vdev.c revision 231141
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. 25229578Smm * Copyright (c) 2011 by Delphix. All rights reserved. 26168404Spjd */ 27168404Spjd 28168404Spjd#include <sys/zfs_context.h> 29168404Spjd#include <sys/fm/fs/zfs.h> 30168404Spjd#include <sys/spa.h> 31168404Spjd#include <sys/spa_impl.h> 32168404Spjd#include <sys/dmu.h> 33168404Spjd#include <sys/dmu_tx.h> 34168404Spjd#include <sys/vdev_impl.h> 35168404Spjd#include <sys/uberblock_impl.h> 36168404Spjd#include <sys/metaslab.h> 37168404Spjd#include <sys/metaslab_impl.h> 38168404Spjd#include <sys/space_map.h> 39168404Spjd#include <sys/zio.h> 40168404Spjd#include <sys/zap.h> 41168404Spjd#include <sys/fs/zfs.h> 42185029Spjd#include <sys/arc.h> 43213197Smm#include <sys/zil.h> 44219089Spjd#include <sys/dsl_scan.h> 45168404Spjd 46168404SpjdSYSCTL_DECL(_vfs_zfs); 47168404SpjdSYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); 48168404Spjd 49168404Spjd/* 50168404Spjd * Virtual device management. 51168404Spjd */ 52168404Spjd 53168404Spjdstatic vdev_ops_t *vdev_ops_table[] = { 54168404Spjd &vdev_root_ops, 55168404Spjd &vdev_raidz_ops, 56168404Spjd &vdev_mirror_ops, 57168404Spjd &vdev_replacing_ops, 58168404Spjd &vdev_spare_ops, 59168404Spjd#ifdef _KERNEL 60168404Spjd &vdev_geom_ops, 61168404Spjd#else 62168404Spjd &vdev_disk_ops, 63185029Spjd#endif 64168404Spjd &vdev_file_ops, 65168404Spjd &vdev_missing_ops, 66219089Spjd &vdev_hole_ops, 67168404Spjd NULL 68168404Spjd}; 69168404Spjd 70185029Spjd/* maximum scrub/resilver I/O queue per leaf vdev */ 71185029Spjdint zfs_scrub_limit = 10; 72168404Spjd 73185029SpjdTUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit); 74185029SpjdSYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0, 75185029Spjd "Maximum scrub/resilver I/O queue"); 76185029Spjd 77168404Spjd/* 78168404Spjd * Given a vdev type, return the appropriate ops vector. 79168404Spjd */ 80168404Spjdstatic vdev_ops_t * 81168404Spjdvdev_getops(const char *type) 82168404Spjd{ 83168404Spjd vdev_ops_t *ops, **opspp; 84168404Spjd 85168404Spjd for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 86168404Spjd if (strcmp(ops->vdev_op_type, type) == 0) 87168404Spjd break; 88168404Spjd 89168404Spjd return (ops); 90168404Spjd} 91168404Spjd 92168404Spjd/* 93168404Spjd * Default asize function: return the MAX of psize with the asize of 94168404Spjd * all children. This is what's used by anything other than RAID-Z. 95168404Spjd */ 96168404Spjduint64_t 97168404Spjdvdev_default_asize(vdev_t *vd, uint64_t psize) 98168404Spjd{ 99168404Spjd uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 100168404Spjd uint64_t csize; 101168404Spjd 102219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 103168404Spjd csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 104168404Spjd asize = MAX(asize, csize); 105168404Spjd } 106168404Spjd 107168404Spjd return (asize); 108168404Spjd} 109168404Spjd 110168404Spjd/* 111219089Spjd * Get the minimum allocatable size. We define the allocatable size as 112219089Spjd * the vdev's asize rounded to the nearest metaslab. This allows us to 113219089Spjd * replace or attach devices which don't have the same physical size but 114219089Spjd * can still satisfy the same number of allocations. 115168404Spjd */ 116168404Spjduint64_t 117219089Spjdvdev_get_min_asize(vdev_t *vd) 118168404Spjd{ 119219089Spjd vdev_t *pvd = vd->vdev_parent; 120168404Spjd 121219089Spjd /* 122219089Spjd * The our parent is NULL (inactive spare or cache) or is the root, 123219089Spjd * just return our own asize. 124219089Spjd */ 125219089Spjd if (pvd == NULL) 126219089Spjd return (vd->vdev_asize); 127168404Spjd 128168404Spjd /* 129219089Spjd * The top-level vdev just returns the allocatable size rounded 130219089Spjd * to the nearest metaslab. 131168404Spjd */ 132219089Spjd if (vd == vd->vdev_top) 133219089Spjd return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 134168404Spjd 135219089Spjd /* 136219089Spjd * The allocatable space for a raidz vdev is N * sizeof(smallest child), 137219089Spjd * so each child must provide at least 1/Nth of its asize. 138219089Spjd */ 139219089Spjd if (pvd->vdev_ops == &vdev_raidz_ops) 140219089Spjd return (pvd->vdev_min_asize / pvd->vdev_children); 141168404Spjd 142219089Spjd return (pvd->vdev_min_asize); 143219089Spjd} 144168404Spjd 145219089Spjdvoid 146219089Spjdvdev_set_min_asize(vdev_t *vd) 147219089Spjd{ 148219089Spjd vd->vdev_min_asize = vdev_get_min_asize(vd); 149219089Spjd 150219089Spjd for (int c = 0; c < vd->vdev_children; c++) 151219089Spjd vdev_set_min_asize(vd->vdev_child[c]); 152168404Spjd} 153168404Spjd 154168404Spjdvdev_t * 155168404Spjdvdev_lookup_top(spa_t *spa, uint64_t vdev) 156168404Spjd{ 157168404Spjd vdev_t *rvd = spa->spa_root_vdev; 158168404Spjd 159185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 160185029Spjd 161185029Spjd if (vdev < rvd->vdev_children) { 162185029Spjd ASSERT(rvd->vdev_child[vdev] != NULL); 163168404Spjd return (rvd->vdev_child[vdev]); 164185029Spjd } 165168404Spjd 166168404Spjd return (NULL); 167168404Spjd} 168168404Spjd 169168404Spjdvdev_t * 170168404Spjdvdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 171168404Spjd{ 172168404Spjd vdev_t *mvd; 173168404Spjd 174168404Spjd if (vd->vdev_guid == guid) 175168404Spjd return (vd); 176168404Spjd 177219089Spjd for (int c = 0; c < vd->vdev_children; c++) 178168404Spjd if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 179168404Spjd NULL) 180168404Spjd return (mvd); 181168404Spjd 182168404Spjd return (NULL); 183168404Spjd} 184168404Spjd 185168404Spjdvoid 186168404Spjdvdev_add_child(vdev_t *pvd, vdev_t *cvd) 187168404Spjd{ 188168404Spjd size_t oldsize, newsize; 189168404Spjd uint64_t id = cvd->vdev_id; 190168404Spjd vdev_t **newchild; 191168404Spjd 192185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 193168404Spjd ASSERT(cvd->vdev_parent == NULL); 194168404Spjd 195168404Spjd cvd->vdev_parent = pvd; 196168404Spjd 197168404Spjd if (pvd == NULL) 198168404Spjd return; 199168404Spjd 200168404Spjd ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 201168404Spjd 202168404Spjd oldsize = pvd->vdev_children * sizeof (vdev_t *); 203168404Spjd pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 204168404Spjd newsize = pvd->vdev_children * sizeof (vdev_t *); 205168404Spjd 206168404Spjd newchild = kmem_zalloc(newsize, KM_SLEEP); 207168404Spjd if (pvd->vdev_child != NULL) { 208168404Spjd bcopy(pvd->vdev_child, newchild, oldsize); 209168404Spjd kmem_free(pvd->vdev_child, oldsize); 210168404Spjd } 211168404Spjd 212168404Spjd pvd->vdev_child = newchild; 213168404Spjd pvd->vdev_child[id] = cvd; 214168404Spjd 215168404Spjd cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 216168404Spjd ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 217168404Spjd 218168404Spjd /* 219168404Spjd * Walk up all ancestors to update guid sum. 220168404Spjd */ 221168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 222168404Spjd pvd->vdev_guid_sum += cvd->vdev_guid_sum; 223168404Spjd} 224168404Spjd 225168404Spjdvoid 226168404Spjdvdev_remove_child(vdev_t *pvd, vdev_t *cvd) 227168404Spjd{ 228168404Spjd int c; 229168404Spjd uint_t id = cvd->vdev_id; 230168404Spjd 231168404Spjd ASSERT(cvd->vdev_parent == pvd); 232168404Spjd 233168404Spjd if (pvd == NULL) 234168404Spjd return; 235168404Spjd 236168404Spjd ASSERT(id < pvd->vdev_children); 237168404Spjd ASSERT(pvd->vdev_child[id] == cvd); 238168404Spjd 239168404Spjd pvd->vdev_child[id] = NULL; 240168404Spjd cvd->vdev_parent = NULL; 241168404Spjd 242168404Spjd for (c = 0; c < pvd->vdev_children; c++) 243168404Spjd if (pvd->vdev_child[c]) 244168404Spjd break; 245168404Spjd 246168404Spjd if (c == pvd->vdev_children) { 247168404Spjd kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 248168404Spjd pvd->vdev_child = NULL; 249168404Spjd pvd->vdev_children = 0; 250168404Spjd } 251168404Spjd 252168404Spjd /* 253168404Spjd * Walk up all ancestors to update guid sum. 254168404Spjd */ 255168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 256168404Spjd pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 257168404Spjd} 258168404Spjd 259168404Spjd/* 260168404Spjd * Remove any holes in the child array. 261168404Spjd */ 262168404Spjdvoid 263168404Spjdvdev_compact_children(vdev_t *pvd) 264168404Spjd{ 265168404Spjd vdev_t **newchild, *cvd; 266168404Spjd int oldc = pvd->vdev_children; 267219089Spjd int newc; 268168404Spjd 269185029Spjd ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 270168404Spjd 271219089Spjd for (int c = newc = 0; c < oldc; c++) 272168404Spjd if (pvd->vdev_child[c]) 273168404Spjd newc++; 274168404Spjd 275168404Spjd newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 276168404Spjd 277219089Spjd for (int c = newc = 0; c < oldc; c++) { 278168404Spjd if ((cvd = pvd->vdev_child[c]) != NULL) { 279168404Spjd newchild[newc] = cvd; 280168404Spjd cvd->vdev_id = newc++; 281168404Spjd } 282168404Spjd } 283168404Spjd 284168404Spjd kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 285168404Spjd pvd->vdev_child = newchild; 286168404Spjd pvd->vdev_children = newc; 287168404Spjd} 288168404Spjd 289168404Spjd/* 290168404Spjd * Allocate and minimally initialize a vdev_t. 291168404Spjd */ 292219089Spjdvdev_t * 293168404Spjdvdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 294168404Spjd{ 295168404Spjd vdev_t *vd; 296168404Spjd 297168404Spjd vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 298168404Spjd 299168404Spjd if (spa->spa_root_vdev == NULL) { 300168404Spjd ASSERT(ops == &vdev_root_ops); 301168404Spjd spa->spa_root_vdev = vd; 302229578Smm spa->spa_load_guid = spa_generate_guid(NULL); 303168404Spjd } 304168404Spjd 305219089Spjd if (guid == 0 && ops != &vdev_hole_ops) { 306168404Spjd if (spa->spa_root_vdev == vd) { 307168404Spjd /* 308168404Spjd * The root vdev's guid will also be the pool guid, 309168404Spjd * which must be unique among all pools. 310168404Spjd */ 311219089Spjd guid = spa_generate_guid(NULL); 312168404Spjd } else { 313168404Spjd /* 314168404Spjd * Any other vdev's guid must be unique within the pool. 315168404Spjd */ 316219089Spjd guid = spa_generate_guid(spa); 317168404Spjd } 318168404Spjd ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 319168404Spjd } 320168404Spjd 321168404Spjd vd->vdev_spa = spa; 322168404Spjd vd->vdev_id = id; 323168404Spjd vd->vdev_guid = guid; 324168404Spjd vd->vdev_guid_sum = guid; 325168404Spjd vd->vdev_ops = ops; 326168404Spjd vd->vdev_state = VDEV_STATE_CLOSED; 327219089Spjd vd->vdev_ishole = (ops == &vdev_hole_ops); 328168404Spjd 329168404Spjd mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 330168404Spjd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 331185029Spjd mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 332209962Smm for (int t = 0; t < DTL_TYPES; t++) { 333209962Smm space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 334209962Smm &vd->vdev_dtl_lock); 335209962Smm } 336168404Spjd txg_list_create(&vd->vdev_ms_list, 337168404Spjd offsetof(struct metaslab, ms_txg_node)); 338168404Spjd txg_list_create(&vd->vdev_dtl_list, 339168404Spjd offsetof(struct vdev, vdev_dtl_node)); 340168404Spjd vd->vdev_stat.vs_timestamp = gethrtime(); 341185029Spjd vdev_queue_init(vd); 342185029Spjd vdev_cache_init(vd); 343168404Spjd 344168404Spjd return (vd); 345168404Spjd} 346168404Spjd 347168404Spjd/* 348168404Spjd * Allocate a new vdev. The 'alloctype' is used to control whether we are 349168404Spjd * creating a new vdev or loading an existing one - the behavior is slightly 350168404Spjd * different for each case. 351168404Spjd */ 352168404Spjdint 353168404Spjdvdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 354168404Spjd int alloctype) 355168404Spjd{ 356168404Spjd vdev_ops_t *ops; 357168404Spjd char *type; 358185029Spjd uint64_t guid = 0, islog, nparity; 359168404Spjd vdev_t *vd; 360168404Spjd 361185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 362168404Spjd 363168404Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 364168404Spjd return (EINVAL); 365168404Spjd 366168404Spjd if ((ops = vdev_getops(type)) == NULL) 367168404Spjd return (EINVAL); 368168404Spjd 369168404Spjd /* 370168404Spjd * If this is a load, get the vdev guid from the nvlist. 371168404Spjd * Otherwise, vdev_alloc_common() will generate one for us. 372168404Spjd */ 373168404Spjd if (alloctype == VDEV_ALLOC_LOAD) { 374168404Spjd uint64_t label_id; 375168404Spjd 376168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 377168404Spjd label_id != id) 378168404Spjd return (EINVAL); 379168404Spjd 380168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 381168404Spjd return (EINVAL); 382168404Spjd } else if (alloctype == VDEV_ALLOC_SPARE) { 383168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 384168404Spjd return (EINVAL); 385185029Spjd } else if (alloctype == VDEV_ALLOC_L2CACHE) { 386185029Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 387185029Spjd return (EINVAL); 388219089Spjd } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 389219089Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 390219089Spjd return (EINVAL); 391168404Spjd } 392168404Spjd 393168404Spjd /* 394168404Spjd * The first allocated vdev must be of type 'root'. 395168404Spjd */ 396168404Spjd if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 397168404Spjd return (EINVAL); 398168404Spjd 399185029Spjd /* 400185029Spjd * Determine whether we're a log vdev. 401185029Spjd */ 402185029Spjd islog = 0; 403185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 404185029Spjd if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 405185029Spjd return (ENOTSUP); 406168404Spjd 407219089Spjd if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 408219089Spjd return (ENOTSUP); 409219089Spjd 410168404Spjd /* 411185029Spjd * Set the nparity property for RAID-Z vdevs. 412168404Spjd */ 413185029Spjd nparity = -1ULL; 414168404Spjd if (ops == &vdev_raidz_ops) { 415168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 416185029Spjd &nparity) == 0) { 417219089Spjd if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 418168404Spjd return (EINVAL); 419168404Spjd /* 420219089Spjd * Previous versions could only support 1 or 2 parity 421219089Spjd * device. 422168404Spjd */ 423219089Spjd if (nparity > 1 && 424219089Spjd spa_version(spa) < SPA_VERSION_RAIDZ2) 425168404Spjd return (ENOTSUP); 426219089Spjd if (nparity > 2 && 427219089Spjd spa_version(spa) < SPA_VERSION_RAIDZ3) 428219089Spjd return (ENOTSUP); 429168404Spjd } else { 430168404Spjd /* 431168404Spjd * We require the parity to be specified for SPAs that 432168404Spjd * support multiple parity levels. 433168404Spjd */ 434219089Spjd if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 435168404Spjd return (EINVAL); 436168404Spjd /* 437168404Spjd * Otherwise, we default to 1 parity device for RAID-Z. 438168404Spjd */ 439185029Spjd nparity = 1; 440168404Spjd } 441168404Spjd } else { 442185029Spjd nparity = 0; 443168404Spjd } 444185029Spjd ASSERT(nparity != -1ULL); 445168404Spjd 446185029Spjd vd = vdev_alloc_common(spa, id, guid, ops); 447185029Spjd 448185029Spjd vd->vdev_islog = islog; 449185029Spjd vd->vdev_nparity = nparity; 450185029Spjd 451185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 452185029Spjd vd->vdev_path = spa_strdup(vd->vdev_path); 453185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 454185029Spjd vd->vdev_devid = spa_strdup(vd->vdev_devid); 455185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 456185029Spjd &vd->vdev_physpath) == 0) 457185029Spjd vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 458209962Smm if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 459209962Smm vd->vdev_fru = spa_strdup(vd->vdev_fru); 460185029Spjd 461168404Spjd /* 462168404Spjd * Set the whole_disk property. If it's not specified, leave the value 463168404Spjd * as -1. 464168404Spjd */ 465168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 466168404Spjd &vd->vdev_wholedisk) != 0) 467168404Spjd vd->vdev_wholedisk = -1ULL; 468168404Spjd 469168404Spjd /* 470168404Spjd * Look for the 'not present' flag. This will only be set if the device 471168404Spjd * was not present at the time of import. 472168404Spjd */ 473209962Smm (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 474209962Smm &vd->vdev_not_present); 475168404Spjd 476168404Spjd /* 477168404Spjd * Get the alignment requirement. 478168404Spjd */ 479168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 480168404Spjd 481168404Spjd /* 482219089Spjd * Retrieve the vdev creation time. 483219089Spjd */ 484219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 485219089Spjd &vd->vdev_crtxg); 486219089Spjd 487219089Spjd /* 488168404Spjd * If we're a top-level vdev, try to load the allocation parameters. 489168404Spjd */ 490219089Spjd if (parent && !parent->vdev_parent && 491219089Spjd (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 492168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 493168404Spjd &vd->vdev_ms_array); 494168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 495168404Spjd &vd->vdev_ms_shift); 496168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 497168404Spjd &vd->vdev_asize); 498219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 499219089Spjd &vd->vdev_removing); 500168404Spjd } 501168404Spjd 502231141Smm if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 503219089Spjd ASSERT(alloctype == VDEV_ALLOC_LOAD || 504219089Spjd alloctype == VDEV_ALLOC_ADD || 505219089Spjd alloctype == VDEV_ALLOC_SPLIT || 506219089Spjd alloctype == VDEV_ALLOC_ROOTPOOL); 507219089Spjd vd->vdev_mg = metaslab_group_create(islog ? 508219089Spjd spa_log_class(spa) : spa_normal_class(spa), vd); 509219089Spjd } 510219089Spjd 511168404Spjd /* 512185029Spjd * If we're a leaf vdev, try to load the DTL object and other state. 513168404Spjd */ 514185029Spjd if (vd->vdev_ops->vdev_op_leaf && 515219089Spjd (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 516219089Spjd alloctype == VDEV_ALLOC_ROOTPOOL)) { 517185029Spjd if (alloctype == VDEV_ALLOC_LOAD) { 518185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 519209962Smm &vd->vdev_dtl_smo.smo_object); 520185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 521185029Spjd &vd->vdev_unspare); 522185029Spjd } 523219089Spjd 524219089Spjd if (alloctype == VDEV_ALLOC_ROOTPOOL) { 525219089Spjd uint64_t spare = 0; 526219089Spjd 527219089Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 528219089Spjd &spare) == 0 && spare) 529219089Spjd spa_spare_add(vd); 530219089Spjd } 531219089Spjd 532168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 533168404Spjd &vd->vdev_offline); 534185029Spjd 535219089Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, 536219089Spjd &vd->vdev_resilvering); 537219089Spjd 538185029Spjd /* 539185029Spjd * When importing a pool, we want to ignore the persistent fault 540185029Spjd * state, as the diagnosis made on another system may not be 541219089Spjd * valid in the current context. Local vdevs will 542219089Spjd * remain in the faulted state. 543185029Spjd */ 544219089Spjd if (spa_load_state(spa) == SPA_LOAD_OPEN) { 545185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 546185029Spjd &vd->vdev_faulted); 547185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 548185029Spjd &vd->vdev_degraded); 549185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 550185029Spjd &vd->vdev_removed); 551219089Spjd 552219089Spjd if (vd->vdev_faulted || vd->vdev_degraded) { 553219089Spjd char *aux; 554219089Spjd 555219089Spjd vd->vdev_label_aux = 556219089Spjd VDEV_AUX_ERR_EXCEEDED; 557219089Spjd if (nvlist_lookup_string(nv, 558219089Spjd ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 559219089Spjd strcmp(aux, "external") == 0) 560219089Spjd vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 561219089Spjd } 562185029Spjd } 563168404Spjd } 564168404Spjd 565168404Spjd /* 566168404Spjd * Add ourselves to the parent's list of children. 567168404Spjd */ 568168404Spjd vdev_add_child(parent, vd); 569168404Spjd 570168404Spjd *vdp = vd; 571168404Spjd 572168404Spjd return (0); 573168404Spjd} 574168404Spjd 575168404Spjdvoid 576168404Spjdvdev_free(vdev_t *vd) 577168404Spjd{ 578185029Spjd spa_t *spa = vd->vdev_spa; 579168404Spjd 580168404Spjd /* 581168404Spjd * vdev_free() implies closing the vdev first. This is simpler than 582168404Spjd * trying to ensure complicated semantics for all callers. 583168404Spjd */ 584168404Spjd vdev_close(vd); 585168404Spjd 586185029Spjd ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 587219089Spjd ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 588168404Spjd 589168404Spjd /* 590168404Spjd * Free all children. 591168404Spjd */ 592219089Spjd for (int c = 0; c < vd->vdev_children; c++) 593168404Spjd vdev_free(vd->vdev_child[c]); 594168404Spjd 595168404Spjd ASSERT(vd->vdev_child == NULL); 596168404Spjd ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 597168404Spjd 598168404Spjd /* 599168404Spjd * Discard allocation state. 600168404Spjd */ 601219089Spjd if (vd->vdev_mg != NULL) { 602168404Spjd vdev_metaslab_fini(vd); 603219089Spjd metaslab_group_destroy(vd->vdev_mg); 604219089Spjd } 605168404Spjd 606168404Spjd ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 607168404Spjd ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 608168404Spjd ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 609168404Spjd 610168404Spjd /* 611168404Spjd * Remove this vdev from its parent's child list. 612168404Spjd */ 613168404Spjd vdev_remove_child(vd->vdev_parent, vd); 614168404Spjd 615168404Spjd ASSERT(vd->vdev_parent == NULL); 616168404Spjd 617185029Spjd /* 618185029Spjd * Clean up vdev structure. 619185029Spjd */ 620185029Spjd vdev_queue_fini(vd); 621185029Spjd vdev_cache_fini(vd); 622185029Spjd 623185029Spjd if (vd->vdev_path) 624185029Spjd spa_strfree(vd->vdev_path); 625185029Spjd if (vd->vdev_devid) 626185029Spjd spa_strfree(vd->vdev_devid); 627185029Spjd if (vd->vdev_physpath) 628185029Spjd spa_strfree(vd->vdev_physpath); 629209962Smm if (vd->vdev_fru) 630209962Smm spa_strfree(vd->vdev_fru); 631185029Spjd 632185029Spjd if (vd->vdev_isspare) 633185029Spjd spa_spare_remove(vd); 634185029Spjd if (vd->vdev_isl2cache) 635185029Spjd spa_l2cache_remove(vd); 636185029Spjd 637185029Spjd txg_list_destroy(&vd->vdev_ms_list); 638185029Spjd txg_list_destroy(&vd->vdev_dtl_list); 639209962Smm 640185029Spjd mutex_enter(&vd->vdev_dtl_lock); 641209962Smm for (int t = 0; t < DTL_TYPES; t++) { 642209962Smm space_map_unload(&vd->vdev_dtl[t]); 643209962Smm space_map_destroy(&vd->vdev_dtl[t]); 644209962Smm } 645185029Spjd mutex_exit(&vd->vdev_dtl_lock); 646209962Smm 647185029Spjd mutex_destroy(&vd->vdev_dtl_lock); 648185029Spjd mutex_destroy(&vd->vdev_stat_lock); 649185029Spjd mutex_destroy(&vd->vdev_probe_lock); 650185029Spjd 651185029Spjd if (vd == spa->spa_root_vdev) 652185029Spjd spa->spa_root_vdev = NULL; 653185029Spjd 654185029Spjd kmem_free(vd, sizeof (vdev_t)); 655168404Spjd} 656168404Spjd 657168404Spjd/* 658168404Spjd * Transfer top-level vdev state from svd to tvd. 659168404Spjd */ 660168404Spjdstatic void 661168404Spjdvdev_top_transfer(vdev_t *svd, vdev_t *tvd) 662168404Spjd{ 663168404Spjd spa_t *spa = svd->vdev_spa; 664168404Spjd metaslab_t *msp; 665168404Spjd vdev_t *vd; 666168404Spjd int t; 667168404Spjd 668168404Spjd ASSERT(tvd == tvd->vdev_top); 669168404Spjd 670168404Spjd tvd->vdev_ms_array = svd->vdev_ms_array; 671168404Spjd tvd->vdev_ms_shift = svd->vdev_ms_shift; 672168404Spjd tvd->vdev_ms_count = svd->vdev_ms_count; 673168404Spjd 674168404Spjd svd->vdev_ms_array = 0; 675168404Spjd svd->vdev_ms_shift = 0; 676168404Spjd svd->vdev_ms_count = 0; 677168404Spjd 678231141Smm if (tvd->vdev_mg) 679231141Smm ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 680168404Spjd tvd->vdev_mg = svd->vdev_mg; 681168404Spjd tvd->vdev_ms = svd->vdev_ms; 682168404Spjd 683168404Spjd svd->vdev_mg = NULL; 684168404Spjd svd->vdev_ms = NULL; 685168404Spjd 686168404Spjd if (tvd->vdev_mg != NULL) 687168404Spjd tvd->vdev_mg->mg_vd = tvd; 688168404Spjd 689168404Spjd tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 690168404Spjd tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 691168404Spjd tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 692168404Spjd 693168404Spjd svd->vdev_stat.vs_alloc = 0; 694168404Spjd svd->vdev_stat.vs_space = 0; 695168404Spjd svd->vdev_stat.vs_dspace = 0; 696168404Spjd 697168404Spjd for (t = 0; t < TXG_SIZE; t++) { 698168404Spjd while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 699168404Spjd (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 700168404Spjd while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 701168404Spjd (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 702168404Spjd if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 703168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 704168404Spjd } 705168404Spjd 706185029Spjd if (list_link_active(&svd->vdev_config_dirty_node)) { 707168404Spjd vdev_config_clean(svd); 708168404Spjd vdev_config_dirty(tvd); 709168404Spjd } 710168404Spjd 711185029Spjd if (list_link_active(&svd->vdev_state_dirty_node)) { 712185029Spjd vdev_state_clean(svd); 713185029Spjd vdev_state_dirty(tvd); 714185029Spjd } 715168404Spjd 716168404Spjd tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 717168404Spjd svd->vdev_deflate_ratio = 0; 718185029Spjd 719185029Spjd tvd->vdev_islog = svd->vdev_islog; 720185029Spjd svd->vdev_islog = 0; 721168404Spjd} 722168404Spjd 723168404Spjdstatic void 724168404Spjdvdev_top_update(vdev_t *tvd, vdev_t *vd) 725168404Spjd{ 726168404Spjd if (vd == NULL) 727168404Spjd return; 728168404Spjd 729168404Spjd vd->vdev_top = tvd; 730168404Spjd 731219089Spjd for (int c = 0; c < vd->vdev_children; c++) 732168404Spjd vdev_top_update(tvd, vd->vdev_child[c]); 733168404Spjd} 734168404Spjd 735168404Spjd/* 736168404Spjd * Add a mirror/replacing vdev above an existing vdev. 737168404Spjd */ 738168404Spjdvdev_t * 739168404Spjdvdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 740168404Spjd{ 741168404Spjd spa_t *spa = cvd->vdev_spa; 742168404Spjd vdev_t *pvd = cvd->vdev_parent; 743168404Spjd vdev_t *mvd; 744168404Spjd 745185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 746168404Spjd 747168404Spjd mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 748168404Spjd 749168404Spjd mvd->vdev_asize = cvd->vdev_asize; 750219089Spjd mvd->vdev_min_asize = cvd->vdev_min_asize; 751168404Spjd mvd->vdev_ashift = cvd->vdev_ashift; 752168404Spjd mvd->vdev_state = cvd->vdev_state; 753219089Spjd mvd->vdev_crtxg = cvd->vdev_crtxg; 754168404Spjd 755168404Spjd vdev_remove_child(pvd, cvd); 756168404Spjd vdev_add_child(pvd, mvd); 757168404Spjd cvd->vdev_id = mvd->vdev_children; 758168404Spjd vdev_add_child(mvd, cvd); 759168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 760168404Spjd 761168404Spjd if (mvd == mvd->vdev_top) 762168404Spjd vdev_top_transfer(cvd, mvd); 763168404Spjd 764168404Spjd return (mvd); 765168404Spjd} 766168404Spjd 767168404Spjd/* 768168404Spjd * Remove a 1-way mirror/replacing vdev from the tree. 769168404Spjd */ 770168404Spjdvoid 771168404Spjdvdev_remove_parent(vdev_t *cvd) 772168404Spjd{ 773168404Spjd vdev_t *mvd = cvd->vdev_parent; 774168404Spjd vdev_t *pvd = mvd->vdev_parent; 775168404Spjd 776185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 777168404Spjd 778168404Spjd ASSERT(mvd->vdev_children == 1); 779168404Spjd ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 780168404Spjd mvd->vdev_ops == &vdev_replacing_ops || 781168404Spjd mvd->vdev_ops == &vdev_spare_ops); 782168404Spjd cvd->vdev_ashift = mvd->vdev_ashift; 783168404Spjd 784168404Spjd vdev_remove_child(mvd, cvd); 785168404Spjd vdev_remove_child(pvd, mvd); 786209962Smm 787185029Spjd /* 788185029Spjd * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 789185029Spjd * Otherwise, we could have detached an offline device, and when we 790185029Spjd * go to import the pool we'll think we have two top-level vdevs, 791185029Spjd * instead of a different version of the same top-level vdev. 792185029Spjd */ 793209962Smm if (mvd->vdev_top == mvd) { 794209962Smm uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 795219089Spjd cvd->vdev_orig_guid = cvd->vdev_guid; 796209962Smm cvd->vdev_guid += guid_delta; 797209962Smm cvd->vdev_guid_sum += guid_delta; 798209962Smm } 799168404Spjd cvd->vdev_id = mvd->vdev_id; 800168404Spjd vdev_add_child(pvd, cvd); 801168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 802168404Spjd 803168404Spjd if (cvd == cvd->vdev_top) 804168404Spjd vdev_top_transfer(mvd, cvd); 805168404Spjd 806168404Spjd ASSERT(mvd->vdev_children == 0); 807168404Spjd vdev_free(mvd); 808168404Spjd} 809168404Spjd 810168404Spjdint 811168404Spjdvdev_metaslab_init(vdev_t *vd, uint64_t txg) 812168404Spjd{ 813168404Spjd spa_t *spa = vd->vdev_spa; 814168404Spjd objset_t *mos = spa->spa_meta_objset; 815168404Spjd uint64_t m; 816168404Spjd uint64_t oldc = vd->vdev_ms_count; 817168404Spjd uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 818168404Spjd metaslab_t **mspp; 819168404Spjd int error; 820168404Spjd 821219089Spjd ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 822219089Spjd 823219089Spjd /* 824219089Spjd * This vdev is not being allocated from yet or is a hole. 825219089Spjd */ 826219089Spjd if (vd->vdev_ms_shift == 0) 827168404Spjd return (0); 828168404Spjd 829219089Spjd ASSERT(!vd->vdev_ishole); 830219089Spjd 831213197Smm /* 832213197Smm * Compute the raidz-deflation ratio. Note, we hard-code 833213197Smm * in 128k (1 << 17) because it is the current "typical" blocksize. 834213197Smm * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 835213197Smm * or we will inconsistently account for existing bp's. 836213197Smm */ 837213197Smm vd->vdev_deflate_ratio = (1 << 17) / 838213197Smm (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 839213197Smm 840168404Spjd ASSERT(oldc <= newc); 841168404Spjd 842168404Spjd mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 843168404Spjd 844168404Spjd if (oldc != 0) { 845168404Spjd bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 846168404Spjd kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 847168404Spjd } 848168404Spjd 849168404Spjd vd->vdev_ms = mspp; 850168404Spjd vd->vdev_ms_count = newc; 851168404Spjd 852168404Spjd for (m = oldc; m < newc; m++) { 853168404Spjd space_map_obj_t smo = { 0, 0, 0 }; 854168404Spjd if (txg == 0) { 855168404Spjd uint64_t object = 0; 856168404Spjd error = dmu_read(mos, vd->vdev_ms_array, 857209962Smm m * sizeof (uint64_t), sizeof (uint64_t), &object, 858209962Smm DMU_READ_PREFETCH); 859168404Spjd if (error) 860168404Spjd return (error); 861168404Spjd if (object != 0) { 862168404Spjd dmu_buf_t *db; 863168404Spjd error = dmu_bonus_hold(mos, object, FTAG, &db); 864168404Spjd if (error) 865168404Spjd return (error); 866185029Spjd ASSERT3U(db->db_size, >=, sizeof (smo)); 867185029Spjd bcopy(db->db_data, &smo, sizeof (smo)); 868168404Spjd ASSERT3U(smo.smo_object, ==, object); 869168404Spjd dmu_buf_rele(db, FTAG); 870168404Spjd } 871168404Spjd } 872168404Spjd vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 873168404Spjd m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 874168404Spjd } 875168404Spjd 876219089Spjd if (txg == 0) 877219089Spjd spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 878219089Spjd 879219089Spjd /* 880219089Spjd * If the vdev is being removed we don't activate 881219089Spjd * the metaslabs since we want to ensure that no new 882219089Spjd * allocations are performed on this device. 883219089Spjd */ 884219089Spjd if (oldc == 0 && !vd->vdev_removing) 885219089Spjd metaslab_group_activate(vd->vdev_mg); 886219089Spjd 887219089Spjd if (txg == 0) 888219089Spjd spa_config_exit(spa, SCL_ALLOC, FTAG); 889219089Spjd 890168404Spjd return (0); 891168404Spjd} 892168404Spjd 893168404Spjdvoid 894168404Spjdvdev_metaslab_fini(vdev_t *vd) 895168404Spjd{ 896168404Spjd uint64_t m; 897168404Spjd uint64_t count = vd->vdev_ms_count; 898168404Spjd 899168404Spjd if (vd->vdev_ms != NULL) { 900219089Spjd metaslab_group_passivate(vd->vdev_mg); 901168404Spjd for (m = 0; m < count; m++) 902168404Spjd if (vd->vdev_ms[m] != NULL) 903168404Spjd metaslab_fini(vd->vdev_ms[m]); 904168404Spjd kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 905168404Spjd vd->vdev_ms = NULL; 906168404Spjd } 907168404Spjd} 908168404Spjd 909185029Spjdtypedef struct vdev_probe_stats { 910185029Spjd boolean_t vps_readable; 911185029Spjd boolean_t vps_writeable; 912185029Spjd int vps_flags; 913185029Spjd} vdev_probe_stats_t; 914185029Spjd 915185029Spjdstatic void 916185029Spjdvdev_probe_done(zio_t *zio) 917185029Spjd{ 918209962Smm spa_t *spa = zio->io_spa; 919209962Smm vdev_t *vd = zio->io_vd; 920185029Spjd vdev_probe_stats_t *vps = zio->io_private; 921185029Spjd 922209962Smm ASSERT(vd->vdev_probe_zio != NULL); 923209962Smm 924185029Spjd if (zio->io_type == ZIO_TYPE_READ) { 925185029Spjd if (zio->io_error == 0) 926185029Spjd vps->vps_readable = 1; 927209962Smm if (zio->io_error == 0 && spa_writeable(spa)) { 928209962Smm zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 929185029Spjd zio->io_offset, zio->io_size, zio->io_data, 930185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 931185029Spjd ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 932185029Spjd } else { 933185029Spjd zio_buf_free(zio->io_data, zio->io_size); 934185029Spjd } 935185029Spjd } else if (zio->io_type == ZIO_TYPE_WRITE) { 936185029Spjd if (zio->io_error == 0) 937185029Spjd vps->vps_writeable = 1; 938185029Spjd zio_buf_free(zio->io_data, zio->io_size); 939185029Spjd } else if (zio->io_type == ZIO_TYPE_NULL) { 940209962Smm zio_t *pio; 941185029Spjd 942185029Spjd vd->vdev_cant_read |= !vps->vps_readable; 943185029Spjd vd->vdev_cant_write |= !vps->vps_writeable; 944185029Spjd 945185029Spjd if (vdev_readable(vd) && 946209962Smm (vdev_writeable(vd) || !spa_writeable(spa))) { 947185029Spjd zio->io_error = 0; 948185029Spjd } else { 949185029Spjd ASSERT(zio->io_error != 0); 950185029Spjd zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 951209962Smm spa, vd, NULL, 0, 0); 952185029Spjd zio->io_error = ENXIO; 953185029Spjd } 954209962Smm 955209962Smm mutex_enter(&vd->vdev_probe_lock); 956209962Smm ASSERT(vd->vdev_probe_zio == zio); 957209962Smm vd->vdev_probe_zio = NULL; 958209962Smm mutex_exit(&vd->vdev_probe_lock); 959209962Smm 960209962Smm while ((pio = zio_walk_parents(zio)) != NULL) 961209962Smm if (!vdev_accessible(vd, pio)) 962209962Smm pio->io_error = ENXIO; 963209962Smm 964185029Spjd kmem_free(vps, sizeof (*vps)); 965185029Spjd } 966185029Spjd} 967185029Spjd 968168404Spjd/* 969185029Spjd * Determine whether this device is accessible by reading and writing 970185029Spjd * to several known locations: the pad regions of each vdev label 971185029Spjd * but the first (which we leave alone in case it contains a VTOC). 972185029Spjd */ 973185029Spjdzio_t * 974209962Smmvdev_probe(vdev_t *vd, zio_t *zio) 975185029Spjd{ 976185029Spjd spa_t *spa = vd->vdev_spa; 977209962Smm vdev_probe_stats_t *vps = NULL; 978209962Smm zio_t *pio; 979185029Spjd 980209962Smm ASSERT(vd->vdev_ops->vdev_op_leaf); 981185029Spjd 982209962Smm /* 983209962Smm * Don't probe the probe. 984209962Smm */ 985209962Smm if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 986209962Smm return (NULL); 987185029Spjd 988209962Smm /* 989209962Smm * To prevent 'probe storms' when a device fails, we create 990209962Smm * just one probe i/o at a time. All zios that want to probe 991209962Smm * this vdev will become parents of the probe io. 992209962Smm */ 993209962Smm mutex_enter(&vd->vdev_probe_lock); 994209962Smm 995209962Smm if ((pio = vd->vdev_probe_zio) == NULL) { 996209962Smm vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 997209962Smm 998209962Smm vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 999209962Smm ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1000213198Smm ZIO_FLAG_TRYHARD; 1001209962Smm 1002209962Smm if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1003209962Smm /* 1004209962Smm * vdev_cant_read and vdev_cant_write can only 1005209962Smm * transition from TRUE to FALSE when we have the 1006209962Smm * SCL_ZIO lock as writer; otherwise they can only 1007209962Smm * transition from FALSE to TRUE. This ensures that 1008209962Smm * any zio looking at these values can assume that 1009209962Smm * failures persist for the life of the I/O. That's 1010209962Smm * important because when a device has intermittent 1011209962Smm * connectivity problems, we want to ensure that 1012209962Smm * they're ascribed to the device (ENXIO) and not 1013209962Smm * the zio (EIO). 1014209962Smm * 1015209962Smm * Since we hold SCL_ZIO as writer here, clear both 1016209962Smm * values so the probe can reevaluate from first 1017209962Smm * principles. 1018209962Smm */ 1019209962Smm vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1020209962Smm vd->vdev_cant_read = B_FALSE; 1021209962Smm vd->vdev_cant_write = B_FALSE; 1022209962Smm } 1023209962Smm 1024209962Smm vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1025209962Smm vdev_probe_done, vps, 1026209962Smm vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1027209962Smm 1028219089Spjd /* 1029219089Spjd * We can't change the vdev state in this context, so we 1030219089Spjd * kick off an async task to do it on our behalf. 1031219089Spjd */ 1032209962Smm if (zio != NULL) { 1033209962Smm vd->vdev_probe_wanted = B_TRUE; 1034209962Smm spa_async_request(spa, SPA_ASYNC_PROBE); 1035209962Smm } 1036185029Spjd } 1037185029Spjd 1038209962Smm if (zio != NULL) 1039209962Smm zio_add_child(zio, pio); 1040185029Spjd 1041209962Smm mutex_exit(&vd->vdev_probe_lock); 1042185029Spjd 1043209962Smm if (vps == NULL) { 1044209962Smm ASSERT(zio != NULL); 1045209962Smm return (NULL); 1046209962Smm } 1047185029Spjd 1048185029Spjd for (int l = 1; l < VDEV_LABELS; l++) { 1049209962Smm zio_nowait(zio_read_phys(pio, vd, 1050185029Spjd vdev_label_offset(vd->vdev_psize, l, 1051209962Smm offsetof(vdev_label_t, vl_pad2)), 1052209962Smm VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1053185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1054185029Spjd ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1055185029Spjd } 1056185029Spjd 1057209962Smm if (zio == NULL) 1058209962Smm return (pio); 1059209962Smm 1060209962Smm zio_nowait(pio); 1061209962Smm return (NULL); 1062185029Spjd} 1063185029Spjd 1064219089Spjdstatic void 1065219089Spjdvdev_open_child(void *arg) 1066219089Spjd{ 1067219089Spjd vdev_t *vd = arg; 1068219089Spjd 1069219089Spjd vd->vdev_open_thread = curthread; 1070219089Spjd vd->vdev_open_error = vdev_open(vd); 1071219089Spjd vd->vdev_open_thread = NULL; 1072219089Spjd} 1073219089Spjd 1074219089Spjdboolean_t 1075219089Spjdvdev_uses_zvols(vdev_t *vd) 1076219089Spjd{ 1077219089Spjd if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1078219089Spjd strlen(ZVOL_DIR)) == 0) 1079219089Spjd return (B_TRUE); 1080219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1081219089Spjd if (vdev_uses_zvols(vd->vdev_child[c])) 1082219089Spjd return (B_TRUE); 1083219089Spjd return (B_FALSE); 1084219089Spjd} 1085219089Spjd 1086219089Spjdvoid 1087219089Spjdvdev_open_children(vdev_t *vd) 1088219089Spjd{ 1089219089Spjd taskq_t *tq; 1090219089Spjd int children = vd->vdev_children; 1091219089Spjd 1092219089Spjd /* 1093219089Spjd * in order to handle pools on top of zvols, do the opens 1094219089Spjd * in a single thread so that the same thread holds the 1095219089Spjd * spa_namespace_lock 1096219089Spjd */ 1097219089Spjd if (B_TRUE || vdev_uses_zvols(vd)) { 1098219089Spjd for (int c = 0; c < children; c++) 1099219089Spjd vd->vdev_child[c]->vdev_open_error = 1100219089Spjd vdev_open(vd->vdev_child[c]); 1101219089Spjd return; 1102219089Spjd } 1103219089Spjd tq = taskq_create("vdev_open", children, minclsyspri, 1104219089Spjd children, children, TASKQ_PREPOPULATE); 1105219089Spjd 1106219089Spjd for (int c = 0; c < children; c++) 1107219089Spjd VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1108219089Spjd TQ_SLEEP) != 0); 1109219089Spjd 1110219089Spjd taskq_destroy(tq); 1111219089Spjd} 1112219089Spjd 1113185029Spjd/* 1114168404Spjd * Prepare a virtual device for access. 1115168404Spjd */ 1116168404Spjdint 1117168404Spjdvdev_open(vdev_t *vd) 1118168404Spjd{ 1119209962Smm spa_t *spa = vd->vdev_spa; 1120168404Spjd int error; 1121168404Spjd uint64_t osize = 0; 1122168404Spjd uint64_t 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); 1146185029Spjd return (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); 1150168404Spjd return (ENXIO); 1151168404Spjd } 1152168404Spjd 1153168404Spjd error = vd->vdev_ops->vdev_op_open(vd, &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); 1185219089Spjd return (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 1202219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 1203168404Spjd if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1204168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1205168404Spjd VDEV_AUX_NONE); 1206168404Spjd break; 1207168404Spjd } 1208219089Spjd } 1209168404Spjd 1210168404Spjd osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1211168404Spjd 1212168404Spjd if (vd->vdev_children == 0) { 1213168404Spjd if (osize < SPA_MINDEVSIZE) { 1214168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1215168404Spjd VDEV_AUX_TOO_SMALL); 1216168404Spjd return (EOVERFLOW); 1217168404Spjd } 1218168404Spjd psize = osize; 1219168404Spjd asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1220168404Spjd } else { 1221168404Spjd if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1222168404Spjd (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1223168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1224168404Spjd VDEV_AUX_TOO_SMALL); 1225168404Spjd return (EOVERFLOW); 1226168404Spjd } 1227168404Spjd psize = 0; 1228168404Spjd asize = osize; 1229168404Spjd } 1230168404Spjd 1231168404Spjd vd->vdev_psize = psize; 1232168404Spjd 1233219089Spjd /* 1234219089Spjd * Make sure the allocatable size hasn't shrunk. 1235219089Spjd */ 1236219089Spjd if (asize < vd->vdev_min_asize) { 1237219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1238219089Spjd VDEV_AUX_BAD_LABEL); 1239219089Spjd return (EINVAL); 1240219089Spjd } 1241219089Spjd 1242168404Spjd if (vd->vdev_asize == 0) { 1243168404Spjd /* 1244168404Spjd * This is the first-ever open, so use the computed values. 1245168404Spjd * For testing purposes, a higher ashift can be requested. 1246168404Spjd */ 1247168404Spjd vd->vdev_asize = asize; 1248168404Spjd vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1249168404Spjd } else { 1250168404Spjd /* 1251168404Spjd * Make sure the alignment requirement hasn't increased. 1252168404Spjd */ 1253168404Spjd if (ashift > vd->vdev_top->vdev_ashift) { 1254168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1255168404Spjd VDEV_AUX_BAD_LABEL); 1256168404Spjd return (EINVAL); 1257168404Spjd } 1258219089Spjd } 1259168404Spjd 1260219089Spjd /* 1261219089Spjd * If all children are healthy and the asize has increased, 1262219089Spjd * then we've experienced dynamic LUN growth. If automatic 1263219089Spjd * expansion is enabled then use the additional space. 1264219089Spjd */ 1265219089Spjd if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1266219089Spjd (vd->vdev_expanding || spa->spa_autoexpand)) 1267219089Spjd vd->vdev_asize = asize; 1268168404Spjd 1269219089Spjd vdev_set_min_asize(vd); 1270168404Spjd 1271168404Spjd /* 1272185029Spjd * Ensure we can issue some IO before declaring the 1273185029Spjd * vdev open for business. 1274185029Spjd */ 1275185029Spjd if (vd->vdev_ops->vdev_op_leaf && 1276185029Spjd (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1277219089Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1278219089Spjd VDEV_AUX_ERR_EXCEEDED); 1279185029Spjd return (error); 1280185029Spjd } 1281185029Spjd 1282185029Spjd /* 1283185029Spjd * If a leaf vdev has a DTL, and seems healthy, then kick off a 1284209962Smm * resilver. But don't do this if we are doing a reopen for a scrub, 1285209962Smm * since this would just restart the scrub we are already doing. 1286168404Spjd */ 1287209962Smm if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1288209962Smm vdev_resilver_needed(vd, NULL, NULL)) 1289209962Smm spa_async_request(spa, SPA_ASYNC_RESILVER); 1290168404Spjd 1291168404Spjd return (0); 1292168404Spjd} 1293168404Spjd 1294168404Spjd/* 1295168404Spjd * Called once the vdevs are all opened, this routine validates the label 1296168404Spjd * contents. This needs to be done before vdev_load() so that we don't 1297185029Spjd * inadvertently do repair I/Os to the wrong device. 1298168404Spjd * 1299231141Smm * If 'strict' is false ignore the spa guid check. This is necessary because 1300231141Smm * if the machine crashed during a re-guid the new guid might have been written 1301231141Smm * to all of the vdev labels, but not the cached config. The strict check 1302231141Smm * will be performed when the pool is opened again using the mos config. 1303231141Smm * 1304168404Spjd * This function will only return failure if one of the vdevs indicates that it 1305168404Spjd * has since been destroyed or exported. This is only possible if 1306168404Spjd * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1307168404Spjd * will be updated but the function will return 0. 1308168404Spjd */ 1309168404Spjdint 1310231141Smmvdev_validate(vdev_t *vd, boolean_t strict) 1311168404Spjd{ 1312168404Spjd spa_t *spa = vd->vdev_spa; 1313168404Spjd nvlist_t *label; 1314219089Spjd uint64_t guid = 0, top_guid; 1315168404Spjd uint64_t state; 1316168404Spjd 1317219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1318231141Smm if (vdev_validate(vd->vdev_child[c], strict) != 0) 1319168926Spjd return (EBADF); 1320168404Spjd 1321168404Spjd /* 1322168404Spjd * If the device has already failed, or was marked offline, don't do 1323168404Spjd * any further validation. Otherwise, label I/O will fail and we will 1324168404Spjd * overwrite the previous state. 1325168404Spjd */ 1326185029Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1327219089Spjd uint64_t aux_guid = 0; 1328219089Spjd nvlist_t *nvl; 1329168404Spjd 1330168404Spjd if ((label = vdev_label_read_config(vd)) == NULL) { 1331168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1332168404Spjd VDEV_AUX_BAD_LABEL); 1333168404Spjd return (0); 1334168404Spjd } 1335168404Spjd 1336219089Spjd /* 1337219089Spjd * Determine if this vdev has been split off into another 1338219089Spjd * pool. If so, then refuse to open it. 1339219089Spjd */ 1340219089Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1341219089Spjd &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1342219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1343219089Spjd VDEV_AUX_SPLIT_POOL); 1344219089Spjd nvlist_free(label); 1345219089Spjd return (0); 1346219089Spjd } 1347219089Spjd 1348231141Smm if (strict && (nvlist_lookup_uint64(label, 1349231141Smm ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1350231141Smm guid != spa_guid(spa))) { 1351168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1352168404Spjd VDEV_AUX_CORRUPT_DATA); 1353168404Spjd nvlist_free(label); 1354168404Spjd return (0); 1355168404Spjd } 1356168404Spjd 1357219089Spjd if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1358219089Spjd != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1359219089Spjd &aux_guid) != 0) 1360219089Spjd aux_guid = 0; 1361219089Spjd 1362185029Spjd /* 1363185029Spjd * If this vdev just became a top-level vdev because its 1364185029Spjd * sibling was detached, it will have adopted the parent's 1365185029Spjd * vdev guid -- but the label may or may not be on disk yet. 1366185029Spjd * Fortunately, either version of the label will have the 1367185029Spjd * same top guid, so if we're a top-level vdev, we can 1368185029Spjd * safely compare to that instead. 1369219089Spjd * 1370219089Spjd * If we split this vdev off instead, then we also check the 1371219089Spjd * original pool's guid. We don't want to consider the vdev 1372219089Spjd * corrupt if it is partway through a split operation. 1373185029Spjd */ 1374168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1375185029Spjd &guid) != 0 || 1376185029Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1377185029Spjd &top_guid) != 0 || 1378219089Spjd ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1379185029Spjd (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1380168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1381168404Spjd VDEV_AUX_CORRUPT_DATA); 1382168404Spjd nvlist_free(label); 1383168404Spjd return (0); 1384168404Spjd } 1385168404Spjd 1386168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1387168404Spjd &state) != 0) { 1388168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1389168404Spjd VDEV_AUX_CORRUPT_DATA); 1390168404Spjd nvlist_free(label); 1391168404Spjd return (0); 1392168404Spjd } 1393168404Spjd 1394168404Spjd nvlist_free(label); 1395168404Spjd 1396209962Smm /* 1397219089Spjd * If this is a verbatim import, no need to check the 1398209962Smm * state of the pool. 1399209962Smm */ 1400219089Spjd if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1401219089Spjd spa_load_state(spa) == SPA_LOAD_OPEN && 1402168404Spjd state != POOL_STATE_ACTIVE) 1403168926Spjd return (EBADF); 1404185029Spjd 1405185029Spjd /* 1406185029Spjd * If we were able to open and validate a vdev that was 1407185029Spjd * previously marked permanently unavailable, clear that state 1408185029Spjd * now. 1409185029Spjd */ 1410185029Spjd if (vd->vdev_not_present) 1411185029Spjd vd->vdev_not_present = 0; 1412168404Spjd } 1413168404Spjd 1414168404Spjd return (0); 1415168404Spjd} 1416168404Spjd 1417168404Spjd/* 1418168404Spjd * Close a virtual device. 1419168404Spjd */ 1420168404Spjdvoid 1421168404Spjdvdev_close(vdev_t *vd) 1422168404Spjd{ 1423209962Smm spa_t *spa = vd->vdev_spa; 1424219089Spjd vdev_t *pvd = vd->vdev_parent; 1425209962Smm 1426209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1427209962Smm 1428219089Spjd /* 1429219089Spjd * If our parent is reopening, then we are as well, unless we are 1430219089Spjd * going offline. 1431219089Spjd */ 1432219089Spjd if (pvd != NULL && pvd->vdev_reopening) 1433219089Spjd vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1434219089Spjd 1435168404Spjd vd->vdev_ops->vdev_op_close(vd); 1436168404Spjd 1437185029Spjd vdev_cache_purge(vd); 1438168404Spjd 1439168404Spjd /* 1440219089Spjd * We record the previous state before we close it, so that if we are 1441168404Spjd * doing a reopen(), we don't generate FMA ereports if we notice that 1442168404Spjd * it's still faulted. 1443168404Spjd */ 1444168404Spjd vd->vdev_prevstate = vd->vdev_state; 1445168404Spjd 1446168404Spjd if (vd->vdev_offline) 1447168404Spjd vd->vdev_state = VDEV_STATE_OFFLINE; 1448168404Spjd else 1449168404Spjd vd->vdev_state = VDEV_STATE_CLOSED; 1450168404Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1451168404Spjd} 1452168404Spjd 1453168404Spjdvoid 1454219089Spjdvdev_hold(vdev_t *vd) 1455219089Spjd{ 1456219089Spjd spa_t *spa = vd->vdev_spa; 1457219089Spjd 1458219089Spjd ASSERT(spa_is_root(spa)); 1459219089Spjd if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1460219089Spjd return; 1461219089Spjd 1462219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1463219089Spjd vdev_hold(vd->vdev_child[c]); 1464219089Spjd 1465219089Spjd if (vd->vdev_ops->vdev_op_leaf) 1466219089Spjd vd->vdev_ops->vdev_op_hold(vd); 1467219089Spjd} 1468219089Spjd 1469219089Spjdvoid 1470219089Spjdvdev_rele(vdev_t *vd) 1471219089Spjd{ 1472219089Spjd spa_t *spa = vd->vdev_spa; 1473219089Spjd 1474219089Spjd ASSERT(spa_is_root(spa)); 1475219089Spjd for (int c = 0; c < vd->vdev_children; c++) 1476219089Spjd vdev_rele(vd->vdev_child[c]); 1477219089Spjd 1478219089Spjd if (vd->vdev_ops->vdev_op_leaf) 1479219089Spjd vd->vdev_ops->vdev_op_rele(vd); 1480219089Spjd} 1481219089Spjd 1482219089Spjd/* 1483219089Spjd * Reopen all interior vdevs and any unopened leaves. We don't actually 1484219089Spjd * reopen leaf vdevs which had previously been opened as they might deadlock 1485219089Spjd * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1486219089Spjd * If the leaf has never been opened then open it, as usual. 1487219089Spjd */ 1488219089Spjdvoid 1489168404Spjdvdev_reopen(vdev_t *vd) 1490168404Spjd{ 1491168404Spjd spa_t *spa = vd->vdev_spa; 1492168404Spjd 1493185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1494168404Spjd 1495219089Spjd /* set the reopening flag unless we're taking the vdev offline */ 1496219089Spjd vd->vdev_reopening = !vd->vdev_offline; 1497168404Spjd vdev_close(vd); 1498168404Spjd (void) vdev_open(vd); 1499168404Spjd 1500168404Spjd /* 1501168404Spjd * Call vdev_validate() here to make sure we have the same device. 1502168404Spjd * Otherwise, a device with an invalid label could be successfully 1503168404Spjd * opened in response to vdev_reopen(). 1504168404Spjd */ 1505185029Spjd if (vd->vdev_aux) { 1506185029Spjd (void) vdev_validate_aux(vd); 1507185029Spjd if (vdev_readable(vd) && vdev_writeable(vd) && 1508209962Smm vd->vdev_aux == &spa->spa_l2cache && 1509219089Spjd !l2arc_vdev_present(vd)) 1510219089Spjd l2arc_add_vdev(spa, vd); 1511185029Spjd } else { 1512231141Smm (void) vdev_validate(vd, B_TRUE); 1513185029Spjd } 1514168404Spjd 1515168404Spjd /* 1516185029Spjd * Reassess parent vdev's health. 1517168404Spjd */ 1518185029Spjd vdev_propagate_state(vd); 1519168404Spjd} 1520168404Spjd 1521168404Spjdint 1522168404Spjdvdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1523168404Spjd{ 1524168404Spjd int error; 1525168404Spjd 1526168404Spjd /* 1527168404Spjd * Normally, partial opens (e.g. of a mirror) are allowed. 1528168404Spjd * For a create, however, we want to fail the request if 1529168404Spjd * there are any components we can't open. 1530168404Spjd */ 1531168404Spjd error = vdev_open(vd); 1532168404Spjd 1533168404Spjd if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1534168404Spjd vdev_close(vd); 1535168404Spjd return (error ? error : ENXIO); 1536168404Spjd } 1537168404Spjd 1538168404Spjd /* 1539168404Spjd * Recursively initialize all labels. 1540168404Spjd */ 1541168404Spjd if ((error = vdev_label_init(vd, txg, isreplacing ? 1542168404Spjd VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1543168404Spjd vdev_close(vd); 1544168404Spjd return (error); 1545168404Spjd } 1546168404Spjd 1547168404Spjd return (0); 1548168404Spjd} 1549168404Spjd 1550168404Spjdvoid 1551219089Spjdvdev_metaslab_set_size(vdev_t *vd) 1552168404Spjd{ 1553168404Spjd /* 1554168404Spjd * Aim for roughly 200 metaslabs per vdev. 1555168404Spjd */ 1556168404Spjd vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1557168404Spjd vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1558168404Spjd} 1559168404Spjd 1560168404Spjdvoid 1561168404Spjdvdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1562168404Spjd{ 1563168404Spjd ASSERT(vd == vd->vdev_top); 1564219089Spjd ASSERT(!vd->vdev_ishole); 1565168404Spjd ASSERT(ISP2(flags)); 1566219089Spjd ASSERT(spa_writeable(vd->vdev_spa)); 1567168404Spjd 1568168404Spjd if (flags & VDD_METASLAB) 1569168404Spjd (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1570168404Spjd 1571168404Spjd if (flags & VDD_DTL) 1572168404Spjd (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1573168404Spjd 1574168404Spjd (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1575168404Spjd} 1576168404Spjd 1577209962Smm/* 1578209962Smm * DTLs. 1579209962Smm * 1580209962Smm * A vdev's DTL (dirty time log) is the set of transaction groups for which 1581219089Spjd * the vdev has less than perfect replication. There are four kinds of DTL: 1582209962Smm * 1583209962Smm * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1584209962Smm * 1585209962Smm * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1586209962Smm * 1587209962Smm * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1588209962Smm * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1589209962Smm * txgs that was scrubbed. 1590209962Smm * 1591209962Smm * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1592209962Smm * persistent errors or just some device being offline. 1593209962Smm * Unlike the other three, the DTL_OUTAGE map is not generally 1594209962Smm * maintained; it's only computed when needed, typically to 1595209962Smm * determine whether a device can be detached. 1596209962Smm * 1597209962Smm * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1598209962Smm * either has the data or it doesn't. 1599209962Smm * 1600209962Smm * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1601209962Smm * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1602209962Smm * if any child is less than fully replicated, then so is its parent. 1603209962Smm * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1604209962Smm * comprising only those txgs which appear in 'maxfaults' or more children; 1605209962Smm * those are the txgs we don't have enough replication to read. For example, 1606209962Smm * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1607209962Smm * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1608209962Smm * two child DTL_MISSING maps. 1609209962Smm * 1610209962Smm * It should be clear from the above that to compute the DTLs and outage maps 1611209962Smm * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1612209962Smm * Therefore, that is all we keep on disk. When loading the pool, or after 1613209962Smm * a configuration change, we generate all other DTLs from first principles. 1614209962Smm */ 1615168404Spjdvoid 1616209962Smmvdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1617168404Spjd{ 1618209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1619209962Smm 1620209962Smm ASSERT(t < DTL_TYPES); 1621209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1622219089Spjd ASSERT(spa_writeable(vd->vdev_spa)); 1623209962Smm 1624168404Spjd mutex_enter(sm->sm_lock); 1625168404Spjd if (!space_map_contains(sm, txg, size)) 1626168404Spjd space_map_add(sm, txg, size); 1627168404Spjd mutex_exit(sm->sm_lock); 1628168404Spjd} 1629168404Spjd 1630209962Smmboolean_t 1631209962Smmvdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1632168404Spjd{ 1633209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1634209962Smm boolean_t dirty = B_FALSE; 1635168404Spjd 1636209962Smm ASSERT(t < DTL_TYPES); 1637209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1638168404Spjd 1639168404Spjd mutex_enter(sm->sm_lock); 1640209962Smm if (sm->sm_space != 0) 1641209962Smm dirty = space_map_contains(sm, txg, size); 1642168404Spjd mutex_exit(sm->sm_lock); 1643168404Spjd 1644168404Spjd return (dirty); 1645168404Spjd} 1646168404Spjd 1647209962Smmboolean_t 1648209962Smmvdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1649209962Smm{ 1650209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1651209962Smm boolean_t empty; 1652209962Smm 1653209962Smm mutex_enter(sm->sm_lock); 1654209962Smm empty = (sm->sm_space == 0); 1655209962Smm mutex_exit(sm->sm_lock); 1656209962Smm 1657209962Smm return (empty); 1658209962Smm} 1659209962Smm 1660168404Spjd/* 1661168404Spjd * Reassess DTLs after a config change or scrub completion. 1662168404Spjd */ 1663168404Spjdvoid 1664168404Spjdvdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1665168404Spjd{ 1666168404Spjd spa_t *spa = vd->vdev_spa; 1667209962Smm avl_tree_t reftree; 1668209962Smm int minref; 1669168404Spjd 1670209962Smm ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1671168404Spjd 1672209962Smm for (int c = 0; c < vd->vdev_children; c++) 1673209962Smm vdev_dtl_reassess(vd->vdev_child[c], txg, 1674209962Smm scrub_txg, scrub_done); 1675209962Smm 1676219089Spjd if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 1677209962Smm return; 1678209962Smm 1679209962Smm if (vd->vdev_ops->vdev_op_leaf) { 1680219089Spjd dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1681219089Spjd 1682168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1683185029Spjd if (scrub_txg != 0 && 1684219089Spjd (spa->spa_scrub_started || 1685219089Spjd (scn && scn->scn_phys.scn_errors == 0))) { 1686185029Spjd /* 1687185029Spjd * We completed a scrub up to scrub_txg. If we 1688185029Spjd * did it without rebooting, then the scrub dtl 1689185029Spjd * will be valid, so excise the old region and 1690185029Spjd * fold in the scrub dtl. Otherwise, leave the 1691185029Spjd * dtl as-is if there was an error. 1692209962Smm * 1693209962Smm * There's little trick here: to excise the beginning 1694209962Smm * of the DTL_MISSING map, we put it into a reference 1695209962Smm * tree and then add a segment with refcnt -1 that 1696209962Smm * covers the range [0, scrub_txg). This means 1697209962Smm * that each txg in that range has refcnt -1 or 0. 1698209962Smm * We then add DTL_SCRUB with a refcnt of 2, so that 1699209962Smm * entries in the range [0, scrub_txg) will have a 1700209962Smm * positive refcnt -- either 1 or 2. We then convert 1701209962Smm * the reference tree into the new DTL_MISSING map. 1702185029Spjd */ 1703209962Smm space_map_ref_create(&reftree); 1704209962Smm space_map_ref_add_map(&reftree, 1705209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1706209962Smm space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 1707209962Smm space_map_ref_add_map(&reftree, 1708209962Smm &vd->vdev_dtl[DTL_SCRUB], 2); 1709209962Smm space_map_ref_generate_map(&reftree, 1710209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1711209962Smm space_map_ref_destroy(&reftree); 1712168404Spjd } 1713209962Smm space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1714209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1715209962Smm space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1716168404Spjd if (scrub_done) 1717209962Smm space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1718209962Smm space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1719209962Smm if (!vdev_readable(vd)) 1720209962Smm space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1721209962Smm else 1722209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1723209962Smm space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1724168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1725185029Spjd 1726168404Spjd if (txg != 0) 1727168404Spjd vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1728168404Spjd return; 1729168404Spjd } 1730168404Spjd 1731168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1732209962Smm for (int t = 0; t < DTL_TYPES; t++) { 1733209962Smm /* account for child's outage in parent's missing map */ 1734209962Smm int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1735209962Smm if (t == DTL_SCRUB) 1736209962Smm continue; /* leaf vdevs only */ 1737209962Smm if (t == DTL_PARTIAL) 1738209962Smm minref = 1; /* i.e. non-zero */ 1739209962Smm else if (vd->vdev_nparity != 0) 1740209962Smm minref = vd->vdev_nparity + 1; /* RAID-Z */ 1741209962Smm else 1742209962Smm minref = vd->vdev_children; /* any kind of mirror */ 1743209962Smm space_map_ref_create(&reftree); 1744209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1745209962Smm vdev_t *cvd = vd->vdev_child[c]; 1746209962Smm mutex_enter(&cvd->vdev_dtl_lock); 1747209962Smm space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 1748209962Smm mutex_exit(&cvd->vdev_dtl_lock); 1749209962Smm } 1750209962Smm space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 1751209962Smm space_map_ref_destroy(&reftree); 1752209962Smm } 1753168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1754168404Spjd} 1755168404Spjd 1756168404Spjdstatic int 1757168404Spjdvdev_dtl_load(vdev_t *vd) 1758168404Spjd{ 1759168404Spjd spa_t *spa = vd->vdev_spa; 1760209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1761168404Spjd objset_t *mos = spa->spa_meta_objset; 1762168404Spjd dmu_buf_t *db; 1763168404Spjd int error; 1764168404Spjd 1765168404Spjd ASSERT(vd->vdev_children == 0); 1766168404Spjd 1767168404Spjd if (smo->smo_object == 0) 1768168404Spjd return (0); 1769168404Spjd 1770219089Spjd ASSERT(!vd->vdev_ishole); 1771219089Spjd 1772168404Spjd if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1773168404Spjd return (error); 1774168404Spjd 1775185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1776185029Spjd bcopy(db->db_data, smo, sizeof (*smo)); 1777168404Spjd dmu_buf_rele(db, FTAG); 1778168404Spjd 1779168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1780209962Smm error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 1781209962Smm NULL, SM_ALLOC, smo, mos); 1782168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1783168404Spjd 1784168404Spjd return (error); 1785168404Spjd} 1786168404Spjd 1787168404Spjdvoid 1788168404Spjdvdev_dtl_sync(vdev_t *vd, uint64_t txg) 1789168404Spjd{ 1790168404Spjd spa_t *spa = vd->vdev_spa; 1791209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1792209962Smm space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1793168404Spjd objset_t *mos = spa->spa_meta_objset; 1794168404Spjd space_map_t smsync; 1795168404Spjd kmutex_t smlock; 1796168404Spjd dmu_buf_t *db; 1797168404Spjd dmu_tx_t *tx; 1798168404Spjd 1799219089Spjd ASSERT(!vd->vdev_ishole); 1800219089Spjd 1801168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1802168404Spjd 1803168404Spjd if (vd->vdev_detached) { 1804168404Spjd if (smo->smo_object != 0) { 1805168404Spjd int err = dmu_object_free(mos, smo->smo_object, tx); 1806168404Spjd ASSERT3U(err, ==, 0); 1807168404Spjd smo->smo_object = 0; 1808168404Spjd } 1809168404Spjd dmu_tx_commit(tx); 1810168404Spjd return; 1811168404Spjd } 1812168404Spjd 1813168404Spjd if (smo->smo_object == 0) { 1814168404Spjd ASSERT(smo->smo_objsize == 0); 1815168404Spjd ASSERT(smo->smo_alloc == 0); 1816168404Spjd smo->smo_object = dmu_object_alloc(mos, 1817168404Spjd DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1818168404Spjd DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1819168404Spjd ASSERT(smo->smo_object != 0); 1820168404Spjd vdev_config_dirty(vd->vdev_top); 1821168404Spjd } 1822168404Spjd 1823168404Spjd mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1824168404Spjd 1825168404Spjd space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1826168404Spjd &smlock); 1827168404Spjd 1828168404Spjd mutex_enter(&smlock); 1829168404Spjd 1830168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1831168404Spjd space_map_walk(sm, space_map_add, &smsync); 1832168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1833168404Spjd 1834168404Spjd space_map_truncate(smo, mos, tx); 1835168404Spjd space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1836168404Spjd 1837168404Spjd space_map_destroy(&smsync); 1838168404Spjd 1839168404Spjd mutex_exit(&smlock); 1840168404Spjd mutex_destroy(&smlock); 1841168404Spjd 1842168404Spjd VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1843168404Spjd dmu_buf_will_dirty(db, tx); 1844185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1845185029Spjd bcopy(smo, db->db_data, sizeof (*smo)); 1846168404Spjd dmu_buf_rele(db, FTAG); 1847168404Spjd 1848168404Spjd dmu_tx_commit(tx); 1849168404Spjd} 1850168404Spjd 1851185029Spjd/* 1852209962Smm * Determine whether the specified vdev can be offlined/detached/removed 1853209962Smm * without losing data. 1854209962Smm */ 1855209962Smmboolean_t 1856209962Smmvdev_dtl_required(vdev_t *vd) 1857209962Smm{ 1858209962Smm spa_t *spa = vd->vdev_spa; 1859209962Smm vdev_t *tvd = vd->vdev_top; 1860209962Smm uint8_t cant_read = vd->vdev_cant_read; 1861209962Smm boolean_t required; 1862209962Smm 1863209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1864209962Smm 1865209962Smm if (vd == spa->spa_root_vdev || vd == tvd) 1866209962Smm return (B_TRUE); 1867209962Smm 1868209962Smm /* 1869209962Smm * Temporarily mark the device as unreadable, and then determine 1870209962Smm * whether this results in any DTL outages in the top-level vdev. 1871209962Smm * If not, we can safely offline/detach/remove the device. 1872209962Smm */ 1873209962Smm vd->vdev_cant_read = B_TRUE; 1874209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1875209962Smm required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 1876209962Smm vd->vdev_cant_read = cant_read; 1877209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1878209962Smm 1879219089Spjd if (!required && zio_injection_enabled) 1880219089Spjd required = !!zio_handle_device_injection(vd, NULL, ECHILD); 1881219089Spjd 1882209962Smm return (required); 1883209962Smm} 1884209962Smm 1885209962Smm/* 1886185029Spjd * Determine if resilver is needed, and if so the txg range. 1887185029Spjd */ 1888185029Spjdboolean_t 1889185029Spjdvdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1890185029Spjd{ 1891185029Spjd boolean_t needed = B_FALSE; 1892185029Spjd uint64_t thismin = UINT64_MAX; 1893185029Spjd uint64_t thismax = 0; 1894185029Spjd 1895185029Spjd if (vd->vdev_children == 0) { 1896185029Spjd mutex_enter(&vd->vdev_dtl_lock); 1897209962Smm if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 1898209962Smm vdev_writeable(vd)) { 1899185029Spjd space_seg_t *ss; 1900185029Spjd 1901209962Smm ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1902185029Spjd thismin = ss->ss_start - 1; 1903209962Smm ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1904185029Spjd thismax = ss->ss_end; 1905185029Spjd needed = B_TRUE; 1906185029Spjd } 1907185029Spjd mutex_exit(&vd->vdev_dtl_lock); 1908185029Spjd } else { 1909209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1910185029Spjd vdev_t *cvd = vd->vdev_child[c]; 1911185029Spjd uint64_t cmin, cmax; 1912185029Spjd 1913185029Spjd if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1914185029Spjd thismin = MIN(thismin, cmin); 1915185029Spjd thismax = MAX(thismax, cmax); 1916185029Spjd needed = B_TRUE; 1917185029Spjd } 1918185029Spjd } 1919185029Spjd } 1920185029Spjd 1921185029Spjd if (needed && minp) { 1922185029Spjd *minp = thismin; 1923185029Spjd *maxp = thismax; 1924185029Spjd } 1925185029Spjd return (needed); 1926185029Spjd} 1927185029Spjd 1928168404Spjdvoid 1929168404Spjdvdev_load(vdev_t *vd) 1930168404Spjd{ 1931168404Spjd /* 1932168404Spjd * Recursively load all children. 1933168404Spjd */ 1934209962Smm for (int c = 0; c < vd->vdev_children; c++) 1935168404Spjd vdev_load(vd->vdev_child[c]); 1936168404Spjd 1937168404Spjd /* 1938168404Spjd * If this is a top-level vdev, initialize its metaslabs. 1939168404Spjd */ 1940219089Spjd if (vd == vd->vdev_top && !vd->vdev_ishole && 1941168404Spjd (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1942168404Spjd vdev_metaslab_init(vd, 0) != 0)) 1943168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1944168404Spjd VDEV_AUX_CORRUPT_DATA); 1945168404Spjd 1946168404Spjd /* 1947168404Spjd * If this is a leaf vdev, load its DTL. 1948168404Spjd */ 1949168404Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1950168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1951168404Spjd VDEV_AUX_CORRUPT_DATA); 1952168404Spjd} 1953168404Spjd 1954168404Spjd/* 1955185029Spjd * The special vdev case is used for hot spares and l2cache devices. Its 1956185029Spjd * sole purpose it to set the vdev state for the associated vdev. To do this, 1957185029Spjd * we make sure that we can open the underlying device, then try to read the 1958185029Spjd * label, and make sure that the label is sane and that it hasn't been 1959185029Spjd * repurposed to another pool. 1960168404Spjd */ 1961168404Spjdint 1962185029Spjdvdev_validate_aux(vdev_t *vd) 1963168404Spjd{ 1964168404Spjd nvlist_t *label; 1965168404Spjd uint64_t guid, version; 1966168404Spjd uint64_t state; 1967168404Spjd 1968185029Spjd if (!vdev_readable(vd)) 1969185029Spjd return (0); 1970185029Spjd 1971168404Spjd if ((label = vdev_label_read_config(vd)) == NULL) { 1972168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1973168404Spjd VDEV_AUX_CORRUPT_DATA); 1974168404Spjd return (-1); 1975168404Spjd } 1976168404Spjd 1977168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1978185029Spjd version > SPA_VERSION || 1979168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 1980168404Spjd guid != vd->vdev_guid || 1981168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 1982168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1983168404Spjd VDEV_AUX_CORRUPT_DATA); 1984168404Spjd nvlist_free(label); 1985168404Spjd return (-1); 1986168404Spjd } 1987168404Spjd 1988168404Spjd /* 1989168404Spjd * We don't actually check the pool state here. If it's in fact in 1990168404Spjd * use by another pool, we update this fact on the fly when requested. 1991168404Spjd */ 1992168404Spjd nvlist_free(label); 1993168404Spjd return (0); 1994168404Spjd} 1995168404Spjd 1996168404Spjdvoid 1997219089Spjdvdev_remove(vdev_t *vd, uint64_t txg) 1998219089Spjd{ 1999219089Spjd spa_t *spa = vd->vdev_spa; 2000219089Spjd objset_t *mos = spa->spa_meta_objset; 2001219089Spjd dmu_tx_t *tx; 2002219089Spjd 2003219089Spjd tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2004219089Spjd 2005219089Spjd if (vd->vdev_dtl_smo.smo_object) { 2006219089Spjd ASSERT3U(vd->vdev_dtl_smo.smo_alloc, ==, 0); 2007219089Spjd (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 2008219089Spjd vd->vdev_dtl_smo.smo_object = 0; 2009219089Spjd } 2010219089Spjd 2011219089Spjd if (vd->vdev_ms != NULL) { 2012219089Spjd for (int m = 0; m < vd->vdev_ms_count; m++) { 2013219089Spjd metaslab_t *msp = vd->vdev_ms[m]; 2014219089Spjd 2015219089Spjd if (msp == NULL || msp->ms_smo.smo_object == 0) 2016219089Spjd continue; 2017219089Spjd 2018219089Spjd ASSERT3U(msp->ms_smo.smo_alloc, ==, 0); 2019219089Spjd (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 2020219089Spjd msp->ms_smo.smo_object = 0; 2021219089Spjd } 2022219089Spjd } 2023219089Spjd 2024219089Spjd if (vd->vdev_ms_array) { 2025219089Spjd (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 2026219089Spjd vd->vdev_ms_array = 0; 2027219089Spjd vd->vdev_ms_shift = 0; 2028219089Spjd } 2029219089Spjd dmu_tx_commit(tx); 2030219089Spjd} 2031219089Spjd 2032219089Spjdvoid 2033168404Spjdvdev_sync_done(vdev_t *vd, uint64_t txg) 2034168404Spjd{ 2035168404Spjd metaslab_t *msp; 2036211931Smm boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2037168404Spjd 2038219089Spjd ASSERT(!vd->vdev_ishole); 2039219089Spjd 2040168404Spjd while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2041168404Spjd metaslab_sync_done(msp, txg); 2042211931Smm 2043211931Smm if (reassess) 2044211931Smm metaslab_sync_reassess(vd->vdev_mg); 2045168404Spjd} 2046168404Spjd 2047168404Spjdvoid 2048168404Spjdvdev_sync(vdev_t *vd, uint64_t txg) 2049168404Spjd{ 2050168404Spjd spa_t *spa = vd->vdev_spa; 2051168404Spjd vdev_t *lvd; 2052168404Spjd metaslab_t *msp; 2053168404Spjd dmu_tx_t *tx; 2054168404Spjd 2055219089Spjd ASSERT(!vd->vdev_ishole); 2056219089Spjd 2057168404Spjd if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2058168404Spjd ASSERT(vd == vd->vdev_top); 2059168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2060168404Spjd vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2061168404Spjd DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2062168404Spjd ASSERT(vd->vdev_ms_array != 0); 2063168404Spjd vdev_config_dirty(vd); 2064168404Spjd dmu_tx_commit(tx); 2065168404Spjd } 2066168404Spjd 2067219089Spjd /* 2068219089Spjd * Remove the metadata associated with this vdev once it's empty. 2069219089Spjd */ 2070219089Spjd if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 2071219089Spjd vdev_remove(vd, txg); 2072219089Spjd 2073168404Spjd while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2074168404Spjd metaslab_sync(msp, txg); 2075168404Spjd (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2076168404Spjd } 2077168404Spjd 2078168404Spjd while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2079168404Spjd vdev_dtl_sync(lvd, txg); 2080168404Spjd 2081168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2082168404Spjd} 2083168404Spjd 2084168404Spjduint64_t 2085168404Spjdvdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2086168404Spjd{ 2087168404Spjd return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2088168404Spjd} 2089168404Spjd 2090185029Spjd/* 2091185029Spjd * Mark the given vdev faulted. A faulted vdev behaves as if the device could 2092185029Spjd * not be opened, and no I/O is attempted. 2093185029Spjd */ 2094185029Spjdint 2095219089Spjdvdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2096168404Spjd{ 2097219089Spjd vdev_t *vd, *tvd; 2098168404Spjd 2099219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2100185029Spjd 2101185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2102185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2103185029Spjd 2104185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 2105185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2106185029Spjd 2107219089Spjd tvd = vd->vdev_top; 2108219089Spjd 2109185029Spjd /* 2110219089Spjd * We don't directly use the aux state here, but if we do a 2111219089Spjd * vdev_reopen(), we need this value to be present to remember why we 2112219089Spjd * were faulted. 2113219089Spjd */ 2114219089Spjd vd->vdev_label_aux = aux; 2115219089Spjd 2116219089Spjd /* 2117185029Spjd * Faulted state takes precedence over degraded. 2118185029Spjd */ 2119219089Spjd vd->vdev_delayed_close = B_FALSE; 2120185029Spjd vd->vdev_faulted = 1ULL; 2121185029Spjd vd->vdev_degraded = 0ULL; 2122219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 2123185029Spjd 2124185029Spjd /* 2125219089Spjd * If this device has the only valid copy of the data, then 2126219089Spjd * back off and simply mark the vdev as degraded instead. 2127185029Spjd */ 2128219089Spjd if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 2129185029Spjd vd->vdev_degraded = 1ULL; 2130185029Spjd vd->vdev_faulted = 0ULL; 2131185029Spjd 2132185029Spjd /* 2133185029Spjd * If we reopen the device and it's not dead, only then do we 2134185029Spjd * mark it degraded. 2135185029Spjd */ 2136219089Spjd vdev_reopen(tvd); 2137185029Spjd 2138219089Spjd if (vdev_readable(vd)) 2139219089Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 2140185029Spjd } 2141185029Spjd 2142185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2143168404Spjd} 2144168404Spjd 2145185029Spjd/* 2146185029Spjd * Mark the given vdev degraded. A degraded vdev is purely an indication to the 2147185029Spjd * user that something is wrong. The vdev continues to operate as normal as far 2148185029Spjd * as I/O is concerned. 2149185029Spjd */ 2150185029Spjdint 2151219089Spjdvdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2152168404Spjd{ 2153185029Spjd vdev_t *vd; 2154168404Spjd 2155219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2156168404Spjd 2157185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2158185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2159168404Spjd 2160185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 2161185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2162185029Spjd 2163185029Spjd /* 2164185029Spjd * If the vdev is already faulted, then don't do anything. 2165185029Spjd */ 2166185029Spjd if (vd->vdev_faulted || vd->vdev_degraded) 2167185029Spjd return (spa_vdev_state_exit(spa, NULL, 0)); 2168185029Spjd 2169185029Spjd vd->vdev_degraded = 1ULL; 2170185029Spjd if (!vdev_is_dead(vd)) 2171185029Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2172219089Spjd aux); 2173185029Spjd 2174185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2175168404Spjd} 2176168404Spjd 2177185029Spjd/* 2178185029Spjd * Online the given vdev. If 'unspare' is set, it implies two things. First, 2179185029Spjd * any attached spare device should be detached when the device finishes 2180185029Spjd * resilvering. Second, the online should be treated like a 'test' online case, 2181185029Spjd * so no FMA events are generated if the device fails to open. 2182185029Spjd */ 2183168404Spjdint 2184185029Spjdvdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 2185168404Spjd{ 2186219089Spjd vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2187168404Spjd 2188219089Spjd spa_vdev_state_enter(spa, SCL_NONE); 2189168404Spjd 2190185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2191185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2192168404Spjd 2193168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 2194185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2195168404Spjd 2196219089Spjd tvd = vd->vdev_top; 2197168404Spjd vd->vdev_offline = B_FALSE; 2198168404Spjd vd->vdev_tmpoffline = B_FALSE; 2199185029Spjd vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2200185029Spjd vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2201219089Spjd 2202219089Spjd /* XXX - L2ARC 1.0 does not support expansion */ 2203219089Spjd if (!vd->vdev_aux) { 2204219089Spjd for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2205219089Spjd pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2206219089Spjd } 2207219089Spjd 2208219089Spjd vdev_reopen(tvd); 2209185029Spjd vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 2210168404Spjd 2211219089Spjd if (!vd->vdev_aux) { 2212219089Spjd for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2213219089Spjd pvd->vdev_expanding = B_FALSE; 2214219089Spjd } 2215219089Spjd 2216185029Spjd if (newstate) 2217185029Spjd *newstate = vd->vdev_state; 2218185029Spjd if ((flags & ZFS_ONLINE_UNSPARE) && 2219185029Spjd !vdev_is_dead(vd) && vd->vdev_parent && 2220185029Spjd vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2221185029Spjd vd->vdev_parent->vdev_child[0] == vd) 2222185029Spjd vd->vdev_unspare = B_TRUE; 2223168404Spjd 2224219089Spjd if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2225219089Spjd 2226219089Spjd /* XXX - L2ARC 1.0 does not support expansion */ 2227219089Spjd if (vd->vdev_aux) 2228219089Spjd return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2229219089Spjd spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2230219089Spjd } 2231209962Smm return (spa_vdev_state_exit(spa, vd, 0)); 2232168404Spjd} 2233168404Spjd 2234219089Spjdstatic int 2235219089Spjdvdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2236168404Spjd{ 2237213197Smm vdev_t *vd, *tvd; 2238219089Spjd int error = 0; 2239219089Spjd uint64_t generation; 2240219089Spjd metaslab_group_t *mg; 2241168404Spjd 2242219089Spjdtop: 2243219089Spjd spa_vdev_state_enter(spa, SCL_ALLOC); 2244168404Spjd 2245185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2246185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2247168404Spjd 2248168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 2249185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2250168404Spjd 2251213197Smm tvd = vd->vdev_top; 2252219089Spjd mg = tvd->vdev_mg; 2253219089Spjd generation = spa->spa_config_generation + 1; 2254213197Smm 2255168404Spjd /* 2256168404Spjd * If the device isn't already offline, try to offline it. 2257168404Spjd */ 2258168404Spjd if (!vd->vdev_offline) { 2259168404Spjd /* 2260209962Smm * If this device has the only valid copy of some data, 2261213197Smm * don't allow it to be offlined. Log devices are always 2262213197Smm * expendable. 2263168404Spjd */ 2264213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2265213197Smm vdev_dtl_required(vd)) 2266185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2267168404Spjd 2268168404Spjd /* 2269219089Spjd * If the top-level is a slog and it has had allocations 2270219089Spjd * then proceed. We check that the vdev's metaslab group 2271219089Spjd * is not NULL since it's possible that we may have just 2272219089Spjd * added this vdev but not yet initialized its metaslabs. 2273219089Spjd */ 2274219089Spjd if (tvd->vdev_islog && mg != NULL) { 2275219089Spjd /* 2276219089Spjd * Prevent any future allocations. 2277219089Spjd */ 2278219089Spjd metaslab_group_passivate(mg); 2279219089Spjd (void) spa_vdev_state_exit(spa, vd, 0); 2280219089Spjd 2281219089Spjd error = spa_offline_log(spa); 2282219089Spjd 2283219089Spjd spa_vdev_state_enter(spa, SCL_ALLOC); 2284219089Spjd 2285219089Spjd /* 2286219089Spjd * Check to see if the config has changed. 2287219089Spjd */ 2288219089Spjd if (error || generation != spa->spa_config_generation) { 2289219089Spjd metaslab_group_activate(mg); 2290219089Spjd if (error) 2291219089Spjd return (spa_vdev_state_exit(spa, 2292219089Spjd vd, error)); 2293219089Spjd (void) spa_vdev_state_exit(spa, vd, 0); 2294219089Spjd goto top; 2295219089Spjd } 2296219089Spjd ASSERT3U(tvd->vdev_stat.vs_alloc, ==, 0); 2297219089Spjd } 2298219089Spjd 2299219089Spjd /* 2300168404Spjd * Offline this device and reopen its top-level vdev. 2301213197Smm * If the top-level vdev is a log device then just offline 2302213197Smm * it. Otherwise, if this action results in the top-level 2303213197Smm * vdev becoming unusable, undo it and fail the request. 2304168404Spjd */ 2305168404Spjd vd->vdev_offline = B_TRUE; 2306213197Smm vdev_reopen(tvd); 2307213197Smm 2308213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2309213197Smm vdev_is_dead(tvd)) { 2310168404Spjd vd->vdev_offline = B_FALSE; 2311213197Smm vdev_reopen(tvd); 2312185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2313168404Spjd } 2314219089Spjd 2315219089Spjd /* 2316219089Spjd * Add the device back into the metaslab rotor so that 2317219089Spjd * once we online the device it's open for business. 2318219089Spjd */ 2319219089Spjd if (tvd->vdev_islog && mg != NULL) 2320219089Spjd metaslab_group_activate(mg); 2321168404Spjd } 2322168404Spjd 2323185029Spjd vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2324168404Spjd 2325219089Spjd return (spa_vdev_state_exit(spa, vd, 0)); 2326219089Spjd} 2327213197Smm 2328219089Spjdint 2329219089Spjdvdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2330219089Spjd{ 2331219089Spjd int error; 2332213197Smm 2333219089Spjd mutex_enter(&spa->spa_vdev_top_lock); 2334219089Spjd error = vdev_offline_locked(spa, guid, flags); 2335219089Spjd mutex_exit(&spa->spa_vdev_top_lock); 2336219089Spjd 2337219089Spjd return (error); 2338168404Spjd} 2339168404Spjd 2340168404Spjd/* 2341168404Spjd * Clear the error counts associated with this vdev. Unlike vdev_online() and 2342168404Spjd * vdev_offline(), we assume the spa config is locked. We also clear all 2343168404Spjd * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2344168404Spjd */ 2345168404Spjdvoid 2346168404Spjdvdev_clear(spa_t *spa, vdev_t *vd) 2347168404Spjd{ 2348185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2349168404Spjd 2350185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2351185029Spjd 2352168404Spjd if (vd == NULL) 2353185029Spjd vd = rvd; 2354168404Spjd 2355168404Spjd vd->vdev_stat.vs_read_errors = 0; 2356168404Spjd vd->vdev_stat.vs_write_errors = 0; 2357168404Spjd vd->vdev_stat.vs_checksum_errors = 0; 2358168404Spjd 2359185029Spjd for (int c = 0; c < vd->vdev_children; c++) 2360168404Spjd vdev_clear(spa, vd->vdev_child[c]); 2361185029Spjd 2362185029Spjd /* 2363185029Spjd * If we're in the FAULTED state or have experienced failed I/O, then 2364185029Spjd * clear the persistent state and attempt to reopen the device. We 2365185029Spjd * also mark the vdev config dirty, so that the new faulted state is 2366185029Spjd * written out to disk. 2367185029Spjd */ 2368185029Spjd if (vd->vdev_faulted || vd->vdev_degraded || 2369185029Spjd !vdev_readable(vd) || !vdev_writeable(vd)) { 2370185029Spjd 2371219089Spjd /* 2372219089Spjd * When reopening in reponse to a clear event, it may be due to 2373219089Spjd * a fmadm repair request. In this case, if the device is 2374219089Spjd * still broken, we want to still post the ereport again. 2375219089Spjd */ 2376219089Spjd vd->vdev_forcefault = B_TRUE; 2377219089Spjd 2378219089Spjd vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2379185029Spjd vd->vdev_cant_read = B_FALSE; 2380185029Spjd vd->vdev_cant_write = B_FALSE; 2381185029Spjd 2382219089Spjd vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 2383185029Spjd 2384219089Spjd vd->vdev_forcefault = B_FALSE; 2385219089Spjd 2386219089Spjd if (vd != rvd && vdev_writeable(vd->vdev_top)) 2387185029Spjd vdev_state_dirty(vd->vdev_top); 2388185029Spjd 2389185029Spjd if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2390185029Spjd spa_async_request(spa, SPA_ASYNC_RESILVER); 2391185029Spjd 2392185029Spjd spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2393185029Spjd } 2394219089Spjd 2395219089Spjd /* 2396219089Spjd * When clearing a FMA-diagnosed fault, we always want to 2397219089Spjd * unspare the device, as we assume that the original spare was 2398219089Spjd * done in response to the FMA fault. 2399219089Spjd */ 2400219089Spjd if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2401219089Spjd vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2402219089Spjd vd->vdev_parent->vdev_child[0] == vd) 2403219089Spjd vd->vdev_unspare = B_TRUE; 2404168404Spjd} 2405168404Spjd 2406185029Spjdboolean_t 2407168404Spjdvdev_is_dead(vdev_t *vd) 2408168404Spjd{ 2409219089Spjd /* 2410219089Spjd * Holes and missing devices are always considered "dead". 2411219089Spjd * This simplifies the code since we don't have to check for 2412219089Spjd * these types of devices in the various code paths. 2413219089Spjd * Instead we rely on the fact that we skip over dead devices 2414219089Spjd * before issuing I/O to them. 2415219089Spjd */ 2416219089Spjd return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 2417219089Spjd vd->vdev_ops == &vdev_missing_ops); 2418168404Spjd} 2419168404Spjd 2420185029Spjdboolean_t 2421185029Spjdvdev_readable(vdev_t *vd) 2422168404Spjd{ 2423185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2424185029Spjd} 2425168404Spjd 2426185029Spjdboolean_t 2427185029Spjdvdev_writeable(vdev_t *vd) 2428185029Spjd{ 2429185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2430185029Spjd} 2431168404Spjd 2432185029Spjdboolean_t 2433208370Smmvdev_allocatable(vdev_t *vd) 2434208370Smm{ 2435209962Smm uint64_t state = vd->vdev_state; 2436209962Smm 2437208370Smm /* 2438209962Smm * We currently allow allocations from vdevs which may be in the 2439208370Smm * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2440208370Smm * fails to reopen then we'll catch it later when we're holding 2441209962Smm * the proper locks. Note that we have to get the vdev state 2442209962Smm * in a local variable because although it changes atomically, 2443209962Smm * we're asking two separate questions about it. 2444208370Smm */ 2445209962Smm return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2446219089Spjd !vd->vdev_cant_write && !vd->vdev_ishole); 2447208370Smm} 2448208370Smm 2449208370Smmboolean_t 2450185029Spjdvdev_accessible(vdev_t *vd, zio_t *zio) 2451185029Spjd{ 2452185029Spjd ASSERT(zio->io_vd == vd); 2453168404Spjd 2454185029Spjd if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2455185029Spjd return (B_FALSE); 2456168404Spjd 2457185029Spjd if (zio->io_type == ZIO_TYPE_READ) 2458185029Spjd return (!vd->vdev_cant_read); 2459168404Spjd 2460185029Spjd if (zio->io_type == ZIO_TYPE_WRITE) 2461185029Spjd return (!vd->vdev_cant_write); 2462168404Spjd 2463185029Spjd return (B_TRUE); 2464168404Spjd} 2465168404Spjd 2466168404Spjd/* 2467168404Spjd * Get statistics for the given vdev. 2468168404Spjd */ 2469168404Spjdvoid 2470168404Spjdvdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2471168404Spjd{ 2472168404Spjd vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2473168404Spjd 2474168404Spjd mutex_enter(&vd->vdev_stat_lock); 2475168404Spjd bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2476168404Spjd vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2477168404Spjd vs->vs_state = vd->vdev_state; 2478219089Spjd vs->vs_rsize = vdev_get_min_asize(vd); 2479219089Spjd if (vd->vdev_ops->vdev_op_leaf) 2480219089Spjd vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2481168404Spjd mutex_exit(&vd->vdev_stat_lock); 2482168404Spjd 2483168404Spjd /* 2484168404Spjd * If we're getting stats on the root vdev, aggregate the I/O counts 2485168404Spjd * over all top-level vdevs (i.e. the direct children of the root). 2486168404Spjd */ 2487168404Spjd if (vd == rvd) { 2488185029Spjd for (int c = 0; c < rvd->vdev_children; c++) { 2489168404Spjd vdev_t *cvd = rvd->vdev_child[c]; 2490168404Spjd vdev_stat_t *cvs = &cvd->vdev_stat; 2491168404Spjd 2492168404Spjd mutex_enter(&vd->vdev_stat_lock); 2493185029Spjd for (int t = 0; t < ZIO_TYPES; t++) { 2494168404Spjd vs->vs_ops[t] += cvs->vs_ops[t]; 2495168404Spjd vs->vs_bytes[t] += cvs->vs_bytes[t]; 2496168404Spjd } 2497219089Spjd cvs->vs_scan_removing = cvd->vdev_removing; 2498168404Spjd mutex_exit(&vd->vdev_stat_lock); 2499168404Spjd } 2500168404Spjd } 2501168404Spjd} 2502168404Spjd 2503168404Spjdvoid 2504185029Spjdvdev_clear_stats(vdev_t *vd) 2505168404Spjd{ 2506185029Spjd mutex_enter(&vd->vdev_stat_lock); 2507185029Spjd vd->vdev_stat.vs_space = 0; 2508185029Spjd vd->vdev_stat.vs_dspace = 0; 2509185029Spjd vd->vdev_stat.vs_alloc = 0; 2510185029Spjd mutex_exit(&vd->vdev_stat_lock); 2511185029Spjd} 2512185029Spjd 2513185029Spjdvoid 2514219089Spjdvdev_scan_stat_init(vdev_t *vd) 2515219089Spjd{ 2516219089Spjd vdev_stat_t *vs = &vd->vdev_stat; 2517219089Spjd 2518219089Spjd for (int c = 0; c < vd->vdev_children; c++) 2519219089Spjd vdev_scan_stat_init(vd->vdev_child[c]); 2520219089Spjd 2521219089Spjd mutex_enter(&vd->vdev_stat_lock); 2522219089Spjd vs->vs_scan_processed = 0; 2523219089Spjd mutex_exit(&vd->vdev_stat_lock); 2524219089Spjd} 2525219089Spjd 2526219089Spjdvoid 2527185029Spjdvdev_stat_update(zio_t *zio, uint64_t psize) 2528185029Spjd{ 2529209962Smm spa_t *spa = zio->io_spa; 2530209962Smm vdev_t *rvd = spa->spa_root_vdev; 2531185029Spjd vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2532168404Spjd vdev_t *pvd; 2533168404Spjd uint64_t txg = zio->io_txg; 2534168404Spjd vdev_stat_t *vs = &vd->vdev_stat; 2535168404Spjd zio_type_t type = zio->io_type; 2536168404Spjd int flags = zio->io_flags; 2537168404Spjd 2538185029Spjd /* 2539185029Spjd * If this i/o is a gang leader, it didn't do any actual work. 2540185029Spjd */ 2541185029Spjd if (zio->io_gang_tree) 2542185029Spjd return; 2543185029Spjd 2544168404Spjd if (zio->io_error == 0) { 2545185029Spjd /* 2546185029Spjd * If this is a root i/o, don't count it -- we've already 2547185029Spjd * counted the top-level vdevs, and vdev_get_stats() will 2548185029Spjd * aggregate them when asked. This reduces contention on 2549185029Spjd * the root vdev_stat_lock and implicitly handles blocks 2550185029Spjd * that compress away to holes, for which there is no i/o. 2551185029Spjd * (Holes never create vdev children, so all the counters 2552185029Spjd * remain zero, which is what we want.) 2553185029Spjd * 2554185029Spjd * Note: this only applies to successful i/o (io_error == 0) 2555185029Spjd * because unlike i/o counts, errors are not additive. 2556185029Spjd * When reading a ditto block, for example, failure of 2557185029Spjd * one top-level vdev does not imply a root-level error. 2558185029Spjd */ 2559185029Spjd if (vd == rvd) 2560185029Spjd return; 2561185029Spjd 2562185029Spjd ASSERT(vd == zio->io_vd); 2563209962Smm 2564209962Smm if (flags & ZIO_FLAG_IO_BYPASS) 2565209962Smm return; 2566209962Smm 2567209962Smm mutex_enter(&vd->vdev_stat_lock); 2568209962Smm 2569185029Spjd if (flags & ZIO_FLAG_IO_REPAIR) { 2570219089Spjd if (flags & ZIO_FLAG_SCAN_THREAD) { 2571219089Spjd dsl_scan_phys_t *scn_phys = 2572219089Spjd &spa->spa_dsl_pool->dp_scan->scn_phys; 2573219089Spjd uint64_t *processed = &scn_phys->scn_processed; 2574219089Spjd 2575219089Spjd /* XXX cleanup? */ 2576219089Spjd if (vd->vdev_ops->vdev_op_leaf) 2577219089Spjd atomic_add_64(processed, psize); 2578219089Spjd vs->vs_scan_processed += psize; 2579219089Spjd } 2580219089Spjd 2581209962Smm if (flags & ZIO_FLAG_SELF_HEAL) 2582185029Spjd vs->vs_self_healed += psize; 2583168404Spjd } 2584209962Smm 2585209962Smm vs->vs_ops[type]++; 2586209962Smm vs->vs_bytes[type] += psize; 2587209962Smm 2588209962Smm mutex_exit(&vd->vdev_stat_lock); 2589168404Spjd return; 2590168404Spjd } 2591168404Spjd 2592168404Spjd if (flags & ZIO_FLAG_SPECULATIVE) 2593168404Spjd return; 2594168404Spjd 2595213198Smm /* 2596213198Smm * If this is an I/O error that is going to be retried, then ignore the 2597213198Smm * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2598213198Smm * hard errors, when in reality they can happen for any number of 2599213198Smm * innocuous reasons (bus resets, MPxIO link failure, etc). 2600213198Smm */ 2601213198Smm if (zio->io_error == EIO && 2602213198Smm !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2603213198Smm return; 2604213198Smm 2605219089Spjd /* 2606219089Spjd * Intent logs writes won't propagate their error to the root 2607219089Spjd * I/O so don't mark these types of failures as pool-level 2608219089Spjd * errors. 2609219089Spjd */ 2610219089Spjd if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 2611219089Spjd return; 2612219089Spjd 2613185029Spjd mutex_enter(&vd->vdev_stat_lock); 2614209962Smm if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2615185029Spjd if (zio->io_error == ECKSUM) 2616185029Spjd vs->vs_checksum_errors++; 2617185029Spjd else 2618185029Spjd vs->vs_read_errors++; 2619168404Spjd } 2620209962Smm if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2621185029Spjd vs->vs_write_errors++; 2622185029Spjd mutex_exit(&vd->vdev_stat_lock); 2623168404Spjd 2624209962Smm if (type == ZIO_TYPE_WRITE && txg != 0 && 2625209962Smm (!(flags & ZIO_FLAG_IO_REPAIR) || 2626219089Spjd (flags & ZIO_FLAG_SCAN_THREAD) || 2627219089Spjd spa->spa_claiming)) { 2628209962Smm /* 2629219089Spjd * This is either a normal write (not a repair), or it's 2630219089Spjd * a repair induced by the scrub thread, or it's a repair 2631219089Spjd * made by zil_claim() during spa_load() in the first txg. 2632219089Spjd * In the normal case, we commit the DTL change in the same 2633219089Spjd * txg as the block was born. In the scrub-induced repair 2634219089Spjd * case, we know that scrubs run in first-pass syncing context, 2635219089Spjd * so we commit the DTL change in spa_syncing_txg(spa). 2636219089Spjd * In the zil_claim() case, we commit in spa_first_txg(spa). 2637209962Smm * 2638209962Smm * We currently do not make DTL entries for failed spontaneous 2639209962Smm * self-healing writes triggered by normal (non-scrubbing) 2640209962Smm * reads, because we have no transactional context in which to 2641209962Smm * do so -- and it's not clear that it'd be desirable anyway. 2642209962Smm */ 2643209962Smm if (vd->vdev_ops->vdev_op_leaf) { 2644209962Smm uint64_t commit_txg = txg; 2645219089Spjd if (flags & ZIO_FLAG_SCAN_THREAD) { 2646209962Smm ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2647209962Smm ASSERT(spa_sync_pass(spa) == 1); 2648209962Smm vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2649219089Spjd commit_txg = spa_syncing_txg(spa); 2650219089Spjd } else if (spa->spa_claiming) { 2651219089Spjd ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2652219089Spjd commit_txg = spa_first_txg(spa); 2653209962Smm } 2654219089Spjd ASSERT(commit_txg >= spa_syncing_txg(spa)); 2655209962Smm if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2656168404Spjd return; 2657209962Smm for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2658209962Smm vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2659209962Smm vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2660168404Spjd } 2661209962Smm if (vd != rvd) 2662209962Smm vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2663168404Spjd } 2664168404Spjd} 2665168404Spjd 2666168404Spjd/* 2667219089Spjd * Update the in-core space usage stats for this vdev, its metaslab class, 2668219089Spjd * and the root vdev. 2669168404Spjd */ 2670168404Spjdvoid 2671219089Spjdvdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2672219089Spjd int64_t space_delta) 2673168404Spjd{ 2674168404Spjd int64_t dspace_delta = space_delta; 2675185029Spjd spa_t *spa = vd->vdev_spa; 2676185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2677219089Spjd metaslab_group_t *mg = vd->vdev_mg; 2678219089Spjd metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2679168404Spjd 2680185029Spjd ASSERT(vd == vd->vdev_top); 2681168404Spjd 2682185029Spjd /* 2683185029Spjd * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2684185029Spjd * factor. We must calculate this here and not at the root vdev 2685185029Spjd * because the root vdev's psize-to-asize is simply the max of its 2686185029Spjd * childrens', thus not accurate enough for us. 2687185029Spjd */ 2688185029Spjd ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2689213197Smm ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2690185029Spjd dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2691185029Spjd vd->vdev_deflate_ratio; 2692185029Spjd 2693185029Spjd mutex_enter(&vd->vdev_stat_lock); 2694219089Spjd vd->vdev_stat.vs_alloc += alloc_delta; 2695185029Spjd vd->vdev_stat.vs_space += space_delta; 2696185029Spjd vd->vdev_stat.vs_dspace += dspace_delta; 2697185029Spjd mutex_exit(&vd->vdev_stat_lock); 2698185029Spjd 2699219089Spjd if (mc == spa_normal_class(spa)) { 2700185029Spjd mutex_enter(&rvd->vdev_stat_lock); 2701219089Spjd rvd->vdev_stat.vs_alloc += alloc_delta; 2702185029Spjd rvd->vdev_stat.vs_space += space_delta; 2703185029Spjd rvd->vdev_stat.vs_dspace += dspace_delta; 2704185029Spjd mutex_exit(&rvd->vdev_stat_lock); 2705185029Spjd } 2706219089Spjd 2707219089Spjd if (mc != NULL) { 2708219089Spjd ASSERT(rvd == vd->vdev_parent); 2709219089Spjd ASSERT(vd->vdev_ms_count != 0); 2710219089Spjd 2711219089Spjd metaslab_class_space_update(mc, 2712219089Spjd alloc_delta, defer_delta, space_delta, dspace_delta); 2713219089Spjd } 2714168404Spjd} 2715168404Spjd 2716168404Spjd/* 2717168404Spjd * Mark a top-level vdev's config as dirty, placing it on the dirty list 2718168404Spjd * so that it will be written out next time the vdev configuration is synced. 2719168404Spjd * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2720168404Spjd */ 2721168404Spjdvoid 2722168404Spjdvdev_config_dirty(vdev_t *vd) 2723168404Spjd{ 2724168404Spjd spa_t *spa = vd->vdev_spa; 2725168404Spjd vdev_t *rvd = spa->spa_root_vdev; 2726168404Spjd int c; 2727168404Spjd 2728219089Spjd ASSERT(spa_writeable(spa)); 2729219089Spjd 2730168404Spjd /* 2731209962Smm * If this is an aux vdev (as with l2cache and spare devices), then we 2732209962Smm * update the vdev config manually and set the sync flag. 2733185029Spjd */ 2734185029Spjd if (vd->vdev_aux != NULL) { 2735185029Spjd spa_aux_vdev_t *sav = vd->vdev_aux; 2736185029Spjd nvlist_t **aux; 2737185029Spjd uint_t naux; 2738185029Spjd 2739185029Spjd for (c = 0; c < sav->sav_count; c++) { 2740185029Spjd if (sav->sav_vdevs[c] == vd) 2741185029Spjd break; 2742185029Spjd } 2743185029Spjd 2744185029Spjd if (c == sav->sav_count) { 2745185029Spjd /* 2746185029Spjd * We're being removed. There's nothing more to do. 2747185029Spjd */ 2748185029Spjd ASSERT(sav->sav_sync == B_TRUE); 2749185029Spjd return; 2750185029Spjd } 2751185029Spjd 2752185029Spjd sav->sav_sync = B_TRUE; 2753185029Spjd 2754209962Smm if (nvlist_lookup_nvlist_array(sav->sav_config, 2755209962Smm ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 2756209962Smm VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2757209962Smm ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 2758209962Smm } 2759185029Spjd 2760185029Spjd ASSERT(c < naux); 2761185029Spjd 2762185029Spjd /* 2763185029Spjd * Setting the nvlist in the middle if the array is a little 2764185029Spjd * sketchy, but it will work. 2765185029Spjd */ 2766185029Spjd nvlist_free(aux[c]); 2767219089Spjd aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2768185029Spjd 2769185029Spjd return; 2770185029Spjd } 2771185029Spjd 2772185029Spjd /* 2773185029Spjd * The dirty list is protected by the SCL_CONFIG lock. The caller 2774185029Spjd * must either hold SCL_CONFIG as writer, or must be the sync thread 2775185029Spjd * (which holds SCL_CONFIG as reader). There's only one sync thread, 2776168404Spjd * so this is sufficient to ensure mutual exclusion. 2777168404Spjd */ 2778185029Spjd ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2779185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2780185029Spjd spa_config_held(spa, SCL_CONFIG, RW_READER))); 2781168404Spjd 2782168404Spjd if (vd == rvd) { 2783168404Spjd for (c = 0; c < rvd->vdev_children; c++) 2784168404Spjd vdev_config_dirty(rvd->vdev_child[c]); 2785168404Spjd } else { 2786168404Spjd ASSERT(vd == vd->vdev_top); 2787168404Spjd 2788219089Spjd if (!list_link_active(&vd->vdev_config_dirty_node) && 2789219089Spjd !vd->vdev_ishole) 2790185029Spjd list_insert_head(&spa->spa_config_dirty_list, vd); 2791168404Spjd } 2792168404Spjd} 2793168404Spjd 2794168404Spjdvoid 2795168404Spjdvdev_config_clean(vdev_t *vd) 2796168404Spjd{ 2797168404Spjd spa_t *spa = vd->vdev_spa; 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 2803185029Spjd ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2804185029Spjd list_remove(&spa->spa_config_dirty_list, vd); 2805168404Spjd} 2806168404Spjd 2807185029Spjd/* 2808185029Spjd * Mark a top-level vdev's state as dirty, so that the next pass of 2809185029Spjd * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2810185029Spjd * the state changes from larger config changes because they require 2811185029Spjd * much less locking, and are often needed for administrative actions. 2812185029Spjd */ 2813168404Spjdvoid 2814185029Spjdvdev_state_dirty(vdev_t *vd) 2815185029Spjd{ 2816185029Spjd spa_t *spa = vd->vdev_spa; 2817185029Spjd 2818219089Spjd ASSERT(spa_writeable(spa)); 2819185029Spjd ASSERT(vd == vd->vdev_top); 2820185029Spjd 2821185029Spjd /* 2822185029Spjd * The state list is protected by the SCL_STATE lock. The caller 2823185029Spjd * must either hold SCL_STATE as writer, or must be the sync thread 2824185029Spjd * (which holds SCL_STATE as reader). There's only one sync thread, 2825185029Spjd * so this is sufficient to ensure mutual exclusion. 2826185029Spjd */ 2827185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2828185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2829185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2830185029Spjd 2831219089Spjd if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2832185029Spjd list_insert_head(&spa->spa_state_dirty_list, vd); 2833185029Spjd} 2834185029Spjd 2835185029Spjdvoid 2836185029Spjdvdev_state_clean(vdev_t *vd) 2837185029Spjd{ 2838185029Spjd spa_t *spa = vd->vdev_spa; 2839185029Spjd 2840185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2841185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2842185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2843185029Spjd 2844185029Spjd ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2845185029Spjd list_remove(&spa->spa_state_dirty_list, vd); 2846185029Spjd} 2847185029Spjd 2848185029Spjd/* 2849185029Spjd * Propagate vdev state up from children to parent. 2850185029Spjd */ 2851185029Spjdvoid 2852168404Spjdvdev_propagate_state(vdev_t *vd) 2853168404Spjd{ 2854209962Smm spa_t *spa = vd->vdev_spa; 2855209962Smm vdev_t *rvd = spa->spa_root_vdev; 2856168404Spjd int degraded = 0, faulted = 0; 2857168404Spjd int corrupted = 0; 2858168404Spjd vdev_t *child; 2859168404Spjd 2860185029Spjd if (vd->vdev_children > 0) { 2861219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 2862185029Spjd child = vd->vdev_child[c]; 2863168404Spjd 2864219089Spjd /* 2865219089Spjd * Don't factor holes into the decision. 2866219089Spjd */ 2867219089Spjd if (child->vdev_ishole) 2868219089Spjd continue; 2869219089Spjd 2870185029Spjd if (!vdev_readable(child) || 2871209962Smm (!vdev_writeable(child) && spa_writeable(spa))) { 2872185029Spjd /* 2873185029Spjd * Root special: if there is a top-level log 2874185029Spjd * device, treat the root vdev as if it were 2875185029Spjd * degraded. 2876185029Spjd */ 2877185029Spjd if (child->vdev_islog && vd == rvd) 2878185029Spjd degraded++; 2879185029Spjd else 2880185029Spjd faulted++; 2881185029Spjd } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 2882185029Spjd degraded++; 2883185029Spjd } 2884185029Spjd 2885185029Spjd if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 2886185029Spjd corrupted++; 2887185029Spjd } 2888185029Spjd 2889185029Spjd vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 2890185029Spjd 2891185029Spjd /* 2892185029Spjd * Root special: if there is a top-level vdev that cannot be 2893185029Spjd * opened due to corrupted metadata, then propagate the root 2894185029Spjd * vdev's aux state as 'corrupt' rather than 'insufficient 2895185029Spjd * replicas'. 2896185029Spjd */ 2897185029Spjd if (corrupted && vd == rvd && 2898185029Spjd rvd->vdev_state == VDEV_STATE_CANT_OPEN) 2899185029Spjd vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 2900185029Spjd VDEV_AUX_CORRUPT_DATA); 2901168404Spjd } 2902168404Spjd 2903185029Spjd if (vd->vdev_parent) 2904185029Spjd vdev_propagate_state(vd->vdev_parent); 2905168404Spjd} 2906168404Spjd 2907168404Spjd/* 2908168404Spjd * Set a vdev's state. If this is during an open, we don't update the parent 2909168404Spjd * state, because we're in the process of opening children depth-first. 2910168404Spjd * Otherwise, we propagate the change to the parent. 2911168404Spjd * 2912168404Spjd * If this routine places a device in a faulted state, an appropriate ereport is 2913168404Spjd * generated. 2914168404Spjd */ 2915168404Spjdvoid 2916168404Spjdvdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2917168404Spjd{ 2918168404Spjd uint64_t save_state; 2919185029Spjd spa_t *spa = vd->vdev_spa; 2920168404Spjd 2921168404Spjd if (state == vd->vdev_state) { 2922168404Spjd vd->vdev_stat.vs_aux = aux; 2923168404Spjd return; 2924168404Spjd } 2925168404Spjd 2926168404Spjd save_state = vd->vdev_state; 2927168404Spjd 2928168404Spjd vd->vdev_state = state; 2929168404Spjd vd->vdev_stat.vs_aux = aux; 2930168404Spjd 2931173373Spjd /* 2932173373Spjd * If we are setting the vdev state to anything but an open state, then 2933219089Spjd * always close the underlying device unless the device has requested 2934219089Spjd * a delayed close (i.e. we're about to remove or fault the device). 2935219089Spjd * Otherwise, we keep accessible but invalid devices open forever. 2936219089Spjd * We don't call vdev_close() itself, because that implies some extra 2937219089Spjd * checks (offline, etc) that we don't want here. This is limited to 2938219089Spjd * leaf devices, because otherwise closing the device will affect other 2939219089Spjd * children. 2940173373Spjd */ 2941219089Spjd if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 2942219089Spjd vd->vdev_ops->vdev_op_leaf) 2943173373Spjd vd->vdev_ops->vdev_op_close(vd); 2944173373Spjd 2945219089Spjd /* 2946219089Spjd * If we have brought this vdev back into service, we need 2947219089Spjd * to notify fmd so that it can gracefully repair any outstanding 2948219089Spjd * cases due to a missing device. We do this in all cases, even those 2949219089Spjd * that probably don't correlate to a repaired fault. This is sure to 2950219089Spjd * catch all cases, and we let the zfs-retire agent sort it out. If 2951219089Spjd * this is a transient state it's OK, as the retire agent will 2952219089Spjd * double-check the state of the vdev before repairing it. 2953219089Spjd */ 2954219089Spjd if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2955219089Spjd vd->vdev_prevstate != state) 2956219089Spjd zfs_post_state_change(spa, vd); 2957219089Spjd 2958185029Spjd if (vd->vdev_removed && 2959185029Spjd state == VDEV_STATE_CANT_OPEN && 2960185029Spjd (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 2961168404Spjd /* 2962185029Spjd * If the previous state is set to VDEV_STATE_REMOVED, then this 2963185029Spjd * device was previously marked removed and someone attempted to 2964185029Spjd * reopen it. If this failed due to a nonexistent device, then 2965185029Spjd * keep the device in the REMOVED state. We also let this be if 2966185029Spjd * it is one of our special test online cases, which is only 2967185029Spjd * attempting to online the device and shouldn't generate an FMA 2968185029Spjd * fault. 2969185029Spjd */ 2970185029Spjd vd->vdev_state = VDEV_STATE_REMOVED; 2971185029Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2972185029Spjd } else if (state == VDEV_STATE_REMOVED) { 2973185029Spjd vd->vdev_removed = B_TRUE; 2974185029Spjd } else if (state == VDEV_STATE_CANT_OPEN) { 2975185029Spjd /* 2976219089Spjd * If we fail to open a vdev during an import or recovery, we 2977219089Spjd * mark it as "not available", which signifies that it was 2978219089Spjd * never there to begin with. Failure to open such a device 2979219089Spjd * is not considered an error. 2980168404Spjd */ 2981219089Spjd if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 2982219089Spjd spa_load_state(spa) == SPA_LOAD_RECOVER) && 2983168404Spjd vd->vdev_ops->vdev_op_leaf) 2984168404Spjd vd->vdev_not_present = 1; 2985168404Spjd 2986168404Spjd /* 2987168404Spjd * Post the appropriate ereport. If the 'prevstate' field is 2988168404Spjd * set to something other than VDEV_STATE_UNKNOWN, it indicates 2989168404Spjd * that this is part of a vdev_reopen(). In this case, we don't 2990168404Spjd * want to post the ereport if the device was already in the 2991168404Spjd * CANT_OPEN state beforehand. 2992185029Spjd * 2993185029Spjd * If the 'checkremove' flag is set, then this is an attempt to 2994185029Spjd * online the device in response to an insertion event. If we 2995185029Spjd * hit this case, then we have detected an insertion event for a 2996185029Spjd * faulted or offline device that wasn't in the removed state. 2997185029Spjd * In this scenario, we don't post an ereport because we are 2998185029Spjd * about to replace the device, or attempt an online with 2999185029Spjd * vdev_forcefault, which will generate the fault for us. 3000168404Spjd */ 3001185029Spjd if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 3002185029Spjd !vd->vdev_not_present && !vd->vdev_checkremove && 3003185029Spjd vd != spa->spa_root_vdev) { 3004168404Spjd const char *class; 3005168404Spjd 3006168404Spjd switch (aux) { 3007168404Spjd case VDEV_AUX_OPEN_FAILED: 3008168404Spjd class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3009168404Spjd break; 3010168404Spjd case VDEV_AUX_CORRUPT_DATA: 3011168404Spjd class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3012168404Spjd break; 3013168404Spjd case VDEV_AUX_NO_REPLICAS: 3014168404Spjd class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3015168404Spjd break; 3016168404Spjd case VDEV_AUX_BAD_GUID_SUM: 3017168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3018168404Spjd break; 3019168404Spjd case VDEV_AUX_TOO_SMALL: 3020168404Spjd class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3021168404Spjd break; 3022168404Spjd case VDEV_AUX_BAD_LABEL: 3023168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3024168404Spjd break; 3025168404Spjd default: 3026168404Spjd class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3027168404Spjd } 3028168404Spjd 3029185029Spjd zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3030168404Spjd } 3031185029Spjd 3032185029Spjd /* Erase any notion of persistent removed state */ 3033185029Spjd vd->vdev_removed = B_FALSE; 3034185029Spjd } else { 3035185029Spjd vd->vdev_removed = B_FALSE; 3036168404Spjd } 3037168404Spjd 3038209962Smm if (!isopen && vd->vdev_parent) 3039209962Smm vdev_propagate_state(vd->vdev_parent); 3040185029Spjd} 3041168404Spjd 3042185029Spjd/* 3043185029Spjd * Check the vdev configuration to ensure that it's capable of supporting 3044193163Sdfr * a root pool. 3045193163Sdfr * 3046193163Sdfr * On Solaris, we do not support RAID-Z or partial configuration. In 3047193163Sdfr * addition, only a single top-level vdev is allowed and none of the 3048193163Sdfr * leaves can be wholedisks. 3049193163Sdfr * 3050193163Sdfr * For FreeBSD, we can boot from any configuration. There is a 3051193163Sdfr * limitation that the boot filesystem must be either uncompressed or 3052193163Sdfr * compresses with lzjb compression but I'm not sure how to enforce 3053193163Sdfr * that here. 3054185029Spjd */ 3055185029Spjdboolean_t 3056185029Spjdvdev_is_bootable(vdev_t *vd) 3057185029Spjd{ 3058213197Smm#ifdef sun 3059185029Spjd if (!vd->vdev_ops->vdev_op_leaf) { 3060185029Spjd char *vdev_type = vd->vdev_ops->vdev_op_type; 3061185029Spjd 3062185029Spjd if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 3063185029Spjd vd->vdev_children > 1) { 3064185029Spjd return (B_FALSE); 3065185029Spjd } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 3066185029Spjd strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 3067185029Spjd return (B_FALSE); 3068185029Spjd } 3069185029Spjd } else if (vd->vdev_wholedisk == 1) { 3070185029Spjd return (B_FALSE); 3071185029Spjd } 3072185029Spjd 3073219089Spjd for (int c = 0; c < vd->vdev_children; c++) { 3074185029Spjd if (!vdev_is_bootable(vd->vdev_child[c])) 3075185029Spjd return (B_FALSE); 3076185029Spjd } 3077213197Smm#endif /* sun */ 3078185029Spjd return (B_TRUE); 3079168404Spjd} 3080213197Smm 3081219089Spjd/* 3082219089Spjd * Load the state from the original vdev tree (ovd) which 3083219089Spjd * we've retrieved from the MOS config object. If the original 3084219089Spjd * vdev was offline or faulted then we transfer that state to the 3085219089Spjd * device in the current vdev tree (nvd). 3086219089Spjd */ 3087213197Smmvoid 3088219089Spjdvdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3089213197Smm{ 3090219089Spjd spa_t *spa = nvd->vdev_spa; 3091213197Smm 3092219089Spjd ASSERT(nvd->vdev_top->vdev_islog); 3093219089Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3094219089Spjd ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3095213197Smm 3096219089Spjd for (int c = 0; c < nvd->vdev_children; c++) 3097219089Spjd vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3098213197Smm 3099219089Spjd if (nvd->vdev_ops->vdev_op_leaf) { 3100213197Smm /* 3101219089Spjd * Restore the persistent vdev state 3102213197Smm */ 3103219089Spjd nvd->vdev_offline = ovd->vdev_offline; 3104219089Spjd nvd->vdev_faulted = ovd->vdev_faulted; 3105219089Spjd nvd->vdev_degraded = ovd->vdev_degraded; 3106219089Spjd nvd->vdev_removed = ovd->vdev_removed; 3107213197Smm } 3108213197Smm} 3109219089Spjd 3110219089Spjd/* 3111219089Spjd * Determine if a log device has valid content. If the vdev was 3112219089Spjd * removed or faulted in the MOS config then we know that 3113219089Spjd * the content on the log device has already been written to the pool. 3114219089Spjd */ 3115219089Spjdboolean_t 3116219089Spjdvdev_log_state_valid(vdev_t *vd) 3117219089Spjd{ 3118219089Spjd if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 3119219089Spjd !vd->vdev_removed) 3120219089Spjd return (B_TRUE); 3121219089Spjd 3122219089Spjd for (int c = 0; c < vd->vdev_children; c++) 3123219089Spjd if (vdev_log_state_valid(vd->vdev_child[c])) 3124219089Spjd return (B_TRUE); 3125219089Spjd 3126219089Spjd return (B_FALSE); 3127219089Spjd} 3128219089Spjd 3129219089Spjd/* 3130219089Spjd * Expand a vdev if possible. 3131219089Spjd */ 3132219089Spjdvoid 3133219089Spjdvdev_expand(vdev_t *vd, uint64_t txg) 3134219089Spjd{ 3135219089Spjd ASSERT(vd->vdev_top == vd); 3136219089Spjd ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3137219089Spjd 3138219089Spjd if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3139219089Spjd VERIFY(vdev_metaslab_init(vd, txg) == 0); 3140219089Spjd vdev_config_dirty(vd); 3141219089Spjd } 3142219089Spjd} 3143219089Spjd 3144219089Spjd/* 3145219089Spjd * Split a vdev. 3146219089Spjd */ 3147219089Spjdvoid 3148219089Spjdvdev_split(vdev_t *vd) 3149219089Spjd{ 3150219089Spjd vdev_t *cvd, *pvd = vd->vdev_parent; 3151219089Spjd 3152219089Spjd vdev_remove_child(pvd, vd); 3153219089Spjd vdev_compact_children(pvd); 3154219089Spjd 3155219089Spjd cvd = pvd->vdev_child[0]; 3156219089Spjd if (pvd->vdev_children == 1) { 3157219089Spjd vdev_remove_parent(cvd); 3158219089Spjd cvd->vdev_splitting = B_TRUE; 3159219089Spjd } 3160219089Spjd vdev_propagate_state(cvd); 3161219089Spjd} 3162