vdev.c revision 213198
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/* 23209962Smm * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24168404Spjd * Use is subject to license terms. 25168404Spjd */ 26168404Spjd 27168404Spjd#include <sys/zfs_context.h> 28168404Spjd#include <sys/fm/fs/zfs.h> 29168404Spjd#include <sys/spa.h> 30168404Spjd#include <sys/spa_impl.h> 31168404Spjd#include <sys/dmu.h> 32168404Spjd#include <sys/dmu_tx.h> 33168404Spjd#include <sys/vdev_impl.h> 34168404Spjd#include <sys/uberblock_impl.h> 35168404Spjd#include <sys/metaslab.h> 36168404Spjd#include <sys/metaslab_impl.h> 37168404Spjd#include <sys/space_map.h> 38168404Spjd#include <sys/zio.h> 39168404Spjd#include <sys/zap.h> 40168404Spjd#include <sys/fs/zfs.h> 41185029Spjd#include <sys/arc.h> 42213197Smm#include <sys/zil.h> 43168404Spjd 44168404SpjdSYSCTL_DECL(_vfs_zfs); 45168404SpjdSYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); 46168404Spjd 47168404Spjd/* 48168404Spjd * Virtual device management. 49168404Spjd */ 50168404Spjd 51168404Spjdstatic vdev_ops_t *vdev_ops_table[] = { 52168404Spjd &vdev_root_ops, 53168404Spjd &vdev_raidz_ops, 54168404Spjd &vdev_mirror_ops, 55168404Spjd &vdev_replacing_ops, 56168404Spjd &vdev_spare_ops, 57168404Spjd#ifdef _KERNEL 58168404Spjd &vdev_geom_ops, 59168404Spjd#else 60168404Spjd &vdev_disk_ops, 61185029Spjd#endif 62168404Spjd &vdev_file_ops, 63168404Spjd &vdev_missing_ops, 64168404Spjd NULL 65168404Spjd}; 66168404Spjd 67185029Spjd/* maximum scrub/resilver I/O queue per leaf vdev */ 68185029Spjdint zfs_scrub_limit = 10; 69168404Spjd 70185029SpjdTUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit); 71185029SpjdSYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0, 72185029Spjd "Maximum scrub/resilver I/O queue"); 73185029Spjd 74168404Spjd/* 75168404Spjd * Given a vdev type, return the appropriate ops vector. 76168404Spjd */ 77168404Spjdstatic vdev_ops_t * 78168404Spjdvdev_getops(const char *type) 79168404Spjd{ 80168404Spjd vdev_ops_t *ops, **opspp; 81168404Spjd 82168404Spjd for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 83168404Spjd if (strcmp(ops->vdev_op_type, type) == 0) 84168404Spjd break; 85168404Spjd 86168404Spjd return (ops); 87168404Spjd} 88168404Spjd 89168404Spjd/* 90168404Spjd * Default asize function: return the MAX of psize with the asize of 91168404Spjd * all children. This is what's used by anything other than RAID-Z. 92168404Spjd */ 93168404Spjduint64_t 94168404Spjdvdev_default_asize(vdev_t *vd, uint64_t psize) 95168404Spjd{ 96168404Spjd uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 97168404Spjd uint64_t csize; 98168404Spjd uint64_t c; 99168404Spjd 100168404Spjd for (c = 0; c < vd->vdev_children; c++) { 101168404Spjd csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 102168404Spjd asize = MAX(asize, csize); 103168404Spjd } 104168404Spjd 105168404Spjd return (asize); 106168404Spjd} 107168404Spjd 108168404Spjd/* 109168404Spjd * Get the replaceable or attachable device size. 110168404Spjd * If the parent is a mirror or raidz, the replaceable size is the minimum 111168404Spjd * psize of all its children. For the rest, just return our own psize. 112168404Spjd * 113168404Spjd * e.g. 114168404Spjd * psize rsize 115168404Spjd * root - - 116168404Spjd * mirror/raidz - - 117168404Spjd * disk1 20g 20g 118168404Spjd * disk2 40g 20g 119168404Spjd * disk3 80g 80g 120168404Spjd */ 121168404Spjduint64_t 122168404Spjdvdev_get_rsize(vdev_t *vd) 123168404Spjd{ 124168404Spjd vdev_t *pvd, *cvd; 125168404Spjd uint64_t c, rsize; 126168404Spjd 127168404Spjd pvd = vd->vdev_parent; 128168404Spjd 129168404Spjd /* 130168404Spjd * If our parent is NULL or the root, just return our own psize. 131168404Spjd */ 132168404Spjd if (pvd == NULL || pvd->vdev_parent == NULL) 133168404Spjd return (vd->vdev_psize); 134168404Spjd 135168404Spjd rsize = 0; 136168404Spjd 137168404Spjd for (c = 0; c < pvd->vdev_children; c++) { 138168404Spjd cvd = pvd->vdev_child[c]; 139168404Spjd rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1; 140168404Spjd } 141168404Spjd 142168404Spjd return (rsize); 143168404Spjd} 144168404Spjd 145168404Spjdvdev_t * 146168404Spjdvdev_lookup_top(spa_t *spa, uint64_t vdev) 147168404Spjd{ 148168404Spjd vdev_t *rvd = spa->spa_root_vdev; 149168404Spjd 150185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 151185029Spjd 152185029Spjd if (vdev < rvd->vdev_children) { 153185029Spjd ASSERT(rvd->vdev_child[vdev] != NULL); 154168404Spjd return (rvd->vdev_child[vdev]); 155185029Spjd } 156168404Spjd 157168404Spjd return (NULL); 158168404Spjd} 159168404Spjd 160168404Spjdvdev_t * 161168404Spjdvdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 162168404Spjd{ 163168404Spjd int c; 164168404Spjd vdev_t *mvd; 165168404Spjd 166168404Spjd if (vd->vdev_guid == guid) 167168404Spjd return (vd); 168168404Spjd 169168404Spjd for (c = 0; c < vd->vdev_children; c++) 170168404Spjd if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 171168404Spjd NULL) 172168404Spjd return (mvd); 173168404Spjd 174168404Spjd return (NULL); 175168404Spjd} 176168404Spjd 177168404Spjdvoid 178168404Spjdvdev_add_child(vdev_t *pvd, vdev_t *cvd) 179168404Spjd{ 180168404Spjd size_t oldsize, newsize; 181168404Spjd uint64_t id = cvd->vdev_id; 182168404Spjd vdev_t **newchild; 183168404Spjd 184185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 185168404Spjd ASSERT(cvd->vdev_parent == NULL); 186168404Spjd 187168404Spjd cvd->vdev_parent = pvd; 188168404Spjd 189168404Spjd if (pvd == NULL) 190168404Spjd return; 191168404Spjd 192168404Spjd ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 193168404Spjd 194168404Spjd oldsize = pvd->vdev_children * sizeof (vdev_t *); 195168404Spjd pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 196168404Spjd newsize = pvd->vdev_children * sizeof (vdev_t *); 197168404Spjd 198168404Spjd newchild = kmem_zalloc(newsize, KM_SLEEP); 199168404Spjd if (pvd->vdev_child != NULL) { 200168404Spjd bcopy(pvd->vdev_child, newchild, oldsize); 201168404Spjd kmem_free(pvd->vdev_child, oldsize); 202168404Spjd } 203168404Spjd 204168404Spjd pvd->vdev_child = newchild; 205168404Spjd pvd->vdev_child[id] = cvd; 206168404Spjd 207168404Spjd cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 208168404Spjd ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 209168404Spjd 210168404Spjd /* 211168404Spjd * Walk up all ancestors to update guid sum. 212168404Spjd */ 213168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 214168404Spjd pvd->vdev_guid_sum += cvd->vdev_guid_sum; 215168404Spjd 216168404Spjd if (cvd->vdev_ops->vdev_op_leaf) 217168404Spjd cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; 218168404Spjd} 219168404Spjd 220168404Spjdvoid 221168404Spjdvdev_remove_child(vdev_t *pvd, vdev_t *cvd) 222168404Spjd{ 223168404Spjd int c; 224168404Spjd uint_t id = cvd->vdev_id; 225168404Spjd 226168404Spjd ASSERT(cvd->vdev_parent == pvd); 227168404Spjd 228168404Spjd if (pvd == NULL) 229168404Spjd return; 230168404Spjd 231168404Spjd ASSERT(id < pvd->vdev_children); 232168404Spjd ASSERT(pvd->vdev_child[id] == cvd); 233168404Spjd 234168404Spjd pvd->vdev_child[id] = NULL; 235168404Spjd cvd->vdev_parent = NULL; 236168404Spjd 237168404Spjd for (c = 0; c < pvd->vdev_children; c++) 238168404Spjd if (pvd->vdev_child[c]) 239168404Spjd break; 240168404Spjd 241168404Spjd if (c == pvd->vdev_children) { 242168404Spjd kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 243168404Spjd pvd->vdev_child = NULL; 244168404Spjd pvd->vdev_children = 0; 245168404Spjd } 246168404Spjd 247168404Spjd /* 248168404Spjd * Walk up all ancestors to update guid sum. 249168404Spjd */ 250168404Spjd for (; pvd != NULL; pvd = pvd->vdev_parent) 251168404Spjd pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 252168404Spjd 253168404Spjd if (cvd->vdev_ops->vdev_op_leaf) 254168404Spjd cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; 255168404Spjd} 256168404Spjd 257168404Spjd/* 258168404Spjd * Remove any holes in the child array. 259168404Spjd */ 260168404Spjdvoid 261168404Spjdvdev_compact_children(vdev_t *pvd) 262168404Spjd{ 263168404Spjd vdev_t **newchild, *cvd; 264168404Spjd int oldc = pvd->vdev_children; 265168404Spjd int newc, c; 266168404Spjd 267185029Spjd ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 268168404Spjd 269168404Spjd for (c = newc = 0; c < oldc; c++) 270168404Spjd if (pvd->vdev_child[c]) 271168404Spjd newc++; 272168404Spjd 273168404Spjd newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 274168404Spjd 275168404Spjd for (c = newc = 0; c < oldc; c++) { 276168404Spjd if ((cvd = pvd->vdev_child[c]) != NULL) { 277168404Spjd newchild[newc] = cvd; 278168404Spjd cvd->vdev_id = newc++; 279168404Spjd } 280168404Spjd } 281168404Spjd 282168404Spjd kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 283168404Spjd pvd->vdev_child = newchild; 284168404Spjd pvd->vdev_children = newc; 285168404Spjd} 286168404Spjd 287168404Spjd/* 288168404Spjd * Allocate and minimally initialize a vdev_t. 289168404Spjd */ 290168404Spjdstatic vdev_t * 291168404Spjdvdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 292168404Spjd{ 293168404Spjd vdev_t *vd; 294168404Spjd 295168404Spjd vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 296168404Spjd 297168404Spjd if (spa->spa_root_vdev == NULL) { 298168404Spjd ASSERT(ops == &vdev_root_ops); 299168404Spjd spa->spa_root_vdev = vd; 300168404Spjd } 301168404Spjd 302168404Spjd if (guid == 0) { 303168404Spjd if (spa->spa_root_vdev == vd) { 304168404Spjd /* 305168404Spjd * The root vdev's guid will also be the pool guid, 306168404Spjd * which must be unique among all pools. 307168404Spjd */ 308168404Spjd while (guid == 0 || spa_guid_exists(guid, 0)) 309168404Spjd guid = spa_get_random(-1ULL); 310168404Spjd } else { 311168404Spjd /* 312168404Spjd * Any other vdev's guid must be unique within the pool. 313168404Spjd */ 314168404Spjd while (guid == 0 || 315168404Spjd spa_guid_exists(spa_guid(spa), guid)) 316168404Spjd guid = spa_get_random(-1ULL); 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; 327168404Spjd 328168404Spjd mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 329168404Spjd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 330185029Spjd mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 331209962Smm for (int t = 0; t < DTL_TYPES; t++) { 332209962Smm space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 333209962Smm &vd->vdev_dtl_lock); 334209962Smm } 335168404Spjd txg_list_create(&vd->vdev_ms_list, 336168404Spjd offsetof(struct metaslab, ms_txg_node)); 337168404Spjd txg_list_create(&vd->vdev_dtl_list, 338168404Spjd offsetof(struct vdev, vdev_dtl_node)); 339168404Spjd vd->vdev_stat.vs_timestamp = gethrtime(); 340185029Spjd vdev_queue_init(vd); 341185029Spjd vdev_cache_init(vd); 342168404Spjd 343168404Spjd return (vd); 344168404Spjd} 345168404Spjd 346168404Spjd/* 347168404Spjd * Allocate a new vdev. The 'alloctype' is used to control whether we are 348168404Spjd * creating a new vdev or loading an existing one - the behavior is slightly 349168404Spjd * different for each case. 350168404Spjd */ 351168404Spjdint 352168404Spjdvdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 353168404Spjd int alloctype) 354168404Spjd{ 355168404Spjd vdev_ops_t *ops; 356168404Spjd char *type; 357185029Spjd uint64_t guid = 0, islog, nparity; 358168404Spjd vdev_t *vd; 359168404Spjd 360185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 361168404Spjd 362168404Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 363168404Spjd return (EINVAL); 364168404Spjd 365168404Spjd if ((ops = vdev_getops(type)) == NULL) 366168404Spjd return (EINVAL); 367168404Spjd 368168404Spjd /* 369168404Spjd * If this is a load, get the vdev guid from the nvlist. 370168404Spjd * Otherwise, vdev_alloc_common() will generate one for us. 371168404Spjd */ 372168404Spjd if (alloctype == VDEV_ALLOC_LOAD) { 373168404Spjd uint64_t label_id; 374168404Spjd 375168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 376168404Spjd label_id != id) 377168404Spjd return (EINVAL); 378168404Spjd 379168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 380168404Spjd return (EINVAL); 381168404Spjd } else if (alloctype == VDEV_ALLOC_SPARE) { 382168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 383168404Spjd return (EINVAL); 384185029Spjd } else if (alloctype == VDEV_ALLOC_L2CACHE) { 385185029Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 386185029Spjd return (EINVAL); 387168404Spjd } 388168404Spjd 389168404Spjd /* 390168404Spjd * The first allocated vdev must be of type 'root'. 391168404Spjd */ 392168404Spjd if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 393168404Spjd return (EINVAL); 394168404Spjd 395185029Spjd /* 396185029Spjd * Determine whether we're a log vdev. 397185029Spjd */ 398185029Spjd islog = 0; 399185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 400185029Spjd if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 401185029Spjd return (ENOTSUP); 402168404Spjd 403168404Spjd /* 404185029Spjd * Set the nparity property for RAID-Z vdevs. 405168404Spjd */ 406185029Spjd nparity = -1ULL; 407168404Spjd if (ops == &vdev_raidz_ops) { 408168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 409185029Spjd &nparity) == 0) { 410168404Spjd /* 411168404Spjd * Currently, we can only support 2 parity devices. 412168404Spjd */ 413185029Spjd if (nparity == 0 || nparity > 2) 414168404Spjd return (EINVAL); 415168404Spjd /* 416168404Spjd * Older versions can only support 1 parity device. 417168404Spjd */ 418185029Spjd if (nparity == 2 && 419185029Spjd spa_version(spa) < SPA_VERSION_RAID6) 420168404Spjd return (ENOTSUP); 421168404Spjd } else { 422168404Spjd /* 423168404Spjd * We require the parity to be specified for SPAs that 424168404Spjd * support multiple parity levels. 425168404Spjd */ 426185029Spjd if (spa_version(spa) >= SPA_VERSION_RAID6) 427168404Spjd return (EINVAL); 428168404Spjd /* 429168404Spjd * Otherwise, we default to 1 parity device for RAID-Z. 430168404Spjd */ 431185029Spjd nparity = 1; 432168404Spjd } 433168404Spjd } else { 434185029Spjd nparity = 0; 435168404Spjd } 436185029Spjd ASSERT(nparity != -1ULL); 437168404Spjd 438185029Spjd vd = vdev_alloc_common(spa, id, guid, ops); 439185029Spjd 440185029Spjd vd->vdev_islog = islog; 441185029Spjd vd->vdev_nparity = nparity; 442185029Spjd 443185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 444185029Spjd vd->vdev_path = spa_strdup(vd->vdev_path); 445185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 446185029Spjd vd->vdev_devid = spa_strdup(vd->vdev_devid); 447185029Spjd if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 448185029Spjd &vd->vdev_physpath) == 0) 449185029Spjd vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 450209962Smm if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 451209962Smm vd->vdev_fru = spa_strdup(vd->vdev_fru); 452185029Spjd 453168404Spjd /* 454168404Spjd * Set the whole_disk property. If it's not specified, leave the value 455168404Spjd * as -1. 456168404Spjd */ 457168404Spjd if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 458168404Spjd &vd->vdev_wholedisk) != 0) 459168404Spjd vd->vdev_wholedisk = -1ULL; 460168404Spjd 461168404Spjd /* 462168404Spjd * Look for the 'not present' flag. This will only be set if the device 463168404Spjd * was not present at the time of import. 464168404Spjd */ 465209962Smm (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 466209962Smm &vd->vdev_not_present); 467168404Spjd 468168404Spjd /* 469168404Spjd * Get the alignment requirement. 470168404Spjd */ 471168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 472168404Spjd 473168404Spjd /* 474168404Spjd * If we're a top-level vdev, try to load the allocation parameters. 475168404Spjd */ 476168404Spjd if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) { 477168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 478168404Spjd &vd->vdev_ms_array); 479168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 480168404Spjd &vd->vdev_ms_shift); 481168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 482168404Spjd &vd->vdev_asize); 483168404Spjd } 484168404Spjd 485168404Spjd /* 486185029Spjd * If we're a leaf vdev, try to load the DTL object and other state. 487168404Spjd */ 488185029Spjd if (vd->vdev_ops->vdev_op_leaf && 489185029Spjd (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE)) { 490185029Spjd if (alloctype == VDEV_ALLOC_LOAD) { 491185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 492209962Smm &vd->vdev_dtl_smo.smo_object); 493185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 494185029Spjd &vd->vdev_unspare); 495185029Spjd } 496168404Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 497168404Spjd &vd->vdev_offline); 498185029Spjd 499185029Spjd /* 500185029Spjd * When importing a pool, we want to ignore the persistent fault 501185029Spjd * state, as the diagnosis made on another system may not be 502185029Spjd * valid in the current context. 503185029Spjd */ 504185029Spjd if (spa->spa_load_state == SPA_LOAD_OPEN) { 505185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 506185029Spjd &vd->vdev_faulted); 507185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 508185029Spjd &vd->vdev_degraded); 509185029Spjd (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 510185029Spjd &vd->vdev_removed); 511185029Spjd } 512168404Spjd } 513168404Spjd 514168404Spjd /* 515168404Spjd * Add ourselves to the parent's list of children. 516168404Spjd */ 517168404Spjd vdev_add_child(parent, vd); 518168404Spjd 519168404Spjd *vdp = vd; 520168404Spjd 521168404Spjd return (0); 522168404Spjd} 523168404Spjd 524168404Spjdvoid 525168404Spjdvdev_free(vdev_t *vd) 526168404Spjd{ 527168404Spjd int c; 528185029Spjd spa_t *spa = vd->vdev_spa; 529168404Spjd 530168404Spjd /* 531168404Spjd * vdev_free() implies closing the vdev first. This is simpler than 532168404Spjd * trying to ensure complicated semantics for all callers. 533168404Spjd */ 534168404Spjd vdev_close(vd); 535168404Spjd 536185029Spjd ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 537168404Spjd 538168404Spjd /* 539168404Spjd * Free all children. 540168404Spjd */ 541168404Spjd for (c = 0; c < vd->vdev_children; c++) 542168404Spjd vdev_free(vd->vdev_child[c]); 543168404Spjd 544168404Spjd ASSERT(vd->vdev_child == NULL); 545168404Spjd ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 546168404Spjd 547168404Spjd /* 548168404Spjd * Discard allocation state. 549168404Spjd */ 550168404Spjd if (vd == vd->vdev_top) 551168404Spjd vdev_metaslab_fini(vd); 552168404Spjd 553168404Spjd ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 554168404Spjd ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 555168404Spjd ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 556168404Spjd 557168404Spjd /* 558168404Spjd * Remove this vdev from its parent's child list. 559168404Spjd */ 560168404Spjd vdev_remove_child(vd->vdev_parent, vd); 561168404Spjd 562168404Spjd ASSERT(vd->vdev_parent == NULL); 563168404Spjd 564185029Spjd /* 565185029Spjd * Clean up vdev structure. 566185029Spjd */ 567185029Spjd vdev_queue_fini(vd); 568185029Spjd vdev_cache_fini(vd); 569185029Spjd 570185029Spjd if (vd->vdev_path) 571185029Spjd spa_strfree(vd->vdev_path); 572185029Spjd if (vd->vdev_devid) 573185029Spjd spa_strfree(vd->vdev_devid); 574185029Spjd if (vd->vdev_physpath) 575185029Spjd spa_strfree(vd->vdev_physpath); 576209962Smm if (vd->vdev_fru) 577209962Smm spa_strfree(vd->vdev_fru); 578185029Spjd 579185029Spjd if (vd->vdev_isspare) 580185029Spjd spa_spare_remove(vd); 581185029Spjd if (vd->vdev_isl2cache) 582185029Spjd spa_l2cache_remove(vd); 583185029Spjd 584185029Spjd txg_list_destroy(&vd->vdev_ms_list); 585185029Spjd txg_list_destroy(&vd->vdev_dtl_list); 586209962Smm 587185029Spjd mutex_enter(&vd->vdev_dtl_lock); 588209962Smm for (int t = 0; t < DTL_TYPES; t++) { 589209962Smm space_map_unload(&vd->vdev_dtl[t]); 590209962Smm space_map_destroy(&vd->vdev_dtl[t]); 591209962Smm } 592185029Spjd mutex_exit(&vd->vdev_dtl_lock); 593209962Smm 594185029Spjd mutex_destroy(&vd->vdev_dtl_lock); 595185029Spjd mutex_destroy(&vd->vdev_stat_lock); 596185029Spjd mutex_destroy(&vd->vdev_probe_lock); 597185029Spjd 598185029Spjd if (vd == spa->spa_root_vdev) 599185029Spjd spa->spa_root_vdev = NULL; 600185029Spjd 601185029Spjd kmem_free(vd, sizeof (vdev_t)); 602168404Spjd} 603168404Spjd 604168404Spjd/* 605168404Spjd * Transfer top-level vdev state from svd to tvd. 606168404Spjd */ 607168404Spjdstatic void 608168404Spjdvdev_top_transfer(vdev_t *svd, vdev_t *tvd) 609168404Spjd{ 610168404Spjd spa_t *spa = svd->vdev_spa; 611168404Spjd metaslab_t *msp; 612168404Spjd vdev_t *vd; 613168404Spjd int t; 614168404Spjd 615168404Spjd ASSERT(tvd == tvd->vdev_top); 616168404Spjd 617168404Spjd tvd->vdev_ms_array = svd->vdev_ms_array; 618168404Spjd tvd->vdev_ms_shift = svd->vdev_ms_shift; 619168404Spjd tvd->vdev_ms_count = svd->vdev_ms_count; 620168404Spjd 621168404Spjd svd->vdev_ms_array = 0; 622168404Spjd svd->vdev_ms_shift = 0; 623168404Spjd svd->vdev_ms_count = 0; 624168404Spjd 625168404Spjd tvd->vdev_mg = svd->vdev_mg; 626168404Spjd tvd->vdev_ms = svd->vdev_ms; 627168404Spjd 628168404Spjd svd->vdev_mg = NULL; 629168404Spjd svd->vdev_ms = NULL; 630168404Spjd 631168404Spjd if (tvd->vdev_mg != NULL) 632168404Spjd tvd->vdev_mg->mg_vd = tvd; 633168404Spjd 634168404Spjd tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 635168404Spjd tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 636168404Spjd tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 637168404Spjd 638168404Spjd svd->vdev_stat.vs_alloc = 0; 639168404Spjd svd->vdev_stat.vs_space = 0; 640168404Spjd svd->vdev_stat.vs_dspace = 0; 641168404Spjd 642168404Spjd for (t = 0; t < TXG_SIZE; t++) { 643168404Spjd while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 644168404Spjd (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 645168404Spjd while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 646168404Spjd (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 647168404Spjd if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 648168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 649168404Spjd } 650168404Spjd 651185029Spjd if (list_link_active(&svd->vdev_config_dirty_node)) { 652168404Spjd vdev_config_clean(svd); 653168404Spjd vdev_config_dirty(tvd); 654168404Spjd } 655168404Spjd 656185029Spjd if (list_link_active(&svd->vdev_state_dirty_node)) { 657185029Spjd vdev_state_clean(svd); 658185029Spjd vdev_state_dirty(tvd); 659185029Spjd } 660168404Spjd 661168404Spjd tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 662168404Spjd svd->vdev_deflate_ratio = 0; 663185029Spjd 664185029Spjd tvd->vdev_islog = svd->vdev_islog; 665185029Spjd svd->vdev_islog = 0; 666168404Spjd} 667168404Spjd 668168404Spjdstatic void 669168404Spjdvdev_top_update(vdev_t *tvd, vdev_t *vd) 670168404Spjd{ 671168404Spjd int c; 672168404Spjd 673168404Spjd if (vd == NULL) 674168404Spjd return; 675168404Spjd 676168404Spjd vd->vdev_top = tvd; 677168404Spjd 678168404Spjd for (c = 0; c < vd->vdev_children; c++) 679168404Spjd vdev_top_update(tvd, vd->vdev_child[c]); 680168404Spjd} 681168404Spjd 682168404Spjd/* 683168404Spjd * Add a mirror/replacing vdev above an existing vdev. 684168404Spjd */ 685168404Spjdvdev_t * 686168404Spjdvdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 687168404Spjd{ 688168404Spjd spa_t *spa = cvd->vdev_spa; 689168404Spjd vdev_t *pvd = cvd->vdev_parent; 690168404Spjd vdev_t *mvd; 691168404Spjd 692185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 693168404Spjd 694168404Spjd mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 695168404Spjd 696168404Spjd mvd->vdev_asize = cvd->vdev_asize; 697168404Spjd mvd->vdev_ashift = cvd->vdev_ashift; 698168404Spjd mvd->vdev_state = cvd->vdev_state; 699168404Spjd 700168404Spjd vdev_remove_child(pvd, cvd); 701168404Spjd vdev_add_child(pvd, mvd); 702168404Spjd cvd->vdev_id = mvd->vdev_children; 703168404Spjd vdev_add_child(mvd, cvd); 704168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 705168404Spjd 706168404Spjd if (mvd == mvd->vdev_top) 707168404Spjd vdev_top_transfer(cvd, mvd); 708168404Spjd 709168404Spjd return (mvd); 710168404Spjd} 711168404Spjd 712168404Spjd/* 713168404Spjd * Remove a 1-way mirror/replacing vdev from the tree. 714168404Spjd */ 715168404Spjdvoid 716168404Spjdvdev_remove_parent(vdev_t *cvd) 717168404Spjd{ 718168404Spjd vdev_t *mvd = cvd->vdev_parent; 719168404Spjd vdev_t *pvd = mvd->vdev_parent; 720168404Spjd 721185029Spjd ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 722168404Spjd 723168404Spjd ASSERT(mvd->vdev_children == 1); 724168404Spjd ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 725168404Spjd mvd->vdev_ops == &vdev_replacing_ops || 726168404Spjd mvd->vdev_ops == &vdev_spare_ops); 727168404Spjd cvd->vdev_ashift = mvd->vdev_ashift; 728168404Spjd 729168404Spjd vdev_remove_child(mvd, cvd); 730168404Spjd vdev_remove_child(pvd, mvd); 731209962Smm 732185029Spjd /* 733185029Spjd * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 734185029Spjd * Otherwise, we could have detached an offline device, and when we 735185029Spjd * go to import the pool we'll think we have two top-level vdevs, 736185029Spjd * instead of a different version of the same top-level vdev. 737185029Spjd */ 738209962Smm if (mvd->vdev_top == mvd) { 739209962Smm uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 740209962Smm cvd->vdev_guid += guid_delta; 741209962Smm cvd->vdev_guid_sum += guid_delta; 742209962Smm } 743168404Spjd cvd->vdev_id = mvd->vdev_id; 744168404Spjd vdev_add_child(pvd, cvd); 745168404Spjd vdev_top_update(cvd->vdev_top, cvd->vdev_top); 746168404Spjd 747168404Spjd if (cvd == cvd->vdev_top) 748168404Spjd vdev_top_transfer(mvd, cvd); 749168404Spjd 750168404Spjd ASSERT(mvd->vdev_children == 0); 751168404Spjd vdev_free(mvd); 752168404Spjd} 753168404Spjd 754168404Spjdint 755168404Spjdvdev_metaslab_init(vdev_t *vd, uint64_t txg) 756168404Spjd{ 757168404Spjd spa_t *spa = vd->vdev_spa; 758168404Spjd objset_t *mos = spa->spa_meta_objset; 759185029Spjd metaslab_class_t *mc; 760168404Spjd uint64_t m; 761168404Spjd uint64_t oldc = vd->vdev_ms_count; 762168404Spjd uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 763168404Spjd metaslab_t **mspp; 764168404Spjd int error; 765168404Spjd 766168404Spjd if (vd->vdev_ms_shift == 0) /* not being allocated from yet */ 767168404Spjd return (0); 768168404Spjd 769213197Smm /* 770213197Smm * Compute the raidz-deflation ratio. Note, we hard-code 771213197Smm * in 128k (1 << 17) because it is the current "typical" blocksize. 772213197Smm * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 773213197Smm * or we will inconsistently account for existing bp's. 774213197Smm */ 775213197Smm vd->vdev_deflate_ratio = (1 << 17) / 776213197Smm (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 777213197Smm 778168404Spjd ASSERT(oldc <= newc); 779168404Spjd 780185029Spjd if (vd->vdev_islog) 781185029Spjd mc = spa->spa_log_class; 782185029Spjd else 783185029Spjd mc = spa->spa_normal_class; 784185029Spjd 785168404Spjd if (vd->vdev_mg == NULL) 786168404Spjd vd->vdev_mg = metaslab_group_create(mc, vd); 787168404Spjd 788168404Spjd mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 789168404Spjd 790168404Spjd if (oldc != 0) { 791168404Spjd bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 792168404Spjd kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 793168404Spjd } 794168404Spjd 795168404Spjd vd->vdev_ms = mspp; 796168404Spjd vd->vdev_ms_count = newc; 797168404Spjd 798168404Spjd for (m = oldc; m < newc; m++) { 799168404Spjd space_map_obj_t smo = { 0, 0, 0 }; 800168404Spjd if (txg == 0) { 801168404Spjd uint64_t object = 0; 802168404Spjd error = dmu_read(mos, vd->vdev_ms_array, 803209962Smm m * sizeof (uint64_t), sizeof (uint64_t), &object, 804209962Smm DMU_READ_PREFETCH); 805168404Spjd if (error) 806168404Spjd return (error); 807168404Spjd if (object != 0) { 808168404Spjd dmu_buf_t *db; 809168404Spjd error = dmu_bonus_hold(mos, object, FTAG, &db); 810168404Spjd if (error) 811168404Spjd return (error); 812185029Spjd ASSERT3U(db->db_size, >=, sizeof (smo)); 813185029Spjd bcopy(db->db_data, &smo, sizeof (smo)); 814168404Spjd ASSERT3U(smo.smo_object, ==, object); 815168404Spjd dmu_buf_rele(db, FTAG); 816168404Spjd } 817168404Spjd } 818168404Spjd vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 819168404Spjd m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 820168404Spjd } 821168404Spjd 822168404Spjd return (0); 823168404Spjd} 824168404Spjd 825168404Spjdvoid 826168404Spjdvdev_metaslab_fini(vdev_t *vd) 827168404Spjd{ 828168404Spjd uint64_t m; 829168404Spjd uint64_t count = vd->vdev_ms_count; 830168404Spjd 831168404Spjd if (vd->vdev_ms != NULL) { 832168404Spjd for (m = 0; m < count; m++) 833168404Spjd if (vd->vdev_ms[m] != NULL) 834168404Spjd metaslab_fini(vd->vdev_ms[m]); 835168404Spjd kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 836168404Spjd vd->vdev_ms = NULL; 837168404Spjd } 838168404Spjd} 839168404Spjd 840185029Spjdtypedef struct vdev_probe_stats { 841185029Spjd boolean_t vps_readable; 842185029Spjd boolean_t vps_writeable; 843185029Spjd int vps_flags; 844185029Spjd} vdev_probe_stats_t; 845185029Spjd 846185029Spjdstatic void 847185029Spjdvdev_probe_done(zio_t *zio) 848185029Spjd{ 849209962Smm spa_t *spa = zio->io_spa; 850209962Smm vdev_t *vd = zio->io_vd; 851185029Spjd vdev_probe_stats_t *vps = zio->io_private; 852185029Spjd 853209962Smm ASSERT(vd->vdev_probe_zio != NULL); 854209962Smm 855185029Spjd if (zio->io_type == ZIO_TYPE_READ) { 856185029Spjd if (zio->io_error == 0) 857185029Spjd vps->vps_readable = 1; 858209962Smm if (zio->io_error == 0 && spa_writeable(spa)) { 859209962Smm zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 860185029Spjd zio->io_offset, zio->io_size, zio->io_data, 861185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 862185029Spjd ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 863185029Spjd } else { 864185029Spjd zio_buf_free(zio->io_data, zio->io_size); 865185029Spjd } 866185029Spjd } else if (zio->io_type == ZIO_TYPE_WRITE) { 867185029Spjd if (zio->io_error == 0) 868185029Spjd vps->vps_writeable = 1; 869185029Spjd zio_buf_free(zio->io_data, zio->io_size); 870185029Spjd } else if (zio->io_type == ZIO_TYPE_NULL) { 871209962Smm zio_t *pio; 872185029Spjd 873185029Spjd vd->vdev_cant_read |= !vps->vps_readable; 874185029Spjd vd->vdev_cant_write |= !vps->vps_writeable; 875185029Spjd 876185029Spjd if (vdev_readable(vd) && 877209962Smm (vdev_writeable(vd) || !spa_writeable(spa))) { 878185029Spjd zio->io_error = 0; 879185029Spjd } else { 880185029Spjd ASSERT(zio->io_error != 0); 881185029Spjd zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 882209962Smm spa, vd, NULL, 0, 0); 883185029Spjd zio->io_error = ENXIO; 884185029Spjd } 885209962Smm 886209962Smm mutex_enter(&vd->vdev_probe_lock); 887209962Smm ASSERT(vd->vdev_probe_zio == zio); 888209962Smm vd->vdev_probe_zio = NULL; 889209962Smm mutex_exit(&vd->vdev_probe_lock); 890209962Smm 891209962Smm while ((pio = zio_walk_parents(zio)) != NULL) 892209962Smm if (!vdev_accessible(vd, pio)) 893209962Smm pio->io_error = ENXIO; 894209962Smm 895185029Spjd kmem_free(vps, sizeof (*vps)); 896185029Spjd } 897185029Spjd} 898185029Spjd 899168404Spjd/* 900185029Spjd * Determine whether this device is accessible by reading and writing 901185029Spjd * to several known locations: the pad regions of each vdev label 902185029Spjd * but the first (which we leave alone in case it contains a VTOC). 903185029Spjd */ 904185029Spjdzio_t * 905209962Smmvdev_probe(vdev_t *vd, zio_t *zio) 906185029Spjd{ 907185029Spjd spa_t *spa = vd->vdev_spa; 908209962Smm vdev_probe_stats_t *vps = NULL; 909209962Smm zio_t *pio; 910185029Spjd 911209962Smm ASSERT(vd->vdev_ops->vdev_op_leaf); 912185029Spjd 913209962Smm /* 914209962Smm * Don't probe the probe. 915209962Smm */ 916209962Smm if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 917209962Smm return (NULL); 918185029Spjd 919209962Smm /* 920209962Smm * To prevent 'probe storms' when a device fails, we create 921209962Smm * just one probe i/o at a time. All zios that want to probe 922209962Smm * this vdev will become parents of the probe io. 923209962Smm */ 924209962Smm mutex_enter(&vd->vdev_probe_lock); 925209962Smm 926209962Smm if ((pio = vd->vdev_probe_zio) == NULL) { 927209962Smm vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 928209962Smm 929209962Smm vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 930209962Smm ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 931213198Smm ZIO_FLAG_TRYHARD; 932209962Smm 933209962Smm if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 934209962Smm /* 935209962Smm * vdev_cant_read and vdev_cant_write can only 936209962Smm * transition from TRUE to FALSE when we have the 937209962Smm * SCL_ZIO lock as writer; otherwise they can only 938209962Smm * transition from FALSE to TRUE. This ensures that 939209962Smm * any zio looking at these values can assume that 940209962Smm * failures persist for the life of the I/O. That's 941209962Smm * important because when a device has intermittent 942209962Smm * connectivity problems, we want to ensure that 943209962Smm * they're ascribed to the device (ENXIO) and not 944209962Smm * the zio (EIO). 945209962Smm * 946209962Smm * Since we hold SCL_ZIO as writer here, clear both 947209962Smm * values so the probe can reevaluate from first 948209962Smm * principles. 949209962Smm */ 950209962Smm vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 951209962Smm vd->vdev_cant_read = B_FALSE; 952209962Smm vd->vdev_cant_write = B_FALSE; 953209962Smm } 954209962Smm 955209962Smm vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 956209962Smm vdev_probe_done, vps, 957209962Smm vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 958209962Smm 959209962Smm if (zio != NULL) { 960209962Smm vd->vdev_probe_wanted = B_TRUE; 961209962Smm spa_async_request(spa, SPA_ASYNC_PROBE); 962209962Smm } 963185029Spjd } 964185029Spjd 965209962Smm if (zio != NULL) 966209962Smm zio_add_child(zio, pio); 967185029Spjd 968209962Smm mutex_exit(&vd->vdev_probe_lock); 969185029Spjd 970209962Smm if (vps == NULL) { 971209962Smm ASSERT(zio != NULL); 972209962Smm return (NULL); 973209962Smm } 974185029Spjd 975185029Spjd for (int l = 1; l < VDEV_LABELS; l++) { 976209962Smm zio_nowait(zio_read_phys(pio, vd, 977185029Spjd vdev_label_offset(vd->vdev_psize, l, 978209962Smm offsetof(vdev_label_t, vl_pad2)), 979209962Smm VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 980185029Spjd ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 981185029Spjd ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 982185029Spjd } 983185029Spjd 984209962Smm if (zio == NULL) 985209962Smm return (pio); 986209962Smm 987209962Smm zio_nowait(pio); 988209962Smm return (NULL); 989185029Spjd} 990185029Spjd 991185029Spjd/* 992168404Spjd * Prepare a virtual device for access. 993168404Spjd */ 994168404Spjdint 995168404Spjdvdev_open(vdev_t *vd) 996168404Spjd{ 997209962Smm spa_t *spa = vd->vdev_spa; 998168404Spjd int error; 999168404Spjd int c; 1000168404Spjd uint64_t osize = 0; 1001168404Spjd uint64_t asize, psize; 1002168404Spjd uint64_t ashift = 0; 1003168404Spjd 1004209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1005209962Smm 1006168404Spjd ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1007168404Spjd vd->vdev_state == VDEV_STATE_CANT_OPEN || 1008168404Spjd vd->vdev_state == VDEV_STATE_OFFLINE); 1009168404Spjd 1010168404Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1011213197Smm vd->vdev_cant_read = B_FALSE; 1012213197Smm vd->vdev_cant_write = B_FALSE; 1013168404Spjd 1014185029Spjd if (!vd->vdev_removed && vd->vdev_faulted) { 1015168404Spjd ASSERT(vd->vdev_children == 0); 1016185029Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1017185029Spjd VDEV_AUX_ERR_EXCEEDED); 1018185029Spjd return (ENXIO); 1019185029Spjd } else if (vd->vdev_offline) { 1020185029Spjd ASSERT(vd->vdev_children == 0); 1021168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1022168404Spjd return (ENXIO); 1023168404Spjd } 1024168404Spjd 1025168404Spjd error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 1026168404Spjd 1027168404Spjd if (zio_injection_enabled && error == 0) 1028213198Smm error = zio_handle_device_injection(vd, NULL, ENXIO); 1029168404Spjd 1030185029Spjd if (error) { 1031185029Spjd if (vd->vdev_removed && 1032185029Spjd vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1033185029Spjd vd->vdev_removed = B_FALSE; 1034168404Spjd 1035168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1036168404Spjd vd->vdev_stat.vs_aux); 1037168404Spjd return (error); 1038168404Spjd } 1039168404Spjd 1040185029Spjd vd->vdev_removed = B_FALSE; 1041168404Spjd 1042185029Spjd if (vd->vdev_degraded) { 1043185029Spjd ASSERT(vd->vdev_children == 0); 1044185029Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1045185029Spjd VDEV_AUX_ERR_EXCEEDED); 1046185029Spjd } else { 1047185029Spjd vd->vdev_state = VDEV_STATE_HEALTHY; 1048185029Spjd } 1049185029Spjd 1050168404Spjd for (c = 0; c < vd->vdev_children; c++) 1051168404Spjd if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1052168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1053168404Spjd VDEV_AUX_NONE); 1054168404Spjd break; 1055168404Spjd } 1056168404Spjd 1057168404Spjd osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1058168404Spjd 1059168404Spjd if (vd->vdev_children == 0) { 1060168404Spjd if (osize < SPA_MINDEVSIZE) { 1061168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1062168404Spjd VDEV_AUX_TOO_SMALL); 1063168404Spjd return (EOVERFLOW); 1064168404Spjd } 1065168404Spjd psize = osize; 1066168404Spjd asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1067168404Spjd } else { 1068168404Spjd if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1069168404Spjd (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1070168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1071168404Spjd VDEV_AUX_TOO_SMALL); 1072168404Spjd return (EOVERFLOW); 1073168404Spjd } 1074168404Spjd psize = 0; 1075168404Spjd asize = osize; 1076168404Spjd } 1077168404Spjd 1078168404Spjd vd->vdev_psize = psize; 1079168404Spjd 1080168404Spjd if (vd->vdev_asize == 0) { 1081168404Spjd /* 1082168404Spjd * This is the first-ever open, so use the computed values. 1083168404Spjd * For testing purposes, a higher ashift can be requested. 1084168404Spjd */ 1085168404Spjd vd->vdev_asize = asize; 1086168404Spjd vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1087168404Spjd } else { 1088168404Spjd /* 1089168404Spjd * Make sure the alignment requirement hasn't increased. 1090168404Spjd */ 1091168404Spjd if (ashift > vd->vdev_top->vdev_ashift) { 1092168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1093168404Spjd VDEV_AUX_BAD_LABEL); 1094168404Spjd return (EINVAL); 1095168404Spjd } 1096168404Spjd 1097168404Spjd /* 1098168404Spjd * Make sure the device hasn't shrunk. 1099168404Spjd */ 1100168404Spjd if (asize < vd->vdev_asize) { 1101168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1102168404Spjd VDEV_AUX_BAD_LABEL); 1103168404Spjd return (EINVAL); 1104168404Spjd } 1105168404Spjd 1106168404Spjd /* 1107168404Spjd * If all children are healthy and the asize has increased, 1108168404Spjd * then we've experienced dynamic LUN growth. 1109168404Spjd */ 1110168404Spjd if (vd->vdev_state == VDEV_STATE_HEALTHY && 1111168404Spjd asize > vd->vdev_asize) { 1112168404Spjd vd->vdev_asize = asize; 1113168404Spjd } 1114168404Spjd } 1115168404Spjd 1116168404Spjd /* 1117185029Spjd * Ensure we can issue some IO before declaring the 1118185029Spjd * vdev open for business. 1119185029Spjd */ 1120185029Spjd if (vd->vdev_ops->vdev_op_leaf && 1121185029Spjd (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1122185029Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1123185029Spjd VDEV_AUX_IO_FAILURE); 1124185029Spjd return (error); 1125185029Spjd } 1126185029Spjd 1127185029Spjd /* 1128185029Spjd * If a leaf vdev has a DTL, and seems healthy, then kick off a 1129209962Smm * resilver. But don't do this if we are doing a reopen for a scrub, 1130209962Smm * since this would just restart the scrub we are already doing. 1131168404Spjd */ 1132209962Smm if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1133209962Smm vdev_resilver_needed(vd, NULL, NULL)) 1134209962Smm spa_async_request(spa, SPA_ASYNC_RESILVER); 1135168404Spjd 1136168404Spjd return (0); 1137168404Spjd} 1138168404Spjd 1139168404Spjd/* 1140168404Spjd * Called once the vdevs are all opened, this routine validates the label 1141168404Spjd * contents. This needs to be done before vdev_load() so that we don't 1142185029Spjd * inadvertently do repair I/Os to the wrong device. 1143168404Spjd * 1144168404Spjd * This function will only return failure if one of the vdevs indicates that it 1145168404Spjd * has since been destroyed or exported. This is only possible if 1146168404Spjd * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1147168404Spjd * will be updated but the function will return 0. 1148168404Spjd */ 1149168404Spjdint 1150168404Spjdvdev_validate(vdev_t *vd) 1151168404Spjd{ 1152168404Spjd spa_t *spa = vd->vdev_spa; 1153168404Spjd int c; 1154168404Spjd nvlist_t *label; 1155185029Spjd uint64_t guid, top_guid; 1156168404Spjd uint64_t state; 1157168404Spjd 1158168404Spjd for (c = 0; c < vd->vdev_children; c++) 1159168404Spjd if (vdev_validate(vd->vdev_child[c]) != 0) 1160168926Spjd return (EBADF); 1161168404Spjd 1162168404Spjd /* 1163168404Spjd * If the device has already failed, or was marked offline, don't do 1164168404Spjd * any further validation. Otherwise, label I/O will fail and we will 1165168404Spjd * overwrite the previous state. 1166168404Spjd */ 1167185029Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1168168404Spjd 1169168404Spjd if ((label = vdev_label_read_config(vd)) == NULL) { 1170168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1171168404Spjd VDEV_AUX_BAD_LABEL); 1172168404Spjd return (0); 1173168404Spjd } 1174168404Spjd 1175168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1176168404Spjd &guid) != 0 || guid != spa_guid(spa)) { 1177168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1178168404Spjd VDEV_AUX_CORRUPT_DATA); 1179168404Spjd nvlist_free(label); 1180168404Spjd return (0); 1181168404Spjd } 1182168404Spjd 1183185029Spjd /* 1184185029Spjd * If this vdev just became a top-level vdev because its 1185185029Spjd * sibling was detached, it will have adopted the parent's 1186185029Spjd * vdev guid -- but the label may or may not be on disk yet. 1187185029Spjd * Fortunately, either version of the label will have the 1188185029Spjd * same top guid, so if we're a top-level vdev, we can 1189185029Spjd * safely compare to that instead. 1190185029Spjd */ 1191168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1192185029Spjd &guid) != 0 || 1193185029Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1194185029Spjd &top_guid) != 0 || 1195185029Spjd (vd->vdev_guid != guid && 1196185029Spjd (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1197168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1198168404Spjd VDEV_AUX_CORRUPT_DATA); 1199168404Spjd nvlist_free(label); 1200168404Spjd return (0); 1201168404Spjd } 1202168404Spjd 1203168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1204168404Spjd &state) != 0) { 1205168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1206168404Spjd VDEV_AUX_CORRUPT_DATA); 1207168404Spjd nvlist_free(label); 1208168404Spjd return (0); 1209168404Spjd } 1210168404Spjd 1211168404Spjd nvlist_free(label); 1212168404Spjd 1213209962Smm /* 1214209962Smm * If spa->spa_load_verbatim is true, no need to check the 1215209962Smm * state of the pool. 1216209962Smm */ 1217209962Smm if (!spa->spa_load_verbatim && 1218209962Smm spa->spa_load_state == SPA_LOAD_OPEN && 1219168404Spjd state != POOL_STATE_ACTIVE) 1220168926Spjd return (EBADF); 1221185029Spjd 1222185029Spjd /* 1223185029Spjd * If we were able to open and validate a vdev that was 1224185029Spjd * previously marked permanently unavailable, clear that state 1225185029Spjd * now. 1226185029Spjd */ 1227185029Spjd if (vd->vdev_not_present) 1228185029Spjd vd->vdev_not_present = 0; 1229168404Spjd } 1230168404Spjd 1231168404Spjd return (0); 1232168404Spjd} 1233168404Spjd 1234168404Spjd/* 1235168404Spjd * Close a virtual device. 1236168404Spjd */ 1237168404Spjdvoid 1238168404Spjdvdev_close(vdev_t *vd) 1239168404Spjd{ 1240209962Smm spa_t *spa = vd->vdev_spa; 1241209962Smm 1242209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1243209962Smm 1244168404Spjd vd->vdev_ops->vdev_op_close(vd); 1245168404Spjd 1246185029Spjd vdev_cache_purge(vd); 1247168404Spjd 1248168404Spjd /* 1249168404Spjd * We record the previous state before we close it, so that if we are 1250168404Spjd * doing a reopen(), we don't generate FMA ereports if we notice that 1251168404Spjd * it's still faulted. 1252168404Spjd */ 1253168404Spjd vd->vdev_prevstate = vd->vdev_state; 1254168404Spjd 1255168404Spjd if (vd->vdev_offline) 1256168404Spjd vd->vdev_state = VDEV_STATE_OFFLINE; 1257168404Spjd else 1258168404Spjd vd->vdev_state = VDEV_STATE_CLOSED; 1259168404Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1260168404Spjd} 1261168404Spjd 1262168404Spjdvoid 1263168404Spjdvdev_reopen(vdev_t *vd) 1264168404Spjd{ 1265168404Spjd spa_t *spa = vd->vdev_spa; 1266168404Spjd 1267185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1268168404Spjd 1269168404Spjd vdev_close(vd); 1270168404Spjd (void) vdev_open(vd); 1271168404Spjd 1272168404Spjd /* 1273168404Spjd * Call vdev_validate() here to make sure we have the same device. 1274168404Spjd * Otherwise, a device with an invalid label could be successfully 1275168404Spjd * opened in response to vdev_reopen(). 1276168404Spjd */ 1277185029Spjd if (vd->vdev_aux) { 1278185029Spjd (void) vdev_validate_aux(vd); 1279185029Spjd if (vdev_readable(vd) && vdev_writeable(vd) && 1280209962Smm vd->vdev_aux == &spa->spa_l2cache && 1281185029Spjd !l2arc_vdev_present(vd)) { 1282185029Spjd uint64_t size = vdev_get_rsize(vd); 1283185029Spjd l2arc_add_vdev(spa, vd, 1284185029Spjd VDEV_LABEL_START_SIZE, 1285185029Spjd size - VDEV_LABEL_START_SIZE); 1286185029Spjd } 1287185029Spjd } else { 1288185029Spjd (void) vdev_validate(vd); 1289185029Spjd } 1290168404Spjd 1291168404Spjd /* 1292185029Spjd * Reassess parent vdev's health. 1293168404Spjd */ 1294185029Spjd vdev_propagate_state(vd); 1295168404Spjd} 1296168404Spjd 1297168404Spjdint 1298168404Spjdvdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1299168404Spjd{ 1300168404Spjd int error; 1301168404Spjd 1302168404Spjd /* 1303168404Spjd * Normally, partial opens (e.g. of a mirror) are allowed. 1304168404Spjd * For a create, however, we want to fail the request if 1305168404Spjd * there are any components we can't open. 1306168404Spjd */ 1307168404Spjd error = vdev_open(vd); 1308168404Spjd 1309168404Spjd if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1310168404Spjd vdev_close(vd); 1311168404Spjd return (error ? error : ENXIO); 1312168404Spjd } 1313168404Spjd 1314168404Spjd /* 1315168404Spjd * Recursively initialize all labels. 1316168404Spjd */ 1317168404Spjd if ((error = vdev_label_init(vd, txg, isreplacing ? 1318168404Spjd VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1319168404Spjd vdev_close(vd); 1320168404Spjd return (error); 1321168404Spjd } 1322168404Spjd 1323168404Spjd return (0); 1324168404Spjd} 1325168404Spjd 1326168404Spjd/* 1327168404Spjd * The is the latter half of vdev_create(). It is distinct because it 1328168404Spjd * involves initiating transactions in order to do metaslab creation. 1329168404Spjd * For creation, we want to try to create all vdevs at once and then undo it 1330168404Spjd * if anything fails; this is much harder if we have pending transactions. 1331168404Spjd */ 1332168404Spjdvoid 1333168404Spjdvdev_init(vdev_t *vd, uint64_t txg) 1334168404Spjd{ 1335168404Spjd /* 1336168404Spjd * Aim for roughly 200 metaslabs per vdev. 1337168404Spjd */ 1338168404Spjd vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1339168404Spjd vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1340168404Spjd 1341168404Spjd /* 1342168404Spjd * Initialize the vdev's metaslabs. This can't fail because 1343168404Spjd * there's nothing to read when creating all new metaslabs. 1344168404Spjd */ 1345168404Spjd VERIFY(vdev_metaslab_init(vd, txg) == 0); 1346168404Spjd} 1347168404Spjd 1348168404Spjdvoid 1349168404Spjdvdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1350168404Spjd{ 1351168404Spjd ASSERT(vd == vd->vdev_top); 1352168404Spjd ASSERT(ISP2(flags)); 1353168404Spjd 1354168404Spjd if (flags & VDD_METASLAB) 1355168404Spjd (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1356168404Spjd 1357168404Spjd if (flags & VDD_DTL) 1358168404Spjd (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1359168404Spjd 1360168404Spjd (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1361168404Spjd} 1362168404Spjd 1363209962Smm/* 1364209962Smm * DTLs. 1365209962Smm * 1366209962Smm * A vdev's DTL (dirty time log) is the set of transaction groups for which 1367209962Smm * the vdev has less than perfect replication. There are three kinds of DTL: 1368209962Smm * 1369209962Smm * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1370209962Smm * 1371209962Smm * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1372209962Smm * 1373209962Smm * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1374209962Smm * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1375209962Smm * txgs that was scrubbed. 1376209962Smm * 1377209962Smm * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1378209962Smm * persistent errors or just some device being offline. 1379209962Smm * Unlike the other three, the DTL_OUTAGE map is not generally 1380209962Smm * maintained; it's only computed when needed, typically to 1381209962Smm * determine whether a device can be detached. 1382209962Smm * 1383209962Smm * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1384209962Smm * either has the data or it doesn't. 1385209962Smm * 1386209962Smm * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1387209962Smm * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1388209962Smm * if any child is less than fully replicated, then so is its parent. 1389209962Smm * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1390209962Smm * comprising only those txgs which appear in 'maxfaults' or more children; 1391209962Smm * those are the txgs we don't have enough replication to read. For example, 1392209962Smm * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1393209962Smm * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1394209962Smm * two child DTL_MISSING maps. 1395209962Smm * 1396209962Smm * It should be clear from the above that to compute the DTLs and outage maps 1397209962Smm * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1398209962Smm * Therefore, that is all we keep on disk. When loading the pool, or after 1399209962Smm * a configuration change, we generate all other DTLs from first principles. 1400209962Smm */ 1401168404Spjdvoid 1402209962Smmvdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1403168404Spjd{ 1404209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1405209962Smm 1406209962Smm ASSERT(t < DTL_TYPES); 1407209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1408209962Smm 1409168404Spjd mutex_enter(sm->sm_lock); 1410168404Spjd if (!space_map_contains(sm, txg, size)) 1411168404Spjd space_map_add(sm, txg, size); 1412168404Spjd mutex_exit(sm->sm_lock); 1413168404Spjd} 1414168404Spjd 1415209962Smmboolean_t 1416209962Smmvdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1417168404Spjd{ 1418209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1419209962Smm boolean_t dirty = B_FALSE; 1420168404Spjd 1421209962Smm ASSERT(t < DTL_TYPES); 1422209962Smm ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1423168404Spjd 1424168404Spjd mutex_enter(sm->sm_lock); 1425209962Smm if (sm->sm_space != 0) 1426209962Smm dirty = space_map_contains(sm, txg, size); 1427168404Spjd mutex_exit(sm->sm_lock); 1428168404Spjd 1429168404Spjd return (dirty); 1430168404Spjd} 1431168404Spjd 1432209962Smmboolean_t 1433209962Smmvdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1434209962Smm{ 1435209962Smm space_map_t *sm = &vd->vdev_dtl[t]; 1436209962Smm boolean_t empty; 1437209962Smm 1438209962Smm mutex_enter(sm->sm_lock); 1439209962Smm empty = (sm->sm_space == 0); 1440209962Smm mutex_exit(sm->sm_lock); 1441209962Smm 1442209962Smm return (empty); 1443209962Smm} 1444209962Smm 1445168404Spjd/* 1446168404Spjd * Reassess DTLs after a config change or scrub completion. 1447168404Spjd */ 1448168404Spjdvoid 1449168404Spjdvdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1450168404Spjd{ 1451168404Spjd spa_t *spa = vd->vdev_spa; 1452209962Smm avl_tree_t reftree; 1453209962Smm int minref; 1454168404Spjd 1455209962Smm ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1456168404Spjd 1457209962Smm for (int c = 0; c < vd->vdev_children; c++) 1458209962Smm vdev_dtl_reassess(vd->vdev_child[c], txg, 1459209962Smm scrub_txg, scrub_done); 1460209962Smm 1461209962Smm if (vd == spa->spa_root_vdev) 1462209962Smm return; 1463209962Smm 1464209962Smm if (vd->vdev_ops->vdev_op_leaf) { 1465168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1466185029Spjd if (scrub_txg != 0 && 1467185029Spjd (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { 1468185029Spjd /* XXX should check scrub_done? */ 1469185029Spjd /* 1470185029Spjd * We completed a scrub up to scrub_txg. If we 1471185029Spjd * did it without rebooting, then the scrub dtl 1472185029Spjd * will be valid, so excise the old region and 1473185029Spjd * fold in the scrub dtl. Otherwise, leave the 1474185029Spjd * dtl as-is if there was an error. 1475209962Smm * 1476209962Smm * There's little trick here: to excise the beginning 1477209962Smm * of the DTL_MISSING map, we put it into a reference 1478209962Smm * tree and then add a segment with refcnt -1 that 1479209962Smm * covers the range [0, scrub_txg). This means 1480209962Smm * that each txg in that range has refcnt -1 or 0. 1481209962Smm * We then add DTL_SCRUB with a refcnt of 2, so that 1482209962Smm * entries in the range [0, scrub_txg) will have a 1483209962Smm * positive refcnt -- either 1 or 2. We then convert 1484209962Smm * the reference tree into the new DTL_MISSING map. 1485185029Spjd */ 1486209962Smm space_map_ref_create(&reftree); 1487209962Smm space_map_ref_add_map(&reftree, 1488209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1489209962Smm space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 1490209962Smm space_map_ref_add_map(&reftree, 1491209962Smm &vd->vdev_dtl[DTL_SCRUB], 2); 1492209962Smm space_map_ref_generate_map(&reftree, 1493209962Smm &vd->vdev_dtl[DTL_MISSING], 1); 1494209962Smm space_map_ref_destroy(&reftree); 1495168404Spjd } 1496209962Smm space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1497209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1498209962Smm space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1499168404Spjd if (scrub_done) 1500209962Smm space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1501209962Smm space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1502209962Smm if (!vdev_readable(vd)) 1503209962Smm space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1504209962Smm else 1505209962Smm space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1506209962Smm space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1507168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1508185029Spjd 1509168404Spjd if (txg != 0) 1510168404Spjd vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1511168404Spjd return; 1512168404Spjd } 1513168404Spjd 1514168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1515209962Smm for (int t = 0; t < DTL_TYPES; t++) { 1516209962Smm /* account for child's outage in parent's missing map */ 1517209962Smm int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1518209962Smm if (t == DTL_SCRUB) 1519209962Smm continue; /* leaf vdevs only */ 1520209962Smm if (t == DTL_PARTIAL) 1521209962Smm minref = 1; /* i.e. non-zero */ 1522209962Smm else if (vd->vdev_nparity != 0) 1523209962Smm minref = vd->vdev_nparity + 1; /* RAID-Z */ 1524209962Smm else 1525209962Smm minref = vd->vdev_children; /* any kind of mirror */ 1526209962Smm space_map_ref_create(&reftree); 1527209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1528209962Smm vdev_t *cvd = vd->vdev_child[c]; 1529209962Smm mutex_enter(&cvd->vdev_dtl_lock); 1530209962Smm space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 1531209962Smm mutex_exit(&cvd->vdev_dtl_lock); 1532209962Smm } 1533209962Smm space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 1534209962Smm space_map_ref_destroy(&reftree); 1535209962Smm } 1536168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1537168404Spjd} 1538168404Spjd 1539168404Spjdstatic int 1540168404Spjdvdev_dtl_load(vdev_t *vd) 1541168404Spjd{ 1542168404Spjd spa_t *spa = vd->vdev_spa; 1543209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1544168404Spjd objset_t *mos = spa->spa_meta_objset; 1545168404Spjd dmu_buf_t *db; 1546168404Spjd int error; 1547168404Spjd 1548168404Spjd ASSERT(vd->vdev_children == 0); 1549168404Spjd 1550168404Spjd if (smo->smo_object == 0) 1551168404Spjd return (0); 1552168404Spjd 1553168404Spjd if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1554168404Spjd return (error); 1555168404Spjd 1556185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1557185029Spjd bcopy(db->db_data, smo, sizeof (*smo)); 1558168404Spjd dmu_buf_rele(db, FTAG); 1559168404Spjd 1560168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1561209962Smm error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 1562209962Smm NULL, SM_ALLOC, smo, mos); 1563168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1564168404Spjd 1565168404Spjd return (error); 1566168404Spjd} 1567168404Spjd 1568168404Spjdvoid 1569168404Spjdvdev_dtl_sync(vdev_t *vd, uint64_t txg) 1570168404Spjd{ 1571168404Spjd spa_t *spa = vd->vdev_spa; 1572209962Smm space_map_obj_t *smo = &vd->vdev_dtl_smo; 1573209962Smm space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1574168404Spjd objset_t *mos = spa->spa_meta_objset; 1575168404Spjd space_map_t smsync; 1576168404Spjd kmutex_t smlock; 1577168404Spjd dmu_buf_t *db; 1578168404Spjd dmu_tx_t *tx; 1579168404Spjd 1580168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1581168404Spjd 1582168404Spjd if (vd->vdev_detached) { 1583168404Spjd if (smo->smo_object != 0) { 1584168404Spjd int err = dmu_object_free(mos, smo->smo_object, tx); 1585168404Spjd ASSERT3U(err, ==, 0); 1586168404Spjd smo->smo_object = 0; 1587168404Spjd } 1588168404Spjd dmu_tx_commit(tx); 1589168404Spjd return; 1590168404Spjd } 1591168404Spjd 1592168404Spjd if (smo->smo_object == 0) { 1593168404Spjd ASSERT(smo->smo_objsize == 0); 1594168404Spjd ASSERT(smo->smo_alloc == 0); 1595168404Spjd smo->smo_object = dmu_object_alloc(mos, 1596168404Spjd DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1597168404Spjd DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1598168404Spjd ASSERT(smo->smo_object != 0); 1599168404Spjd vdev_config_dirty(vd->vdev_top); 1600168404Spjd } 1601168404Spjd 1602168404Spjd mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1603168404Spjd 1604168404Spjd space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1605168404Spjd &smlock); 1606168404Spjd 1607168404Spjd mutex_enter(&smlock); 1608168404Spjd 1609168404Spjd mutex_enter(&vd->vdev_dtl_lock); 1610168404Spjd space_map_walk(sm, space_map_add, &smsync); 1611168404Spjd mutex_exit(&vd->vdev_dtl_lock); 1612168404Spjd 1613168404Spjd space_map_truncate(smo, mos, tx); 1614168404Spjd space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1615168404Spjd 1616168404Spjd space_map_destroy(&smsync); 1617168404Spjd 1618168404Spjd mutex_exit(&smlock); 1619168404Spjd mutex_destroy(&smlock); 1620168404Spjd 1621168404Spjd VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1622168404Spjd dmu_buf_will_dirty(db, tx); 1623185029Spjd ASSERT3U(db->db_size, >=, sizeof (*smo)); 1624185029Spjd bcopy(smo, db->db_data, sizeof (*smo)); 1625168404Spjd dmu_buf_rele(db, FTAG); 1626168404Spjd 1627168404Spjd dmu_tx_commit(tx); 1628168404Spjd} 1629168404Spjd 1630185029Spjd/* 1631209962Smm * Determine whether the specified vdev can be offlined/detached/removed 1632209962Smm * without losing data. 1633209962Smm */ 1634209962Smmboolean_t 1635209962Smmvdev_dtl_required(vdev_t *vd) 1636209962Smm{ 1637209962Smm spa_t *spa = vd->vdev_spa; 1638209962Smm vdev_t *tvd = vd->vdev_top; 1639209962Smm uint8_t cant_read = vd->vdev_cant_read; 1640209962Smm boolean_t required; 1641209962Smm 1642209962Smm ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1643209962Smm 1644209962Smm if (vd == spa->spa_root_vdev || vd == tvd) 1645209962Smm return (B_TRUE); 1646209962Smm 1647209962Smm /* 1648209962Smm * Temporarily mark the device as unreadable, and then determine 1649209962Smm * whether this results in any DTL outages in the top-level vdev. 1650209962Smm * If not, we can safely offline/detach/remove the device. 1651209962Smm */ 1652209962Smm vd->vdev_cant_read = B_TRUE; 1653209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1654209962Smm required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 1655209962Smm vd->vdev_cant_read = cant_read; 1656209962Smm vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1657209962Smm 1658209962Smm return (required); 1659209962Smm} 1660209962Smm 1661209962Smm/* 1662185029Spjd * Determine if resilver is needed, and if so the txg range. 1663185029Spjd */ 1664185029Spjdboolean_t 1665185029Spjdvdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1666185029Spjd{ 1667185029Spjd boolean_t needed = B_FALSE; 1668185029Spjd uint64_t thismin = UINT64_MAX; 1669185029Spjd uint64_t thismax = 0; 1670185029Spjd 1671185029Spjd if (vd->vdev_children == 0) { 1672185029Spjd mutex_enter(&vd->vdev_dtl_lock); 1673209962Smm if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 1674209962Smm vdev_writeable(vd)) { 1675185029Spjd space_seg_t *ss; 1676185029Spjd 1677209962Smm ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1678185029Spjd thismin = ss->ss_start - 1; 1679209962Smm ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1680185029Spjd thismax = ss->ss_end; 1681185029Spjd needed = B_TRUE; 1682185029Spjd } 1683185029Spjd mutex_exit(&vd->vdev_dtl_lock); 1684185029Spjd } else { 1685209962Smm for (int c = 0; c < vd->vdev_children; c++) { 1686185029Spjd vdev_t *cvd = vd->vdev_child[c]; 1687185029Spjd uint64_t cmin, cmax; 1688185029Spjd 1689185029Spjd if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1690185029Spjd thismin = MIN(thismin, cmin); 1691185029Spjd thismax = MAX(thismax, cmax); 1692185029Spjd needed = B_TRUE; 1693185029Spjd } 1694185029Spjd } 1695185029Spjd } 1696185029Spjd 1697185029Spjd if (needed && minp) { 1698185029Spjd *minp = thismin; 1699185029Spjd *maxp = thismax; 1700185029Spjd } 1701185029Spjd return (needed); 1702185029Spjd} 1703185029Spjd 1704168404Spjdvoid 1705168404Spjdvdev_load(vdev_t *vd) 1706168404Spjd{ 1707168404Spjd /* 1708168404Spjd * Recursively load all children. 1709168404Spjd */ 1710209962Smm for (int c = 0; c < vd->vdev_children; c++) 1711168404Spjd vdev_load(vd->vdev_child[c]); 1712168404Spjd 1713168404Spjd /* 1714168404Spjd * If this is a top-level vdev, initialize its metaslabs. 1715168404Spjd */ 1716168404Spjd if (vd == vd->vdev_top && 1717168404Spjd (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1718168404Spjd vdev_metaslab_init(vd, 0) != 0)) 1719168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1720168404Spjd VDEV_AUX_CORRUPT_DATA); 1721168404Spjd 1722168404Spjd /* 1723168404Spjd * If this is a leaf vdev, load its DTL. 1724168404Spjd */ 1725168404Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1726168404Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1727168404Spjd VDEV_AUX_CORRUPT_DATA); 1728168404Spjd} 1729168404Spjd 1730168404Spjd/* 1731185029Spjd * The special vdev case is used for hot spares and l2cache devices. Its 1732185029Spjd * sole purpose it to set the vdev state for the associated vdev. To do this, 1733185029Spjd * we make sure that we can open the underlying device, then try to read the 1734185029Spjd * label, and make sure that the label is sane and that it hasn't been 1735185029Spjd * repurposed to another pool. 1736168404Spjd */ 1737168404Spjdint 1738185029Spjdvdev_validate_aux(vdev_t *vd) 1739168404Spjd{ 1740168404Spjd nvlist_t *label; 1741168404Spjd uint64_t guid, version; 1742168404Spjd uint64_t state; 1743168404Spjd 1744185029Spjd if (!vdev_readable(vd)) 1745185029Spjd return (0); 1746185029Spjd 1747168404Spjd if ((label = vdev_label_read_config(vd)) == NULL) { 1748168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1749168404Spjd VDEV_AUX_CORRUPT_DATA); 1750168404Spjd return (-1); 1751168404Spjd } 1752168404Spjd 1753168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1754185029Spjd version > SPA_VERSION || 1755168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 1756168404Spjd guid != vd->vdev_guid || 1757168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 1758168404Spjd vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1759168404Spjd VDEV_AUX_CORRUPT_DATA); 1760168404Spjd nvlist_free(label); 1761168404Spjd return (-1); 1762168404Spjd } 1763168404Spjd 1764168404Spjd /* 1765168404Spjd * We don't actually check the pool state here. If it's in fact in 1766168404Spjd * use by another pool, we update this fact on the fly when requested. 1767168404Spjd */ 1768168404Spjd nvlist_free(label); 1769168404Spjd return (0); 1770168404Spjd} 1771168404Spjd 1772168404Spjdvoid 1773168404Spjdvdev_sync_done(vdev_t *vd, uint64_t txg) 1774168404Spjd{ 1775168404Spjd metaslab_t *msp; 1776211931Smm boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 1777168404Spjd 1778168404Spjd while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 1779168404Spjd metaslab_sync_done(msp, txg); 1780211931Smm 1781211931Smm if (reassess) 1782211931Smm metaslab_sync_reassess(vd->vdev_mg); 1783168404Spjd} 1784168404Spjd 1785168404Spjdvoid 1786168404Spjdvdev_sync(vdev_t *vd, uint64_t txg) 1787168404Spjd{ 1788168404Spjd spa_t *spa = vd->vdev_spa; 1789168404Spjd vdev_t *lvd; 1790168404Spjd metaslab_t *msp; 1791168404Spjd dmu_tx_t *tx; 1792168404Spjd 1793168404Spjd if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 1794168404Spjd ASSERT(vd == vd->vdev_top); 1795168404Spjd tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1796168404Spjd vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 1797168404Spjd DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 1798168404Spjd ASSERT(vd->vdev_ms_array != 0); 1799168404Spjd vdev_config_dirty(vd); 1800168404Spjd dmu_tx_commit(tx); 1801168404Spjd } 1802168404Spjd 1803168404Spjd while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 1804168404Spjd metaslab_sync(msp, txg); 1805168404Spjd (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 1806168404Spjd } 1807168404Spjd 1808168404Spjd while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 1809168404Spjd vdev_dtl_sync(lvd, txg); 1810168404Spjd 1811168404Spjd (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 1812168404Spjd} 1813168404Spjd 1814168404Spjduint64_t 1815168404Spjdvdev_psize_to_asize(vdev_t *vd, uint64_t psize) 1816168404Spjd{ 1817168404Spjd return (vd->vdev_ops->vdev_op_asize(vd, psize)); 1818168404Spjd} 1819168404Spjd 1820185029Spjd/* 1821185029Spjd * Mark the given vdev faulted. A faulted vdev behaves as if the device could 1822185029Spjd * not be opened, and no I/O is attempted. 1823185029Spjd */ 1824185029Spjdint 1825185029Spjdvdev_fault(spa_t *spa, uint64_t guid) 1826168404Spjd{ 1827185029Spjd vdev_t *vd; 1828168404Spjd 1829185029Spjd spa_vdev_state_enter(spa); 1830185029Spjd 1831185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1832185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1833185029Spjd 1834185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 1835185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1836185029Spjd 1837185029Spjd /* 1838185029Spjd * Faulted state takes precedence over degraded. 1839185029Spjd */ 1840185029Spjd vd->vdev_faulted = 1ULL; 1841185029Spjd vd->vdev_degraded = 0ULL; 1842185029Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED); 1843185029Spjd 1844185029Spjd /* 1845185029Spjd * If marking the vdev as faulted cause the top-level vdev to become 1846185029Spjd * unavailable, then back off and simply mark the vdev as degraded 1847185029Spjd * instead. 1848185029Spjd */ 1849185029Spjd if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) { 1850185029Spjd vd->vdev_degraded = 1ULL; 1851185029Spjd vd->vdev_faulted = 0ULL; 1852185029Spjd 1853185029Spjd /* 1854185029Spjd * If we reopen the device and it's not dead, only then do we 1855185029Spjd * mark it degraded. 1856185029Spjd */ 1857185029Spjd vdev_reopen(vd); 1858185029Spjd 1859185029Spjd if (vdev_readable(vd)) { 1860185029Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 1861185029Spjd VDEV_AUX_ERR_EXCEEDED); 1862185029Spjd } 1863185029Spjd } 1864185029Spjd 1865185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 1866168404Spjd} 1867168404Spjd 1868185029Spjd/* 1869185029Spjd * Mark the given vdev degraded. A degraded vdev is purely an indication to the 1870185029Spjd * user that something is wrong. The vdev continues to operate as normal as far 1871185029Spjd * as I/O is concerned. 1872185029Spjd */ 1873185029Spjdint 1874185029Spjdvdev_degrade(spa_t *spa, uint64_t guid) 1875168404Spjd{ 1876185029Spjd vdev_t *vd; 1877168404Spjd 1878185029Spjd spa_vdev_state_enter(spa); 1879168404Spjd 1880185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1881185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1882168404Spjd 1883185029Spjd if (!vd->vdev_ops->vdev_op_leaf) 1884185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1885185029Spjd 1886185029Spjd /* 1887185029Spjd * If the vdev is already faulted, then don't do anything. 1888185029Spjd */ 1889185029Spjd if (vd->vdev_faulted || vd->vdev_degraded) 1890185029Spjd return (spa_vdev_state_exit(spa, NULL, 0)); 1891185029Spjd 1892185029Spjd vd->vdev_degraded = 1ULL; 1893185029Spjd if (!vdev_is_dead(vd)) 1894185029Spjd vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 1895185029Spjd VDEV_AUX_ERR_EXCEEDED); 1896185029Spjd 1897185029Spjd return (spa_vdev_state_exit(spa, vd, 0)); 1898168404Spjd} 1899168404Spjd 1900185029Spjd/* 1901185029Spjd * Online the given vdev. If 'unspare' is set, it implies two things. First, 1902185029Spjd * any attached spare device should be detached when the device finishes 1903185029Spjd * resilvering. Second, the online should be treated like a 'test' online case, 1904185029Spjd * so no FMA events are generated if the device fails to open. 1905185029Spjd */ 1906168404Spjdint 1907185029Spjdvdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 1908168404Spjd{ 1909185029Spjd vdev_t *vd; 1910168404Spjd 1911185029Spjd spa_vdev_state_enter(spa); 1912168404Spjd 1913185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1914185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1915168404Spjd 1916168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1917185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1918168404Spjd 1919168404Spjd vd->vdev_offline = B_FALSE; 1920168404Spjd vd->vdev_tmpoffline = B_FALSE; 1921185029Spjd vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 1922185029Spjd vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 1923168404Spjd vdev_reopen(vd->vdev_top); 1924185029Spjd vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 1925168404Spjd 1926185029Spjd if (newstate) 1927185029Spjd *newstate = vd->vdev_state; 1928185029Spjd if ((flags & ZFS_ONLINE_UNSPARE) && 1929185029Spjd !vdev_is_dead(vd) && vd->vdev_parent && 1930185029Spjd vd->vdev_parent->vdev_ops == &vdev_spare_ops && 1931185029Spjd vd->vdev_parent->vdev_child[0] == vd) 1932185029Spjd vd->vdev_unspare = B_TRUE; 1933168404Spjd 1934209962Smm return (spa_vdev_state_exit(spa, vd, 0)); 1935168404Spjd} 1936168404Spjd 1937168404Spjdint 1938185029Spjdvdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 1939168404Spjd{ 1940213197Smm vdev_t *vd, *tvd; 1941213197Smm int error; 1942168404Spjd 1943185029Spjd spa_vdev_state_enter(spa); 1944168404Spjd 1945185029Spjd if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1946185029Spjd return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1947168404Spjd 1948168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1949185029Spjd return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1950168404Spjd 1951213197Smm tvd = vd->vdev_top; 1952213197Smm 1953168404Spjd /* 1954168404Spjd * If the device isn't already offline, try to offline it. 1955168404Spjd */ 1956168404Spjd if (!vd->vdev_offline) { 1957168404Spjd /* 1958209962Smm * If this device has the only valid copy of some data, 1959213197Smm * don't allow it to be offlined. Log devices are always 1960213197Smm * expendable. 1961168404Spjd */ 1962213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 1963213197Smm vdev_dtl_required(vd)) 1964185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 1965168404Spjd 1966168404Spjd /* 1967168404Spjd * Offline this device and reopen its top-level vdev. 1968213197Smm * If the top-level vdev is a log device then just offline 1969213197Smm * it. Otherwise, if this action results in the top-level 1970213197Smm * vdev becoming unusable, undo it and fail the request. 1971168404Spjd */ 1972168404Spjd vd->vdev_offline = B_TRUE; 1973213197Smm vdev_reopen(tvd); 1974213197Smm 1975213197Smm if (!tvd->vdev_islog && vd->vdev_aux == NULL && 1976213197Smm vdev_is_dead(tvd)) { 1977168404Spjd vd->vdev_offline = B_FALSE; 1978213197Smm vdev_reopen(tvd); 1979185029Spjd return (spa_vdev_state_exit(spa, NULL, EBUSY)); 1980168404Spjd } 1981168404Spjd } 1982168404Spjd 1983185029Spjd vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 1984168404Spjd 1985213197Smm if (!tvd->vdev_islog || !vdev_is_dead(tvd)) 1986213197Smm return (spa_vdev_state_exit(spa, vd, 0)); 1987213197Smm 1988213197Smm (void) spa_vdev_state_exit(spa, vd, 0); 1989213197Smm 1990213197Smm error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1991213197Smm NULL, DS_FIND_CHILDREN); 1992213197Smm if (error) { 1993213197Smm (void) vdev_online(spa, guid, 0, NULL); 1994213197Smm return (error); 1995213197Smm } 1996213197Smm /* 1997213197Smm * If we successfully offlined the log device then we need to 1998213197Smm * sync out the current txg so that the "stubby" block can be 1999213197Smm * removed by zil_sync(). 2000213197Smm */ 2001213197Smm txg_wait_synced(spa->spa_dsl_pool, 0); 2002213197Smm return (0); 2003168404Spjd} 2004168404Spjd 2005168404Spjd/* 2006168404Spjd * Clear the error counts associated with this vdev. Unlike vdev_online() and 2007168404Spjd * vdev_offline(), we assume the spa config is locked. We also clear all 2008168404Spjd * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2009168404Spjd */ 2010168404Spjdvoid 2011168404Spjdvdev_clear(spa_t *spa, vdev_t *vd) 2012168404Spjd{ 2013185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2014168404Spjd 2015185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2016185029Spjd 2017168404Spjd if (vd == NULL) 2018185029Spjd vd = rvd; 2019168404Spjd 2020168404Spjd vd->vdev_stat.vs_read_errors = 0; 2021168404Spjd vd->vdev_stat.vs_write_errors = 0; 2022168404Spjd vd->vdev_stat.vs_checksum_errors = 0; 2023168404Spjd 2024185029Spjd for (int c = 0; c < vd->vdev_children; c++) 2025168404Spjd vdev_clear(spa, vd->vdev_child[c]); 2026185029Spjd 2027185029Spjd /* 2028185029Spjd * If we're in the FAULTED state or have experienced failed I/O, then 2029185029Spjd * clear the persistent state and attempt to reopen the device. We 2030185029Spjd * also mark the vdev config dirty, so that the new faulted state is 2031185029Spjd * written out to disk. 2032185029Spjd */ 2033185029Spjd if (vd->vdev_faulted || vd->vdev_degraded || 2034185029Spjd !vdev_readable(vd) || !vdev_writeable(vd)) { 2035185029Spjd 2036185029Spjd vd->vdev_faulted = vd->vdev_degraded = 0; 2037185029Spjd vd->vdev_cant_read = B_FALSE; 2038185029Spjd vd->vdev_cant_write = B_FALSE; 2039185029Spjd 2040185029Spjd vdev_reopen(vd); 2041185029Spjd 2042185029Spjd if (vd != rvd) 2043185029Spjd vdev_state_dirty(vd->vdev_top); 2044185029Spjd 2045185029Spjd if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2046185029Spjd spa_async_request(spa, SPA_ASYNC_RESILVER); 2047185029Spjd 2048185029Spjd spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2049185029Spjd } 2050168404Spjd} 2051168404Spjd 2052185029Spjdboolean_t 2053168404Spjdvdev_is_dead(vdev_t *vd) 2054168404Spjd{ 2055185029Spjd return (vd->vdev_state < VDEV_STATE_DEGRADED); 2056168404Spjd} 2057168404Spjd 2058185029Spjdboolean_t 2059185029Spjdvdev_readable(vdev_t *vd) 2060168404Spjd{ 2061185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2062185029Spjd} 2063168404Spjd 2064185029Spjdboolean_t 2065185029Spjdvdev_writeable(vdev_t *vd) 2066185029Spjd{ 2067185029Spjd return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2068185029Spjd} 2069168404Spjd 2070185029Spjdboolean_t 2071208370Smmvdev_allocatable(vdev_t *vd) 2072208370Smm{ 2073209962Smm uint64_t state = vd->vdev_state; 2074209962Smm 2075208370Smm /* 2076209962Smm * We currently allow allocations from vdevs which may be in the 2077208370Smm * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2078208370Smm * fails to reopen then we'll catch it later when we're holding 2079209962Smm * the proper locks. Note that we have to get the vdev state 2080209962Smm * in a local variable because although it changes atomically, 2081209962Smm * we're asking two separate questions about it. 2082208370Smm */ 2083209962Smm return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2084208370Smm !vd->vdev_cant_write); 2085208370Smm} 2086208370Smm 2087208370Smmboolean_t 2088185029Spjdvdev_accessible(vdev_t *vd, zio_t *zio) 2089185029Spjd{ 2090185029Spjd ASSERT(zio->io_vd == vd); 2091168404Spjd 2092185029Spjd if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2093185029Spjd return (B_FALSE); 2094168404Spjd 2095185029Spjd if (zio->io_type == ZIO_TYPE_READ) 2096185029Spjd return (!vd->vdev_cant_read); 2097168404Spjd 2098185029Spjd if (zio->io_type == ZIO_TYPE_WRITE) 2099185029Spjd return (!vd->vdev_cant_write); 2100168404Spjd 2101185029Spjd return (B_TRUE); 2102168404Spjd} 2103168404Spjd 2104168404Spjd/* 2105168404Spjd * Get statistics for the given vdev. 2106168404Spjd */ 2107168404Spjdvoid 2108168404Spjdvdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2109168404Spjd{ 2110168404Spjd vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2111168404Spjd 2112168404Spjd mutex_enter(&vd->vdev_stat_lock); 2113168404Spjd bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2114185029Spjd vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; 2115168404Spjd vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2116168404Spjd vs->vs_state = vd->vdev_state; 2117168404Spjd vs->vs_rsize = vdev_get_rsize(vd); 2118168404Spjd mutex_exit(&vd->vdev_stat_lock); 2119168404Spjd 2120168404Spjd /* 2121168404Spjd * If we're getting stats on the root vdev, aggregate the I/O counts 2122168404Spjd * over all top-level vdevs (i.e. the direct children of the root). 2123168404Spjd */ 2124168404Spjd if (vd == rvd) { 2125185029Spjd for (int c = 0; c < rvd->vdev_children; c++) { 2126168404Spjd vdev_t *cvd = rvd->vdev_child[c]; 2127168404Spjd vdev_stat_t *cvs = &cvd->vdev_stat; 2128168404Spjd 2129168404Spjd mutex_enter(&vd->vdev_stat_lock); 2130185029Spjd for (int t = 0; t < ZIO_TYPES; t++) { 2131168404Spjd vs->vs_ops[t] += cvs->vs_ops[t]; 2132168404Spjd vs->vs_bytes[t] += cvs->vs_bytes[t]; 2133168404Spjd } 2134168404Spjd vs->vs_scrub_examined += cvs->vs_scrub_examined; 2135168404Spjd mutex_exit(&vd->vdev_stat_lock); 2136168404Spjd } 2137168404Spjd } 2138168404Spjd} 2139168404Spjd 2140168404Spjdvoid 2141185029Spjdvdev_clear_stats(vdev_t *vd) 2142168404Spjd{ 2143185029Spjd mutex_enter(&vd->vdev_stat_lock); 2144185029Spjd vd->vdev_stat.vs_space = 0; 2145185029Spjd vd->vdev_stat.vs_dspace = 0; 2146185029Spjd vd->vdev_stat.vs_alloc = 0; 2147185029Spjd mutex_exit(&vd->vdev_stat_lock); 2148185029Spjd} 2149185029Spjd 2150185029Spjdvoid 2151185029Spjdvdev_stat_update(zio_t *zio, uint64_t psize) 2152185029Spjd{ 2153209962Smm spa_t *spa = zio->io_spa; 2154209962Smm vdev_t *rvd = spa->spa_root_vdev; 2155185029Spjd vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2156168404Spjd vdev_t *pvd; 2157168404Spjd uint64_t txg = zio->io_txg; 2158168404Spjd vdev_stat_t *vs = &vd->vdev_stat; 2159168404Spjd zio_type_t type = zio->io_type; 2160168404Spjd int flags = zio->io_flags; 2161168404Spjd 2162185029Spjd /* 2163185029Spjd * If this i/o is a gang leader, it didn't do any actual work. 2164185029Spjd */ 2165185029Spjd if (zio->io_gang_tree) 2166185029Spjd return; 2167185029Spjd 2168168404Spjd if (zio->io_error == 0) { 2169185029Spjd /* 2170185029Spjd * If this is a root i/o, don't count it -- we've already 2171185029Spjd * counted the top-level vdevs, and vdev_get_stats() will 2172185029Spjd * aggregate them when asked. This reduces contention on 2173185029Spjd * the root vdev_stat_lock and implicitly handles blocks 2174185029Spjd * that compress away to holes, for which there is no i/o. 2175185029Spjd * (Holes never create vdev children, so all the counters 2176185029Spjd * remain zero, which is what we want.) 2177185029Spjd * 2178185029Spjd * Note: this only applies to successful i/o (io_error == 0) 2179185029Spjd * because unlike i/o counts, errors are not additive. 2180185029Spjd * When reading a ditto block, for example, failure of 2181185029Spjd * one top-level vdev does not imply a root-level error. 2182185029Spjd */ 2183185029Spjd if (vd == rvd) 2184185029Spjd return; 2185185029Spjd 2186185029Spjd ASSERT(vd == zio->io_vd); 2187209962Smm 2188209962Smm if (flags & ZIO_FLAG_IO_BYPASS) 2189209962Smm return; 2190209962Smm 2191209962Smm mutex_enter(&vd->vdev_stat_lock); 2192209962Smm 2193185029Spjd if (flags & ZIO_FLAG_IO_REPAIR) { 2194168404Spjd if (flags & ZIO_FLAG_SCRUB_THREAD) 2195185029Spjd vs->vs_scrub_repaired += psize; 2196209962Smm if (flags & ZIO_FLAG_SELF_HEAL) 2197185029Spjd vs->vs_self_healed += psize; 2198168404Spjd } 2199209962Smm 2200209962Smm vs->vs_ops[type]++; 2201209962Smm vs->vs_bytes[type] += psize; 2202209962Smm 2203209962Smm mutex_exit(&vd->vdev_stat_lock); 2204168404Spjd return; 2205168404Spjd } 2206168404Spjd 2207168404Spjd if (flags & ZIO_FLAG_SPECULATIVE) 2208168404Spjd return; 2209168404Spjd 2210213198Smm /* 2211213198Smm * If this is an I/O error that is going to be retried, then ignore the 2212213198Smm * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2213213198Smm * hard errors, when in reality they can happen for any number of 2214213198Smm * innocuous reasons (bus resets, MPxIO link failure, etc). 2215213198Smm */ 2216213198Smm if (zio->io_error == EIO && 2217213198Smm !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2218213198Smm return; 2219213198Smm 2220185029Spjd mutex_enter(&vd->vdev_stat_lock); 2221209962Smm if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2222185029Spjd if (zio->io_error == ECKSUM) 2223185029Spjd vs->vs_checksum_errors++; 2224185029Spjd else 2225185029Spjd vs->vs_read_errors++; 2226168404Spjd } 2227209962Smm if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2228185029Spjd vs->vs_write_errors++; 2229185029Spjd mutex_exit(&vd->vdev_stat_lock); 2230168404Spjd 2231209962Smm if (type == ZIO_TYPE_WRITE && txg != 0 && 2232209962Smm (!(flags & ZIO_FLAG_IO_REPAIR) || 2233209962Smm (flags & ZIO_FLAG_SCRUB_THREAD))) { 2234209962Smm /* 2235209962Smm * This is either a normal write (not a repair), or it's a 2236209962Smm * repair induced by the scrub thread. In the normal case, 2237209962Smm * we commit the DTL change in the same txg as the block 2238209962Smm * was born. In the scrub-induced repair case, we know that 2239209962Smm * scrubs run in first-pass syncing context, so we commit 2240209962Smm * the DTL change in spa->spa_syncing_txg. 2241209962Smm * 2242209962Smm * We currently do not make DTL entries for failed spontaneous 2243209962Smm * self-healing writes triggered by normal (non-scrubbing) 2244209962Smm * reads, because we have no transactional context in which to 2245209962Smm * do so -- and it's not clear that it'd be desirable anyway. 2246209962Smm */ 2247209962Smm if (vd->vdev_ops->vdev_op_leaf) { 2248209962Smm uint64_t commit_txg = txg; 2249209962Smm if (flags & ZIO_FLAG_SCRUB_THREAD) { 2250209962Smm ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2251209962Smm ASSERT(spa_sync_pass(spa) == 1); 2252209962Smm vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2253209962Smm commit_txg = spa->spa_syncing_txg; 2254209962Smm } 2255209962Smm ASSERT(commit_txg >= spa->spa_syncing_txg); 2256209962Smm if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2257168404Spjd return; 2258209962Smm for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2259209962Smm vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2260209962Smm vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2261168404Spjd } 2262209962Smm if (vd != rvd) 2263209962Smm vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2264168404Spjd } 2265168404Spjd} 2266168404Spjd 2267168404Spjdvoid 2268168404Spjdvdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) 2269168404Spjd{ 2270168404Spjd int c; 2271168404Spjd vdev_stat_t *vs = &vd->vdev_stat; 2272168404Spjd 2273168404Spjd for (c = 0; c < vd->vdev_children; c++) 2274168404Spjd vdev_scrub_stat_update(vd->vdev_child[c], type, complete); 2275168404Spjd 2276168404Spjd mutex_enter(&vd->vdev_stat_lock); 2277168404Spjd 2278168404Spjd if (type == POOL_SCRUB_NONE) { 2279168404Spjd /* 2280168404Spjd * Update completion and end time. Leave everything else alone 2281168404Spjd * so we can report what happened during the previous scrub. 2282168404Spjd */ 2283168404Spjd vs->vs_scrub_complete = complete; 2284168404Spjd vs->vs_scrub_end = gethrestime_sec(); 2285168404Spjd } else { 2286168404Spjd vs->vs_scrub_type = type; 2287168404Spjd vs->vs_scrub_complete = 0; 2288168404Spjd vs->vs_scrub_examined = 0; 2289168404Spjd vs->vs_scrub_repaired = 0; 2290168404Spjd vs->vs_scrub_start = gethrestime_sec(); 2291168404Spjd vs->vs_scrub_end = 0; 2292168404Spjd } 2293168404Spjd 2294168404Spjd mutex_exit(&vd->vdev_stat_lock); 2295168404Spjd} 2296168404Spjd 2297168404Spjd/* 2298168404Spjd * Update the in-core space usage stats for this vdev and the root vdev. 2299168404Spjd */ 2300168404Spjdvoid 2301185029Spjdvdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta, 2302185029Spjd boolean_t update_root) 2303168404Spjd{ 2304168404Spjd int64_t dspace_delta = space_delta; 2305185029Spjd spa_t *spa = vd->vdev_spa; 2306185029Spjd vdev_t *rvd = spa->spa_root_vdev; 2307168404Spjd 2308185029Spjd ASSERT(vd == vd->vdev_top); 2309168404Spjd 2310185029Spjd /* 2311185029Spjd * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2312185029Spjd * factor. We must calculate this here and not at the root vdev 2313185029Spjd * because the root vdev's psize-to-asize is simply the max of its 2314185029Spjd * childrens', thus not accurate enough for us. 2315185029Spjd */ 2316185029Spjd ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2317213197Smm ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2318185029Spjd dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2319185029Spjd vd->vdev_deflate_ratio; 2320185029Spjd 2321185029Spjd mutex_enter(&vd->vdev_stat_lock); 2322185029Spjd vd->vdev_stat.vs_space += space_delta; 2323185029Spjd vd->vdev_stat.vs_alloc += alloc_delta; 2324185029Spjd vd->vdev_stat.vs_dspace += dspace_delta; 2325185029Spjd mutex_exit(&vd->vdev_stat_lock); 2326185029Spjd 2327185029Spjd if (update_root) { 2328185029Spjd ASSERT(rvd == vd->vdev_parent); 2329185029Spjd ASSERT(vd->vdev_ms_count != 0); 2330185029Spjd 2331185029Spjd /* 2332185029Spjd * Don't count non-normal (e.g. intent log) space as part of 2333185029Spjd * the pool's capacity. 2334185029Spjd */ 2335185029Spjd if (vd->vdev_mg->mg_class != spa->spa_normal_class) 2336185029Spjd return; 2337185029Spjd 2338185029Spjd mutex_enter(&rvd->vdev_stat_lock); 2339185029Spjd rvd->vdev_stat.vs_space += space_delta; 2340185029Spjd rvd->vdev_stat.vs_alloc += alloc_delta; 2341185029Spjd rvd->vdev_stat.vs_dspace += dspace_delta; 2342185029Spjd mutex_exit(&rvd->vdev_stat_lock); 2343185029Spjd } 2344168404Spjd} 2345168404Spjd 2346168404Spjd/* 2347168404Spjd * Mark a top-level vdev's config as dirty, placing it on the dirty list 2348168404Spjd * so that it will be written out next time the vdev configuration is synced. 2349168404Spjd * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2350168404Spjd */ 2351168404Spjdvoid 2352168404Spjdvdev_config_dirty(vdev_t *vd) 2353168404Spjd{ 2354168404Spjd spa_t *spa = vd->vdev_spa; 2355168404Spjd vdev_t *rvd = spa->spa_root_vdev; 2356168404Spjd int c; 2357168404Spjd 2358168404Spjd /* 2359209962Smm * If this is an aux vdev (as with l2cache and spare devices), then we 2360209962Smm * update the vdev config manually and set the sync flag. 2361185029Spjd */ 2362185029Spjd if (vd->vdev_aux != NULL) { 2363185029Spjd spa_aux_vdev_t *sav = vd->vdev_aux; 2364185029Spjd nvlist_t **aux; 2365185029Spjd uint_t naux; 2366185029Spjd 2367185029Spjd for (c = 0; c < sav->sav_count; c++) { 2368185029Spjd if (sav->sav_vdevs[c] == vd) 2369185029Spjd break; 2370185029Spjd } 2371185029Spjd 2372185029Spjd if (c == sav->sav_count) { 2373185029Spjd /* 2374185029Spjd * We're being removed. There's nothing more to do. 2375185029Spjd */ 2376185029Spjd ASSERT(sav->sav_sync == B_TRUE); 2377185029Spjd return; 2378185029Spjd } 2379185029Spjd 2380185029Spjd sav->sav_sync = B_TRUE; 2381185029Spjd 2382209962Smm if (nvlist_lookup_nvlist_array(sav->sav_config, 2383209962Smm ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 2384209962Smm VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2385209962Smm ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 2386209962Smm } 2387185029Spjd 2388185029Spjd ASSERT(c < naux); 2389185029Spjd 2390185029Spjd /* 2391185029Spjd * Setting the nvlist in the middle if the array is a little 2392185029Spjd * sketchy, but it will work. 2393185029Spjd */ 2394185029Spjd nvlist_free(aux[c]); 2395185029Spjd aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); 2396185029Spjd 2397185029Spjd return; 2398185029Spjd } 2399185029Spjd 2400185029Spjd /* 2401185029Spjd * The dirty list is protected by the SCL_CONFIG lock. The caller 2402185029Spjd * must either hold SCL_CONFIG as writer, or must be the sync thread 2403185029Spjd * (which holds SCL_CONFIG as reader). There's only one sync thread, 2404168404Spjd * so this is sufficient to ensure mutual exclusion. 2405168404Spjd */ 2406185029Spjd ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2407185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2408185029Spjd spa_config_held(spa, SCL_CONFIG, RW_READER))); 2409168404Spjd 2410168404Spjd if (vd == rvd) { 2411168404Spjd for (c = 0; c < rvd->vdev_children; c++) 2412168404Spjd vdev_config_dirty(rvd->vdev_child[c]); 2413168404Spjd } else { 2414168404Spjd ASSERT(vd == vd->vdev_top); 2415168404Spjd 2416185029Spjd if (!list_link_active(&vd->vdev_config_dirty_node)) 2417185029Spjd list_insert_head(&spa->spa_config_dirty_list, vd); 2418168404Spjd } 2419168404Spjd} 2420168404Spjd 2421168404Spjdvoid 2422168404Spjdvdev_config_clean(vdev_t *vd) 2423168404Spjd{ 2424168404Spjd spa_t *spa = vd->vdev_spa; 2425168404Spjd 2426185029Spjd ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2427185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2428185029Spjd spa_config_held(spa, SCL_CONFIG, RW_READER))); 2429168404Spjd 2430185029Spjd ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2431185029Spjd list_remove(&spa->spa_config_dirty_list, vd); 2432168404Spjd} 2433168404Spjd 2434185029Spjd/* 2435185029Spjd * Mark a top-level vdev's state as dirty, so that the next pass of 2436185029Spjd * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2437185029Spjd * the state changes from larger config changes because they require 2438185029Spjd * much less locking, and are often needed for administrative actions. 2439185029Spjd */ 2440168404Spjdvoid 2441185029Spjdvdev_state_dirty(vdev_t *vd) 2442185029Spjd{ 2443185029Spjd spa_t *spa = vd->vdev_spa; 2444185029Spjd 2445185029Spjd ASSERT(vd == vd->vdev_top); 2446185029Spjd 2447185029Spjd /* 2448185029Spjd * The state list is protected by the SCL_STATE lock. The caller 2449185029Spjd * must either hold SCL_STATE as writer, or must be the sync thread 2450185029Spjd * (which holds SCL_STATE as reader). There's only one sync thread, 2451185029Spjd * so this is sufficient to ensure mutual exclusion. 2452185029Spjd */ 2453185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2454185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2455185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2456185029Spjd 2457185029Spjd if (!list_link_active(&vd->vdev_state_dirty_node)) 2458185029Spjd list_insert_head(&spa->spa_state_dirty_list, vd); 2459185029Spjd} 2460185029Spjd 2461185029Spjdvoid 2462185029Spjdvdev_state_clean(vdev_t *vd) 2463185029Spjd{ 2464185029Spjd spa_t *spa = vd->vdev_spa; 2465185029Spjd 2466185029Spjd ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2467185029Spjd (dsl_pool_sync_context(spa_get_dsl(spa)) && 2468185029Spjd spa_config_held(spa, SCL_STATE, RW_READER))); 2469185029Spjd 2470185029Spjd ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2471185029Spjd list_remove(&spa->spa_state_dirty_list, vd); 2472185029Spjd} 2473185029Spjd 2474185029Spjd/* 2475185029Spjd * Propagate vdev state up from children to parent. 2476185029Spjd */ 2477185029Spjdvoid 2478168404Spjdvdev_propagate_state(vdev_t *vd) 2479168404Spjd{ 2480209962Smm spa_t *spa = vd->vdev_spa; 2481209962Smm vdev_t *rvd = spa->spa_root_vdev; 2482168404Spjd int degraded = 0, faulted = 0; 2483168404Spjd int corrupted = 0; 2484168404Spjd int c; 2485168404Spjd vdev_t *child; 2486168404Spjd 2487185029Spjd if (vd->vdev_children > 0) { 2488185029Spjd for (c = 0; c < vd->vdev_children; c++) { 2489185029Spjd child = vd->vdev_child[c]; 2490168404Spjd 2491185029Spjd if (!vdev_readable(child) || 2492209962Smm (!vdev_writeable(child) && spa_writeable(spa))) { 2493185029Spjd /* 2494185029Spjd * Root special: if there is a top-level log 2495185029Spjd * device, treat the root vdev as if it were 2496185029Spjd * degraded. 2497185029Spjd */ 2498185029Spjd if (child->vdev_islog && vd == rvd) 2499185029Spjd degraded++; 2500185029Spjd else 2501185029Spjd faulted++; 2502185029Spjd } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 2503185029Spjd degraded++; 2504185029Spjd } 2505185029Spjd 2506185029Spjd if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 2507185029Spjd corrupted++; 2508185029Spjd } 2509185029Spjd 2510185029Spjd vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 2511185029Spjd 2512185029Spjd /* 2513185029Spjd * Root special: if there is a top-level vdev that cannot be 2514185029Spjd * opened due to corrupted metadata, then propagate the root 2515185029Spjd * vdev's aux state as 'corrupt' rather than 'insufficient 2516185029Spjd * replicas'. 2517185029Spjd */ 2518185029Spjd if (corrupted && vd == rvd && 2519185029Spjd rvd->vdev_state == VDEV_STATE_CANT_OPEN) 2520185029Spjd vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 2521185029Spjd VDEV_AUX_CORRUPT_DATA); 2522168404Spjd } 2523168404Spjd 2524185029Spjd if (vd->vdev_parent) 2525185029Spjd vdev_propagate_state(vd->vdev_parent); 2526168404Spjd} 2527168404Spjd 2528168404Spjd/* 2529168404Spjd * Set a vdev's state. If this is during an open, we don't update the parent 2530168404Spjd * state, because we're in the process of opening children depth-first. 2531168404Spjd * Otherwise, we propagate the change to the parent. 2532168404Spjd * 2533168404Spjd * If this routine places a device in a faulted state, an appropriate ereport is 2534168404Spjd * generated. 2535168404Spjd */ 2536168404Spjdvoid 2537168404Spjdvdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2538168404Spjd{ 2539168404Spjd uint64_t save_state; 2540185029Spjd spa_t *spa = vd->vdev_spa; 2541168404Spjd 2542168404Spjd if (state == vd->vdev_state) { 2543168404Spjd vd->vdev_stat.vs_aux = aux; 2544168404Spjd return; 2545168404Spjd } 2546168404Spjd 2547168404Spjd save_state = vd->vdev_state; 2548168404Spjd 2549168404Spjd vd->vdev_state = state; 2550168404Spjd vd->vdev_stat.vs_aux = aux; 2551168404Spjd 2552173373Spjd /* 2553173373Spjd * If we are setting the vdev state to anything but an open state, then 2554173373Spjd * always close the underlying device. Otherwise, we keep accessible 2555173373Spjd * but invalid devices open forever. We don't call vdev_close() itself, 2556173373Spjd * because that implies some extra checks (offline, etc) that we don't 2557173373Spjd * want here. This is limited to leaf devices, because otherwise 2558173373Spjd * closing the device will affect other children. 2559173373Spjd */ 2560173373Spjd if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) 2561173373Spjd vd->vdev_ops->vdev_op_close(vd); 2562173373Spjd 2563185029Spjd if (vd->vdev_removed && 2564185029Spjd state == VDEV_STATE_CANT_OPEN && 2565185029Spjd (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 2566168404Spjd /* 2567185029Spjd * If the previous state is set to VDEV_STATE_REMOVED, then this 2568185029Spjd * device was previously marked removed and someone attempted to 2569185029Spjd * reopen it. If this failed due to a nonexistent device, then 2570185029Spjd * keep the device in the REMOVED state. We also let this be if 2571185029Spjd * it is one of our special test online cases, which is only 2572185029Spjd * attempting to online the device and shouldn't generate an FMA 2573185029Spjd * fault. 2574185029Spjd */ 2575185029Spjd vd->vdev_state = VDEV_STATE_REMOVED; 2576185029Spjd vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2577185029Spjd } else if (state == VDEV_STATE_REMOVED) { 2578185029Spjd /* 2579185029Spjd * Indicate to the ZFS DE that this device has been removed, and 2580185029Spjd * any recent errors should be ignored. 2581185029Spjd */ 2582185029Spjd zfs_post_remove(spa, vd); 2583185029Spjd vd->vdev_removed = B_TRUE; 2584185029Spjd } else if (state == VDEV_STATE_CANT_OPEN) { 2585185029Spjd /* 2586168404Spjd * If we fail to open a vdev during an import, we mark it as 2587168404Spjd * "not available", which signifies that it was never there to 2588168404Spjd * begin with. Failure to open such a device is not considered 2589168404Spjd * an error. 2590168404Spjd */ 2591185029Spjd if (spa->spa_load_state == SPA_LOAD_IMPORT && 2592168404Spjd vd->vdev_ops->vdev_op_leaf) 2593168404Spjd vd->vdev_not_present = 1; 2594168404Spjd 2595168404Spjd /* 2596168404Spjd * Post the appropriate ereport. If the 'prevstate' field is 2597168404Spjd * set to something other than VDEV_STATE_UNKNOWN, it indicates 2598168404Spjd * that this is part of a vdev_reopen(). In this case, we don't 2599168404Spjd * want to post the ereport if the device was already in the 2600168404Spjd * CANT_OPEN state beforehand. 2601185029Spjd * 2602185029Spjd * If the 'checkremove' flag is set, then this is an attempt to 2603185029Spjd * online the device in response to an insertion event. If we 2604185029Spjd * hit this case, then we have detected an insertion event for a 2605185029Spjd * faulted or offline device that wasn't in the removed state. 2606185029Spjd * In this scenario, we don't post an ereport because we are 2607185029Spjd * about to replace the device, or attempt an online with 2608185029Spjd * vdev_forcefault, which will generate the fault for us. 2609168404Spjd */ 2610185029Spjd if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 2611185029Spjd !vd->vdev_not_present && !vd->vdev_checkremove && 2612185029Spjd vd != spa->spa_root_vdev) { 2613168404Spjd const char *class; 2614168404Spjd 2615168404Spjd switch (aux) { 2616168404Spjd case VDEV_AUX_OPEN_FAILED: 2617168404Spjd class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2618168404Spjd break; 2619168404Spjd case VDEV_AUX_CORRUPT_DATA: 2620168404Spjd class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2621168404Spjd break; 2622168404Spjd case VDEV_AUX_NO_REPLICAS: 2623168404Spjd class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2624168404Spjd break; 2625168404Spjd case VDEV_AUX_BAD_GUID_SUM: 2626168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2627168404Spjd break; 2628168404Spjd case VDEV_AUX_TOO_SMALL: 2629168404Spjd class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 2630168404Spjd break; 2631168404Spjd case VDEV_AUX_BAD_LABEL: 2632168404Spjd class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 2633168404Spjd break; 2634185029Spjd case VDEV_AUX_IO_FAILURE: 2635185029Spjd class = FM_EREPORT_ZFS_IO_FAILURE; 2636185029Spjd break; 2637168404Spjd default: 2638168404Spjd class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 2639168404Spjd } 2640168404Spjd 2641185029Spjd zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 2642168404Spjd } 2643185029Spjd 2644185029Spjd /* Erase any notion of persistent removed state */ 2645185029Spjd vd->vdev_removed = B_FALSE; 2646185029Spjd } else { 2647185029Spjd vd->vdev_removed = B_FALSE; 2648168404Spjd } 2649168404Spjd 2650209962Smm if (!isopen && vd->vdev_parent) 2651209962Smm vdev_propagate_state(vd->vdev_parent); 2652185029Spjd} 2653168404Spjd 2654185029Spjd/* 2655185029Spjd * Check the vdev configuration to ensure that it's capable of supporting 2656193163Sdfr * a root pool. 2657193163Sdfr * 2658193163Sdfr * On Solaris, we do not support RAID-Z or partial configuration. In 2659193163Sdfr * addition, only a single top-level vdev is allowed and none of the 2660193163Sdfr * leaves can be wholedisks. 2661193163Sdfr * 2662193163Sdfr * For FreeBSD, we can boot from any configuration. There is a 2663193163Sdfr * limitation that the boot filesystem must be either uncompressed or 2664193163Sdfr * compresses with lzjb compression but I'm not sure how to enforce 2665193163Sdfr * that here. 2666185029Spjd */ 2667185029Spjdboolean_t 2668185029Spjdvdev_is_bootable(vdev_t *vd) 2669185029Spjd{ 2670213197Smm#ifdef sun 2671185029Spjd if (!vd->vdev_ops->vdev_op_leaf) { 2672185029Spjd char *vdev_type = vd->vdev_ops->vdev_op_type; 2673185029Spjd 2674185029Spjd if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 2675185029Spjd vd->vdev_children > 1) { 2676185029Spjd return (B_FALSE); 2677185029Spjd } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 2678185029Spjd strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 2679185029Spjd return (B_FALSE); 2680185029Spjd } 2681185029Spjd } else if (vd->vdev_wholedisk == 1) { 2682185029Spjd return (B_FALSE); 2683185029Spjd } 2684185029Spjd 2685185029Spjd for (c = 0; c < vd->vdev_children; c++) { 2686185029Spjd if (!vdev_is_bootable(vd->vdev_child[c])) 2687185029Spjd return (B_FALSE); 2688185029Spjd } 2689213197Smm#endif /* sun */ 2690185029Spjd return (B_TRUE); 2691168404Spjd} 2692213197Smm 2693213197Smmvoid 2694213197Smmvdev_load_log_state(vdev_t *vd, nvlist_t *nv) 2695213197Smm{ 2696213197Smm uint_t c, children; 2697213197Smm nvlist_t **child; 2698213197Smm uint64_t val; 2699213197Smm spa_t *spa = vd->vdev_spa; 2700213197Smm 2701213197Smm if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 2702213197Smm &child, &children) == 0) { 2703213197Smm for (c = 0; c < children; c++) 2704213197Smm vdev_load_log_state(vd->vdev_child[c], child[c]); 2705213197Smm } 2706213197Smm 2707213197Smm if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv, 2708213197Smm ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) { 2709213197Smm 2710213197Smm /* 2711213197Smm * It would be nice to call vdev_offline() 2712213197Smm * directly but the pool isn't fully loaded and 2713213197Smm * the txg threads have not been started yet. 2714213197Smm */ 2715213197Smm spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER); 2716213197Smm vd->vdev_offline = val; 2717213197Smm vdev_reopen(vd->vdev_top); 2718213197Smm spa_config_exit(spa, SCL_STATE_ALL, FTAG); 2719213197Smm } 2720213197Smm} 2721