vdev.c revision 252751
1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd 22168404Spjd/* 23219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24229578Smm * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25249643Smm * Copyright (c) 2013 by Delphix. All rights reserved. 26248369Smm * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27168404Spjd */ 28168404Spjd 29168404Spjd#include <sys/zfs_context.h> 30168404Spjd#include <sys/fm/fs/zfs.h> 31168404Spjd#include <sys/spa.h> 32168404Spjd#include <sys/spa_impl.h> 33168404Spjd#include <sys/dmu.h> 34168404Spjd#include <sys/dmu_tx.h> 35168404Spjd#include <sys/vdev_impl.h> 36168404Spjd#include <sys/uberblock_impl.h> 37168404Spjd#include <sys/metaslab.h> 38168404Spjd#include <sys/metaslab_impl.h> 39168404Spjd#include <sys/space_map.h> 40168404Spjd#include <sys/zio.h> 41168404Spjd#include <sys/zap.h> 42168404Spjd#include <sys/fs/zfs.h> 43185029Spjd#include <sys/arc.h> 44213197Smm#include <sys/zil.h> 45219089Spjd#include <sys/dsl_scan.h> 46251419Ssmh#include <sys/trim_map.h> 47168404Spjd 48168404SpjdSYSCTL_DECL(_vfs_zfs); 49168404SpjdSYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); 50168404Spjd 51168404Spjd/* 52168404Spjd * Virtual device management. 53168404Spjd */ 54168404Spjd 55168404Spjdstatic vdev_ops_t *vdev_ops_table[] = { 56168404Spjd &vdev_root_ops, 57168404Spjd &vdev_raidz_ops, 58168404Spjd &vdev_mirror_ops, 59168404Spjd &vdev_replacing_ops, 60168404Spjd &vdev_spare_ops, 61168404Spjd#ifdef _KERNEL 62168404Spjd &vdev_geom_ops, 63168404Spjd#else 64168404Spjd &vdev_disk_ops, 65185029Spjd#endif 66168404Spjd &vdev_file_ops, 67168404Spjd &vdev_missing_ops, 68219089Spjd &vdev_hole_ops, 69168404Spjd NULL 70168404Spjd}; 71168404Spjd 72168404Spjd 73168404Spjd/* 74168404Spjd * Given a vdev type, return the appropriate ops vector. 75168404Spjd */ 76168404Spjdstatic vdev_ops_t * 77168404Spjdvdev_getops(const char *type) 78168404Spjd{ 79168404Spjd vdev_ops_t *ops, **opspp; 80168404Spjd 81168404Spjd for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 82168404Spjd if (strcmp(ops->vdev_op_type, type) == 0) 83168404Spjd break; 84168404Spjd 85168404Spjd return (ops); 86168404Spjd} 87168404Spjd 88168404Spjd/* 89168404Spjd * Default asize function: return the MAX of psize with the asize of 90168404Spjd * all children. This is what's used by anything other than RAID-Z. 91168404Spjd */ 92168404Spjduint64_t 93168404Spjdvdev_default_asize(vdev_t *vd, uint64_t psize) 94168404Spjd{ 95168404Spjd uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 96168404Spjd uint64_t csize; 97168404Spjd 98219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 99168404Spjd csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 100168404Spjd asize = MAX(asize, csize); 101168404Spjd } 102168404Spjd 103168404Spjd return (asize); 104168404Spjd} 105168404Spjd 106168404Spjd/* 107219089Spjd * Get the minimum allocatable size. We define the allocatable size as 108219089Spjd * the vdev's asize rounded to the nearest metaslab. This allows us to 109219089Spjd * replace or attach devices which don't have the same physical size but 110219089Spjd * can still satisfy the same number of allocations. 111168404Spjd */ 112168404Spjduint64_t 113219089Spjdvdev_get_min_asize(vdev_t *vd) 114168404Spjd{ 115219089Spjd vdev_t *pvd = vd->vdev_parent; 116168404Spjd 117219089Spjd /* 118236839Smm * If our parent is NULL (inactive spare or cache) or is the root, 119219089Spjd * just return our own asize. 120219089Spjd */ 121219089Spjd if (pvd == NULL) 122219089Spjd return (vd->vdev_asize); 123168404Spjd 124168404Spjd /* 125219089Spjd * The top-level vdev just returns the allocatable size rounded 126219089Spjd * to the nearest metaslab. 127168404Spjd */ 128219089Spjd if (vd == vd->vdev_top) 129219089Spjd return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 130168404Spjd 131219089Spjd /* 132219089Spjd * The allocatable space for a raidz vdev is N * sizeof(smallest child), 133219089Spjd * so each child must provide at least 1/Nth of its asize. 134219089Spjd */ 135219089Spjd if (pvd->vdev_ops == &vdev_raidz_ops) 136219089Spjd return (pvd->vdev_min_asize / pvd->vdev_children); 137168404Spjd 138219089Spjd return (pvd->vdev_min_asize); 139219089Spjd} 140168404Spjd 141219089Spjdvoid 142219089Spjdvdev_set_min_asize(vdev_t *vd) 143219089Spjd{ 144219089Spjd vd->vdev_min_asize = vdev_get_min_asize(vd); 145219089Spjd 146219089Spjd for (int c = 0; c < vd->vdev_children; c++) 147219089Spjd vdev_set_min_asize(vd->vdev_child[c]); 148168404Spjd} 149168404Spjd 150168404Spjdvdev_t * 151168404Spjdvdev_lookup_top(spa_t *spa, uint64_t vdev) 152168404Spjd{ 153168404Spjd vdev_t *rvd = spa->spa_root_vdev; 154168404Spjd 155185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 156185029Spjd 157185029Spjd if (vdev < rvd->vdev_children) { 158185029Spjd ASSERT(rvd->vdev_child[vdev] != NULL); 159168404Spjd return (rvd->vdev_child[vdev]); 160185029Spjd } 161168404Spjd 162168404Spjd return (NULL); 163168404Spjd} 164168404Spjd 165168404Spjdvdev_t * 166168404Spjdvdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 167168404Spjd{ 168168404Spjd vdev_t *mvd; 169168404Spjd 170168404Spjd if (vd->vdev_guid == guid) 171168404Spjd return (vd); 172168404Spjd 173219089Spjd for (int c = 0; c < vd->vdev_children; c++) 174168404Spjd if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 175168404Spjd NULL) 176168404Spjd return (mvd); 177168404Spjd 178168404Spjd return (NULL); 179168404Spjd} 180168404Spjd 181168404Spjdvoid 182168404Spjdvdev_add_child(vdev_t *pvd, vdev_t *cvd) 183168404Spjd{ 184168404Spjd size_t oldsize, newsize; 185168404Spjd uint64_t id = cvd->vdev_id; 186168404Spjd vdev_t **newchild; 187168404Spjd 188185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 189168404Spjd ASSERT(cvd->vdev_parent == NULL); 190168404Spjd 191168404Spjd cvd->vdev_parent = pvd; 192168404Spjd 193168404Spjd if (pvd == NULL) 194168404Spjd return; 195168404Spjd 196168404Spjd ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 197168404Spjd 198168404Spjd oldsize = pvd->vdev_children * sizeof (vdev_t *); 199168404Spjd pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 200168404Spjd newsize = pvd->vdev_children * sizeof (vdev_t *); 201168404Spjd 202168404Spjd newchild = kmem_zalloc(newsize, KM_SLEEP); 203168404Spjd if (pvd->vdev_child != NULL) { 204168404Spjd bcopy(pvd->vdev_child, newchild, oldsize); 205168404Spjd kmem_free(pvd->vdev_child, oldsize); 206168404Spjd } 207168404Spjd 208168404Spjd pvd->vdev_child = newchild; 209168404Spjd pvd->vdev_child[id] = cvd; 210168404Spjd 211168404Spjd cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 212168404Spjd ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 213168404Spjd 214168404Spjd /* 215168404Spjd * Walk up all ancestors to update guid sum. 216168404Spjd */ 217168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 218168404Spjd pvd->vdev_guid_sum += cvd->vdev_guid_sum; 219168404Spjd} 220168404Spjd 221168404Spjdvoid 222168404Spjdvdev_remove_child(vdev_t *pvd, vdev_t *cvd) 223168404Spjd{ 224168404Spjd int c; 225168404Spjd uint_t id = cvd->vdev_id; 226168404Spjd 227168404Spjd ASSERT(cvd->vdev_parent == pvd); 228168404Spjd 229168404Spjd if (pvd == NULL) 230168404Spjd return; 231168404Spjd 232168404Spjd ASSERT(id < pvd->vdev_children); 233168404Spjd ASSERT(pvd->vdev_child[id] == cvd); 234168404Spjd 235168404Spjd pvd->vdev_child[id] = NULL; 236168404Spjd cvd->vdev_parent = NULL; 237168404Spjd 238168404Spjd for (c = 0; c < pvd->vdev_children; c++) 239168404Spjd if (pvd->vdev_child[c]) 240168404Spjd break; 241168404Spjd 242168404Spjd if (c == pvd->vdev_children) { 243168404Spjd kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 244168404Spjd pvd->vdev_child = NULL; 245168404Spjd pvd->vdev_children = 0; 246168404Spjd } 247168404Spjd 248168404Spjd /* 249168404Spjd * Walk up all ancestors to update guid sum. 250168404Spjd */ 251168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 252168404Spjd pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 253168404Spjd} 254168404Spjd 255168404Spjd/* 256168404Spjd * Remove any holes in the child array. 257168404Spjd */ 258168404Spjdvoid 259168404Spjdvdev_compact_children(vdev_t *pvd) 260168404Spjd{ 261168404Spjd vdev_t **newchild, *cvd; 262168404Spjd int oldc = pvd->vdev_children; 263219089Spjd int newc; 264168404Spjd 265185029Spjd ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 266168404Spjd 267219089Spjd for (int c = newc = 0; c < oldc; c++) 268168404Spjd if (pvd->vdev_child[c]) 269168404Spjd newc++; 270168404Spjd 271168404Spjd newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 272168404Spjd 273219089Spjd for (int c = newc = 0; c < oldc; c++) { 274168404Spjd if ((cvd = pvd->vdev_child[c]) != NULL) { 275168404Spjd newchild[newc] = cvd; 276168404Spjd cvd->vdev_id = newc++; 277168404Spjd } 278168404Spjd } 279168404Spjd 280168404Spjd kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 281168404Spjd pvd->vdev_child = newchild; 282168404Spjd pvd->vdev_children = newc; 283168404Spjd} 284168404Spjd 285168404Spjd/* 286168404Spjd * Allocate and minimally initialize a vdev_t. 287168404Spjd */ 288219089Spjdvdev_t * 289168404Spjdvdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 290168404Spjd{ 291168404Spjd vdev_t *vd; 292168404Spjd 293168404Spjd vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 294168404Spjd 295168404Spjd if (spa->spa_root_vdev == NULL) { 296168404Spjd ASSERT(ops == &vdev_root_ops); 297168404Spjd spa->spa_root_vdev = vd; 298229578Smm spa->spa_load_guid = spa_generate_guid(NULL); 299168404Spjd } 300168404Spjd 301219089Spjd if (guid == 0 && ops != &vdev_hole_ops) { 302168404Spjd if (spa->spa_root_vdev == vd) { 303168404Spjd /* 304168404Spjd * The root vdev's guid will also be the pool guid, 305168404Spjd * which must be unique among all pools. 306168404Spjd */ 307219089Spjd guid = spa_generate_guid(NULL); 308168404Spjd } else { 309168404Spjd /* 310168404Spjd * Any other vdev's guid must be unique within the pool. 311168404Spjd */ 312219089Spjd guid = spa_generate_guid(spa); 313168404Spjd } 314168404Spjd ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 315168404Spjd } 316168404Spjd 317168404Spjd vd->vdev_spa = spa; 318168404Spjd vd->vdev_id = id; 319168404Spjd vd->vdev_guid = guid; 320168404Spjd vd->vdev_guid_sum = guid; 321168404Spjd vd->vdev_ops = ops; 322168404Spjd vd->vdev_state = VDEV_STATE_CLOSED; 323219089Spjd vd->vdev_ishole = (ops == &vdev_hole_ops); 324168404Spjd 325168404Spjd mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 326168404Spjd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 327185029Spjd mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 328209962Smm for (int t = 0; t < DTL_TYPES; t++) { 329209962Smm space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 330209962Smm &vd->vdev_dtl_lock); 331209962Smm } 332168404Spjd txg_list_create(&vd->vdev_ms_list, 333168404Spjd offsetof(struct metaslab, ms_txg_node)); 334168404Spjd txg_list_create(&vd->vdev_dtl_list, 335168404Spjd offsetof(struct vdev, vdev_dtl_node)); 336168404Spjd vd->vdev_stat.vs_timestamp = gethrtime(); 337185029Spjd vdev_queue_init(vd); 338185029Spjd vdev_cache_init(vd); 339168404Spjd 340168404Spjd return (vd); 341168404Spjd} 342168404Spjd 343168404Spjd/* 344168404Spjd * Allocate a new vdev. The 'alloctype' is used to control whether we are 345168404Spjd * creating a new vdev or loading an existing one - the behavior is slightly 346168404Spjd * different for each case. 347168404Spjd */ 348168404Spjdint 349168404Spjdvdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 350168404Spjd int alloctype) 351168404Spjd{ 352168404Spjd vdev_ops_t *ops; 353168404Spjd char *type; 354185029Spjd uint64_t guid = 0, islog, nparity; 355168404Spjd vdev_t *vd; 356168404Spjd 357185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 358168404Spjd 359168404Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 360249643Smm return (SET_ERROR(EINVAL)); 361168404Spjd 362168404Spjd if ((ops = vdev_getops(type)) == NULL) 363249643Smm return (SET_ERROR(EINVAL)); 364168404Spjd 365168404Spjd /* 366168404Spjd * If this is a load, get the vdev guid from the nvlist. 367168404Spjd * Otherwise, vdev_alloc_common() will generate one for us. 368168404Spjd */ 369168404Spjd if (alloctype == VDEV_ALLOC_LOAD) { 370168404Spjd uint64_t label_id; 371168404Spjd 372168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 373168404Spjd label_id != id) 374249643Smm return (SET_ERROR(EINVAL)); 375168404Spjd 376168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 377249643Smm return (SET_ERROR(EINVAL)); 378168404Spjd } else if (alloctype == VDEV_ALLOC_SPARE) { 379168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 380249643Smm return (SET_ERROR(EINVAL)); 381185029Spjd } else if (alloctype == VDEV_ALLOC_L2CACHE) { 382185029Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 383249643Smm return (SET_ERROR(EINVAL)); 384219089Spjd } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 385219089Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 386249643Smm return (SET_ERROR(EINVAL)); 387168404Spjd } 388168404Spjd 389168404Spjd /* 390168404Spjd * The first allocated vdev must be of type 'root'. 391168404Spjd */ 392168404Spjd if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 393249643Smm return (SET_ERROR(EINVAL)); 394168404Spjd 395185029Spjd /* 396185029Spjd * Determine whether we're a log vdev. 397185029Spjd */ 398185029Spjd islog = 0; 399185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 400185029Spjd if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 401249643Smm return (SET_ERROR(ENOTSUP)); 402168404Spjd 403219089Spjd if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 404249643Smm return (SET_ERROR(ENOTSUP)); 405219089Spjd 406168404Spjd /* 407185029Spjd * Set the nparity property for RAID-Z vdevs. 408168404Spjd */ 409185029Spjd nparity = -1ULL; 410168404Spjd if (ops == &vdev_raidz_ops) { 411168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 412185029Spjd &nparity) == 0) { 413219089Spjd if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 414249643Smm return (SET_ERROR(EINVAL)); 415168404Spjd /* 416219089Spjd * Previous versions could only support 1 or 2 parity 417219089Spjd * device. 418168404Spjd */ 419219089Spjd if (nparity > 1 && 420219089Spjd spa_version(spa) < SPA_VERSION_RAIDZ2) 421249643Smm return (SET_ERROR(ENOTSUP)); 422219089Spjd if (nparity > 2 && 423219089Spjd spa_version(spa) < SPA_VERSION_RAIDZ3) 424249643Smm return (SET_ERROR(ENOTSUP)); 425168404Spjd } else { 426168404Spjd /* 427168404Spjd * We require the parity to be specified for SPAs that 428168404Spjd * support multiple parity levels. 429168404Spjd */ 430219089Spjd if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 431249643Smm return (SET_ERROR(EINVAL)); 432168404Spjd /* 433168404Spjd * Otherwise, we default to 1 parity device for RAID-Z. 434168404Spjd */ 435185029Spjd nparity = 1; 436168404Spjd } 437168404Spjd } else { 438185029Spjd nparity = 0; 439168404Spjd } 440185029Spjd ASSERT(nparity != -1ULL); 441168404Spjd 442185029Spjd vd = vdev_alloc_common(spa, id, guid, ops); 443185029Spjd 444185029Spjd vd->vdev_islog = islog; 445185029Spjd vd->vdev_nparity = nparity; 446185029Spjd 447185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 448185029Spjd vd->vdev_path = spa_strdup(vd->vdev_path); 449185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 450185029Spjd vd->vdev_devid = spa_strdup(vd->vdev_devid); 451185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 452185029Spjd &vd->vdev_physpath) == 0) 453185029Spjd vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 454209962Smm if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 455209962Smm vd->vdev_fru = spa_strdup(vd->vdev_fru); 456185029Spjd 457168404Spjd /* 458168404Spjd * Set the whole_disk property. If it's not specified, leave the value 459168404Spjd * as -1. 460168404Spjd */ 461168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 462168404Spjd &vd->vdev_wholedisk) != 0) 463168404Spjd vd->vdev_wholedisk = -1ULL; 464168404Spjd 465168404Spjd /* 466168404Spjd * Look for the 'not present' flag. This will only be set if the device 467168404Spjd * was not present at the time of import. 468168404Spjd */ 469209962Smm (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 470209962Smm &vd->vdev_not_present); 471168404Spjd 472168404Spjd /* 473168404Spjd * Get the alignment requirement. 474168404Spjd */ 475168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 476168404Spjd 477168404Spjd /* 478219089Spjd * Retrieve the vdev creation time. 479219089Spjd */ 480219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 481219089Spjd &vd->vdev_crtxg); 482219089Spjd 483219089Spjd /* 484168404Spjd * If we're a top-level vdev, try to load the allocation parameters. 485168404Spjd */ 486219089Spjd if (parent && !parent->vdev_parent && 487219089Spjd (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 488168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 489168404Spjd &vd->vdev_ms_array); 490168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 491168404Spjd &vd->vdev_ms_shift); 492168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 493168404Spjd &vd->vdev_asize); 494219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 495219089Spjd &vd->vdev_removing); 496168404Spjd } 497168404Spjd 498231141Smm if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 499219089Spjd ASSERT(alloctype == VDEV_ALLOC_LOAD || 500219089Spjd alloctype == VDEV_ALLOC_ADD || 501219089Spjd alloctype == VDEV_ALLOC_SPLIT || 502219089Spjd alloctype == VDEV_ALLOC_ROOTPOOL); 503219089Spjd vd->vdev_mg = metaslab_group_create(islog ? 504219089Spjd spa_log_class(spa) : spa_normal_class(spa), vd); 505219089Spjd } 506219089Spjd 507168404Spjd /* 508185029Spjd * If we're a leaf vdev, try to load the DTL object and other state. 509168404Spjd */ 510185029Spjd if (vd->vdev_ops->vdev_op_leaf && 511219089Spjd (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 512219089Spjd alloctype == VDEV_ALLOC_ROOTPOOL)) { 513185029Spjd if (alloctype == VDEV_ALLOC_LOAD) { 514185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 515209962Smm &vd->vdev_dtl_smo.smo_object); 516185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 517185029Spjd &vd->vdev_unspare); 518185029Spjd } 519219089Spjd 520219089Spjd if (alloctype == VDEV_ALLOC_ROOTPOOL) { 521219089Spjd uint64_t spare = 0; 522219089Spjd 523219089Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 524219089Spjd &spare) == 0 && spare) 525219089Spjd spa_spare_add(vd); 526219089Spjd } 527219089Spjd 528168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 529168404Spjd &vd->vdev_offline); 530185029Spjd 531219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, 532219089Spjd &vd->vdev_resilvering); 533219089Spjd 534185029Spjd /* 535185029Spjd * When importing a pool, we want to ignore the persistent fault 536185029Spjd * state, as the diagnosis made on another system may not be 537219089Spjd * valid in the current context. Local vdevs will 538219089Spjd * remain in the faulted state. 539185029Spjd */ 540219089Spjd if (spa_load_state(spa) == SPA_LOAD_OPEN) { 541185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 542185029Spjd &vd->vdev_faulted); 543185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 544185029Spjd &vd->vdev_degraded); 545185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 546185029Spjd &vd->vdev_removed); 547219089Spjd 548219089Spjd if (vd->vdev_faulted || vd->vdev_degraded) { 549219089Spjd char *aux; 550219089Spjd 551219089Spjd vd->vdev_label_aux = 552219089Spjd VDEV_AUX_ERR_EXCEEDED; 553219089Spjd if (nvlist_lookup_string(nv, 554219089Spjd ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 555219089Spjd strcmp(aux, "external") == 0) 556219089Spjd vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 557219089Spjd } 558185029Spjd } 559168404Spjd } 560168404Spjd 561168404Spjd /* 562168404Spjd * Add ourselves to the parent's list of children. 563168404Spjd */ 564168404Spjd vdev_add_child(parent, vd); 565168404Spjd 566168404Spjd *vdp = vd; 567168404Spjd 568168404Spjd return (0); 569168404Spjd} 570168404Spjd 571168404Spjdvoid 572168404Spjdvdev_free(vdev_t *vd) 573168404Spjd{ 574185029Spjd spa_t *spa = vd->vdev_spa; 575168404Spjd 576168404Spjd /* 577168404Spjd * vdev_free() implies closing the vdev first. This is simpler than 578168404Spjd * trying to ensure complicated semantics for all callers. 579168404Spjd */ 580168404Spjd vdev_close(vd); 581168404Spjd 582185029Spjd ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 583219089Spjd ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 584168404Spjd 585168404Spjd /* 586168404Spjd * Free all children. 587168404Spjd */ 588219089Spjd for (int c = 0; c < vd->vdev_children; c++) 589168404Spjd vdev_free(vd->vdev_child[c]); 590168404Spjd 591168404Spjd ASSERT(vd->vdev_child == NULL); 592168404Spjd ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 593168404Spjd 594168404Spjd /* 595168404Spjd * Discard allocation state. 596168404Spjd */ 597219089Spjd if (vd->vdev_mg != NULL) { 598168404Spjd vdev_metaslab_fini(vd); 599219089Spjd metaslab_group_destroy(vd->vdev_mg); 600219089Spjd } 601168404Spjd 602243674Smm ASSERT0(vd->vdev_stat.vs_space); 603243674Smm ASSERT0(vd->vdev_stat.vs_dspace); 604243674Smm ASSERT0(vd->vdev_stat.vs_alloc); 605168404Spjd 606168404Spjd /* 607168404Spjd * Remove this vdev from its parent's child list. 608168404Spjd */ 609168404Spjd vdev_remove_child(vd->vdev_parent, vd); 610168404Spjd 611168404Spjd ASSERT(vd->vdev_parent == NULL); 612168404Spjd 613185029Spjd /* 614185029Spjd * Clean up vdev structure. 615185029Spjd */ 616185029Spjd vdev_queue_fini(vd); 617185029Spjd vdev_cache_fini(vd); 618185029Spjd 619185029Spjd if (vd->vdev_path) 620185029Spjd spa_strfree(vd->vdev_path); 621185029Spjd if (vd->vdev_devid) 622185029Spjd spa_strfree(vd->vdev_devid); 623185029Spjd if (vd->vdev_physpath) 624185029Spjd spa_strfree(vd->vdev_physpath); 625209962Smm if (vd->vdev_fru) 626209962Smm spa_strfree(vd->vdev_fru); 627185029Spjd 628185029Spjd if (vd->vdev_isspare) 629185029Spjd spa_spare_remove(vd); 630185029Spjd if (vd->vdev_isl2cache) 631185029Spjd spa_l2cache_remove(vd); 632185029Spjd 633185029Spjd txg_list_destroy(&vd->vdev_ms_list); 634185029Spjd txg_list_destroy(&vd->vdev_dtl_list); 635209962Smm 636185029Spjd mutex_enter(&vd->vdev_dtl_lock); 637209962Smm for (int t = 0; t < DTL_TYPES; t++) { 638209962Smm space_map_unload(&vd->vdev_dtl[t]); 639209962Smm space_map_destroy(&vd->vdev_dtl[t]); 640209962Smm } 641185029Spjd mutex_exit(&vd->vdev_dtl_lock); 642209962Smm 643185029Spjd mutex_destroy(&vd->vdev_dtl_lock); 644185029Spjd mutex_destroy(&vd->vdev_stat_lock); 645185029Spjd mutex_destroy(&vd->vdev_probe_lock); 646185029Spjd 647185029Spjd if (vd == spa->spa_root_vdev) 648185029Spjd spa->spa_root_vdev = NULL; 649185029Spjd 650185029Spjd kmem_free(vd, sizeof (vdev_t)); 651168404Spjd} 652168404Spjd 653168404Spjd/* 654168404Spjd * Transfer top-level vdev state from svd to tvd. 655168404Spjd */ 656168404Spjdstatic void 657168404Spjdvdev_top_transfer(vdev_t *svd, vdev_t *tvd) 658168404Spjd{ 659168404Spjd spa_t *spa = svd->vdev_spa; 660168404Spjd metaslab_t *msp; 661168404Spjd vdev_t *vd; 662168404Spjd int t; 663168404Spjd 664168404Spjd ASSERT(tvd == tvd->vdev_top); 665168404Spjd 666168404Spjd tvd->vdev_ms_array = svd->vdev_ms_array; 667168404Spjd tvd->vdev_ms_shift = svd->vdev_ms_shift; 668168404Spjd tvd->vdev_ms_count = svd->vdev_ms_count; 669168404Spjd 670168404Spjd svd->vdev_ms_array = 0; 671168404Spjd svd->vdev_ms_shift = 0; 672168404Spjd svd->vdev_ms_count = 0; 673168404Spjd 674231141Smm if (tvd->vdev_mg) 675231141Smm ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 676168404Spjd tvd->vdev_mg = svd->vdev_mg; 677168404Spjd tvd->vdev_ms = svd->vdev_ms; 678168404Spjd 679168404Spjd svd->vdev_mg = NULL; 680168404Spjd svd->vdev_ms = NULL; 681168404Spjd 682168404Spjd if (tvd->vdev_mg != NULL) 683168404Spjd tvd->vdev_mg->mg_vd = tvd; 684168404Spjd 685168404Spjd tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 686168404Spjd tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 687168404Spjd tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 688168404Spjd 689168404Spjd svd->vdev_stat.vs_alloc = 0; 690168404Spjd svd->vdev_stat.vs_space = 0; 691168404Spjd svd->vdev_stat.vs_dspace = 0; 692168404Spjd 693168404Spjd for (t = 0; t < TXG_SIZE; t++) { 694168404Spjd while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 695168404Spjd (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 696168404Spjd while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 697168404Spjd (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 698168404Spjd if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 699168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 700168404Spjd } 701168404Spjd 702185029Spjd if (list_link_active(&svd->vdev_config_dirty_node)) { 703168404Spjd vdev_config_clean(svd); 704168404Spjd vdev_config_dirty(tvd); 705168404Spjd } 706168404Spjd 707185029Spjd if (list_link_active(&svd->vdev_state_dirty_node)) { 708185029Spjd vdev_state_clean(svd); 709185029Spjd vdev_state_dirty(tvd); 710185029Spjd } 711168404Spjd 712168404Spjd tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 713168404Spjd svd->vdev_deflate_ratio = 0; 714185029Spjd 715185029Spjd tvd->vdev_islog = svd->vdev_islog; 716185029Spjd svd->vdev_islog = 0; 717168404Spjd} 718168404Spjd 719168404Spjdstatic void 720168404Spjdvdev_top_update(vdev_t *tvd, vdev_t *vd) 721168404Spjd{ 722168404Spjd if (vd == NULL) 723168404Spjd return; 724168404Spjd 725168404Spjd vd->vdev_top = tvd; 726168404Spjd 727219089Spjd for (int c = 0; c < vd->vdev_children; c++) 728168404Spjd vdev_top_update(tvd, vd->vdev_child[c]); 729168404Spjd} 730168404Spjd 731168404Spjd/* 732168404Spjd * Add a mirror/replacing vdev above an existing vdev. 733168404Spjd */ 734168404Spjdvdev_t * 735168404Spjdvdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 736168404Spjd{ 737168404Spjd spa_t *spa = cvd->vdev_spa; 738168404Spjd vdev_t *pvd = cvd->vdev_parent; 739168404Spjd vdev_t *mvd; 740168404Spjd 741185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 742168404Spjd 743168404Spjd mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 744168404Spjd 745168404Spjd mvd->vdev_asize = cvd->vdev_asize; 746219089Spjd mvd->vdev_min_asize = cvd->vdev_min_asize; 747236839Smm mvd->vdev_max_asize = cvd->vdev_max_asize; 748168404Spjd mvd->vdev_ashift = cvd->vdev_ashift; 749168404Spjd mvd->vdev_state = cvd->vdev_state; 750219089Spjd mvd->vdev_crtxg = cvd->vdev_crtxg; 751168404Spjd 752168404Spjd vdev_remove_child(pvd, cvd); 753168404Spjd vdev_add_child(pvd, mvd); 754168404Spjd cvd->vdev_id = mvd->vdev_children; 755168404Spjd vdev_add_child(mvd, cvd); 756168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 757168404Spjd 758168404Spjd if (mvd == mvd->vdev_top) 759168404Spjd vdev_top_transfer(cvd, mvd); 760168404Spjd 761168404Spjd return (mvd); 762168404Spjd} 763168404Spjd 764168404Spjd/* 765168404Spjd * Remove a 1-way mirror/replacing vdev from the tree. 766168404Spjd */ 767168404Spjdvoid 768168404Spjdvdev_remove_parent(vdev_t *cvd) 769168404Spjd{ 770168404Spjd vdev_t *mvd = cvd->vdev_parent; 771168404Spjd vdev_t *pvd = mvd->vdev_parent; 772168404Spjd 773185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 774168404Spjd 775168404Spjd ASSERT(mvd->vdev_children == 1); 776168404Spjd ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 777168404Spjd mvd->vdev_ops == &vdev_replacing_ops || 778168404Spjd mvd->vdev_ops == &vdev_spare_ops); 779168404Spjd cvd->vdev_ashift = mvd->vdev_ashift; 780168404Spjd 781168404Spjd vdev_remove_child(mvd, cvd); 782168404Spjd vdev_remove_child(pvd, mvd); 783209962Smm 784185029Spjd /* 785185029Spjd * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 786185029Spjd * Otherwise, we could have detached an offline device, and when we 787185029Spjd * go to import the pool we'll think we have two top-level vdevs, 788185029Spjd * instead of a different version of the same top-level vdev. 789185029Spjd */ 790209962Smm if (mvd->vdev_top == mvd) { 791209962Smm uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 792219089Spjd cvd->vdev_orig_guid = cvd->vdev_guid; 793209962Smm cvd->vdev_guid += guid_delta; 794209962Smm cvd->vdev_guid_sum += guid_delta; 795209962Smm } 796168404Spjd cvd->vdev_id = mvd->vdev_id; 797168404Spjd vdev_add_child(pvd, cvd); 798168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 799168404Spjd 800168404Spjd if (cvd == cvd->vdev_top) 801168404Spjd vdev_top_transfer(mvd, cvd); 802168404Spjd 803168404Spjd ASSERT(mvd->vdev_children == 0); 804168404Spjd vdev_free(mvd); 805168404Spjd} 806168404Spjd 807168404Spjdint 808168404Spjdvdev_metaslab_init(vdev_t *vd, uint64_t txg) 809168404Spjd{ 810168404Spjd spa_t *spa = vd->vdev_spa; 811168404Spjd objset_t *mos = spa->spa_meta_objset; 812168404Spjd uint64_t m; 813168404Spjd uint64_t oldc = vd->vdev_ms_count; 814168404Spjd uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 815168404Spjd metaslab_t **mspp; 816168404Spjd int error; 817168404Spjd 818219089Spjd ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 819219089Spjd 820219089Spjd /* 821219089Spjd * This vdev is not being allocated from yet or is a hole. 822219089Spjd */ 823219089Spjd if (vd->vdev_ms_shift == 0) 824168404Spjd return (0); 825168404Spjd 826219089Spjd ASSERT(!vd->vdev_ishole); 827219089Spjd 828213197Smm /* 829213197Smm * Compute the raidz-deflation ratio. Note, we hard-code 830213197Smm * in 128k (1 << 17) because it is the current "typical" blocksize. 831213197Smm * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 832213197Smm * or we will inconsistently account for existing bp's. 833213197Smm */ 834213197Smm vd->vdev_deflate_ratio = (1 << 17) / 835213197Smm (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 836213197Smm 837168404Spjd ASSERT(oldc <= newc); 838168404Spjd 839168404Spjd mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 840168404Spjd 841168404Spjd if (oldc != 0) { 842168404Spjd bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 843168404Spjd kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 844168404Spjd } 845168404Spjd 846168404Spjd vd->vdev_ms = mspp; 847168404Spjd vd->vdev_ms_count = newc; 848168404Spjd 849168404Spjd for (m = oldc; m < newc; m++) { 850168404Spjd space_map_obj_t smo = { 0, 0, 0 }; 851168404Spjd if (txg == 0) { 852168404Spjd uint64_t object = 0; 853168404Spjd error = dmu_read(mos, vd->vdev_ms_array, 854209962Smm m * sizeof (uint64_t), sizeof (uint64_t), &object, 855209962Smm DMU_READ_PREFETCH); 856168404Spjd if (error) 857168404Spjd return (error); 858168404Spjd if (object != 0) { 859168404Spjd dmu_buf_t *db; 860168404Spjd error = dmu_bonus_hold(mos, object, FTAG, &db); 861168404Spjd if (error) 862168404Spjd return (error); 863185029Spjd ASSERT3U(db->db_size, >=, sizeof (smo)); 864185029Spjd bcopy(db->db_data, &smo, sizeof (smo)); 865168404Spjd ASSERT3U(smo.smo_object, ==, object); 866168404Spjd dmu_buf_rele(db, FTAG); 867168404Spjd } 868168404Spjd } 869168404Spjd vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 870168404Spjd m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 871168404Spjd } 872168404Spjd 873219089Spjd if (txg == 0) 874219089Spjd spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 875219089Spjd 876219089Spjd /* 877219089Spjd * If the vdev is being removed we don't activate 878219089Spjd * the metaslabs since we want to ensure that no new 879219089Spjd * allocations are performed on this device. 880219089Spjd */ 881219089Spjd if (oldc == 0 && !vd->vdev_removing) 882219089Spjd metaslab_group_activate(vd->vdev_mg); 883219089Spjd 884219089Spjd if (txg == 0) 885219089Spjd spa_config_exit(spa, SCL_ALLOC, FTAG); 886219089Spjd 887168404Spjd return (0); 888168404Spjd} 889168404Spjd 890168404Spjdvoid 891168404Spjdvdev_metaslab_fini(vdev_t *vd) 892168404Spjd{ 893168404Spjd uint64_t m; 894168404Spjd uint64_t count = vd->vdev_ms_count; 895168404Spjd 896168404Spjd if (vd->vdev_ms != NULL) { 897219089Spjd metaslab_group_passivate(vd->vdev_mg); 898168404Spjd for (m = 0; m < count; m++) 899168404Spjd if (vd->vdev_ms[m] != NULL) 900168404Spjd metaslab_fini(vd->vdev_ms[m]); 901168404Spjd kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 902168404Spjd vd->vdev_ms = NULL; 903168404Spjd } 904168404Spjd} 905168404Spjd 906185029Spjdtypedef struct vdev_probe_stats { 907185029Spjd boolean_t vps_readable; 908185029Spjd boolean_t vps_writeable; 909185029Spjd int vps_flags; 910185029Spjd} vdev_probe_stats_t; 911185029Spjd 912185029Spjdstatic void 913185029Spjdvdev_probe_done(zio_t *zio) 914185029Spjd{ 915209962Smm spa_t *spa = zio->io_spa; 916209962Smm vdev_t *vd = zio->io_vd; 917185029Spjd vdev_probe_stats_t *vps = zio->io_private; 918185029Spjd 919209962Smm ASSERT(vd->vdev_probe_zio != NULL); 920209962Smm 921185029Spjd if (zio->io_type == ZIO_TYPE_READ) { 922185029Spjd if (zio->io_error == 0) 923185029Spjd vps->vps_readable = 1; 924209962Smm if (zio->io_error == 0 && spa_writeable(spa)) { 925209962Smm zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 926185029Spjd zio->io_offset, zio->io_size, zio->io_data, 927185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 928185029Spjd ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 929185029Spjd } else { 930185029Spjd zio_buf_free(zio->io_data, zio->io_size); 931185029Spjd } 932185029Spjd } else if (zio->io_type == ZIO_TYPE_WRITE) { 933185029Spjd if (zio->io_error == 0) 934185029Spjd vps->vps_writeable = 1; 935185029Spjd zio_buf_free(zio->io_data, zio->io_size); 936185029Spjd } else if (zio->io_type == ZIO_TYPE_NULL) { 937209962Smm zio_t *pio; 938185029Spjd 939185029Spjd vd->vdev_cant_read |= !vps->vps_readable; 940185029Spjd vd->vdev_cant_write |= !vps->vps_writeable; 941185029Spjd 942185029Spjd if (vdev_readable(vd) && 943209962Smm (vdev_writeable(vd) || !spa_writeable(spa))) { 944185029Spjd zio->io_error = 0; 945185029Spjd } else { 946185029Spjd ASSERT(zio->io_error != 0); 947185029Spjd zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 948209962Smm spa, vd, NULL, 0, 0); 949249643Smm zio->io_error = SET_ERROR(ENXIO); 950185029Spjd } 951209962Smm 952209962Smm mutex_enter(&vd->vdev_probe_lock); 953209962Smm ASSERT(vd->vdev_probe_zio == zio); 954209962Smm vd->vdev_probe_zio = NULL; 955209962Smm mutex_exit(&vd->vdev_probe_lock); 956209962Smm 957209962Smm while ((pio = zio_walk_parents(zio)) != NULL) 958209962Smm if (!vdev_accessible(vd, pio)) 959249643Smm pio->io_error = SET_ERROR(ENXIO); 960209962Smm 961185029Spjd kmem_free(vps, sizeof (*vps)); 962185029Spjd } 963185029Spjd} 964185029Spjd 965168404Spjd/* 966252751Sdelphij * Determine whether this device is accessible. 967252751Sdelphij * 968252751Sdelphij * Read and write to several known locations: the pad regions of each 969252751Sdelphij * vdev label but the first, which we leave alone in case it contains 970252751Sdelphij * a VTOC. 971185029Spjd */ 972185029Spjdzio_t * 973209962Smmvdev_probe(vdev_t *vd, zio_t *zio) 974185029Spjd{ 975185029Spjd spa_t *spa = vd->vdev_spa; 976209962Smm vdev_probe_stats_t *vps = NULL; 977209962Smm zio_t *pio; 978185029Spjd 979209962Smm ASSERT(vd->vdev_ops->vdev_op_leaf); 980185029Spjd 981209962Smm /* 982209962Smm * Don't probe the probe. 983209962Smm */ 984209962Smm if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 985209962Smm return (NULL); 986185029Spjd 987209962Smm /* 988209962Smm * To prevent 'probe storms' when a device fails, we create 989209962Smm * just one probe i/o at a time. All zios that want to probe 990209962Smm * this vdev will become parents of the probe io. 991209962Smm */ 992209962Smm mutex_enter(&vd->vdev_probe_lock); 993209962Smm 994209962Smm if ((pio = vd->vdev_probe_zio) == NULL) { 995209962Smm vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 996209962Smm 997209962Smm vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 998209962Smm ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 999213198Smm ZIO_FLAG_TRYHARD; 1000209962Smm 1001209962Smm if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1002209962Smm /* 1003209962Smm * vdev_cant_read and vdev_cant_write can only 1004209962Smm * transition from TRUE to FALSE when we have the 1005209962Smm * SCL_ZIO lock as writer; otherwise they can only 1006209962Smm * transition from FALSE to TRUE. This ensures that 1007209962Smm * any zio looking at these values can assume that 1008209962Smm * failures persist for the life of the I/O. That's 1009209962Smm * important because when a device has intermittent 1010209962Smm * connectivity problems, we want to ensure that 1011209962Smm * they're ascribed to the device (ENXIO) and not 1012209962Smm * the zio (EIO). 1013209962Smm * 1014209962Smm * Since we hold SCL_ZIO as writer here, clear both 1015209962Smm * values so the probe can reevaluate from first 1016209962Smm * principles. 1017209962Smm */ 1018209962Smm vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1019209962Smm vd->vdev_cant_read = B_FALSE; 1020209962Smm vd->vdev_cant_write = B_FALSE; 1021209962Smm } 1022209962Smm 1023209962Smm vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1024209962Smm vdev_probe_done, vps, 1025209962Smm vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1026209962Smm 1027219089Spjd /* 1028219089Spjd * We can't change the vdev state in this context, so we 1029219089Spjd * kick off an async task to do it on our behalf. 1030219089Spjd */ 1031209962Smm if (zio != NULL) { 1032209962Smm vd->vdev_probe_wanted = B_TRUE; 1033209962Smm spa_async_request(spa, SPA_ASYNC_PROBE); 1034209962Smm } 1035185029Spjd } 1036185029Spjd 1037209962Smm if (zio != NULL) 1038209962Smm zio_add_child(zio, pio); 1039185029Spjd 1040209962Smm mutex_exit(&vd->vdev_probe_lock); 1041185029Spjd 1042209962Smm if (vps == NULL) { 1043209962Smm ASSERT(zio != NULL); 1044209962Smm return (NULL); 1045209962Smm } 1046185029Spjd 1047185029Spjd for (int l = 1; l < VDEV_LABELS; l++) { 1048209962Smm zio_nowait(zio_read_phys(pio, vd, 1049185029Spjd vdev_label_offset(vd->vdev_psize, l, 1050209962Smm offsetof(vdev_label_t, vl_pad2)), 1051209962Smm VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1052185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1053185029Spjd ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1054185029Spjd } 1055185029Spjd 1056209962Smm if (zio == NULL) 1057209962Smm return (pio); 1058209962Smm 1059209962Smm zio_nowait(pio); 1060209962Smm return (NULL); 1061185029Spjd} 1062185029Spjd 1063219089Spjdstatic void 1064219089Spjdvdev_open_child(void *arg) 1065219089Spjd{ 1066219089Spjd vdev_t *vd = arg; 1067219089Spjd 1068219089Spjd vd->vdev_open_thread = curthread; 1069219089Spjd vd->vdev_open_error = vdev_open(vd); 1070219089Spjd vd->vdev_open_thread = NULL; 1071219089Spjd} 1072219089Spjd 1073219089Spjdboolean_t 1074219089Spjdvdev_uses_zvols(vdev_t *vd) 1075219089Spjd{ 1076219089Spjd if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1077219089Spjd strlen(ZVOL_DIR)) == 0) 1078219089Spjd return (B_TRUE); 1079219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1080219089Spjd if (vdev_uses_zvols(vd->vdev_child[c])) 1081219089Spjd return (B_TRUE); 1082219089Spjd return (B_FALSE); 1083219089Spjd} 1084219089Spjd 1085219089Spjdvoid 1086219089Spjdvdev_open_children(vdev_t *vd) 1087219089Spjd{ 1088219089Spjd taskq_t *tq; 1089219089Spjd int children = vd->vdev_children; 1090219089Spjd 1091219089Spjd /* 1092219089Spjd * in order to handle pools on top of zvols, do the opens 1093219089Spjd * in a single thread so that the same thread holds the 1094219089Spjd * spa_namespace_lock 1095219089Spjd */ 1096219089Spjd if (B_TRUE || vdev_uses_zvols(vd)) { 1097219089Spjd for (int c = 0; c < children; c++) 1098219089Spjd vd->vdev_child[c]->vdev_open_error = 1099219089Spjd vdev_open(vd->vdev_child[c]); 1100219089Spjd return; 1101219089Spjd } 1102219089Spjd tq = taskq_create("vdev_open", children, minclsyspri, 1103219089Spjd children, children, TASKQ_PREPOPULATE); 1104219089Spjd 1105219089Spjd for (int c = 0; c < children; c++) 1106219089Spjd VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1107219089Spjd TQ_SLEEP) != 0); 1108219089Spjd 1109219089Spjd taskq_destroy(tq); 1110219089Spjd} 1111219089Spjd 1112185029Spjd/* 1113168404Spjd * Prepare a virtual device for access. 1114168404Spjd */ 1115168404Spjdint 1116168404Spjdvdev_open(vdev_t *vd) 1117168404Spjd{ 1118209962Smm spa_t *spa = vd->vdev_spa; 1119168404Spjd int error; 1120168404Spjd uint64_t osize = 0; 1121236839Smm uint64_t max_osize = 0; 1122236839Smm uint64_t asize, max_asize, psize; 1123168404Spjd uint64_t ashift = 0; 1124168404Spjd 1125219089Spjd ASSERT(vd->vdev_open_thread == curthread || 1126219089Spjd spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1127168404Spjd ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1128168404Spjd vd->vdev_state == VDEV_STATE_CANT_OPEN || 1129168404Spjd vd->vdev_state == VDEV_STATE_OFFLINE); 1130168404Spjd 1131168404Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1132213197Smm vd->vdev_cant_read = B_FALSE; 1133213197Smm vd->vdev_cant_write = B_FALSE; 1134219089Spjd vd->vdev_min_asize = vdev_get_min_asize(vd); 1135168404Spjd 1136219089Spjd /* 1137219089Spjd * If this vdev is not removed, check its fault status. If it's 1138219089Spjd * faulted, bail out of the open. 1139219089Spjd */ 1140185029Spjd if (!vd->vdev_removed && vd->vdev_faulted) { 1141168404Spjd ASSERT(vd->vdev_children == 0); 1142219089Spjd ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1143219089Spjd vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1144185029Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1145219089Spjd vd->vdev_label_aux); 1146249643Smm return (SET_ERROR(ENXIO)); 1147185029Spjd } else if (vd->vdev_offline) { 1148185029Spjd ASSERT(vd->vdev_children == 0); 1149168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1150249643Smm return (SET_ERROR(ENXIO)); 1151168404Spjd } 1152168404Spjd 1153236839Smm error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1154168404Spjd 1155219089Spjd /* 1156219089Spjd * Reset the vdev_reopening flag so that we actually close 1157219089Spjd * the vdev on error. 1158219089Spjd */ 1159219089Spjd vd->vdev_reopening = B_FALSE; 1160168404Spjd if (zio_injection_enabled && error == 0) 1161213198Smm error = zio_handle_device_injection(vd, NULL, ENXIO); 1162168404Spjd 1163185029Spjd if (error) { 1164185029Spjd if (vd->vdev_removed && 1165185029Spjd vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1166185029Spjd vd->vdev_removed = B_FALSE; 1167168404Spjd 1168168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1169168404Spjd vd->vdev_stat.vs_aux); 1170168404Spjd return (error); 1171168404Spjd } 1172168404Spjd 1173185029Spjd vd->vdev_removed = B_FALSE; 1174168404Spjd 1175219089Spjd /* 1176219089Spjd * Recheck the faulted flag now that we have confirmed that 1177219089Spjd * the vdev is accessible. If we're faulted, bail. 1178219089Spjd */ 1179219089Spjd if (vd->vdev_faulted) { 1180219089Spjd ASSERT(vd->vdev_children == 0); 1181219089Spjd ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1182219089Spjd vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1183219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1184219089Spjd vd->vdev_label_aux); 1185249643Smm return (SET_ERROR(ENXIO)); 1186219089Spjd } 1187219089Spjd 1188185029Spjd if (vd->vdev_degraded) { 1189185029Spjd ASSERT(vd->vdev_children == 0); 1190185029Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1191185029Spjd VDEV_AUX_ERR_EXCEEDED); 1192185029Spjd } else { 1193219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1194185029Spjd } 1195185029Spjd 1196219089Spjd /* 1197219089Spjd * For hole or missing vdevs we just return success. 1198219089Spjd */ 1199219089Spjd if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1200219089Spjd return (0); 1201219089Spjd 1202251419Ssmh if (vd->vdev_ops->vdev_op_leaf) { 1203251419Ssmh vd->vdev_notrim = B_FALSE; 1204251419Ssmh trim_map_create(vd); 1205251419Ssmh } 1206251419Ssmh 1207219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 1208168404Spjd if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1209168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1210168404Spjd VDEV_AUX_NONE); 1211168404Spjd break; 1212168404Spjd } 1213219089Spjd } 1214168404Spjd 1215168404Spjd osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1216236839Smm max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1217168404Spjd 1218168404Spjd if (vd->vdev_children == 0) { 1219168404Spjd if (osize < SPA_MINDEVSIZE) { 1220168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1221168404Spjd VDEV_AUX_TOO_SMALL); 1222249643Smm return (SET_ERROR(EOVERFLOW)); 1223168404Spjd } 1224168404Spjd psize = osize; 1225168404Spjd asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1226236839Smm max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1227236839Smm VDEV_LABEL_END_SIZE); 1228168404Spjd } else { 1229168404Spjd if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1230168404Spjd (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1231168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1232168404Spjd VDEV_AUX_TOO_SMALL); 1233249643Smm return (SET_ERROR(EOVERFLOW)); 1234168404Spjd } 1235168404Spjd psize = 0; 1236168404Spjd asize = osize; 1237236839Smm max_asize = max_osize; 1238168404Spjd } 1239168404Spjd 1240168404Spjd vd->vdev_psize = psize; 1241168404Spjd 1242219089Spjd /* 1243219089Spjd * Make sure the allocatable size hasn't shrunk. 1244219089Spjd */ 1245219089Spjd if (asize < vd->vdev_min_asize) { 1246219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1247219089Spjd VDEV_AUX_BAD_LABEL); 1248249643Smm return (SET_ERROR(EINVAL)); 1249219089Spjd } 1250219089Spjd 1251168404Spjd if (vd->vdev_asize == 0) { 1252168404Spjd /* 1253168404Spjd * This is the first-ever open, so use the computed values. 1254168404Spjd * For testing purposes, a higher ashift can be requested. 1255168404Spjd */ 1256168404Spjd vd->vdev_asize = asize; 1257236839Smm vd->vdev_max_asize = max_asize; 1258168404Spjd vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1259168404Spjd } else { 1260168404Spjd /* 1261168404Spjd * Make sure the alignment requirement hasn't increased. 1262168404Spjd */ 1263168404Spjd if (ashift > vd->vdev_top->vdev_ashift) { 1264168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1265168404Spjd VDEV_AUX_BAD_LABEL); 1266168404Spjd return (EINVAL); 1267168404Spjd } 1268236839Smm vd->vdev_max_asize = max_asize; 1269219089Spjd } 1270168404Spjd 1271219089Spjd /* 1272219089Spjd * If all children are healthy and the asize has increased, 1273219089Spjd * then we've experienced dynamic LUN growth. If automatic 1274219089Spjd * expansion is enabled then use the additional space. 1275219089Spjd */ 1276219089Spjd if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1277219089Spjd (vd->vdev_expanding || spa->spa_autoexpand)) 1278219089Spjd vd->vdev_asize = asize; 1279168404Spjd 1280219089Spjd vdev_set_min_asize(vd); 1281168404Spjd 1282168404Spjd /* 1283185029Spjd * Ensure we can issue some IO before declaring the 1284185029Spjd * vdev open for business. 1285185029Spjd */ 1286185029Spjd if (vd->vdev_ops->vdev_op_leaf && 1287185029Spjd (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1288219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1289219089Spjd VDEV_AUX_ERR_EXCEEDED); 1290185029Spjd return (error); 1291185029Spjd } 1292185029Spjd 1293185029Spjd /* 1294185029Spjd * If a leaf vdev has a DTL, and seems healthy, then kick off a 1295209962Smm * resilver. But don't do this if we are doing a reopen for a scrub, 1296209962Smm * since this would just restart the scrub we are already doing. 1297168404Spjd */ 1298209962Smm if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1299209962Smm vdev_resilver_needed(vd, NULL, NULL)) 1300209962Smm spa_async_request(spa, SPA_ASYNC_RESILVER); 1301168404Spjd 1302168404Spjd return (0); 1303168404Spjd} 1304168404Spjd 1305168404Spjd/* 1306168404Spjd * Called once the vdevs are all opened, this routine validates the label 1307168404Spjd * contents. This needs to be done before vdev_load() so that we don't 1308185029Spjd * inadvertently do repair I/Os to the wrong device. 1309168404Spjd * 1310231141Smm * If 'strict' is false ignore the spa guid check. This is necessary because 1311231141Smm * if the machine crashed during a re-guid the new guid might have been written 1312231141Smm * to all of the vdev labels, but not the cached config. The strict check 1313231141Smm * will be performed when the pool is opened again using the mos config. 1314231141Smm * 1315168404Spjd * This function will only return failure if one of the vdevs indicates that it 1316168404Spjd * has since been destroyed or exported. This is only possible if 1317168404Spjd * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1318168404Spjd * will be updated but the function will return 0. 1319168404Spjd */ 1320168404Spjdint 1321231141Smmvdev_validate(vdev_t *vd, boolean_t strict) 1322168404Spjd{ 1323168404Spjd spa_t *spa = vd->vdev_spa; 1324168404Spjd nvlist_t *label; 1325219089Spjd uint64_t guid = 0, top_guid; 1326168404Spjd uint64_t state; 1327168404Spjd 1328219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1329231141Smm if (vdev_validate(vd->vdev_child[c], strict) != 0) 1330249643Smm return (SET_ERROR(EBADF)); 1331168404Spjd 1332168404Spjd /* 1333168404Spjd * If the device has already failed, or was marked offline, don't do 1334168404Spjd * any further validation. Otherwise, label I/O will fail and we will 1335168404Spjd * overwrite the previous state. 1336168404Spjd */ 1337185029Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1338219089Spjd uint64_t aux_guid = 0; 1339219089Spjd nvlist_t *nvl; 1340247406Smm uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1341247406Smm spa_last_synced_txg(spa) : -1ULL; 1342168404Spjd 1343243674Smm if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1344168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1345168404Spjd VDEV_AUX_BAD_LABEL); 1346168404Spjd return (0); 1347168404Spjd } 1348168404Spjd 1349219089Spjd /* 1350219089Spjd * Determine if this vdev has been split off into another 1351219089Spjd * pool. If so, then refuse to open it. 1352219089Spjd */ 1353219089Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1354219089Spjd &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1355219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1356219089Spjd VDEV_AUX_SPLIT_POOL); 1357219089Spjd nvlist_free(label); 1358219089Spjd return (0); 1359219089Spjd } 1360219089Spjd 1361231141Smm if (strict && (nvlist_lookup_uint64(label, 1362231141Smm ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1363231141Smm guid != spa_guid(spa))) { 1364168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1365168404Spjd VDEV_AUX_CORRUPT_DATA); 1366168404Spjd nvlist_free(label); 1367168404Spjd return (0); 1368168404Spjd } 1369168404Spjd 1370219089Spjd if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1371219089Spjd != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1372219089Spjd &aux_guid) != 0) 1373219089Spjd aux_guid = 0; 1374219089Spjd 1375185029Spjd /* 1376185029Spjd * If this vdev just became a top-level vdev because its 1377185029Spjd * sibling was detached, it will have adopted the parent's 1378185029Spjd * vdev guid -- but the label may or may not be on disk yet. 1379185029Spjd * Fortunately, either version of the label will have the 1380185029Spjd * same top guid, so if we're a top-level vdev, we can 1381185029Spjd * safely compare to that instead. 1382219089Spjd * 1383219089Spjd * If we split this vdev off instead, then we also check the 1384219089Spjd * original pool's guid. We don't want to consider the vdev 1385219089Spjd * corrupt if it is partway through a split operation. 1386185029Spjd */ 1387168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1388185029Spjd &guid) != 0 || 1389185029Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1390185029Spjd &top_guid) != 0 || 1391219089Spjd ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1392185029Spjd (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1393168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1394168404Spjd VDEV_AUX_CORRUPT_DATA); 1395168404Spjd nvlist_free(label); 1396168404Spjd return (0); 1397168404Spjd } 1398168404Spjd 1399168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1400168404Spjd &state) != 0) { 1401168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1402168404Spjd VDEV_AUX_CORRUPT_DATA); 1403168404Spjd nvlist_free(label); 1404168404Spjd return (0); 1405168404Spjd } 1406168404Spjd 1407168404Spjd nvlist_free(label); 1408168404Spjd 1409209962Smm /* 1410219089Spjd * If this is a verbatim import, no need to check the 1411209962Smm * state of the pool. 1412209962Smm */ 1413219089Spjd if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1414219089Spjd spa_load_state(spa) == SPA_LOAD_OPEN && 1415168404Spjd state != POOL_STATE_ACTIVE) 1416249643Smm return (SET_ERROR(EBADF)); 1417185029Spjd 1418185029Spjd /* 1419185029Spjd * If we were able to open and validate a vdev that was 1420185029Spjd * previously marked permanently unavailable, clear that state 1421185029Spjd * now. 1422185029Spjd */ 1423185029Spjd if (vd->vdev_not_present) 1424185029Spjd vd->vdev_not_present = 0; 1425168404Spjd } 1426168404Spjd 1427168404Spjd return (0); 1428168404Spjd} 1429168404Spjd 1430168404Spjd/* 1431168404Spjd * Close a virtual device. 1432168404Spjd */ 1433168404Spjdvoid 1434168404Spjdvdev_close(vdev_t *vd) 1435168404Spjd{ 1436209962Smm spa_t *spa = vd->vdev_spa; 1437219089Spjd vdev_t *pvd = vd->vdev_parent; 1438209962Smm 1439209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1440209962Smm 1441219089Spjd /* 1442219089Spjd * If our parent is reopening, then we are as well, unless we are 1443219089Spjd * going offline. 1444219089Spjd */ 1445219089Spjd if (pvd != NULL && pvd->vdev_reopening) 1446219089Spjd vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1447219089Spjd 1448168404Spjd vd->vdev_ops->vdev_op_close(vd); 1449168404Spjd 1450185029Spjd vdev_cache_purge(vd); 1451168404Spjd 1452251419Ssmh if (vd->vdev_ops->vdev_op_leaf) 1453251419Ssmh trim_map_destroy(vd); 1454251419Ssmh 1455168404Spjd /* 1456219089Spjd * We record the previous state before we close it, so that if we are 1457168404Spjd * doing a reopen(), we don't generate FMA ereports if we notice that 1458168404Spjd * it's still faulted. 1459168404Spjd */ 1460168404Spjd vd->vdev_prevstate = vd->vdev_state; 1461168404Spjd 1462168404Spjd if (vd->vdev_offline) 1463168404Spjd vd->vdev_state = VDEV_STATE_OFFLINE; 1464168404Spjd else 1465168404Spjd vd->vdev_state = VDEV_STATE_CLOSED; 1466168404Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1467168404Spjd} 1468168404Spjd 1469168404Spjdvoid 1470219089Spjdvdev_hold(vdev_t *vd) 1471219089Spjd{ 1472219089Spjd spa_t *spa = vd->vdev_spa; 1473219089Spjd 1474219089Spjd ASSERT(spa_is_root(spa)); 1475219089Spjd if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1476219089Spjd return; 1477219089Spjd 1478219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1479219089Spjd vdev_hold(vd->vdev_child[c]); 1480219089Spjd 1481219089Spjd if (vd->vdev_ops->vdev_op_leaf) 1482219089Spjd vd->vdev_ops->vdev_op_hold(vd); 1483219089Spjd} 1484219089Spjd 1485219089Spjdvoid 1486219089Spjdvdev_rele(vdev_t *vd) 1487219089Spjd{ 1488219089Spjd spa_t *spa = vd->vdev_spa; 1489219089Spjd 1490219089Spjd ASSERT(spa_is_root(spa)); 1491219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1492219089Spjd vdev_rele(vd->vdev_child[c]); 1493219089Spjd 1494219089Spjd if (vd->vdev_ops->vdev_op_leaf) 1495219089Spjd vd->vdev_ops->vdev_op_rele(vd); 1496219089Spjd} 1497219089Spjd 1498219089Spjd/* 1499219089Spjd * Reopen all interior vdevs and any unopened leaves. We don't actually 1500219089Spjd * reopen leaf vdevs which had previously been opened as they might deadlock 1501219089Spjd * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1502219089Spjd * If the leaf has never been opened then open it, as usual. 1503219089Spjd */ 1504219089Spjdvoid 1505168404Spjdvdev_reopen(vdev_t *vd) 1506168404Spjd{ 1507168404Spjd spa_t *spa = vd->vdev_spa; 1508168404Spjd 1509185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1510168404Spjd 1511219089Spjd /* set the reopening flag unless we're taking the vdev offline */ 1512219089Spjd vd->vdev_reopening = !vd->vdev_offline; 1513168404Spjd vdev_close(vd); 1514168404Spjd (void) vdev_open(vd); 1515168404Spjd 1516168404Spjd /* 1517168404Spjd * Call vdev_validate() here to make sure we have the same device. 1518168404Spjd * Otherwise, a device with an invalid label could be successfully 1519168404Spjd * opened in response to vdev_reopen(). 1520168404Spjd */ 1521185029Spjd if (vd->vdev_aux) { 1522185029Spjd (void) vdev_validate_aux(vd); 1523185029Spjd if (vdev_readable(vd) && vdev_writeable(vd) && 1524209962Smm vd->vdev_aux == &spa->spa_l2cache && 1525219089Spjd !l2arc_vdev_present(vd)) 1526219089Spjd l2arc_add_vdev(spa, vd); 1527185029Spjd } else { 1528247406Smm (void) vdev_validate(vd, B_TRUE); 1529185029Spjd } 1530168404Spjd 1531168404Spjd /* 1532185029Spjd * Reassess parent vdev's health. 1533168404Spjd */ 1534185029Spjd vdev_propagate_state(vd); 1535168404Spjd} 1536168404Spjd 1537168404Spjdint 1538168404Spjdvdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1539168404Spjd{ 1540168404Spjd int error; 1541168404Spjd 1542168404Spjd /* 1543168404Spjd * Normally, partial opens (e.g. of a mirror) are allowed. 1544168404Spjd * For a create, however, we want to fail the request if 1545168404Spjd * there are any components we can't open. 1546168404Spjd */ 1547168404Spjd error = vdev_open(vd); 1548168404Spjd 1549168404Spjd if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1550168404Spjd vdev_close(vd); 1551168404Spjd return (error ? error : ENXIO); 1552168404Spjd } 1553168404Spjd 1554168404Spjd /* 1555168404Spjd * Recursively initialize all labels. 1556168404Spjd */ 1557168404Spjd if ((error = vdev_label_init(vd, txg, isreplacing ? 1558168404Spjd VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1559168404Spjd vdev_close(vd); 1560168404Spjd return (error); 1561168404Spjd } 1562168404Spjd 1563168404Spjd return (0); 1564168404Spjd} 1565168404Spjd 1566168404Spjdvoid 1567219089Spjdvdev_metaslab_set_size(vdev_t *vd) 1568168404Spjd{ 1569168404Spjd /* 1570168404Spjd * Aim for roughly 200 metaslabs per vdev. 1571168404Spjd */ 1572168404Spjd vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1573168404Spjd vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1574168404Spjd} 1575168404Spjd 1576168404Spjdvoid 1577168404Spjdvdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1578168404Spjd{ 1579168404Spjd ASSERT(vd == vd->vdev_top); 1580219089Spjd ASSERT(!vd->vdev_ishole); 1581168404Spjd ASSERT(ISP2(flags)); 1582219089Spjd ASSERT(spa_writeable(vd->vdev_spa)); 1583168404Spjd 1584168404Spjd if (flags & VDD_METASLAB) 1585168404Spjd (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1586168404Spjd 1587168404Spjd if (flags & VDD_DTL) 1588168404Spjd (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1589168404Spjd 1590168404Spjd (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1591168404Spjd} 1592168404Spjd 1593209962Smm/* 1594209962Smm * DTLs. 1595209962Smm * 1596209962Smm * A vdev's DTL (dirty time log) is the set of transaction groups for which 1597219089Spjd * the vdev has less than perfect replication. There are four kinds of DTL: 1598209962Smm * 1599209962Smm * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1600209962Smm * 1601209962Smm * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1602209962Smm * 1603209962Smm * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1604209962Smm * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1605209962Smm * txgs that was scrubbed. 1606209962Smm * 1607209962Smm * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1608209962Smm * persistent errors or just some device being offline. 1609209962Smm * Unlike the other three, the DTL_OUTAGE map is not generally 1610209962Smm * maintained; it's only computed when needed, typically to 1611209962Smm * determine whether a device can be detached. 1612209962Smm * 1613209962Smm * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1614209962Smm * either has the data or it doesn't. 1615209962Smm * 1616209962Smm * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1617209962Smm * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1618209962Smm * if any child is less than fully replicated, then so is its parent. 1619209962Smm * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1620209962Smm * comprising only those txgs which appear in 'maxfaults' or more children; 1621209962Smm * those are the txgs we don't have enough replication to read. For example, 1622209962Smm * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1623209962Smm * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1624209962Smm * two child DTL_MISSING maps. 1625209962Smm * 1626209962Smm * It should be clear from the above that to compute the DTLs and outage maps 1627209962Smm * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1628209962Smm * Therefore, that is all we keep on disk. When loading the pool, or after 1629209962Smm * a configuration change, we generate all other DTLs from first principles. 1630209962Smm */ 1631168404Spjdvoid 1632209962Smmvdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1633168404Spjd{ 1634209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1635209962Smm 1636209962Smm ASSERT(t < DTL_TYPES); 1637209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1638219089Spjd ASSERT(spa_writeable(vd->vdev_spa)); 1639209962Smm 1640168404Spjd mutex_enter(sm->sm_lock); 1641168404Spjd if (!space_map_contains(sm, txg, size)) 1642168404Spjd space_map_add(sm, txg, size); 1643168404Spjd mutex_exit(sm->sm_lock); 1644168404Spjd} 1645168404Spjd 1646209962Smmboolean_t 1647209962Smmvdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1648168404Spjd{ 1649209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1650209962Smm boolean_t dirty = B_FALSE; 1651168404Spjd 1652209962Smm ASSERT(t < DTL_TYPES); 1653209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1654168404Spjd 1655168404Spjd mutex_enter(sm->sm_lock); 1656209962Smm if (sm->sm_space != 0) 1657209962Smm dirty = space_map_contains(sm, txg, size); 1658168404Spjd mutex_exit(sm->sm_lock); 1659168404Spjd 1660168404Spjd return (dirty); 1661168404Spjd} 1662168404Spjd 1663209962Smmboolean_t 1664209962Smmvdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1665209962Smm{ 1666209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1667209962Smm boolean_t empty; 1668209962Smm 1669209962Smm mutex_enter(sm->sm_lock); 1670209962Smm empty = (sm->sm_space == 0); 1671209962Smm mutex_exit(sm->sm_lock); 1672209962Smm 1673209962Smm return (empty); 1674209962Smm} 1675209962Smm 1676168404Spjd/* 1677168404Spjd * Reassess DTLs after a config change or scrub completion. 1678168404Spjd */ 1679168404Spjdvoid 1680168404Spjdvdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1681168404Spjd{ 1682168404Spjd spa_t *spa = vd->vdev_spa; 1683209962Smm avl_tree_t reftree; 1684209962Smm int minref; 1685168404Spjd 1686209962Smm ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1687168404Spjd 1688209962Smm for (int c = 0; c < vd->vdev_children; c++) 1689209962Smm vdev_dtl_reassess(vd->vdev_child[c], txg, 1690209962Smm scrub_txg, scrub_done); 1691209962Smm 1692219089Spjd if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 1693209962Smm return; 1694209962Smm 1695209962Smm if (vd->vdev_ops->vdev_op_leaf) { 1696219089Spjd dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1697219089Spjd 1698168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1699185029Spjd if (scrub_txg != 0 && 1700219089Spjd (spa->spa_scrub_started || 1701219089Spjd (scn && scn->scn_phys.scn_errors == 0))) { 1702185029Spjd /* 1703185029Spjd * We completed a scrub up to scrub_txg. If we 1704185029Spjd * did it without rebooting, then the scrub dtl 1705185029Spjd * will be valid, so excise the old region and 1706185029Spjd * fold in the scrub dtl. Otherwise, leave the 1707185029Spjd * dtl as-is if there was an error. 1708209962Smm * 1709209962Smm * There's little trick here: to excise the beginning 1710209962Smm * of the DTL_MISSING map, we put it into a reference 1711209962Smm * tree and then add a segment with refcnt -1 that 1712209962Smm * covers the range [0, scrub_txg). This means 1713209962Smm * that each txg in that range has refcnt -1 or 0. 1714209962Smm * We then add DTL_SCRUB with a refcnt of 2, so that 1715209962Smm * entries in the range [0, scrub_txg) will have a 1716209962Smm * positive refcnt -- either 1 or 2. We then convert 1717209962Smm * the reference tree into the new DTL_MISSING map. 1718185029Spjd */ 1719209962Smm space_map_ref_create(&reftree); 1720209962Smm space_map_ref_add_map(&reftree, 1721209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1722209962Smm space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 1723209962Smm space_map_ref_add_map(&reftree, 1724209962Smm &vd->vdev_dtl[DTL_SCRUB], 2); 1725209962Smm space_map_ref_generate_map(&reftree, 1726209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1727209962Smm space_map_ref_destroy(&reftree); 1728168404Spjd } 1729209962Smm space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1730209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1731209962Smm space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1732168404Spjd if (scrub_done) 1733209962Smm space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1734209962Smm space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1735209962Smm if (!vdev_readable(vd)) 1736209962Smm space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1737209962Smm else 1738209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1739209962Smm space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1740168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1741185029Spjd 1742168404Spjd if (txg != 0) 1743168404Spjd vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1744168404Spjd return; 1745168404Spjd } 1746168404Spjd 1747168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1748209962Smm for (int t = 0; t < DTL_TYPES; t++) { 1749209962Smm /* account for child's outage in parent's missing map */ 1750209962Smm int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1751209962Smm if (t == DTL_SCRUB) 1752209962Smm continue; /* leaf vdevs only */ 1753209962Smm if (t == DTL_PARTIAL) 1754209962Smm minref = 1; /* i.e. non-zero */ 1755209962Smm else if (vd->vdev_nparity != 0) 1756209962Smm minref = vd->vdev_nparity + 1; /* RAID-Z */ 1757209962Smm else 1758209962Smm minref = vd->vdev_children; /* any kind of mirror */ 1759209962Smm space_map_ref_create(&reftree); 1760209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1761209962Smm vdev_t *cvd = vd->vdev_child[c]; 1762209962Smm mutex_enter(&cvd->vdev_dtl_lock); 1763209962Smm space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 1764209962Smm mutex_exit(&cvd->vdev_dtl_lock); 1765209962Smm } 1766209962Smm space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 1767209962Smm space_map_ref_destroy(&reftree); 1768209962Smm } 1769168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1770168404Spjd} 1771168404Spjd 1772168404Spjdstatic int 1773168404Spjdvdev_dtl_load(vdev_t *vd) 1774168404Spjd{ 1775168404Spjd spa_t *spa = vd->vdev_spa; 1776209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1777168404Spjd objset_t *mos = spa->spa_meta_objset; 1778168404Spjd dmu_buf_t *db; 1779168404Spjd int error; 1780168404Spjd 1781168404Spjd ASSERT(vd->vdev_children == 0); 1782168404Spjd 1783168404Spjd if (smo->smo_object == 0) 1784168404Spjd return (0); 1785168404Spjd 1786219089Spjd ASSERT(!vd->vdev_ishole); 1787219089Spjd 1788168404Spjd if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1789168404Spjd return (error); 1790168404Spjd 1791185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1792185029Spjd bcopy(db->db_data, smo, sizeof (*smo)); 1793168404Spjd dmu_buf_rele(db, FTAG); 1794168404Spjd 1795168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1796209962Smm error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 1797209962Smm NULL, SM_ALLOC, smo, mos); 1798168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1799168404Spjd 1800168404Spjd return (error); 1801168404Spjd} 1802168404Spjd 1803168404Spjdvoid 1804168404Spjdvdev_dtl_sync(vdev_t *vd, uint64_t txg) 1805168404Spjd{ 1806168404Spjd spa_t *spa = vd->vdev_spa; 1807209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1808209962Smm space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1809168404Spjd objset_t *mos = spa->spa_meta_objset; 1810168404Spjd space_map_t smsync; 1811168404Spjd kmutex_t smlock; 1812168404Spjd dmu_buf_t *db; 1813168404Spjd dmu_tx_t *tx; 1814168404Spjd 1815219089Spjd ASSERT(!vd->vdev_ishole); 1816219089Spjd 1817168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1818168404Spjd 1819168404Spjd if (vd->vdev_detached) { 1820168404Spjd if (smo->smo_object != 0) { 1821168404Spjd int err = dmu_object_free(mos, smo->smo_object, tx); 1822243674Smm ASSERT0(err); 1823168404Spjd smo->smo_object = 0; 1824168404Spjd } 1825168404Spjd dmu_tx_commit(tx); 1826168404Spjd return; 1827168404Spjd } 1828168404Spjd 1829168404Spjd if (smo->smo_object == 0) { 1830168404Spjd ASSERT(smo->smo_objsize == 0); 1831168404Spjd ASSERT(smo->smo_alloc == 0); 1832168404Spjd smo->smo_object = dmu_object_alloc(mos, 1833168404Spjd DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1834168404Spjd DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1835168404Spjd ASSERT(smo->smo_object != 0); 1836168404Spjd vdev_config_dirty(vd->vdev_top); 1837168404Spjd } 1838168404Spjd 1839252308Ssmh bzero(&smlock, sizeof (smlock)); 1840168404Spjd mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1841168404Spjd 1842168404Spjd space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1843168404Spjd &smlock); 1844168404Spjd 1845168404Spjd mutex_enter(&smlock); 1846168404Spjd 1847168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1848168404Spjd space_map_walk(sm, space_map_add, &smsync); 1849168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1850168404Spjd 1851168404Spjd space_map_truncate(smo, mos, tx); 1852168404Spjd space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1853248369Smm space_map_vacate(&smsync, NULL, NULL); 1854168404Spjd 1855168404Spjd space_map_destroy(&smsync); 1856168404Spjd 1857168404Spjd mutex_exit(&smlock); 1858168404Spjd mutex_destroy(&smlock); 1859168404Spjd 1860168404Spjd VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1861168404Spjd dmu_buf_will_dirty(db, tx); 1862185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1863185029Spjd bcopy(smo, db->db_data, sizeof (*smo)); 1864168404Spjd dmu_buf_rele(db, FTAG); 1865168404Spjd 1866168404Spjd dmu_tx_commit(tx); 1867168404Spjd} 1868168404Spjd 1869185029Spjd/* 1870209962Smm * Determine whether the specified vdev can be offlined/detached/removed 1871209962Smm * without losing data. 1872209962Smm */ 1873209962Smmboolean_t 1874209962Smmvdev_dtl_required(vdev_t *vd) 1875209962Smm{ 1876209962Smm spa_t *spa = vd->vdev_spa; 1877209962Smm vdev_t *tvd = vd->vdev_top; 1878209962Smm uint8_t cant_read = vd->vdev_cant_read; 1879209962Smm boolean_t required; 1880209962Smm 1881209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1882209962Smm 1883209962Smm if (vd == spa->spa_root_vdev || vd == tvd) 1884209962Smm return (B_TRUE); 1885209962Smm 1886209962Smm /* 1887209962Smm * Temporarily mark the device as unreadable, and then determine 1888209962Smm * whether this results in any DTL outages in the top-level vdev. 1889209962Smm * If not, we can safely offline/detach/remove the device. 1890209962Smm */ 1891209962Smm vd->vdev_cant_read = B_TRUE; 1892209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1893209962Smm required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 1894209962Smm vd->vdev_cant_read = cant_read; 1895209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1896209962Smm 1897219089Spjd if (!required && zio_injection_enabled) 1898219089Spjd required = !!zio_handle_device_injection(vd, NULL, ECHILD); 1899219089Spjd 1900209962Smm return (required); 1901209962Smm} 1902209962Smm 1903209962Smm/* 1904185029Spjd * Determine if resilver is needed, and if so the txg range. 1905185029Spjd */ 1906185029Spjdboolean_t 1907185029Spjdvdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1908185029Spjd{ 1909185029Spjd boolean_t needed = B_FALSE; 1910185029Spjd uint64_t thismin = UINT64_MAX; 1911185029Spjd uint64_t thismax = 0; 1912185029Spjd 1913185029Spjd if (vd->vdev_children == 0) { 1914185029Spjd mutex_enter(&vd->vdev_dtl_lock); 1915209962Smm if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 1916209962Smm vdev_writeable(vd)) { 1917185029Spjd space_seg_t *ss; 1918185029Spjd 1919209962Smm ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1920185029Spjd thismin = ss->ss_start - 1; 1921209962Smm ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1922185029Spjd thismax = ss->ss_end; 1923185029Spjd needed = B_TRUE; 1924185029Spjd } 1925185029Spjd mutex_exit(&vd->vdev_dtl_lock); 1926185029Spjd } else { 1927209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1928185029Spjd vdev_t *cvd = vd->vdev_child[c]; 1929185029Spjd uint64_t cmin, cmax; 1930185029Spjd 1931185029Spjd if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1932185029Spjd thismin = MIN(thismin, cmin); 1933185029Spjd thismax = MAX(thismax, cmax); 1934185029Spjd needed = B_TRUE; 1935185029Spjd } 1936185029Spjd } 1937185029Spjd } 1938185029Spjd 1939185029Spjd if (needed && minp) { 1940185029Spjd *minp = thismin; 1941185029Spjd *maxp = thismax; 1942185029Spjd } 1943185029Spjd return (needed); 1944185029Spjd} 1945185029Spjd 1946168404Spjdvoid 1947168404Spjdvdev_load(vdev_t *vd) 1948168404Spjd{ 1949168404Spjd /* 1950168404Spjd * Recursively load all children. 1951168404Spjd */ 1952209962Smm for (int c = 0; c < vd->vdev_children; c++) 1953168404Spjd vdev_load(vd->vdev_child[c]); 1954168404Spjd 1955168404Spjd /* 1956168404Spjd * If this is a top-level vdev, initialize its metaslabs. 1957168404Spjd */ 1958219089Spjd if (vd == vd->vdev_top && !vd->vdev_ishole && 1959168404Spjd (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1960168404Spjd vdev_metaslab_init(vd, 0) != 0)) 1961168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1962168404Spjd VDEV_AUX_CORRUPT_DATA); 1963168404Spjd 1964168404Spjd /* 1965168404Spjd * If this is a leaf vdev, load its DTL. 1966168404Spjd */ 1967168404Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1968168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1969168404Spjd VDEV_AUX_CORRUPT_DATA); 1970168404Spjd} 1971168404Spjd 1972168404Spjd/* 1973185029Spjd * The special vdev case is used for hot spares and l2cache devices. Its 1974185029Spjd * sole purpose it to set the vdev state for the associated vdev. To do this, 1975185029Spjd * we make sure that we can open the underlying device, then try to read the 1976185029Spjd * label, and make sure that the label is sane and that it hasn't been 1977185029Spjd * repurposed to another pool. 1978168404Spjd */ 1979168404Spjdint 1980185029Spjdvdev_validate_aux(vdev_t *vd) 1981168404Spjd{ 1982168404Spjd nvlist_t *label; 1983168404Spjd uint64_t guid, version; 1984168404Spjd uint64_t state; 1985168404Spjd 1986185029Spjd if (!vdev_readable(vd)) 1987185029Spjd return (0); 1988185029Spjd 1989243674Smm if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 1990168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1991168404Spjd VDEV_AUX_CORRUPT_DATA); 1992168404Spjd return (-1); 1993168404Spjd } 1994168404Spjd 1995168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1996243674Smm !SPA_VERSION_IS_SUPPORTED(version) || 1997168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 1998168404Spjd guid != vd->vdev_guid || 1999168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 2000168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2001168404Spjd VDEV_AUX_CORRUPT_DATA); 2002168404Spjd nvlist_free(label); 2003168404Spjd return (-1); 2004168404Spjd } 2005168404Spjd 2006168404Spjd /* 2007168404Spjd * We don't actually check the pool state here. If it's in fact in 2008168404Spjd * use by another pool, we update this fact on the fly when requested. 2009168404Spjd */ 2010168404Spjd nvlist_free(label); 2011168404Spjd return (0); 2012168404Spjd} 2013168404Spjd 2014168404Spjdvoid 2015219089Spjdvdev_remove(vdev_t *vd, uint64_t txg) 2016219089Spjd{ 2017219089Spjd spa_t *spa = vd->vdev_spa; 2018219089Spjd objset_t *mos = spa->spa_meta_objset; 2019219089Spjd dmu_tx_t *tx; 2020219089Spjd 2021219089Spjd tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2022219089Spjd 2023219089Spjd if (vd->vdev_dtl_smo.smo_object) { 2024243674Smm ASSERT0(vd->vdev_dtl_smo.smo_alloc); 2025219089Spjd (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 2026219089Spjd vd->vdev_dtl_smo.smo_object = 0; 2027219089Spjd } 2028219089Spjd 2029219089Spjd if (vd->vdev_ms != NULL) { 2030219089Spjd for (int m = 0; m < vd->vdev_ms_count; m++) { 2031219089Spjd metaslab_t *msp = vd->vdev_ms[m]; 2032219089Spjd 2033219089Spjd if (msp == NULL || msp->ms_smo.smo_object == 0) 2034219089Spjd continue; 2035219089Spjd 2036243674Smm ASSERT0(msp->ms_smo.smo_alloc); 2037219089Spjd (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 2038219089Spjd msp->ms_smo.smo_object = 0; 2039219089Spjd } 2040219089Spjd } 2041219089Spjd 2042219089Spjd if (vd->vdev_ms_array) { 2043219089Spjd (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 2044219089Spjd vd->vdev_ms_array = 0; 2045219089Spjd vd->vdev_ms_shift = 0; 2046219089Spjd } 2047219089Spjd dmu_tx_commit(tx); 2048219089Spjd} 2049219089Spjd 2050219089Spjdvoid 2051168404Spjdvdev_sync_done(vdev_t *vd, uint64_t txg) 2052168404Spjd{ 2053168404Spjd metaslab_t *msp; 2054211931Smm boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2055168404Spjd 2056219089Spjd ASSERT(!vd->vdev_ishole); 2057219089Spjd 2058168404Spjd while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2059168404Spjd metaslab_sync_done(msp, txg); 2060211931Smm 2061211931Smm if (reassess) 2062211931Smm metaslab_sync_reassess(vd->vdev_mg); 2063168404Spjd} 2064168404Spjd 2065168404Spjdvoid 2066168404Spjdvdev_sync(vdev_t *vd, uint64_t txg) 2067168404Spjd{ 2068168404Spjd spa_t *spa = vd->vdev_spa; 2069168404Spjd vdev_t *lvd; 2070168404Spjd metaslab_t *msp; 2071168404Spjd dmu_tx_t *tx; 2072168404Spjd 2073219089Spjd ASSERT(!vd->vdev_ishole); 2074219089Spjd 2075168404Spjd if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2076168404Spjd ASSERT(vd == vd->vdev_top); 2077168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2078168404Spjd vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2079168404Spjd DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2080168404Spjd ASSERT(vd->vdev_ms_array != 0); 2081168404Spjd vdev_config_dirty(vd); 2082168404Spjd dmu_tx_commit(tx); 2083168404Spjd } 2084168404Spjd 2085219089Spjd /* 2086219089Spjd * Remove the metadata associated with this vdev once it's empty. 2087219089Spjd */ 2088219089Spjd if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 2089219089Spjd vdev_remove(vd, txg); 2090219089Spjd 2091168404Spjd while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2092168404Spjd metaslab_sync(msp, txg); 2093168404Spjd (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2094168404Spjd } 2095168404Spjd 2096168404Spjd while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2097168404Spjd vdev_dtl_sync(lvd, txg); 2098168404Spjd 2099168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2100168404Spjd} 2101168404Spjd 2102168404Spjduint64_t 2103168404Spjdvdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2104168404Spjd{ 2105168404Spjd return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2106168404Spjd} 2107168404Spjd 2108185029Spjd/* 2109185029Spjd * Mark the given vdev faulted. A faulted vdev behaves as if the device could 2110185029Spjd * not be opened, and no I/O is attempted. 2111185029Spjd */ 2112185029Spjdint 2113219089Spjdvdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2114168404Spjd{ 2115219089Spjd vdev_t *vd, *tvd; 2116168404Spjd 2117219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2118185029Spjd 2119185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2120185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2121185029Spjd 2122185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 2123185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2124185029Spjd 2125219089Spjd tvd = vd->vdev_top; 2126219089Spjd 2127185029Spjd /* 2128219089Spjd * We don't directly use the aux state here, but if we do a 2129219089Spjd * vdev_reopen(), we need this value to be present to remember why we 2130219089Spjd * were faulted. 2131219089Spjd */ 2132219089Spjd vd->vdev_label_aux = aux; 2133219089Spjd 2134219089Spjd /* 2135185029Spjd * Faulted state takes precedence over degraded. 2136185029Spjd */ 2137219089Spjd vd->vdev_delayed_close = B_FALSE; 2138185029Spjd vd->vdev_faulted = 1ULL; 2139185029Spjd vd->vdev_degraded = 0ULL; 2140219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 2141185029Spjd 2142185029Spjd /* 2143219089Spjd * If this device has the only valid copy of the data, then 2144219089Spjd * back off and simply mark the vdev as degraded instead. 2145185029Spjd */ 2146219089Spjd if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 2147185029Spjd vd->vdev_degraded = 1ULL; 2148185029Spjd vd->vdev_faulted = 0ULL; 2149185029Spjd 2150185029Spjd /* 2151185029Spjd * If we reopen the device and it's not dead, only then do we 2152185029Spjd * mark it degraded. 2153185029Spjd */ 2154219089Spjd vdev_reopen(tvd); 2155185029Spjd 2156219089Spjd if (vdev_readable(vd)) 2157219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 2158185029Spjd } 2159185029Spjd 2160185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2161168404Spjd} 2162168404Spjd 2163185029Spjd/* 2164185029Spjd * Mark the given vdev degraded. A degraded vdev is purely an indication to the 2165185029Spjd * user that something is wrong. The vdev continues to operate as normal as far 2166185029Spjd * as I/O is concerned. 2167185029Spjd */ 2168185029Spjdint 2169219089Spjdvdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2170168404Spjd{ 2171185029Spjd vdev_t *vd; 2172168404Spjd 2173219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2174168404Spjd 2175185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2176185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2177168404Spjd 2178185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 2179185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2180185029Spjd 2181185029Spjd /* 2182185029Spjd * If the vdev is already faulted, then don't do anything. 2183185029Spjd */ 2184185029Spjd if (vd->vdev_faulted || vd->vdev_degraded) 2185185029Spjd return (spa_vdev_state_exit(spa, NULL, 0)); 2186185029Spjd 2187185029Spjd vd->vdev_degraded = 1ULL; 2188185029Spjd if (!vdev_is_dead(vd)) 2189185029Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2190219089Spjd aux); 2191185029Spjd 2192185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2193168404Spjd} 2194168404Spjd 2195185029Spjd/* 2196252751Sdelphij * Online the given vdev. 2197252751Sdelphij * 2198252751Sdelphij * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2199252751Sdelphij * spare device should be detached when the device finishes resilvering. 2200252751Sdelphij * Second, the online should be treated like a 'test' online case, so no FMA 2201252751Sdelphij * events are generated if the device fails to open. 2202185029Spjd */ 2203168404Spjdint 2204185029Spjdvdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 2205168404Spjd{ 2206219089Spjd vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2207168404Spjd 2208219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2209168404Spjd 2210185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2211185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2212168404Spjd 2213168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 2214185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2215168404Spjd 2216219089Spjd tvd = vd->vdev_top; 2217168404Spjd vd->vdev_offline = B_FALSE; 2218168404Spjd vd->vdev_tmpoffline = B_FALSE; 2219185029Spjd vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2220185029Spjd vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2221219089Spjd 2222219089Spjd /* XXX - L2ARC 1.0 does not support expansion */ 2223219089Spjd if (!vd->vdev_aux) { 2224219089Spjd for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2225219089Spjd pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2226219089Spjd } 2227219089Spjd 2228219089Spjd vdev_reopen(tvd); 2229185029Spjd vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 2230168404Spjd 2231219089Spjd if (!vd->vdev_aux) { 2232219089Spjd for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2233219089Spjd pvd->vdev_expanding = B_FALSE; 2234219089Spjd } 2235219089Spjd 2236185029Spjd if (newstate) 2237185029Spjd *newstate = vd->vdev_state; 2238185029Spjd if ((flags & ZFS_ONLINE_UNSPARE) && 2239185029Spjd !vdev_is_dead(vd) && vd->vdev_parent && 2240185029Spjd vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2241185029Spjd vd->vdev_parent->vdev_child[0] == vd) 2242185029Spjd vd->vdev_unspare = B_TRUE; 2243168404Spjd 2244219089Spjd if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2245219089Spjd 2246219089Spjd /* XXX - L2ARC 1.0 does not support expansion */ 2247219089Spjd if (vd->vdev_aux) 2248219089Spjd return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2249219089Spjd spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2250219089Spjd } 2251209962Smm return (spa_vdev_state_exit(spa, vd, 0)); 2252168404Spjd} 2253168404Spjd 2254219089Spjdstatic int 2255219089Spjdvdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2256168404Spjd{ 2257213197Smm vdev_t *vd, *tvd; 2258219089Spjd int error = 0; 2259219089Spjd uint64_t generation; 2260219089Spjd metaslab_group_t *mg; 2261168404Spjd 2262219089Spjdtop: 2263219089Spjd spa_vdev_state_enter(spa, SCL_ALLOC); 2264168404Spjd 2265185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2266185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2267168404Spjd 2268168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 2269185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2270168404Spjd 2271213197Smm tvd = vd->vdev_top; 2272219089Spjd mg = tvd->vdev_mg; 2273219089Spjd generation = spa->spa_config_generation + 1; 2274213197Smm 2275168404Spjd /* 2276168404Spjd * If the device isn't already offline, try to offline it. 2277168404Spjd */ 2278168404Spjd if (!vd->vdev_offline) { 2279168404Spjd /* 2280209962Smm * If this device has the only valid copy of some data, 2281213197Smm * don't allow it to be offlined. Log devices are always 2282213197Smm * expendable. 2283168404Spjd */ 2284213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2285213197Smm vdev_dtl_required(vd)) 2286185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2287168404Spjd 2288168404Spjd /* 2289219089Spjd * If the top-level is a slog and it has had allocations 2290219089Spjd * then proceed. We check that the vdev's metaslab group 2291219089Spjd * is not NULL since it's possible that we may have just 2292219089Spjd * added this vdev but not yet initialized its metaslabs. 2293219089Spjd */ 2294219089Spjd if (tvd->vdev_islog && mg != NULL) { 2295219089Spjd /* 2296219089Spjd * Prevent any future allocations. 2297219089Spjd */ 2298219089Spjd metaslab_group_passivate(mg); 2299219089Spjd (void) spa_vdev_state_exit(spa, vd, 0); 2300219089Spjd 2301219089Spjd error = spa_offline_log(spa); 2302219089Spjd 2303219089Spjd spa_vdev_state_enter(spa, SCL_ALLOC); 2304219089Spjd 2305219089Spjd /* 2306219089Spjd * Check to see if the config has changed. 2307219089Spjd */ 2308219089Spjd if (error || generation != spa->spa_config_generation) { 2309219089Spjd metaslab_group_activate(mg); 2310219089Spjd if (error) 2311219089Spjd return (spa_vdev_state_exit(spa, 2312219089Spjd vd, error)); 2313219089Spjd (void) spa_vdev_state_exit(spa, vd, 0); 2314219089Spjd goto top; 2315219089Spjd } 2316243674Smm ASSERT0(tvd->vdev_stat.vs_alloc); 2317219089Spjd } 2318219089Spjd 2319219089Spjd /* 2320168404Spjd * Offline this device and reopen its top-level vdev. 2321213197Smm * If the top-level vdev is a log device then just offline 2322213197Smm * it. Otherwise, if this action results in the top-level 2323213197Smm * vdev becoming unusable, undo it and fail the request. 2324168404Spjd */ 2325168404Spjd vd->vdev_offline = B_TRUE; 2326213197Smm vdev_reopen(tvd); 2327213197Smm 2328213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2329213197Smm vdev_is_dead(tvd)) { 2330168404Spjd vd->vdev_offline = B_FALSE; 2331213197Smm vdev_reopen(tvd); 2332185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2333168404Spjd } 2334219089Spjd 2335219089Spjd /* 2336219089Spjd * Add the device back into the metaslab rotor so that 2337219089Spjd * once we online the device it's open for business. 2338219089Spjd */ 2339219089Spjd if (tvd->vdev_islog && mg != NULL) 2340219089Spjd metaslab_group_activate(mg); 2341168404Spjd } 2342168404Spjd 2343185029Spjd vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2344168404Spjd 2345219089Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2346219089Spjd} 2347213197Smm 2348219089Spjdint 2349219089Spjdvdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2350219089Spjd{ 2351219089Spjd int error; 2352213197Smm 2353219089Spjd mutex_enter(&spa->spa_vdev_top_lock); 2354219089Spjd error = vdev_offline_locked(spa, guid, flags); 2355219089Spjd mutex_exit(&spa->spa_vdev_top_lock); 2356219089Spjd 2357219089Spjd return (error); 2358168404Spjd} 2359168404Spjd 2360168404Spjd/* 2361168404Spjd * Clear the error counts associated with this vdev. Unlike vdev_online() and 2362168404Spjd * vdev_offline(), we assume the spa config is locked. We also clear all 2363168404Spjd * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2364168404Spjd */ 2365168404Spjdvoid 2366168404Spjdvdev_clear(spa_t *spa, vdev_t *vd) 2367168404Spjd{ 2368185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2369168404Spjd 2370185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2371185029Spjd 2372168404Spjd if (vd == NULL) 2373185029Spjd vd = rvd; 2374168404Spjd 2375168404Spjd vd->vdev_stat.vs_read_errors = 0; 2376168404Spjd vd->vdev_stat.vs_write_errors = 0; 2377168404Spjd vd->vdev_stat.vs_checksum_errors = 0; 2378168404Spjd 2379185029Spjd for (int c = 0; c < vd->vdev_children; c++) 2380168404Spjd vdev_clear(spa, vd->vdev_child[c]); 2381185029Spjd 2382185029Spjd /* 2383185029Spjd * If we're in the FAULTED state or have experienced failed I/O, then 2384185029Spjd * clear the persistent state and attempt to reopen the device. We 2385185029Spjd * also mark the vdev config dirty, so that the new faulted state is 2386185029Spjd * written out to disk. 2387185029Spjd */ 2388185029Spjd if (vd->vdev_faulted || vd->vdev_degraded || 2389185029Spjd !vdev_readable(vd) || !vdev_writeable(vd)) { 2390185029Spjd 2391219089Spjd /* 2392219089Spjd * When reopening in reponse to a clear event, it may be due to 2393219089Spjd * a fmadm repair request. In this case, if the device is 2394219089Spjd * still broken, we want to still post the ereport again. 2395219089Spjd */ 2396219089Spjd vd->vdev_forcefault = B_TRUE; 2397219089Spjd 2398219089Spjd vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2399185029Spjd vd->vdev_cant_read = B_FALSE; 2400185029Spjd vd->vdev_cant_write = B_FALSE; 2401185029Spjd 2402219089Spjd vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 2403185029Spjd 2404219089Spjd vd->vdev_forcefault = B_FALSE; 2405219089Spjd 2406219089Spjd if (vd != rvd && vdev_writeable(vd->vdev_top)) 2407185029Spjd vdev_state_dirty(vd->vdev_top); 2408185029Spjd 2409185029Spjd if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2410185029Spjd spa_async_request(spa, SPA_ASYNC_RESILVER); 2411185029Spjd 2412185029Spjd spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2413185029Spjd } 2414219089Spjd 2415219089Spjd /* 2416219089Spjd * When clearing a FMA-diagnosed fault, we always want to 2417219089Spjd * unspare the device, as we assume that the original spare was 2418219089Spjd * done in response to the FMA fault. 2419219089Spjd */ 2420219089Spjd if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2421219089Spjd vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2422219089Spjd vd->vdev_parent->vdev_child[0] == vd) 2423219089Spjd vd->vdev_unspare = B_TRUE; 2424168404Spjd} 2425168404Spjd 2426185029Spjdboolean_t 2427168404Spjdvdev_is_dead(vdev_t *vd) 2428168404Spjd{ 2429219089Spjd /* 2430219089Spjd * Holes and missing devices are always considered "dead". 2431219089Spjd * This simplifies the code since we don't have to check for 2432219089Spjd * these types of devices in the various code paths. 2433219089Spjd * Instead we rely on the fact that we skip over dead devices 2434219089Spjd * before issuing I/O to them. 2435219089Spjd */ 2436219089Spjd return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 2437219089Spjd vd->vdev_ops == &vdev_missing_ops); 2438168404Spjd} 2439168404Spjd 2440185029Spjdboolean_t 2441185029Spjdvdev_readable(vdev_t *vd) 2442168404Spjd{ 2443185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2444185029Spjd} 2445168404Spjd 2446185029Spjdboolean_t 2447185029Spjdvdev_writeable(vdev_t *vd) 2448185029Spjd{ 2449185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2450185029Spjd} 2451168404Spjd 2452185029Spjdboolean_t 2453208370Smmvdev_allocatable(vdev_t *vd) 2454208370Smm{ 2455209962Smm uint64_t state = vd->vdev_state; 2456209962Smm 2457208370Smm /* 2458209962Smm * We currently allow allocations from vdevs which may be in the 2459208370Smm * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2460208370Smm * fails to reopen then we'll catch it later when we're holding 2461209962Smm * the proper locks. Note that we have to get the vdev state 2462209962Smm * in a local variable because although it changes atomically, 2463209962Smm * we're asking two separate questions about it. 2464208370Smm */ 2465209962Smm return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2466219089Spjd !vd->vdev_cant_write && !vd->vdev_ishole); 2467208370Smm} 2468208370Smm 2469208370Smmboolean_t 2470185029Spjdvdev_accessible(vdev_t *vd, zio_t *zio) 2471185029Spjd{ 2472185029Spjd ASSERT(zio->io_vd == vd); 2473168404Spjd 2474185029Spjd if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2475185029Spjd return (B_FALSE); 2476168404Spjd 2477185029Spjd if (zio->io_type == ZIO_TYPE_READ) 2478185029Spjd return (!vd->vdev_cant_read); 2479168404Spjd 2480185029Spjd if (zio->io_type == ZIO_TYPE_WRITE) 2481185029Spjd return (!vd->vdev_cant_write); 2482168404Spjd 2483185029Spjd return (B_TRUE); 2484168404Spjd} 2485168404Spjd 2486168404Spjd/* 2487168404Spjd * Get statistics for the given vdev. 2488168404Spjd */ 2489168404Spjdvoid 2490168404Spjdvdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2491168404Spjd{ 2492168404Spjd vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2493168404Spjd 2494168404Spjd mutex_enter(&vd->vdev_stat_lock); 2495168404Spjd bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2496168404Spjd vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2497168404Spjd vs->vs_state = vd->vdev_state; 2498219089Spjd vs->vs_rsize = vdev_get_min_asize(vd); 2499219089Spjd if (vd->vdev_ops->vdev_op_leaf) 2500219089Spjd vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2501236839Smm vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 2502168404Spjd mutex_exit(&vd->vdev_stat_lock); 2503168404Spjd 2504168404Spjd /* 2505168404Spjd * If we're getting stats on the root vdev, aggregate the I/O counts 2506168404Spjd * over all top-level vdevs (i.e. the direct children of the root). 2507168404Spjd */ 2508168404Spjd if (vd == rvd) { 2509185029Spjd for (int c = 0; c < rvd->vdev_children; c++) { 2510168404Spjd vdev_t *cvd = rvd->vdev_child[c]; 2511168404Spjd vdev_stat_t *cvs = &cvd->vdev_stat; 2512168404Spjd 2513168404Spjd mutex_enter(&vd->vdev_stat_lock); 2514185029Spjd for (int t = 0; t < ZIO_TYPES; t++) { 2515168404Spjd vs->vs_ops[t] += cvs->vs_ops[t]; 2516168404Spjd vs->vs_bytes[t] += cvs->vs_bytes[t]; 2517168404Spjd } 2518219089Spjd cvs->vs_scan_removing = cvd->vdev_removing; 2519168404Spjd mutex_exit(&vd->vdev_stat_lock); 2520168404Spjd } 2521168404Spjd } 2522168404Spjd} 2523168404Spjd 2524168404Spjdvoid 2525185029Spjdvdev_clear_stats(vdev_t *vd) 2526168404Spjd{ 2527185029Spjd mutex_enter(&vd->vdev_stat_lock); 2528185029Spjd vd->vdev_stat.vs_space = 0; 2529185029Spjd vd->vdev_stat.vs_dspace = 0; 2530185029Spjd vd->vdev_stat.vs_alloc = 0; 2531185029Spjd mutex_exit(&vd->vdev_stat_lock); 2532185029Spjd} 2533185029Spjd 2534185029Spjdvoid 2535219089Spjdvdev_scan_stat_init(vdev_t *vd) 2536219089Spjd{ 2537219089Spjd vdev_stat_t *vs = &vd->vdev_stat; 2538219089Spjd 2539219089Spjd for (int c = 0; c < vd->vdev_children; c++) 2540219089Spjd vdev_scan_stat_init(vd->vdev_child[c]); 2541219089Spjd 2542219089Spjd mutex_enter(&vd->vdev_stat_lock); 2543219089Spjd vs->vs_scan_processed = 0; 2544219089Spjd mutex_exit(&vd->vdev_stat_lock); 2545219089Spjd} 2546219089Spjd 2547219089Spjdvoid 2548185029Spjdvdev_stat_update(zio_t *zio, uint64_t psize) 2549185029Spjd{ 2550209962Smm spa_t *spa = zio->io_spa; 2551209962Smm vdev_t *rvd = spa->spa_root_vdev; 2552185029Spjd vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2553168404Spjd vdev_t *pvd; 2554168404Spjd uint64_t txg = zio->io_txg; 2555168404Spjd vdev_stat_t *vs = &vd->vdev_stat; 2556168404Spjd zio_type_t type = zio->io_type; 2557168404Spjd int flags = zio->io_flags; 2558168404Spjd 2559185029Spjd /* 2560185029Spjd * If this i/o is a gang leader, it didn't do any actual work. 2561185029Spjd */ 2562185029Spjd if (zio->io_gang_tree) 2563185029Spjd return; 2564185029Spjd 2565168404Spjd if (zio->io_error == 0) { 2566185029Spjd /* 2567185029Spjd * If this is a root i/o, don't count it -- we've already 2568185029Spjd * counted the top-level vdevs, and vdev_get_stats() will 2569185029Spjd * aggregate them when asked. This reduces contention on 2570185029Spjd * the root vdev_stat_lock and implicitly handles blocks 2571185029Spjd * that compress away to holes, for which there is no i/o. 2572185029Spjd * (Holes never create vdev children, so all the counters 2573185029Spjd * remain zero, which is what we want.) 2574185029Spjd * 2575185029Spjd * Note: this only applies to successful i/o (io_error == 0) 2576185029Spjd * because unlike i/o counts, errors are not additive. 2577185029Spjd * When reading a ditto block, for example, failure of 2578185029Spjd * one top-level vdev does not imply a root-level error. 2579185029Spjd */ 2580185029Spjd if (vd == rvd) 2581185029Spjd return; 2582185029Spjd 2583185029Spjd ASSERT(vd == zio->io_vd); 2584209962Smm 2585209962Smm if (flags & ZIO_FLAG_IO_BYPASS) 2586209962Smm return; 2587209962Smm 2588209962Smm mutex_enter(&vd->vdev_stat_lock); 2589209962Smm 2590185029Spjd if (flags & ZIO_FLAG_IO_REPAIR) { 2591219089Spjd if (flags & ZIO_FLAG_SCAN_THREAD) { 2592219089Spjd dsl_scan_phys_t *scn_phys = 2593219089Spjd &spa->spa_dsl_pool->dp_scan->scn_phys; 2594219089Spjd uint64_t *processed = &scn_phys->scn_processed; 2595219089Spjd 2596219089Spjd /* XXX cleanup? */ 2597219089Spjd if (vd->vdev_ops->vdev_op_leaf) 2598219089Spjd atomic_add_64(processed, psize); 2599219089Spjd vs->vs_scan_processed += psize; 2600219089Spjd } 2601219089Spjd 2602209962Smm if (flags & ZIO_FLAG_SELF_HEAL) 2603185029Spjd vs->vs_self_healed += psize; 2604168404Spjd } 2605209962Smm 2606209962Smm vs->vs_ops[type]++; 2607209962Smm vs->vs_bytes[type] += psize; 2608209962Smm 2609209962Smm mutex_exit(&vd->vdev_stat_lock); 2610168404Spjd return; 2611168404Spjd } 2612168404Spjd 2613168404Spjd if (flags & ZIO_FLAG_SPECULATIVE) 2614168404Spjd return; 2615168404Spjd 2616213198Smm /* 2617213198Smm * If this is an I/O error that is going to be retried, then ignore the 2618213198Smm * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2619213198Smm * hard errors, when in reality they can happen for any number of 2620213198Smm * innocuous reasons (bus resets, MPxIO link failure, etc). 2621213198Smm */ 2622213198Smm if (zio->io_error == EIO && 2623213198Smm !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2624213198Smm return; 2625213198Smm 2626219089Spjd /* 2627219089Spjd * Intent logs writes won't propagate their error to the root 2628219089Spjd * I/O so don't mark these types of failures as pool-level 2629219089Spjd * errors. 2630219089Spjd */ 2631219089Spjd if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 2632219089Spjd return; 2633219089Spjd 2634185029Spjd mutex_enter(&vd->vdev_stat_lock); 2635209962Smm if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2636185029Spjd if (zio->io_error == ECKSUM) 2637185029Spjd vs->vs_checksum_errors++; 2638185029Spjd else 2639185029Spjd vs->vs_read_errors++; 2640168404Spjd } 2641209962Smm if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2642185029Spjd vs->vs_write_errors++; 2643185029Spjd mutex_exit(&vd->vdev_stat_lock); 2644168404Spjd 2645209962Smm if (type == ZIO_TYPE_WRITE && txg != 0 && 2646209962Smm (!(flags & ZIO_FLAG_IO_REPAIR) || 2647219089Spjd (flags & ZIO_FLAG_SCAN_THREAD) || 2648219089Spjd spa->spa_claiming)) { 2649209962Smm /* 2650219089Spjd * This is either a normal write (not a repair), or it's 2651219089Spjd * a repair induced by the scrub thread, or it's a repair 2652219089Spjd * made by zil_claim() during spa_load() in the first txg. 2653219089Spjd * In the normal case, we commit the DTL change in the same 2654219089Spjd * txg as the block was born. In the scrub-induced repair 2655219089Spjd * case, we know that scrubs run in first-pass syncing context, 2656219089Spjd * so we commit the DTL change in spa_syncing_txg(spa). 2657219089Spjd * In the zil_claim() case, we commit in spa_first_txg(spa). 2658209962Smm * 2659209962Smm * We currently do not make DTL entries for failed spontaneous 2660209962Smm * self-healing writes triggered by normal (non-scrubbing) 2661209962Smm * reads, because we have no transactional context in which to 2662209962Smm * do so -- and it's not clear that it'd be desirable anyway. 2663209962Smm */ 2664209962Smm if (vd->vdev_ops->vdev_op_leaf) { 2665209962Smm uint64_t commit_txg = txg; 2666219089Spjd if (flags & ZIO_FLAG_SCAN_THREAD) { 2667209962Smm ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2668209962Smm ASSERT(spa_sync_pass(spa) == 1); 2669209962Smm vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2670219089Spjd commit_txg = spa_syncing_txg(spa); 2671219089Spjd } else if (spa->spa_claiming) { 2672219089Spjd ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2673219089Spjd commit_txg = spa_first_txg(spa); 2674209962Smm } 2675219089Spjd ASSERT(commit_txg >= spa_syncing_txg(spa)); 2676209962Smm if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2677168404Spjd return; 2678209962Smm for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2679209962Smm vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2680209962Smm vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2681168404Spjd } 2682209962Smm if (vd != rvd) 2683209962Smm vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2684168404Spjd } 2685168404Spjd} 2686168404Spjd 2687168404Spjd/* 2688219089Spjd * Update the in-core space usage stats for this vdev, its metaslab class, 2689219089Spjd * and the root vdev. 2690168404Spjd */ 2691168404Spjdvoid 2692219089Spjdvdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2693219089Spjd int64_t space_delta) 2694168404Spjd{ 2695168404Spjd int64_t dspace_delta = space_delta; 2696185029Spjd spa_t *spa = vd->vdev_spa; 2697185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2698219089Spjd metaslab_group_t *mg = vd->vdev_mg; 2699219089Spjd metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2700168404Spjd 2701185029Spjd ASSERT(vd == vd->vdev_top); 2702168404Spjd 2703185029Spjd /* 2704185029Spjd * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2705185029Spjd * factor. We must calculate this here and not at the root vdev 2706185029Spjd * because the root vdev's psize-to-asize is simply the max of its 2707185029Spjd * childrens', thus not accurate enough for us. 2708185029Spjd */ 2709185029Spjd ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2710213197Smm ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2711185029Spjd dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2712185029Spjd vd->vdev_deflate_ratio; 2713185029Spjd 2714185029Spjd mutex_enter(&vd->vdev_stat_lock); 2715219089Spjd vd->vdev_stat.vs_alloc += alloc_delta; 2716185029Spjd vd->vdev_stat.vs_space += space_delta; 2717185029Spjd vd->vdev_stat.vs_dspace += dspace_delta; 2718185029Spjd mutex_exit(&vd->vdev_stat_lock); 2719185029Spjd 2720219089Spjd if (mc == spa_normal_class(spa)) { 2721185029Spjd mutex_enter(&rvd->vdev_stat_lock); 2722219089Spjd rvd->vdev_stat.vs_alloc += alloc_delta; 2723185029Spjd rvd->vdev_stat.vs_space += space_delta; 2724185029Spjd rvd->vdev_stat.vs_dspace += dspace_delta; 2725185029Spjd mutex_exit(&rvd->vdev_stat_lock); 2726185029Spjd } 2727219089Spjd 2728219089Spjd if (mc != NULL) { 2729219089Spjd ASSERT(rvd == vd->vdev_parent); 2730219089Spjd ASSERT(vd->vdev_ms_count != 0); 2731219089Spjd 2732219089Spjd metaslab_class_space_update(mc, 2733219089Spjd alloc_delta, defer_delta, space_delta, dspace_delta); 2734219089Spjd } 2735168404Spjd} 2736168404Spjd 2737168404Spjd/* 2738168404Spjd * Mark a top-level vdev's config as dirty, placing it on the dirty list 2739168404Spjd * so that it will be written out next time the vdev configuration is synced. 2740168404Spjd * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2741168404Spjd */ 2742168404Spjdvoid 2743168404Spjdvdev_config_dirty(vdev_t *vd) 2744168404Spjd{ 2745168404Spjd spa_t *spa = vd->vdev_spa; 2746168404Spjd vdev_t *rvd = spa->spa_root_vdev; 2747168404Spjd int c; 2748168404Spjd 2749219089Spjd ASSERT(spa_writeable(spa)); 2750219089Spjd 2751168404Spjd /* 2752209962Smm * If this is an aux vdev (as with l2cache and spare devices), then we 2753209962Smm * update the vdev config manually and set the sync flag. 2754185029Spjd */ 2755185029Spjd if (vd->vdev_aux != NULL) { 2756185029Spjd spa_aux_vdev_t *sav = vd->vdev_aux; 2757185029Spjd nvlist_t **aux; 2758185029Spjd uint_t naux; 2759185029Spjd 2760185029Spjd for (c = 0; c < sav->sav_count; c++) { 2761185029Spjd if (sav->sav_vdevs[c] == vd) 2762185029Spjd break; 2763185029Spjd } 2764185029Spjd 2765185029Spjd if (c == sav->sav_count) { 2766185029Spjd /* 2767185029Spjd * We're being removed. There's nothing more to do. 2768185029Spjd */ 2769185029Spjd ASSERT(sav->sav_sync == B_TRUE); 2770185029Spjd return; 2771185029Spjd } 2772185029Spjd 2773185029Spjd sav->sav_sync = B_TRUE; 2774185029Spjd 2775209962Smm if (nvlist_lookup_nvlist_array(sav->sav_config, 2776209962Smm ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 2777209962Smm VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2778209962Smm ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 2779209962Smm } 2780185029Spjd 2781185029Spjd ASSERT(c < naux); 2782185029Spjd 2783185029Spjd /* 2784185029Spjd * Setting the nvlist in the middle if the array is a little 2785185029Spjd * sketchy, but it will work. 2786185029Spjd */ 2787185029Spjd nvlist_free(aux[c]); 2788219089Spjd aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2789185029Spjd 2790185029Spjd return; 2791185029Spjd } 2792185029Spjd 2793185029Spjd /* 2794185029Spjd * The dirty list is protected by the SCL_CONFIG lock. The caller 2795185029Spjd * must either hold SCL_CONFIG as writer, or must be the sync thread 2796185029Spjd * (which holds SCL_CONFIG as reader). There's only one sync thread, 2797168404Spjd * so this is sufficient to ensure mutual exclusion. 2798168404Spjd */ 2799185029Spjd ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2800185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2801185029Spjd spa_config_held(spa, SCL_CONFIG, RW_READER))); 2802168404Spjd 2803168404Spjd if (vd == rvd) { 2804168404Spjd for (c = 0; c < rvd->vdev_children; c++) 2805168404Spjd vdev_config_dirty(rvd->vdev_child[c]); 2806168404Spjd } else { 2807168404Spjd ASSERT(vd == vd->vdev_top); 2808168404Spjd 2809219089Spjd if (!list_link_active(&vd->vdev_config_dirty_node) && 2810219089Spjd !vd->vdev_ishole) 2811185029Spjd list_insert_head(&spa->spa_config_dirty_list, vd); 2812168404Spjd } 2813168404Spjd} 2814168404Spjd 2815168404Spjdvoid 2816168404Spjdvdev_config_clean(vdev_t *vd) 2817168404Spjd{ 2818168404Spjd spa_t *spa = vd->vdev_spa; 2819168404Spjd 2820185029Spjd ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2821185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2822185029Spjd spa_config_held(spa, SCL_CONFIG, RW_READER))); 2823168404Spjd 2824185029Spjd ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2825185029Spjd list_remove(&spa->spa_config_dirty_list, vd); 2826168404Spjd} 2827168404Spjd 2828185029Spjd/* 2829185029Spjd * Mark a top-level vdev's state as dirty, so that the next pass of 2830185029Spjd * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2831185029Spjd * the state changes from larger config changes because they require 2832185029Spjd * much less locking, and are often needed for administrative actions. 2833185029Spjd */ 2834168404Spjdvoid 2835185029Spjdvdev_state_dirty(vdev_t *vd) 2836185029Spjd{ 2837185029Spjd spa_t *spa = vd->vdev_spa; 2838185029Spjd 2839219089Spjd ASSERT(spa_writeable(spa)); 2840185029Spjd ASSERT(vd == vd->vdev_top); 2841185029Spjd 2842185029Spjd /* 2843185029Spjd * The state list is protected by the SCL_STATE lock. The caller 2844185029Spjd * must either hold SCL_STATE as writer, or must be the sync thread 2845185029Spjd * (which holds SCL_STATE as reader). There's only one sync thread, 2846185029Spjd * so this is sufficient to ensure mutual exclusion. 2847185029Spjd */ 2848185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2849185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2850185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2851185029Spjd 2852219089Spjd if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2853185029Spjd list_insert_head(&spa->spa_state_dirty_list, vd); 2854185029Spjd} 2855185029Spjd 2856185029Spjdvoid 2857185029Spjdvdev_state_clean(vdev_t *vd) 2858185029Spjd{ 2859185029Spjd spa_t *spa = vd->vdev_spa; 2860185029Spjd 2861185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2862185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2863185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2864185029Spjd 2865185029Spjd ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2866185029Spjd list_remove(&spa->spa_state_dirty_list, vd); 2867185029Spjd} 2868185029Spjd 2869185029Spjd/* 2870185029Spjd * Propagate vdev state up from children to parent. 2871185029Spjd */ 2872185029Spjdvoid 2873168404Spjdvdev_propagate_state(vdev_t *vd) 2874168404Spjd{ 2875209962Smm spa_t *spa = vd->vdev_spa; 2876209962Smm vdev_t *rvd = spa->spa_root_vdev; 2877168404Spjd int degraded = 0, faulted = 0; 2878168404Spjd int corrupted = 0; 2879168404Spjd vdev_t *child; 2880168404Spjd 2881185029Spjd if (vd->vdev_children > 0) { 2882219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 2883185029Spjd child = vd->vdev_child[c]; 2884168404Spjd 2885219089Spjd /* 2886219089Spjd * Don't factor holes into the decision. 2887219089Spjd */ 2888219089Spjd if (child->vdev_ishole) 2889219089Spjd continue; 2890219089Spjd 2891185029Spjd if (!vdev_readable(child) || 2892209962Smm (!vdev_writeable(child) && spa_writeable(spa))) { 2893185029Spjd /* 2894185029Spjd * Root special: if there is a top-level log 2895185029Spjd * device, treat the root vdev as if it were 2896185029Spjd * degraded. 2897185029Spjd */ 2898185029Spjd if (child->vdev_islog && vd == rvd) 2899185029Spjd degraded++; 2900185029Spjd else 2901185029Spjd faulted++; 2902185029Spjd } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 2903185029Spjd degraded++; 2904185029Spjd } 2905185029Spjd 2906185029Spjd if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 2907185029Spjd corrupted++; 2908185029Spjd } 2909185029Spjd 2910185029Spjd vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 2911185029Spjd 2912185029Spjd /* 2913185029Spjd * Root special: if there is a top-level vdev that cannot be 2914185029Spjd * opened due to corrupted metadata, then propagate the root 2915185029Spjd * vdev's aux state as 'corrupt' rather than 'insufficient 2916185029Spjd * replicas'. 2917185029Spjd */ 2918185029Spjd if (corrupted && vd == rvd && 2919185029Spjd rvd->vdev_state == VDEV_STATE_CANT_OPEN) 2920185029Spjd vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 2921185029Spjd VDEV_AUX_CORRUPT_DATA); 2922168404Spjd } 2923168404Spjd 2924185029Spjd if (vd->vdev_parent) 2925185029Spjd vdev_propagate_state(vd->vdev_parent); 2926168404Spjd} 2927168404Spjd 2928168404Spjd/* 2929168404Spjd * Set a vdev's state. If this is during an open, we don't update the parent 2930168404Spjd * state, because we're in the process of opening children depth-first. 2931168404Spjd * Otherwise, we propagate the change to the parent. 2932168404Spjd * 2933168404Spjd * If this routine places a device in a faulted state, an appropriate ereport is 2934168404Spjd * generated. 2935168404Spjd */ 2936168404Spjdvoid 2937168404Spjdvdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2938168404Spjd{ 2939168404Spjd uint64_t save_state; 2940185029Spjd spa_t *spa = vd->vdev_spa; 2941168404Spjd 2942168404Spjd if (state == vd->vdev_state) { 2943168404Spjd vd->vdev_stat.vs_aux = aux; 2944168404Spjd return; 2945168404Spjd } 2946168404Spjd 2947168404Spjd save_state = vd->vdev_state; 2948168404Spjd 2949168404Spjd vd->vdev_state = state; 2950168404Spjd vd->vdev_stat.vs_aux = aux; 2951168404Spjd 2952173373Spjd /* 2953173373Spjd * If we are setting the vdev state to anything but an open state, then 2954219089Spjd * always close the underlying device unless the device has requested 2955219089Spjd * a delayed close (i.e. we're about to remove or fault the device). 2956219089Spjd * Otherwise, we keep accessible but invalid devices open forever. 2957219089Spjd * We don't call vdev_close() itself, because that implies some extra 2958219089Spjd * checks (offline, etc) that we don't want here. This is limited to 2959219089Spjd * leaf devices, because otherwise closing the device will affect other 2960219089Spjd * children. 2961173373Spjd */ 2962219089Spjd if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 2963219089Spjd vd->vdev_ops->vdev_op_leaf) 2964173373Spjd vd->vdev_ops->vdev_op_close(vd); 2965173373Spjd 2966219089Spjd /* 2967219089Spjd * If we have brought this vdev back into service, we need 2968219089Spjd * to notify fmd so that it can gracefully repair any outstanding 2969219089Spjd * cases due to a missing device. We do this in all cases, even those 2970219089Spjd * that probably don't correlate to a repaired fault. This is sure to 2971219089Spjd * catch all cases, and we let the zfs-retire agent sort it out. If 2972219089Spjd * this is a transient state it's OK, as the retire agent will 2973219089Spjd * double-check the state of the vdev before repairing it. 2974219089Spjd */ 2975219089Spjd if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2976219089Spjd vd->vdev_prevstate != state) 2977219089Spjd zfs_post_state_change(spa, vd); 2978219089Spjd 2979185029Spjd if (vd->vdev_removed && 2980185029Spjd state == VDEV_STATE_CANT_OPEN && 2981185029Spjd (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 2982168404Spjd /* 2983185029Spjd * If the previous state is set to VDEV_STATE_REMOVED, then this 2984185029Spjd * device was previously marked removed and someone attempted to 2985185029Spjd * reopen it. If this failed due to a nonexistent device, then 2986185029Spjd * keep the device in the REMOVED state. We also let this be if 2987185029Spjd * it is one of our special test online cases, which is only 2988185029Spjd * attempting to online the device and shouldn't generate an FMA 2989185029Spjd * fault. 2990185029Spjd */ 2991185029Spjd vd->vdev_state = VDEV_STATE_REMOVED; 2992185029Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2993185029Spjd } else if (state == VDEV_STATE_REMOVED) { 2994185029Spjd vd->vdev_removed = B_TRUE; 2995185029Spjd } else if (state == VDEV_STATE_CANT_OPEN) { 2996185029Spjd /* 2997219089Spjd * If we fail to open a vdev during an import or recovery, we 2998219089Spjd * mark it as "not available", which signifies that it was 2999219089Spjd * never there to begin with. Failure to open such a device 3000219089Spjd * is not considered an error. 3001168404Spjd */ 3002219089Spjd if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3003219089Spjd spa_load_state(spa) == SPA_LOAD_RECOVER) && 3004168404Spjd vd->vdev_ops->vdev_op_leaf) 3005168404Spjd vd->vdev_not_present = 1; 3006168404Spjd 3007168404Spjd /* 3008168404Spjd * Post the appropriate ereport. If the 'prevstate' field is 3009168404Spjd * set to something other than VDEV_STATE_UNKNOWN, it indicates 3010168404Spjd * that this is part of a vdev_reopen(). In this case, we don't 3011168404Spjd * want to post the ereport if the device was already in the 3012168404Spjd * CANT_OPEN state beforehand. 3013185029Spjd * 3014185029Spjd * If the 'checkremove' flag is set, then this is an attempt to 3015185029Spjd * online the device in response to an insertion event. If we 3016185029Spjd * hit this case, then we have detected an insertion event for a 3017185029Spjd * faulted or offline device that wasn't in the removed state. 3018185029Spjd * In this scenario, we don't post an ereport because we are 3019185029Spjd * about to replace the device, or attempt an online with 3020185029Spjd * vdev_forcefault, which will generate the fault for us. 3021168404Spjd */ 3022185029Spjd if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 3023185029Spjd !vd->vdev_not_present && !vd->vdev_checkremove && 3024185029Spjd vd != spa->spa_root_vdev) { 3025168404Spjd const char *class; 3026168404Spjd 3027168404Spjd switch (aux) { 3028168404Spjd case VDEV_AUX_OPEN_FAILED: 3029168404Spjd class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3030168404Spjd break; 3031168404Spjd case VDEV_AUX_CORRUPT_DATA: 3032168404Spjd class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3033168404Spjd break; 3034168404Spjd case VDEV_AUX_NO_REPLICAS: 3035168404Spjd class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3036168404Spjd break; 3037168404Spjd case VDEV_AUX_BAD_GUID_SUM: 3038168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3039168404Spjd break; 3040168404Spjd case VDEV_AUX_TOO_SMALL: 3041168404Spjd class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3042168404Spjd break; 3043168404Spjd case VDEV_AUX_BAD_LABEL: 3044168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3045168404Spjd break; 3046168404Spjd default: 3047168404Spjd class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3048168404Spjd } 3049168404Spjd 3050185029Spjd zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3051168404Spjd } 3052185029Spjd 3053185029Spjd /* Erase any notion of persistent removed state */ 3054185029Spjd vd->vdev_removed = B_FALSE; 3055185029Spjd } else { 3056185029Spjd vd->vdev_removed = B_FALSE; 3057168404Spjd } 3058168404Spjd 3059209962Smm if (!isopen && vd->vdev_parent) 3060209962Smm vdev_propagate_state(vd->vdev_parent); 3061185029Spjd} 3062168404Spjd 3063185029Spjd/* 3064185029Spjd * Check the vdev configuration to ensure that it's capable of supporting 3065193163Sdfr * a root pool. 3066193163Sdfr * 3067193163Sdfr * On Solaris, we do not support RAID-Z or partial configuration. In 3068193163Sdfr * addition, only a single top-level vdev is allowed and none of the 3069193163Sdfr * leaves can be wholedisks. 3070193163Sdfr * 3071193163Sdfr * For FreeBSD, we can boot from any configuration. There is a 3072193163Sdfr * limitation that the boot filesystem must be either uncompressed or 3073193163Sdfr * compresses with lzjb compression but I'm not sure how to enforce 3074193163Sdfr * that here. 3075185029Spjd */ 3076185029Spjdboolean_t 3077185029Spjdvdev_is_bootable(vdev_t *vd) 3078185029Spjd{ 3079213197Smm#ifdef sun 3080185029Spjd if (!vd->vdev_ops->vdev_op_leaf) { 3081185029Spjd char *vdev_type = vd->vdev_ops->vdev_op_type; 3082185029Spjd 3083185029Spjd if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 3084185029Spjd vd->vdev_children > 1) { 3085185029Spjd return (B_FALSE); 3086185029Spjd } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 3087185029Spjd strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 3088185029Spjd return (B_FALSE); 3089185029Spjd } 3090185029Spjd } else if (vd->vdev_wholedisk == 1) { 3091185029Spjd return (B_FALSE); 3092185029Spjd } 3093185029Spjd 3094219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 3095185029Spjd if (!vdev_is_bootable(vd->vdev_child[c])) 3096185029Spjd return (B_FALSE); 3097185029Spjd } 3098213197Smm#endif /* sun */ 3099185029Spjd return (B_TRUE); 3100168404Spjd} 3101213197Smm 3102219089Spjd/* 3103219089Spjd * Load the state from the original vdev tree (ovd) which 3104219089Spjd * we've retrieved from the MOS config object. If the original 3105219089Spjd * vdev was offline or faulted then we transfer that state to the 3106219089Spjd * device in the current vdev tree (nvd). 3107219089Spjd */ 3108213197Smmvoid 3109219089Spjdvdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3110213197Smm{ 3111219089Spjd spa_t *spa = nvd->vdev_spa; 3112213197Smm 3113219089Spjd ASSERT(nvd->vdev_top->vdev_islog); 3114219089Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3115219089Spjd ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3116213197Smm 3117219089Spjd for (int c = 0; c < nvd->vdev_children; c++) 3118219089Spjd vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3119213197Smm 3120219089Spjd if (nvd->vdev_ops->vdev_op_leaf) { 3121213197Smm /* 3122219089Spjd * Restore the persistent vdev state 3123213197Smm */ 3124219089Spjd nvd->vdev_offline = ovd->vdev_offline; 3125219089Spjd nvd->vdev_faulted = ovd->vdev_faulted; 3126219089Spjd nvd->vdev_degraded = ovd->vdev_degraded; 3127219089Spjd nvd->vdev_removed = ovd->vdev_removed; 3128213197Smm } 3129213197Smm} 3130219089Spjd 3131219089Spjd/* 3132219089Spjd * Determine if a log device has valid content. If the vdev was 3133219089Spjd * removed or faulted in the MOS config then we know that 3134219089Spjd * the content on the log device has already been written to the pool. 3135219089Spjd */ 3136219089Spjdboolean_t 3137219089Spjdvdev_log_state_valid(vdev_t *vd) 3138219089Spjd{ 3139219089Spjd if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 3140219089Spjd !vd->vdev_removed) 3141219089Spjd return (B_TRUE); 3142219089Spjd 3143219089Spjd for (int c = 0; c < vd->vdev_children; c++) 3144219089Spjd if (vdev_log_state_valid(vd->vdev_child[c])) 3145219089Spjd return (B_TRUE); 3146219089Spjd 3147219089Spjd return (B_FALSE); 3148219089Spjd} 3149219089Spjd 3150219089Spjd/* 3151219089Spjd * Expand a vdev if possible. 3152219089Spjd */ 3153219089Spjdvoid 3154219089Spjdvdev_expand(vdev_t *vd, uint64_t txg) 3155219089Spjd{ 3156219089Spjd ASSERT(vd->vdev_top == vd); 3157219089Spjd ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3158219089Spjd 3159219089Spjd if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3160219089Spjd VERIFY(vdev_metaslab_init(vd, txg) == 0); 3161219089Spjd vdev_config_dirty(vd); 3162219089Spjd } 3163219089Spjd} 3164219089Spjd 3165219089Spjd/* 3166219089Spjd * Split a vdev. 3167219089Spjd */ 3168219089Spjdvoid 3169219089Spjdvdev_split(vdev_t *vd) 3170219089Spjd{ 3171219089Spjd vdev_t *cvd, *pvd = vd->vdev_parent; 3172219089Spjd 3173219089Spjd vdev_remove_child(pvd, vd); 3174219089Spjd vdev_compact_children(pvd); 3175219089Spjd 3176219089Spjd cvd = pvd->vdev_child[0]; 3177219089Spjd if (pvd->vdev_children == 1) { 3178219089Spjd vdev_remove_parent(cvd); 3179219089Spjd cvd->vdev_splitting = B_TRUE; 3180219089Spjd } 3181219089Spjd vdev_propagate_state(cvd); 3182219089Spjd} 3183248369Smm 3184248369Smmvoid 3185248369Smmvdev_deadman(vdev_t *vd) 3186248369Smm{ 3187248369Smm for (int c = 0; c < vd->vdev_children; c++) { 3188248369Smm vdev_t *cvd = vd->vdev_child[c]; 3189248369Smm 3190248369Smm vdev_deadman(cvd); 3191248369Smm } 3192248369Smm 3193248369Smm if (vd->vdev_ops->vdev_op_leaf) { 3194248369Smm vdev_queue_t *vq = &vd->vdev_queue; 3195248369Smm 3196248369Smm mutex_enter(&vq->vq_lock); 3197248369Smm if (avl_numnodes(&vq->vq_pending_tree) > 0) { 3198248369Smm spa_t *spa = vd->vdev_spa; 3199248369Smm zio_t *fio; 3200248369Smm uint64_t delta; 3201248369Smm 3202248369Smm /* 3203248369Smm * Look at the head of all the pending queues, 3204248369Smm * if any I/O has been outstanding for longer than 3205248369Smm * the spa_deadman_synctime we panic the system. 3206248369Smm */ 3207248369Smm fio = avl_first(&vq->vq_pending_tree); 3208249643Smm delta = gethrtime() - fio->io_timestamp; 3209249643Smm if (delta > spa_deadman_synctime(spa)) { 3210249643Smm zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3211249643Smm "delta %lluns, last io %lluns", 3212248369Smm fio->io_timestamp, delta, 3213248369Smm vq->vq_io_complete_ts); 3214248369Smm fm_panic("I/O to pool '%s' appears to be " 3215248369Smm "hung on vdev guid %llu at '%s'.", 3216248369Smm spa_name(spa), 3217248369Smm (long long unsigned int) vd->vdev_guid, 3218248369Smm vd->vdev_path); 3219248369Smm } 3220248369Smm } 3221248369Smm mutex_exit(&vq->vq_lock); 3222248369Smm } 3223248369Smm} 3224