1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 23a3f829aeSBill Moore * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24fa9e4066Sahrens * Use is subject to license terms. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/zfs_context.h> 28ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 29fa9e4066Sahrens #include <sys/spa.h> 30fa9e4066Sahrens #include <sys/spa_impl.h> 31fa9e4066Sahrens #include <sys/dmu.h> 32fa9e4066Sahrens #include <sys/dmu_tx.h> 33fa9e4066Sahrens #include <sys/vdev_impl.h> 34fa9e4066Sahrens #include <sys/uberblock_impl.h> 35fa9e4066Sahrens #include <sys/metaslab.h> 36fa9e4066Sahrens #include <sys/metaslab_impl.h> 37fa9e4066Sahrens #include <sys/space_map.h> 38fa9e4066Sahrens #include <sys/zio.h> 39fa9e4066Sahrens #include <sys/zap.h> 40fa9e4066Sahrens #include <sys/fs/zfs.h> 41c5904d13Seschrock #include <sys/arc.h> 42*e6ca193dSGeorge Wilson #include <sys/zil.h> 43fa9e4066Sahrens 44fa9e4066Sahrens /* 45fa9e4066Sahrens * Virtual device management. 46fa9e4066Sahrens */ 47fa9e4066Sahrens 48fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 49fa9e4066Sahrens &vdev_root_ops, 50fa9e4066Sahrens &vdev_raidz_ops, 51fa9e4066Sahrens &vdev_mirror_ops, 52fa9e4066Sahrens &vdev_replacing_ops, 5399653d4eSeschrock &vdev_spare_ops, 54fa9e4066Sahrens &vdev_disk_ops, 55fa9e4066Sahrens &vdev_file_ops, 56fa9e4066Sahrens &vdev_missing_ops, 57fa9e4066Sahrens NULL 58fa9e4066Sahrens }; 59fa9e4066Sahrens 60088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 61088f3894Sahrens int zfs_scrub_limit = 10; 6205b2b3b8Smishra 63fa9e4066Sahrens /* 64fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 65fa9e4066Sahrens */ 66fa9e4066Sahrens static vdev_ops_t * 67fa9e4066Sahrens vdev_getops(const char *type) 68fa9e4066Sahrens { 69fa9e4066Sahrens vdev_ops_t *ops, **opspp; 70fa9e4066Sahrens 71fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 72fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 73fa9e4066Sahrens break; 74fa9e4066Sahrens 75fa9e4066Sahrens return (ops); 76fa9e4066Sahrens } 77fa9e4066Sahrens 78fa9e4066Sahrens /* 79fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 80fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 81fa9e4066Sahrens */ 82fa9e4066Sahrens uint64_t 83fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 84fa9e4066Sahrens { 85ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 86fa9e4066Sahrens uint64_t csize; 87fa9e4066Sahrens uint64_t c; 88fa9e4066Sahrens 89fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) { 90fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 91fa9e4066Sahrens asize = MAX(asize, csize); 92fa9e4066Sahrens } 93fa9e4066Sahrens 94fa9e4066Sahrens return (asize); 95fa9e4066Sahrens } 96fa9e4066Sahrens 972a79c5feSlling /* 982a79c5feSlling * Get the replaceable or attachable device size. 992a79c5feSlling * If the parent is a mirror or raidz, the replaceable size is the minimum 1002a79c5feSlling * psize of all its children. For the rest, just return our own psize. 1012a79c5feSlling * 1022a79c5feSlling * e.g. 1032a79c5feSlling * psize rsize 1042a79c5feSlling * root - - 1052a79c5feSlling * mirror/raidz - - 1062a79c5feSlling * disk1 20g 20g 1072a79c5feSlling * disk2 40g 20g 1082a79c5feSlling * disk3 80g 80g 1092a79c5feSlling */ 1102a79c5feSlling uint64_t 1112a79c5feSlling vdev_get_rsize(vdev_t *vd) 1122a79c5feSlling { 1132a79c5feSlling vdev_t *pvd, *cvd; 1142a79c5feSlling uint64_t c, rsize; 1152a79c5feSlling 1162a79c5feSlling pvd = vd->vdev_parent; 1172a79c5feSlling 1182a79c5feSlling /* 1192a79c5feSlling * If our parent is NULL or the root, just return our own psize. 1202a79c5feSlling */ 1212a79c5feSlling if (pvd == NULL || pvd->vdev_parent == NULL) 1222a79c5feSlling return (vd->vdev_psize); 1232a79c5feSlling 1242a79c5feSlling rsize = 0; 1252a79c5feSlling 1262a79c5feSlling for (c = 0; c < pvd->vdev_children; c++) { 1272a79c5feSlling cvd = pvd->vdev_child[c]; 1282a79c5feSlling rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1; 1292a79c5feSlling } 1302a79c5feSlling 1312a79c5feSlling return (rsize); 1322a79c5feSlling } 1332a79c5feSlling 134fa9e4066Sahrens vdev_t * 135fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 136fa9e4066Sahrens { 137fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 138fa9e4066Sahrens 139e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 140e05725b1Sbonwick 141088f3894Sahrens if (vdev < rvd->vdev_children) { 142088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 143fa9e4066Sahrens return (rvd->vdev_child[vdev]); 144088f3894Sahrens } 145fa9e4066Sahrens 146fa9e4066Sahrens return (NULL); 147fa9e4066Sahrens } 148fa9e4066Sahrens 149fa9e4066Sahrens vdev_t * 150fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 151fa9e4066Sahrens { 152fa9e4066Sahrens int c; 153fa9e4066Sahrens vdev_t *mvd; 154fa9e4066Sahrens 1550e34b6a7Sbonwick if (vd->vdev_guid == guid) 156fa9e4066Sahrens return (vd); 157fa9e4066Sahrens 158fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 159fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 160fa9e4066Sahrens NULL) 161fa9e4066Sahrens return (mvd); 162fa9e4066Sahrens 163fa9e4066Sahrens return (NULL); 164fa9e4066Sahrens } 165fa9e4066Sahrens 166fa9e4066Sahrens void 167fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 168fa9e4066Sahrens { 169fa9e4066Sahrens size_t oldsize, newsize; 170fa9e4066Sahrens uint64_t id = cvd->vdev_id; 171fa9e4066Sahrens vdev_t **newchild; 172fa9e4066Sahrens 173e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 174fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 175fa9e4066Sahrens 176fa9e4066Sahrens cvd->vdev_parent = pvd; 177fa9e4066Sahrens 178fa9e4066Sahrens if (pvd == NULL) 179fa9e4066Sahrens return; 180fa9e4066Sahrens 181fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 182fa9e4066Sahrens 183fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 184fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 185fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 186fa9e4066Sahrens 187fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 188fa9e4066Sahrens if (pvd->vdev_child != NULL) { 189fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 190fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 191fa9e4066Sahrens } 192fa9e4066Sahrens 193fa9e4066Sahrens pvd->vdev_child = newchild; 194fa9e4066Sahrens pvd->vdev_child[id] = cvd; 195fa9e4066Sahrens 196fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 197fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 198fa9e4066Sahrens 199fa9e4066Sahrens /* 200fa9e4066Sahrens * Walk up all ancestors to update guid sum. 201fa9e4066Sahrens */ 202fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 203fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 20405b2b3b8Smishra 20505b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 20605b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; 207fa9e4066Sahrens } 208fa9e4066Sahrens 209fa9e4066Sahrens void 210fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 211fa9e4066Sahrens { 212fa9e4066Sahrens int c; 213fa9e4066Sahrens uint_t id = cvd->vdev_id; 214fa9e4066Sahrens 215fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 216fa9e4066Sahrens 217fa9e4066Sahrens if (pvd == NULL) 218fa9e4066Sahrens return; 219fa9e4066Sahrens 220fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 221fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 222fa9e4066Sahrens 223fa9e4066Sahrens pvd->vdev_child[id] = NULL; 224fa9e4066Sahrens cvd->vdev_parent = NULL; 225fa9e4066Sahrens 226fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 227fa9e4066Sahrens if (pvd->vdev_child[c]) 228fa9e4066Sahrens break; 229fa9e4066Sahrens 230fa9e4066Sahrens if (c == pvd->vdev_children) { 231fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 232fa9e4066Sahrens pvd->vdev_child = NULL; 233fa9e4066Sahrens pvd->vdev_children = 0; 234fa9e4066Sahrens } 235fa9e4066Sahrens 236fa9e4066Sahrens /* 237fa9e4066Sahrens * Walk up all ancestors to update guid sum. 238fa9e4066Sahrens */ 239fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 240fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 24105b2b3b8Smishra 24205b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 24305b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; 244fa9e4066Sahrens } 245fa9e4066Sahrens 246fa9e4066Sahrens /* 247fa9e4066Sahrens * Remove any holes in the child array. 248fa9e4066Sahrens */ 249fa9e4066Sahrens void 250fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 251fa9e4066Sahrens { 252fa9e4066Sahrens vdev_t **newchild, *cvd; 253fa9e4066Sahrens int oldc = pvd->vdev_children; 254fa9e4066Sahrens int newc, c; 255fa9e4066Sahrens 256e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 257fa9e4066Sahrens 258fa9e4066Sahrens for (c = newc = 0; c < oldc; c++) 259fa9e4066Sahrens if (pvd->vdev_child[c]) 260fa9e4066Sahrens newc++; 261fa9e4066Sahrens 262fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 263fa9e4066Sahrens 264fa9e4066Sahrens for (c = newc = 0; c < oldc; c++) { 265fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 266fa9e4066Sahrens newchild[newc] = cvd; 267fa9e4066Sahrens cvd->vdev_id = newc++; 268fa9e4066Sahrens } 269fa9e4066Sahrens } 270fa9e4066Sahrens 271fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 272fa9e4066Sahrens pvd->vdev_child = newchild; 273fa9e4066Sahrens pvd->vdev_children = newc; 274fa9e4066Sahrens } 275fa9e4066Sahrens 276fa9e4066Sahrens /* 277fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 278fa9e4066Sahrens */ 279fa9e4066Sahrens static vdev_t * 280fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 281fa9e4066Sahrens { 282fa9e4066Sahrens vdev_t *vd; 283fa9e4066Sahrens 284fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 285fa9e4066Sahrens 2860e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2870e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2880e34b6a7Sbonwick spa->spa_root_vdev = vd; 2890e34b6a7Sbonwick } 2900e34b6a7Sbonwick 2910e34b6a7Sbonwick if (guid == 0) { 2920e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 2930e34b6a7Sbonwick /* 2940e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 2950e34b6a7Sbonwick * which must be unique among all pools. 2960e34b6a7Sbonwick */ 2970e34b6a7Sbonwick while (guid == 0 || spa_guid_exists(guid, 0)) 2980e34b6a7Sbonwick guid = spa_get_random(-1ULL); 2990e34b6a7Sbonwick } else { 3000e34b6a7Sbonwick /* 3010e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3020e34b6a7Sbonwick */ 3030e34b6a7Sbonwick while (guid == 0 || 3040e34b6a7Sbonwick spa_guid_exists(spa_guid(spa), guid)) 3050e34b6a7Sbonwick guid = spa_get_random(-1ULL); 3060e34b6a7Sbonwick } 3070e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3080e34b6a7Sbonwick } 3090e34b6a7Sbonwick 310fa9e4066Sahrens vd->vdev_spa = spa; 311fa9e4066Sahrens vd->vdev_id = id; 312fa9e4066Sahrens vd->vdev_guid = guid; 313fa9e4066Sahrens vd->vdev_guid_sum = guid; 314fa9e4066Sahrens vd->vdev_ops = ops; 315fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 316fa9e4066Sahrens 317fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3185ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 319e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 3208ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 3218ad4d6ddSJeff Bonwick space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 3228ad4d6ddSJeff Bonwick &vd->vdev_dtl_lock); 3238ad4d6ddSJeff Bonwick } 324fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 325fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 326fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 327fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 328fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3293d7072f8Seschrock vdev_queue_init(vd); 3303d7072f8Seschrock vdev_cache_init(vd); 331fa9e4066Sahrens 332fa9e4066Sahrens return (vd); 333fa9e4066Sahrens } 334fa9e4066Sahrens 335fa9e4066Sahrens /* 336fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 337fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 338fa9e4066Sahrens * different for each case. 339fa9e4066Sahrens */ 34099653d4eSeschrock int 34199653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 34299653d4eSeschrock int alloctype) 343fa9e4066Sahrens { 344fa9e4066Sahrens vdev_ops_t *ops; 345fa9e4066Sahrens char *type; 3468654d025Sperrin uint64_t guid = 0, islog, nparity; 347fa9e4066Sahrens vdev_t *vd; 348fa9e4066Sahrens 349e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 350fa9e4066Sahrens 351fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 35299653d4eSeschrock return (EINVAL); 353fa9e4066Sahrens 354fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 35599653d4eSeschrock return (EINVAL); 356fa9e4066Sahrens 357fa9e4066Sahrens /* 358fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 359fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 360fa9e4066Sahrens */ 361fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 362fa9e4066Sahrens uint64_t label_id; 363fa9e4066Sahrens 364fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 365fa9e4066Sahrens label_id != id) 36699653d4eSeschrock return (EINVAL); 367fa9e4066Sahrens 368fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 36999653d4eSeschrock return (EINVAL); 37099653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 37199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37299653d4eSeschrock return (EINVAL); 373fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 374fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 375fa94a07fSbrendan return (EINVAL); 376fa9e4066Sahrens } 377fa9e4066Sahrens 37899653d4eSeschrock /* 37999653d4eSeschrock * The first allocated vdev must be of type 'root'. 38099653d4eSeschrock */ 38199653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 38299653d4eSeschrock return (EINVAL); 38399653d4eSeschrock 3848654d025Sperrin /* 3858654d025Sperrin * Determine whether we're a log vdev. 3868654d025Sperrin */ 3878654d025Sperrin islog = 0; 3888654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 389990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 3908654d025Sperrin return (ENOTSUP); 391fa9e4066Sahrens 39299653d4eSeschrock /* 3938654d025Sperrin * Set the nparity property for RAID-Z vdevs. 39499653d4eSeschrock */ 3958654d025Sperrin nparity = -1ULL; 39699653d4eSeschrock if (ops == &vdev_raidz_ops) { 39799653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 3988654d025Sperrin &nparity) == 0) { 39999653d4eSeschrock /* 40099653d4eSeschrock * Currently, we can only support 2 parity devices. 40199653d4eSeschrock */ 4028654d025Sperrin if (nparity == 0 || nparity > 2) 40399653d4eSeschrock return (EINVAL); 40499653d4eSeschrock /* 40599653d4eSeschrock * Older versions can only support 1 parity device. 40699653d4eSeschrock */ 4078654d025Sperrin if (nparity == 2 && 408e7437265Sahrens spa_version(spa) < SPA_VERSION_RAID6) 40999653d4eSeschrock return (ENOTSUP); 41099653d4eSeschrock } else { 41199653d4eSeschrock /* 41299653d4eSeschrock * We require the parity to be specified for SPAs that 41399653d4eSeschrock * support multiple parity levels. 41499653d4eSeschrock */ 415e7437265Sahrens if (spa_version(spa) >= SPA_VERSION_RAID6) 41699653d4eSeschrock return (EINVAL); 41799653d4eSeschrock /* 41899653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 41999653d4eSeschrock */ 4208654d025Sperrin nparity = 1; 42199653d4eSeschrock } 42299653d4eSeschrock } else { 4238654d025Sperrin nparity = 0; 42499653d4eSeschrock } 4258654d025Sperrin ASSERT(nparity != -1ULL); 4268654d025Sperrin 4278654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4288654d025Sperrin 4298654d025Sperrin vd->vdev_islog = islog; 4308654d025Sperrin vd->vdev_nparity = nparity; 4318654d025Sperrin 4328654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4338654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4348654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4358654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4368654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4378654d025Sperrin &vd->vdev_physpath) == 0) 4388654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 4396809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 4406809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 44199653d4eSeschrock 442afefbcddSeschrock /* 443afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 444afefbcddSeschrock * as -1. 445afefbcddSeschrock */ 446afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 447afefbcddSeschrock &vd->vdev_wholedisk) != 0) 448afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 449afefbcddSeschrock 450ea8dc4b6Seschrock /* 451ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 452ea8dc4b6Seschrock * was not present at the time of import. 453ea8dc4b6Seschrock */ 4546809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 4556809eb4eSEric Schrock &vd->vdev_not_present); 456ea8dc4b6Seschrock 457ecc2d604Sbonwick /* 458ecc2d604Sbonwick * Get the alignment requirement. 459ecc2d604Sbonwick */ 460ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 461ecc2d604Sbonwick 462fa9e4066Sahrens /* 463fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 464fa9e4066Sahrens */ 465fa9e4066Sahrens if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) { 466fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 467fa9e4066Sahrens &vd->vdev_ms_array); 468fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 469fa9e4066Sahrens &vd->vdev_ms_shift); 470fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 471fa9e4066Sahrens &vd->vdev_asize); 472fa9e4066Sahrens } 473fa9e4066Sahrens 474fa9e4066Sahrens /* 4753d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 476fa9e4066Sahrens */ 477c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 478c5904d13Seschrock (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE)) { 479c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 480c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 4818ad4d6ddSJeff Bonwick &vd->vdev_dtl_smo.smo_object); 482c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 483c5904d13Seschrock &vd->vdev_unspare); 484c5904d13Seschrock } 485ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 486ecc2d604Sbonwick &vd->vdev_offline); 487c5904d13Seschrock 4883d7072f8Seschrock /* 4893d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 4903d7072f8Seschrock * state, as the diagnosis made on another system may not be 4913d7072f8Seschrock * valid in the current context. 4923d7072f8Seschrock */ 4933d7072f8Seschrock if (spa->spa_load_state == SPA_LOAD_OPEN) { 4943d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 4953d7072f8Seschrock &vd->vdev_faulted); 4963d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 4973d7072f8Seschrock &vd->vdev_degraded); 4983d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 4993d7072f8Seschrock &vd->vdev_removed); 5003d7072f8Seschrock } 501fa9e4066Sahrens } 502fa9e4066Sahrens 503fa9e4066Sahrens /* 504fa9e4066Sahrens * Add ourselves to the parent's list of children. 505fa9e4066Sahrens */ 506fa9e4066Sahrens vdev_add_child(parent, vd); 507fa9e4066Sahrens 50899653d4eSeschrock *vdp = vd; 50999653d4eSeschrock 51099653d4eSeschrock return (0); 511fa9e4066Sahrens } 512fa9e4066Sahrens 513fa9e4066Sahrens void 514fa9e4066Sahrens vdev_free(vdev_t *vd) 515fa9e4066Sahrens { 516fa9e4066Sahrens int c; 5173d7072f8Seschrock spa_t *spa = vd->vdev_spa; 518fa9e4066Sahrens 519fa9e4066Sahrens /* 520fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 521fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 522fa9e4066Sahrens */ 523fa9e4066Sahrens vdev_close(vd); 524fa9e4066Sahrens 525e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 526fa9e4066Sahrens 527fa9e4066Sahrens /* 528fa9e4066Sahrens * Free all children. 529fa9e4066Sahrens */ 530fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 531fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 532fa9e4066Sahrens 533fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 534fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 535fa9e4066Sahrens 536fa9e4066Sahrens /* 537fa9e4066Sahrens * Discard allocation state. 538fa9e4066Sahrens */ 539fa9e4066Sahrens if (vd == vd->vdev_top) 540fa9e4066Sahrens vdev_metaslab_fini(vd); 541fa9e4066Sahrens 542fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 54399653d4eSeschrock ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 544fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 545fa9e4066Sahrens 546fa9e4066Sahrens /* 547fa9e4066Sahrens * Remove this vdev from its parent's child list. 548fa9e4066Sahrens */ 549fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 550fa9e4066Sahrens 551fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 552fa9e4066Sahrens 5533d7072f8Seschrock /* 5543d7072f8Seschrock * Clean up vdev structure. 5553d7072f8Seschrock */ 5563d7072f8Seschrock vdev_queue_fini(vd); 5573d7072f8Seschrock vdev_cache_fini(vd); 5583d7072f8Seschrock 5593d7072f8Seschrock if (vd->vdev_path) 5603d7072f8Seschrock spa_strfree(vd->vdev_path); 5613d7072f8Seschrock if (vd->vdev_devid) 5623d7072f8Seschrock spa_strfree(vd->vdev_devid); 5633d7072f8Seschrock if (vd->vdev_physpath) 5643d7072f8Seschrock spa_strfree(vd->vdev_physpath); 5656809eb4eSEric Schrock if (vd->vdev_fru) 5666809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 5673d7072f8Seschrock 5683d7072f8Seschrock if (vd->vdev_isspare) 5693d7072f8Seschrock spa_spare_remove(vd); 570fa94a07fSbrendan if (vd->vdev_isl2cache) 571fa94a07fSbrendan spa_l2cache_remove(vd); 5723d7072f8Seschrock 5733d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 5743d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 5758ad4d6ddSJeff Bonwick 5763d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 5778ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 5788ad4d6ddSJeff Bonwick space_map_unload(&vd->vdev_dtl[t]); 5798ad4d6ddSJeff Bonwick space_map_destroy(&vd->vdev_dtl[t]); 5808ad4d6ddSJeff Bonwick } 5813d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 5828ad4d6ddSJeff Bonwick 5833d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 5843d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 585e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 5863d7072f8Seschrock 5873d7072f8Seschrock if (vd == spa->spa_root_vdev) 5883d7072f8Seschrock spa->spa_root_vdev = NULL; 5893d7072f8Seschrock 5903d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 591fa9e4066Sahrens } 592fa9e4066Sahrens 593fa9e4066Sahrens /* 594fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 595fa9e4066Sahrens */ 596fa9e4066Sahrens static void 597fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 598fa9e4066Sahrens { 599fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 600fa9e4066Sahrens metaslab_t *msp; 601fa9e4066Sahrens vdev_t *vd; 602fa9e4066Sahrens int t; 603fa9e4066Sahrens 604fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 605fa9e4066Sahrens 606fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 607fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 608fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 609fa9e4066Sahrens 610fa9e4066Sahrens svd->vdev_ms_array = 0; 611fa9e4066Sahrens svd->vdev_ms_shift = 0; 612fa9e4066Sahrens svd->vdev_ms_count = 0; 613fa9e4066Sahrens 614fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 615fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 616fa9e4066Sahrens 617fa9e4066Sahrens svd->vdev_mg = NULL; 618fa9e4066Sahrens svd->vdev_ms = NULL; 619ecc2d604Sbonwick 620ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 621ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 622fa9e4066Sahrens 623fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 624fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 62599653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 626fa9e4066Sahrens 627fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 628fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 62999653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 630fa9e4066Sahrens 631fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 632fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 633fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 634fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 635fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 636fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 637fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 638fa9e4066Sahrens } 639fa9e4066Sahrens 640e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 641fa9e4066Sahrens vdev_config_clean(svd); 642fa9e4066Sahrens vdev_config_dirty(tvd); 643fa9e4066Sahrens } 644fa9e4066Sahrens 645e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 646e14bb325SJeff Bonwick vdev_state_clean(svd); 647e14bb325SJeff Bonwick vdev_state_dirty(tvd); 648e14bb325SJeff Bonwick } 649e14bb325SJeff Bonwick 65099653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 65199653d4eSeschrock svd->vdev_deflate_ratio = 0; 6528654d025Sperrin 6538654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 6548654d025Sperrin svd->vdev_islog = 0; 655fa9e4066Sahrens } 656fa9e4066Sahrens 657fa9e4066Sahrens static void 658fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 659fa9e4066Sahrens { 660fa9e4066Sahrens int c; 661fa9e4066Sahrens 662fa9e4066Sahrens if (vd == NULL) 663fa9e4066Sahrens return; 664fa9e4066Sahrens 665fa9e4066Sahrens vd->vdev_top = tvd; 666fa9e4066Sahrens 667fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 668fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 669fa9e4066Sahrens } 670fa9e4066Sahrens 671fa9e4066Sahrens /* 672fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 673fa9e4066Sahrens */ 674fa9e4066Sahrens vdev_t * 675fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 676fa9e4066Sahrens { 677fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 678fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 679fa9e4066Sahrens vdev_t *mvd; 680fa9e4066Sahrens 681e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 682fa9e4066Sahrens 683fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 684ecc2d604Sbonwick 685ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 686ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 687ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 688ecc2d604Sbonwick 689fa9e4066Sahrens vdev_remove_child(pvd, cvd); 690fa9e4066Sahrens vdev_add_child(pvd, mvd); 691fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 692fa9e4066Sahrens vdev_add_child(mvd, cvd); 693fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 694fa9e4066Sahrens 695fa9e4066Sahrens if (mvd == mvd->vdev_top) 696fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 697fa9e4066Sahrens 698fa9e4066Sahrens return (mvd); 699fa9e4066Sahrens } 700fa9e4066Sahrens 701fa9e4066Sahrens /* 702fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 703fa9e4066Sahrens */ 704fa9e4066Sahrens void 705fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 706fa9e4066Sahrens { 707fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 708fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 709fa9e4066Sahrens 710e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 711fa9e4066Sahrens 712fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 713fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 71499653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 71599653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 716ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 717fa9e4066Sahrens 718fa9e4066Sahrens vdev_remove_child(mvd, cvd); 719fa9e4066Sahrens vdev_remove_child(pvd, mvd); 7208ad4d6ddSJeff Bonwick 72199653d4eSeschrock /* 722e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 723e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 724e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 725e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 72699653d4eSeschrock */ 7278ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 7288ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 7298ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 7308ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 7318ad4d6ddSJeff Bonwick } 732e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 733e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 734fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 735fa9e4066Sahrens 736fa9e4066Sahrens if (cvd == cvd->vdev_top) 737fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 738fa9e4066Sahrens 739fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 740fa9e4066Sahrens vdev_free(mvd); 741fa9e4066Sahrens } 742fa9e4066Sahrens 743ea8dc4b6Seschrock int 744fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 745fa9e4066Sahrens { 746fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 747ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 7488654d025Sperrin metaslab_class_t *mc; 749ecc2d604Sbonwick uint64_t m; 750fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 751fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 752ecc2d604Sbonwick metaslab_t **mspp; 753ecc2d604Sbonwick int error; 754fa9e4066Sahrens 7550e34b6a7Sbonwick if (vd->vdev_ms_shift == 0) /* not being allocated from yet */ 7560e34b6a7Sbonwick return (0); 7570e34b6a7Sbonwick 758*e6ca193dSGeorge Wilson /* 759*e6ca193dSGeorge Wilson * Compute the raidz-deflation ratio. Note, we hard-code 760*e6ca193dSGeorge Wilson * in 128k (1 << 17) because it is the current "typical" blocksize. 761*e6ca193dSGeorge Wilson * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 762*e6ca193dSGeorge Wilson * or we will inconsistently account for existing bp's. 763*e6ca193dSGeorge Wilson */ 764*e6ca193dSGeorge Wilson vd->vdev_deflate_ratio = (1 << 17) / 765*e6ca193dSGeorge Wilson (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 766*e6ca193dSGeorge Wilson 767fa9e4066Sahrens ASSERT(oldc <= newc); 768fa9e4066Sahrens 7698654d025Sperrin if (vd->vdev_islog) 7708654d025Sperrin mc = spa->spa_log_class; 7718654d025Sperrin else 7728654d025Sperrin mc = spa->spa_normal_class; 7738654d025Sperrin 774ecc2d604Sbonwick if (vd->vdev_mg == NULL) 775ecc2d604Sbonwick vd->vdev_mg = metaslab_group_create(mc, vd); 776fa9e4066Sahrens 777ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 778fa9e4066Sahrens 779ecc2d604Sbonwick if (oldc != 0) { 780ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 781ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 782ecc2d604Sbonwick } 783fa9e4066Sahrens 784ecc2d604Sbonwick vd->vdev_ms = mspp; 785ecc2d604Sbonwick vd->vdev_ms_count = newc; 786fa9e4066Sahrens 787ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 788ecc2d604Sbonwick space_map_obj_t smo = { 0, 0, 0 }; 789ecc2d604Sbonwick if (txg == 0) { 790ecc2d604Sbonwick uint64_t object = 0; 791ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 7927bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 7937bfdf011SNeil Perrin DMU_READ_PREFETCH); 794ecc2d604Sbonwick if (error) 795ecc2d604Sbonwick return (error); 796ecc2d604Sbonwick if (object != 0) { 797ecc2d604Sbonwick dmu_buf_t *db; 798ecc2d604Sbonwick error = dmu_bonus_hold(mos, object, FTAG, &db); 799ecc2d604Sbonwick if (error) 800ecc2d604Sbonwick return (error); 8011934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (smo)); 8021934e92fSmaybee bcopy(db->db_data, &smo, sizeof (smo)); 803ecc2d604Sbonwick ASSERT3U(smo.smo_object, ==, object); 804ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 805fa9e4066Sahrens } 806fa9e4066Sahrens } 807ecc2d604Sbonwick vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 808ecc2d604Sbonwick m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 809fa9e4066Sahrens } 810fa9e4066Sahrens 811ea8dc4b6Seschrock return (0); 812fa9e4066Sahrens } 813fa9e4066Sahrens 814fa9e4066Sahrens void 815fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 816fa9e4066Sahrens { 817fa9e4066Sahrens uint64_t m; 818fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 819fa9e4066Sahrens 820fa9e4066Sahrens if (vd->vdev_ms != NULL) { 821fa9e4066Sahrens for (m = 0; m < count; m++) 822ecc2d604Sbonwick if (vd->vdev_ms[m] != NULL) 823ecc2d604Sbonwick metaslab_fini(vd->vdev_ms[m]); 824fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 825fa9e4066Sahrens vd->vdev_ms = NULL; 826fa9e4066Sahrens } 827fa9e4066Sahrens } 828fa9e4066Sahrens 829e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 830e14bb325SJeff Bonwick boolean_t vps_readable; 831e14bb325SJeff Bonwick boolean_t vps_writeable; 832e14bb325SJeff Bonwick int vps_flags; 833e14bb325SJeff Bonwick } vdev_probe_stats_t; 834e14bb325SJeff Bonwick 835e14bb325SJeff Bonwick static void 836e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 8370a4e9518Sgw { 8388ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 839a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 840e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 841a3f829aeSBill Moore 842a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 843e14bb325SJeff Bonwick 844e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 845e14bb325SJeff Bonwick if (zio->io_error == 0) 846e14bb325SJeff Bonwick vps->vps_readable = 1; 8478ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 848a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 849e14bb325SJeff Bonwick zio->io_offset, zio->io_size, zio->io_data, 850e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 851e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 852e14bb325SJeff Bonwick } else { 853e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 854e14bb325SJeff Bonwick } 855e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 856e14bb325SJeff Bonwick if (zio->io_error == 0) 857e14bb325SJeff Bonwick vps->vps_writeable = 1; 858e14bb325SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 859e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 860a3f829aeSBill Moore zio_t *pio; 861e14bb325SJeff Bonwick 862e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 863e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 864e14bb325SJeff Bonwick 865e14bb325SJeff Bonwick if (vdev_readable(vd) && 8668ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 867e14bb325SJeff Bonwick zio->io_error = 0; 868e14bb325SJeff Bonwick } else { 869e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 870e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 8718ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 872e14bb325SJeff Bonwick zio->io_error = ENXIO; 873e14bb325SJeff Bonwick } 874a3f829aeSBill Moore 875a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 876a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 877a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 878a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 879a3f829aeSBill Moore 880a3f829aeSBill Moore while ((pio = zio_walk_parents(zio)) != NULL) 881a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 882a3f829aeSBill Moore pio->io_error = ENXIO; 883a3f829aeSBill Moore 884e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 885e14bb325SJeff Bonwick } 886e14bb325SJeff Bonwick } 8870a4e9518Sgw 888e14bb325SJeff Bonwick /* 889e14bb325SJeff Bonwick * Determine whether this device is accessible by reading and writing 890e14bb325SJeff Bonwick * to several known locations: the pad regions of each vdev label 891e14bb325SJeff Bonwick * but the first (which we leave alone in case it contains a VTOC). 892e14bb325SJeff Bonwick */ 893e14bb325SJeff Bonwick zio_t * 894a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 895e14bb325SJeff Bonwick { 896e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 897a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 898a3f829aeSBill Moore zio_t *pio; 899a3f829aeSBill Moore 900a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 9010a4e9518Sgw 902a3f829aeSBill Moore /* 903a3f829aeSBill Moore * Don't probe the probe. 904a3f829aeSBill Moore */ 905a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 906a3f829aeSBill Moore return (NULL); 907e14bb325SJeff Bonwick 908a3f829aeSBill Moore /* 909a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 910a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 911a3f829aeSBill Moore * this vdev will become parents of the probe io. 912a3f829aeSBill Moore */ 913a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 914e14bb325SJeff Bonwick 915a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 916a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 917a3f829aeSBill Moore 918a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 919a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 920a3f829aeSBill Moore ZIO_FLAG_DONT_RETRY; 921a3f829aeSBill Moore 922a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 923a3f829aeSBill Moore /* 924a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 925a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 926a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 927a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 928a3f829aeSBill Moore * any zio looking at these values can assume that 929a3f829aeSBill Moore * failures persist for the life of the I/O. That's 930a3f829aeSBill Moore * important because when a device has intermittent 931a3f829aeSBill Moore * connectivity problems, we want to ensure that 932a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 933a3f829aeSBill Moore * the zio (EIO). 934a3f829aeSBill Moore * 935a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 936a3f829aeSBill Moore * values so the probe can reevaluate from first 937a3f829aeSBill Moore * principles. 938a3f829aeSBill Moore */ 939a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 940a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 941a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 942a3f829aeSBill Moore } 943a3f829aeSBill Moore 944a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 945a3f829aeSBill Moore vdev_probe_done, vps, 946a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 947a3f829aeSBill Moore 948a3f829aeSBill Moore if (zio != NULL) { 949a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 950a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 951a3f829aeSBill Moore } 952e14bb325SJeff Bonwick } 953e14bb325SJeff Bonwick 954a3f829aeSBill Moore if (zio != NULL) 955a3f829aeSBill Moore zio_add_child(zio, pio); 956e14bb325SJeff Bonwick 957a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 958e14bb325SJeff Bonwick 959a3f829aeSBill Moore if (vps == NULL) { 960a3f829aeSBill Moore ASSERT(zio != NULL); 961a3f829aeSBill Moore return (NULL); 962a3f829aeSBill Moore } 963e14bb325SJeff Bonwick 964e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 965a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 966e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 967f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2)), 968f83ffe1aSLin Ling VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 969e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 970e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 971e14bb325SJeff Bonwick } 972e14bb325SJeff Bonwick 973a3f829aeSBill Moore if (zio == NULL) 974a3f829aeSBill Moore return (pio); 975a3f829aeSBill Moore 976a3f829aeSBill Moore zio_nowait(pio); 977a3f829aeSBill Moore return (NULL); 9780a4e9518Sgw } 9790a4e9518Sgw 980fa9e4066Sahrens /* 981fa9e4066Sahrens * Prepare a virtual device for access. 982fa9e4066Sahrens */ 983fa9e4066Sahrens int 984fa9e4066Sahrens vdev_open(vdev_t *vd) 985fa9e4066Sahrens { 9868ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 987fa9e4066Sahrens int error; 988fa9e4066Sahrens int c; 989fa9e4066Sahrens uint64_t osize = 0; 990fa9e4066Sahrens uint64_t asize, psize; 991ecc2d604Sbonwick uint64_t ashift = 0; 992fa9e4066Sahrens 9938ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 9948ad4d6ddSJeff Bonwick 995fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 996fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 997fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 998fa9e4066Sahrens 999fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1000*e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1001*e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1002fa9e4066Sahrens 10033d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 10043d7072f8Seschrock ASSERT(vd->vdev_children == 0); 10053d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 10063d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 10073d7072f8Seschrock return (ENXIO); 10083d7072f8Seschrock } else if (vd->vdev_offline) { 1009fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1010ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1011fa9e4066Sahrens return (ENXIO); 1012fa9e4066Sahrens } 1013fa9e4066Sahrens 1014fa9e4066Sahrens error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 1015fa9e4066Sahrens 1016ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 1017ea8dc4b6Seschrock error = zio_handle_device_injection(vd, ENXIO); 1018ea8dc4b6Seschrock 1019fa9e4066Sahrens if (error) { 10203d7072f8Seschrock if (vd->vdev_removed && 10213d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 10223d7072f8Seschrock vd->vdev_removed = B_FALSE; 10233d7072f8Seschrock 1024ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1025fa9e4066Sahrens vd->vdev_stat.vs_aux); 1026fa9e4066Sahrens return (error); 1027fa9e4066Sahrens } 1028fa9e4066Sahrens 10293d7072f8Seschrock vd->vdev_removed = B_FALSE; 10303d7072f8Seschrock 10313d7072f8Seschrock if (vd->vdev_degraded) { 10323d7072f8Seschrock ASSERT(vd->vdev_children == 0); 10333d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 10343d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 10353d7072f8Seschrock } else { 10363d7072f8Seschrock vd->vdev_state = VDEV_STATE_HEALTHY; 10373d7072f8Seschrock } 1038fa9e4066Sahrens 1039fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 1040ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1041ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1042ea8dc4b6Seschrock VDEV_AUX_NONE); 1043ea8dc4b6Seschrock break; 1044ea8dc4b6Seschrock } 1045fa9e4066Sahrens 1046fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1047fa9e4066Sahrens 1048fa9e4066Sahrens if (vd->vdev_children == 0) { 1049fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1050ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1051ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1052fa9e4066Sahrens return (EOVERFLOW); 1053fa9e4066Sahrens } 1054fa9e4066Sahrens psize = osize; 1055fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1056fa9e4066Sahrens } else { 1057ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1058fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1059ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1060ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1061fa9e4066Sahrens return (EOVERFLOW); 1062fa9e4066Sahrens } 1063fa9e4066Sahrens psize = 0; 1064fa9e4066Sahrens asize = osize; 1065fa9e4066Sahrens } 1066fa9e4066Sahrens 1067fa9e4066Sahrens vd->vdev_psize = psize; 1068fa9e4066Sahrens 1069fa9e4066Sahrens if (vd->vdev_asize == 0) { 1070fa9e4066Sahrens /* 1071fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1072ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1073fa9e4066Sahrens */ 1074fa9e4066Sahrens vd->vdev_asize = asize; 1075ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1076fa9e4066Sahrens } else { 1077fa9e4066Sahrens /* 1078fa9e4066Sahrens * Make sure the alignment requirement hasn't increased. 1079fa9e4066Sahrens */ 1080ecc2d604Sbonwick if (ashift > vd->vdev_top->vdev_ashift) { 1081ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1082ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 1083fa9e4066Sahrens return (EINVAL); 1084fa9e4066Sahrens } 1085fa9e4066Sahrens 1086fa9e4066Sahrens /* 1087fa9e4066Sahrens * Make sure the device hasn't shrunk. 1088fa9e4066Sahrens */ 1089fa9e4066Sahrens if (asize < vd->vdev_asize) { 1090ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1091ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 1092fa9e4066Sahrens return (EINVAL); 1093fa9e4066Sahrens } 1094fa9e4066Sahrens 1095fa9e4066Sahrens /* 1096fa9e4066Sahrens * If all children are healthy and the asize has increased, 1097fa9e4066Sahrens * then we've experienced dynamic LUN growth. 1098fa9e4066Sahrens */ 1099fa9e4066Sahrens if (vd->vdev_state == VDEV_STATE_HEALTHY && 1100fa9e4066Sahrens asize > vd->vdev_asize) { 1101fa9e4066Sahrens vd->vdev_asize = asize; 1102fa9e4066Sahrens } 1103fa9e4066Sahrens } 1104fa9e4066Sahrens 11050a4e9518Sgw /* 11060a4e9518Sgw * Ensure we can issue some IO before declaring the 11070a4e9518Sgw * vdev open for business. 11080a4e9518Sgw */ 1109e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1110e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 11110a4e9518Sgw vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1112e14bb325SJeff Bonwick VDEV_AUX_IO_FAILURE); 11130a4e9518Sgw return (error); 11140a4e9518Sgw } 11150a4e9518Sgw 1116088f3894Sahrens /* 1117088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 11188ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 11198ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1120088f3894Sahrens */ 11218ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 11228ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 11238ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1124088f3894Sahrens 1125fa9e4066Sahrens return (0); 1126fa9e4066Sahrens } 1127fa9e4066Sahrens 1128560e6e96Seschrock /* 1129560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1130560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 11313d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1132560e6e96Seschrock * 1133560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1134560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1135560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1136560e6e96Seschrock * will be updated but the function will return 0. 1137560e6e96Seschrock */ 1138560e6e96Seschrock int 1139560e6e96Seschrock vdev_validate(vdev_t *vd) 1140560e6e96Seschrock { 1141560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1142560e6e96Seschrock int c; 1143560e6e96Seschrock nvlist_t *label; 1144e14bb325SJeff Bonwick uint64_t guid, top_guid; 1145560e6e96Seschrock uint64_t state; 1146560e6e96Seschrock 1147560e6e96Seschrock for (c = 0; c < vd->vdev_children; c++) 1148560e6e96Seschrock if (vdev_validate(vd->vdev_child[c]) != 0) 11490bf246f5Smc return (EBADF); 1150560e6e96Seschrock 1151b5989ec7Seschrock /* 1152b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1153b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1154b5989ec7Seschrock * overwrite the previous state. 1155b5989ec7Seschrock */ 1156e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1157560e6e96Seschrock 1158560e6e96Seschrock if ((label = vdev_label_read_config(vd)) == NULL) { 1159560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1160560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1161560e6e96Seschrock return (0); 1162560e6e96Seschrock } 1163560e6e96Seschrock 1164560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1165560e6e96Seschrock &guid) != 0 || guid != spa_guid(spa)) { 1166560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1167560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1168560e6e96Seschrock nvlist_free(label); 1169560e6e96Seschrock return (0); 1170560e6e96Seschrock } 1171560e6e96Seschrock 1172e14bb325SJeff Bonwick /* 1173e14bb325SJeff Bonwick * If this vdev just became a top-level vdev because its 1174e14bb325SJeff Bonwick * sibling was detached, it will have adopted the parent's 1175e14bb325SJeff Bonwick * vdev guid -- but the label may or may not be on disk yet. 1176e14bb325SJeff Bonwick * Fortunately, either version of the label will have the 1177e14bb325SJeff Bonwick * same top guid, so if we're a top-level vdev, we can 1178e14bb325SJeff Bonwick * safely compare to that instead. 1179e14bb325SJeff Bonwick */ 1180560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1181e14bb325SJeff Bonwick &guid) != 0 || 1182e14bb325SJeff Bonwick nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1183e14bb325SJeff Bonwick &top_guid) != 0 || 1184e14bb325SJeff Bonwick (vd->vdev_guid != guid && 1185e14bb325SJeff Bonwick (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1186560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1187560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1188560e6e96Seschrock nvlist_free(label); 1189560e6e96Seschrock return (0); 1190560e6e96Seschrock } 1191560e6e96Seschrock 1192560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1193560e6e96Seschrock &state) != 0) { 1194560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1195560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1196560e6e96Seschrock nvlist_free(label); 1197560e6e96Seschrock return (0); 1198560e6e96Seschrock } 1199560e6e96Seschrock 1200560e6e96Seschrock nvlist_free(label); 1201560e6e96Seschrock 1202560e6e96Seschrock if (spa->spa_load_state == SPA_LOAD_OPEN && 1203560e6e96Seschrock state != POOL_STATE_ACTIVE) 12040bf246f5Smc return (EBADF); 1205560e6e96Seschrock 120651ece835Seschrock /* 120751ece835Seschrock * If we were able to open and validate a vdev that was 120851ece835Seschrock * previously marked permanently unavailable, clear that state 120951ece835Seschrock * now. 121051ece835Seschrock */ 121151ece835Seschrock if (vd->vdev_not_present) 121251ece835Seschrock vd->vdev_not_present = 0; 121351ece835Seschrock } 1214560e6e96Seschrock 1215560e6e96Seschrock return (0); 1216560e6e96Seschrock } 1217560e6e96Seschrock 1218fa9e4066Sahrens /* 1219fa9e4066Sahrens * Close a virtual device. 1220fa9e4066Sahrens */ 1221fa9e4066Sahrens void 1222fa9e4066Sahrens vdev_close(vdev_t *vd) 1223fa9e4066Sahrens { 12248ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 12258ad4d6ddSJeff Bonwick 12268ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 12278ad4d6ddSJeff Bonwick 1228fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1229fa9e4066Sahrens 12303d7072f8Seschrock vdev_cache_purge(vd); 1231fa9e4066Sahrens 1232560e6e96Seschrock /* 1233560e6e96Seschrock * We record the previous state before we close it, so that if we are 1234560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1235560e6e96Seschrock * it's still faulted. 1236560e6e96Seschrock */ 1237560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1238560e6e96Seschrock 1239fa9e4066Sahrens if (vd->vdev_offline) 1240fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1241fa9e4066Sahrens else 1242fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1243ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1244fa9e4066Sahrens } 1245fa9e4066Sahrens 1246fa9e4066Sahrens void 1247ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1248fa9e4066Sahrens { 1249ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1250fa9e4066Sahrens 1251e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1252ea8dc4b6Seschrock 1253fa9e4066Sahrens vdev_close(vd); 1254fa9e4066Sahrens (void) vdev_open(vd); 1255fa9e4066Sahrens 125639c23413Seschrock /* 125739c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 125839c23413Seschrock * Otherwise, a device with an invalid label could be successfully 125939c23413Seschrock * opened in response to vdev_reopen(). 126039c23413Seschrock */ 1261c5904d13Seschrock if (vd->vdev_aux) { 1262c5904d13Seschrock (void) vdev_validate_aux(vd); 1263e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 12646809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1265c5904d13Seschrock !l2arc_vdev_present(vd)) { 1266c5904d13Seschrock uint64_t size = vdev_get_rsize(vd); 1267c5904d13Seschrock l2arc_add_vdev(spa, vd, 1268c5904d13Seschrock VDEV_LABEL_START_SIZE, 1269c5904d13Seschrock size - VDEV_LABEL_START_SIZE); 1270c5904d13Seschrock } 1271c5904d13Seschrock } else { 1272c5904d13Seschrock (void) vdev_validate(vd); 1273c5904d13Seschrock } 127439c23413Seschrock 1275fa9e4066Sahrens /* 12763d7072f8Seschrock * Reassess parent vdev's health. 1277fa9e4066Sahrens */ 12783d7072f8Seschrock vdev_propagate_state(vd); 1279fa9e4066Sahrens } 1280fa9e4066Sahrens 1281fa9e4066Sahrens int 128299653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1283fa9e4066Sahrens { 1284fa9e4066Sahrens int error; 1285fa9e4066Sahrens 1286fa9e4066Sahrens /* 1287fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1288fa9e4066Sahrens * For a create, however, we want to fail the request if 1289fa9e4066Sahrens * there are any components we can't open. 1290fa9e4066Sahrens */ 1291fa9e4066Sahrens error = vdev_open(vd); 1292fa9e4066Sahrens 1293fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1294fa9e4066Sahrens vdev_close(vd); 1295fa9e4066Sahrens return (error ? error : ENXIO); 1296fa9e4066Sahrens } 1297fa9e4066Sahrens 1298fa9e4066Sahrens /* 1299fa9e4066Sahrens * Recursively initialize all labels. 1300fa9e4066Sahrens */ 130139c23413Seschrock if ((error = vdev_label_init(vd, txg, isreplacing ? 130239c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1303fa9e4066Sahrens vdev_close(vd); 1304fa9e4066Sahrens return (error); 1305fa9e4066Sahrens } 1306fa9e4066Sahrens 1307fa9e4066Sahrens return (0); 1308fa9e4066Sahrens } 1309fa9e4066Sahrens 1310fa9e4066Sahrens /* 1311fa9e4066Sahrens * The is the latter half of vdev_create(). It is distinct because it 1312fa9e4066Sahrens * involves initiating transactions in order to do metaslab creation. 1313fa9e4066Sahrens * For creation, we want to try to create all vdevs at once and then undo it 1314fa9e4066Sahrens * if anything fails; this is much harder if we have pending transactions. 1315fa9e4066Sahrens */ 13160e34b6a7Sbonwick void 1317fa9e4066Sahrens vdev_init(vdev_t *vd, uint64_t txg) 1318fa9e4066Sahrens { 1319fa9e4066Sahrens /* 1320fa9e4066Sahrens * Aim for roughly 200 metaslabs per vdev. 1321fa9e4066Sahrens */ 1322fa9e4066Sahrens vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1323fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1324fa9e4066Sahrens 1325fa9e4066Sahrens /* 13260e34b6a7Sbonwick * Initialize the vdev's metaslabs. This can't fail because 13270e34b6a7Sbonwick * there's nothing to read when creating all new metaslabs. 1328fa9e4066Sahrens */ 13290e34b6a7Sbonwick VERIFY(vdev_metaslab_init(vd, txg) == 0); 1330fa9e4066Sahrens } 1331fa9e4066Sahrens 1332fa9e4066Sahrens void 1333ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1334fa9e4066Sahrens { 1335ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 1336ecc2d604Sbonwick ASSERT(ISP2(flags)); 1337fa9e4066Sahrens 1338ecc2d604Sbonwick if (flags & VDD_METASLAB) 1339ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1340ecc2d604Sbonwick 1341ecc2d604Sbonwick if (flags & VDD_DTL) 1342ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1343ecc2d604Sbonwick 1344ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1345fa9e4066Sahrens } 1346fa9e4066Sahrens 13478ad4d6ddSJeff Bonwick /* 13488ad4d6ddSJeff Bonwick * DTLs. 13498ad4d6ddSJeff Bonwick * 13508ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 13518ad4d6ddSJeff Bonwick * the vdev has less than perfect replication. There are three kinds of DTL: 13528ad4d6ddSJeff Bonwick * 13538ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 13548ad4d6ddSJeff Bonwick * 13558ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 13568ad4d6ddSJeff Bonwick * 13578ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 13588ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 13598ad4d6ddSJeff Bonwick * txgs that was scrubbed. 13608ad4d6ddSJeff Bonwick * 13618ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 13628ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 13638ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 13648ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 13658ad4d6ddSJeff Bonwick * determine whether a device can be detached. 13668ad4d6ddSJeff Bonwick * 13678ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 13688ad4d6ddSJeff Bonwick * either has the data or it doesn't. 13698ad4d6ddSJeff Bonwick * 13708ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 13718ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 13728ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 13738ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 13748ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 13758ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 13768ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 13778ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 13788ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 13798ad4d6ddSJeff Bonwick * 13808ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 13818ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 13828ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 13838ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 13848ad4d6ddSJeff Bonwick */ 1385fa9e4066Sahrens void 13868ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1387fa9e4066Sahrens { 13888ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 13898ad4d6ddSJeff Bonwick 13908ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 13918ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 13928ad4d6ddSJeff Bonwick 1393fa9e4066Sahrens mutex_enter(sm->sm_lock); 1394fa9e4066Sahrens if (!space_map_contains(sm, txg, size)) 1395fa9e4066Sahrens space_map_add(sm, txg, size); 1396fa9e4066Sahrens mutex_exit(sm->sm_lock); 1397fa9e4066Sahrens } 1398fa9e4066Sahrens 13998ad4d6ddSJeff Bonwick boolean_t 14008ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1401fa9e4066Sahrens { 14028ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 14038ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 1404fa9e4066Sahrens 14058ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 14068ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1407fa9e4066Sahrens 1408fa9e4066Sahrens mutex_enter(sm->sm_lock); 14098ad4d6ddSJeff Bonwick if (sm->sm_space != 0) 14108ad4d6ddSJeff Bonwick dirty = space_map_contains(sm, txg, size); 1411fa9e4066Sahrens mutex_exit(sm->sm_lock); 1412fa9e4066Sahrens 1413fa9e4066Sahrens return (dirty); 1414fa9e4066Sahrens } 1415fa9e4066Sahrens 14168ad4d6ddSJeff Bonwick boolean_t 14178ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 14188ad4d6ddSJeff Bonwick { 14198ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[t]; 14208ad4d6ddSJeff Bonwick boolean_t empty; 14218ad4d6ddSJeff Bonwick 14228ad4d6ddSJeff Bonwick mutex_enter(sm->sm_lock); 14238ad4d6ddSJeff Bonwick empty = (sm->sm_space == 0); 14248ad4d6ddSJeff Bonwick mutex_exit(sm->sm_lock); 14258ad4d6ddSJeff Bonwick 14268ad4d6ddSJeff Bonwick return (empty); 14278ad4d6ddSJeff Bonwick } 14288ad4d6ddSJeff Bonwick 1429fa9e4066Sahrens /* 1430fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1431fa9e4066Sahrens */ 1432fa9e4066Sahrens void 1433fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1434fa9e4066Sahrens { 1435ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 14368ad4d6ddSJeff Bonwick avl_tree_t reftree; 14378ad4d6ddSJeff Bonwick int minref; 1438fa9e4066Sahrens 14398ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1440fa9e4066Sahrens 14418ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 14428ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 14438ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 14448ad4d6ddSJeff Bonwick 14458ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev) 14468ad4d6ddSJeff Bonwick return; 14478ad4d6ddSJeff Bonwick 14488ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 1449fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1450088f3894Sahrens if (scrub_txg != 0 && 1451088f3894Sahrens (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { 1452088f3894Sahrens /* XXX should check scrub_done? */ 1453088f3894Sahrens /* 1454088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1455088f3894Sahrens * did it without rebooting, then the scrub dtl 1456088f3894Sahrens * will be valid, so excise the old region and 1457088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1458088f3894Sahrens * dtl as-is if there was an error. 14598ad4d6ddSJeff Bonwick * 14608ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 14618ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 14628ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 14638ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 14648ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 14658ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 14668ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 14678ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 14688ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 1469088f3894Sahrens */ 14708ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 14718ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 14728ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 14738ad4d6ddSJeff Bonwick space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 14748ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, 14758ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_SCRUB], 2); 14768ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, 14778ad4d6ddSJeff Bonwick &vd->vdev_dtl[DTL_MISSING], 1); 14788ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1479fa9e4066Sahrens } 14808ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 14818ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 14828ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1483fa9e4066Sahrens if (scrub_done) 14848ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 14858ad4d6ddSJeff Bonwick space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 14868ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 14878ad4d6ddSJeff Bonwick space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 14888ad4d6ddSJeff Bonwick else 14898ad4d6ddSJeff Bonwick space_map_walk(&vd->vdev_dtl[DTL_MISSING], 14908ad4d6ddSJeff Bonwick space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1491fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1492088f3894Sahrens 1493ecc2d604Sbonwick if (txg != 0) 1494ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1495fa9e4066Sahrens return; 1496fa9e4066Sahrens } 1497fa9e4066Sahrens 1498fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 14998ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 15008ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 15018ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 15028ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 15038ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 15048ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 15058ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 15068ad4d6ddSJeff Bonwick else 15078ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 15088ad4d6ddSJeff Bonwick space_map_ref_create(&reftree); 15098ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 15108ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 15118ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 15128ad4d6ddSJeff Bonwick space_map_ref_add_map(&reftree, &cvd->vdev_dtl[t], 1); 15138ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 15148ad4d6ddSJeff Bonwick } 15158ad4d6ddSJeff Bonwick space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 15168ad4d6ddSJeff Bonwick space_map_ref_destroy(&reftree); 1517fa9e4066Sahrens } 15188ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 1519fa9e4066Sahrens } 1520fa9e4066Sahrens 1521fa9e4066Sahrens static int 1522fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1523fa9e4066Sahrens { 1524fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 15258ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 1526ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1527fa9e4066Sahrens dmu_buf_t *db; 1528fa9e4066Sahrens int error; 1529fa9e4066Sahrens 1530fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1531fa9e4066Sahrens 1532fa9e4066Sahrens if (smo->smo_object == 0) 1533fa9e4066Sahrens return (0); 1534fa9e4066Sahrens 1535ecc2d604Sbonwick if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1536ea8dc4b6Seschrock return (error); 1537ecc2d604Sbonwick 15381934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 15391934e92fSmaybee bcopy(db->db_data, smo, sizeof (*smo)); 1540ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1541fa9e4066Sahrens 1542fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 15438ad4d6ddSJeff Bonwick error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 15448ad4d6ddSJeff Bonwick NULL, SM_ALLOC, smo, mos); 1545fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1546fa9e4066Sahrens 1547fa9e4066Sahrens return (error); 1548fa9e4066Sahrens } 1549fa9e4066Sahrens 1550fa9e4066Sahrens void 1551fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1552fa9e4066Sahrens { 1553fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 15548ad4d6ddSJeff Bonwick space_map_obj_t *smo = &vd->vdev_dtl_smo; 15558ad4d6ddSJeff Bonwick space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1556ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1557fa9e4066Sahrens space_map_t smsync; 1558fa9e4066Sahrens kmutex_t smlock; 1559fa9e4066Sahrens dmu_buf_t *db; 1560fa9e4066Sahrens dmu_tx_t *tx; 1561fa9e4066Sahrens 1562fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1563fa9e4066Sahrens 1564fa9e4066Sahrens if (vd->vdev_detached) { 1565fa9e4066Sahrens if (smo->smo_object != 0) { 1566ecc2d604Sbonwick int err = dmu_object_free(mos, smo->smo_object, tx); 1567fa9e4066Sahrens ASSERT3U(err, ==, 0); 1568fa9e4066Sahrens smo->smo_object = 0; 1569fa9e4066Sahrens } 1570fa9e4066Sahrens dmu_tx_commit(tx); 1571fa9e4066Sahrens return; 1572fa9e4066Sahrens } 1573fa9e4066Sahrens 1574fa9e4066Sahrens if (smo->smo_object == 0) { 1575fa9e4066Sahrens ASSERT(smo->smo_objsize == 0); 1576fa9e4066Sahrens ASSERT(smo->smo_alloc == 0); 1577ecc2d604Sbonwick smo->smo_object = dmu_object_alloc(mos, 1578fa9e4066Sahrens DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1579fa9e4066Sahrens DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1580fa9e4066Sahrens ASSERT(smo->smo_object != 0); 1581fa9e4066Sahrens vdev_config_dirty(vd->vdev_top); 1582fa9e4066Sahrens } 1583fa9e4066Sahrens 1584fa9e4066Sahrens mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1585fa9e4066Sahrens 1586fa9e4066Sahrens space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1587fa9e4066Sahrens &smlock); 1588fa9e4066Sahrens 1589fa9e4066Sahrens mutex_enter(&smlock); 1590fa9e4066Sahrens 1591fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1592ecc2d604Sbonwick space_map_walk(sm, space_map_add, &smsync); 1593fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1594fa9e4066Sahrens 1595ecc2d604Sbonwick space_map_truncate(smo, mos, tx); 1596ecc2d604Sbonwick space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1597fa9e4066Sahrens 1598fa9e4066Sahrens space_map_destroy(&smsync); 1599fa9e4066Sahrens 1600fa9e4066Sahrens mutex_exit(&smlock); 1601fa9e4066Sahrens mutex_destroy(&smlock); 1602fa9e4066Sahrens 1603ecc2d604Sbonwick VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1604fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 16051934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 16061934e92fSmaybee bcopy(smo, db->db_data, sizeof (*smo)); 1607ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1608fa9e4066Sahrens 1609fa9e4066Sahrens dmu_tx_commit(tx); 1610fa9e4066Sahrens } 1611fa9e4066Sahrens 16128ad4d6ddSJeff Bonwick /* 16138ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 16148ad4d6ddSJeff Bonwick * without losing data. 16158ad4d6ddSJeff Bonwick */ 16168ad4d6ddSJeff Bonwick boolean_t 16178ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 16188ad4d6ddSJeff Bonwick { 16198ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 16208ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 16218ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 16228ad4d6ddSJeff Bonwick boolean_t required; 16238ad4d6ddSJeff Bonwick 16248ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 16258ad4d6ddSJeff Bonwick 16268ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 16278ad4d6ddSJeff Bonwick return (B_TRUE); 16288ad4d6ddSJeff Bonwick 16298ad4d6ddSJeff Bonwick /* 16308ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 16318ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 16328ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 16338ad4d6ddSJeff Bonwick */ 16348ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 16358ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 16368ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 16378ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 16388ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 16398ad4d6ddSJeff Bonwick 16408ad4d6ddSJeff Bonwick return (required); 16418ad4d6ddSJeff Bonwick } 16428ad4d6ddSJeff Bonwick 1643088f3894Sahrens /* 1644088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 1645088f3894Sahrens */ 1646088f3894Sahrens boolean_t 1647088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1648088f3894Sahrens { 1649088f3894Sahrens boolean_t needed = B_FALSE; 1650088f3894Sahrens uint64_t thismin = UINT64_MAX; 1651088f3894Sahrens uint64_t thismax = 0; 1652088f3894Sahrens 1653088f3894Sahrens if (vd->vdev_children == 0) { 1654088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 16558ad4d6ddSJeff Bonwick if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 16568ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 1657088f3894Sahrens space_seg_t *ss; 1658088f3894Sahrens 16598ad4d6ddSJeff Bonwick ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1660088f3894Sahrens thismin = ss->ss_start - 1; 16618ad4d6ddSJeff Bonwick ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1662088f3894Sahrens thismax = ss->ss_end; 1663088f3894Sahrens needed = B_TRUE; 1664088f3894Sahrens } 1665088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 1666088f3894Sahrens } else { 16678ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 1668088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 1669088f3894Sahrens uint64_t cmin, cmax; 1670088f3894Sahrens 1671088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1672088f3894Sahrens thismin = MIN(thismin, cmin); 1673088f3894Sahrens thismax = MAX(thismax, cmax); 1674088f3894Sahrens needed = B_TRUE; 1675088f3894Sahrens } 1676088f3894Sahrens } 1677088f3894Sahrens } 1678088f3894Sahrens 1679088f3894Sahrens if (needed && minp) { 1680088f3894Sahrens *minp = thismin; 1681088f3894Sahrens *maxp = thismax; 1682088f3894Sahrens } 1683088f3894Sahrens return (needed); 1684088f3894Sahrens } 1685088f3894Sahrens 1686560e6e96Seschrock void 1687ea8dc4b6Seschrock vdev_load(vdev_t *vd) 1688fa9e4066Sahrens { 1689fa9e4066Sahrens /* 1690fa9e4066Sahrens * Recursively load all children. 1691fa9e4066Sahrens */ 16928ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1693560e6e96Seschrock vdev_load(vd->vdev_child[c]); 1694fa9e4066Sahrens 1695fa9e4066Sahrens /* 16960e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 1697fa9e4066Sahrens */ 1698560e6e96Seschrock if (vd == vd->vdev_top && 1699560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1700560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 1701560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1702560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1703fa9e4066Sahrens 1704fa9e4066Sahrens /* 1705fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 1706fa9e4066Sahrens */ 1707560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1708560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1709560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1710fa9e4066Sahrens } 1711fa9e4066Sahrens 171299653d4eSeschrock /* 1713fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 1714fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 1715fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 1716fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 1717fa94a07fSbrendan * repurposed to another pool. 171899653d4eSeschrock */ 171999653d4eSeschrock int 1720fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 172199653d4eSeschrock { 172299653d4eSeschrock nvlist_t *label; 172399653d4eSeschrock uint64_t guid, version; 172499653d4eSeschrock uint64_t state; 172599653d4eSeschrock 1726e14bb325SJeff Bonwick if (!vdev_readable(vd)) 1727c5904d13Seschrock return (0); 1728c5904d13Seschrock 172999653d4eSeschrock if ((label = vdev_label_read_config(vd)) == NULL) { 173099653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 173199653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 173299653d4eSeschrock return (-1); 173399653d4eSeschrock } 173499653d4eSeschrock 173599653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1736e7437265Sahrens version > SPA_VERSION || 173799653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 173899653d4eSeschrock guid != vd->vdev_guid || 173999653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 174099653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 174199653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 174299653d4eSeschrock nvlist_free(label); 174399653d4eSeschrock return (-1); 174499653d4eSeschrock } 174599653d4eSeschrock 174699653d4eSeschrock /* 174799653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 174899653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 174999653d4eSeschrock */ 175099653d4eSeschrock nvlist_free(label); 175199653d4eSeschrock return (0); 175299653d4eSeschrock } 175399653d4eSeschrock 1754fa9e4066Sahrens void 1755fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 1756fa9e4066Sahrens { 1757fa9e4066Sahrens metaslab_t *msp; 1758fa9e4066Sahrens 1759fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 1760fa9e4066Sahrens metaslab_sync_done(msp, txg); 1761fa9e4066Sahrens } 1762fa9e4066Sahrens 1763fa9e4066Sahrens void 1764fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 1765fa9e4066Sahrens { 1766fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1767fa9e4066Sahrens vdev_t *lvd; 1768fa9e4066Sahrens metaslab_t *msp; 1769ecc2d604Sbonwick dmu_tx_t *tx; 1770fa9e4066Sahrens 1771ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 1772ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 1773ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1774ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 1775ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 1776ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 1777ecc2d604Sbonwick vdev_config_dirty(vd); 1778ecc2d604Sbonwick dmu_tx_commit(tx); 1779ecc2d604Sbonwick } 1780fa9e4066Sahrens 1781ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 1782fa9e4066Sahrens metaslab_sync(msp, txg); 1783ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 1784ecc2d604Sbonwick } 1785fa9e4066Sahrens 1786fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 1787fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 1788fa9e4066Sahrens 1789fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 1790fa9e4066Sahrens } 1791fa9e4066Sahrens 1792fa9e4066Sahrens uint64_t 1793fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 1794fa9e4066Sahrens { 1795fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 1796fa9e4066Sahrens } 1797fa9e4066Sahrens 17983d7072f8Seschrock /* 17993d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 18003d7072f8Seschrock * not be opened, and no I/O is attempted. 18013d7072f8Seschrock */ 1802fa9e4066Sahrens int 18033d7072f8Seschrock vdev_fault(spa_t *spa, uint64_t guid) 1804fa9e4066Sahrens { 1805c5904d13Seschrock vdev_t *vd; 1806fa9e4066Sahrens 1807e14bb325SJeff Bonwick spa_vdev_state_enter(spa); 1808fa9e4066Sahrens 1809c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1810e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1811e14bb325SJeff Bonwick 18123d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 1813e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1814fa9e4066Sahrens 18153d7072f8Seschrock /* 18163d7072f8Seschrock * Faulted state takes precedence over degraded. 18173d7072f8Seschrock */ 18183d7072f8Seschrock vd->vdev_faulted = 1ULL; 18193d7072f8Seschrock vd->vdev_degraded = 0ULL; 1820e14bb325SJeff Bonwick vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED); 18213d7072f8Seschrock 18223d7072f8Seschrock /* 18236988b9faSDavid Marker * If marking the vdev as faulted cause the top-level vdev to become 18243d7072f8Seschrock * unavailable, then back off and simply mark the vdev as degraded 18253d7072f8Seschrock * instead. 18263d7072f8Seschrock */ 1827c5904d13Seschrock if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) { 18283d7072f8Seschrock vd->vdev_degraded = 1ULL; 18293d7072f8Seschrock vd->vdev_faulted = 0ULL; 18303d7072f8Seschrock 18313d7072f8Seschrock /* 18323d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 18333d7072f8Seschrock * mark it degraded. 18343d7072f8Seschrock */ 18353d7072f8Seschrock vdev_reopen(vd); 18363d7072f8Seschrock 18370a4e9518Sgw if (vdev_readable(vd)) { 18383d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 18393d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 18403d7072f8Seschrock } 18413d7072f8Seschrock } 18423d7072f8Seschrock 1843e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 18443d7072f8Seschrock } 18453d7072f8Seschrock 18463d7072f8Seschrock /* 18473d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 18483d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 18493d7072f8Seschrock * as I/O is concerned. 18503d7072f8Seschrock */ 18513d7072f8Seschrock int 18523d7072f8Seschrock vdev_degrade(spa_t *spa, uint64_t guid) 18533d7072f8Seschrock { 1854c5904d13Seschrock vdev_t *vd; 18550a4e9518Sgw 1856e14bb325SJeff Bonwick spa_vdev_state_enter(spa); 18573d7072f8Seschrock 1858c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1859e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1860e14bb325SJeff Bonwick 18610e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 1862e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 18630e34b6a7Sbonwick 18643d7072f8Seschrock /* 18653d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 18663d7072f8Seschrock */ 1867e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 1868e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 18693d7072f8Seschrock 18703d7072f8Seschrock vd->vdev_degraded = 1ULL; 18713d7072f8Seschrock if (!vdev_is_dead(vd)) 18723d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 18733d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 18743d7072f8Seschrock 1875e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 18763d7072f8Seschrock } 18773d7072f8Seschrock 18783d7072f8Seschrock /* 18793d7072f8Seschrock * Online the given vdev. If 'unspare' is set, it implies two things. First, 18803d7072f8Seschrock * any attached spare device should be detached when the device finishes 18813d7072f8Seschrock * resilvering. Second, the online should be treated like a 'test' online case, 18823d7072f8Seschrock * so no FMA events are generated if the device fails to open. 18833d7072f8Seschrock */ 18843d7072f8Seschrock int 1885e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 18863d7072f8Seschrock { 1887c5904d13Seschrock vdev_t *vd; 18883d7072f8Seschrock 1889e14bb325SJeff Bonwick spa_vdev_state_enter(spa); 18903d7072f8Seschrock 1891c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1892e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 18933d7072f8Seschrock 18943d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 1895e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 1896fa9e4066Sahrens 1897fa9e4066Sahrens vd->vdev_offline = B_FALSE; 1898441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 1899e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 1900e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 1901ea8dc4b6Seschrock vdev_reopen(vd->vdev_top); 19023d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 19033d7072f8Seschrock 19043d7072f8Seschrock if (newstate) 19053d7072f8Seschrock *newstate = vd->vdev_state; 19063d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 19073d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 19083d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 19093d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 19103d7072f8Seschrock vd->vdev_unspare = B_TRUE; 1911fa9e4066Sahrens 19128ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 1913fa9e4066Sahrens } 1914fa9e4066Sahrens 1915fa9e4066Sahrens int 19163d7072f8Seschrock vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 1917fa9e4066Sahrens { 1918*e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 1919*e6ca193dSGeorge Wilson int error; 19200a4e9518Sgw 1921e14bb325SJeff Bonwick spa_vdev_state_enter(spa); 1922fa9e4066Sahrens 1923c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1924e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 1925fa9e4066Sahrens 19260e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 1927e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 19280e34b6a7Sbonwick 1929*e6ca193dSGeorge Wilson tvd = vd->vdev_top; 1930*e6ca193dSGeorge Wilson 1931fa9e4066Sahrens /* 1932ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 1933fa9e4066Sahrens */ 1934ecc2d604Sbonwick if (!vd->vdev_offline) { 1935ecc2d604Sbonwick /* 19368ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 1937*e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 1938*e6ca193dSGeorge Wilson * expendable. 1939ecc2d604Sbonwick */ 1940*e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 1941*e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 1942e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 1943fa9e4066Sahrens 1944ecc2d604Sbonwick /* 1945ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 1946*e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 1947*e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 1948*e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 1949ecc2d604Sbonwick */ 1950ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 1951*e6ca193dSGeorge Wilson vdev_reopen(tvd); 1952*e6ca193dSGeorge Wilson 1953*e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 1954*e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 1955ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 1956*e6ca193dSGeorge Wilson vdev_reopen(tvd); 1957e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 1958ecc2d604Sbonwick } 1959fa9e4066Sahrens } 1960fa9e4066Sahrens 1961e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 1962ecc2d604Sbonwick 1963*e6ca193dSGeorge Wilson if (!tvd->vdev_islog || !vdev_is_dead(tvd)) 1964*e6ca193dSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 1965*e6ca193dSGeorge Wilson 1966*e6ca193dSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 1967*e6ca193dSGeorge Wilson 1968*e6ca193dSGeorge Wilson error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1969*e6ca193dSGeorge Wilson NULL, DS_FIND_CHILDREN); 1970*e6ca193dSGeorge Wilson if (error) { 1971*e6ca193dSGeorge Wilson (void) vdev_online(spa, guid, 0, NULL); 1972*e6ca193dSGeorge Wilson return (error); 1973*e6ca193dSGeorge Wilson } 1974*e6ca193dSGeorge Wilson /* 1975*e6ca193dSGeorge Wilson * If we successfully offlined the log device then we need to 1976*e6ca193dSGeorge Wilson * sync out the current txg so that the "stubby" block can be 1977*e6ca193dSGeorge Wilson * removed by zil_sync(). 1978*e6ca193dSGeorge Wilson */ 1979*e6ca193dSGeorge Wilson txg_wait_synced(spa->spa_dsl_pool, 0); 1980*e6ca193dSGeorge Wilson return (0); 1981fa9e4066Sahrens } 1982fa9e4066Sahrens 1983ea8dc4b6Seschrock /* 1984ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 1985ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 1986ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 1987ea8dc4b6Seschrock */ 1988ea8dc4b6Seschrock void 1989e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 1990fa9e4066Sahrens { 1991e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 1992e14bb325SJeff Bonwick 1993e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1994fa9e4066Sahrens 1995ea8dc4b6Seschrock if (vd == NULL) 1996e14bb325SJeff Bonwick vd = rvd; 1997fa9e4066Sahrens 1998ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 1999ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 2000ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 2001fa9e4066Sahrens 2002e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 2003e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 20043d7072f8Seschrock 20053d7072f8Seschrock /* 20068a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 20078a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 20088a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 20098a79c1b5Sek * written out to disk. 20103d7072f8Seschrock */ 2011e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 2012e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 20138a79c1b5Sek 20143d7072f8Seschrock vd->vdev_faulted = vd->vdev_degraded = 0; 2015e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 2016e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 2017e14bb325SJeff Bonwick 20183d7072f8Seschrock vdev_reopen(vd); 20193d7072f8Seschrock 2020e14bb325SJeff Bonwick if (vd != rvd) 2021e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 2022e14bb325SJeff Bonwick 2023e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2024bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 20253d7072f8Seschrock 20263d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 20273d7072f8Seschrock } 2028fa9e4066Sahrens } 2029fa9e4066Sahrens 2030e14bb325SJeff Bonwick boolean_t 2031e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 20320a4e9518Sgw { 2033e14bb325SJeff Bonwick return (vd->vdev_state < VDEV_STATE_DEGRADED); 20340a4e9518Sgw } 20350a4e9518Sgw 2036e14bb325SJeff Bonwick boolean_t 2037e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 20380a4e9518Sgw { 2039e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 20400a4e9518Sgw } 20410a4e9518Sgw 2042e14bb325SJeff Bonwick boolean_t 2043e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 2044fa9e4066Sahrens { 2045e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2046fa9e4066Sahrens } 2047fa9e4066Sahrens 2048a31e6787SGeorge Wilson boolean_t 2049a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 2050a31e6787SGeorge Wilson { 20518ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 20528ad4d6ddSJeff Bonwick 2053a31e6787SGeorge Wilson /* 20548ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 2055a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2056a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 20578ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 20588ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 20598ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 2060a31e6787SGeorge Wilson */ 20618ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2062a31e6787SGeorge Wilson !vd->vdev_cant_write); 2063a31e6787SGeorge Wilson } 2064a31e6787SGeorge Wilson 2065e14bb325SJeff Bonwick boolean_t 2066e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 2067fa9e4066Sahrens { 2068e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 2069fa9e4066Sahrens 2070e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2071e14bb325SJeff Bonwick return (B_FALSE); 2072fa9e4066Sahrens 2073e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 2074e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 2075fa9e4066Sahrens 2076e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2077e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 2078fa9e4066Sahrens 2079e14bb325SJeff Bonwick return (B_TRUE); 2080fa9e4066Sahrens } 2081fa9e4066Sahrens 2082fa9e4066Sahrens /* 2083fa9e4066Sahrens * Get statistics for the given vdev. 2084fa9e4066Sahrens */ 2085fa9e4066Sahrens void 2086fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2087fa9e4066Sahrens { 2088fa9e4066Sahrens vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2089fa9e4066Sahrens 2090fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2091fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2092088f3894Sahrens vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; 2093fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2094fa9e4066Sahrens vs->vs_state = vd->vdev_state; 20952a79c5feSlling vs->vs_rsize = vdev_get_rsize(vd); 2096fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2097fa9e4066Sahrens 2098fa9e4066Sahrens /* 2099fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 2100fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 2101fa9e4066Sahrens */ 2102fa9e4066Sahrens if (vd == rvd) { 2103e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 2104fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 2105fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 2106fa9e4066Sahrens 2107fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2108e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 2109fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 2110fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 2111fa9e4066Sahrens } 2112fa9e4066Sahrens vs->vs_scrub_examined += cvs->vs_scrub_examined; 2113fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2114fa9e4066Sahrens } 2115fa9e4066Sahrens } 2116fa9e4066Sahrens } 2117fa9e4066Sahrens 2118fa94a07fSbrendan void 2119fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 2120fa94a07fSbrendan { 2121fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 2122fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 2123fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 2124fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 2125fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 2126fa94a07fSbrendan } 2127fa94a07fSbrendan 2128fa9e4066Sahrens void 2129e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 2130fa9e4066Sahrens { 21318ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 21328ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 2133e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2134fa9e4066Sahrens vdev_t *pvd; 2135fa9e4066Sahrens uint64_t txg = zio->io_txg; 2136fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2137fa9e4066Sahrens zio_type_t type = zio->io_type; 2138fa9e4066Sahrens int flags = zio->io_flags; 2139fa9e4066Sahrens 2140e14bb325SJeff Bonwick /* 2141e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 2142e14bb325SJeff Bonwick */ 2143e14bb325SJeff Bonwick if (zio->io_gang_tree) 2144e14bb325SJeff Bonwick return; 2145e14bb325SJeff Bonwick 2146fa9e4066Sahrens if (zio->io_error == 0) { 2147e14bb325SJeff Bonwick /* 2148e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 2149e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 2150e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 2151e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 2152e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 2153e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 2154e14bb325SJeff Bonwick * remain zero, which is what we want.) 2155e14bb325SJeff Bonwick * 2156e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 2157e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 2158e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 2159e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 2160e14bb325SJeff Bonwick */ 2161e14bb325SJeff Bonwick if (vd == rvd) 2162e14bb325SJeff Bonwick return; 2163e14bb325SJeff Bonwick 2164e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 21658ad4d6ddSJeff Bonwick 21668ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 21678ad4d6ddSJeff Bonwick return; 21688ad4d6ddSJeff Bonwick 21698ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 21708ad4d6ddSJeff Bonwick 2171e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 2172d80c45e0Sbonwick if (flags & ZIO_FLAG_SCRUB_THREAD) 2173e14bb325SJeff Bonwick vs->vs_scrub_repaired += psize; 21748ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 2175e14bb325SJeff Bonwick vs->vs_self_healed += psize; 2176fa9e4066Sahrens } 21778ad4d6ddSJeff Bonwick 21788ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 21798ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 21808ad4d6ddSJeff Bonwick 21818ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2182fa9e4066Sahrens return; 2183fa9e4066Sahrens } 2184fa9e4066Sahrens 2185fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 2186fa9e4066Sahrens return; 2187fa9e4066Sahrens 2188e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 2189b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2190e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 2191e14bb325SJeff Bonwick vs->vs_checksum_errors++; 2192e14bb325SJeff Bonwick else 2193e14bb325SJeff Bonwick vs->vs_read_errors++; 2194fa9e4066Sahrens } 2195b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2196e14bb325SJeff Bonwick vs->vs_write_errors++; 2197e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 2198fa9e4066Sahrens 21998ad4d6ddSJeff Bonwick if (type == ZIO_TYPE_WRITE && txg != 0 && 22008ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 22018ad4d6ddSJeff Bonwick (flags & ZIO_FLAG_SCRUB_THREAD))) { 22028ad4d6ddSJeff Bonwick /* 22038ad4d6ddSJeff Bonwick * This is either a normal write (not a repair), or it's a 22048ad4d6ddSJeff Bonwick * repair induced by the scrub thread. In the normal case, 22058ad4d6ddSJeff Bonwick * we commit the DTL change in the same txg as the block 22068ad4d6ddSJeff Bonwick * was born. In the scrub-induced repair case, we know that 22078ad4d6ddSJeff Bonwick * scrubs run in first-pass syncing context, so we commit 22088ad4d6ddSJeff Bonwick * the DTL change in spa->spa_syncing_txg. 22098ad4d6ddSJeff Bonwick * 22108ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 22118ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 22128ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 22138ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 22148ad4d6ddSJeff Bonwick */ 22158ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 22168ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 22178ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SCRUB_THREAD) { 22188ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 22198ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 22208ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 22218ad4d6ddSJeff Bonwick commit_txg = spa->spa_syncing_txg; 22228ad4d6ddSJeff Bonwick } 22238ad4d6ddSJeff Bonwick ASSERT(commit_txg >= spa->spa_syncing_txg); 22248ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2225fa9e4066Sahrens return; 22268ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 22278ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 22288ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2229fa9e4066Sahrens } 22308ad4d6ddSJeff Bonwick if (vd != rvd) 22318ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2232fa9e4066Sahrens } 2233fa9e4066Sahrens } 2234fa9e4066Sahrens 2235fa9e4066Sahrens void 2236fa9e4066Sahrens vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) 2237fa9e4066Sahrens { 2238fa9e4066Sahrens int c; 2239fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2240fa9e4066Sahrens 2241fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 2242fa9e4066Sahrens vdev_scrub_stat_update(vd->vdev_child[c], type, complete); 2243fa9e4066Sahrens 2244fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2245fa9e4066Sahrens 2246fa9e4066Sahrens if (type == POOL_SCRUB_NONE) { 2247fa9e4066Sahrens /* 2248fa9e4066Sahrens * Update completion and end time. Leave everything else alone 2249fa9e4066Sahrens * so we can report what happened during the previous scrub. 2250fa9e4066Sahrens */ 2251fa9e4066Sahrens vs->vs_scrub_complete = complete; 2252fa9e4066Sahrens vs->vs_scrub_end = gethrestime_sec(); 2253fa9e4066Sahrens } else { 2254fa9e4066Sahrens vs->vs_scrub_type = type; 2255fa9e4066Sahrens vs->vs_scrub_complete = 0; 2256fa9e4066Sahrens vs->vs_scrub_examined = 0; 2257fa9e4066Sahrens vs->vs_scrub_repaired = 0; 2258fa9e4066Sahrens vs->vs_scrub_start = gethrestime_sec(); 2259fa9e4066Sahrens vs->vs_scrub_end = 0; 2260fa9e4066Sahrens } 2261fa9e4066Sahrens 2262fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2263fa9e4066Sahrens } 2264fa9e4066Sahrens 2265fa9e4066Sahrens /* 2266fa9e4066Sahrens * Update the in-core space usage stats for this vdev and the root vdev. 2267fa9e4066Sahrens */ 2268fa9e4066Sahrens void 2269fa94a07fSbrendan vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta, 2270fa94a07fSbrendan boolean_t update_root) 2271fa9e4066Sahrens { 227299653d4eSeschrock int64_t dspace_delta = space_delta; 22738654d025Sperrin spa_t *spa = vd->vdev_spa; 22748654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2275fa9e4066Sahrens 22768654d025Sperrin ASSERT(vd == vd->vdev_top); 227799653d4eSeschrock 22788654d025Sperrin /* 22798654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 22808654d025Sperrin * factor. We must calculate this here and not at the root vdev 22818654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 22828654d025Sperrin * childrens', thus not accurate enough for us. 22838654d025Sperrin */ 22848654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2285*e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 22868654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 22878654d025Sperrin vd->vdev_deflate_ratio; 22888654d025Sperrin 22898654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 22908654d025Sperrin vd->vdev_stat.vs_space += space_delta; 22918654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 22928654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 22938654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 22948654d025Sperrin 2295fa94a07fSbrendan if (update_root) { 2296fa94a07fSbrendan ASSERT(rvd == vd->vdev_parent); 2297fa94a07fSbrendan ASSERT(vd->vdev_ms_count != 0); 2298fa94a07fSbrendan 2299fa94a07fSbrendan /* 2300fa94a07fSbrendan * Don't count non-normal (e.g. intent log) space as part of 2301fa94a07fSbrendan * the pool's capacity. 2302fa94a07fSbrendan */ 2303fa94a07fSbrendan if (vd->vdev_mg->mg_class != spa->spa_normal_class) 2304fa94a07fSbrendan return; 23058654d025Sperrin 2306fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2307fa94a07fSbrendan rvd->vdev_stat.vs_space += space_delta; 2308fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2309fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2310fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2311fa94a07fSbrendan } 2312fa9e4066Sahrens } 2313fa9e4066Sahrens 2314fa9e4066Sahrens /* 2315fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2316fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2317fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2318fa9e4066Sahrens */ 2319fa9e4066Sahrens void 2320fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2321fa9e4066Sahrens { 2322fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2323fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2324fa9e4066Sahrens int c; 2325fa9e4066Sahrens 2326c5904d13Seschrock /* 23276809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 23286809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 2329c5904d13Seschrock */ 2330c5904d13Seschrock if (vd->vdev_aux != NULL) { 2331c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2332c5904d13Seschrock nvlist_t **aux; 2333c5904d13Seschrock uint_t naux; 2334c5904d13Seschrock 2335c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2336c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2337c5904d13Seschrock break; 2338c5904d13Seschrock } 2339c5904d13Seschrock 2340e14bb325SJeff Bonwick if (c == sav->sav_count) { 2341e14bb325SJeff Bonwick /* 2342e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 2343e14bb325SJeff Bonwick */ 2344e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 2345e14bb325SJeff Bonwick return; 2346e14bb325SJeff Bonwick } 2347e14bb325SJeff Bonwick 2348c5904d13Seschrock sav->sav_sync = B_TRUE; 2349c5904d13Seschrock 23506809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 23516809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 23526809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 23536809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 23546809eb4eSEric Schrock } 2355c5904d13Seschrock 2356c5904d13Seschrock ASSERT(c < naux); 2357c5904d13Seschrock 2358c5904d13Seschrock /* 2359c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2360c5904d13Seschrock * sketchy, but it will work. 2361c5904d13Seschrock */ 2362c5904d13Seschrock nvlist_free(aux[c]); 2363c5904d13Seschrock aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); 2364c5904d13Seschrock 2365c5904d13Seschrock return; 2366c5904d13Seschrock } 2367c5904d13Seschrock 23685dabedeeSbonwick /* 2369e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 2370e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 2371e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 23725dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 23735dabedeeSbonwick */ 2374e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2375e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2376e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 23775dabedeeSbonwick 2378fa9e4066Sahrens if (vd == rvd) { 2379fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2380fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2381fa9e4066Sahrens } else { 2382fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2383fa9e4066Sahrens 2384e14bb325SJeff Bonwick if (!list_link_active(&vd->vdev_config_dirty_node)) 2385e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 2386fa9e4066Sahrens } 2387fa9e4066Sahrens } 2388fa9e4066Sahrens 2389fa9e4066Sahrens void 2390fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2391fa9e4066Sahrens { 23925dabedeeSbonwick spa_t *spa = vd->vdev_spa; 23935dabedeeSbonwick 2394e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2395e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2396e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 23975dabedeeSbonwick 2398e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2399e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 2400e14bb325SJeff Bonwick } 2401e14bb325SJeff Bonwick 2402e14bb325SJeff Bonwick /* 2403e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 2404e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2405e14bb325SJeff Bonwick * the state changes from larger config changes because they require 2406e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 2407e14bb325SJeff Bonwick */ 2408e14bb325SJeff Bonwick void 2409e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 2410e14bb325SJeff Bonwick { 2411e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2412e14bb325SJeff Bonwick 2413e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2414e14bb325SJeff Bonwick 2415e14bb325SJeff Bonwick /* 2416e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 2417e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 2418e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 2419e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 2420e14bb325SJeff Bonwick */ 2421e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2422e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2423e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2424e14bb325SJeff Bonwick 2425e14bb325SJeff Bonwick if (!list_link_active(&vd->vdev_state_dirty_node)) 2426e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 2427e14bb325SJeff Bonwick } 2428e14bb325SJeff Bonwick 2429e14bb325SJeff Bonwick void 2430e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 2431e14bb325SJeff Bonwick { 2432e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 2433e14bb325SJeff Bonwick 2434e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2435e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 2436e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 2437e14bb325SJeff Bonwick 2438e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2439e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 2440fa9e4066Sahrens } 2441fa9e4066Sahrens 244232b87932Sek /* 244332b87932Sek * Propagate vdev state up from children to parent. 244432b87932Sek */ 244544cd46caSbillm void 244644cd46caSbillm vdev_propagate_state(vdev_t *vd) 244744cd46caSbillm { 24488ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 24498ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 245044cd46caSbillm int degraded = 0, faulted = 0; 245144cd46caSbillm int corrupted = 0; 245244cd46caSbillm int c; 245344cd46caSbillm vdev_t *child; 245444cd46caSbillm 24553d7072f8Seschrock if (vd->vdev_children > 0) { 24563d7072f8Seschrock for (c = 0; c < vd->vdev_children; c++) { 24573d7072f8Seschrock child = vd->vdev_child[c]; 245851ece835Seschrock 2459e14bb325SJeff Bonwick if (!vdev_readable(child) || 24608ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 246151ece835Seschrock /* 246251ece835Seschrock * Root special: if there is a top-level log 246351ece835Seschrock * device, treat the root vdev as if it were 246451ece835Seschrock * degraded. 246551ece835Seschrock */ 246651ece835Seschrock if (child->vdev_islog && vd == rvd) 246751ece835Seschrock degraded++; 246851ece835Seschrock else 246951ece835Seschrock faulted++; 247051ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 24713d7072f8Seschrock degraded++; 247251ece835Seschrock } 247344cd46caSbillm 24743d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 24753d7072f8Seschrock corrupted++; 24763d7072f8Seschrock } 247744cd46caSbillm 24783d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 24793d7072f8Seschrock 24803d7072f8Seschrock /* 2481e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 24823d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 24833d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 24843d7072f8Seschrock * replicas'. 24853d7072f8Seschrock */ 24863d7072f8Seschrock if (corrupted && vd == rvd && 24873d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 24883d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 24893d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 24903d7072f8Seschrock } 24913d7072f8Seschrock 249251ece835Seschrock if (vd->vdev_parent) 24933d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 249444cd46caSbillm } 249544cd46caSbillm 2496fa9e4066Sahrens /* 2497ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 2498ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 2499ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 2500ea8dc4b6Seschrock * 2501ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 2502ea8dc4b6Seschrock * generated. 2503fa9e4066Sahrens */ 2504fa9e4066Sahrens void 2505ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2506fa9e4066Sahrens { 2507560e6e96Seschrock uint64_t save_state; 2508c5904d13Seschrock spa_t *spa = vd->vdev_spa; 2509ea8dc4b6Seschrock 2510ea8dc4b6Seschrock if (state == vd->vdev_state) { 2511ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 2512fa9e4066Sahrens return; 2513ea8dc4b6Seschrock } 2514ea8dc4b6Seschrock 2515560e6e96Seschrock save_state = vd->vdev_state; 2516fa9e4066Sahrens 2517fa9e4066Sahrens vd->vdev_state = state; 2518fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 2519fa9e4066Sahrens 25203d7072f8Seschrock /* 25213d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 25223d7072f8Seschrock * always close the underlying device. Otherwise, we keep accessible 25233d7072f8Seschrock * but invalid devices open forever. We don't call vdev_close() itself, 25243d7072f8Seschrock * because that implies some extra checks (offline, etc) that we don't 25253d7072f8Seschrock * want here. This is limited to leaf devices, because otherwise 25263d7072f8Seschrock * closing the device will affect other children. 25273d7072f8Seschrock */ 2528cbd2b15eSJeff Bonwick if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) 25293d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 25303d7072f8Seschrock 25313d7072f8Seschrock if (vd->vdev_removed && 25323d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 25333d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 25343d7072f8Seschrock /* 25353d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 25363d7072f8Seschrock * device was previously marked removed and someone attempted to 25373d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 25383d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 25393d7072f8Seschrock * it is one of our special test online cases, which is only 25403d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 25413d7072f8Seschrock * fault. 25423d7072f8Seschrock */ 25433d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 25443d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 25453d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 25463d7072f8Seschrock /* 25473d7072f8Seschrock * Indicate to the ZFS DE that this device has been removed, and 25483d7072f8Seschrock * any recent errors should be ignored. 25493d7072f8Seschrock */ 2550c5904d13Seschrock zfs_post_remove(spa, vd); 25513d7072f8Seschrock vd->vdev_removed = B_TRUE; 25523d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 2553ea8dc4b6Seschrock /* 2554ea8dc4b6Seschrock * If we fail to open a vdev during an import, we mark it as 2555ea8dc4b6Seschrock * "not available", which signifies that it was never there to 2556ea8dc4b6Seschrock * begin with. Failure to open such a device is not considered 2557ea8dc4b6Seschrock * an error. 2558ea8dc4b6Seschrock */ 2559c5904d13Seschrock if (spa->spa_load_state == SPA_LOAD_IMPORT && 2560560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 2561560e6e96Seschrock vd->vdev_not_present = 1; 2562560e6e96Seschrock 2563560e6e96Seschrock /* 2564560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 2565560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 2566560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 2567560e6e96Seschrock * want to post the ereport if the device was already in the 2568560e6e96Seschrock * CANT_OPEN state beforehand. 25693d7072f8Seschrock * 25703d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 25713d7072f8Seschrock * online the device in response to an insertion event. If we 25723d7072f8Seschrock * hit this case, then we have detected an insertion event for a 25733d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 25743d7072f8Seschrock * In this scenario, we don't post an ereport because we are 25753d7072f8Seschrock * about to replace the device, or attempt an online with 25763d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 2577560e6e96Seschrock */ 25783d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 25793d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 2580c5904d13Seschrock vd != spa->spa_root_vdev) { 2581ea8dc4b6Seschrock const char *class; 2582ea8dc4b6Seschrock 2583ea8dc4b6Seschrock switch (aux) { 2584ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 2585ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2586ea8dc4b6Seschrock break; 2587ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 2588ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2589ea8dc4b6Seschrock break; 2590ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 2591ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2592ea8dc4b6Seschrock break; 2593ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 2594ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2595ea8dc4b6Seschrock break; 2596ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 2597ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 2598ea8dc4b6Seschrock break; 2599ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 2600ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 2601ea8dc4b6Seschrock break; 2602e14bb325SJeff Bonwick case VDEV_AUX_IO_FAILURE: 2603e14bb325SJeff Bonwick class = FM_EREPORT_ZFS_IO_FAILURE; 2604e14bb325SJeff Bonwick break; 2605ea8dc4b6Seschrock default: 2606ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 2607ea8dc4b6Seschrock } 2608ea8dc4b6Seschrock 2609c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 2610ea8dc4b6Seschrock } 2611ea8dc4b6Seschrock 26123d7072f8Seschrock /* Erase any notion of persistent removed state */ 26133d7072f8Seschrock vd->vdev_removed = B_FALSE; 26143d7072f8Seschrock } else { 26153d7072f8Seschrock vd->vdev_removed = B_FALSE; 26163d7072f8Seschrock } 2617ea8dc4b6Seschrock 26188b33d774STim Haley if (!isopen && vd->vdev_parent) 26198b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 2620fa9e4066Sahrens } 262115e6edf1Sgw 262215e6edf1Sgw /* 262315e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 262415e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 262515e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 262615e6edf1Sgw * can be wholedisks. 262715e6edf1Sgw */ 262815e6edf1Sgw boolean_t 262915e6edf1Sgw vdev_is_bootable(vdev_t *vd) 263015e6edf1Sgw { 263115e6edf1Sgw int c; 263215e6edf1Sgw 263315e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 263415e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 263515e6edf1Sgw 263615e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 263715e6edf1Sgw vd->vdev_children > 1) { 263815e6edf1Sgw return (B_FALSE); 263915e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 264015e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 264115e6edf1Sgw return (B_FALSE); 264215e6edf1Sgw } 264315e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 264415e6edf1Sgw return (B_FALSE); 264515e6edf1Sgw } 264615e6edf1Sgw 264715e6edf1Sgw for (c = 0; c < vd->vdev_children; c++) { 264815e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 264915e6edf1Sgw return (B_FALSE); 265015e6edf1Sgw } 265115e6edf1Sgw return (B_TRUE); 265215e6edf1Sgw } 2653*e6ca193dSGeorge Wilson 2654*e6ca193dSGeorge Wilson void 2655*e6ca193dSGeorge Wilson vdev_load_log_state(vdev_t *vd, nvlist_t *nv) 2656*e6ca193dSGeorge Wilson { 2657*e6ca193dSGeorge Wilson uint_t c, children; 2658*e6ca193dSGeorge Wilson nvlist_t **child; 2659*e6ca193dSGeorge Wilson uint64_t val; 2660*e6ca193dSGeorge Wilson spa_t *spa = vd->vdev_spa; 2661*e6ca193dSGeorge Wilson 2662*e6ca193dSGeorge Wilson if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 2663*e6ca193dSGeorge Wilson &child, &children) == 0) { 2664*e6ca193dSGeorge Wilson for (c = 0; c < children; c++) 2665*e6ca193dSGeorge Wilson vdev_load_log_state(vd->vdev_child[c], child[c]); 2666*e6ca193dSGeorge Wilson } 2667*e6ca193dSGeorge Wilson 2668*e6ca193dSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && nvlist_lookup_uint64(nv, 2669*e6ca193dSGeorge Wilson ZPOOL_CONFIG_OFFLINE, &val) == 0 && val) { 2670*e6ca193dSGeorge Wilson 2671*e6ca193dSGeorge Wilson /* 2672*e6ca193dSGeorge Wilson * It would be nice to call vdev_offline() 2673*e6ca193dSGeorge Wilson * directly but the pool isn't fully loaded and 2674*e6ca193dSGeorge Wilson * the txg threads have not been started yet. 2675*e6ca193dSGeorge Wilson */ 2676*e6ca193dSGeorge Wilson spa_config_enter(spa, SCL_STATE_ALL, FTAG, RW_WRITER); 2677*e6ca193dSGeorge Wilson vd->vdev_offline = val; 2678*e6ca193dSGeorge Wilson vdev_reopen(vd->vdev_top); 2679*e6ca193dSGeorge Wilson spa_config_exit(spa, SCL_STATE_ALL, FTAG); 2680*e6ca193dSGeorge Wilson } 2681*e6ca193dSGeorge Wilson } 2682