 1282560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1283560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 12843d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1285560e6e96Seschrock * 1286560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1287560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1288560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1289560e6e96Seschrock * will be updated but the function will return 0. 1290560e6e96Seschrock */ 1291560e6e96Seschrock int 1292560e6e96Seschrock vdev_validate(vdev_t *vd) 1293560e6e96Seschrock { 1294560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1295560e6e96Seschrock nvlist_t *label; 12961195e687SMark J Musante uint64_t guid = 0, top_guid; 1297560e6e96Seschrock uint64_t state; 1298560e6e96Seschrock 1299573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1300560e6e96Seschrock if (vdev_validate(vd->vdev_child[c]) != 0) 13010bf246f5Smc return (EBADF); 1302560e6e96Seschrock 1303b5989ec7Seschrock /* 1304b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1305b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1306b5989ec7Seschrock * overwrite the previous state. 1307b5989ec7Seschrock */ 1308e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 13091195e687SMark J Musante uint64_t aux_guid = 0; 13101195e687SMark J Musante nvlist_t *nvl; 1311560e6e96Seschrock 1312560e6e96Seschrock if ((label = vdev_label_read_config(vd)) == NULL) { 1313560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1314560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1315560e6e96Seschrock return (0); 1316560e6e96Seschrock } 1317560e6e96Seschrock 13181195e687SMark J Musante /* 13191195e687SMark J Musante * Determine if this vdev has been split off into another 13201195e687SMark J Musante * pool. If so, then refuse to open it. 13211195e687SMark J Musante */ 13221195e687SMark J Musante if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 13231195e687SMark J Musante &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 13241195e687SMark J Musante vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 13251195e687SMark J Musante VDEV_AUX_SPLIT_POOL); 13261195e687SMark J Musante nvlist_free(label); 13271195e687SMark J Musante return (0); 13281195e687SMark J Musante } 13291195e687SMark J Musante 1330560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1331560e6e96Seschrock &guid) != 0 || guid != spa_guid(spa)) { 1332560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1333560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1334560e6e96Seschrock nvlist_free(label); 1335560e6e96Seschrock return (0); 1336560e6e96Seschrock } 1337560e6e96Seschrock 13381195e687SMark J Musante if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 13391195e687SMark J Musante != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 13401195e687SMark J Musante &aux_guid) != 0) 13411195e687SMark J Musante aux_guid = 0; 13421195e687SMark J Musante 1343e14bb325SJeff Bonwick /* 1344e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1345e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1346e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1347e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1348e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1349e14bb325SJeff Bonwick * safely compare to that instead. 13501195e687SMark J Musante * 13511195e687SMark J Musante * If we split this vdev off instead, then we also check the 13521195e687SMark J Musante * original pool's guid. We don't want to consider the vdev 13531195e687SMark J Musante * corrupt if it is partway through a split operation. 1354e14bb325SJeff Bonwick */ 1355560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1356e14bb325SJeff Bonwick &guid) != 0 || 1357e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1358e14bb325SJeff Bonwick &top_guid) != 0 || 13591195e687SMark J Musante ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1360e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1361560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1362560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1363560e6e96Seschrock nvlist_free(label); 1364560e6e96Seschrock return (0); 1365560e6e96Seschrock } 1366560e6e96Seschrock 1367560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1368560e6e96Seschrock &state) != 0) { 1369560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1370560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1371560e6e96Seschrock nvlist_free(label); 1372560e6e96Seschrock return (0); 1373560e6e96Seschrock } 1374560e6e96Seschrock 1375560e6e96Seschrock nvlist_free(label); 1376560e6e96Seschrock 1377bc758434SLin Ling /* 13784b964adaSGeorge Wilson * If this is a verbatim import, no need to check the 1379bc758434SLin Ling * state of the pool. 1380bc758434SLin Ling */ 13814b964adaSGeorge Wilson if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1382b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_OPEN && 1383bc758434SLin Ling state != POOL_STATE_ACTIVE) 13840bf246f5Smc return (EBADF); 1385560e6e96Seschrock 138651ece835Seschrock /* 138751ece835Seschrock * If we were able to open and validate a vdev that was 138851ece835Seschrock * previously marked permanently unavailable, clear that state 138951ece835Seschrock * now. 139051ece835Seschrock */ 139151ece835Seschrock if (vd->vdev_not_present) 139251ece835Seschrock vd->vdev_not_present = 0; 139351ece835Seschrock } 1394560e6e96Seschrock 1395560e6e96Seschrock return (0); 1396560e6e96Seschrock } 1397560e6e96Seschrock 1398fa9e4066Sahrens /* 1399fa9e4066Sahrens * Close a virtual device. 1400fa9e4066Sahrens */ 1401fa9e4066Sahrens void 1402fa9e4066Sahrens vdev_close(vdev_t *vd) 1403fa9e4066Sahrens { 14048ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1405095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 14068ad4d6ddSJeff Bonwick 14078ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 14088ad4d6ddSJeff Bonwick 14091195e687SMark J Musante /* 14101195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 14111195e687SMark J Musante * going offline. 14121195e687SMark J Musante */ 1413095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 14141195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1415095bcd66SGeorge Wilson 1416fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1417fa9e4066Sahrens 14183d7072f8Seschrock vdev_cache_purge(vd); 1419fa9e4066Sahrens 1420560e6e96Seschrock /* 1421573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1422560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1423560e6e96Seschrock * it's still faulted. 1424560e6e96Seschrock */ 1425560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1426560e6e96Seschrock 1427fa9e4066Sahrens if (vd->vdev_offline) 1428fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1429fa9e4066Sahrens else 1430fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1431ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1432fa9e4066Sahrens } 1433fa9e4066Sahrens 1434dcba9f3fSGeorge Wilson void 1435dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1436dcba9f3fSGeorge Wilson { 1437dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1438dcba9f3fSGeorge Wilson 1439dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1440dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1441dcba9f3fSGeorge Wilson return; 1442dcba9f3fSGeorge Wilson 1443dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1444dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1445dcba9f3fSGeorge Wilson 1446dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1447dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1448dcba9f3fSGeorge Wilson } 1449dcba9f3fSGeorge Wilson 1450dcba9f3fSGeorge Wilson void 1451dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1452dcba9f3fSGeorge Wilson { 1453dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1454dcba9f3fSGeorge Wilson 1455dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1456dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1457dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1458dcba9f3fSGeorge Wilson 1459dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1460dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1461dcba9f3fSGeorge Wilson } 1462dcba9f3fSGeorge Wilson 1463095bcd66SGeorge Wilson /* 1464095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1465095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1466095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1467095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1468095bcd66SGeorge Wilson */ 1469fa9e4066Sahrens void 1470ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1471fa9e4066Sahrens { 1472ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1473fa9e4066Sahrens 1474e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1475ea8dc4b6Seschrock 14761195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 14771195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1478fa9e4066Sahrens vdev_close(vd); 1479fa9e4066Sahrens (void) vdev_open(vd); 1480fa9e4066Sahrens 148139c23413Seschrock /* 148239c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 148339c23413Seschrock * Otherwise, a device with an invalid label could be successfully 148439c23413Seschrock * opened in response to vdev_reopen(). 148539c23413Seschrock */ 1486c5904d13Seschrock if (vd->vdev_aux) { 1487c5904d13Seschrock (void) vdev_validate_aux(vd); 1488e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 14896809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1490573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1491573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1492c5904d13Seschrock } else { 1493c5904d13Seschrock (void) vdev_validate(vd); 1494c5904d13Seschrock } 149539c23413Seschrock 1496fa9e4066Sahrens /* 14973d7072f8Seschrock * Reassess parent vdev's health. 1498fa9e4066Sahrens */ 14993d7072f8Seschrock vdev_propagate_state(vd); 1500fa9e4066Sahrens } 1501fa9e4066Sahrens 1502fa9e4066Sahrens int 150399653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1504fa9e4066Sahrens { 1505fa9e4066Sahrens int error; 1506fa9e4066Sahrens 1507fa9e4066Sahrens /* 1508fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1509fa9e4066Sahrens * For a create, however, we want to fail the request if 1510fa9e4066Sahrens * there are any components we can't open. 1511fa9e4066Sahrens */ 1512fa9e4066Sahrens error = vdev_open(vd); 1513fa9e4066Sahrens 1514fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1515fa9e4066Sahrens vdev_close(vd); 1516fa9e4066Sahrens return (error ? error : ENXIO); 1517fa9e4066Sahrens } 1518fa9e4066Sahrens 1519fa9e4066Sahrens /* 1520fa9e4066Sahrens * Recursively initialize all labels. 1521fa9e4066Sahrens */ 152239c23413Seschrock if ((error = vdev_label_init(vd, txg, isreplacing ? 152339c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1524fa9e4066Sahrens vdev_close(vd); 1525fa9e4066Sahrens return (error); 1526fa9e4066Sahrens } 1527fa9e4066Sahrens 1528fa9e4066Sahrens return (0); 1529fa9e4066Sahrens } 1530fa9e4066Sahrens 15310e34b6a7Sbonwick void 1532573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1533fa9e4066Sahrens { 1534fa9e4066Sahrens /* 1535fa9e4066Sahrens * Aim for roughly 200 metaslabs per vdev. 1536fa9e4066Sahrens */ 1537fa9e4066Sahrens vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1538fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1539fa9e4066Sahrens } 1540fa9e4066Sahrens 1541fa9e4066Sahrens void 1542ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1543fa9e4066Sahrens { 1544ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 154588ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 1546ecc2d604Sbonwick ASSERT(ISP2(flags)); 1547f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1548fa9e4066Sahrens 1549ecc2d604Sbonwick if (flags & VDD_METASLAB) 1550ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1551ecc2d604Sbonwick 1552ecc2d604Sbonwick if (flags & VDD_DTL) 1553ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1554ecc2d604Sbonwick 1555ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1556fa9e4066Sahrens } 1557fa9e4066Sahrens 15588ad4d6ddSJeff Bonwick /* 15598ad4d6ddSJeff Bonwick * DTLs. 15608ad4d6ddSJeff Bonwick * 15618ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 15629fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 15638ad4d6ddSJeff Bonwick * 15648ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 15658ad4d6ddSJeff Bonwick * 15668ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 15678ad4d6ddSJeff Bonwick * 15688ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 15698ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 15708ad4d6ddSJeff Bonwick * txgs that was scrubbed. 15718ad4d6ddSJeff Bonwick * 15728ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 15738ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 15748ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 15758ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 15768ad4d6ddSJeff Bonwick * determine whether a device can be detached. 15778ad4d6ddSJeff Bonwick * 15788ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 15798ad4d6ddSJeff Bonwick * either has the data or it doesn't. 15808ad4d6ddSJeff Bonwick * 15818ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 15828ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 15838ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 15848ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 15858ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 15868ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 15878ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 15888ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 15898ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 15908ad4d6ddSJeff Bonwick * 15918ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 15928ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 15938ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 15948ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 15958ad4d6ddSJeff Bonwick */ 1596fa9e4066Sahrens void 15978ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1598fa9e4066Sahrens { 15998ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 16008ad4d6ddSJeff Bonwick 16018ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16028ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1603f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 16048ad4d6ddSJeff Bonwick 1605fa9e4066Sahrens mutex_enter(sm->sm_lock); 1606fa9e4066Sahrens if (!space_map_contains(sm, txg, size)) 1607fa9e4066Sahrens space_map_add(sm, txg, size); 1608fa9e4066Sahrens mutex_exit(sm->sm_lock); 1609fa9e4066Sahrens } 1610fa9e4066Sahrens 16118ad4d6ddSJeff Bonwick boolean_t 16128ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1613fa9e4066Sahrens { 16148ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 16158ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1616fa9e4066Sahrens 16178ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 16188ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1619fa9e4066Sahrens 1620fa9e4066Sahrens mutex_enter(sm->sm_lock); 16218ad4d6ddSJeff Bonwick if (sm->sm_space != 0) 16228ad4d6ddSJeff Bonwick dirty = space_map_contains(sm, txg, size); 1623fa9e4066Sahrens mutex_exit(sm->sm_lock); 1624fa9e4066Sahrens 1625fa9e4066Sahrens return (dirty); 1626fa9e4066Sahrens } 1627fa9e4066Sahrens 16288ad4d6ddSJeff Bonwick boolean_t 16298ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 16308ad4d6ddSJeff Bonwick { 16318ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 16328ad4d6ddSJeff Bonwick boolean_t empty; 16338ad4d6ddSJeff Bonwick 16348ad4d6ddSJeff Bonwick mutex_enter(sm->sm_lock); 16358ad4d6ddSJeff Bonwick empty = (sm->sm_space == 0); 16368ad4d6ddSJeff Bonwick mutex_exit(sm->sm_lock); 16378ad4d6ddSJeff Bonwick 16388ad4d6ddSJeff Bonwick return (empty); 16398ad4d6ddSJeff Bonwick } 16408ad4d6ddSJeff Bonwick 1641fa9e4066Sahrens /* 1642fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1643fa9e4066Sahrens */ 1644fa9e4066Sahrens void 1645fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1646fa9e4066Sahrens { 1647ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 16488ad4d6ddSJeff Bonwick avl_tree_t reftree; 16498ad4d6ddSJeff Bonwick int minref; 1650fa9e4066Sahrens 16518ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1652fa9e4066Sahrens 16538ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 16548ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 16558ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 16568ad4d6ddSJeff Bonwick 1657b24ab676SJeff Bonwick if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 16588ad4d6ddSJeff Bonwick return; 16598ad4d6ddSJeff Bonwick 16608ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 16613f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 16623f9d6ad7SLin Ling 1663fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1664088f3894Sahrens if (scrub_txg != 0 && 16653f9d6ad7SLin Ling (spa->spa_scrub_started || 16663f9d6ad7SLin Ling (scn && scn->scn_phys.scn_errors == 0))) { 1667088f3894Sahrens /* 1668088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1669088f3894Sahrens * did it without rebooting, then the scrub dtl 1670088f3894Sahrens * will be valid, so excise the old region and 1671088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1672088f3894Sahrens * dtl as-is if there was an error. 16738ad4d6ddSJeff Bonwick * 16748ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 16758ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 16768ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 16778ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 16788ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 16798ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 16808ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 16818ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 16828ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1683088f3894Sahrens */ 16848ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 16858ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16868ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16878ad4d6ddSJeff Bonwick space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 16888ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 16898ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_SCRUB], 2); 16908ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, 16918ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 16928ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1693fa9e4066Sahrens } 16948ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 16958ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 16968ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1697fa9e4066Sahrens if (scrub_done) 16988ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 16998ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 17008ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 17018ad4d6ddSJeff Bonwick space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 17028ad4d6ddSJeff Bonwick else 17038ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 17048ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1705fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1706088f3894Sahrens 1707ecc2d604Sbonwick if (txg != 0) 1708ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1709fa9e4066Sahrens return; 1710fa9e4066Sahrens } 1711fa9e4066Sahrens 1712fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 17138ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 171499bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 171599bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 17168ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 17178ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 17188ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 17198ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 17208ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 17218ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 17228ad4d6ddSJeff Bonwick else 17238ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 17248ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 17258ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 17268ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 17278ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 172899bb17e2SEric Taylor space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 17298ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 17308ad4d6ddSJeff Bonwick } 17318ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 17328ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1733fa9e4066Sahrens } 17348ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1735fa9e4066Sahrens } 1736fa9e4066Sahrens 1737fa9e4066Sahrens static int 1738fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1739fa9e4066Sahrens { 1740fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 17418ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 1742ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1743fa9e4066Sahrens dmu_buf_t *db; 1744fa9e4066Sahrens int error; 1745fa9e4066Sahrens 1746fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1747fa9e4066Sahrens 1748fa9e4066Sahrens if (smo->smo_object == 0) 1749fa9e4066Sahrens return (0); 1750fa9e4066Sahrens 175188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 175288ecc943SGeorge Wilson 1753ecc2d604Sbonwick if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1754ea8dc4b6Seschrock return (error); 1755ecc2d604Sbonwick 17561934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 17571934e92fSmaybee bcopy(db->db_data, smo, sizeof (*smo)); 1758ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1759fa9e4066Sahrens 1760fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 17618ad4d6ddSJeff Bonwick error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 17628ad4d6ddSJeff Bonwick NULL, SM_ALLOC, smo, mos); 1763fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1764fa9e4066Sahrens 1765fa9e4066Sahrens return (error); 1766fa9e4066Sahrens } 1767fa9e4066Sahrens 1768fa9e4066Sahrens void 1769fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1770fa9e4066Sahrens { 1771fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 17728ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 17738ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1774ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1775fa9e4066Sahrens space_map_t smsync; 1776fa9e4066Sahrens kmutex_t smlock; 1777fa9e4066Sahrens dmu_buf_t *db; 1778fa9e4066Sahrens dmu_tx_t *tx; 1779fa9e4066Sahrens 178088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 178188ecc943SGeorge Wilson 1782fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1783fa9e4066Sahrens 1784fa9e4066Sahrens if (vd->vdev_detached) { 1785fa9e4066Sahrens if (smo->smo_object != 0) { 1786ecc2d604Sbonwick int err = dmu_object_free(mos, smo->smo_object, tx); 1787fa9e4066Sahrens ASSERT3U(err, ==, 0); 1788fa9e4066Sahrens smo->smo_object = 0; 1789fa9e4066Sahrens } 1790fa9e4066Sahrens dmu_tx_commit(tx); 1791fa9e4066Sahrens return; 1792fa9e4066Sahrens } 1793fa9e4066Sahrens 1794fa9e4066Sahrens if (smo->smo_object == 0) { 1795fa9e4066Sahrens ASSERT(smo->smo_objsize == 0); 1796fa9e4066Sahrens ASSERT(smo->smo_alloc == 0); 1797ecc2d604Sbonwick smo->smo_object = dmu_object_alloc(mos, 1798fa9e4066Sahrens DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1799fa9e4066Sahrens DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1800fa9e4066Sahrens ASSERT(smo->smo_object != 0); 1801fa9e4066Sahrens vdev_config_dirty(vd->vdev_top); 1802fa9e4066Sahrens } 1803fa9e4066Sahrens 1804fa9e4066Sahrens mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1805fa9e4066Sahrens 1806fa9e4066Sahrens space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1807fa9e4066Sahrens &smlock); 1808fa9e4066Sahrens 1809fa9e4066Sahrens mutex_enter(&smlock); 1810fa9e4066Sahrens 1811fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1812ecc2d604Sbonwick space_map_walk(sm, space_map_add, &smsync); 1813fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1814fa9e4066Sahrens 1815ecc2d604Sbonwick space_map_truncate(smo, mos, tx); 1816ecc2d604Sbonwick space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1817fa9e4066Sahrens 1818fa9e4066Sahrens space_map_destroy(&smsync); 1819fa9e4066Sahrens 1820fa9e4066Sahrens mutex_exit(&smlock); 1821fa9e4066Sahrens mutex_destroy(&smlock); 1822fa9e4066Sahrens 1823ecc2d604Sbonwick VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1824fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 18251934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 18261934e92fSmaybee bcopy(smo, db->db_data, sizeof (*smo)); 1827ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1828fa9e4066Sahrens 1829fa9e4066Sahrens dmu_tx_commit(tx); 1830fa9e4066Sahrens } 1831fa9e4066Sahrens 18328ad4d6ddSJeff Bonwick /* 18338ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 18348ad4d6ddSJeff Bonwick * without losing data. 18358ad4d6ddSJeff Bonwick */ 18368ad4d6ddSJeff Bonwick boolean_t 18378ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 18388ad4d6ddSJeff Bonwick { 18398ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 18408ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 18418ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 18428ad4d6ddSJeff Bonwick boolean_t required; 18438ad4d6ddSJeff Bonwick 18448ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 18458ad4d6ddSJeff Bonwick 18468ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 18478ad4d6ddSJeff Bonwick return (B_TRUE); 18488ad4d6ddSJeff Bonwick 18498ad4d6ddSJeff Bonwick /* 18508ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 18518ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 18528ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 18538ad4d6ddSJeff Bonwick */ 18548ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 18558ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 18568ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 18578ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 18588ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 18598ad4d6ddSJeff Bonwick 1860cb04b873SMark J Musante if (!required && zio_injection_enabled) 1861cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 1862cb04b873SMark J Musante 18638ad4d6ddSJeff Bonwick return (required); 18648ad4d6ddSJeff Bonwick } 18658ad4d6ddSJeff Bonwick 1866088f3894Sahrens /* 1867088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 1868088f3894Sahrens */ 1869088f3894Sahrens boolean_t 1870088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1871088f3894Sahrens { 1872088f3894Sahrens boolean_t needed = B_FALSE; 1873088f3894Sahrens uint64_t thismin = UINT64_MAX; 1874088f3894Sahrens uint64_t thismax = 0; 1875088f3894Sahrens 1876088f3894Sahrens if (vd->vdev_children == 0) { 1877088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 18788ad4d6ddSJeff Bonwick if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 18798ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 1880088f3894Sahrens space_seg_t *ss; 1881088f3894Sahrens 18828ad4d6ddSJeff Bonwick ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1883088f3894Sahrens thismin = ss->ss_start - 1; 18848ad4d6ddSJeff Bonwick ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1885088f3894Sahrens thismax = ss->ss_end; 1886088f3894Sahrens needed = B_TRUE; 1887088f3894Sahrens } 1888088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 1889088f3894Sahrens } else { 18908ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 1891088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 1892088f3894Sahrens uint64_t cmin, cmax; 1893088f3894Sahrens 1894088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1895088f3894Sahrens thismin = MIN(thismin, cmin); 1896088f3894Sahrens thismax = MAX(thismax, cmax); 1897088f3894Sahrens needed = B_TRUE; 1898088f3894Sahrens } 1899088f3894Sahrens } 1900088f3894Sahrens } 1901088f3894Sahrens 1902088f3894Sahrens if (needed && minp) { 1903088f3894Sahrens *minp = thismin; 1904088f3894Sahrens *maxp = thismax; 1905088f3894Sahrens } 1906088f3894Sahrens return (needed); 1907088f3894Sahrens } 1908088f3894Sahrens 1909560e6e96Seschrock void 1910ea8dc4b6Seschrock vdev_load(vdev_t *vd) 1911fa9e4066Sahrens { 1912fa9e4066Sahrens /* 1913fa9e4066Sahrens * Recursively load all children. 1914fa9e4066Sahrens */ 19158ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1916560e6e96Seschrock vdev_load(vd->vdev_child[c]); 1917fa9e4066Sahrens 1918fa9e4066Sahrens /* 19190e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 1920fa9e4066Sahrens */ 192188ecc943SGeorge Wilson if (vd == vd->vdev_top && !vd->vdev_ishole && 1922560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1923560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 1924560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1925560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1926fa9e4066Sahrens 1927fa9e4066Sahrens /* 1928fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 1929fa9e4066Sahrens */ 1930560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1931560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1932560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1933fa9e4066Sahrens } 1934fa9e4066Sahrens 193599653d4eSeschrock /* 1936fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 1937fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 1938fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 1939fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 1940fa94a07fSbrendan * repurposed to another pool. 194199653d4eSeschrock */ 194299653d4eSeschrock int 1943fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 194499653d4eSeschrock { 194599653d4eSeschrock nvlist_t *label; 194699653d4eSeschrock uint64_t guid, version; 194799653d4eSeschrock uint64_t state; 194899653d4eSeschrock 1949e14bb325SJeff Bonwick if (!vdev_readable(vd)) 1950c5904d13Seschrock return (0); 1951c5904d13Seschrock 195299653d4eSeschrock if ((label = vdev_label_read_config(vd)) == NULL) { 195399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 195499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 195599653d4eSeschrock return (-1); 195699653d4eSeschrock } 195799653d4eSeschrock 195899653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1959e7437265Sahrens version > SPA_VERSION || 196099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 196199653d4eSeschrock guid != vd->vdev_guid || 196299653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 196399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 196499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 196599653d4eSeschrock nvlist_free(label); 196699653d4eSeschrock return (-1); 196799653d4eSeschrock } 196899653d4eSeschrock 196999653d4eSeschrock /* 197099653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 197199653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 197299653d4eSeschrock */ 197399653d4eSeschrock nvlist_free(label); 197499653d4eSeschrock return (0); 197599653d4eSeschrock } 197699653d4eSeschrock 197788ecc943SGeorge Wilson void 197888ecc943SGeorge Wilson vdev_remove(vdev_t *vd, uint64_t txg) 197988ecc943SGeorge Wilson { 198088ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 198188ecc943SGeorge Wilson objset_t *mos = spa->spa_meta_objset; 198288ecc943SGeorge Wilson dmu_tx_t *tx; 198388ecc943SGeorge Wilson 198488ecc943SGeorge Wilson tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 198588ecc943SGeorge Wilson 198688ecc943SGeorge Wilson if (vd->vdev_dtl_smo.smo_object) { 198788ecc943SGeorge Wilson ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); 198888ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 198988ecc943SGeorge Wilson vd->vdev_dtl_smo.smo_object = 0; 199088ecc943SGeorge Wilson } 199188ecc943SGeorge Wilson 199288ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 199388ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 199488ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 199588ecc943SGeorge Wilson 199688ecc943SGeorge Wilson if (msp == NULL || msp->ms_smo.smo_object == 0) 199788ecc943SGeorge Wilson continue; 199888ecc943SGeorge Wilson 199988ecc943SGeorge Wilson ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); 200088ecc943SGeorge Wilson (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 200188ecc943SGeorge Wilson msp->ms_smo.smo_object = 0; 200288ecc943SGeorge Wilson } 200388ecc943SGeorge Wilson } 200488ecc943SGeorge Wilson 200588ecc943SGeorge Wilson if (vd->vdev_ms_array) { 200688ecc943SGeorge Wilson (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 200788ecc943SGeorge Wilson vd->vdev_ms_array = 0; 200888ecc943SGeorge Wilson vd->vdev_ms_shift = 0; 200988ecc943SGeorge Wilson } 201088ecc943SGeorge Wilson dmu_tx_commit(tx); 201188ecc943SGeorge Wilson } 201288ecc943SGeorge Wilson 2013fa9e4066Sahrens void 2014fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2015fa9e4066Sahrens { 2016fa9e4066Sahrens metaslab_t *msp; 201780eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2018fa9e4066Sahrens 201988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 202088ecc943SGeorge Wilson 2021fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2022fa9e4066Sahrens metaslab_sync_done(msp, txg); 202380eb36f2SGeorge Wilson 202480eb36f2SGeorge Wilson if (reassess) 202580eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2026fa9e4066Sahrens } 2027fa9e4066Sahrens 2028fa9e4066Sahrens void 2029fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2030fa9e4066Sahrens { 2031fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2032fa9e4066Sahrens vdev_t *lvd; 2033fa9e4066Sahrens metaslab_t *msp; 2034ecc2d604Sbonwick dmu_tx_t *tx; 2035fa9e4066Sahrens 203688ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 203788ecc943SGeorge Wilson 2038ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2039ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 2040ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2041ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2042ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2043ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2044ecc2d604Sbonwick vdev_config_dirty(vd); 2045ecc2d604Sbonwick dmu_tx_commit(tx); 2046ecc2d604Sbonwick } 2047fa9e4066Sahrens 20483f9d6ad7SLin Ling /* 20493f9d6ad7SLin Ling * Remove the metadata associated with this vdev once it's empty. 20503f9d6ad7SLin Ling */ 20513f9d6ad7SLin Ling if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 205288ecc943SGeorge Wilson vdev_remove(vd, txg); 205388ecc943SGeorge Wilson 2054ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2055fa9e4066Sahrens metaslab_sync(msp, txg); 2056ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2057ecc2d604Sbonwick } 2058fa9e4066Sahrens 2059fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2060fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2061fa9e4066Sahrens 2062fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2063fa9e4066Sahrens } 2064fa9e4066Sahrens 2065fa9e4066Sahrens uint64_t 2066fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2067fa9e4066Sahrens { 2068fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2069fa9e4066Sahrens } 2070fa9e4066Sahrens 20713d7072f8Seschrock /* 20723d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 20733d7072f8Seschrock * not be opened, and no I/O is attempted. 20743d7072f8Seschrock */ 2075fa9e4066Sahrens int 2076069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2077fa9e4066Sahrens { 20784b964adaSGeorge Wilson vdev_t *vd, *tvd; 2079fa9e4066Sahrens 20808f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2081fa9e4066Sahrens 2082c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2083e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2084e14bb325SJeff Bonwick 20853d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2086e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2087fa9e4066Sahrens 20884b964adaSGeorge Wilson tvd = vd->vdev_top; 20894b964adaSGeorge Wilson 2090069f55e2SEric Schrock /* 2091069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2092069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2093069f55e2SEric Schrock * were faulted. 2094069f55e2SEric Schrock */ 2095069f55e2SEric Schrock vd->vdev_label_aux = aux; 2096069f55e2SEric Schrock 20973d7072f8Seschrock /* 20983d7072f8Seschrock * Faulted state takes precedence over degraded. 20993d7072f8Seschrock */ 210098d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 21013d7072f8Seschrock vd->vdev_faulted = 1ULL; 21023d7072f8Seschrock vd->vdev_degraded = 0ULL; 2103069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 21043d7072f8Seschrock 21053d7072f8Seschrock /* 2106c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2107c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 21083d7072f8Seschrock */ 21094b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 21103d7072f8Seschrock vd->vdev_degraded = 1ULL; 21113d7072f8Seschrock vd->vdev_faulted = 0ULL; 21123d7072f8Seschrock 21133d7072f8Seschrock /* 21143d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 21153d7072f8Seschrock * mark it degraded. 21163d7072f8Seschrock */ 21174b964adaSGeorge Wilson vdev_reopen(tvd); 21183d7072f8Seschrock 2119069f55e2SEric Schrock if (vdev_readable(vd)) 2120069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 21213d7072f8Seschrock } 21223d7072f8Seschrock 2123e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 21243d7072f8Seschrock } 21253d7072f8Seschrock 21263d7072f8Seschrock /* 21273d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 21283d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 21293d7072f8Seschrock * as I/O is concerned. 21303d7072f8Seschrock */ 21313d7072f8Seschrock int 2132069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 21333d7072f8Seschrock { 2134c5904d13Seschrock vdev_t *vd; 21350a4e9518Sgw 21368f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 21373d7072f8Seschrock 2138c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2139e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2140e14bb325SJeff Bonwick 21410e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2142e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 21430e34b6a7Sbonwick 21443d7072f8Seschrock /* 21453d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 21463d7072f8Seschrock */ 2147e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2148e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 21493d7072f8Seschrock 21503d7072f8Seschrock vd->vdev_degraded = 1ULL; 21513d7072f8Seschrock if (!vdev_is_dead(vd)) 21523d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2153069f55e2SEric Schrock aux); 21543d7072f8Seschrock 2155e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 21563d7072f8Seschrock } 21573d7072f8Seschrock 21583d7072f8Seschrock /* 21593d7072f8Seschrock * Online the given vdev. If 'unspare' is set, it implies two things. First, 21603d7072f8Seschrock * any attached spare device should be detached when the device finishes 21613d7072f8Seschrock * resilvering. Second, the online should be treated like a 'test' online case, 21623d7072f8Seschrock * so no FMA events are generated if the device fails to open. 21633d7072f8Seschrock */ 21643d7072f8Seschrock int 2165e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 21663d7072f8Seschrock { 2167573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 21683d7072f8Seschrock 21698f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 21703d7072f8Seschrock 2171c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2172e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 21733d7072f8Seschrock 21743d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2175e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2176fa9e4066Sahrens 2177573ca77eSGeorge Wilson tvd = vd->vdev_top; 2178fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2179441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2180e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2181e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2182573ca77eSGeorge Wilson 2183573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2184573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2185573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2186573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2187573ca77eSGeorge Wilson } 2188573ca77eSGeorge Wilson 2189573ca77eSGeorge Wilson vdev_reopen(tvd); 21903d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 21913d7072f8Seschrock 2192573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2193573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2194573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2195573ca77eSGeorge Wilson } 2196573ca77eSGeorge Wilson 21973d7072f8Seschrock if (newstate) 21983d7072f8Seschrock *newstate = vd->vdev_state; 21993d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 22003d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 22013d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 22023d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 22033d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2204fa9e4066Sahrens 2205573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2206573ca77eSGeorge Wilson 2207573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2208573ca77eSGeorge Wilson if (vd->vdev_aux) 2209573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2210573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2211573ca77eSGeorge Wilson } 22128ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2213fa9e4066Sahrens } 2214fa9e4066Sahrens 2215a1521560SJeff Bonwick static int 2216a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2217fa9e4066Sahrens { 2218e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 22198f18d1faSGeorge Wilson int error = 0; 22208f18d1faSGeorge Wilson uint64_t generation; 22218f18d1faSGeorge Wilson metaslab_group_t *mg; 22220a4e9518Sgw 22238f18d1faSGeorge Wilson top: 22248f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2225fa9e4066Sahrens 2226c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2227e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2228fa9e4066Sahrens 22290e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2230e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 22310e34b6a7Sbonwick 2232e6ca193dSGeorge Wilson tvd = vd->vdev_top; 22338f18d1faSGeorge Wilson mg = tvd->vdev_mg; 22348f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2235e6ca193dSGeorge Wilson 2236fa9e4066Sahrens /* 2237ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2238fa9e4066Sahrens */ 2239ecc2d604Sbonwick if (!vd->vdev_offline) { 2240ecc2d604Sbonwick /* 22418ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2242e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2243e6ca193dSGeorge Wilson * expendable. 2244ecc2d604Sbonwick */ 2245e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2246e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2247e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2248fa9e4066Sahrens 22498f18d1faSGeorge Wilson /* 2250b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2251b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2252b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2253b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 22548f18d1faSGeorge Wilson */ 22558f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 22568f18d1faSGeorge Wilson /* 22578f18d1faSGeorge Wilson * Prevent any future allocations. 22588f18d1faSGeorge Wilson */ 2259a1521560SJeff Bonwick metaslab_group_passivate(mg); 22608f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 22618f18d1faSGeorge Wilson 22621195e687SMark J Musante error = spa_offline_log(spa); 22638f18d1faSGeorge Wilson 22648f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 22658f18d1faSGeorge Wilson 22668f18d1faSGeorge Wilson /* 22678f18d1faSGeorge Wilson * Check to see if the config has changed. 22688f18d1faSGeorge Wilson */ 22698f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2270a1521560SJeff Bonwick metaslab_group_activate(mg); 22718f18d1faSGeorge Wilson if (error) 22728f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 22738f18d1faSGeorge Wilson vd, error)); 22748f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 22758f18d1faSGeorge Wilson goto top; 22768f18d1faSGeorge Wilson } 22778f18d1faSGeorge Wilson ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); 22788f18d1faSGeorge Wilson } 22798f18d1faSGeorge Wilson 2280ecc2d604Sbonwick /* 2281ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 2282e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 2283e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 2284e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 2285ecc2d604Sbonwick */ 2286ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 2287e6ca193dSGeorge Wilson vdev_reopen(tvd); 2288e6ca193dSGeorge Wilson 2289e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2290e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 2291ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 2292e6ca193dSGeorge Wilson vdev_reopen(tvd); 2293e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2294ecc2d604Sbonwick } 22958f18d1faSGeorge Wilson 22968f18d1faSGeorge Wilson /* 22978f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 22988f18d1faSGeorge Wilson * once we online the device it's open for business. 22998f18d1faSGeorge Wilson */ 23008f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 2301a1521560SJeff Bonwick metaslab_group_activate(mg); 2302fa9e4066Sahrens } 2303fa9e4066Sahrens 2304e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2305ecc2d604Sbonwick 23068f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 2307fa9e4066Sahrens } 2308fa9e4066Sahrens 2309a1521560SJeff Bonwick int 2310a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2311a1521560SJeff Bonwick { 2312a1521560SJeff Bonwick int error; 2313a1521560SJeff Bonwick 2314a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 2315a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 2316a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 2317a1521560SJeff Bonwick 2318a1521560SJeff Bonwick return (error); 2319a1521560SJeff Bonwick } 2320a1521560SJeff Bonwick 2321ea8dc4b6Seschrock /* 2322ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 2323ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 2324ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2325ea8dc4b6Seschrock */ 2326ea8dc4b6Seschrock void 2327e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 2328fa9e4066Sahrens { 2329e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2330e14bb325SJeff Bonwick 2331e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2332fa9e4066Sahrens 2333ea8dc4b6Seschrock if (vd == NULL) 2334e14bb325SJeff Bonwick vd = rvd; 2335fa9e4066Sahrens 2336ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 2337ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2338ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2339fa9e4066Sahrens 2340e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2341e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 23423d7072f8Seschrock 23433d7072f8Seschrock /* 23448a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 23458a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 23468a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 23478a79c1b5Sek * written out to disk. 23483d7072f8Seschrock */ 2349e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2350e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 23518a79c1b5Sek 2352096d22d4SEric Schrock /* 2353096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 2354096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 2355096d22d4SEric Schrock * still broken, we want to still post the ereport again. 2356096d22d4SEric Schrock */ 2357096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 2358096d22d4SEric Schrock 23594b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2360e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2361e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2362e14bb325SJeff Bonwick 2363f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 23643d7072f8Seschrock 2365096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 2366096d22d4SEric Schrock 2367f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 2368e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2369e14bb325SJeff Bonwick 2370e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2371bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 23723d7072f8Seschrock 23733d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 23743d7072f8Seschrock } 2375096d22d4SEric Schrock 2376096d22d4SEric Schrock /* 2377096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 2378096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 2379096d22d4SEric Schrock * done in response to the FMA fault. 2380096d22d4SEric Schrock */ 2381096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2382096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2383096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 2384096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 2385fa9e4066Sahrens } 2386fa9e4066Sahrens 2387e14bb325SJeff Bonwick boolean_t 2388e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 23890a4e9518Sgw { 239088ecc943SGeorge Wilson /* 239188ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 239288ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 239388ecc943SGeorge Wilson * these types of devices in the various code paths. 239488ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 239588ecc943SGeorge Wilson * before issuing I/O to them. 239688ecc943SGeorge Wilson */ 239788ecc943SGeorge Wilson return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 239888ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 23990a4e9518Sgw } 24000a4e9518Sgw 2401e14bb325SJeff Bonwick boolean_t 2402e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 24030a4e9518Sgw { 2404e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 24050a4e9518Sgw } 24060a4e9518Sgw 2407e14bb325SJeff Bonwick boolean_t 2408e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2409fa9e4066Sahrens { 2410e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2411fa9e4066Sahrens } 2412fa9e4066Sahrens 2413a31e6787SGeorge Wilson boolean_t 2414a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2415a31e6787SGeorge Wilson { 24168ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 24178ad4d6ddSJeff Bonwick 2418a31e6787SGeorge Wilson /* 24198ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2420a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2421a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 24228ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 24238ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 24248ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2425a31e6787SGeorge Wilson */ 24268ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 24273f9d6ad7SLin Ling !vd->vdev_cant_write && !vd->vdev_ishole); 2428a31e6787SGeorge Wilson } 2429a31e6787SGeorge Wilson 2430e14bb325SJeff Bonwick boolean_t 2431e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2432fa9e4066Sahrens { 2433e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2434fa9e4066Sahrens 2435e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2436e14bb325SJeff Bonwick return (B_FALSE); 2437fa9e4066Sahrens 2438e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2439e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2440fa9e4066Sahrens 2441e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2442e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2443fa9e4066Sahrens 2444e14bb325SJeff Bonwick return (B_TRUE); 2445fa9e4066Sahrens } 2446fa9e4066Sahrens 2447fa9e4066Sahrens /* 2448fa9e4066Sahrens * Get statistics for the given vdev. 2449fa9e4066Sahrens */ 2450fa9e4066Sahrens void 2451fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2452fa9e4066Sahrens { 2453fa9e4066Sahrens vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2454fa9e4066Sahrens 2455fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2456fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2457fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2458fa9e4066Sahrens vs->vs_state = vd->vdev_state; 2459573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 2460573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2461573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2462fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2463fa9e4066Sahrens 2464fa9e4066Sahrens /* 2465fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2466fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2467fa9e4066Sahrens */ 2468fa9e4066Sahrens if (vd == rvd) { 2469e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2470fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2471fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2472fa9e4066Sahrens 2473fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2474e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2475fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2476fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2477fa9e4066Sahrens } 24783f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 2479fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2480fa9e4066Sahrens } 2481fa9e4066Sahrens } 2482fa9e4066Sahrens } 2483fa9e4066Sahrens 2484fa94a07fSbrendan void 2485fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2486fa94a07fSbrendan { 2487fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2488fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2489fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2490fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2491fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2492fa94a07fSbrendan } 2493fa94a07fSbrendan 24943f9d6ad7SLin Ling void 24953f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 24963f9d6ad7SLin Ling { 24973f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 24983f9d6ad7SLin Ling 24993f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 25003f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 25013f9d6ad7SLin Ling 25023f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 25033f9d6ad7SLin Ling vs->vs_scan_processed = 0; 25043f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 25053f9d6ad7SLin Ling } 25063f9d6ad7SLin Ling 2507fa9e4066Sahrens void 2508e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2509fa9e4066Sahrens { 25108ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 25118ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2512e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2513fa9e4066Sahrens vdev_t *pvd; 2514fa9e4066Sahrens uint64_t txg = zio->io_txg; 2515fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2516fa9e4066Sahrens zio_type_t type = zio->io_type; 2517fa9e4066Sahrens int flags = zio->io_flags; 2518fa9e4066Sahrens 2519e14bb325SJeff Bonwick /* 2520e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2521e14bb325SJeff Bonwick */ 2522e14bb325SJeff Bonwick if (zio->io_gang_tree) 2523e14bb325SJeff Bonwick return; 2524e14bb325SJeff Bonwick 2525fa9e4066Sahrens if (zio->io_error == 0) { 2526e14bb325SJeff Bonwick /* 2527e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2528e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2529e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2530e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2531e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2532e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2533e14bb325SJeff Bonwick * remain zero, which is what we want.) 2534e14bb325SJeff Bonwick * 2535e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2536e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2537e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2538e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2539e14bb325SJeff Bonwick */ 2540e14bb325SJeff Bonwick if (vd == rvd) 2541e14bb325SJeff Bonwick return; 2542e14bb325SJeff Bonwick 2543e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 25448ad4d6ddSJeff Bonwick 25458ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 25468ad4d6ddSJeff Bonwick return; 25478ad4d6ddSJeff Bonwick 25488ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 25498ad4d6ddSJeff Bonwick 2550e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 255144ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 25523f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 25533f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 25543f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 25553f9d6ad7SLin Ling 25563f9d6ad7SLin Ling /* XXX cleanup? */ 25573f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 25583f9d6ad7SLin Ling atomic_add_64(processed, psize); 25593f9d6ad7SLin Ling vs->vs_scan_processed += psize; 25603f9d6ad7SLin Ling } 25613f9d6ad7SLin Ling 25628ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2563e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2564fa9e4066Sahrens } 25658ad4d6ddSJeff Bonwick 25668ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 25678ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 25688ad4d6ddSJeff Bonwick 25698ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2570fa9e4066Sahrens return; 2571fa9e4066Sahrens } 2572fa9e4066Sahrens 2573fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2574fa9e4066Sahrens return; 2575fa9e4066Sahrens 25768956713aSEric Schrock /* 25778956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 25788956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 25798956713aSEric Schrock * hard errors, when in reality they can happen for any number of 25808956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 25818956713aSEric Schrock */ 25828956713aSEric Schrock if (zio->io_error == EIO && 25838956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 25848956713aSEric Schrock return; 25858956713aSEric Schrock 25868f18d1faSGeorge Wilson /* 25878f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 25888f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 25898f18d1faSGeorge Wilson * errors. 25908f18d1faSGeorge Wilson */ 25918f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 25928f18d1faSGeorge Wilson return; 25938f18d1faSGeorge Wilson 2594e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2595b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2596e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2597e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2598e14bb325SJeff Bonwick else 2599e14bb325SJeff Bonwick vs->vs_read_errors++; 2600fa9e4066Sahrens } 2601b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2602e14bb325SJeff Bonwick vs->vs_write_errors++; 2603e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2604fa9e4066Sahrens 26058ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 26068ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 260744ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 2608b24ab676SJeff Bonwick spa->spa_claiming)) { 26098ad4d6ddSJeff Bonwick /* 2610b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 2611b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 2612b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 2613b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 2614b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 2615b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 2616b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 2617b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 26188ad4d6ddSJeff Bonwick * 26198ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 26208ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 26218ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 26228ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 26238ad4d6ddSJeff Bonwick */ 26248ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 26258ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 262644ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 26278ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 26288ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 26298ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2630b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 2631b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 2632b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2633b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 26348ad4d6ddSJeff Bonwick } 2635b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 26368ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2637fa9e4066Sahrens return; 26388ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 26398ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 26408ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2641fa9e4066Sahrens } 26428ad4d6ddSJeff Bonwick if (vd != rvd) 26438ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2644fa9e4066Sahrens } 2645fa9e4066Sahrens } 2646fa9e4066Sahrens 2647fa9e4066Sahrens /* 2648b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 2649b24ab676SJeff Bonwick * and the root vdev. 2650fa9e4066Sahrens */ 2651fa9e4066Sahrens void 2652b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2653b24ab676SJeff Bonwick int64_t space_delta) 2654fa9e4066Sahrens { 265599653d4eSeschrock int64_t dspace_delta = space_delta; 26568654d025Sperrin spa_t *spa = vd->vdev_spa; 26578654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2658b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 2659b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2660fa9e4066Sahrens 26618654d025Sperrin ASSERT(vd == vd->vdev_top); 266299653d4eSeschrock 26638654d025Sperrin /* 26648654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 26658654d025Sperrin * factor. We must calculate this here and not at the root vdev 26668654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 26678654d025Sperrin * childrens', thus not accurate enough for us. 26688654d025Sperrin */ 26698654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2670e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 26718654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 26728654d025Sperrin vd->vdev_deflate_ratio; 26738654d025Sperrin 26748654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 26758654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 2676b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 26778654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 26788654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 26798654d025Sperrin 2680b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 2681fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2682fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2683b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 2684fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2685fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2686fa94a07fSbrendan } 2687b24ab676SJeff Bonwick 2688b24ab676SJeff Bonwick if (mc != NULL) { 2689b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 2690b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 2691b24ab676SJeff Bonwick 2692b24ab676SJeff Bonwick metaslab_class_space_update(mc, 2693b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 2694b24ab676SJeff Bonwick } 2695fa9e4066Sahrens } 2696fa9e4066Sahrens 2697fa9e4066Sahrens /* 2698fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2699fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2700fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2701fa9e4066Sahrens */ 2702fa9e4066Sahrens void 2703fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2704fa9e4066Sahrens { 2705fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2706fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2707fa9e4066Sahrens int c; 2708fa9e4066Sahrens 2709f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2710f9af39baSGeorge Wilson 2711c5904d13Seschrock /* 27126809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 27136809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2714c5904d13Seschrock */ 2715c5904d13Seschrock if (vd->vdev_aux != NULL) { 2716c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2717c5904d13Seschrock nvlist_t **aux; 2718c5904d13Seschrock uint_t naux; 2719c5904d13Seschrock 2720c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2721c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2722c5904d13Seschrock break; 2723c5904d13Seschrock } 2724c5904d13Seschrock 2725e14bb325SJeff Bonwick if (c == sav->sav_count) { 2726e14bb325SJeff Bonwick /* 2727e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2728e14bb325SJeff Bonwick */ 2729e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2730e14bb325SJeff Bonwick return; 2731e14bb325SJeff Bonwick } 2732e14bb325SJeff Bonwick 2733c5904d13Seschrock sav->sav_sync = B_TRUE; 2734c5904d13Seschrock 27356809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 27366809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 27376809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 27386809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 27396809eb4eSEric Schrock } 2740c5904d13Seschrock 2741c5904d13Seschrock ASSERT(c < naux); 2742c5904d13Seschrock 2743c5904d13Seschrock /* 2744c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2745c5904d13Seschrock * sketchy, but it will work. 2746c5904d13Seschrock */ 2747c5904d13Seschrock nvlist_free(aux[c]); 27483f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2749c5904d13Seschrock 2750c5904d13Seschrock return; 2751c5904d13Seschrock } 2752c5904d13Seschrock 27535dabedeeSbonwick /* 2754e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2755e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2756e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 27575dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 27585dabedeeSbonwick */ 2759e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2760e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2761e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 27625dabedeeSbonwick 2763fa9e4066Sahrens if (vd == rvd) { 2764fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2765fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2766fa9e4066Sahrens } else { 2767fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2768fa9e4066Sahrens 276988ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 277088ecc943SGeorge Wilson !vd->vdev_ishole) 2771e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2772fa9e4066Sahrens } 2773fa9e4066Sahrens } 2774fa9e4066Sahrens 2775fa9e4066Sahrens void 2776fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2777fa9e4066Sahrens { 27785dabedeeSbonwick spa_t *spa = vd->vdev_spa; 27795dabedeeSbonwick 2780e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2781e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2782e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 27835dabedeeSbonwick 2784e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2785e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2786e14bb325SJeff Bonwick } 2787e14bb325SJeff Bonwick 2788e14bb325SJeff Bonwick /* 2789e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2790e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2791e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2792e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2793e14bb325SJeff Bonwick */ 2794e14bb325SJeff Bonwick void 2795e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2796e14bb325SJeff Bonwick { 2797e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2798e14bb325SJeff Bonwick 2799f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 2800e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2801e14bb325SJeff Bonwick 2802e14bb325SJeff Bonwick /* 2803e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 2804e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 2805e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 2806e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 2807e14bb325SJeff Bonwick */ 2808e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2809e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2810e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2811e14bb325SJeff Bonwick 2812b24ab676SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2813e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 2814e14bb325SJeff Bonwick } 2815e14bb325SJeff Bonwick 2816e14bb325SJeff Bonwick void 2817e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 2818e14bb325SJeff Bonwick { 2819e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2820e14bb325SJeff Bonwick 2821e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2822e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2823e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2824e14bb325SJeff Bonwick 2825e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2826e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 2827fa9e4066Sahrens } 2828fa9e4066Sahrens 282932b87932Sek /* 283032b87932Sek * Propagate vdev state up from children to parent. 283132b87932Sek */ 283244cd46caSbillm void 283344cd46caSbillm vdev_propagate_state(vdev_t *vd) 283444cd46caSbillm { 28358ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 28368ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 283744cd46caSbillm int degraded = 0, faulted = 0; 283844cd46caSbillm int corrupted = 0; 283944cd46caSbillm vdev_t *child; 284044cd46caSbillm 28413d7072f8Seschrock if (vd->vdev_children > 0) { 2842573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 28433d7072f8Seschrock child = vd->vdev_child[c]; 284451ece835Seschrock 284588ecc943SGeorge Wilson /* 284688ecc943SGeorge Wilson * Don't factor holes into the decision. 284788ecc943SGeorge Wilson */ 284888ecc943SGeorge Wilson if (child->vdev_ishole) 284988ecc943SGeorge Wilson continue; 285088ecc943SGeorge Wilson 2851e14bb325SJeff Bonwick if (!vdev_readable(child) || 28528ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 285351ece835Seschrock /* 285451ece835Seschrock * Root special: if there is a top-level log 285551ece835Seschrock * device, treat the root vdev as if it were 285651ece835Seschrock * degraded. 285751ece835Seschrock */ 285851ece835Seschrock if (child->vdev_islog && vd == rvd) 285951ece835Seschrock degraded++; 286051ece835Seschrock else 286151ece835Seschrock faulted++; 286251ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 28633d7072f8Seschrock degraded++; 286451ece835Seschrock } 286544cd46caSbillm 28663d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 28673d7072f8Seschrock corrupted++; 28683d7072f8Seschrock } 286944cd46caSbillm 28703d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 28713d7072f8Seschrock 28723d7072f8Seschrock /* 2873e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 28743d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 28753d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 28763d7072f8Seschrock * replicas'. 28773d7072f8Seschrock */ 28783d7072f8Seschrock if (corrupted && vd == rvd && 28793d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 28803d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 28813d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 28823d7072f8Seschrock } 28833d7072f8Seschrock 288451ece835Seschrock if (vd->vdev_parent) 28853d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 288644cd46caSbillm } 288744cd46caSbillm 2888fa9e4066Sahrens /* 2889ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 2890ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 2891ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 2892ea8dc4b6Seschrock * 2893ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 2894ea8dc4b6Seschrock * generated. 2895fa9e4066Sahrens */ 2896fa9e4066Sahrens void 2897ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2898fa9e4066Sahrens { 2899560e6e96Seschrock uint64_t save_state; 2900c5904d13Seschrock spa_t *spa = vd->vdev_spa; 2901ea8dc4b6Seschrock 2902ea8dc4b6Seschrock if (state == vd->vdev_state) { 2903ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 2904fa9e4066Sahrens return; 2905ea8dc4b6Seschrock } 2906ea8dc4b6Seschrock 2907560e6e96Seschrock save_state = vd->vdev_state; 2908fa9e4066Sahrens 2909fa9e4066Sahrens vd->vdev_state = state; 2910fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 2911fa9e4066Sahrens 29123d7072f8Seschrock /* 29133d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 291498d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 291598d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 291698d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 291798d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 291898d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 291998d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 292098d1cbfeSGeorge Wilson * children. 292198d1cbfeSGeorge Wilson */ 292298d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 292398d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 29243d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 29253d7072f8Seschrock 2926069f55e2SEric Schrock /* 2927069f55e2SEric Schrock * If we have brought this vdev back into service, we need 2928069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 2929069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 2930069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 2931069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 2932069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 2933069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 2934069f55e2SEric Schrock */ 2935069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2936069f55e2SEric Schrock vd->vdev_prevstate != state) 2937069f55e2SEric Schrock zfs_post_state_change(spa, vd); 2938069f55e2SEric Schrock 29393d7072f8Seschrock if (vd->vdev_removed && 29403d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 29413d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 29423d7072f8Seschrock /* 29433d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 29443d7072f8Seschrock * device was previously marked removed and someone attempted to 29453d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 29463d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 29473d7072f8Seschrock * it is one of our special test online cases, which is only 29483d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 29493d7072f8Seschrock * fault. 29503d7072f8Seschrock */ 29513d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 29523d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 29533d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 29543d7072f8Seschrock vd->vdev_removed = B_TRUE; 29553d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 2956ea8dc4b6Seschrock /* 2957cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 2958cb04b873SMark J Musante * mark it as "not available", which signifies that it was 2959cb04b873SMark J Musante * never there to begin with. Failure to open such a device 2960cb04b873SMark J Musante * is not considered an error. 2961ea8dc4b6Seschrock */ 2962cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 2963cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 2964560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 2965560e6e96Seschrock vd->vdev_not_present = 1; 2966560e6e96Seschrock 2967560e6e96Seschrock /* 2968560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 2969560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 2970560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 2971560e6e96Seschrock * want to post the ereport if the device was already in the 2972560e6e96Seschrock * CANT_OPEN state beforehand. 29733d7072f8Seschrock * 29743d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 29753d7072f8Seschrock * online the device in response to an insertion event. If we 29763d7072f8Seschrock * hit this case, then we have detected an insertion event for a 29773d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 29783d7072f8Seschrock * In this scenario, we don't post an ereport because we are 29793d7072f8Seschrock * about to replace the device, or attempt an online with 29803d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 2981560e6e96Seschrock */ 29823d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 29833d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 2984c5904d13Seschrock vd != spa->spa_root_vdev) { 2985ea8dc4b6Seschrock const char *class; 2986ea8dc4b6Seschrock 2987ea8dc4b6Seschrock switch (aux) { 2988ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 2989ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2990ea8dc4b6Seschrock break; 2991ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 2992ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2993ea8dc4b6Seschrock break; 2994ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 2995ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2996ea8dc4b6Seschrock break; 2997ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 2998ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2999ea8dc4b6Seschrock break; 3000ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3001ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3002ea8dc4b6Seschrock break; 3003ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3004ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3005ea8dc4b6Seschrock break; 3006ea8dc4b6Seschrock default: 3007ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3008ea8dc4b6Seschrock } 3009ea8dc4b6Seschrock 3010c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3011ea8dc4b6Seschrock } 3012ea8dc4b6Seschrock 30133d7072f8Seschrock /* Erase any notion of persistent removed state */ 30143d7072f8Seschrock vd->vdev_removed = B_FALSE; 30153d7072f8Seschrock } else { 30163d7072f8Seschrock vd->vdev_removed = B_FALSE; 30173d7072f8Seschrock } 3018ea8dc4b6Seschrock 30198b33d774STim Haley if (!isopen && vd->vdev_parent) 30208b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3021fa9e4066Sahrens } 302215e6edf1Sgw 302315e6edf1Sgw /* 302415e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 302515e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 302615e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 302715e6edf1Sgw * can be wholedisks. 302815e6edf1Sgw */ 302915e6edf1Sgw boolean_t 303015e6edf1Sgw vdev_is_bootable(vdev_t *vd) 303115e6edf1Sgw { 303215e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 303315e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 303415e6edf1Sgw 303515e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 303615e6edf1Sgw vd->vdev_children > 1) { 303715e6edf1Sgw return (B_FALSE); 303815e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 303915e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 304015e6edf1Sgw return (B_FALSE); 304115e6edf1Sgw } 304215e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 304315e6edf1Sgw return (B_FALSE); 304415e6edf1Sgw } 304515e6edf1Sgw 3046573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 304715e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 304815e6edf1Sgw return (B_FALSE); 304915e6edf1Sgw } 305015e6edf1Sgw return (B_TRUE); 305115e6edf1Sgw } 3052e6ca193dSGeorge Wilson 305388ecc943SGeorge Wilson /* 305488ecc943SGeorge Wilson * Load the state from the original vdev tree (ovd) which 305588ecc943SGeorge Wilson * we've retrieved from the MOS config object. If the original 30564b964adaSGeorge Wilson * vdev was offline or faulted then we transfer that state to the 30574b964adaSGeorge Wilson * device in the current vdev tree (nvd). 305888ecc943SGeorge Wilson */ 3059e6ca193dSGeorge Wilson void 306088ecc943SGeorge Wilson vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3061e6ca193dSGeorge Wilson { 306288ecc943SGeorge Wilson spa_t *spa = nvd->vdev_spa; 3063e6ca193dSGeorge Wilson 30644b964adaSGeorge Wilson ASSERT(nvd->vdev_top->vdev_islog); 306588ecc943SGeorge Wilson ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 306688ecc943SGeorge Wilson ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3067e6ca193dSGeorge Wilson 306888ecc943SGeorge Wilson for (int c = 0; c < nvd->vdev_children; c++) 306988ecc943SGeorge Wilson vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3070e6ca193dSGeorge Wilson 30714b964adaSGeorge Wilson if (nvd->vdev_ops->vdev_op_leaf) { 3072e6ca193dSGeorge Wilson /* 30734b964adaSGeorge Wilson * Restore the persistent vdev state 3074e6ca193dSGeorge Wilson */ 307588ecc943SGeorge Wilson nvd->vdev_offline = ovd->vdev_offline; 30764b964adaSGeorge Wilson nvd->vdev_faulted = ovd->vdev_faulted; 30774b964adaSGeorge Wilson nvd->vdev_degraded = ovd->vdev_degraded; 30784b964adaSGeorge Wilson nvd->vdev_removed = ovd->vdev_removed; 3079e6ca193dSGeorge Wilson } 3080e6ca193dSGeorge Wilson } 3081573ca77eSGeorge Wilson 30824b964adaSGeorge Wilson /* 30834b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 30844b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 30854b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 30864b964adaSGeorge Wilson */ 30874b964adaSGeorge Wilson boolean_t 30884b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 30894b964adaSGeorge Wilson { 30904b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 30914b964adaSGeorge Wilson !vd->vdev_removed) 30924b964adaSGeorge Wilson return (B_TRUE); 30934b964adaSGeorge Wilson 30944b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 30954b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 30964b964adaSGeorge Wilson return (B_TRUE); 30974b964adaSGeorge Wilson 30984b964adaSGeorge Wilson return (B_FALSE); 30994b964adaSGeorge Wilson } 31004b964adaSGeorge Wilson 3101573ca77eSGeorge Wilson /* 3102573ca77eSGeorge Wilson * Expand a vdev if possible. 3103573ca77eSGeorge Wilson */ 3104573ca77eSGeorge Wilson void 3105573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3106573ca77eSGeorge Wilson { 3107573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3108573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3109573ca77eSGeorge Wilson 3110573ca77eSGeorge Wilson if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3111573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3112573ca77eSGeorge Wilson vdev_config_dirty(vd); 3113573ca77eSGeorge Wilson } 3114573ca77eSGeorge Wilson } 31151195e687SMark J Musante 31161195e687SMark J Musante /* 31171195e687SMark J Musante * Split a vdev. 31181195e687SMark J Musante */ 31191195e687SMark J Musante void 31201195e687SMark J Musante vdev_split(vdev_t *vd) 31211195e687SMark J Musante { 31221195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 31231195e687SMark J Musante 31241195e687SMark J Musante vdev_remove_child(pvd, vd); 31251195e687SMark J Musante vdev_compact_children(pvd); 31261195e687SMark J Musante 31271195e687SMark J Musante cvd = pvd->vdev_child[0]; 31281195e687SMark J Musante if (pvd->vdev_children == 1) { 31291195e687SMark J Musante vdev_remove_parent(cvd); 31301195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 31311195e687SMark J Musante } 31321195e687SMark J Musante vdev_propagate_state(cvd); 31331195e687SMark J Musante } 3134