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 /* 2332b87932Sek * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24fa9e4066Sahrens * Use is subject to license terms. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #pragma ident "%Z%%M% %I% %E% SMI" 28fa9e4066Sahrens 29fa9e4066Sahrens #include <sys/zfs_context.h> 30ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 31fa9e4066Sahrens #include <sys/spa.h> 32fa9e4066Sahrens #include <sys/spa_impl.h> 33fa9e4066Sahrens #include <sys/dmu.h> 34fa9e4066Sahrens #include <sys/dmu_tx.h> 35fa9e4066Sahrens #include <sys/vdev_impl.h> 36fa9e4066Sahrens #include <sys/uberblock_impl.h> 37fa9e4066Sahrens #include <sys/metaslab.h> 38fa9e4066Sahrens #include <sys/metaslab_impl.h> 39fa9e4066Sahrens #include <sys/space_map.h> 40fa9e4066Sahrens #include <sys/zio.h> 41fa9e4066Sahrens #include <sys/zap.h> 42fa9e4066Sahrens #include <sys/fs/zfs.h> 43c5904d13Seschrock #include <sys/arc.h> 44fa9e4066Sahrens 45fa9e4066Sahrens /* 46fa9e4066Sahrens * Virtual device management. 47fa9e4066Sahrens */ 48fa9e4066Sahrens 49fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 50fa9e4066Sahrens &vdev_root_ops, 51fa9e4066Sahrens &vdev_raidz_ops, 52fa9e4066Sahrens &vdev_mirror_ops, 53fa9e4066Sahrens &vdev_replacing_ops, 5499653d4eSeschrock &vdev_spare_ops, 55fa9e4066Sahrens &vdev_disk_ops, 56fa9e4066Sahrens &vdev_file_ops, 57fa9e4066Sahrens &vdev_missing_ops, 58fa9e4066Sahrens NULL 59fa9e4066Sahrens }; 60fa9e4066Sahrens 61*088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 62*088f3894Sahrens int zfs_scrub_limit = 10; 6305b2b3b8Smishra 64fa9e4066Sahrens /* 65fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 66fa9e4066Sahrens */ 67fa9e4066Sahrens static vdev_ops_t * 68fa9e4066Sahrens vdev_getops(const char *type) 69fa9e4066Sahrens { 70fa9e4066Sahrens vdev_ops_t *ops, **opspp; 71fa9e4066Sahrens 72fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 73fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 74fa9e4066Sahrens break; 75fa9e4066Sahrens 76fa9e4066Sahrens return (ops); 77fa9e4066Sahrens } 78fa9e4066Sahrens 79fa9e4066Sahrens /* 80fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 81fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 82fa9e4066Sahrens */ 83fa9e4066Sahrens uint64_t 84fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 85fa9e4066Sahrens { 86ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 87fa9e4066Sahrens uint64_t csize; 88fa9e4066Sahrens uint64_t c; 89fa9e4066Sahrens 90fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) { 91fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 92fa9e4066Sahrens asize = MAX(asize, csize); 93fa9e4066Sahrens } 94fa9e4066Sahrens 95fa9e4066Sahrens return (asize); 96fa9e4066Sahrens } 97fa9e4066Sahrens 982a79c5feSlling /* 992a79c5feSlling * Get the replaceable or attachable device size. 1002a79c5feSlling * If the parent is a mirror or raidz, the replaceable size is the minimum 1012a79c5feSlling * psize of all its children. For the rest, just return our own psize. 1022a79c5feSlling * 1032a79c5feSlling * e.g. 1042a79c5feSlling * psize rsize 1052a79c5feSlling * root - - 1062a79c5feSlling * mirror/raidz - - 1072a79c5feSlling * disk1 20g 20g 1082a79c5feSlling * disk2 40g 20g 1092a79c5feSlling * disk3 80g 80g 1102a79c5feSlling */ 1112a79c5feSlling uint64_t 1122a79c5feSlling vdev_get_rsize(vdev_t *vd) 1132a79c5feSlling { 1142a79c5feSlling vdev_t *pvd, *cvd; 1152a79c5feSlling uint64_t c, rsize; 1162a79c5feSlling 1172a79c5feSlling pvd = vd->vdev_parent; 1182a79c5feSlling 1192a79c5feSlling /* 1202a79c5feSlling * If our parent is NULL or the root, just return our own psize. 1212a79c5feSlling */ 1222a79c5feSlling if (pvd == NULL || pvd->vdev_parent == NULL) 1232a79c5feSlling return (vd->vdev_psize); 1242a79c5feSlling 1252a79c5feSlling rsize = 0; 1262a79c5feSlling 1272a79c5feSlling for (c = 0; c < pvd->vdev_children; c++) { 1282a79c5feSlling cvd = pvd->vdev_child[c]; 1292a79c5feSlling rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1; 1302a79c5feSlling } 1312a79c5feSlling 1322a79c5feSlling return (rsize); 1332a79c5feSlling } 1342a79c5feSlling 135fa9e4066Sahrens vdev_t * 136fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 137fa9e4066Sahrens { 138fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 139fa9e4066Sahrens 140*088f3894Sahrens ASSERT(spa_config_held(spa, RW_READER)); 141e05725b1Sbonwick 142*088f3894Sahrens if (vdev < rvd->vdev_children) { 143*088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 144fa9e4066Sahrens return (rvd->vdev_child[vdev]); 145*088f3894Sahrens } 146fa9e4066Sahrens 147fa9e4066Sahrens return (NULL); 148fa9e4066Sahrens } 149fa9e4066Sahrens 150fa9e4066Sahrens vdev_t * 151fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 152fa9e4066Sahrens { 153fa9e4066Sahrens int c; 154fa9e4066Sahrens vdev_t *mvd; 155fa9e4066Sahrens 1560e34b6a7Sbonwick if (vd->vdev_guid == guid) 157fa9e4066Sahrens return (vd); 158fa9e4066Sahrens 159fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 160fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 161fa9e4066Sahrens NULL) 162fa9e4066Sahrens return (mvd); 163fa9e4066Sahrens 164fa9e4066Sahrens return (NULL); 165fa9e4066Sahrens } 166fa9e4066Sahrens 167fa9e4066Sahrens void 168fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 169fa9e4066Sahrens { 170fa9e4066Sahrens size_t oldsize, newsize; 171fa9e4066Sahrens uint64_t id = cvd->vdev_id; 172fa9e4066Sahrens vdev_t **newchild; 173fa9e4066Sahrens 174fa9e4066Sahrens ASSERT(spa_config_held(cvd->vdev_spa, RW_WRITER)); 175fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 176fa9e4066Sahrens 177fa9e4066Sahrens cvd->vdev_parent = pvd; 178fa9e4066Sahrens 179fa9e4066Sahrens if (pvd == NULL) 180fa9e4066Sahrens return; 181fa9e4066Sahrens 182fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 183fa9e4066Sahrens 184fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 185fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 186fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 187fa9e4066Sahrens 188fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 189fa9e4066Sahrens if (pvd->vdev_child != NULL) { 190fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 191fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 192fa9e4066Sahrens } 193fa9e4066Sahrens 194fa9e4066Sahrens pvd->vdev_child = newchild; 195fa9e4066Sahrens pvd->vdev_child[id] = cvd; 196fa9e4066Sahrens 197fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 198fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 199fa9e4066Sahrens 200fa9e4066Sahrens /* 201fa9e4066Sahrens * Walk up all ancestors to update guid sum. 202fa9e4066Sahrens */ 203fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 204fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 20505b2b3b8Smishra 20605b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 20705b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit; 208fa9e4066Sahrens } 209fa9e4066Sahrens 210fa9e4066Sahrens void 211fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 212fa9e4066Sahrens { 213fa9e4066Sahrens int c; 214fa9e4066Sahrens uint_t id = cvd->vdev_id; 215fa9e4066Sahrens 216fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 217fa9e4066Sahrens 218fa9e4066Sahrens if (pvd == NULL) 219fa9e4066Sahrens return; 220fa9e4066Sahrens 221fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 222fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 223fa9e4066Sahrens 224fa9e4066Sahrens pvd->vdev_child[id] = NULL; 225fa9e4066Sahrens cvd->vdev_parent = NULL; 226fa9e4066Sahrens 227fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 228fa9e4066Sahrens if (pvd->vdev_child[c]) 229fa9e4066Sahrens break; 230fa9e4066Sahrens 231fa9e4066Sahrens if (c == pvd->vdev_children) { 232fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 233fa9e4066Sahrens pvd->vdev_child = NULL; 234fa9e4066Sahrens pvd->vdev_children = 0; 235fa9e4066Sahrens } 236fa9e4066Sahrens 237fa9e4066Sahrens /* 238fa9e4066Sahrens * Walk up all ancestors to update guid sum. 239fa9e4066Sahrens */ 240fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 241fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 24205b2b3b8Smishra 24305b2b3b8Smishra if (cvd->vdev_ops->vdev_op_leaf) 24405b2b3b8Smishra cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit; 245fa9e4066Sahrens } 246fa9e4066Sahrens 247fa9e4066Sahrens /* 248fa9e4066Sahrens * Remove any holes in the child array. 249fa9e4066Sahrens */ 250fa9e4066Sahrens void 251fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 252fa9e4066Sahrens { 253fa9e4066Sahrens vdev_t **newchild, *cvd; 254fa9e4066Sahrens int oldc = pvd->vdev_children; 255fa9e4066Sahrens int newc, c; 256fa9e4066Sahrens 257fa9e4066Sahrens ASSERT(spa_config_held(pvd->vdev_spa, RW_WRITER)); 258fa9e4066Sahrens 259fa9e4066Sahrens for (c = newc = 0; c < oldc; c++) 260fa9e4066Sahrens if (pvd->vdev_child[c]) 261fa9e4066Sahrens newc++; 262fa9e4066Sahrens 263fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 264fa9e4066Sahrens 265fa9e4066Sahrens for (c = newc = 0; c < oldc; c++) { 266fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 267fa9e4066Sahrens newchild[newc] = cvd; 268fa9e4066Sahrens cvd->vdev_id = newc++; 269fa9e4066Sahrens } 270fa9e4066Sahrens } 271fa9e4066Sahrens 272fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 273fa9e4066Sahrens pvd->vdev_child = newchild; 274fa9e4066Sahrens pvd->vdev_children = newc; 275fa9e4066Sahrens } 276fa9e4066Sahrens 277fa9e4066Sahrens /* 278fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 279fa9e4066Sahrens */ 280fa9e4066Sahrens static vdev_t * 281fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 282fa9e4066Sahrens { 283fa9e4066Sahrens vdev_t *vd; 284fa9e4066Sahrens 285fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 286fa9e4066Sahrens 2870e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 2880e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 2890e34b6a7Sbonwick spa->spa_root_vdev = vd; 2900e34b6a7Sbonwick } 2910e34b6a7Sbonwick 2920e34b6a7Sbonwick if (guid == 0) { 2930e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 2940e34b6a7Sbonwick /* 2950e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 2960e34b6a7Sbonwick * which must be unique among all pools. 2970e34b6a7Sbonwick */ 2980e34b6a7Sbonwick while (guid == 0 || spa_guid_exists(guid, 0)) 2990e34b6a7Sbonwick guid = spa_get_random(-1ULL); 3000e34b6a7Sbonwick } else { 3010e34b6a7Sbonwick /* 3020e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 3030e34b6a7Sbonwick */ 3040e34b6a7Sbonwick while (guid == 0 || 3050e34b6a7Sbonwick spa_guid_exists(spa_guid(spa), guid)) 3060e34b6a7Sbonwick guid = spa_get_random(-1ULL); 3070e34b6a7Sbonwick } 3080e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 3090e34b6a7Sbonwick } 3100e34b6a7Sbonwick 311fa9e4066Sahrens vd->vdev_spa = spa; 312fa9e4066Sahrens vd->vdev_id = id; 313fa9e4066Sahrens vd->vdev_guid = guid; 314fa9e4066Sahrens vd->vdev_guid_sum = guid; 315fa9e4066Sahrens vd->vdev_ops = ops; 316fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 317fa9e4066Sahrens 318fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 3195ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 320fa9e4066Sahrens space_map_create(&vd->vdev_dtl_map, 0, -1ULL, 0, &vd->vdev_dtl_lock); 321fa9e4066Sahrens space_map_create(&vd->vdev_dtl_scrub, 0, -1ULL, 0, &vd->vdev_dtl_lock); 322fa9e4066Sahrens txg_list_create(&vd->vdev_ms_list, 323fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 324fa9e4066Sahrens txg_list_create(&vd->vdev_dtl_list, 325fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 326fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 3273d7072f8Seschrock vdev_queue_init(vd); 3283d7072f8Seschrock vdev_cache_init(vd); 329fa9e4066Sahrens 330fa9e4066Sahrens return (vd); 331fa9e4066Sahrens } 332fa9e4066Sahrens 333fa9e4066Sahrens /* 334fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 335fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 336fa9e4066Sahrens * different for each case. 337fa9e4066Sahrens */ 33899653d4eSeschrock int 33999653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 34099653d4eSeschrock int alloctype) 341fa9e4066Sahrens { 342fa9e4066Sahrens vdev_ops_t *ops; 343fa9e4066Sahrens char *type; 3448654d025Sperrin uint64_t guid = 0, islog, nparity; 345fa9e4066Sahrens vdev_t *vd; 346fa9e4066Sahrens 347fa9e4066Sahrens ASSERT(spa_config_held(spa, RW_WRITER)); 348fa9e4066Sahrens 349fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 35099653d4eSeschrock return (EINVAL); 351fa9e4066Sahrens 352fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 35399653d4eSeschrock return (EINVAL); 354fa9e4066Sahrens 355fa9e4066Sahrens /* 356fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 357fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 358fa9e4066Sahrens */ 359fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 360fa9e4066Sahrens uint64_t label_id; 361fa9e4066Sahrens 362fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 363fa9e4066Sahrens label_id != id) 36499653d4eSeschrock return (EINVAL); 365fa9e4066Sahrens 366fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 36799653d4eSeschrock return (EINVAL); 36899653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 36999653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 37099653d4eSeschrock return (EINVAL); 371fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 372fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 373fa94a07fSbrendan return (EINVAL); 374fa9e4066Sahrens } 375fa9e4066Sahrens 37699653d4eSeschrock /* 37799653d4eSeschrock * The first allocated vdev must be of type 'root'. 37899653d4eSeschrock */ 37999653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 38099653d4eSeschrock return (EINVAL); 38199653d4eSeschrock 3828654d025Sperrin /* 3838654d025Sperrin * Determine whether we're a log vdev. 3848654d025Sperrin */ 3858654d025Sperrin islog = 0; 3868654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 387990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 3888654d025Sperrin return (ENOTSUP); 389fa9e4066Sahrens 39099653d4eSeschrock /* 3918654d025Sperrin * Set the nparity property for RAID-Z vdevs. 39299653d4eSeschrock */ 3938654d025Sperrin nparity = -1ULL; 39499653d4eSeschrock if (ops == &vdev_raidz_ops) { 39599653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 3968654d025Sperrin &nparity) == 0) { 39799653d4eSeschrock /* 39899653d4eSeschrock * Currently, we can only support 2 parity devices. 39999653d4eSeschrock */ 4008654d025Sperrin if (nparity == 0 || nparity > 2) 40199653d4eSeschrock return (EINVAL); 40299653d4eSeschrock /* 40399653d4eSeschrock * Older versions can only support 1 parity device. 40499653d4eSeschrock */ 4058654d025Sperrin if (nparity == 2 && 406e7437265Sahrens spa_version(spa) < SPA_VERSION_RAID6) 40799653d4eSeschrock return (ENOTSUP); 40899653d4eSeschrock } else { 40999653d4eSeschrock /* 41099653d4eSeschrock * We require the parity to be specified for SPAs that 41199653d4eSeschrock * support multiple parity levels. 41299653d4eSeschrock */ 413e7437265Sahrens if (spa_version(spa) >= SPA_VERSION_RAID6) 41499653d4eSeschrock return (EINVAL); 41599653d4eSeschrock /* 41699653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 41799653d4eSeschrock */ 4188654d025Sperrin nparity = 1; 41999653d4eSeschrock } 42099653d4eSeschrock } else { 4218654d025Sperrin nparity = 0; 42299653d4eSeschrock } 4238654d025Sperrin ASSERT(nparity != -1ULL); 4248654d025Sperrin 4258654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 4268654d025Sperrin 4278654d025Sperrin vd->vdev_islog = islog; 4288654d025Sperrin vd->vdev_nparity = nparity; 4298654d025Sperrin 4308654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 4318654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 4328654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 4338654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 4348654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 4358654d025Sperrin &vd->vdev_physpath) == 0) 4368654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 43799653d4eSeschrock 438afefbcddSeschrock /* 439afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 440afefbcddSeschrock * as -1. 441afefbcddSeschrock */ 442afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 443afefbcddSeschrock &vd->vdev_wholedisk) != 0) 444afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 445afefbcddSeschrock 446ea8dc4b6Seschrock /* 447ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 448ea8dc4b6Seschrock * was not present at the time of import. 449ea8dc4b6Seschrock */ 450c5904d13Seschrock if (!spa->spa_import_faulted) 451c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 452c5904d13Seschrock &vd->vdev_not_present); 453ea8dc4b6Seschrock 454ecc2d604Sbonwick /* 455ecc2d604Sbonwick * Get the alignment requirement. 456ecc2d604Sbonwick */ 457ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 458ecc2d604Sbonwick 459fa9e4066Sahrens /* 460fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 461fa9e4066Sahrens */ 462fa9e4066Sahrens if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) { 463fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 464fa9e4066Sahrens &vd->vdev_ms_array); 465fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 466fa9e4066Sahrens &vd->vdev_ms_shift); 467fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 468fa9e4066Sahrens &vd->vdev_asize); 469fa9e4066Sahrens } 470fa9e4066Sahrens 471fa9e4066Sahrens /* 4723d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 473fa9e4066Sahrens */ 474c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 475c5904d13Seschrock (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE)) { 476c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 477c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 478c5904d13Seschrock &vd->vdev_dtl.smo_object); 479c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 480c5904d13Seschrock &vd->vdev_unspare); 481c5904d13Seschrock } 482ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 483ecc2d604Sbonwick &vd->vdev_offline); 484c5904d13Seschrock 4853d7072f8Seschrock /* 4863d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 4873d7072f8Seschrock * state, as the diagnosis made on another system may not be 4883d7072f8Seschrock * valid in the current context. 4893d7072f8Seschrock */ 4903d7072f8Seschrock if (spa->spa_load_state == SPA_LOAD_OPEN) { 4913d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 4923d7072f8Seschrock &vd->vdev_faulted); 4933d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 4943d7072f8Seschrock &vd->vdev_degraded); 4953d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 4963d7072f8Seschrock &vd->vdev_removed); 4973d7072f8Seschrock } 498fa9e4066Sahrens } 499fa9e4066Sahrens 500fa9e4066Sahrens /* 501fa9e4066Sahrens * Add ourselves to the parent's list of children. 502fa9e4066Sahrens */ 503fa9e4066Sahrens vdev_add_child(parent, vd); 504fa9e4066Sahrens 50599653d4eSeschrock *vdp = vd; 50699653d4eSeschrock 50799653d4eSeschrock return (0); 508fa9e4066Sahrens } 509fa9e4066Sahrens 510fa9e4066Sahrens void 511fa9e4066Sahrens vdev_free(vdev_t *vd) 512fa9e4066Sahrens { 513fa9e4066Sahrens int c; 5143d7072f8Seschrock spa_t *spa = vd->vdev_spa; 515fa9e4066Sahrens 516fa9e4066Sahrens /* 517fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 518fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 519fa9e4066Sahrens */ 520fa9e4066Sahrens vdev_close(vd); 521fa9e4066Sahrens 5223d7072f8Seschrock 523ecc2d604Sbonwick ASSERT(!list_link_active(&vd->vdev_dirty_node)); 524fa9e4066Sahrens 525fa9e4066Sahrens /* 526fa9e4066Sahrens * Free all children. 527fa9e4066Sahrens */ 528fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 529fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 530fa9e4066Sahrens 531fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 532fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 533fa9e4066Sahrens 534fa9e4066Sahrens /* 535fa9e4066Sahrens * Discard allocation state. 536fa9e4066Sahrens */ 537fa9e4066Sahrens if (vd == vd->vdev_top) 538fa9e4066Sahrens vdev_metaslab_fini(vd); 539fa9e4066Sahrens 540fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_space, ==, 0); 54199653d4eSeschrock ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0); 542fa9e4066Sahrens ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0); 543fa9e4066Sahrens 544fa9e4066Sahrens /* 545fa9e4066Sahrens * Remove this vdev from its parent's child list. 546fa9e4066Sahrens */ 547fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 548fa9e4066Sahrens 549fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 550fa9e4066Sahrens 5513d7072f8Seschrock /* 5523d7072f8Seschrock * Clean up vdev structure. 5533d7072f8Seschrock */ 5543d7072f8Seschrock vdev_queue_fini(vd); 5553d7072f8Seschrock vdev_cache_fini(vd); 5563d7072f8Seschrock 5573d7072f8Seschrock if (vd->vdev_path) 5583d7072f8Seschrock spa_strfree(vd->vdev_path); 5593d7072f8Seschrock if (vd->vdev_devid) 5603d7072f8Seschrock spa_strfree(vd->vdev_devid); 5613d7072f8Seschrock if (vd->vdev_physpath) 5623d7072f8Seschrock spa_strfree(vd->vdev_physpath); 5633d7072f8Seschrock 5643d7072f8Seschrock if (vd->vdev_isspare) 5653d7072f8Seschrock spa_spare_remove(vd); 566fa94a07fSbrendan if (vd->vdev_isl2cache) 567fa94a07fSbrendan spa_l2cache_remove(vd); 5683d7072f8Seschrock 5693d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 5703d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 5713d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 5723d7072f8Seschrock space_map_unload(&vd->vdev_dtl_map); 5733d7072f8Seschrock space_map_destroy(&vd->vdev_dtl_map); 5743d7072f8Seschrock space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL); 5753d7072f8Seschrock space_map_destroy(&vd->vdev_dtl_scrub); 5763d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 5773d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 5783d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 5793d7072f8Seschrock 5803d7072f8Seschrock if (vd == spa->spa_root_vdev) 5813d7072f8Seschrock spa->spa_root_vdev = NULL; 5823d7072f8Seschrock 5833d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 584fa9e4066Sahrens } 585fa9e4066Sahrens 586fa9e4066Sahrens /* 587fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 588fa9e4066Sahrens */ 589fa9e4066Sahrens static void 590fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 591fa9e4066Sahrens { 592fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 593fa9e4066Sahrens metaslab_t *msp; 594fa9e4066Sahrens vdev_t *vd; 595fa9e4066Sahrens int t; 596fa9e4066Sahrens 597fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 598fa9e4066Sahrens 599fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 600fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 601fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 602fa9e4066Sahrens 603fa9e4066Sahrens svd->vdev_ms_array = 0; 604fa9e4066Sahrens svd->vdev_ms_shift = 0; 605fa9e4066Sahrens svd->vdev_ms_count = 0; 606fa9e4066Sahrens 607fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 608fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 609fa9e4066Sahrens 610fa9e4066Sahrens svd->vdev_mg = NULL; 611fa9e4066Sahrens svd->vdev_ms = NULL; 612ecc2d604Sbonwick 613ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 614ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 615fa9e4066Sahrens 616fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 617fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 61899653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 619fa9e4066Sahrens 620fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 621fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 62299653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 623fa9e4066Sahrens 624fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 625fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 626fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 627fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 628fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 629fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 630fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 631fa9e4066Sahrens } 632fa9e4066Sahrens 633ecc2d604Sbonwick if (list_link_active(&svd->vdev_dirty_node)) { 634fa9e4066Sahrens vdev_config_clean(svd); 635fa9e4066Sahrens vdev_config_dirty(tvd); 636fa9e4066Sahrens } 637fa9e4066Sahrens 63899653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 63999653d4eSeschrock svd->vdev_deflate_ratio = 0; 6408654d025Sperrin 6418654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 6428654d025Sperrin svd->vdev_islog = 0; 643fa9e4066Sahrens } 644fa9e4066Sahrens 645fa9e4066Sahrens static void 646fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 647fa9e4066Sahrens { 648fa9e4066Sahrens int c; 649fa9e4066Sahrens 650fa9e4066Sahrens if (vd == NULL) 651fa9e4066Sahrens return; 652fa9e4066Sahrens 653fa9e4066Sahrens vd->vdev_top = tvd; 654fa9e4066Sahrens 655fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 656fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 657fa9e4066Sahrens } 658fa9e4066Sahrens 659fa9e4066Sahrens /* 660fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 661fa9e4066Sahrens */ 662fa9e4066Sahrens vdev_t * 663fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 664fa9e4066Sahrens { 665fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 666fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 667fa9e4066Sahrens vdev_t *mvd; 668fa9e4066Sahrens 669fa9e4066Sahrens ASSERT(spa_config_held(spa, RW_WRITER)); 670fa9e4066Sahrens 671fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 672ecc2d604Sbonwick 673ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 674ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 675ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 676ecc2d604Sbonwick 677fa9e4066Sahrens vdev_remove_child(pvd, cvd); 678fa9e4066Sahrens vdev_add_child(pvd, mvd); 679fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 680fa9e4066Sahrens vdev_add_child(mvd, cvd); 681fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 682fa9e4066Sahrens 683fa9e4066Sahrens if (mvd == mvd->vdev_top) 684fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 685fa9e4066Sahrens 686fa9e4066Sahrens return (mvd); 687fa9e4066Sahrens } 688fa9e4066Sahrens 689fa9e4066Sahrens /* 690fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 691fa9e4066Sahrens */ 692fa9e4066Sahrens void 693fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 694fa9e4066Sahrens { 695fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 696fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 697fa9e4066Sahrens 698fa9e4066Sahrens ASSERT(spa_config_held(cvd->vdev_spa, RW_WRITER)); 699fa9e4066Sahrens 700fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 701fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 70299653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 70399653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 704ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 705fa9e4066Sahrens 706fa9e4066Sahrens vdev_remove_child(mvd, cvd); 707fa9e4066Sahrens vdev_remove_child(pvd, mvd); 708fa9e4066Sahrens cvd->vdev_id = mvd->vdev_id; 709fa9e4066Sahrens vdev_add_child(pvd, cvd); 71099653d4eSeschrock /* 71199653d4eSeschrock * If we created a new toplevel vdev, then we need to change the child's 71299653d4eSeschrock * vdev GUID to match the old toplevel vdev. Otherwise, we could have 71399653d4eSeschrock * detached an offline device, and when we go to import the pool we'll 71499653d4eSeschrock * think we have two toplevel vdevs, instead of a different version of 71599653d4eSeschrock * the same toplevel vdev. 71699653d4eSeschrock */ 71799653d4eSeschrock if (cvd->vdev_top == cvd) { 71899653d4eSeschrock pvd->vdev_guid_sum -= cvd->vdev_guid; 71999653d4eSeschrock cvd->vdev_guid_sum -= cvd->vdev_guid; 72099653d4eSeschrock cvd->vdev_guid = mvd->vdev_guid; 72199653d4eSeschrock cvd->vdev_guid_sum += mvd->vdev_guid; 72299653d4eSeschrock pvd->vdev_guid_sum += cvd->vdev_guid; 72399653d4eSeschrock } 724fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 725fa9e4066Sahrens 726fa9e4066Sahrens if (cvd == cvd->vdev_top) 727fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 728fa9e4066Sahrens 729fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 730fa9e4066Sahrens vdev_free(mvd); 731fa9e4066Sahrens } 732fa9e4066Sahrens 733ea8dc4b6Seschrock int 734fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 735fa9e4066Sahrens { 736fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 737ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 7388654d025Sperrin metaslab_class_t *mc; 739ecc2d604Sbonwick uint64_t m; 740fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 741fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 742ecc2d604Sbonwick metaslab_t **mspp; 743ecc2d604Sbonwick int error; 744fa9e4066Sahrens 7450e34b6a7Sbonwick if (vd->vdev_ms_shift == 0) /* not being allocated from yet */ 7460e34b6a7Sbonwick return (0); 7470e34b6a7Sbonwick 748fa9e4066Sahrens dprintf("%s oldc %llu newc %llu\n", vdev_description(vd), oldc, newc); 749fa9e4066Sahrens 750fa9e4066Sahrens ASSERT(oldc <= newc); 751fa9e4066Sahrens 7528654d025Sperrin if (vd->vdev_islog) 7538654d025Sperrin mc = spa->spa_log_class; 7548654d025Sperrin else 7558654d025Sperrin mc = spa->spa_normal_class; 7568654d025Sperrin 757ecc2d604Sbonwick if (vd->vdev_mg == NULL) 758ecc2d604Sbonwick vd->vdev_mg = metaslab_group_create(mc, vd); 759fa9e4066Sahrens 760ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 761fa9e4066Sahrens 762ecc2d604Sbonwick if (oldc != 0) { 763ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 764ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 765ecc2d604Sbonwick } 766fa9e4066Sahrens 767ecc2d604Sbonwick vd->vdev_ms = mspp; 768ecc2d604Sbonwick vd->vdev_ms_count = newc; 769fa9e4066Sahrens 770ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 771ecc2d604Sbonwick space_map_obj_t smo = { 0, 0, 0 }; 772ecc2d604Sbonwick if (txg == 0) { 773ecc2d604Sbonwick uint64_t object = 0; 774ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 775ecc2d604Sbonwick m * sizeof (uint64_t), sizeof (uint64_t), &object); 776ecc2d604Sbonwick if (error) 777ecc2d604Sbonwick return (error); 778ecc2d604Sbonwick if (object != 0) { 779ecc2d604Sbonwick dmu_buf_t *db; 780ecc2d604Sbonwick error = dmu_bonus_hold(mos, object, FTAG, &db); 781ecc2d604Sbonwick if (error) 782ecc2d604Sbonwick return (error); 7831934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (smo)); 7841934e92fSmaybee bcopy(db->db_data, &smo, sizeof (smo)); 785ecc2d604Sbonwick ASSERT3U(smo.smo_object, ==, object); 786ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 787fa9e4066Sahrens } 788fa9e4066Sahrens } 789ecc2d604Sbonwick vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 790ecc2d604Sbonwick m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 791fa9e4066Sahrens } 792fa9e4066Sahrens 793ea8dc4b6Seschrock return (0); 794fa9e4066Sahrens } 795fa9e4066Sahrens 796fa9e4066Sahrens void 797fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 798fa9e4066Sahrens { 799fa9e4066Sahrens uint64_t m; 800fa9e4066Sahrens uint64_t count = vd->vdev_ms_count; 801fa9e4066Sahrens 802fa9e4066Sahrens if (vd->vdev_ms != NULL) { 803fa9e4066Sahrens for (m = 0; m < count; m++) 804ecc2d604Sbonwick if (vd->vdev_ms[m] != NULL) 805ecc2d604Sbonwick metaslab_fini(vd->vdev_ms[m]); 806fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 807fa9e4066Sahrens vd->vdev_ms = NULL; 808fa9e4066Sahrens } 809fa9e4066Sahrens } 810fa9e4066Sahrens 8110a4e9518Sgw int 8120a4e9518Sgw vdev_probe(vdev_t *vd) 8130a4e9518Sgw { 8140a4e9518Sgw if (vd == NULL) 8150a4e9518Sgw return (EINVAL); 8160a4e9518Sgw 8170a4e9518Sgw /* 8180a4e9518Sgw * Right now we only support status checks on the leaf vdevs. 8190a4e9518Sgw */ 8200a4e9518Sgw if (vd->vdev_ops->vdev_op_leaf) 8210a4e9518Sgw return (vd->vdev_ops->vdev_op_probe(vd)); 8220a4e9518Sgw 8230a4e9518Sgw return (0); 8240a4e9518Sgw } 8250a4e9518Sgw 826fa9e4066Sahrens /* 827fa9e4066Sahrens * Prepare a virtual device for access. 828fa9e4066Sahrens */ 829fa9e4066Sahrens int 830fa9e4066Sahrens vdev_open(vdev_t *vd) 831fa9e4066Sahrens { 832fa9e4066Sahrens int error; 833fa9e4066Sahrens int c; 834fa9e4066Sahrens uint64_t osize = 0; 835fa9e4066Sahrens uint64_t asize, psize; 836ecc2d604Sbonwick uint64_t ashift = 0; 837fa9e4066Sahrens 838fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 839fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 840fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 841fa9e4066Sahrens 842fa9e4066Sahrens if (vd->vdev_fault_mode == VDEV_FAULT_COUNT) 843fa9e4066Sahrens vd->vdev_fault_arg >>= 1; 844fa9e4066Sahrens else 845fa9e4066Sahrens vd->vdev_fault_mode = VDEV_FAULT_NONE; 846fa9e4066Sahrens 847fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 848fa9e4066Sahrens 8493d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 8503d7072f8Seschrock ASSERT(vd->vdev_children == 0); 8513d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 8523d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 8533d7072f8Seschrock return (ENXIO); 8543d7072f8Seschrock } else if (vd->vdev_offline) { 855fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 856ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 857fa9e4066Sahrens return (ENXIO); 858fa9e4066Sahrens } 859fa9e4066Sahrens 860fa9e4066Sahrens error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift); 861fa9e4066Sahrens 862ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 863ea8dc4b6Seschrock error = zio_handle_device_injection(vd, ENXIO); 864ea8dc4b6Seschrock 865fa9e4066Sahrens if (error) { 8663d7072f8Seschrock if (vd->vdev_removed && 8673d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 8683d7072f8Seschrock vd->vdev_removed = B_FALSE; 8693d7072f8Seschrock 870ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 871fa9e4066Sahrens vd->vdev_stat.vs_aux); 872fa9e4066Sahrens return (error); 873fa9e4066Sahrens } 874fa9e4066Sahrens 8753d7072f8Seschrock vd->vdev_removed = B_FALSE; 8763d7072f8Seschrock 8773d7072f8Seschrock if (vd->vdev_degraded) { 8783d7072f8Seschrock ASSERT(vd->vdev_children == 0); 8793d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 8803d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 8813d7072f8Seschrock } else { 8823d7072f8Seschrock vd->vdev_state = VDEV_STATE_HEALTHY; 8833d7072f8Seschrock } 884fa9e4066Sahrens 885fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 886ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 887ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 888ea8dc4b6Seschrock VDEV_AUX_NONE); 889ea8dc4b6Seschrock break; 890ea8dc4b6Seschrock } 891fa9e4066Sahrens 892fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 893fa9e4066Sahrens 894fa9e4066Sahrens if (vd->vdev_children == 0) { 895fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 896ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 897ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 898fa9e4066Sahrens return (EOVERFLOW); 899fa9e4066Sahrens } 900fa9e4066Sahrens psize = osize; 901fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 902fa9e4066Sahrens } else { 903ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 904fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 905ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 906ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 907fa9e4066Sahrens return (EOVERFLOW); 908fa9e4066Sahrens } 909fa9e4066Sahrens psize = 0; 910fa9e4066Sahrens asize = osize; 911fa9e4066Sahrens } 912fa9e4066Sahrens 913fa9e4066Sahrens vd->vdev_psize = psize; 914fa9e4066Sahrens 915fa9e4066Sahrens if (vd->vdev_asize == 0) { 916fa9e4066Sahrens /* 917fa9e4066Sahrens * This is the first-ever open, so use the computed values. 918ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 919fa9e4066Sahrens */ 920fa9e4066Sahrens vd->vdev_asize = asize; 921ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 922fa9e4066Sahrens } else { 923fa9e4066Sahrens /* 924fa9e4066Sahrens * Make sure the alignment requirement hasn't increased. 925fa9e4066Sahrens */ 926ecc2d604Sbonwick if (ashift > vd->vdev_top->vdev_ashift) { 927ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 928ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 929fa9e4066Sahrens return (EINVAL); 930fa9e4066Sahrens } 931fa9e4066Sahrens 932fa9e4066Sahrens /* 933fa9e4066Sahrens * Make sure the device hasn't shrunk. 934fa9e4066Sahrens */ 935fa9e4066Sahrens if (asize < vd->vdev_asize) { 936ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 937ea8dc4b6Seschrock VDEV_AUX_BAD_LABEL); 938fa9e4066Sahrens return (EINVAL); 939fa9e4066Sahrens } 940fa9e4066Sahrens 941fa9e4066Sahrens /* 942fa9e4066Sahrens * If all children are healthy and the asize has increased, 943fa9e4066Sahrens * then we've experienced dynamic LUN growth. 944fa9e4066Sahrens */ 945fa9e4066Sahrens if (vd->vdev_state == VDEV_STATE_HEALTHY && 946fa9e4066Sahrens asize > vd->vdev_asize) { 947fa9e4066Sahrens vd->vdev_asize = asize; 948fa9e4066Sahrens } 949fa9e4066Sahrens } 950fa9e4066Sahrens 9510a4e9518Sgw /* 9520a4e9518Sgw * Ensure we can issue some IO before declaring the 9530a4e9518Sgw * vdev open for business. 9540a4e9518Sgw */ 9550a4e9518Sgw error = vdev_probe(vd); 9560a4e9518Sgw if (error) { 9570a4e9518Sgw vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 9580a4e9518Sgw VDEV_AUX_OPEN_FAILED); 9590a4e9518Sgw return (error); 9600a4e9518Sgw } 9610a4e9518Sgw 96299653d4eSeschrock /* 96399653d4eSeschrock * If this is a top-level vdev, compute the raidz-deflation 96499653d4eSeschrock * ratio. Note, we hard-code in 128k (1<<17) because it is the 96599653d4eSeschrock * current "typical" blocksize. Even if SPA_MAXBLOCKSIZE 96699653d4eSeschrock * changes, this algorithm must never change, or we will 96799653d4eSeschrock * inconsistently account for existing bp's. 96899653d4eSeschrock */ 96999653d4eSeschrock if (vd->vdev_top == vd) { 97099653d4eSeschrock vd->vdev_deflate_ratio = (1<<17) / 97199653d4eSeschrock (vdev_psize_to_asize(vd, 1<<17) >> SPA_MINBLOCKSHIFT); 97299653d4eSeschrock } 97399653d4eSeschrock 974*088f3894Sahrens /* 975*088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 976*088f3894Sahrens * resilver. But don't do this if we are doing a reopen for a 977*088f3894Sahrens * scrub, since this would just restart the scrub we are already 978*088f3894Sahrens * doing. 979*088f3894Sahrens */ 980*088f3894Sahrens if (vd->vdev_children == 0 && !vd->vdev_spa->spa_scrub_reopen) { 981*088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 982*088f3894Sahrens if (vd->vdev_dtl_map.sm_space != 0 && vdev_writeable(vd)) 983*088f3894Sahrens spa_async_request(vd->vdev_spa, SPA_ASYNC_RESILVER); 984*088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 985*088f3894Sahrens } 986*088f3894Sahrens 987fa9e4066Sahrens return (0); 988fa9e4066Sahrens } 989fa9e4066Sahrens 990560e6e96Seschrock /* 991560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 992560e6e96Seschrock * contents. This needs to be done before vdev_load() so that we don't 9933d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 994560e6e96Seschrock * 995560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 996560e6e96Seschrock * has since been destroyed or exported. This is only possible if 997560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 998560e6e96Seschrock * will be updated but the function will return 0. 999560e6e96Seschrock */ 1000560e6e96Seschrock int 1001560e6e96Seschrock vdev_validate(vdev_t *vd) 1002560e6e96Seschrock { 1003560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1004560e6e96Seschrock int c; 1005560e6e96Seschrock nvlist_t *label; 1006560e6e96Seschrock uint64_t guid; 1007560e6e96Seschrock uint64_t state; 1008560e6e96Seschrock 1009560e6e96Seschrock for (c = 0; c < vd->vdev_children; c++) 1010560e6e96Seschrock if (vdev_validate(vd->vdev_child[c]) != 0) 10110bf246f5Smc return (EBADF); 1012560e6e96Seschrock 1013b5989ec7Seschrock /* 1014b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1015b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1016b5989ec7Seschrock * overwrite the previous state. 1017b5989ec7Seschrock */ 1018b5989ec7Seschrock if (vd->vdev_ops->vdev_op_leaf && !vdev_is_dead(vd)) { 1019560e6e96Seschrock 1020560e6e96Seschrock if ((label = vdev_label_read_config(vd)) == NULL) { 1021560e6e96Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1022560e6e96Seschrock VDEV_AUX_BAD_LABEL); 1023560e6e96Seschrock return (0); 1024560e6e96Seschrock } 1025560e6e96Seschrock 1026560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 1027560e6e96Seschrock &guid) != 0 || guid != spa_guid(spa)) { 1028560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1029560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1030560e6e96Seschrock nvlist_free(label); 1031560e6e96Seschrock return (0); 1032560e6e96Seschrock } 1033560e6e96Seschrock 1034560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1035560e6e96Seschrock &guid) != 0 || guid != vd->vdev_guid) { 1036560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1037560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1038560e6e96Seschrock nvlist_free(label); 1039560e6e96Seschrock return (0); 1040560e6e96Seschrock } 1041560e6e96Seschrock 1042560e6e96Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1043560e6e96Seschrock &state) != 0) { 1044560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1045560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1046560e6e96Seschrock nvlist_free(label); 1047560e6e96Seschrock return (0); 1048560e6e96Seschrock } 1049560e6e96Seschrock 1050560e6e96Seschrock nvlist_free(label); 1051560e6e96Seschrock 1052560e6e96Seschrock if (spa->spa_load_state == SPA_LOAD_OPEN && 1053560e6e96Seschrock state != POOL_STATE_ACTIVE) 10540bf246f5Smc return (EBADF); 1055560e6e96Seschrock 105651ece835Seschrock /* 105751ece835Seschrock * If we were able to open and validate a vdev that was 105851ece835Seschrock * previously marked permanently unavailable, clear that state 105951ece835Seschrock * now. 106051ece835Seschrock */ 106151ece835Seschrock if (vd->vdev_not_present) 106251ece835Seschrock vd->vdev_not_present = 0; 106351ece835Seschrock } 1064560e6e96Seschrock 1065560e6e96Seschrock return (0); 1066560e6e96Seschrock } 1067560e6e96Seschrock 1068fa9e4066Sahrens /* 1069fa9e4066Sahrens * Close a virtual device. 1070fa9e4066Sahrens */ 1071fa9e4066Sahrens void 1072fa9e4066Sahrens vdev_close(vdev_t *vd) 1073fa9e4066Sahrens { 1074fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1075fa9e4066Sahrens 10763d7072f8Seschrock vdev_cache_purge(vd); 1077fa9e4066Sahrens 1078560e6e96Seschrock /* 1079560e6e96Seschrock * We record the previous state before we close it, so that if we are 1080560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1081560e6e96Seschrock * it's still faulted. 1082560e6e96Seschrock */ 1083560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1084560e6e96Seschrock 1085fa9e4066Sahrens if (vd->vdev_offline) 1086fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1087fa9e4066Sahrens else 1088fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1089ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1090fa9e4066Sahrens } 1091fa9e4066Sahrens 1092fa9e4066Sahrens void 1093ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1094fa9e4066Sahrens { 1095ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1096fa9e4066Sahrens 1097ea8dc4b6Seschrock ASSERT(spa_config_held(spa, RW_WRITER)); 1098ea8dc4b6Seschrock 1099fa9e4066Sahrens vdev_close(vd); 1100fa9e4066Sahrens (void) vdev_open(vd); 1101fa9e4066Sahrens 110239c23413Seschrock /* 110339c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 110439c23413Seschrock * Otherwise, a device with an invalid label could be successfully 110539c23413Seschrock * opened in response to vdev_reopen(). 110639c23413Seschrock */ 1107c5904d13Seschrock if (vd->vdev_aux) { 1108c5904d13Seschrock (void) vdev_validate_aux(vd); 1109c5904d13Seschrock if (!vdev_is_dead(vd) && 1110c5904d13Seschrock !l2arc_vdev_present(vd)) { 1111c5904d13Seschrock uint64_t size = vdev_get_rsize(vd); 1112c5904d13Seschrock l2arc_add_vdev(spa, vd, 1113c5904d13Seschrock VDEV_LABEL_START_SIZE, 1114c5904d13Seschrock size - VDEV_LABEL_START_SIZE); 1115c5904d13Seschrock } 1116c5904d13Seschrock } else { 1117c5904d13Seschrock (void) vdev_validate(vd); 1118c5904d13Seschrock } 111939c23413Seschrock 1120fa9e4066Sahrens /* 11213d7072f8Seschrock * Reassess parent vdev's health. 1122fa9e4066Sahrens */ 11233d7072f8Seschrock vdev_propagate_state(vd); 1124fa9e4066Sahrens } 1125fa9e4066Sahrens 1126fa9e4066Sahrens int 112799653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1128fa9e4066Sahrens { 1129fa9e4066Sahrens int error; 1130fa9e4066Sahrens 1131fa9e4066Sahrens /* 1132fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1133fa9e4066Sahrens * For a create, however, we want to fail the request if 1134fa9e4066Sahrens * there are any components we can't open. 1135fa9e4066Sahrens */ 1136fa9e4066Sahrens error = vdev_open(vd); 1137fa9e4066Sahrens 1138fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1139fa9e4066Sahrens vdev_close(vd); 1140fa9e4066Sahrens return (error ? error : ENXIO); 1141fa9e4066Sahrens } 1142fa9e4066Sahrens 1143fa9e4066Sahrens /* 1144fa9e4066Sahrens * Recursively initialize all labels. 1145fa9e4066Sahrens */ 114639c23413Seschrock if ((error = vdev_label_init(vd, txg, isreplacing ? 114739c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1148fa9e4066Sahrens vdev_close(vd); 1149fa9e4066Sahrens return (error); 1150fa9e4066Sahrens } 1151fa9e4066Sahrens 1152fa9e4066Sahrens return (0); 1153fa9e4066Sahrens } 1154fa9e4066Sahrens 1155fa9e4066Sahrens /* 1156fa9e4066Sahrens * The is the latter half of vdev_create(). It is distinct because it 1157fa9e4066Sahrens * involves initiating transactions in order to do metaslab creation. 1158fa9e4066Sahrens * For creation, we want to try to create all vdevs at once and then undo it 1159fa9e4066Sahrens * if anything fails; this is much harder if we have pending transactions. 1160fa9e4066Sahrens */ 11610e34b6a7Sbonwick void 1162fa9e4066Sahrens vdev_init(vdev_t *vd, uint64_t txg) 1163fa9e4066Sahrens { 1164fa9e4066Sahrens /* 1165fa9e4066Sahrens * Aim for roughly 200 metaslabs per vdev. 1166fa9e4066Sahrens */ 1167fa9e4066Sahrens vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1168fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1169fa9e4066Sahrens 1170fa9e4066Sahrens /* 11710e34b6a7Sbonwick * Initialize the vdev's metaslabs. This can't fail because 11720e34b6a7Sbonwick * there's nothing to read when creating all new metaslabs. 1173fa9e4066Sahrens */ 11740e34b6a7Sbonwick VERIFY(vdev_metaslab_init(vd, txg) == 0); 1175fa9e4066Sahrens } 1176fa9e4066Sahrens 1177fa9e4066Sahrens void 1178ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1179fa9e4066Sahrens { 1180ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 1181ecc2d604Sbonwick ASSERT(ISP2(flags)); 1182fa9e4066Sahrens 1183ecc2d604Sbonwick if (flags & VDD_METASLAB) 1184ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1185ecc2d604Sbonwick 1186ecc2d604Sbonwick if (flags & VDD_DTL) 1187ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1188ecc2d604Sbonwick 1189ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1190fa9e4066Sahrens } 1191fa9e4066Sahrens 1192fa9e4066Sahrens void 1193fa9e4066Sahrens vdev_dtl_dirty(space_map_t *sm, uint64_t txg, uint64_t size) 1194fa9e4066Sahrens { 1195fa9e4066Sahrens mutex_enter(sm->sm_lock); 1196fa9e4066Sahrens if (!space_map_contains(sm, txg, size)) 1197fa9e4066Sahrens space_map_add(sm, txg, size); 1198fa9e4066Sahrens mutex_exit(sm->sm_lock); 1199fa9e4066Sahrens } 1200fa9e4066Sahrens 1201fa9e4066Sahrens int 1202fa9e4066Sahrens vdev_dtl_contains(space_map_t *sm, uint64_t txg, uint64_t size) 1203fa9e4066Sahrens { 1204fa9e4066Sahrens int dirty; 1205fa9e4066Sahrens 1206fa9e4066Sahrens /* 1207fa9e4066Sahrens * Quick test without the lock -- covers the common case that 1208fa9e4066Sahrens * there are no dirty time segments. 1209fa9e4066Sahrens */ 1210fa9e4066Sahrens if (sm->sm_space == 0) 1211fa9e4066Sahrens return (0); 1212fa9e4066Sahrens 1213fa9e4066Sahrens mutex_enter(sm->sm_lock); 1214fa9e4066Sahrens dirty = space_map_contains(sm, txg, size); 1215fa9e4066Sahrens mutex_exit(sm->sm_lock); 1216fa9e4066Sahrens 1217fa9e4066Sahrens return (dirty); 1218fa9e4066Sahrens } 1219fa9e4066Sahrens 1220fa9e4066Sahrens /* 1221fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 1222fa9e4066Sahrens */ 1223fa9e4066Sahrens void 1224fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1225fa9e4066Sahrens { 1226ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1227fa9e4066Sahrens int c; 1228fa9e4066Sahrens 1229*088f3894Sahrens ASSERT(spa_config_held(spa, RW_READER)); 1230fa9e4066Sahrens 1231fa9e4066Sahrens if (vd->vdev_children == 0) { 1232fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1233*088f3894Sahrens if (scrub_txg != 0 && 1234*088f3894Sahrens (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) { 1235*088f3894Sahrens /* XXX should check scrub_done? */ 1236*088f3894Sahrens /* 1237*088f3894Sahrens * We completed a scrub up to scrub_txg. If we 1238*088f3894Sahrens * did it without rebooting, then the scrub dtl 1239*088f3894Sahrens * will be valid, so excise the old region and 1240*088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 1241*088f3894Sahrens * dtl as-is if there was an error. 1242*088f3894Sahrens */ 1243fa9e4066Sahrens space_map_excise(&vd->vdev_dtl_map, 0, scrub_txg); 1244fa9e4066Sahrens space_map_union(&vd->vdev_dtl_map, &vd->vdev_dtl_scrub); 1245fa9e4066Sahrens } 1246fa9e4066Sahrens if (scrub_done) 1247fa9e4066Sahrens space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL); 1248fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1249*088f3894Sahrens 1250ecc2d604Sbonwick if (txg != 0) 1251ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1252fa9e4066Sahrens return; 1253fa9e4066Sahrens } 1254fa9e4066Sahrens 1255ea8dc4b6Seschrock /* 1256ea8dc4b6Seschrock * Make sure the DTLs are always correct under the scrub lock. 1257ea8dc4b6Seschrock */ 1258ea8dc4b6Seschrock if (vd == spa->spa_root_vdev) 1259ea8dc4b6Seschrock mutex_enter(&spa->spa_scrub_lock); 1260ea8dc4b6Seschrock 1261fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1262fa9e4066Sahrens space_map_vacate(&vd->vdev_dtl_map, NULL, NULL); 1263fa9e4066Sahrens space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL); 1264fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1265fa9e4066Sahrens 1266fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) { 1267fa9e4066Sahrens vdev_t *cvd = vd->vdev_child[c]; 1268fa9e4066Sahrens vdev_dtl_reassess(cvd, txg, scrub_txg, scrub_done); 1269fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1270fa9e4066Sahrens space_map_union(&vd->vdev_dtl_map, &cvd->vdev_dtl_map); 1271fa9e4066Sahrens space_map_union(&vd->vdev_dtl_scrub, &cvd->vdev_dtl_scrub); 1272fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1273fa9e4066Sahrens } 1274ea8dc4b6Seschrock 1275ea8dc4b6Seschrock if (vd == spa->spa_root_vdev) 1276ea8dc4b6Seschrock mutex_exit(&spa->spa_scrub_lock); 1277fa9e4066Sahrens } 1278fa9e4066Sahrens 1279fa9e4066Sahrens static int 1280fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 1281fa9e4066Sahrens { 1282fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1283fa9e4066Sahrens space_map_obj_t *smo = &vd->vdev_dtl; 1284ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1285fa9e4066Sahrens dmu_buf_t *db; 1286fa9e4066Sahrens int error; 1287fa9e4066Sahrens 1288fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1289fa9e4066Sahrens 1290fa9e4066Sahrens if (smo->smo_object == 0) 1291fa9e4066Sahrens return (0); 1292fa9e4066Sahrens 1293ecc2d604Sbonwick if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1294ea8dc4b6Seschrock return (error); 1295ecc2d604Sbonwick 12961934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 12971934e92fSmaybee bcopy(db->db_data, smo, sizeof (*smo)); 1298ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1299fa9e4066Sahrens 1300fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1301ecc2d604Sbonwick error = space_map_load(&vd->vdev_dtl_map, NULL, SM_ALLOC, smo, mos); 1302fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1303fa9e4066Sahrens 1304fa9e4066Sahrens return (error); 1305fa9e4066Sahrens } 1306fa9e4066Sahrens 1307fa9e4066Sahrens void 1308fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1309fa9e4066Sahrens { 1310fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1311fa9e4066Sahrens space_map_obj_t *smo = &vd->vdev_dtl; 1312fa9e4066Sahrens space_map_t *sm = &vd->vdev_dtl_map; 1313ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1314fa9e4066Sahrens space_map_t smsync; 1315fa9e4066Sahrens kmutex_t smlock; 1316fa9e4066Sahrens dmu_buf_t *db; 1317fa9e4066Sahrens dmu_tx_t *tx; 1318fa9e4066Sahrens 1319fa9e4066Sahrens dprintf("%s in txg %llu pass %d\n", 1320fa9e4066Sahrens vdev_description(vd), (u_longlong_t)txg, spa_sync_pass(spa)); 1321fa9e4066Sahrens 1322fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1323fa9e4066Sahrens 1324fa9e4066Sahrens if (vd->vdev_detached) { 1325fa9e4066Sahrens if (smo->smo_object != 0) { 1326ecc2d604Sbonwick int err = dmu_object_free(mos, smo->smo_object, tx); 1327fa9e4066Sahrens ASSERT3U(err, ==, 0); 1328fa9e4066Sahrens smo->smo_object = 0; 1329fa9e4066Sahrens } 1330fa9e4066Sahrens dmu_tx_commit(tx); 1331ecc2d604Sbonwick dprintf("detach %s committed in txg %llu\n", 1332ecc2d604Sbonwick vdev_description(vd), txg); 1333fa9e4066Sahrens return; 1334fa9e4066Sahrens } 1335fa9e4066Sahrens 1336fa9e4066Sahrens if (smo->smo_object == 0) { 1337fa9e4066Sahrens ASSERT(smo->smo_objsize == 0); 1338fa9e4066Sahrens ASSERT(smo->smo_alloc == 0); 1339ecc2d604Sbonwick smo->smo_object = dmu_object_alloc(mos, 1340fa9e4066Sahrens DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1341fa9e4066Sahrens DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1342fa9e4066Sahrens ASSERT(smo->smo_object != 0); 1343fa9e4066Sahrens vdev_config_dirty(vd->vdev_top); 1344fa9e4066Sahrens } 1345fa9e4066Sahrens 1346fa9e4066Sahrens mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1347fa9e4066Sahrens 1348fa9e4066Sahrens space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1349fa9e4066Sahrens &smlock); 1350fa9e4066Sahrens 1351fa9e4066Sahrens mutex_enter(&smlock); 1352fa9e4066Sahrens 1353fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 1354ecc2d604Sbonwick space_map_walk(sm, space_map_add, &smsync); 1355fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 1356fa9e4066Sahrens 1357ecc2d604Sbonwick space_map_truncate(smo, mos, tx); 1358ecc2d604Sbonwick space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1359fa9e4066Sahrens 1360fa9e4066Sahrens space_map_destroy(&smsync); 1361fa9e4066Sahrens 1362fa9e4066Sahrens mutex_exit(&smlock); 1363fa9e4066Sahrens mutex_destroy(&smlock); 1364fa9e4066Sahrens 1365ecc2d604Sbonwick VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1366fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 13671934e92fSmaybee ASSERT3U(db->db_size, >=, sizeof (*smo)); 13681934e92fSmaybee bcopy(smo, db->db_data, sizeof (*smo)); 1369ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 1370fa9e4066Sahrens 1371fa9e4066Sahrens dmu_tx_commit(tx); 1372fa9e4066Sahrens } 1373fa9e4066Sahrens 1374*088f3894Sahrens /* 1375*088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 1376*088f3894Sahrens */ 1377*088f3894Sahrens boolean_t 1378*088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1379*088f3894Sahrens { 1380*088f3894Sahrens boolean_t needed = B_FALSE; 1381*088f3894Sahrens uint64_t thismin = UINT64_MAX; 1382*088f3894Sahrens uint64_t thismax = 0; 1383*088f3894Sahrens 1384*088f3894Sahrens if (vd->vdev_children == 0) { 1385*088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 1386*088f3894Sahrens if (vd->vdev_dtl_map.sm_space != 0 && vdev_writeable(vd)) { 1387*088f3894Sahrens space_seg_t *ss; 1388*088f3894Sahrens 1389*088f3894Sahrens ss = avl_first(&vd->vdev_dtl_map.sm_root); 1390*088f3894Sahrens thismin = ss->ss_start - 1; 1391*088f3894Sahrens ss = avl_last(&vd->vdev_dtl_map.sm_root); 1392*088f3894Sahrens thismax = ss->ss_end; 1393*088f3894Sahrens needed = B_TRUE; 1394*088f3894Sahrens } 1395*088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 1396*088f3894Sahrens } else { 1397*088f3894Sahrens int c; 1398*088f3894Sahrens for (c = 0; c < vd->vdev_children; c++) { 1399*088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 1400*088f3894Sahrens uint64_t cmin, cmax; 1401*088f3894Sahrens 1402*088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1403*088f3894Sahrens thismin = MIN(thismin, cmin); 1404*088f3894Sahrens thismax = MAX(thismax, cmax); 1405*088f3894Sahrens needed = B_TRUE; 1406*088f3894Sahrens } 1407*088f3894Sahrens } 1408*088f3894Sahrens } 1409*088f3894Sahrens 1410*088f3894Sahrens if (needed && minp) { 1411*088f3894Sahrens *minp = thismin; 1412*088f3894Sahrens *maxp = thismax; 1413*088f3894Sahrens } 1414*088f3894Sahrens return (needed); 1415*088f3894Sahrens } 1416*088f3894Sahrens 1417560e6e96Seschrock void 1418ea8dc4b6Seschrock vdev_load(vdev_t *vd) 1419fa9e4066Sahrens { 1420560e6e96Seschrock int c; 1421fa9e4066Sahrens 1422fa9e4066Sahrens /* 1423fa9e4066Sahrens * Recursively load all children. 1424fa9e4066Sahrens */ 1425fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 1426560e6e96Seschrock vdev_load(vd->vdev_child[c]); 1427fa9e4066Sahrens 1428fa9e4066Sahrens /* 14290e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 1430fa9e4066Sahrens */ 1431560e6e96Seschrock if (vd == vd->vdev_top && 1432560e6e96Seschrock (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1433560e6e96Seschrock vdev_metaslab_init(vd, 0) != 0)) 1434560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1435560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1436fa9e4066Sahrens 1437fa9e4066Sahrens /* 1438fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 1439fa9e4066Sahrens */ 1440560e6e96Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1441560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1442560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1443fa9e4066Sahrens } 1444fa9e4066Sahrens 144599653d4eSeschrock /* 1446fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 1447fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 1448fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 1449fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 1450fa94a07fSbrendan * repurposed to another pool. 145199653d4eSeschrock */ 145299653d4eSeschrock int 1453fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 145499653d4eSeschrock { 145599653d4eSeschrock nvlist_t *label; 145699653d4eSeschrock uint64_t guid, version; 145799653d4eSeschrock uint64_t state; 145899653d4eSeschrock 1459c5904d13Seschrock if (vdev_is_dead(vd)) 1460c5904d13Seschrock return (0); 1461c5904d13Seschrock 146299653d4eSeschrock if ((label = vdev_label_read_config(vd)) == NULL) { 146399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 146499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 146599653d4eSeschrock return (-1); 146699653d4eSeschrock } 146799653d4eSeschrock 146899653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1469e7437265Sahrens version > SPA_VERSION || 147099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 147199653d4eSeschrock guid != vd->vdev_guid || 147299653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 147399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 147499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 147599653d4eSeschrock nvlist_free(label); 147699653d4eSeschrock return (-1); 147799653d4eSeschrock } 147899653d4eSeschrock 147999653d4eSeschrock /* 148099653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 148199653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 148299653d4eSeschrock */ 148399653d4eSeschrock nvlist_free(label); 148499653d4eSeschrock return (0); 148599653d4eSeschrock } 148699653d4eSeschrock 1487fa9e4066Sahrens void 1488fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 1489fa9e4066Sahrens { 1490fa9e4066Sahrens metaslab_t *msp; 1491fa9e4066Sahrens 1492fa9e4066Sahrens dprintf("%s txg %llu\n", vdev_description(vd), txg); 1493fa9e4066Sahrens 1494fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 1495fa9e4066Sahrens metaslab_sync_done(msp, txg); 1496fa9e4066Sahrens } 1497fa9e4066Sahrens 1498fa9e4066Sahrens void 1499fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 1500fa9e4066Sahrens { 1501fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1502fa9e4066Sahrens vdev_t *lvd; 1503fa9e4066Sahrens metaslab_t *msp; 1504ecc2d604Sbonwick dmu_tx_t *tx; 1505fa9e4066Sahrens 1506fa9e4066Sahrens dprintf("%s txg %llu pass %d\n", 1507fa9e4066Sahrens vdev_description(vd), (u_longlong_t)txg, spa_sync_pass(spa)); 1508fa9e4066Sahrens 1509ecc2d604Sbonwick if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 1510ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 1511ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1512ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 1513ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 1514ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 1515ecc2d604Sbonwick vdev_config_dirty(vd); 1516ecc2d604Sbonwick dmu_tx_commit(tx); 1517ecc2d604Sbonwick } 1518fa9e4066Sahrens 1519ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 1520fa9e4066Sahrens metaslab_sync(msp, txg); 1521ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 1522ecc2d604Sbonwick } 1523fa9e4066Sahrens 1524fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 1525fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 1526fa9e4066Sahrens 1527fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 1528fa9e4066Sahrens } 1529fa9e4066Sahrens 1530fa9e4066Sahrens uint64_t 1531fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 1532fa9e4066Sahrens { 1533fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 1534fa9e4066Sahrens } 1535fa9e4066Sahrens 1536fa9e4066Sahrens const char * 1537fa9e4066Sahrens vdev_description(vdev_t *vd) 1538fa9e4066Sahrens { 1539fa9e4066Sahrens if (vd == NULL || vd->vdev_ops == NULL) 1540fa9e4066Sahrens return ("<unknown>"); 1541fa9e4066Sahrens 1542fa9e4066Sahrens if (vd->vdev_path != NULL) 1543fa9e4066Sahrens return (vd->vdev_path); 1544fa9e4066Sahrens 1545fa9e4066Sahrens if (vd->vdev_parent == NULL) 1546fa9e4066Sahrens return (spa_name(vd->vdev_spa)); 1547fa9e4066Sahrens 1548fa9e4066Sahrens return (vd->vdev_ops->vdev_op_type); 1549fa9e4066Sahrens } 1550fa9e4066Sahrens 15513d7072f8Seschrock /* 15523d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 15533d7072f8Seschrock * not be opened, and no I/O is attempted. 15543d7072f8Seschrock */ 1555fa9e4066Sahrens int 15563d7072f8Seschrock vdev_fault(spa_t *spa, uint64_t guid) 1557fa9e4066Sahrens { 1558c5904d13Seschrock vdev_t *vd; 1559441d80aaSlling uint64_t txg; 1560fa9e4066Sahrens 15610a4e9518Sgw /* 15620a4e9518Sgw * Disregard a vdev fault request if the pool has 15630a4e9518Sgw * experienced a complete failure. 15640a4e9518Sgw * 15650a4e9518Sgw * XXX - We do this here so that we don't hold the 15660a4e9518Sgw * spa_namespace_lock in the event that we can't get 15670a4e9518Sgw * the RW_WRITER spa_config_lock. 15680a4e9518Sgw */ 15690a4e9518Sgw if (spa_state(spa) == POOL_STATE_IO_FAILURE) 15700a4e9518Sgw return (EIO); 15710a4e9518Sgw 1572441d80aaSlling txg = spa_vdev_enter(spa); 1573fa9e4066Sahrens 1574c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1575441d80aaSlling return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 15763d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 15773d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 1578fa9e4066Sahrens 15793d7072f8Seschrock /* 15803d7072f8Seschrock * Faulted state takes precedence over degraded. 15813d7072f8Seschrock */ 15823d7072f8Seschrock vd->vdev_faulted = 1ULL; 15833d7072f8Seschrock vd->vdev_degraded = 0ULL; 15843d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, 15853d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 15863d7072f8Seschrock 15873d7072f8Seschrock /* 15883d7072f8Seschrock * If marking the vdev as faulted cause the toplevel vdev to become 15893d7072f8Seschrock * unavailable, then back off and simply mark the vdev as degraded 15903d7072f8Seschrock * instead. 15913d7072f8Seschrock */ 1592c5904d13Seschrock if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) { 15933d7072f8Seschrock vd->vdev_degraded = 1ULL; 15943d7072f8Seschrock vd->vdev_faulted = 0ULL; 15953d7072f8Seschrock 15963d7072f8Seschrock /* 15973d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 15983d7072f8Seschrock * mark it degraded. 15993d7072f8Seschrock */ 16003d7072f8Seschrock vdev_reopen(vd); 16013d7072f8Seschrock 16020a4e9518Sgw if (vdev_readable(vd)) { 16033d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 16043d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 16053d7072f8Seschrock } 16063d7072f8Seschrock } 16073d7072f8Seschrock 16083d7072f8Seschrock vdev_config_dirty(vd->vdev_top); 16093d7072f8Seschrock 16103d7072f8Seschrock (void) spa_vdev_exit(spa, NULL, txg, 0); 16113d7072f8Seschrock 16123d7072f8Seschrock return (0); 16133d7072f8Seschrock } 16143d7072f8Seschrock 16153d7072f8Seschrock /* 16163d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 16173d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 16183d7072f8Seschrock * as I/O is concerned. 16193d7072f8Seschrock */ 16203d7072f8Seschrock int 16213d7072f8Seschrock vdev_degrade(spa_t *spa, uint64_t guid) 16223d7072f8Seschrock { 1623c5904d13Seschrock vdev_t *vd; 16243d7072f8Seschrock uint64_t txg; 16253d7072f8Seschrock 16260a4e9518Sgw /* 16270a4e9518Sgw * Disregard a vdev fault request if the pool has 16280a4e9518Sgw * experienced a complete failure. 16290a4e9518Sgw * 16300a4e9518Sgw * XXX - We do this here so that we don't hold the 16310a4e9518Sgw * spa_namespace_lock in the event that we can't get 16320a4e9518Sgw * the RW_WRITER spa_config_lock. 16330a4e9518Sgw */ 16340a4e9518Sgw if (spa_state(spa) == POOL_STATE_IO_FAILURE) 16350a4e9518Sgw return (EIO); 16360a4e9518Sgw 16373d7072f8Seschrock txg = spa_vdev_enter(spa); 16383d7072f8Seschrock 1639c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 16403d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 16410e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 16420e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 16430e34b6a7Sbonwick 16443d7072f8Seschrock /* 16453d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 16463d7072f8Seschrock */ 16473d7072f8Seschrock if (vd->vdev_faulted || vd->vdev_degraded) { 16483d7072f8Seschrock (void) spa_vdev_exit(spa, NULL, txg, 0); 16493d7072f8Seschrock return (0); 16503d7072f8Seschrock } 16513d7072f8Seschrock 16523d7072f8Seschrock vd->vdev_degraded = 1ULL; 16533d7072f8Seschrock if (!vdev_is_dead(vd)) 16543d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 16553d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 16563d7072f8Seschrock vdev_config_dirty(vd->vdev_top); 16573d7072f8Seschrock 16583d7072f8Seschrock (void) spa_vdev_exit(spa, NULL, txg, 0); 16593d7072f8Seschrock 16603d7072f8Seschrock return (0); 16613d7072f8Seschrock } 16623d7072f8Seschrock 16633d7072f8Seschrock /* 16643d7072f8Seschrock * Online the given vdev. If 'unspare' is set, it implies two things. First, 16653d7072f8Seschrock * any attached spare device should be detached when the device finishes 16663d7072f8Seschrock * resilvering. Second, the online should be treated like a 'test' online case, 16673d7072f8Seschrock * so no FMA events are generated if the device fails to open. 16683d7072f8Seschrock */ 16693d7072f8Seschrock int 16703d7072f8Seschrock vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, 16713d7072f8Seschrock vdev_state_t *newstate) 16723d7072f8Seschrock { 1673c5904d13Seschrock vdev_t *vd; 16743d7072f8Seschrock uint64_t txg; 16753d7072f8Seschrock 16760a4e9518Sgw /* 16770a4e9518Sgw * Disregard a vdev fault request if the pool has 16780a4e9518Sgw * experienced a complete failure. 16790a4e9518Sgw * 16800a4e9518Sgw * XXX - We do this here so that we don't hold the 16810a4e9518Sgw * spa_namespace_lock in the event that we can't get 16820a4e9518Sgw * the RW_WRITER spa_config_lock. 16830a4e9518Sgw */ 16840a4e9518Sgw if (spa_state(spa) == POOL_STATE_IO_FAILURE) 16850a4e9518Sgw return (EIO); 16860a4e9518Sgw 16873d7072f8Seschrock txg = spa_vdev_enter(spa); 16883d7072f8Seschrock 1689c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 16903d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 16913d7072f8Seschrock 16923d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 16933d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 1694fa9e4066Sahrens 1695fa9e4066Sahrens vd->vdev_offline = B_FALSE; 1696441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 16973d7072f8Seschrock vd->vdev_checkremove = (flags & ZFS_ONLINE_CHECKREMOVE) ? 16983d7072f8Seschrock B_TRUE : B_FALSE; 16993d7072f8Seschrock vd->vdev_forcefault = (flags & ZFS_ONLINE_FORCEFAULT) ? 17003d7072f8Seschrock B_TRUE : B_FALSE; 1701ea8dc4b6Seschrock vdev_reopen(vd->vdev_top); 17023d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 17033d7072f8Seschrock 17043d7072f8Seschrock if (newstate) 17053d7072f8Seschrock *newstate = vd->vdev_state; 17063d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 17073d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 17083d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 17093d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 17103d7072f8Seschrock vd->vdev_unspare = B_TRUE; 1711fa9e4066Sahrens 1712441d80aaSlling vdev_config_dirty(vd->vdev_top); 1713441d80aaSlling 1714441d80aaSlling (void) spa_vdev_exit(spa, NULL, txg, 0); 1715fa9e4066Sahrens 17163d7072f8Seschrock /* 17173d7072f8Seschrock * Must hold spa_namespace_lock in order to post resilver sysevent 17183d7072f8Seschrock * w/pool name. 17193d7072f8Seschrock */ 17203d7072f8Seschrock mutex_enter(&spa_namespace_lock); 1721*088f3894Sahrens VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0); 17223d7072f8Seschrock mutex_exit(&spa_namespace_lock); 1723fa9e4066Sahrens 1724fa9e4066Sahrens return (0); 1725fa9e4066Sahrens } 1726fa9e4066Sahrens 1727fa9e4066Sahrens int 17283d7072f8Seschrock vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 1729fa9e4066Sahrens { 1730c5904d13Seschrock vdev_t *vd; 1731441d80aaSlling uint64_t txg; 1732fa9e4066Sahrens 17330a4e9518Sgw /* 17340a4e9518Sgw * Disregard a vdev fault request if the pool has 17350a4e9518Sgw * experienced a complete failure. 17360a4e9518Sgw * 17370a4e9518Sgw * XXX - We do this here so that we don't hold the 17380a4e9518Sgw * spa_namespace_lock in the event that we can't get 17390a4e9518Sgw * the RW_WRITER spa_config_lock. 17400a4e9518Sgw */ 17410a4e9518Sgw if (spa_state(spa) == POOL_STATE_IO_FAILURE) 17420a4e9518Sgw return (EIO); 17430a4e9518Sgw 1744441d80aaSlling txg = spa_vdev_enter(spa); 1745fa9e4066Sahrens 1746c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 1747441d80aaSlling return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 1748fa9e4066Sahrens 17490e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 17500e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 17510e34b6a7Sbonwick 1752fa9e4066Sahrens /* 1753ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 1754fa9e4066Sahrens */ 1755ecc2d604Sbonwick if (!vd->vdev_offline) { 1756ecc2d604Sbonwick /* 1757ecc2d604Sbonwick * If this device's top-level vdev has a non-empty DTL, 1758ecc2d604Sbonwick * don't allow the device to be offlined. 1759ecc2d604Sbonwick * 1760ecc2d604Sbonwick * XXX -- make this more precise by allowing the offline 1761ecc2d604Sbonwick * as long as the remaining devices don't have any DTL holes. 1762ecc2d604Sbonwick */ 1763ecc2d604Sbonwick if (vd->vdev_top->vdev_dtl_map.sm_space != 0) 1764ecc2d604Sbonwick return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 1765fa9e4066Sahrens 1766ecc2d604Sbonwick /* 1767ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 1768ecc2d604Sbonwick * If this action results in the top-level vdev becoming 1769ecc2d604Sbonwick * unusable, undo it and fail the request. 1770ecc2d604Sbonwick */ 1771ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 1772ea8dc4b6Seschrock vdev_reopen(vd->vdev_top); 1773c5904d13Seschrock if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) { 1774ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 1775ecc2d604Sbonwick vdev_reopen(vd->vdev_top); 1776ecc2d604Sbonwick return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 1777ecc2d604Sbonwick } 1778fa9e4066Sahrens } 1779fa9e4066Sahrens 17803d7072f8Seschrock vd->vdev_tmpoffline = (flags & ZFS_OFFLINE_TEMPORARY) ? 17813d7072f8Seschrock B_TRUE : B_FALSE; 1782ecc2d604Sbonwick 1783ecc2d604Sbonwick vdev_config_dirty(vd->vdev_top); 1784441d80aaSlling 1785441d80aaSlling return (spa_vdev_exit(spa, NULL, txg, 0)); 1786fa9e4066Sahrens } 1787fa9e4066Sahrens 1788ea8dc4b6Seschrock /* 1789ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 1790ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 1791ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 17920a4e9518Sgw * If reopen is specified then attempt to reopen the vdev if the vdev is 17930a4e9518Sgw * faulted or degraded. 1794ea8dc4b6Seschrock */ 1795ea8dc4b6Seschrock void 17960a4e9518Sgw vdev_clear(spa_t *spa, vdev_t *vd, boolean_t reopen_wanted) 1797fa9e4066Sahrens { 1798ea8dc4b6Seschrock int c; 1799fa9e4066Sahrens 1800ea8dc4b6Seschrock if (vd == NULL) 1801ea8dc4b6Seschrock vd = spa->spa_root_vdev; 1802fa9e4066Sahrens 1803ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 1804ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 1805ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 18060a4e9518Sgw vd->vdev_is_failing = B_FALSE; 1807fa9e4066Sahrens 1808ea8dc4b6Seschrock for (c = 0; c < vd->vdev_children; c++) 18090a4e9518Sgw vdev_clear(spa, vd->vdev_child[c], reopen_wanted); 18103d7072f8Seschrock 18113d7072f8Seschrock /* 18128a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 18138a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 18148a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 18158a79c1b5Sek * written out to disk. 18163d7072f8Seschrock */ 18178a79c1b5Sek if (reopen_wanted && (vd->vdev_faulted || vd->vdev_degraded || 18188a79c1b5Sek vd->vdev_stat.vs_aux == VDEV_AUX_IO_FAILURE)) { 18198a79c1b5Sek boolean_t resilver = (vd->vdev_faulted || vd->vdev_degraded); 18208a79c1b5Sek 18213d7072f8Seschrock vd->vdev_faulted = vd->vdev_degraded = 0; 18223d7072f8Seschrock vdev_reopen(vd); 18233d7072f8Seschrock vdev_config_dirty(vd->vdev_top); 18243d7072f8Seschrock 18258a79c1b5Sek if (resilver && vd->vdev_aux == NULL && !vdev_is_dead(vd)) 1826bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 18273d7072f8Seschrock 18283d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 18293d7072f8Seschrock } 1830fa9e4066Sahrens } 1831fa9e4066Sahrens 18320a4e9518Sgw int 18330a4e9518Sgw vdev_readable(vdev_t *vd) 18340a4e9518Sgw { 18350a4e9518Sgw /* XXPOLICY */ 18360a4e9518Sgw return (!vdev_is_dead(vd)); 18370a4e9518Sgw } 18380a4e9518Sgw 18390a4e9518Sgw int 18400a4e9518Sgw vdev_writeable(vdev_t *vd) 18410a4e9518Sgw { 18421f7ad2e1Sgw return (!vdev_is_dead(vd) && !vd->vdev_is_failing); 18430a4e9518Sgw } 18440a4e9518Sgw 1845fa9e4066Sahrens int 1846fa9e4066Sahrens vdev_is_dead(vdev_t *vd) 1847fa9e4066Sahrens { 184832b87932Sek /* 184932b87932Sek * If the vdev experienced I/O failures, then the vdev is marked 185032b87932Sek * as faulted (VDEV_STATE_FAULTED) for status output and FMA; however, 185132b87932Sek * we need to allow access to the vdev for resumed I/Os (see 185232b87932Sek * zio_vdev_resume_io() ). 185332b87932Sek */ 185432b87932Sek return (vd->vdev_state < VDEV_STATE_DEGRADED && 185532b87932Sek vd->vdev_stat.vs_aux != VDEV_AUX_IO_FAILURE); 1856fa9e4066Sahrens } 1857fa9e4066Sahrens 1858fa9e4066Sahrens int 1859fa9e4066Sahrens vdev_error_inject(vdev_t *vd, zio_t *zio) 1860fa9e4066Sahrens { 1861fa9e4066Sahrens int error = 0; 1862fa9e4066Sahrens 1863fa9e4066Sahrens if (vd->vdev_fault_mode == VDEV_FAULT_NONE) 1864fa9e4066Sahrens return (0); 1865fa9e4066Sahrens 1866fa9e4066Sahrens if (((1ULL << zio->io_type) & vd->vdev_fault_mask) == 0) 1867fa9e4066Sahrens return (0); 1868fa9e4066Sahrens 1869fa9e4066Sahrens switch (vd->vdev_fault_mode) { 1870fa9e4066Sahrens case VDEV_FAULT_RANDOM: 1871fa9e4066Sahrens if (spa_get_random(vd->vdev_fault_arg) == 0) 1872fa9e4066Sahrens error = EIO; 1873fa9e4066Sahrens break; 1874fa9e4066Sahrens 1875fa9e4066Sahrens case VDEV_FAULT_COUNT: 1876fa9e4066Sahrens if ((int64_t)--vd->vdev_fault_arg <= 0) 1877fa9e4066Sahrens vd->vdev_fault_mode = VDEV_FAULT_NONE; 1878fa9e4066Sahrens error = EIO; 1879fa9e4066Sahrens break; 1880fa9e4066Sahrens } 1881fa9e4066Sahrens 1882fa9e4066Sahrens return (error); 1883fa9e4066Sahrens } 1884fa9e4066Sahrens 1885fa9e4066Sahrens /* 1886fa9e4066Sahrens * Get statistics for the given vdev. 1887fa9e4066Sahrens */ 1888fa9e4066Sahrens void 1889fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 1890fa9e4066Sahrens { 1891fa9e4066Sahrens vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 1892fa9e4066Sahrens int c, t; 1893fa9e4066Sahrens 1894fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 1895fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 1896*088f3894Sahrens vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors; 1897fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 1898fa9e4066Sahrens vs->vs_state = vd->vdev_state; 18992a79c5feSlling vs->vs_rsize = vdev_get_rsize(vd); 1900fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 1901fa9e4066Sahrens 1902fa9e4066Sahrens /* 1903fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 1904fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 1905fa9e4066Sahrens */ 1906fa9e4066Sahrens if (vd == rvd) { 1907fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) { 1908fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 1909fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 1910fa9e4066Sahrens 1911fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 1912fa9e4066Sahrens for (t = 0; t < ZIO_TYPES; t++) { 1913fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 1914fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 1915fa9e4066Sahrens } 1916fa9e4066Sahrens vs->vs_read_errors += cvs->vs_read_errors; 1917fa9e4066Sahrens vs->vs_write_errors += cvs->vs_write_errors; 1918fa9e4066Sahrens vs->vs_checksum_errors += cvs->vs_checksum_errors; 1919fa9e4066Sahrens vs->vs_scrub_examined += cvs->vs_scrub_examined; 1920fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 1921fa9e4066Sahrens } 1922fa9e4066Sahrens } 1923fa9e4066Sahrens } 1924fa9e4066Sahrens 1925fa94a07fSbrendan void 1926fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 1927fa94a07fSbrendan { 1928fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 1929fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 1930fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 1931fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 1932fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 1933fa94a07fSbrendan } 1934fa94a07fSbrendan 1935fa9e4066Sahrens void 1936fa9e4066Sahrens vdev_stat_update(zio_t *zio) 1937fa9e4066Sahrens { 1938fa9e4066Sahrens vdev_t *vd = zio->io_vd; 1939fa9e4066Sahrens vdev_t *pvd; 1940fa9e4066Sahrens uint64_t txg = zio->io_txg; 1941fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 1942fa9e4066Sahrens zio_type_t type = zio->io_type; 1943fa9e4066Sahrens int flags = zio->io_flags; 1944fa9e4066Sahrens 1945fa9e4066Sahrens if (zio->io_error == 0) { 1946fa9e4066Sahrens if (!(flags & ZIO_FLAG_IO_BYPASS)) { 1947fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 1948fa9e4066Sahrens vs->vs_ops[type]++; 1949fa9e4066Sahrens vs->vs_bytes[type] += zio->io_size; 1950fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 1951fa9e4066Sahrens } 1952fa9e4066Sahrens if ((flags & ZIO_FLAG_IO_REPAIR) && 1953fa9e4066Sahrens zio->io_delegate_list == NULL) { 1954fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 1955d80c45e0Sbonwick if (flags & ZIO_FLAG_SCRUB_THREAD) 1956fa9e4066Sahrens vs->vs_scrub_repaired += zio->io_size; 1957fa9e4066Sahrens else 1958fa9e4066Sahrens vs->vs_self_healed += zio->io_size; 1959fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 1960fa9e4066Sahrens } 1961fa9e4066Sahrens return; 1962fa9e4066Sahrens } 1963fa9e4066Sahrens 1964fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 1965fa9e4066Sahrens return; 1966fa9e4066Sahrens 19670a4e9518Sgw if (vdev_readable(vd)) { 1968fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 1969fa9e4066Sahrens if (type == ZIO_TYPE_READ) { 1970fa9e4066Sahrens if (zio->io_error == ECKSUM) 1971fa9e4066Sahrens vs->vs_checksum_errors++; 1972fa9e4066Sahrens else 1973fa9e4066Sahrens vs->vs_read_errors++; 1974fa9e4066Sahrens } 1975fa9e4066Sahrens if (type == ZIO_TYPE_WRITE) 1976fa9e4066Sahrens vs->vs_write_errors++; 1977fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 1978fa9e4066Sahrens } 1979fa9e4066Sahrens 1980fa9e4066Sahrens if (type == ZIO_TYPE_WRITE) { 1981fa9e4066Sahrens if (txg == 0 || vd->vdev_children != 0) 1982fa9e4066Sahrens return; 1983d80c45e0Sbonwick if (flags & ZIO_FLAG_SCRUB_THREAD) { 1984fa9e4066Sahrens ASSERT(flags & ZIO_FLAG_IO_REPAIR); 1985fa9e4066Sahrens for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1986fa9e4066Sahrens vdev_dtl_dirty(&pvd->vdev_dtl_scrub, txg, 1); 1987fa9e4066Sahrens } 1988fa9e4066Sahrens if (!(flags & ZIO_FLAG_IO_REPAIR)) { 1989fa9e4066Sahrens if (vdev_dtl_contains(&vd->vdev_dtl_map, txg, 1)) 1990fa9e4066Sahrens return; 1991ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1992fa9e4066Sahrens for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 1993fa9e4066Sahrens vdev_dtl_dirty(&pvd->vdev_dtl_map, txg, 1); 1994fa9e4066Sahrens } 1995fa9e4066Sahrens } 1996fa9e4066Sahrens } 1997fa9e4066Sahrens 1998fa9e4066Sahrens void 1999fa9e4066Sahrens vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete) 2000fa9e4066Sahrens { 2001fa9e4066Sahrens int c; 2002fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 2003fa9e4066Sahrens 2004fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 2005fa9e4066Sahrens vdev_scrub_stat_update(vd->vdev_child[c], type, complete); 2006fa9e4066Sahrens 2007fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 2008fa9e4066Sahrens 2009fa9e4066Sahrens if (type == POOL_SCRUB_NONE) { 2010fa9e4066Sahrens /* 2011fa9e4066Sahrens * Update completion and end time. Leave everything else alone 2012fa9e4066Sahrens * so we can report what happened during the previous scrub. 2013fa9e4066Sahrens */ 2014fa9e4066Sahrens vs->vs_scrub_complete = complete; 2015fa9e4066Sahrens vs->vs_scrub_end = gethrestime_sec(); 2016fa9e4066Sahrens } else { 2017fa9e4066Sahrens vs->vs_scrub_type = type; 2018fa9e4066Sahrens vs->vs_scrub_complete = 0; 2019fa9e4066Sahrens vs->vs_scrub_examined = 0; 2020fa9e4066Sahrens vs->vs_scrub_repaired = 0; 2021fa9e4066Sahrens vs->vs_scrub_start = gethrestime_sec(); 2022fa9e4066Sahrens vs->vs_scrub_end = 0; 2023fa9e4066Sahrens } 2024fa9e4066Sahrens 2025fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 2026fa9e4066Sahrens } 2027fa9e4066Sahrens 2028fa9e4066Sahrens /* 2029fa9e4066Sahrens * Update the in-core space usage stats for this vdev and the root vdev. 2030fa9e4066Sahrens */ 2031fa9e4066Sahrens void 2032fa94a07fSbrendan vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta, 2033fa94a07fSbrendan boolean_t update_root) 2034fa9e4066Sahrens { 203599653d4eSeschrock int64_t dspace_delta = space_delta; 20368654d025Sperrin spa_t *spa = vd->vdev_spa; 20378654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 2038fa9e4066Sahrens 20398654d025Sperrin ASSERT(vd == vd->vdev_top); 204099653d4eSeschrock 20418654d025Sperrin /* 20428654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 20438654d025Sperrin * factor. We must calculate this here and not at the root vdev 20448654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 20458654d025Sperrin * childrens', thus not accurate enough for us. 20468654d025Sperrin */ 20478654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 20488654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 20498654d025Sperrin vd->vdev_deflate_ratio; 20508654d025Sperrin 20518654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 20528654d025Sperrin vd->vdev_stat.vs_space += space_delta; 20538654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 20548654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 20558654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 20568654d025Sperrin 2057fa94a07fSbrendan if (update_root) { 2058fa94a07fSbrendan ASSERT(rvd == vd->vdev_parent); 2059fa94a07fSbrendan ASSERT(vd->vdev_ms_count != 0); 2060fa94a07fSbrendan 2061fa94a07fSbrendan /* 2062fa94a07fSbrendan * Don't count non-normal (e.g. intent log) space as part of 2063fa94a07fSbrendan * the pool's capacity. 2064fa94a07fSbrendan */ 2065fa94a07fSbrendan if (vd->vdev_mg->mg_class != spa->spa_normal_class) 2066fa94a07fSbrendan return; 20678654d025Sperrin 2068fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 2069fa94a07fSbrendan rvd->vdev_stat.vs_space += space_delta; 2070fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 2071fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 2072fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 2073fa94a07fSbrendan } 2074fa9e4066Sahrens } 2075fa9e4066Sahrens 2076fa9e4066Sahrens /* 2077fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 2078fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 2079fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2080fa9e4066Sahrens */ 2081fa9e4066Sahrens void 2082fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 2083fa9e4066Sahrens { 2084fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2085fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2086fa9e4066Sahrens int c; 2087fa9e4066Sahrens 2088c5904d13Seschrock /* 2089c5904d13Seschrock * If this is an aux vdev (as with l2cache devices), then we update the 2090c5904d13Seschrock * vdev config manually and set the sync flag. 2091c5904d13Seschrock */ 2092c5904d13Seschrock if (vd->vdev_aux != NULL) { 2093c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 2094c5904d13Seschrock nvlist_t **aux; 2095c5904d13Seschrock uint_t naux; 2096c5904d13Seschrock 2097c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 2098c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 2099c5904d13Seschrock break; 2100c5904d13Seschrock } 2101c5904d13Seschrock 2102c5904d13Seschrock ASSERT(c < sav->sav_count); 2103c5904d13Seschrock sav->sav_sync = B_TRUE; 2104c5904d13Seschrock 2105c5904d13Seschrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2106c5904d13Seschrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) == 0); 2107c5904d13Seschrock 2108c5904d13Seschrock ASSERT(c < naux); 2109c5904d13Seschrock 2110c5904d13Seschrock /* 2111c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 2112c5904d13Seschrock * sketchy, but it will work. 2113c5904d13Seschrock */ 2114c5904d13Seschrock nvlist_free(aux[c]); 2115c5904d13Seschrock aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE); 2116c5904d13Seschrock 2117c5904d13Seschrock return; 2118c5904d13Seschrock } 2119c5904d13Seschrock 21205dabedeeSbonwick /* 21215dabedeeSbonwick * The dirty list is protected by the config lock. The caller must 21225dabedeeSbonwick * either hold the config lock as writer, or must be the sync thread 21235dabedeeSbonwick * (which holds the lock as reader). There's only one sync thread, 21245dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 21255dabedeeSbonwick */ 21265dabedeeSbonwick ASSERT(spa_config_held(spa, RW_WRITER) || 21275dabedeeSbonwick dsl_pool_sync_context(spa_get_dsl(spa))); 21285dabedeeSbonwick 2129fa9e4066Sahrens if (vd == rvd) { 2130fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 2131fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 2132fa9e4066Sahrens } else { 2133fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 2134fa9e4066Sahrens 2135ecc2d604Sbonwick if (!list_link_active(&vd->vdev_dirty_node)) 2136fa9e4066Sahrens list_insert_head(&spa->spa_dirty_list, vd); 2137fa9e4066Sahrens } 2138fa9e4066Sahrens } 2139fa9e4066Sahrens 2140fa9e4066Sahrens void 2141fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 2142fa9e4066Sahrens { 21435dabedeeSbonwick spa_t *spa = vd->vdev_spa; 21445dabedeeSbonwick 21455dabedeeSbonwick ASSERT(spa_config_held(spa, RW_WRITER) || 21465dabedeeSbonwick dsl_pool_sync_context(spa_get_dsl(spa))); 21475dabedeeSbonwick 2148ecc2d604Sbonwick ASSERT(list_link_active(&vd->vdev_dirty_node)); 21495dabedeeSbonwick list_remove(&spa->spa_dirty_list, vd); 2150fa9e4066Sahrens } 2151fa9e4066Sahrens 215232b87932Sek /* 215332b87932Sek * Propagate vdev state up from children to parent. 215432b87932Sek */ 215544cd46caSbillm void 215644cd46caSbillm vdev_propagate_state(vdev_t *vd) 215744cd46caSbillm { 215844cd46caSbillm vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 215944cd46caSbillm int degraded = 0, faulted = 0; 216044cd46caSbillm int corrupted = 0; 216144cd46caSbillm int c; 216244cd46caSbillm vdev_t *child; 216344cd46caSbillm 21643d7072f8Seschrock if (vd->vdev_children > 0) { 21653d7072f8Seschrock for (c = 0; c < vd->vdev_children; c++) { 21663d7072f8Seschrock child = vd->vdev_child[c]; 216751ece835Seschrock 216851ece835Seschrock if ((vdev_is_dead(child) && !vdev_readable(child)) || 216951ece835Seschrock child->vdev_stat.vs_aux == VDEV_AUX_IO_FAILURE) { 217051ece835Seschrock /* 217151ece835Seschrock * Root special: if there is a top-level log 217251ece835Seschrock * device, treat the root vdev as if it were 217351ece835Seschrock * degraded. 217451ece835Seschrock */ 217551ece835Seschrock if (child->vdev_islog && vd == rvd) 217651ece835Seschrock degraded++; 217751ece835Seschrock else 217851ece835Seschrock faulted++; 217951ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 21803d7072f8Seschrock degraded++; 218151ece835Seschrock } 218244cd46caSbillm 21833d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 21843d7072f8Seschrock corrupted++; 21853d7072f8Seschrock } 218644cd46caSbillm 21873d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 21883d7072f8Seschrock 21893d7072f8Seschrock /* 21903d7072f8Seschrock * Root special: if there is a toplevel vdev that cannot be 21913d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 21923d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 21933d7072f8Seschrock * replicas'. 21943d7072f8Seschrock */ 21953d7072f8Seschrock if (corrupted && vd == rvd && 21963d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 21973d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 21983d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 21993d7072f8Seschrock } 22003d7072f8Seschrock 220151ece835Seschrock if (vd->vdev_parent) 22023d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 220344cd46caSbillm } 220444cd46caSbillm 2205fa9e4066Sahrens /* 2206ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 2207ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 2208ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 2209ea8dc4b6Seschrock * 2210ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 2211ea8dc4b6Seschrock * generated. 2212fa9e4066Sahrens */ 2213fa9e4066Sahrens void 2214ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2215fa9e4066Sahrens { 2216560e6e96Seschrock uint64_t save_state; 2217c5904d13Seschrock spa_t *spa = vd->vdev_spa; 2218ea8dc4b6Seschrock 2219ea8dc4b6Seschrock if (state == vd->vdev_state) { 2220ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 2221fa9e4066Sahrens return; 2222ea8dc4b6Seschrock } 2223ea8dc4b6Seschrock 2224560e6e96Seschrock save_state = vd->vdev_state; 2225fa9e4066Sahrens 2226fa9e4066Sahrens vd->vdev_state = state; 2227fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 2228fa9e4066Sahrens 22293d7072f8Seschrock /* 22303d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 22313d7072f8Seschrock * always close the underlying device. Otherwise, we keep accessible 22323d7072f8Seschrock * but invalid devices open forever. We don't call vdev_close() itself, 22333d7072f8Seschrock * because that implies some extra checks (offline, etc) that we don't 22343d7072f8Seschrock * want here. This is limited to leaf devices, because otherwise 22353d7072f8Seschrock * closing the device will affect other children. 22363d7072f8Seschrock */ 22370a4e9518Sgw if (!vdev_readable(vd) && vd->vdev_ops->vdev_op_leaf) 22383d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 22393d7072f8Seschrock 22403d7072f8Seschrock if (vd->vdev_removed && 22413d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 22423d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 22433d7072f8Seschrock /* 22443d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 22453d7072f8Seschrock * device was previously marked removed and someone attempted to 22463d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 22473d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 22483d7072f8Seschrock * it is one of our special test online cases, which is only 22493d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 22503d7072f8Seschrock * fault. 22513d7072f8Seschrock */ 22523d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 22533d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 22543d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 22553d7072f8Seschrock /* 22563d7072f8Seschrock * Indicate to the ZFS DE that this device has been removed, and 22573d7072f8Seschrock * any recent errors should be ignored. 22583d7072f8Seschrock */ 2259c5904d13Seschrock zfs_post_remove(spa, vd); 22603d7072f8Seschrock vd->vdev_removed = B_TRUE; 22613d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 2262ea8dc4b6Seschrock /* 2263ea8dc4b6Seschrock * If we fail to open a vdev during an import, we mark it as 2264ea8dc4b6Seschrock * "not available", which signifies that it was never there to 2265ea8dc4b6Seschrock * begin with. Failure to open such a device is not considered 2266ea8dc4b6Seschrock * an error. 2267ea8dc4b6Seschrock */ 2268c5904d13Seschrock if (spa->spa_load_state == SPA_LOAD_IMPORT && 2269c5904d13Seschrock !spa->spa_import_faulted && 2270560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 2271560e6e96Seschrock vd->vdev_not_present = 1; 2272560e6e96Seschrock 2273560e6e96Seschrock /* 2274560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 2275560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 2276560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 2277560e6e96Seschrock * want to post the ereport if the device was already in the 2278560e6e96Seschrock * CANT_OPEN state beforehand. 22793d7072f8Seschrock * 22803d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 22813d7072f8Seschrock * online the device in response to an insertion event. If we 22823d7072f8Seschrock * hit this case, then we have detected an insertion event for a 22833d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 22843d7072f8Seschrock * In this scenario, we don't post an ereport because we are 22853d7072f8Seschrock * about to replace the device, or attempt an online with 22863d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 2287560e6e96Seschrock */ 22883d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 22893d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 2290c5904d13Seschrock vd != spa->spa_root_vdev) { 2291ea8dc4b6Seschrock const char *class; 2292ea8dc4b6Seschrock 2293ea8dc4b6Seschrock switch (aux) { 2294ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 2295ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 2296ea8dc4b6Seschrock break; 2297ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 2298ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 2299ea8dc4b6Seschrock break; 2300ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 2301ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 2302ea8dc4b6Seschrock break; 2303ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 2304ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 2305ea8dc4b6Seschrock break; 2306ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 2307ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 2308ea8dc4b6Seschrock break; 2309ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 2310ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 2311ea8dc4b6Seschrock break; 2312ea8dc4b6Seschrock default: 2313ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 2314ea8dc4b6Seschrock } 2315ea8dc4b6Seschrock 2316c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 2317ea8dc4b6Seschrock } 2318ea8dc4b6Seschrock 23193d7072f8Seschrock /* Erase any notion of persistent removed state */ 23203d7072f8Seschrock vd->vdev_removed = B_FALSE; 23213d7072f8Seschrock } else { 23223d7072f8Seschrock vd->vdev_removed = B_FALSE; 23233d7072f8Seschrock } 2324ea8dc4b6Seschrock 23253d7072f8Seschrock if (!isopen) 23263d7072f8Seschrock vdev_propagate_state(vd); 2327fa9e4066Sahrens } 232815e6edf1Sgw 232915e6edf1Sgw /* 233015e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 233115e6edf1Sgw * a root pool. Currently, we do not support RAID-Z or partial configuration. 233215e6edf1Sgw * In addition, only a single top-level vdev is allowed and none of the leaves 233315e6edf1Sgw * can be wholedisks. 233415e6edf1Sgw */ 233515e6edf1Sgw boolean_t 233615e6edf1Sgw vdev_is_bootable(vdev_t *vd) 233715e6edf1Sgw { 233815e6edf1Sgw int c; 233915e6edf1Sgw 234015e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 234115e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 234215e6edf1Sgw 234315e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 234415e6edf1Sgw vd->vdev_children > 1) { 234515e6edf1Sgw return (B_FALSE); 234615e6edf1Sgw } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 234715e6edf1Sgw strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 234815e6edf1Sgw return (B_FALSE); 234915e6edf1Sgw } 235015e6edf1Sgw } else if (vd->vdev_wholedisk == 1) { 235115e6edf1Sgw return (B_FALSE); 235215e6edf1Sgw } 235315e6edf1Sgw 235415e6edf1Sgw for (c = 0; c < vd->vdev_children; c++) { 235515e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 235615e6edf1Sgw return (B_FALSE); 235715e6edf1Sgw } 235815e6edf1Sgw return (B_TRUE); 235915e6edf1Sgw } 2360