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 /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 245cabbc6bSPrashanth Sreenivasa * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 255f368aefSYuri Pankov * Copyright 2017 Nexenta Systems, Inc. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27c8811bd3SToomas Soome * Copyright 2016 Toomas Soome <tsoome@me.com> 28c1064fd7SJerry Jelinek * Copyright 2019 Joyent, Inc. 29663207adSDon Brady * Copyright (c) 2017, Intel Corporation. 30fa9e4066Sahrens */ 31fa9e4066Sahrens 32fa9e4066Sahrens #include <sys/zfs_context.h> 33ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 34fa9e4066Sahrens #include <sys/spa.h> 35fa9e4066Sahrens #include <sys/spa_impl.h> 365cabbc6bSPrashanth Sreenivasa #include <sys/bpobj.h> 37fa9e4066Sahrens #include <sys/dmu.h> 38fa9e4066Sahrens #include <sys/dmu_tx.h> 395cabbc6bSPrashanth Sreenivasa #include <sys/dsl_dir.h> 40fa9e4066Sahrens #include <sys/vdev_impl.h> 41fa9e4066Sahrens #include <sys/uberblock_impl.h> 42fa9e4066Sahrens #include <sys/metaslab.h> 43fa9e4066Sahrens #include <sys/metaslab_impl.h> 44fa9e4066Sahrens #include <sys/space_map.h> 450713e232SGeorge Wilson #include <sys/space_reftree.h> 46fa9e4066Sahrens #include <sys/zio.h> 47fa9e4066Sahrens #include <sys/zap.h> 48fa9e4066Sahrens #include <sys/fs/zfs.h> 49c5904d13Seschrock #include <sys/arc.h> 50e6ca193dSGeorge Wilson #include <sys/zil.h> 513f9d6ad7SLin Ling #include <sys/dsl_scan.h> 52770499e1SDan Kimmel #include <sys/abd.h> 53094e47e9SGeorge Wilson #include <sys/vdev_initialize.h> 54*084fd14fSBrian Behlendorf #include <sys/vdev_trim.h> 55fa9e4066Sahrens 56fa9e4066Sahrens /* 57fa9e4066Sahrens * Virtual device management. 58fa9e4066Sahrens */ 59fa9e4066Sahrens 60fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 61fa9e4066Sahrens &vdev_root_ops, 62fa9e4066Sahrens &vdev_raidz_ops, 63fa9e4066Sahrens &vdev_mirror_ops, 64fa9e4066Sahrens &vdev_replacing_ops, 6599653d4eSeschrock &vdev_spare_ops, 66fa9e4066Sahrens &vdev_disk_ops, 67fa9e4066Sahrens &vdev_file_ops, 68fa9e4066Sahrens &vdev_missing_ops, 6988ecc943SGeorge Wilson &vdev_hole_ops, 705cabbc6bSPrashanth Sreenivasa &vdev_indirect_ops, 71fa9e4066Sahrens NULL 72fa9e4066Sahrens }; 73fa9e4066Sahrens 74088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 75088f3894Sahrens int zfs_scrub_limit = 10; 7605b2b3b8Smishra 77a0b03b16SSerapheim Dimitropoulos /* default target for number of metaslabs per top-level vdev */ 78a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_count = 200; 7986714001SSerapheim Dimitropoulos 80b4bf0cf0SDon Brady /* minimum number of metaslabs per top-level vdev */ 81a0b03b16SSerapheim Dimitropoulos int zfs_vdev_min_ms_count = 16; 8286714001SSerapheim Dimitropoulos 83b4bf0cf0SDon Brady /* practical upper limit of total metaslabs per top-level vdev */ 84a0b03b16SSerapheim Dimitropoulos int zfs_vdev_ms_count_limit = 1ULL << 17; 85b4bf0cf0SDon Brady 86b4bf0cf0SDon Brady /* lower limit for metaslab size (512M) */ 87a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_shift = 29; 8886714001SSerapheim Dimitropoulos 89a0b03b16SSerapheim Dimitropoulos /* upper limit for metaslab size (16G) */ 90a0b03b16SSerapheim Dimitropoulos int zfs_vdev_max_ms_shift = 34; 91b4bf0cf0SDon Brady 9286714001SSerapheim Dimitropoulos boolean_t vdev_validate_skip = B_FALSE; 9386714001SSerapheim Dimitropoulos 94bf3e216cSMatthew Ahrens /* 9586714001SSerapheim Dimitropoulos * Since the DTL space map of a vdev is not expected to have a lot of 9686714001SSerapheim Dimitropoulos * entries, we default its block size to 4K. 97bf3e216cSMatthew Ahrens */ 9886714001SSerapheim Dimitropoulos int vdev_dtl_sm_blksz = (1 << 12); 99bf3e216cSMatthew Ahrens 10086714001SSerapheim Dimitropoulos /* 10186714001SSerapheim Dimitropoulos * vdev-wide space maps that have lots of entries written to them at 10286714001SSerapheim Dimitropoulos * the end of each transaction can benefit from a higher I/O bandwidth 10386714001SSerapheim Dimitropoulos * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 10486714001SSerapheim Dimitropoulos */ 10586714001SSerapheim Dimitropoulos int vdev_standard_sm_blksz = (1 << 17); 1066f793812SPavel Zakharov 10793a1902eSMatthew Ahrens int zfs_ashift_min; 10893a1902eSMatthew Ahrens 1093ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 1103ee8c80cSPavel Zakharov void 1113ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 1123ee8c80cSPavel Zakharov { 1133ee8c80cSPavel Zakharov va_list adx; 1143ee8c80cSPavel Zakharov char buf[256]; 1153ee8c80cSPavel Zakharov 1163ee8c80cSPavel Zakharov va_start(adx, fmt); 1173ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 1183ee8c80cSPavel Zakharov va_end(adx); 1193ee8c80cSPavel Zakharov 1203ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 1213ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 1223ee8c80cSPavel Zakharov vd->vdev_path, buf); 1233ee8c80cSPavel Zakharov } else { 1243ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 1253ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 1263ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1273ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1283ee8c80cSPavel Zakharov } 1293ee8c80cSPavel Zakharov } 1303ee8c80cSPavel Zakharov 1316f793812SPavel Zakharov void 1326f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 1336f793812SPavel Zakharov { 1346f793812SPavel Zakharov char state[20]; 1356f793812SPavel Zakharov 1366f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 1376f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 1386f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 1396f793812SPavel Zakharov return; 1406f793812SPavel Zakharov } 1416f793812SPavel Zakharov 1426f793812SPavel Zakharov switch (vd->vdev_state) { 1436f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 1446f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 1456f793812SPavel Zakharov break; 1466f793812SPavel Zakharov case VDEV_STATE_CLOSED: 1476f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 1486f793812SPavel Zakharov break; 1496f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 1506f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 1516f793812SPavel Zakharov break; 1526f793812SPavel Zakharov case VDEV_STATE_REMOVED: 1536f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 1546f793812SPavel Zakharov break; 1556f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 1566f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 1576f793812SPavel Zakharov break; 1586f793812SPavel Zakharov case VDEV_STATE_FAULTED: 1596f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 1606f793812SPavel Zakharov break; 1616f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 1626f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 1636f793812SPavel Zakharov break; 1646f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 1656f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 1666f793812SPavel Zakharov break; 1676f793812SPavel Zakharov default: 1686f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 1696f793812SPavel Zakharov (uint_t)vd->vdev_state); 1706f793812SPavel Zakharov } 1716f793812SPavel Zakharov 1726f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 173c7a7b2faSAndriy Gapon "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 1746f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 1756f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1766f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 1776f793812SPavel Zakharov 1786f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 1796f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 1806f793812SPavel Zakharov } 1816f793812SPavel Zakharov 182fa9e4066Sahrens /* 183fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 184fa9e4066Sahrens */ 185fa9e4066Sahrens static vdev_ops_t * 186fa9e4066Sahrens vdev_getops(const char *type) 187fa9e4066Sahrens { 188fa9e4066Sahrens vdev_ops_t *ops, **opspp; 189fa9e4066Sahrens 190fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 191fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 192fa9e4066Sahrens break; 193fa9e4066Sahrens 194fa9e4066Sahrens return (ops); 195fa9e4066Sahrens } 196fa9e4066Sahrens 197663207adSDon Brady /* 198663207adSDon Brady * Derive the enumerated alloction bias from string input. 199663207adSDon Brady * String origin is either the per-vdev zap or zpool(1M). 200663207adSDon Brady */ 201663207adSDon Brady static vdev_alloc_bias_t 202663207adSDon Brady vdev_derive_alloc_bias(const char *bias) 203663207adSDon Brady { 204663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 205663207adSDon Brady 206663207adSDon Brady if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 207663207adSDon Brady alloc_bias = VDEV_BIAS_LOG; 208663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 209663207adSDon Brady alloc_bias = VDEV_BIAS_SPECIAL; 210663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 211663207adSDon Brady alloc_bias = VDEV_BIAS_DEDUP; 212663207adSDon Brady 213663207adSDon Brady return (alloc_bias); 214663207adSDon Brady } 215663207adSDon Brady 216094e47e9SGeorge Wilson /* ARGSUSED */ 217094e47e9SGeorge Wilson void 218094e47e9SGeorge Wilson vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res) 219094e47e9SGeorge Wilson { 220094e47e9SGeorge Wilson res->rs_start = in->rs_start; 221094e47e9SGeorge Wilson res->rs_end = in->rs_end; 222094e47e9SGeorge Wilson } 223094e47e9SGeorge Wilson 224fa9e4066Sahrens /* 225fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 226fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 227fa9e4066Sahrens */ 228fa9e4066Sahrens uint64_t 229fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 230fa9e4066Sahrens { 231ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 232fa9e4066Sahrens uint64_t csize; 233fa9e4066Sahrens 234573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 235fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 236fa9e4066Sahrens asize = MAX(asize, csize); 237fa9e4066Sahrens } 238fa9e4066Sahrens 239fa9e4066Sahrens return (asize); 240fa9e4066Sahrens } 241fa9e4066Sahrens 2422a79c5feSlling /* 243573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 244573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 245573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 246573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 2472a79c5feSlling */ 2482a79c5feSlling uint64_t 249573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 2502a79c5feSlling { 251573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 2522a79c5feSlling 253573ca77eSGeorge Wilson /* 2544263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 255573ca77eSGeorge Wilson * just return our own asize. 256573ca77eSGeorge Wilson */ 257573ca77eSGeorge Wilson if (pvd == NULL) 258573ca77eSGeorge Wilson return (vd->vdev_asize); 2592a79c5feSlling 2602a79c5feSlling /* 261573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 262573ca77eSGeorge Wilson * to the nearest metaslab. 2632a79c5feSlling */ 264573ca77eSGeorge Wilson if (vd == vd->vdev_top) 265573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2662a79c5feSlling 267573ca77eSGeorge Wilson /* 268573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 269573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 270573ca77eSGeorge Wilson */ 271573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 272c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 273c040c10cSSteven Hartland pvd->vdev_children); 2742a79c5feSlling 275573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 276573ca77eSGeorge Wilson } 2772a79c5feSlling 278573ca77eSGeorge Wilson void 279573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 280573ca77eSGeorge Wilson { 281573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 282573ca77eSGeorge Wilson 283573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 284573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2852a79c5feSlling } 2862a79c5feSlling 287fa9e4066Sahrens vdev_t * 288fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 289fa9e4066Sahrens { 290fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 291fa9e4066Sahrens 292e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 293e05725b1Sbonwick 294088f3894Sahrens if (vdev < rvd->vdev_children) { 295088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 296fa9e4066Sahrens return (rvd->vdev_child[vdev]); 297088f3894Sahrens } 298fa9e4066Sahrens 299fa9e4066Sahrens return (NULL); 300fa9e4066Sahrens } 301fa9e4066Sahrens 302fa9e4066Sahrens vdev_t * 303fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 304fa9e4066Sahrens { 305fa9e4066Sahrens vdev_t *mvd; 306fa9e4066Sahrens 3070e34b6a7Sbonwick if (vd->vdev_guid == guid) 308fa9e4066Sahrens return (vd); 309fa9e4066Sahrens 310573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 311fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 312fa9e4066Sahrens NULL) 313fa9e4066Sahrens return (mvd); 314fa9e4066Sahrens 315fa9e4066Sahrens return (NULL); 316fa9e4066Sahrens } 317fa9e4066Sahrens 31812380e1eSArne Jansen static int 31912380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 32012380e1eSArne Jansen { 32112380e1eSArne Jansen int n = 0; 32212380e1eSArne Jansen 32312380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 32412380e1eSArne Jansen return (1); 32512380e1eSArne Jansen 32612380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 32712380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 32812380e1eSArne Jansen 32912380e1eSArne Jansen return (n); 33012380e1eSArne Jansen } 33112380e1eSArne Jansen 33212380e1eSArne Jansen int 33312380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 33412380e1eSArne Jansen { 33512380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 33612380e1eSArne Jansen } 33712380e1eSArne Jansen 338fa9e4066Sahrens void 339fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 340fa9e4066Sahrens { 341fa9e4066Sahrens size_t oldsize, newsize; 342fa9e4066Sahrens uint64_t id = cvd->vdev_id; 343fa9e4066Sahrens vdev_t **newchild; 34481cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 345fa9e4066Sahrens 34681cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 347fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 348fa9e4066Sahrens 349fa9e4066Sahrens cvd->vdev_parent = pvd; 350fa9e4066Sahrens 351fa9e4066Sahrens if (pvd == NULL) 352fa9e4066Sahrens return; 353fa9e4066Sahrens 354fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 355fa9e4066Sahrens 356fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 357fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 358fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 359fa9e4066Sahrens 360fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 361fa9e4066Sahrens if (pvd->vdev_child != NULL) { 362fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 363fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 364fa9e4066Sahrens } 365fa9e4066Sahrens 366fa9e4066Sahrens pvd->vdev_child = newchild; 367fa9e4066Sahrens pvd->vdev_child[id] = cvd; 368fa9e4066Sahrens 369fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 370fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 371fa9e4066Sahrens 372fa9e4066Sahrens /* 373fa9e4066Sahrens * Walk up all ancestors to update guid sum. 374fa9e4066Sahrens */ 375fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 376fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 377e0f1c0afSOlaf Faaland 378e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 379e0f1c0afSOlaf Faaland list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 380e0f1c0afSOlaf Faaland cvd->vdev_spa->spa_leaf_list_gen++; 381e0f1c0afSOlaf Faaland } 382fa9e4066Sahrens } 383fa9e4066Sahrens 384fa9e4066Sahrens void 385fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 386fa9e4066Sahrens { 387fa9e4066Sahrens int c; 388fa9e4066Sahrens uint_t id = cvd->vdev_id; 389fa9e4066Sahrens 390fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 391fa9e4066Sahrens 392fa9e4066Sahrens if (pvd == NULL) 393fa9e4066Sahrens return; 394fa9e4066Sahrens 395fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 396fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 397fa9e4066Sahrens 398fa9e4066Sahrens pvd->vdev_child[id] = NULL; 399fa9e4066Sahrens cvd->vdev_parent = NULL; 400fa9e4066Sahrens 401fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 402fa9e4066Sahrens if (pvd->vdev_child[c]) 403fa9e4066Sahrens break; 404fa9e4066Sahrens 405fa9e4066Sahrens if (c == pvd->vdev_children) { 406fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 407fa9e4066Sahrens pvd->vdev_child = NULL; 408fa9e4066Sahrens pvd->vdev_children = 0; 409fa9e4066Sahrens } 410fa9e4066Sahrens 411e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 412e0f1c0afSOlaf Faaland spa_t *spa = cvd->vdev_spa; 413e0f1c0afSOlaf Faaland list_remove(&spa->spa_leaf_list, cvd); 414e0f1c0afSOlaf Faaland spa->spa_leaf_list_gen++; 415e0f1c0afSOlaf Faaland } 416e0f1c0afSOlaf Faaland 417fa9e4066Sahrens /* 418fa9e4066Sahrens * Walk up all ancestors to update guid sum. 419fa9e4066Sahrens */ 420fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 421fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 422fa9e4066Sahrens } 423fa9e4066Sahrens 424fa9e4066Sahrens /* 425fa9e4066Sahrens * Remove any holes in the child array. 426fa9e4066Sahrens */ 427fa9e4066Sahrens void 428fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 429fa9e4066Sahrens { 430fa9e4066Sahrens vdev_t **newchild, *cvd; 431fa9e4066Sahrens int oldc = pvd->vdev_children; 432573ca77eSGeorge Wilson int newc; 433fa9e4066Sahrens 434e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 435fa9e4066Sahrens 436573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 437fa9e4066Sahrens if (pvd->vdev_child[c]) 438fa9e4066Sahrens newc++; 439fa9e4066Sahrens 440fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 441fa9e4066Sahrens 442573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 443fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 444fa9e4066Sahrens newchild[newc] = cvd; 445fa9e4066Sahrens cvd->vdev_id = newc++; 446fa9e4066Sahrens } 447fa9e4066Sahrens } 448fa9e4066Sahrens 449fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 450fa9e4066Sahrens pvd->vdev_child = newchild; 451fa9e4066Sahrens pvd->vdev_children = newc; 452fa9e4066Sahrens } 453fa9e4066Sahrens 454fa9e4066Sahrens /* 455fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 456fa9e4066Sahrens */ 45788ecc943SGeorge Wilson vdev_t * 458fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 459fa9e4066Sahrens { 460fa9e4066Sahrens vdev_t *vd; 4615cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 462fa9e4066Sahrens 463fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 4645cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 465fa9e4066Sahrens 4660e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4670e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4680e34b6a7Sbonwick spa->spa_root_vdev = vd; 469e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4700e34b6a7Sbonwick } 4710e34b6a7Sbonwick 47288ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4730e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4740e34b6a7Sbonwick /* 4750e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4760e34b6a7Sbonwick * which must be unique among all pools. 4770e34b6a7Sbonwick */ 4781195e687SMark J Musante guid = spa_generate_guid(NULL); 4790e34b6a7Sbonwick } else { 4800e34b6a7Sbonwick /* 4810e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4820e34b6a7Sbonwick */ 4831195e687SMark J Musante guid = spa_generate_guid(spa); 4840e34b6a7Sbonwick } 4850e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4860e34b6a7Sbonwick } 4870e34b6a7Sbonwick 488fa9e4066Sahrens vd->vdev_spa = spa; 489fa9e4066Sahrens vd->vdev_id = id; 490fa9e4066Sahrens vd->vdev_guid = guid; 491fa9e4066Sahrens vd->vdev_guid_sum = guid; 492fa9e4066Sahrens vd->vdev_ops = ops; 493fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 49488ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 4955cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 4965cabbc6bSPrashanth Sreenivasa 4975cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 4985cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 4995cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); 500fa9e4066Sahrens 501*084fd14fSBrian Behlendorf list_link_init(&vd->vdev_initialize_node); 502e0f1c0afSOlaf Faaland list_link_init(&vd->vdev_leaf_node); 503*084fd14fSBrian Behlendorf list_link_init(&vd->vdev_trim_node); 504fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 5055ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 506e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 507a3874b8bSToomas Soome mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 508094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 509094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 510094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 511094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 512*084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 513*084fd14fSBrian Behlendorf mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 514*084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 515*084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 516*084fd14fSBrian Behlendorf cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 517*084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 518094e47e9SGeorge Wilson 5198ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 5205cabbc6bSPrashanth Sreenivasa vd->vdev_dtl[t] = range_tree_create(NULL, NULL); 5218ad4d6ddSJeff Bonwick } 522b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 523fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 524b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 525fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 526fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 5273d7072f8Seschrock vdev_queue_init(vd); 5283d7072f8Seschrock vdev_cache_init(vd); 529fa9e4066Sahrens 530fa9e4066Sahrens return (vd); 531fa9e4066Sahrens } 532fa9e4066Sahrens 533fa9e4066Sahrens /* 534fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 535fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 536fa9e4066Sahrens * different for each case. 537fa9e4066Sahrens */ 53899653d4eSeschrock int 53999653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 54099653d4eSeschrock int alloctype) 541fa9e4066Sahrens { 542fa9e4066Sahrens vdev_ops_t *ops; 543fa9e4066Sahrens char *type; 5448654d025Sperrin uint64_t guid = 0, islog, nparity; 545fa9e4066Sahrens vdev_t *vd; 5465cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 547663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 548663207adSDon Brady boolean_t top_level = (parent && !parent->vdev_parent); 549fa9e4066Sahrens 550e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 551fa9e4066Sahrens 552fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 553be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 554fa9e4066Sahrens 555fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 556be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 557fa9e4066Sahrens 558fa9e4066Sahrens /* 559fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 560fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 561fa9e4066Sahrens */ 562fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 563fa9e4066Sahrens uint64_t label_id; 564fa9e4066Sahrens 565fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 566fa9e4066Sahrens label_id != id) 567be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 568fa9e4066Sahrens 569fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 570be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 57199653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 57299653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 573be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 574fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 575fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 576be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 57721ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 57821ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 579be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 580fa9e4066Sahrens } 581fa9e4066Sahrens 58299653d4eSeschrock /* 58399653d4eSeschrock * The first allocated vdev must be of type 'root'. 58499653d4eSeschrock */ 58599653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 586be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 58799653d4eSeschrock 5888654d025Sperrin /* 5898654d025Sperrin * Determine whether we're a log vdev. 5908654d025Sperrin */ 5918654d025Sperrin islog = 0; 5928654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 593990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 594be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 595fa9e4066Sahrens 59688ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 597be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 59888ecc943SGeorge Wilson 59999653d4eSeschrock /* 6008654d025Sperrin * Set the nparity property for RAID-Z vdevs. 60199653d4eSeschrock */ 6028654d025Sperrin nparity = -1ULL; 60399653d4eSeschrock if (ops == &vdev_raidz_ops) { 60499653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 6058654d025Sperrin &nparity) == 0) { 606b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 607be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 60899653d4eSeschrock /* 609f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 610f94275ceSAdam Leventhal * device. 61199653d4eSeschrock */ 612f94275ceSAdam Leventhal if (nparity > 1 && 613f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 614be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 615f94275ceSAdam Leventhal if (nparity > 2 && 616f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 617be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 61899653d4eSeschrock } else { 61999653d4eSeschrock /* 62099653d4eSeschrock * We require the parity to be specified for SPAs that 62199653d4eSeschrock * support multiple parity levels. 62299653d4eSeschrock */ 623f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 624be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 62599653d4eSeschrock /* 62699653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 62799653d4eSeschrock */ 6288654d025Sperrin nparity = 1; 62999653d4eSeschrock } 63099653d4eSeschrock } else { 6318654d025Sperrin nparity = 0; 63299653d4eSeschrock } 6338654d025Sperrin ASSERT(nparity != -1ULL); 6348654d025Sperrin 635663207adSDon Brady /* 636663207adSDon Brady * If creating a top-level vdev, check for allocation classes input 637663207adSDon Brady */ 638663207adSDon Brady if (top_level && alloctype == VDEV_ALLOC_ADD) { 639663207adSDon Brady char *bias; 640663207adSDon Brady 641663207adSDon Brady if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 642663207adSDon Brady &bias) == 0) { 643663207adSDon Brady alloc_bias = vdev_derive_alloc_bias(bias); 644663207adSDon Brady 645663207adSDon Brady /* spa_vdev_add() expects feature to be enabled */ 646c1064fd7SJerry Jelinek if (alloc_bias != VDEV_BIAS_LOG && 647c1064fd7SJerry Jelinek spa->spa_load_state != SPA_LOAD_CREATE && 648663207adSDon Brady !spa_feature_is_enabled(spa, 649663207adSDon Brady SPA_FEATURE_ALLOCATION_CLASSES)) { 650663207adSDon Brady return (SET_ERROR(ENOTSUP)); 651663207adSDon Brady } 652663207adSDon Brady } 653663207adSDon Brady } 654663207adSDon Brady 6558654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 6565cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 6578654d025Sperrin 6588654d025Sperrin vd->vdev_islog = islog; 6598654d025Sperrin vd->vdev_nparity = nparity; 660663207adSDon Brady if (top_level && alloc_bias != VDEV_BIAS_NONE) 661663207adSDon Brady vd->vdev_alloc_bias = alloc_bias; 6628654d025Sperrin 6638654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 6648654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 6658654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 6668654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 6678654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 6688654d025Sperrin &vd->vdev_physpath) == 0) 6698654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 6706809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 6716809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 67299653d4eSeschrock 673afefbcddSeschrock /* 674afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 675afefbcddSeschrock * as -1. 676afefbcddSeschrock */ 677afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 678afefbcddSeschrock &vd->vdev_wholedisk) != 0) 679afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 680afefbcddSeschrock 6815cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 6825cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 6835cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 6845cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 6855cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 6865cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 6875cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 6885cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 6895cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 6905cabbc6bSPrashanth Sreenivasa 691ea8dc4b6Seschrock /* 692ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 693ea8dc4b6Seschrock * was not present at the time of import. 694ea8dc4b6Seschrock */ 6956809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 6966809eb4eSEric Schrock &vd->vdev_not_present); 697ea8dc4b6Seschrock 698ecc2d604Sbonwick /* 699ecc2d604Sbonwick * Get the alignment requirement. 700ecc2d604Sbonwick */ 701ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 702ecc2d604Sbonwick 70388ecc943SGeorge Wilson /* 70488ecc943SGeorge Wilson * Retrieve the vdev creation time. 70588ecc943SGeorge Wilson */ 70688ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 70788ecc943SGeorge Wilson &vd->vdev_crtxg); 70888ecc943SGeorge Wilson 709fa9e4066Sahrens /* 710fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 711fa9e4066Sahrens */ 712663207adSDon Brady if (top_level && 7131195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 714fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 715fa9e4066Sahrens &vd->vdev_ms_array); 716fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 717fa9e4066Sahrens &vd->vdev_ms_shift); 718fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 719fa9e4066Sahrens &vd->vdev_asize); 7203f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 7213f9d6ad7SLin Ling &vd->vdev_removing); 722215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 723215198a6SJoe Stein &vd->vdev_top_zap); 724215198a6SJoe Stein } else { 725215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 726fa9e4066Sahrens } 727fa9e4066Sahrens 728663207adSDon Brady if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 729a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 7309f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 7311195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 7329f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 733663207adSDon Brady /* Note: metaslab_group_create() is now deferred */ 734a1521560SJeff Bonwick } 735a1521560SJeff Bonwick 736215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 737215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 738215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 739215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 740215198a6SJoe Stein } else { 741215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 742215198a6SJoe Stein } 743215198a6SJoe Stein 744fa9e4066Sahrens /* 7453d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 746fa9e4066Sahrens */ 747215198a6SJoe Stein 748c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 74921ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 75021ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 751c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 752c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 7530713e232SGeorge Wilson &vd->vdev_dtl_object); 754c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 755c5904d13Seschrock &vd->vdev_unspare); 756c5904d13Seschrock } 75721ecdf64SLin Ling 75821ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 75921ecdf64SLin Ling uint64_t spare = 0; 76021ecdf64SLin Ling 76121ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 76221ecdf64SLin Ling &spare) == 0 && spare) 76321ecdf64SLin Ling spa_spare_add(vd); 76421ecdf64SLin Ling } 76521ecdf64SLin Ling 766ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 767ecc2d604Sbonwick &vd->vdev_offline); 768c5904d13Seschrock 769b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 770b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 771cb04b873SMark J Musante 772e4c795beSTom Caputi if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 773e4c795beSTom Caputi vdev_set_deferred_resilver(spa, vd); 774e4c795beSTom Caputi 7753d7072f8Seschrock /* 7763d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 7773d7072f8Seschrock * state, as the diagnosis made on another system may not be 778069f55e2SEric Schrock * valid in the current context. Local vdevs will 779069f55e2SEric Schrock * remain in the faulted state. 7803d7072f8Seschrock */ 781b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 7823d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 7833d7072f8Seschrock &vd->vdev_faulted); 7843d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 7853d7072f8Seschrock &vd->vdev_degraded); 7863d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 7873d7072f8Seschrock &vd->vdev_removed); 788069f55e2SEric Schrock 789069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 790069f55e2SEric Schrock char *aux; 791069f55e2SEric Schrock 792069f55e2SEric Schrock vd->vdev_label_aux = 793069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 794069f55e2SEric Schrock if (nvlist_lookup_string(nv, 795069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 796069f55e2SEric Schrock strcmp(aux, "external") == 0) 797069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 798069f55e2SEric Schrock } 7993d7072f8Seschrock } 800fa9e4066Sahrens } 801fa9e4066Sahrens 802fa9e4066Sahrens /* 803fa9e4066Sahrens * Add ourselves to the parent's list of children. 804fa9e4066Sahrens */ 805fa9e4066Sahrens vdev_add_child(parent, vd); 806fa9e4066Sahrens 80799653d4eSeschrock *vdp = vd; 80899653d4eSeschrock 80999653d4eSeschrock return (0); 810fa9e4066Sahrens } 811fa9e4066Sahrens 812fa9e4066Sahrens void 813fa9e4066Sahrens vdev_free(vdev_t *vd) 814fa9e4066Sahrens { 8153d7072f8Seschrock spa_t *spa = vd->vdev_spa; 816*084fd14fSBrian Behlendorf 817094e47e9SGeorge Wilson ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 818*084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_trim_thread, ==, NULL); 819*084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 820fa9e4066Sahrens 821a3874b8bSToomas Soome /* 822a3874b8bSToomas Soome * Scan queues are normally destroyed at the end of a scan. If the 823a3874b8bSToomas Soome * queue exists here, that implies the vdev is being removed while 824a3874b8bSToomas Soome * the scan is still running. 825a3874b8bSToomas Soome */ 826a3874b8bSToomas Soome if (vd->vdev_scan_io_queue != NULL) { 827a3874b8bSToomas Soome mutex_enter(&vd->vdev_scan_io_queue_lock); 828a3874b8bSToomas Soome dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 829a3874b8bSToomas Soome vd->vdev_scan_io_queue = NULL; 830a3874b8bSToomas Soome mutex_exit(&vd->vdev_scan_io_queue_lock); 831a3874b8bSToomas Soome } 832a3874b8bSToomas Soome 833fa9e4066Sahrens /* 834fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 835fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 836fa9e4066Sahrens */ 837fa9e4066Sahrens vdev_close(vd); 838fa9e4066Sahrens 839e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 840b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 841fa9e4066Sahrens 842fa9e4066Sahrens /* 843fa9e4066Sahrens * Free all children. 844fa9e4066Sahrens */ 845573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 846fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 847fa9e4066Sahrens 848fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 849fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 850fa9e4066Sahrens 851fa9e4066Sahrens /* 852fa9e4066Sahrens * Discard allocation state. 853fa9e4066Sahrens */ 854a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 855fa9e4066Sahrens vdev_metaslab_fini(vd); 856a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 857a1521560SJeff Bonwick } 858fa9e4066Sahrens 859fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 860fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 861fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 862fa9e4066Sahrens 863fa9e4066Sahrens /* 864fa9e4066Sahrens * Remove this vdev from its parent's child list. 865fa9e4066Sahrens */ 866fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 867fa9e4066Sahrens 868fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 869e0f1c0afSOlaf Faaland ASSERT(!list_link_active(&vd->vdev_leaf_node)); 870fa9e4066Sahrens 8713d7072f8Seschrock /* 8723d7072f8Seschrock * Clean up vdev structure. 8733d7072f8Seschrock */ 8743d7072f8Seschrock vdev_queue_fini(vd); 8753d7072f8Seschrock vdev_cache_fini(vd); 8763d7072f8Seschrock 8773d7072f8Seschrock if (vd->vdev_path) 8783d7072f8Seschrock spa_strfree(vd->vdev_path); 8793d7072f8Seschrock if (vd->vdev_devid) 8803d7072f8Seschrock spa_strfree(vd->vdev_devid); 8813d7072f8Seschrock if (vd->vdev_physpath) 8823d7072f8Seschrock spa_strfree(vd->vdev_physpath); 8836809eb4eSEric Schrock if (vd->vdev_fru) 8846809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 8853d7072f8Seschrock 8863d7072f8Seschrock if (vd->vdev_isspare) 8873d7072f8Seschrock spa_spare_remove(vd); 888fa94a07fSbrendan if (vd->vdev_isl2cache) 889fa94a07fSbrendan spa_l2cache_remove(vd); 8903d7072f8Seschrock 8913d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 8923d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 8938ad4d6ddSJeff Bonwick 8943d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 8950713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 8968ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 8970713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 8980713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 8998ad4d6ddSJeff Bonwick } 9003d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 9018ad4d6ddSJeff Bonwick 9025cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 9035cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 9045cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 9055cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 9065cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 9075cabbc6bSPrashanth Sreenivasa } 9085cabbc6bSPrashanth Sreenivasa 9095cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 9105cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 9115cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 9125cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 9135cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 9145cabbc6bSPrashanth Sreenivasa } 9155cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 9165cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 9175cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 9185cabbc6bSPrashanth Sreenivasa 9193d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 9203d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 921e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 922a3874b8bSToomas Soome mutex_destroy(&vd->vdev_scan_io_queue_lock); 923094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_lock); 924094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_io_lock); 925094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_io_cv); 926094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_cv); 927*084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_lock); 928*084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_autotrim_lock); 929*084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_io_lock); 930*084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_cv); 931*084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_autotrim_cv); 932*084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_io_cv); 9333d7072f8Seschrock 9343d7072f8Seschrock if (vd == spa->spa_root_vdev) 9353d7072f8Seschrock spa->spa_root_vdev = NULL; 9363d7072f8Seschrock 9373d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 938fa9e4066Sahrens } 939fa9e4066Sahrens 940fa9e4066Sahrens /* 941fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 942fa9e4066Sahrens */ 943fa9e4066Sahrens static void 944fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 945fa9e4066Sahrens { 946fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 947fa9e4066Sahrens metaslab_t *msp; 948fa9e4066Sahrens vdev_t *vd; 949fa9e4066Sahrens int t; 950fa9e4066Sahrens 951fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 952fa9e4066Sahrens 953fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 954fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 955fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 956215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 957fa9e4066Sahrens 958fa9e4066Sahrens svd->vdev_ms_array = 0; 959fa9e4066Sahrens svd->vdev_ms_shift = 0; 960fa9e4066Sahrens svd->vdev_ms_count = 0; 961215198a6SJoe Stein svd->vdev_top_zap = 0; 962fa9e4066Sahrens 963cd0837ccSGeorge Wilson if (tvd->vdev_mg) 964cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 965fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 966fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 967fa9e4066Sahrens 968fa9e4066Sahrens svd->vdev_mg = NULL; 969fa9e4066Sahrens svd->vdev_ms = NULL; 970ecc2d604Sbonwick 971ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 972ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 973fa9e4066Sahrens 97486714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 97586714001SSerapheim Dimitropoulos svd->vdev_checkpoint_sm = NULL; 97686714001SSerapheim Dimitropoulos 977663207adSDon Brady tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 978663207adSDon Brady svd->vdev_alloc_bias = VDEV_BIAS_NONE; 979663207adSDon Brady 980fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 981fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 98299653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 983fa9e4066Sahrens 984fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 985fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 98699653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 987fa9e4066Sahrens 9883a4b1be9SMatthew Ahrens /* 9893a4b1be9SMatthew Ahrens * State which may be set on a top-level vdev that's in the 9903a4b1be9SMatthew Ahrens * process of being removed. 9913a4b1be9SMatthew Ahrens */ 9923a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_births_object); 9933a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 9943a4b1be9SMatthew Ahrens ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 9953a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 9963a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 9973a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 9983a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_removing); 9993a4b1be9SMatthew Ahrens tvd->vdev_removing = svd->vdev_removing; 10003a4b1be9SMatthew Ahrens tvd->vdev_indirect_config = svd->vdev_indirect_config; 10013a4b1be9SMatthew Ahrens tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 10023a4b1be9SMatthew Ahrens tvd->vdev_indirect_births = svd->vdev_indirect_births; 10033a4b1be9SMatthew Ahrens range_tree_swap(&svd->vdev_obsolete_segments, 10043a4b1be9SMatthew Ahrens &tvd->vdev_obsolete_segments); 10053a4b1be9SMatthew Ahrens tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 10063a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_mapping_object = 0; 10073a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_births_object = 0; 10083a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 10093a4b1be9SMatthew Ahrens svd->vdev_indirect_mapping = NULL; 10103a4b1be9SMatthew Ahrens svd->vdev_indirect_births = NULL; 10113a4b1be9SMatthew Ahrens svd->vdev_obsolete_sm = NULL; 10123a4b1be9SMatthew Ahrens svd->vdev_removing = 0; 10133a4b1be9SMatthew Ahrens 1014fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 1015fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1016fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1017fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1018fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1019fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1020fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1021fa9e4066Sahrens } 1022fa9e4066Sahrens 1023e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 1024fa9e4066Sahrens vdev_config_clean(svd); 1025fa9e4066Sahrens vdev_config_dirty(tvd); 1026fa9e4066Sahrens } 1027fa9e4066Sahrens 1028e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 1029e14bb325SJeff Bonwick vdev_state_clean(svd); 1030e14bb325SJeff Bonwick vdev_state_dirty(tvd); 1031e14bb325SJeff Bonwick } 1032e14bb325SJeff Bonwick 103399653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 103499653d4eSeschrock svd->vdev_deflate_ratio = 0; 10358654d025Sperrin 10368654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 10378654d025Sperrin svd->vdev_islog = 0; 1038a3874b8bSToomas Soome 1039a3874b8bSToomas Soome dsl_scan_io_queue_vdev_xfer(svd, tvd); 1040fa9e4066Sahrens } 1041fa9e4066Sahrens 1042fa9e4066Sahrens static void 1043fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 1044fa9e4066Sahrens { 1045fa9e4066Sahrens if (vd == NULL) 1046fa9e4066Sahrens return; 1047fa9e4066Sahrens 1048fa9e4066Sahrens vd->vdev_top = tvd; 1049fa9e4066Sahrens 1050573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1051fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 1052fa9e4066Sahrens } 1053fa9e4066Sahrens 1054fa9e4066Sahrens /* 1055fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 1056fa9e4066Sahrens */ 1057fa9e4066Sahrens vdev_t * 1058fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1059fa9e4066Sahrens { 1060fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 1061fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 1062fa9e4066Sahrens vdev_t *mvd; 1063fa9e4066Sahrens 1064e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1065fa9e4066Sahrens 1066fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1067ecc2d604Sbonwick 1068ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 1069573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 10704263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 10715cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 1072ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 1073ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 107488ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 1075ecc2d604Sbonwick 1076fa9e4066Sahrens vdev_remove_child(pvd, cvd); 1077fa9e4066Sahrens vdev_add_child(pvd, mvd); 1078fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 1079fa9e4066Sahrens vdev_add_child(mvd, cvd); 1080fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1081fa9e4066Sahrens 1082fa9e4066Sahrens if (mvd == mvd->vdev_top) 1083fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 1084fa9e4066Sahrens 1085fa9e4066Sahrens return (mvd); 1086fa9e4066Sahrens } 1087fa9e4066Sahrens 1088fa9e4066Sahrens /* 1089fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 1090fa9e4066Sahrens */ 1091fa9e4066Sahrens void 1092fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 1093fa9e4066Sahrens { 1094fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 1095fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 1096fa9e4066Sahrens 1097e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1098fa9e4066Sahrens 1099fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 1100fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 110199653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 110299653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 1103ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 1104fa9e4066Sahrens 1105fa9e4066Sahrens vdev_remove_child(mvd, cvd); 1106fa9e4066Sahrens vdev_remove_child(pvd, mvd); 11078ad4d6ddSJeff Bonwick 110899653d4eSeschrock /* 1109e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1110e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 1111e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 1112e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 111399653d4eSeschrock */ 11148ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 11158ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 11161195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 11178ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 11188ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 11198ad4d6ddSJeff Bonwick } 1120e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 1121e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 1122fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1123fa9e4066Sahrens 1124fa9e4066Sahrens if (cvd == cvd->vdev_top) 1125fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 1126fa9e4066Sahrens 1127fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 1128fa9e4066Sahrens vdev_free(mvd); 1129fa9e4066Sahrens } 1130fa9e4066Sahrens 1131663207adSDon Brady static void 1132663207adSDon Brady vdev_metaslab_group_create(vdev_t *vd) 1133663207adSDon Brady { 1134663207adSDon Brady spa_t *spa = vd->vdev_spa; 1135663207adSDon Brady 1136663207adSDon Brady /* 1137663207adSDon Brady * metaslab_group_create was delayed until allocation bias was available 1138663207adSDon Brady */ 1139663207adSDon Brady if (vd->vdev_mg == NULL) { 1140663207adSDon Brady metaslab_class_t *mc; 1141663207adSDon Brady 1142663207adSDon Brady if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1143663207adSDon Brady vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1144663207adSDon Brady 1145663207adSDon Brady ASSERT3U(vd->vdev_islog, ==, 1146663207adSDon Brady (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1147663207adSDon Brady 1148663207adSDon Brady switch (vd->vdev_alloc_bias) { 1149663207adSDon Brady case VDEV_BIAS_LOG: 1150663207adSDon Brady mc = spa_log_class(spa); 1151663207adSDon Brady break; 1152663207adSDon Brady case VDEV_BIAS_SPECIAL: 1153663207adSDon Brady mc = spa_special_class(spa); 1154663207adSDon Brady break; 1155663207adSDon Brady case VDEV_BIAS_DEDUP: 1156663207adSDon Brady mc = spa_dedup_class(spa); 1157663207adSDon Brady break; 1158663207adSDon Brady default: 1159663207adSDon Brady mc = spa_normal_class(spa); 1160663207adSDon Brady } 1161663207adSDon Brady 1162663207adSDon Brady vd->vdev_mg = metaslab_group_create(mc, vd, 1163663207adSDon Brady spa->spa_alloc_count); 1164663207adSDon Brady 1165663207adSDon Brady /* 1166663207adSDon Brady * The spa ashift values currently only reflect the 1167663207adSDon Brady * general vdev classes. Class destination is late 1168663207adSDon Brady * binding so ashift checking had to wait until now 1169663207adSDon Brady */ 1170663207adSDon Brady if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1171663207adSDon Brady mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1172663207adSDon Brady if (vd->vdev_ashift > spa->spa_max_ashift) 1173663207adSDon Brady spa->spa_max_ashift = vd->vdev_ashift; 1174663207adSDon Brady if (vd->vdev_ashift < spa->spa_min_ashift) 1175663207adSDon Brady spa->spa_min_ashift = vd->vdev_ashift; 1176663207adSDon Brady } 1177663207adSDon Brady } 1178663207adSDon Brady } 1179663207adSDon Brady 1180ea8dc4b6Seschrock int 1181fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1182fa9e4066Sahrens { 1183fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1184ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1185ecc2d604Sbonwick uint64_t m; 1186fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 1187fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1188ecc2d604Sbonwick metaslab_t **mspp; 1189ecc2d604Sbonwick int error; 1190663207adSDon Brady boolean_t expanding = (oldc != 0); 1191fa9e4066Sahrens 1192a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1193a1521560SJeff Bonwick 119488ecc943SGeorge Wilson /* 119588ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 119688ecc943SGeorge Wilson */ 119788ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 11980e34b6a7Sbonwick return (0); 11990e34b6a7Sbonwick 120088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 120188ecc943SGeorge Wilson 1202fa9e4066Sahrens ASSERT(oldc <= newc); 1203fa9e4066Sahrens 1204ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1205fa9e4066Sahrens 1206663207adSDon Brady if (expanding) { 1207ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1208ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1209ecc2d604Sbonwick } 1210fa9e4066Sahrens 1211ecc2d604Sbonwick vd->vdev_ms = mspp; 1212ecc2d604Sbonwick vd->vdev_ms_count = newc; 1213ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 12140713e232SGeorge Wilson uint64_t object = 0; 12150713e232SGeorge Wilson 12165cabbc6bSPrashanth Sreenivasa /* 12175cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 12185cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 12195cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 12205cabbc6bSPrashanth Sreenivasa */ 12215cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1222ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 12237bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 12247bfdf011SNeil Perrin DMU_READ_PREFETCH); 12253ee8c80cSPavel Zakharov if (error != 0) { 12263ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 12273ee8c80cSPavel Zakharov "array [error=%d]", error); 1228ecc2d604Sbonwick return (error); 12293ee8c80cSPavel Zakharov } 1230fa9e4066Sahrens } 12311e9bd7ecSPrakash Surya 1232663207adSDon Brady #ifndef _KERNEL 1233663207adSDon Brady /* 1234663207adSDon Brady * To accomodate zdb_leak_init() fake indirect 1235663207adSDon Brady * metaslabs, we allocate a metaslab group for 1236663207adSDon Brady * indirect vdevs which normally don't have one. 1237663207adSDon Brady */ 1238663207adSDon Brady if (vd->vdev_mg == NULL) { 1239663207adSDon Brady ASSERT0(vdev_is_concrete(vd)); 1240663207adSDon Brady vdev_metaslab_group_create(vd); 1241663207adSDon Brady } 1242663207adSDon Brady #endif 12431e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 12441e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 12453ee8c80cSPavel Zakharov if (error != 0) { 12463ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 12473ee8c80cSPavel Zakharov error); 12481e9bd7ecSPrakash Surya return (error); 12493ee8c80cSPavel Zakharov } 1250fa9e4066Sahrens } 1251fa9e4066Sahrens 1252a1521560SJeff Bonwick if (txg == 0) 1253a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1254a1521560SJeff Bonwick 12553f9d6ad7SLin Ling /* 12563f9d6ad7SLin Ling * If the vdev is being removed we don't activate 12573f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 12583f9d6ad7SLin Ling * allocations are performed on this device. 12593f9d6ad7SLin Ling */ 1260663207adSDon Brady if (!expanding && !vd->vdev_removing) { 1261a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1262663207adSDon Brady } 1263a1521560SJeff Bonwick 1264a1521560SJeff Bonwick if (txg == 0) 1265a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1266a1521560SJeff Bonwick 1267ea8dc4b6Seschrock return (0); 1268fa9e4066Sahrens } 1269fa9e4066Sahrens 1270fa9e4066Sahrens void 1271fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1272fa9e4066Sahrens { 127386714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 127486714001SSerapheim Dimitropoulos ASSERT(spa_feature_is_active(vd->vdev_spa, 127586714001SSerapheim Dimitropoulos SPA_FEATURE_POOL_CHECKPOINT)); 127686714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 127786714001SSerapheim Dimitropoulos /* 127886714001SSerapheim Dimitropoulos * Even though we close the space map, we need to set its 127986714001SSerapheim Dimitropoulos * pointer to NULL. The reason is that vdev_metaslab_fini() 128086714001SSerapheim Dimitropoulos * may be called multiple times for certain operations 128186714001SSerapheim Dimitropoulos * (i.e. when destroying a pool) so we need to ensure that 128286714001SSerapheim Dimitropoulos * this clause never executes twice. This logic is similar 128386714001SSerapheim Dimitropoulos * to the one used for the vdev_ms clause below. 128486714001SSerapheim Dimitropoulos */ 128586714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 128686714001SSerapheim Dimitropoulos } 128786714001SSerapheim Dimitropoulos 1288fa9e4066Sahrens if (vd->vdev_ms != NULL) { 1289555d674dSSerapheim Dimitropoulos metaslab_group_t *mg = vd->vdev_mg; 1290555d674dSSerapheim Dimitropoulos metaslab_group_passivate(mg); 12915cabbc6bSPrashanth Sreenivasa 1292555d674dSSerapheim Dimitropoulos uint64_t count = vd->vdev_ms_count; 12935cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 12940713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 12950713e232SGeorge Wilson if (msp != NULL) 12960713e232SGeorge Wilson metaslab_fini(msp); 12970713e232SGeorge Wilson } 1298fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1299fa9e4066Sahrens vd->vdev_ms = NULL; 13005cabbc6bSPrashanth Sreenivasa 13015cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1302555d674dSSerapheim Dimitropoulos 1303555d674dSSerapheim Dimitropoulos for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 1304555d674dSSerapheim Dimitropoulos ASSERT0(mg->mg_histogram[i]); 1305fa9e4066Sahrens } 13065cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1307fa9e4066Sahrens } 1308fa9e4066Sahrens 1309e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1310e14bb325SJeff Bonwick boolean_t vps_readable; 1311e14bb325SJeff Bonwick boolean_t vps_writeable; 1312e14bb325SJeff Bonwick int vps_flags; 1313e14bb325SJeff Bonwick } vdev_probe_stats_t; 1314e14bb325SJeff Bonwick 1315e14bb325SJeff Bonwick static void 1316e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 13170a4e9518Sgw { 13188ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1319a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1320e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1321a3f829aeSBill Moore 1322a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1323e14bb325SJeff Bonwick 1324e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1325e14bb325SJeff Bonwick if (zio->io_error == 0) 1326e14bb325SJeff Bonwick vps->vps_readable = 1; 13278ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1328a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1329770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1330e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1331e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1332e14bb325SJeff Bonwick } else { 1333770499e1SDan Kimmel abd_free(zio->io_abd); 1334e14bb325SJeff Bonwick } 1335e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1336e14bb325SJeff Bonwick if (zio->io_error == 0) 1337e14bb325SJeff Bonwick vps->vps_writeable = 1; 1338770499e1SDan Kimmel abd_free(zio->io_abd); 1339e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1340a3f829aeSBill Moore zio_t *pio; 1341e14bb325SJeff Bonwick 1342e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1343e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1344e14bb325SJeff Bonwick 1345e14bb325SJeff Bonwick if (vdev_readable(vd) && 13468ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1347e14bb325SJeff Bonwick zio->io_error = 0; 1348e14bb325SJeff Bonwick } else { 1349e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 13503ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1351e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1352eb633035STom Caputi spa, vd, NULL, NULL, 0, 0); 1353be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1354e14bb325SJeff Bonwick } 1355a3f829aeSBill Moore 1356a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1357a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1358a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1359a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1360a3f829aeSBill Moore 13610f7643c7SGeorge Wilson zio_link_t *zl = NULL; 13620f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1363a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1364be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1365a3f829aeSBill Moore 1366e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1367e14bb325SJeff Bonwick } 1368e14bb325SJeff Bonwick } 13690a4e9518Sgw 1370e14bb325SJeff Bonwick /* 1371f7170741SWill Andrews * Determine whether this device is accessible. 1372f7170741SWill Andrews * 1373f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1374f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1375f7170741SWill Andrews * a VTOC. 1376e14bb325SJeff Bonwick */ 1377e14bb325SJeff Bonwick zio_t * 1378a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1379e14bb325SJeff Bonwick { 1380e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1381a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1382a3f829aeSBill Moore zio_t *pio; 1383a3f829aeSBill Moore 1384a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 13850a4e9518Sgw 1386a3f829aeSBill Moore /* 1387a3f829aeSBill Moore * Don't probe the probe. 1388a3f829aeSBill Moore */ 1389a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1390a3f829aeSBill Moore return (NULL); 1391e14bb325SJeff Bonwick 1392a3f829aeSBill Moore /* 1393a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1394a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1395a3f829aeSBill Moore * this vdev will become parents of the probe io. 1396a3f829aeSBill Moore */ 1397a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1398e14bb325SJeff Bonwick 1399a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1400a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1401a3f829aeSBill Moore 1402a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1403a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 14048956713aSEric Schrock ZIO_FLAG_TRYHARD; 1405a3f829aeSBill Moore 1406a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1407a3f829aeSBill Moore /* 1408a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1409a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1410a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1411a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1412a3f829aeSBill Moore * any zio looking at these values can assume that 1413a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1414a3f829aeSBill Moore * important because when a device has intermittent 1415a3f829aeSBill Moore * connectivity problems, we want to ensure that 1416a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1417a3f829aeSBill Moore * the zio (EIO). 1418a3f829aeSBill Moore * 1419a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1420a3f829aeSBill Moore * values so the probe can reevaluate from first 1421a3f829aeSBill Moore * principles. 1422a3f829aeSBill Moore */ 1423a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1424a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1425a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1426a3f829aeSBill Moore } 1427a3f829aeSBill Moore 1428a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1429a3f829aeSBill Moore vdev_probe_done, vps, 1430a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1431a3f829aeSBill Moore 143298d1cbfeSGeorge Wilson /* 143398d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 143498d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 143598d1cbfeSGeorge Wilson */ 1436a3f829aeSBill Moore if (zio != NULL) { 1437a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1438a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1439a3f829aeSBill Moore } 1440e14bb325SJeff Bonwick } 1441e14bb325SJeff Bonwick 1442a3f829aeSBill Moore if (zio != NULL) 1443a3f829aeSBill Moore zio_add_child(zio, pio); 1444e14bb325SJeff Bonwick 1445a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1446e14bb325SJeff Bonwick 1447a3f829aeSBill Moore if (vps == NULL) { 1448a3f829aeSBill Moore ASSERT(zio != NULL); 1449a3f829aeSBill Moore return (NULL); 1450a3f829aeSBill Moore } 1451e14bb325SJeff Bonwick 1452e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1453a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1454e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1455770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1456770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1457e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1458e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1459e14bb325SJeff Bonwick } 1460e14bb325SJeff Bonwick 1461a3f829aeSBill Moore if (zio == NULL) 1462a3f829aeSBill Moore return (pio); 1463a3f829aeSBill Moore 1464a3f829aeSBill Moore zio_nowait(pio); 1465a3f829aeSBill Moore return (NULL); 14660a4e9518Sgw } 14670a4e9518Sgw 1468f64c0e34SEric Taylor static void 1469f64c0e34SEric Taylor vdev_open_child(void *arg) 1470f64c0e34SEric Taylor { 1471f64c0e34SEric Taylor vdev_t *vd = arg; 1472f64c0e34SEric Taylor 1473f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1474f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1475f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1476f64c0e34SEric Taylor } 1477f64c0e34SEric Taylor 1478681d9761SEric Taylor boolean_t 1479681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1480681d9761SEric Taylor { 1481681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1482681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1483681d9761SEric Taylor return (B_TRUE); 1484681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1485681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1486681d9761SEric Taylor return (B_TRUE); 1487681d9761SEric Taylor return (B_FALSE); 1488681d9761SEric Taylor } 1489681d9761SEric Taylor 1490f64c0e34SEric Taylor void 1491f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1492f64c0e34SEric Taylor { 1493f64c0e34SEric Taylor taskq_t *tq; 1494f64c0e34SEric Taylor int children = vd->vdev_children; 1495f64c0e34SEric Taylor 1496681d9761SEric Taylor /* 1497681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1498681d9761SEric Taylor * in a single thread so that the same thread holds the 1499681d9761SEric Taylor * spa_namespace_lock 1500681d9761SEric Taylor */ 1501681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 150212a8814cSTom Caputi retry_sync: 1503681d9761SEric Taylor for (int c = 0; c < children; c++) 1504681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1505681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 150612a8814cSTom Caputi } else { 150712a8814cSTom Caputi tq = taskq_create("vdev_open", children, minclsyspri, 150812a8814cSTom Caputi children, children, TASKQ_PREPOPULATE); 150912a8814cSTom Caputi if (tq == NULL) 151012a8814cSTom Caputi goto retry_sync; 151112a8814cSTom Caputi 151212a8814cSTom Caputi for (int c = 0; c < children; c++) 151312a8814cSTom Caputi VERIFY(taskq_dispatch(tq, vdev_open_child, 151412a8814cSTom Caputi vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); 151512a8814cSTom Caputi 151612a8814cSTom Caputi taskq_destroy(tq); 1517681d9761SEric Taylor } 1518f64c0e34SEric Taylor 151912a8814cSTom Caputi vd->vdev_nonrot = B_TRUE; 1520f64c0e34SEric Taylor 152112a8814cSTom Caputi for (int c = 0; c < children; c++) 152212a8814cSTom Caputi vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; 1523f64c0e34SEric Taylor } 1524f64c0e34SEric Taylor 15255cabbc6bSPrashanth Sreenivasa /* 15265cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 15275cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 15285cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 15295cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 15305cabbc6bSPrashanth Sreenivasa */ 15315cabbc6bSPrashanth Sreenivasa static void 15325cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 15335cabbc6bSPrashanth Sreenivasa { 15345cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 15355cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 15365cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 15375cabbc6bSPrashanth Sreenivasa } 15385cabbc6bSPrashanth Sreenivasa } 15395cabbc6bSPrashanth Sreenivasa 1540fa9e4066Sahrens /* 1541fa9e4066Sahrens * Prepare a virtual device for access. 1542fa9e4066Sahrens */ 1543fa9e4066Sahrens int 1544fa9e4066Sahrens vdev_open(vdev_t *vd) 1545fa9e4066Sahrens { 15468ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1547fa9e4066Sahrens int error; 1548fa9e4066Sahrens uint64_t osize = 0; 15494263d13fSGeorge Wilson uint64_t max_osize = 0; 15504263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1551ecc2d604Sbonwick uint64_t ashift = 0; 1552fa9e4066Sahrens 1553f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1554f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1555fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1556fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1557fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1558fa9e4066Sahrens 1559fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1560e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1561e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1562573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1563fa9e4066Sahrens 1564069f55e2SEric Schrock /* 1565069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1566069f55e2SEric Schrock * faulted, bail out of the open. 1567069f55e2SEric Schrock */ 15683d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 15693d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1570069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1571069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 15723d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1573069f55e2SEric Schrock vd->vdev_label_aux); 1574be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 15753d7072f8Seschrock } else if (vd->vdev_offline) { 1576fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1577ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1578be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1579fa9e4066Sahrens } 1580fa9e4066Sahrens 15814263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1582fa9e4066Sahrens 1583095bcd66SGeorge Wilson /* 1584095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1585095bcd66SGeorge Wilson * the vdev on error. 1586095bcd66SGeorge Wilson */ 1587095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1588ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 15898956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1590ea8dc4b6Seschrock 1591fa9e4066Sahrens if (error) { 15923d7072f8Seschrock if (vd->vdev_removed && 15933d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 15943d7072f8Seschrock vd->vdev_removed = B_FALSE; 15953d7072f8Seschrock 15966f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 15976f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 15986f793812SPavel Zakharov vd->vdev_stat.vs_aux); 15996f793812SPavel Zakharov } else { 16006f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 16016f793812SPavel Zakharov vd->vdev_stat.vs_aux); 16026f793812SPavel Zakharov } 1603fa9e4066Sahrens return (error); 1604fa9e4066Sahrens } 1605fa9e4066Sahrens 16063d7072f8Seschrock vd->vdev_removed = B_FALSE; 16073d7072f8Seschrock 1608096d22d4SEric Schrock /* 1609096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1610096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1611096d22d4SEric Schrock */ 1612096d22d4SEric Schrock if (vd->vdev_faulted) { 1613096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1614096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1615096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1616096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1617096d22d4SEric Schrock vd->vdev_label_aux); 1618be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1619096d22d4SEric Schrock } 1620096d22d4SEric Schrock 16213d7072f8Seschrock if (vd->vdev_degraded) { 16223d7072f8Seschrock ASSERT(vd->vdev_children == 0); 16233d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 16243d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 16253d7072f8Seschrock } else { 1626069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 16273d7072f8Seschrock } 1628fa9e4066Sahrens 162988ecc943SGeorge Wilson /* 163088ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 163188ecc943SGeorge Wilson */ 163288ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 163388ecc943SGeorge Wilson return (0); 163488ecc943SGeorge Wilson 1635573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1636ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1637ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1638ea8dc4b6Seschrock VDEV_AUX_NONE); 1639ea8dc4b6Seschrock break; 1640ea8dc4b6Seschrock } 1641573ca77eSGeorge Wilson } 1642fa9e4066Sahrens 1643fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 16444263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1645fa9e4066Sahrens 1646fa9e4066Sahrens if (vd->vdev_children == 0) { 1647fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1648ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1649ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1650be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1651fa9e4066Sahrens } 1652fa9e4066Sahrens psize = osize; 1653fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 16544263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 16554263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1656fa9e4066Sahrens } else { 1657ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1658fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1659ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1660ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1661be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1662fa9e4066Sahrens } 1663fa9e4066Sahrens psize = 0; 1664fa9e4066Sahrens asize = osize; 16654263d13fSGeorge Wilson max_asize = max_osize; 1666fa9e4066Sahrens } 1667fa9e4066Sahrens 1668fa9e4066Sahrens vd->vdev_psize = psize; 1669fa9e4066Sahrens 1670573ca77eSGeorge Wilson /* 1671c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1672573ca77eSGeorge Wilson */ 1673573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1674573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1675573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1676be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1677573ca77eSGeorge Wilson } 1678573ca77eSGeorge Wilson 1679fa9e4066Sahrens if (vd->vdev_asize == 0) { 1680fa9e4066Sahrens /* 1681fa9e4066Sahrens * This is the first-ever open, so use the computed values. 16825711d393Sloli * For compatibility, a different ashift can be requested. 1683fa9e4066Sahrens */ 1684fa9e4066Sahrens vd->vdev_asize = asize; 16854263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 16865711d393Sloli if (vd->vdev_ashift == 0) { 16875711d393Sloli vd->vdev_ashift = ashift; /* use detected value */ 16885711d393Sloli } 16895711d393Sloli if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 16905711d393Sloli vd->vdev_ashift > ASHIFT_MAX)) { 16915711d393Sloli vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 16925711d393Sloli VDEV_AUX_BAD_ASHIFT); 16935711d393Sloli return (SET_ERROR(EDOM)); 16945711d393Sloli } 1695fa9e4066Sahrens } else { 1696fa9e4066Sahrens /* 16972384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 16982384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 16995711d393Sloli * post an event instead. 1700fa9e4066Sahrens */ 17012384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 17022384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 17035711d393Sloli zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 17045711d393Sloli spa, vd, NULL, NULL, 0, 0); 1705fa9e4066Sahrens } 17065711d393Sloli 17074263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1708573ca77eSGeorge Wilson } 1709fa9e4066Sahrens 1710573ca77eSGeorge Wilson /* 1711c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1712c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1713c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1714c040c10cSSteven Hartland * making the additional space available. 1715c040c10cSSteven Hartland * 1716c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1717c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1718c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1719c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1720c040c10cSSteven Hartland * vdev_min_asize. 1721573ca77eSGeorge Wilson */ 1722c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1723c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1724c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1725c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1726573ca77eSGeorge Wilson vd->vdev_asize = asize; 1727fa9e4066Sahrens 1728573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1729fa9e4066Sahrens 17300a4e9518Sgw /* 17310a4e9518Sgw * Ensure we can issue some IO before declaring the 17320a4e9518Sgw * vdev open for business. 17330a4e9518Sgw */ 1734e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1735e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 173698d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 173798d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 17380a4e9518Sgw return (error); 17390a4e9518Sgw } 17400a4e9518Sgw 174181cd5c55SMatthew Ahrens /* 174281cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 1743663207adSDon Brady * 1744663207adSDon Brady * DJB - TBD these should perhaps be tracked per allocation class 1745663207adSDon Brady * (e.g. spa_min_ashift is used to round up post compression buffers) 174681cd5c55SMatthew Ahrens */ 174781cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1748663207adSDon Brady vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1749663207adSDon Brady vd->vdev_aux == NULL) { 175081cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 175181cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 175281cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 175381cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 175481cd5c55SMatthew Ahrens } 175581cd5c55SMatthew Ahrens 1756088f3894Sahrens /* 1757088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 17588ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 17598ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1760088f3894Sahrens */ 17618ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1762e4c795beSTom Caputi vdev_resilver_needed(vd, NULL, NULL)) { 1763e4c795beSTom Caputi if (dsl_scan_resilvering(spa->spa_dsl_pool) && 1764e4c795beSTom Caputi spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) 1765e4c795beSTom Caputi vdev_set_deferred_resilver(spa, vd); 1766e4c795beSTom Caputi else 1767e4c795beSTom Caputi spa_async_request(spa, SPA_ASYNC_RESILVER); 1768e4c795beSTom Caputi } 1769088f3894Sahrens 1770fa9e4066Sahrens return (0); 1771fa9e4066Sahrens } 1772fa9e4066Sahrens 1773560e6e96Seschrock /* 1774560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 17756f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 17763d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1777560e6e96Seschrock * 1778560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1779560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1780560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1781560e6e96Seschrock * will be updated but the function will return 0. 1782560e6e96Seschrock */ 1783560e6e96Seschrock int 17846f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1785560e6e96Seschrock { 1786560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1787560e6e96Seschrock nvlist_t *label; 17886f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1789560e6e96Seschrock uint64_t state; 17906f793812SPavel Zakharov nvlist_t *nvl; 17916f793812SPavel Zakharov uint64_t txg; 1792560e6e96Seschrock 17936f793812SPavel Zakharov if (vdev_validate_skip) 17946f793812SPavel Zakharov return (0); 17956f793812SPavel Zakharov 17966f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 17976f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1798be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1799560e6e96Seschrock 1800b5989ec7Seschrock /* 1801b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1802b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1803b5989ec7Seschrock * overwrite the previous state. 1804b5989ec7Seschrock */ 18056f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 18066f793812SPavel Zakharov return (0); 1807560e6e96Seschrock 18086f793812SPavel Zakharov /* 18096f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 18106f793812SPavel Zakharov * was modified at a point after the current txg. 1811d1de72cfSPavel Zakharov * If config lock is not held do not check for the txg. spa_sync could 1812d1de72cfSPavel Zakharov * be updating the vdev's label before updating spa_last_synced_txg. 18136f793812SPavel Zakharov */ 1814d1de72cfSPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1815d1de72cfSPavel Zakharov spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 18166f793812SPavel Zakharov txg = UINT64_MAX; 18176f793812SPavel Zakharov else 18186f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1819560e6e96Seschrock 18206f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 18216f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 18226f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 1823b6bf6e15SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1824b6bf6e15SPavel Zakharov "txg %llu", (u_longlong_t)txg); 18256f793812SPavel Zakharov return (0); 18266f793812SPavel Zakharov } 18271195e687SMark J Musante 18286f793812SPavel Zakharov /* 18296f793812SPavel Zakharov * Determine if this vdev has been split off into another 18306f793812SPavel Zakharov * pool. If so, then refuse to open it. 18316f793812SPavel Zakharov */ 18326f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 18336f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 18346f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18356f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 18366f793812SPavel Zakharov nvlist_free(label); 18376f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 18386f793812SPavel Zakharov return (0); 18396f793812SPavel Zakharov } 1840560e6e96Seschrock 18416f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 18426f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18436f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18446f793812SPavel Zakharov nvlist_free(label); 18456f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18466f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 18476f793812SPavel Zakharov return (0); 18486f793812SPavel Zakharov } 18491195e687SMark J Musante 18506f793812SPavel Zakharov /* 18516f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 18526f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 18536f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 18546f793812SPavel Zakharov * cached config. The check will be performed again once we have the 18556f793812SPavel Zakharov * trusted config from the MOS. 18566f793812SPavel Zakharov */ 18576f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 18586f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18596f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18606f793812SPavel Zakharov nvlist_free(label); 18616f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 18626f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 18636f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 18646f793812SPavel Zakharov return (0); 18656f793812SPavel Zakharov } 18666f793812SPavel Zakharov 18676f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 18686f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 18696f793812SPavel Zakharov &aux_guid) != 0) 18706f793812SPavel Zakharov aux_guid = 0; 18716f793812SPavel Zakharov 18726f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 18736f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18746f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18756f793812SPavel Zakharov nvlist_free(label); 18766f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18776f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 18786f793812SPavel Zakharov return (0); 18796f793812SPavel Zakharov } 18806f793812SPavel Zakharov 18816f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 18826f793812SPavel Zakharov != 0) { 18836f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18846f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18856f793812SPavel Zakharov nvlist_free(label); 18866f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18876f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 18886f793812SPavel Zakharov return (0); 18896f793812SPavel Zakharov } 18906f793812SPavel Zakharov 18916f793812SPavel Zakharov /* 18926f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 18936f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 18946f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 18956f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 18966f793812SPavel Zakharov * vdev, we can safely compare to that instead. 18976f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 18986f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 18996f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 19006f793812SPavel Zakharov * extreme rewind. 19016f793812SPavel Zakharov * 19026f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 19036f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 19046f793812SPavel Zakharov * corrupt if it is partway through a split operation. 19056f793812SPavel Zakharov */ 19066f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 19076f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 19086f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 19096f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 19106f793812SPavel Zakharov mismatch = B_TRUE; 19116f793812SPavel Zakharov } else { 19126f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 19136f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 19146f793812SPavel Zakharov mismatch = B_TRUE; 1915560e6e96Seschrock } 1916560e6e96Seschrock 19176f793812SPavel Zakharov if (mismatch) { 1918560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1919560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1920560e6e96Seschrock nvlist_free(label); 19216f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 19226f793812SPavel Zakharov "doesn't match label guid"); 19236f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 19246f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 19256f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 19266f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 19276f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 19286f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1929560e6e96Seschrock return (0); 1930560e6e96Seschrock } 19316f793812SPavel Zakharov } 1932560e6e96Seschrock 19336f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 19346f793812SPavel Zakharov &state) != 0) { 19356f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 19366f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1937560e6e96Seschrock nvlist_free(label); 19386f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 19396f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 19406f793812SPavel Zakharov return (0); 19416f793812SPavel Zakharov } 1942560e6e96Seschrock 19436f793812SPavel Zakharov nvlist_free(label); 19446f793812SPavel Zakharov 19456f793812SPavel Zakharov /* 19466f793812SPavel Zakharov * If this is a verbatim import, no need to check the 19476f793812SPavel Zakharov * state of the pool. 19486f793812SPavel Zakharov */ 19496f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 19506f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 19516f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 19526f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 19536f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 19546f793812SPavel Zakharov return (SET_ERROR(EBADF)); 19556f793812SPavel Zakharov } 19566f793812SPavel Zakharov 19576f793812SPavel Zakharov /* 19586f793812SPavel Zakharov * If we were able to open and validate a vdev that was 19596f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 19606f793812SPavel Zakharov * now. 19616f793812SPavel Zakharov */ 19626f793812SPavel Zakharov if (vd->vdev_not_present) 19636f793812SPavel Zakharov vd->vdev_not_present = 0; 19646f793812SPavel Zakharov 19656f793812SPavel Zakharov return (0); 19666f793812SPavel Zakharov } 19676f793812SPavel Zakharov 19686f793812SPavel Zakharov static void 19696f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 19706f793812SPavel Zakharov { 19716f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 19726f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 19736f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 19746f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 19756f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 19766f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 19776f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19783ee8c80cSPavel Zakharov } 19796f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 19806f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19816f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 19826f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 19836f793812SPavel Zakharov } 19846f793812SPavel Zakharov } 1985560e6e96Seschrock 19866f793812SPavel Zakharov /* 19876f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 19886f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 19896f793812SPavel Zakharov * paths from userland config into MOS config. 19906f793812SPavel Zakharov */ 19916f793812SPavel Zakharov int 19926f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 19936f793812SPavel Zakharov { 19946f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 19956f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 19966f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 19976f793812SPavel Zakharov return (0); 19986f793812SPavel Zakharov 19996f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 20006f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 20016f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 20026f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20036f793812SPavel Zakharov } 20046f793812SPavel Zakharov 20056f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 20066f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 20076f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 20086f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 20096f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20106f793812SPavel Zakharov } 20116f793812SPavel Zakharov 20126f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 20136f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 20146f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 20156f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 20166f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 201751ece835Seschrock } 2018560e6e96Seschrock 20196f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 20206f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 20216f793812SPavel Zakharov dvd->vdev_child[i]); 20226f793812SPavel Zakharov if (error != 0) 20236f793812SPavel Zakharov return (error); 20246f793812SPavel Zakharov } 20256f793812SPavel Zakharov 20266f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 20276f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 20286f793812SPavel Zakharov 2029560e6e96Seschrock return (0); 2030560e6e96Seschrock } 2031560e6e96Seschrock 20326f793812SPavel Zakharov static void 20336f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 20346f793812SPavel Zakharov { 20356f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 20366f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 20376f793812SPavel Zakharov 20386f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 20396f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 20406f793812SPavel Zakharov } 20416f793812SPavel Zakharov 20426f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 20436f793812SPavel Zakharov return; 20446f793812SPavel Zakharov 20456f793812SPavel Zakharov /* 20466f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 20476f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 20486f793812SPavel Zakharov * step outside of it. 20496f793812SPavel Zakharov */ 20506f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 20516f793812SPavel Zakharov 20526f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 20536f793812SPavel Zakharov return; 20546f793812SPavel Zakharov 20556f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 20566f793812SPavel Zakharov 20576f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 20586f793812SPavel Zakharov } 20596f793812SPavel Zakharov 20606f793812SPavel Zakharov /* 20616f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 20626f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 20636f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 20646f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 20656f793812SPavel Zakharov */ 20666f793812SPavel Zakharov void 20676f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 20686f793812SPavel Zakharov { 20696f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 20706f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 20716f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 20726f793812SPavel Zakharov 20736f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 20746f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 20756f793812SPavel Zakharov drvd->vdev_child[i]); 20766f793812SPavel Zakharov } 20776f793812SPavel Zakharov } 20786f793812SPavel Zakharov 2079fa9e4066Sahrens /* 2080fa9e4066Sahrens * Close a virtual device. 2081fa9e4066Sahrens */ 2082fa9e4066Sahrens void 2083fa9e4066Sahrens vdev_close(vdev_t *vd) 2084fa9e4066Sahrens { 20858ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 2086095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 20878ad4d6ddSJeff Bonwick 20888ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 20898ad4d6ddSJeff Bonwick 20901195e687SMark J Musante /* 20911195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 20921195e687SMark J Musante * going offline. 20931195e687SMark J Musante */ 2094095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 20951195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2096095bcd66SGeorge Wilson 2097fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 2098fa9e4066Sahrens 20993d7072f8Seschrock vdev_cache_purge(vd); 2100fa9e4066Sahrens 2101560e6e96Seschrock /* 2102573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 2103560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 2104560e6e96Seschrock * it's still faulted. 2105560e6e96Seschrock */ 2106560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 2107560e6e96Seschrock 2108fa9e4066Sahrens if (vd->vdev_offline) 2109fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 2110fa9e4066Sahrens else 2111fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 2112ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2113fa9e4066Sahrens } 2114fa9e4066Sahrens 2115dcba9f3fSGeorge Wilson void 2116dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 2117dcba9f3fSGeorge Wilson { 2118dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2119dcba9f3fSGeorge Wilson 2120dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2121dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2122dcba9f3fSGeorge Wilson return; 2123dcba9f3fSGeorge Wilson 2124dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2125dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 2126dcba9f3fSGeorge Wilson 2127dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2128dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 2129dcba9f3fSGeorge Wilson } 2130dcba9f3fSGeorge Wilson 2131dcba9f3fSGeorge Wilson void 2132dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 2133dcba9f3fSGeorge Wilson { 2134dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2135dcba9f3fSGeorge Wilson 2136dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2137dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2138dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 2139dcba9f3fSGeorge Wilson 2140dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2141dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 2142dcba9f3fSGeorge Wilson } 2143dcba9f3fSGeorge Wilson 2144095bcd66SGeorge Wilson /* 2145095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 2146095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 2147095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2148095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 2149095bcd66SGeorge Wilson */ 2150fa9e4066Sahrens void 2151ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 2152fa9e4066Sahrens { 2153ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 2154fa9e4066Sahrens 2155e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2156ea8dc4b6Seschrock 21571195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 21581195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 2159fa9e4066Sahrens vdev_close(vd); 2160fa9e4066Sahrens (void) vdev_open(vd); 2161fa9e4066Sahrens 216239c23413Seschrock /* 216339c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 216439c23413Seschrock * Otherwise, a device with an invalid label could be successfully 216539c23413Seschrock * opened in response to vdev_reopen(). 216639c23413Seschrock */ 2167c5904d13Seschrock if (vd->vdev_aux) { 2168c5904d13Seschrock (void) vdev_validate_aux(vd); 2169e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 21706809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 2171573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 2172573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 2173c5904d13Seschrock } else { 21746f793812SPavel Zakharov (void) vdev_validate(vd); 2175c5904d13Seschrock } 217639c23413Seschrock 2177fa9e4066Sahrens /* 21783d7072f8Seschrock * Reassess parent vdev's health. 2179fa9e4066Sahrens */ 21803d7072f8Seschrock vdev_propagate_state(vd); 2181fa9e4066Sahrens } 2182fa9e4066Sahrens 2183fa9e4066Sahrens int 218499653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2185fa9e4066Sahrens { 2186fa9e4066Sahrens int error; 2187fa9e4066Sahrens 2188fa9e4066Sahrens /* 2189fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 2190fa9e4066Sahrens * For a create, however, we want to fail the request if 2191fa9e4066Sahrens * there are any components we can't open. 2192fa9e4066Sahrens */ 2193fa9e4066Sahrens error = vdev_open(vd); 2194fa9e4066Sahrens 2195fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2196fa9e4066Sahrens vdev_close(vd); 2197fa9e4066Sahrens return (error ? error : ENXIO); 2198fa9e4066Sahrens } 2199fa9e4066Sahrens 2200fa9e4066Sahrens /* 22010713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 2202fa9e4066Sahrens */ 22030713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 22040713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 220539c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2206fa9e4066Sahrens vdev_close(vd); 2207fa9e4066Sahrens return (error); 2208fa9e4066Sahrens } 2209fa9e4066Sahrens 2210fa9e4066Sahrens return (0); 2211fa9e4066Sahrens } 2212fa9e4066Sahrens 22130e34b6a7Sbonwick void 2214573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 2215fa9e4066Sahrens { 221686714001SSerapheim Dimitropoulos uint64_t asize = vd->vdev_asize; 2217a0b03b16SSerapheim Dimitropoulos uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2218b4bf0cf0SDon Brady uint64_t ms_shift; 221986714001SSerapheim Dimitropoulos 2220a3874b8bSToomas Soome /* BEGIN CSTYLED */ 2221fa9e4066Sahrens /* 2222b4bf0cf0SDon Brady * There are two dimensions to the metaslab sizing calculation: 2223b4bf0cf0SDon Brady * the size of the metaslab and the count of metaslabs per vdev. 2224b4bf0cf0SDon Brady * 2225a0b03b16SSerapheim Dimitropoulos * The default values used below are a good balance between memory 2226a0b03b16SSerapheim Dimitropoulos * usage (larger metaslab size means more memory needed for loaded 2227a0b03b16SSerapheim Dimitropoulos * metaslabs; more metaslabs means more memory needed for the 2228a0b03b16SSerapheim Dimitropoulos * metaslab_t structs), metaslab load time (larger metaslabs take 2229a0b03b16SSerapheim Dimitropoulos * longer to load), and metaslab sync time (more metaslabs means 2230a0b03b16SSerapheim Dimitropoulos * more time spent syncing all of them). 2231a0b03b16SSerapheim Dimitropoulos * 2232a0b03b16SSerapheim Dimitropoulos * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2233a0b03b16SSerapheim Dimitropoulos * The range of the dimensions are as follows: 2234a0b03b16SSerapheim Dimitropoulos * 2235a0b03b16SSerapheim Dimitropoulos * 2^29 <= ms_size <= 2^34 2236b4bf0cf0SDon Brady * 16 <= ms_count <= 131,072 2237b4bf0cf0SDon Brady * 2238b4bf0cf0SDon Brady * On the lower end of vdev sizes, we aim for metaslabs sizes of 2239b4bf0cf0SDon Brady * at least 512MB (2^29) to minimize fragmentation effects when 2240b4bf0cf0SDon Brady * testing with smaller devices. However, the count constraint 2241b4bf0cf0SDon Brady * of at least 16 metaslabs will override this minimum size goal. 2242b4bf0cf0SDon Brady * 2243b4bf0cf0SDon Brady * On the upper end of vdev sizes, we aim for a maximum metaslab 2244a0b03b16SSerapheim Dimitropoulos * size of 16GB. However, we will cap the total count to 2^17 2245a0b03b16SSerapheim Dimitropoulos * metaslabs to keep our memory footprint in check and let the 2246a0b03b16SSerapheim Dimitropoulos * metaslab size grow from there if that limit is hit. 2247b4bf0cf0SDon Brady * 2248b4bf0cf0SDon Brady * The net effect of applying above constrains is summarized below. 2249b4bf0cf0SDon Brady * 2250663207adSDon Brady * vdev size metaslab count 2251a0b03b16SSerapheim Dimitropoulos * --------------|----------------- 2252663207adSDon Brady * < 8GB ~16 2253663207adSDon Brady * 8GB - 100GB one per 512MB 2254663207adSDon Brady * 100GB - 3TB ~200 2255663207adSDon Brady * 3TB - 2PB one per 16GB 2256663207adSDon Brady * > 2PB ~131,072 2257a0b03b16SSerapheim Dimitropoulos * -------------------------------- 2258a0b03b16SSerapheim Dimitropoulos * 2259a0b03b16SSerapheim Dimitropoulos * Finally, note that all of the above calculate the initial 2260a0b03b16SSerapheim Dimitropoulos * number of metaslabs. Expanding a top-level vdev will result 2261a0b03b16SSerapheim Dimitropoulos * in additional metaslabs being allocated making it possible 2262a0b03b16SSerapheim Dimitropoulos * to exceed the zfs_vdev_ms_count_limit. 2263a0b03b16SSerapheim Dimitropoulos */ 2264a3874b8bSToomas Soome /* END CSTYLED */ 2265a0b03b16SSerapheim Dimitropoulos 2266a0b03b16SSerapheim Dimitropoulos if (ms_count < zfs_vdev_min_ms_count) 2267a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2268a0b03b16SSerapheim Dimitropoulos else if (ms_count > zfs_vdev_default_ms_count) 2269a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2270b4bf0cf0SDon Brady else 2271a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_default_ms_shift; 2272b4bf0cf0SDon Brady 2273b4bf0cf0SDon Brady if (ms_shift < SPA_MAXBLOCKSHIFT) { 2274b4bf0cf0SDon Brady ms_shift = SPA_MAXBLOCKSHIFT; 2275a0b03b16SSerapheim Dimitropoulos } else if (ms_shift > zfs_vdev_max_ms_shift) { 2276a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_max_ms_shift; 2277b4bf0cf0SDon Brady /* cap the total count to constrain memory footprint */ 2278a0b03b16SSerapheim Dimitropoulos if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2279a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 228086714001SSerapheim Dimitropoulos } 228186714001SSerapheim Dimitropoulos 228286714001SSerapheim Dimitropoulos vd->vdev_ms_shift = ms_shift; 228386714001SSerapheim Dimitropoulos ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2284fa9e4066Sahrens } 2285fa9e4066Sahrens 2286fa9e4066Sahrens void 2287ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2288fa9e4066Sahrens { 2289ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 22905cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 22915cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 2292ecc2d604Sbonwick ASSERT(ISP2(flags)); 2293f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 2294fa9e4066Sahrens 2295ecc2d604Sbonwick if (flags & VDD_METASLAB) 2296ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2297ecc2d604Sbonwick 2298ecc2d604Sbonwick if (flags & VDD_DTL) 2299ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2300ecc2d604Sbonwick 2301ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2302fa9e4066Sahrens } 2303fa9e4066Sahrens 23040713e232SGeorge Wilson void 23050713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 23060713e232SGeorge Wilson { 23070713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 23080713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 23090713e232SGeorge Wilson 23100713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 23110713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 23120713e232SGeorge Wilson } 23130713e232SGeorge Wilson 23148ad4d6ddSJeff Bonwick /* 23158ad4d6ddSJeff Bonwick * DTLs. 23168ad4d6ddSJeff Bonwick * 23178ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 23189fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 23198ad4d6ddSJeff Bonwick * 23208ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 23218ad4d6ddSJeff Bonwick * 23228ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 23238ad4d6ddSJeff Bonwick * 23248ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 23258ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 23268ad4d6ddSJeff Bonwick * txgs that was scrubbed. 23278ad4d6ddSJeff Bonwick * 23288ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 23298ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 23308ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 23318ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 23328ad4d6ddSJeff Bonwick * determine whether a device can be detached. 23338ad4d6ddSJeff Bonwick * 23348ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 23358ad4d6ddSJeff Bonwick * either has the data or it doesn't. 23368ad4d6ddSJeff Bonwick * 23378ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 23388ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 23398ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 23408ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 23418ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 23428ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 23438ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 23448ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 23458ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 23468ad4d6ddSJeff Bonwick * 23478ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 23488ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 23498ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 23508ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 23518ad4d6ddSJeff Bonwick */ 2352fa9e4066Sahrens void 23538ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2354fa9e4066Sahrens { 23550713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23568ad4d6ddSJeff Bonwick 23578ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23588ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2359f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 23608ad4d6ddSJeff Bonwick 23615cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 23620713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 23630713e232SGeorge Wilson range_tree_add(rt, txg, size); 23645cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2365fa9e4066Sahrens } 2366fa9e4066Sahrens 23678ad4d6ddSJeff Bonwick boolean_t 23688ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2369fa9e4066Sahrens { 23700713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23718ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2372fa9e4066Sahrens 23738ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23748ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2375fa9e4066Sahrens 23765cabbc6bSPrashanth Sreenivasa /* 23775cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 23785cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 23795cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 23805cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 23815cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 23825cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 23835cabbc6bSPrashanth Sreenivasa */ 23845cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 23855cabbc6bSPrashanth Sreenivasa return (B_FALSE); 23865cabbc6bSPrashanth Sreenivasa 23875cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 238886714001SSerapheim Dimitropoulos if (!range_tree_is_empty(rt)) 23890713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 23905cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2391fa9e4066Sahrens 2392fa9e4066Sahrens return (dirty); 2393fa9e4066Sahrens } 2394fa9e4066Sahrens 23958ad4d6ddSJeff Bonwick boolean_t 23968ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 23978ad4d6ddSJeff Bonwick { 23980713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23998ad4d6ddSJeff Bonwick boolean_t empty; 24008ad4d6ddSJeff Bonwick 24015cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 240286714001SSerapheim Dimitropoulos empty = range_tree_is_empty(rt); 24035cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 24048ad4d6ddSJeff Bonwick 24058ad4d6ddSJeff Bonwick return (empty); 24068ad4d6ddSJeff Bonwick } 24078ad4d6ddSJeff Bonwick 2408a3874b8bSToomas Soome /* 2409a3874b8bSToomas Soome * Returns B_TRUE if vdev determines offset needs to be resilvered. 2410a3874b8bSToomas Soome */ 2411a3874b8bSToomas Soome boolean_t 2412a3874b8bSToomas Soome vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2413a3874b8bSToomas Soome { 2414a3874b8bSToomas Soome ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2415a3874b8bSToomas Soome 2416a3874b8bSToomas Soome if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2417a3874b8bSToomas Soome vd->vdev_ops->vdev_op_leaf) 2418a3874b8bSToomas Soome return (B_TRUE); 2419a3874b8bSToomas Soome 2420a3874b8bSToomas Soome return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2421a3874b8bSToomas Soome } 2422a3874b8bSToomas Soome 2423b4952e17SGeorge Wilson /* 2424b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2425b4952e17SGeorge Wilson */ 2426b4952e17SGeorge Wilson static uint64_t 2427b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2428b4952e17SGeorge Wilson { 24290713e232SGeorge Wilson range_seg_t *rs; 2430b4952e17SGeorge Wilson 2431b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 24320713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2433b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2434b4952e17SGeorge Wilson 24350713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 24360713e232SGeorge Wilson return (rs->rs_start - 1); 2437b4952e17SGeorge Wilson } 2438b4952e17SGeorge Wilson 2439b4952e17SGeorge Wilson /* 2440b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2441b4952e17SGeorge Wilson */ 2442b4952e17SGeorge Wilson static uint64_t 2443b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 2444b4952e17SGeorge Wilson { 24450713e232SGeorge Wilson range_seg_t *rs; 2446b4952e17SGeorge Wilson 2447b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 24480713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2449b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2450b4952e17SGeorge Wilson 24510713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 24520713e232SGeorge Wilson return (rs->rs_end); 2453b4952e17SGeorge Wilson } 2454b4952e17SGeorge Wilson 2455b4952e17SGeorge Wilson /* 2456b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2457b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2458b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2459b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2460b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2461b4952e17SGeorge Wilson * excise the DTLs. 2462b4952e17SGeorge Wilson */ 2463b4952e17SGeorge Wilson static boolean_t 2464b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2465b4952e17SGeorge Wilson { 2466b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2467b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2468b4952e17SGeorge Wilson 2469b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 2470b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2471b4952e17SGeorge Wilson 24722d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 24732d2f193aSMatthew Ahrens return (B_FALSE); 24742d2f193aSMatthew Ahrens 2475e4c795beSTom Caputi if (vd->vdev_resilver_deferred) 2476e4c795beSTom Caputi return (B_FALSE); 2477e4c795beSTom Caputi 2478b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 247986714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2480b4952e17SGeorge Wilson return (B_TRUE); 2481b4952e17SGeorge Wilson 2482b4952e17SGeorge Wilson /* 2483b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2484b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2485b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2486b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2487b4952e17SGeorge Wilson * for excision. 2488b4952e17SGeorge Wilson */ 2489b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2490b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2491b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2492b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2493b4952e17SGeorge Wilson return (B_TRUE); 2494b4952e17SGeorge Wilson } 2495b4952e17SGeorge Wilson return (B_FALSE); 2496b4952e17SGeorge Wilson } 2497b4952e17SGeorge Wilson 2498fa9e4066Sahrens /* 2499fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2500fa9e4066Sahrens */ 2501fa9e4066Sahrens void 2502fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2503fa9e4066Sahrens { 2504ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 25058ad4d6ddSJeff Bonwick avl_tree_t reftree; 25068ad4d6ddSJeff Bonwick int minref; 2507fa9e4066Sahrens 25088ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2509fa9e4066Sahrens 25108ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 25118ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 25128ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 25138ad4d6ddSJeff Bonwick 25145cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 25158ad4d6ddSJeff Bonwick return; 25168ad4d6ddSJeff Bonwick 25178ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 25183f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 25193f9d6ad7SLin Ling 2520fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2521b4952e17SGeorge Wilson 2522b4952e17SGeorge Wilson /* 2523b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2524b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2525b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2526b4952e17SGeorge Wilson * the entire duration of this scan. 2527b4952e17SGeorge Wilson */ 2528088f3894Sahrens if (scrub_txg != 0 && 25293f9d6ad7SLin Ling (spa->spa_scrub_started || 2530b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2531b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2532088f3894Sahrens /* 2533088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2534088f3894Sahrens * did it without rebooting, then the scrub dtl 2535088f3894Sahrens * will be valid, so excise the old region and 2536088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2537088f3894Sahrens * dtl as-is if there was an error. 25388ad4d6ddSJeff Bonwick * 25398ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 25408ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 25418ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 25428ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 25438ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 25448ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 25458ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 25468ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 25478ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2548088f3894Sahrens */ 25490713e232SGeorge Wilson space_reftree_create(&reftree); 25500713e232SGeorge Wilson space_reftree_add_map(&reftree, 25510713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25520713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 25530713e232SGeorge Wilson space_reftree_add_map(&reftree, 25540713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 25550713e232SGeorge Wilson space_reftree_generate_map(&reftree, 25560713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25570713e232SGeorge Wilson space_reftree_destroy(&reftree); 2558fa9e4066Sahrens } 25590713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 25600713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 25610713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2562fa9e4066Sahrens if (scrub_done) 25630713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 25640713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 25658ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 25660713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 25678ad4d6ddSJeff Bonwick else 25680713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 25690713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2570b4952e17SGeorge Wilson 2571b4952e17SGeorge Wilson /* 2572b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2573b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2574b4952e17SGeorge Wilson */ 2575b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 257686714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 257786714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2578b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2579b4952e17SGeorge Wilson 2580fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2581088f3894Sahrens 2582ecc2d604Sbonwick if (txg != 0) 2583ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2584fa9e4066Sahrens return; 2585fa9e4066Sahrens } 2586fa9e4066Sahrens 2587fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 25888ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 258999bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 259099bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 25918ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 25928ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 25938ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 25948ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 25958ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 25968ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 25978ad4d6ddSJeff Bonwick else 25988ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 25990713e232SGeorge Wilson space_reftree_create(&reftree); 26008ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 26018ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 26028ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 26030713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 26048ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 26058ad4d6ddSJeff Bonwick } 26060713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 26070713e232SGeorge Wilson space_reftree_destroy(&reftree); 2608fa9e4066Sahrens } 26098ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2610fa9e4066Sahrens } 2611fa9e4066Sahrens 26120713e232SGeorge Wilson int 2613fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2614fa9e4066Sahrens { 2615fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2616ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 26170713e232SGeorge Wilson int error = 0; 2618fa9e4066Sahrens 26190713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 26205cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2621fa9e4066Sahrens 26220713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 26235cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 26240713e232SGeorge Wilson if (error) 26250713e232SGeorge Wilson return (error); 26260713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2627fa9e4066Sahrens 26280713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 26290713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 26300713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 26310713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2632fa9e4066Sahrens 26330713e232SGeorge Wilson return (error); 26340713e232SGeorge Wilson } 26350713e232SGeorge Wilson 26360713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 26370713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 26380713e232SGeorge Wilson if (error != 0) 26390713e232SGeorge Wilson break; 26400713e232SGeorge Wilson } 2641fa9e4066Sahrens 2642fa9e4066Sahrens return (error); 2643fa9e4066Sahrens } 2644fa9e4066Sahrens 2645663207adSDon Brady static void 2646663207adSDon Brady vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2647663207adSDon Brady { 2648663207adSDon Brady spa_t *spa = vd->vdev_spa; 2649663207adSDon Brady objset_t *mos = spa->spa_meta_objset; 2650663207adSDon Brady vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2651663207adSDon Brady const char *string; 2652663207adSDon Brady 2653663207adSDon Brady ASSERT(alloc_bias != VDEV_BIAS_NONE); 2654663207adSDon Brady 2655663207adSDon Brady string = 2656663207adSDon Brady (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2657663207adSDon Brady (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2658663207adSDon Brady (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2659663207adSDon Brady 2660663207adSDon Brady ASSERT(string != NULL); 2661663207adSDon Brady VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2662663207adSDon Brady 1, strlen(string) + 1, string, tx)); 2663663207adSDon Brady 2664663207adSDon Brady if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2665663207adSDon Brady spa_activate_allocation_classes(spa, tx); 2666663207adSDon Brady } 2667663207adSDon Brady } 2668663207adSDon Brady 2669215198a6SJoe Stein void 2670215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2671215198a6SJoe Stein { 2672215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2673215198a6SJoe Stein 2674215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2675215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2676215198a6SJoe Stein zapobj, tx)); 2677215198a6SJoe Stein } 2678215198a6SJoe Stein 2679215198a6SJoe Stein uint64_t 2680215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2681215198a6SJoe Stein { 2682215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2683215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2684215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2685215198a6SJoe Stein 2686215198a6SJoe Stein ASSERT(zap != 0); 2687215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2688215198a6SJoe Stein zap, tx)); 2689215198a6SJoe Stein 2690215198a6SJoe Stein return (zap); 2691215198a6SJoe Stein } 2692215198a6SJoe Stein 2693215198a6SJoe Stein void 2694215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2695215198a6SJoe Stein { 2696215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2697215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2698215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2699215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2700215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2701215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2702215198a6SJoe Stein } 2703215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2704215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2705663207adSDon Brady if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2706663207adSDon Brady vdev_zap_allocation_data(vd, tx); 2707215198a6SJoe Stein } 2708215198a6SJoe Stein } 2709663207adSDon Brady 2710215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2711215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2712215198a6SJoe Stein } 2713215198a6SJoe Stein } 2714215198a6SJoe Stein 2715fa9e4066Sahrens void 2716fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2717fa9e4066Sahrens { 2718fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 27190713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2720ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 27210713e232SGeorge Wilson range_tree_t *rtsync; 2722fa9e4066Sahrens dmu_tx_t *tx; 27230713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2724fa9e4066Sahrens 27255cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 27260713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 272788ecc943SGeorge Wilson 2728fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2729fa9e4066Sahrens 27300713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 27310713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 27320713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 27330713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 27340713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 27350713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2736215198a6SJoe Stein 2737215198a6SJoe Stein /* 2738215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2739215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2740215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2741215198a6SJoe Stein */ 2742215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2743215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2744215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2745215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2746215198a6SJoe Stein } 2747215198a6SJoe Stein 2748fa9e4066Sahrens dmu_tx_commit(tx); 2749fa9e4066Sahrens return; 2750fa9e4066Sahrens } 2751fa9e4066Sahrens 27520713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 27530713e232SGeorge Wilson uint64_t new_object; 27540713e232SGeorge Wilson 275586714001SSerapheim Dimitropoulos new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); 27560713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 27570713e232SGeorge Wilson 27580713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 27595cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 27600713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2761fa9e4066Sahrens } 2762fa9e4066Sahrens 27635cabbc6bSPrashanth Sreenivasa rtsync = range_tree_create(NULL, NULL); 2764fa9e4066Sahrens 2765fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 27660713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2767fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2768fa9e4066Sahrens 276986714001SSerapheim Dimitropoulos space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); 277017f11284SSerapheim Dimitropoulos space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 27710713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2772fa9e4066Sahrens 27730713e232SGeorge Wilson range_tree_destroy(rtsync); 2774fa9e4066Sahrens 27750713e232SGeorge Wilson /* 27760713e232SGeorge Wilson * If the object for the space map has changed then dirty 27770713e232SGeorge Wilson * the top level so that we update the config. 27780713e232SGeorge Wilson */ 27790713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 27803ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 27813ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 27823ee8c80cSPavel Zakharov (u_longlong_t)object, 27833ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 27840713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 27850713e232SGeorge Wilson } 2786fa9e4066Sahrens 2787fa9e4066Sahrens dmu_tx_commit(tx); 2788fa9e4066Sahrens } 2789fa9e4066Sahrens 27908ad4d6ddSJeff Bonwick /* 27918ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 27928ad4d6ddSJeff Bonwick * without losing data. 27938ad4d6ddSJeff Bonwick */ 27948ad4d6ddSJeff Bonwick boolean_t 27958ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 27968ad4d6ddSJeff Bonwick { 27978ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 27988ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 27998ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 28008ad4d6ddSJeff Bonwick boolean_t required; 28018ad4d6ddSJeff Bonwick 28028ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 28038ad4d6ddSJeff Bonwick 28048ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 28058ad4d6ddSJeff Bonwick return (B_TRUE); 28068ad4d6ddSJeff Bonwick 28078ad4d6ddSJeff Bonwick /* 28088ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 28098ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 28108ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 28118ad4d6ddSJeff Bonwick */ 28128ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 28138ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28148ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 28158ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 28168ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28178ad4d6ddSJeff Bonwick 2818cb04b873SMark J Musante if (!required && zio_injection_enabled) 2819cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2820cb04b873SMark J Musante 28218ad4d6ddSJeff Bonwick return (required); 28228ad4d6ddSJeff Bonwick } 28238ad4d6ddSJeff Bonwick 2824088f3894Sahrens /* 2825088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2826088f3894Sahrens */ 2827088f3894Sahrens boolean_t 2828088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2829088f3894Sahrens { 2830088f3894Sahrens boolean_t needed = B_FALSE; 2831088f3894Sahrens uint64_t thismin = UINT64_MAX; 2832088f3894Sahrens uint64_t thismax = 0; 2833088f3894Sahrens 2834088f3894Sahrens if (vd->vdev_children == 0) { 2835088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 283686714001SSerapheim Dimitropoulos if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 28378ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2838088f3894Sahrens 2839b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2840b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2841088f3894Sahrens needed = B_TRUE; 2842088f3894Sahrens } 2843088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2844088f3894Sahrens } else { 28458ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2846088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2847088f3894Sahrens uint64_t cmin, cmax; 2848088f3894Sahrens 2849088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2850088f3894Sahrens thismin = MIN(thismin, cmin); 2851088f3894Sahrens thismax = MAX(thismax, cmax); 2852088f3894Sahrens needed = B_TRUE; 2853088f3894Sahrens } 2854088f3894Sahrens } 2855088f3894Sahrens } 2856088f3894Sahrens 2857088f3894Sahrens if (needed && minp) { 2858088f3894Sahrens *minp = thismin; 2859088f3894Sahrens *maxp = thismax; 2860088f3894Sahrens } 2861088f3894Sahrens return (needed); 2862088f3894Sahrens } 2863088f3894Sahrens 286486714001SSerapheim Dimitropoulos /* 286586714001SSerapheim Dimitropoulos * Gets the checkpoint space map object from the vdev's ZAP. 286686714001SSerapheim Dimitropoulos * Returns the spacemap object, or 0 if it wasn't in the ZAP 286786714001SSerapheim Dimitropoulos * or the ZAP doesn't exist yet. 286886714001SSerapheim Dimitropoulos */ 286986714001SSerapheim Dimitropoulos int 287086714001SSerapheim Dimitropoulos vdev_checkpoint_sm_object(vdev_t *vd) 287186714001SSerapheim Dimitropoulos { 287286714001SSerapheim Dimitropoulos ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 287386714001SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) { 287486714001SSerapheim Dimitropoulos return (0); 287586714001SSerapheim Dimitropoulos } 287686714001SSerapheim Dimitropoulos 287786714001SSerapheim Dimitropoulos uint64_t sm_obj = 0; 287886714001SSerapheim Dimitropoulos int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 287986714001SSerapheim Dimitropoulos VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 288086714001SSerapheim Dimitropoulos 288186714001SSerapheim Dimitropoulos ASSERT(err == 0 || err == ENOENT); 288286714001SSerapheim Dimitropoulos 288386714001SSerapheim Dimitropoulos return (sm_obj); 288486714001SSerapheim Dimitropoulos } 288586714001SSerapheim Dimitropoulos 28865cabbc6bSPrashanth Sreenivasa int 2887ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2888fa9e4066Sahrens { 28895cabbc6bSPrashanth Sreenivasa int error = 0; 2890fa9e4066Sahrens /* 2891fa9e4066Sahrens * Recursively load all children. 2892fa9e4066Sahrens */ 28935cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 28945cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 28955cabbc6bSPrashanth Sreenivasa if (error != 0) { 28965cabbc6bSPrashanth Sreenivasa return (error); 28975cabbc6bSPrashanth Sreenivasa } 28985cabbc6bSPrashanth Sreenivasa } 28995cabbc6bSPrashanth Sreenivasa 29005cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2901fa9e4066Sahrens 2902663207adSDon Brady /* 2903663207adSDon Brady * On spa_load path, grab the allocation bias from our zap 2904663207adSDon Brady */ 2905663207adSDon Brady if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2906663207adSDon Brady spa_t *spa = vd->vdev_spa; 2907663207adSDon Brady char bias_str[64]; 2908663207adSDon Brady 2909663207adSDon Brady if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2910663207adSDon Brady VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2911663207adSDon Brady bias_str) == 0) { 2912663207adSDon Brady ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2913663207adSDon Brady vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2914663207adSDon Brady } 2915663207adSDon Brady } 2916663207adSDon Brady 2917fa9e4066Sahrens /* 29180e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2919fa9e4066Sahrens */ 29205cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2921663207adSDon Brady vdev_metaslab_group_create(vd); 2922663207adSDon Brady 29235cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 29245cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29255cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29263ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 29273ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 29283ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 29295cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 2930555d674dSSerapheim Dimitropoulos } 2931555d674dSSerapheim Dimitropoulos 2932555d674dSSerapheim Dimitropoulos error = vdev_metaslab_init(vd, 0); 2933555d674dSSerapheim Dimitropoulos if (error != 0) { 29343ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 29353ee8c80cSPavel Zakharov "[error=%d]", error); 29365cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29375cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29385cabbc6bSPrashanth Sreenivasa return (error); 29395cabbc6bSPrashanth Sreenivasa } 294086714001SSerapheim Dimitropoulos 294186714001SSerapheim Dimitropoulos uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 294286714001SSerapheim Dimitropoulos if (checkpoint_sm_obj != 0) { 294386714001SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 294486714001SSerapheim Dimitropoulos ASSERT(vd->vdev_asize != 0); 294586714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 294686714001SSerapheim Dimitropoulos 2947555d674dSSerapheim Dimitropoulos error = space_map_open(&vd->vdev_checkpoint_sm, 294886714001SSerapheim Dimitropoulos mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2949555d674dSSerapheim Dimitropoulos vd->vdev_ashift); 2950555d674dSSerapheim Dimitropoulos if (error != 0) { 295186714001SSerapheim Dimitropoulos vdev_dbgmsg(vd, "vdev_load: space_map_open " 295286714001SSerapheim Dimitropoulos "failed for checkpoint spacemap (obj %llu) " 295386714001SSerapheim Dimitropoulos "[error=%d]", 295486714001SSerapheim Dimitropoulos (u_longlong_t)checkpoint_sm_obj, error); 295586714001SSerapheim Dimitropoulos return (error); 295686714001SSerapheim Dimitropoulos } 295786714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 295886714001SSerapheim Dimitropoulos 295986714001SSerapheim Dimitropoulos /* 296086714001SSerapheim Dimitropoulos * Since the checkpoint_sm contains free entries 2961555d674dSSerapheim Dimitropoulos * exclusively we can use space_map_allocated() to 2962555d674dSSerapheim Dimitropoulos * indicate the cumulative checkpointed space that 2963555d674dSSerapheim Dimitropoulos * has been freed. 296486714001SSerapheim Dimitropoulos */ 296586714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space = 2966555d674dSSerapheim Dimitropoulos -space_map_allocated(vd->vdev_checkpoint_sm); 296786714001SSerapheim Dimitropoulos vd->vdev_spa->spa_checkpoint_info.sci_dspace += 296886714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space; 296986714001SSerapheim Dimitropoulos } 29705cabbc6bSPrashanth Sreenivasa } 2971fa9e4066Sahrens 2972fa9e4066Sahrens /* 2973fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2974fa9e4066Sahrens */ 29755cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2976560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2977560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 29783ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 29793ee8c80cSPavel Zakharov "[error=%d]", error); 29805cabbc6bSPrashanth Sreenivasa return (error); 29815cabbc6bSPrashanth Sreenivasa } 29825cabbc6bSPrashanth Sreenivasa 29835cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 29845cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 29855cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 29865cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 298786714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 29885cabbc6bSPrashanth Sreenivasa 29895cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 29905cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 29915cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29925cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29933ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 29943ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 29953ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 29965cabbc6bSPrashanth Sreenivasa return (error); 29975cabbc6bSPrashanth Sreenivasa } 29985cabbc6bSPrashanth Sreenivasa } 29995cabbc6bSPrashanth Sreenivasa 30005cabbc6bSPrashanth Sreenivasa return (0); 3001fa9e4066Sahrens } 3002fa9e4066Sahrens 300399653d4eSeschrock /* 3004fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 3005fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 3006fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 3007fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 3008fa94a07fSbrendan * repurposed to another pool. 300999653d4eSeschrock */ 301099653d4eSeschrock int 3011fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 301299653d4eSeschrock { 301399653d4eSeschrock nvlist_t *label; 301499653d4eSeschrock uint64_t guid, version; 301599653d4eSeschrock uint64_t state; 301699653d4eSeschrock 3017e14bb325SJeff Bonwick if (!vdev_readable(vd)) 3018c5904d13Seschrock return (0); 3019c5904d13Seschrock 3020dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 302199653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 302299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 302399653d4eSeschrock return (-1); 302499653d4eSeschrock } 302599653d4eSeschrock 302699653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3027ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 302899653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 302999653d4eSeschrock guid != vd->vdev_guid || 303099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 303199653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 303299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 303399653d4eSeschrock nvlist_free(label); 303499653d4eSeschrock return (-1); 303599653d4eSeschrock } 303699653d4eSeschrock 303799653d4eSeschrock /* 303899653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 303999653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 304099653d4eSeschrock */ 304199653d4eSeschrock nvlist_free(label); 304299653d4eSeschrock return (0); 304399653d4eSeschrock } 304499653d4eSeschrock 30455cabbc6bSPrashanth Sreenivasa /* 30465cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 30475cabbc6bSPrashanth Sreenivasa * that points to them. 30485cabbc6bSPrashanth Sreenivasa */ 304988ecc943SGeorge Wilson void 30505cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 30515cabbc6bSPrashanth Sreenivasa { 30525cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 30535cabbc6bSPrashanth Sreenivasa return; 30545cabbc6bSPrashanth Sreenivasa 30555cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 30565cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 30575cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 30585cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 30595cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 30605cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 30615cabbc6bSPrashanth Sreenivasa 30625cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 30635cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 30645cabbc6bSPrashanth Sreenivasa if (smobj == 0) 30655cabbc6bSPrashanth Sreenivasa continue; 30665cabbc6bSPrashanth Sreenivasa 30675cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 30685cabbc6bSPrashanth Sreenivasa } 30695cabbc6bSPrashanth Sreenivasa 30705cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 30715cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 30725cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 30735cabbc6bSPrashanth Sreenivasa } 30745cabbc6bSPrashanth Sreenivasa 30755cabbc6bSPrashanth Sreenivasa static void 30764e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 307788ecc943SGeorge Wilson { 307888ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 307988ecc943SGeorge Wilson 30804e75ba68SSerapheim Dimitropoulos ASSERT(vd->vdev_islog); 3081215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 3082215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 308388ecc943SGeorge Wilson 30844e75ba68SSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3085215198a6SJoe Stein 30864e75ba68SSerapheim Dimitropoulos vdev_destroy_spacemaps(vd, tx); 30874e75ba68SSerapheim Dimitropoulos if (vd->vdev_top_zap != 0) { 3088215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3089215198a6SJoe Stein vd->vdev_top_zap = 0; 3090215198a6SJoe Stein } 30914e75ba68SSerapheim Dimitropoulos 309288ecc943SGeorge Wilson dmu_tx_commit(tx); 309388ecc943SGeorge Wilson } 309488ecc943SGeorge Wilson 3095fa9e4066Sahrens void 3096fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 3097fa9e4066Sahrens { 3098fa9e4066Sahrens metaslab_t *msp; 309980eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3100fa9e4066Sahrens 31015cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 310288ecc943SGeorge Wilson 3103094e47e9SGeorge Wilson while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3104094e47e9SGeorge Wilson != NULL) 3105fa9e4066Sahrens metaslab_sync_done(msp, txg); 310680eb36f2SGeorge Wilson 310780eb36f2SGeorge Wilson if (reassess) 310880eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 3109fa9e4066Sahrens } 3110fa9e4066Sahrens 3111fa9e4066Sahrens void 3112fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 3113fa9e4066Sahrens { 3114fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3115fa9e4066Sahrens vdev_t *lvd; 3116fa9e4066Sahrens metaslab_t *msp; 3117fa9e4066Sahrens 3118555d674dSSerapheim Dimitropoulos ASSERT3U(txg, ==, spa->spa_syncing_txg); 3119555d674dSSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 31205cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 31215cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 31225cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 312388ecc943SGeorge Wilson 31245cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 31255cabbc6bSPrashanth Sreenivasa 31265cabbc6bSPrashanth Sreenivasa /* 31275cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 31285cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 31295cabbc6bSPrashanth Sreenivasa */ 31305cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 31315cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 31325cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3133555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 31345cabbc6bSPrashanth Sreenivasa return; 31355cabbc6bSPrashanth Sreenivasa } 31365cabbc6bSPrashanth Sreenivasa } 31375cabbc6bSPrashanth Sreenivasa 31385cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 31395cabbc6bSPrashanth Sreenivasa 31405cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 31415cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 3142ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 31435cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3144ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3145ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3146ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 3147ecc2d604Sbonwick vdev_config_dirty(vd); 3148ecc2d604Sbonwick } 3149fa9e4066Sahrens 3150ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3151fa9e4066Sahrens metaslab_sync(msp, txg); 3152ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3153ecc2d604Sbonwick } 3154fa9e4066Sahrens 3155fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3156fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 3157fa9e4066Sahrens 31585cabbc6bSPrashanth Sreenivasa /* 31594e75ba68SSerapheim Dimitropoulos * If this is an empty log device being removed, destroy the 31604e75ba68SSerapheim Dimitropoulos * metadata associated with it. 31615cabbc6bSPrashanth Sreenivasa */ 31624e75ba68SSerapheim Dimitropoulos if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 31634e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vd, txg); 31645cabbc6bSPrashanth Sreenivasa 3165fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3166555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 3167fa9e4066Sahrens } 3168fa9e4066Sahrens 3169fa9e4066Sahrens uint64_t 3170fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3171fa9e4066Sahrens { 3172fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3173fa9e4066Sahrens } 3174fa9e4066Sahrens 31753d7072f8Seschrock /* 31763d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 31773d7072f8Seschrock * not be opened, and no I/O is attempted. 31783d7072f8Seschrock */ 3179fa9e4066Sahrens int 3180069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3181fa9e4066Sahrens { 31824b964adaSGeorge Wilson vdev_t *vd, *tvd; 3183fa9e4066Sahrens 31848f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 3185fa9e4066Sahrens 3186c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3187e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3188e14bb325SJeff Bonwick 31893d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3190e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3191fa9e4066Sahrens 31924b964adaSGeorge Wilson tvd = vd->vdev_top; 31934b964adaSGeorge Wilson 3194069f55e2SEric Schrock /* 3195069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 3196069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 3197069f55e2SEric Schrock * were faulted. 3198069f55e2SEric Schrock */ 3199069f55e2SEric Schrock vd->vdev_label_aux = aux; 3200069f55e2SEric Schrock 32013d7072f8Seschrock /* 32023d7072f8Seschrock * Faulted state takes precedence over degraded. 32033d7072f8Seschrock */ 320498d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 32053d7072f8Seschrock vd->vdev_faulted = 1ULL; 32063d7072f8Seschrock vd->vdev_degraded = 0ULL; 3207069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 32083d7072f8Seschrock 32093d7072f8Seschrock /* 3210c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 3211c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 32123d7072f8Seschrock */ 32134b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 32143d7072f8Seschrock vd->vdev_degraded = 1ULL; 32153d7072f8Seschrock vd->vdev_faulted = 0ULL; 32163d7072f8Seschrock 32173d7072f8Seschrock /* 32183d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 32193d7072f8Seschrock * mark it degraded. 32203d7072f8Seschrock */ 32214b964adaSGeorge Wilson vdev_reopen(tvd); 32223d7072f8Seschrock 3223069f55e2SEric Schrock if (vdev_readable(vd)) 3224069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 32253d7072f8Seschrock } 32263d7072f8Seschrock 3227e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 32283d7072f8Seschrock } 32293d7072f8Seschrock 32303d7072f8Seschrock /* 32313d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 32323d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 32333d7072f8Seschrock * as I/O is concerned. 32343d7072f8Seschrock */ 32353d7072f8Seschrock int 3236069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 32373d7072f8Seschrock { 3238c5904d13Seschrock vdev_t *vd; 32390a4e9518Sgw 32408f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 32413d7072f8Seschrock 3242c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3243e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3244e14bb325SJeff Bonwick 32450e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3246e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 32470e34b6a7Sbonwick 32483d7072f8Seschrock /* 32493d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 32503d7072f8Seschrock */ 3251e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 3252e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 32533d7072f8Seschrock 32543d7072f8Seschrock vd->vdev_degraded = 1ULL; 32553d7072f8Seschrock if (!vdev_is_dead(vd)) 32563d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3257069f55e2SEric Schrock aux); 32583d7072f8Seschrock 3259e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 32603d7072f8Seschrock } 32613d7072f8Seschrock 32623d7072f8Seschrock /* 3263f7170741SWill Andrews * Online the given vdev. 3264f7170741SWill Andrews * 3265f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3266f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 3267f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 3268f7170741SWill Andrews * events are generated if the device fails to open. 32693d7072f8Seschrock */ 32703d7072f8Seschrock int 3271e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 32723d7072f8Seschrock { 3273573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 32745f368aefSYuri Pankov boolean_t wasoffline; 32755f368aefSYuri Pankov vdev_state_t oldstate; 32763d7072f8Seschrock 32778f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 32783d7072f8Seschrock 3279c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3280e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 32813d7072f8Seschrock 32823d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3283e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3284fa9e4066Sahrens 32855f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 32865f368aefSYuri Pankov oldstate = vd->vdev_state; 328714372834SHans Rosenfeld 3288573ca77eSGeorge Wilson tvd = vd->vdev_top; 3289fa9e4066Sahrens vd->vdev_offline = B_FALSE; 3290441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 3291e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3292e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3293573ca77eSGeorge Wilson 3294573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3295573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3296573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3297573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3298573ca77eSGeorge Wilson } 3299573ca77eSGeorge Wilson 3300573ca77eSGeorge Wilson vdev_reopen(tvd); 33013d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 33023d7072f8Seschrock 3303573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3304573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3305573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 3306573ca77eSGeorge Wilson } 3307573ca77eSGeorge Wilson 33083d7072f8Seschrock if (newstate) 33093d7072f8Seschrock *newstate = vd->vdev_state; 33103d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 33113d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 33123d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 33133d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 33143d7072f8Seschrock vd->vdev_unspare = B_TRUE; 3315fa9e4066Sahrens 3316573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3317573ca77eSGeorge Wilson 3318573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3319573ca77eSGeorge Wilson if (vd->vdev_aux) 3320573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3321573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3322573ca77eSGeorge Wilson } 332314372834SHans Rosenfeld 3324094e47e9SGeorge Wilson /* Restart initializing if necessary */ 3325094e47e9SGeorge Wilson mutex_enter(&vd->vdev_initialize_lock); 3326094e47e9SGeorge Wilson if (vdev_writeable(vd) && 3327094e47e9SGeorge Wilson vd->vdev_initialize_thread == NULL && 3328094e47e9SGeorge Wilson vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3329094e47e9SGeorge Wilson (void) vdev_initialize(vd); 3330094e47e9SGeorge Wilson } 3331094e47e9SGeorge Wilson mutex_exit(&vd->vdev_initialize_lock); 3332094e47e9SGeorge Wilson 3333*084fd14fSBrian Behlendorf /* Restart trimming if necessary */ 3334*084fd14fSBrian Behlendorf mutex_enter(&vd->vdev_trim_lock); 3335*084fd14fSBrian Behlendorf if (vdev_writeable(vd) && 3336*084fd14fSBrian Behlendorf vd->vdev_trim_thread == NULL && 3337*084fd14fSBrian Behlendorf vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 3338*084fd14fSBrian Behlendorf (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 3339*084fd14fSBrian Behlendorf vd->vdev_trim_secure); 3340*084fd14fSBrian Behlendorf } 3341*084fd14fSBrian Behlendorf mutex_exit(&vd->vdev_trim_lock); 3342*084fd14fSBrian Behlendorf 33435f368aefSYuri Pankov if (wasoffline || 33445f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 33455f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 3346ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 334714372834SHans Rosenfeld 33488ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 3349fa9e4066Sahrens } 3350fa9e4066Sahrens 3351a1521560SJeff Bonwick static int 3352a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3353fa9e4066Sahrens { 3354e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 33558f18d1faSGeorge Wilson int error = 0; 33568f18d1faSGeorge Wilson uint64_t generation; 33578f18d1faSGeorge Wilson metaslab_group_t *mg; 33580a4e9518Sgw 33598f18d1faSGeorge Wilson top: 33608f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 3361fa9e4066Sahrens 3362c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3363e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3364fa9e4066Sahrens 33650e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3366e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 33670e34b6a7Sbonwick 3368e6ca193dSGeorge Wilson tvd = vd->vdev_top; 33698f18d1faSGeorge Wilson mg = tvd->vdev_mg; 33708f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 3371e6ca193dSGeorge Wilson 3372fa9e4066Sahrens /* 3373ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 3374fa9e4066Sahrens */ 3375ecc2d604Sbonwick if (!vd->vdev_offline) { 3376ecc2d604Sbonwick /* 33778ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 3378e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 3379e6ca193dSGeorge Wilson * expendable. 3380ecc2d604Sbonwick */ 3381e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3382e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 3383e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3384fa9e4066Sahrens 33858f18d1faSGeorge Wilson /* 3386b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 3387b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 3388b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 3389b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 33908f18d1faSGeorge Wilson */ 33918f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 33928f18d1faSGeorge Wilson /* 33938f18d1faSGeorge Wilson * Prevent any future allocations. 33948f18d1faSGeorge Wilson */ 3395a1521560SJeff Bonwick metaslab_group_passivate(mg); 33968f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 33978f18d1faSGeorge Wilson 33985cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 33998f18d1faSGeorge Wilson 340086714001SSerapheim Dimitropoulos /* 340186714001SSerapheim Dimitropoulos * If the log device was successfully reset but has 340286714001SSerapheim Dimitropoulos * checkpointed data, do not offline it. 340386714001SSerapheim Dimitropoulos */ 340486714001SSerapheim Dimitropoulos if (error == 0 && 340586714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm != NULL) { 340686714001SSerapheim Dimitropoulos error = ZFS_ERR_CHECKPOINT_EXISTS; 340786714001SSerapheim Dimitropoulos } 340886714001SSerapheim Dimitropoulos 34098f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 34108f18d1faSGeorge Wilson 34118f18d1faSGeorge Wilson /* 34128f18d1faSGeorge Wilson * Check to see if the config has changed. 34138f18d1faSGeorge Wilson */ 34148f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 3415a1521560SJeff Bonwick metaslab_group_activate(mg); 34168f18d1faSGeorge Wilson if (error) 34178f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 34188f18d1faSGeorge Wilson vd, error)); 34198f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 34208f18d1faSGeorge Wilson goto top; 34218f18d1faSGeorge Wilson } 3422fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 34238f18d1faSGeorge Wilson } 34248f18d1faSGeorge Wilson 3425ecc2d604Sbonwick /* 3426ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3427e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3428e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3429e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3430ecc2d604Sbonwick */ 3431ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3432e6ca193dSGeorge Wilson vdev_reopen(tvd); 3433e6ca193dSGeorge Wilson 3434e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3435e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3436ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3437e6ca193dSGeorge Wilson vdev_reopen(tvd); 3438e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3439ecc2d604Sbonwick } 34408f18d1faSGeorge Wilson 34418f18d1faSGeorge Wilson /* 34428f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 34438f18d1faSGeorge Wilson * once we online the device it's open for business. 34448f18d1faSGeorge Wilson */ 34458f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3446a1521560SJeff Bonwick metaslab_group_activate(mg); 3447fa9e4066Sahrens } 3448fa9e4066Sahrens 3449e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3450ecc2d604Sbonwick 34518f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3452fa9e4066Sahrens } 3453fa9e4066Sahrens 3454a1521560SJeff Bonwick int 3455a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3456a1521560SJeff Bonwick { 3457a1521560SJeff Bonwick int error; 3458a1521560SJeff Bonwick 3459a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3460a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3461a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3462a1521560SJeff Bonwick 3463a1521560SJeff Bonwick return (error); 3464a1521560SJeff Bonwick } 3465a1521560SJeff Bonwick 3466ea8dc4b6Seschrock /* 3467ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3468ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3469ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3470ea8dc4b6Seschrock */ 3471ea8dc4b6Seschrock void 3472e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3473fa9e4066Sahrens { 3474e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3475e14bb325SJeff Bonwick 3476e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3477fa9e4066Sahrens 3478ea8dc4b6Seschrock if (vd == NULL) 3479e14bb325SJeff Bonwick vd = rvd; 3480fa9e4066Sahrens 3481ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3482ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3483ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3484fa9e4066Sahrens 3485e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3486e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 34873d7072f8Seschrock 34885cabbc6bSPrashanth Sreenivasa /* 34895cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 34905cabbc6bSPrashanth Sreenivasa */ 34915cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 34925cabbc6bSPrashanth Sreenivasa return; 34935cabbc6bSPrashanth Sreenivasa 34943d7072f8Seschrock /* 34958a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 34968a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 34978a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 34988a79c1b5Sek * written out to disk. 34993d7072f8Seschrock */ 3500e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3501e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 35028a79c1b5Sek 3503096d22d4SEric Schrock /* 3504096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3505096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3506096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3507096d22d4SEric Schrock */ 3508096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3509096d22d4SEric Schrock 35104b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3511e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3512e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3513e14bb325SJeff Bonwick 3514f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 35153d7072f8Seschrock 3516096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3517096d22d4SEric Schrock 3518f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3519e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3520e14bb325SJeff Bonwick 3521e4c795beSTom Caputi if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) { 3522e4c795beSTom Caputi if (dsl_scan_resilvering(spa->spa_dsl_pool) && 3523e4c795beSTom Caputi spa_feature_is_enabled(spa, 3524e4c795beSTom Caputi SPA_FEATURE_RESILVER_DEFER)) 3525e4c795beSTom Caputi vdev_set_deferred_resilver(spa, vd); 3526e4c795beSTom Caputi else 3527e4c795beSTom Caputi spa_async_request(spa, SPA_ASYNC_RESILVER); 3528e4c795beSTom Caputi } 35293d7072f8Seschrock 3530ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 35313d7072f8Seschrock } 3532096d22d4SEric Schrock 3533096d22d4SEric Schrock /* 3534096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3535096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3536096d22d4SEric Schrock * done in response to the FMA fault. 3537096d22d4SEric Schrock */ 3538096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3539096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3540096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3541096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3542fa9e4066Sahrens } 3543fa9e4066Sahrens 3544e14bb325SJeff Bonwick boolean_t 3545e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 35460a4e9518Sgw { 354788ecc943SGeorge Wilson /* 354888ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 354988ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 355088ecc943SGeorge Wilson * these types of devices in the various code paths. 355188ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 355288ecc943SGeorge Wilson * before issuing I/O to them. 355388ecc943SGeorge Wilson */ 35545cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 35555cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 355688ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 35570a4e9518Sgw } 35580a4e9518Sgw 3559e14bb325SJeff Bonwick boolean_t 3560e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 35610a4e9518Sgw { 3562e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 35630a4e9518Sgw } 35640a4e9518Sgw 3565e14bb325SJeff Bonwick boolean_t 3566e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3567fa9e4066Sahrens { 35685cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 35695cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3570fa9e4066Sahrens } 3571fa9e4066Sahrens 3572a31e6787SGeorge Wilson boolean_t 3573a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3574a31e6787SGeorge Wilson { 35758ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 35768ad4d6ddSJeff Bonwick 3577a31e6787SGeorge Wilson /* 35788ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3579a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3580a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 35818ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 35828ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 35838ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3584a31e6787SGeorge Wilson */ 35858ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 35865cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 35870f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3588a31e6787SGeorge Wilson } 3589a31e6787SGeorge Wilson 3590e14bb325SJeff Bonwick boolean_t 3591e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3592fa9e4066Sahrens { 3593e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3594fa9e4066Sahrens 3595e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3596e14bb325SJeff Bonwick return (B_FALSE); 3597fa9e4066Sahrens 3598e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3599e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3600fa9e4066Sahrens 3601e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3602e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3603fa9e4066Sahrens 3604e14bb325SJeff Bonwick return (B_TRUE); 3605fa9e4066Sahrens } 3606fa9e4066Sahrens 360786714001SSerapheim Dimitropoulos boolean_t 360886714001SSerapheim Dimitropoulos vdev_is_spacemap_addressable(vdev_t *vd) 360986714001SSerapheim Dimitropoulos { 3610a0b03b16SSerapheim Dimitropoulos if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3611a0b03b16SSerapheim Dimitropoulos return (B_TRUE); 3612a0b03b16SSerapheim Dimitropoulos 361386714001SSerapheim Dimitropoulos /* 3614a0b03b16SSerapheim Dimitropoulos * If double-word space map entries are not enabled we assume 3615a0b03b16SSerapheim Dimitropoulos * 47 bits of the space map entry are dedicated to the entry's 3616a0b03b16SSerapheim Dimitropoulos * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3617a0b03b16SSerapheim Dimitropoulos * to calculate the maximum address that can be described by a 3618a0b03b16SSerapheim Dimitropoulos * space map entry for the given device. 361986714001SSerapheim Dimitropoulos */ 3620a0b03b16SSerapheim Dimitropoulos uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 362186714001SSerapheim Dimitropoulos 362286714001SSerapheim Dimitropoulos if (shift >= 63) /* detect potential overflow */ 362386714001SSerapheim Dimitropoulos return (B_TRUE); 362486714001SSerapheim Dimitropoulos 362586714001SSerapheim Dimitropoulos return (vd->vdev_asize < (1ULL << shift)); 362686714001SSerapheim Dimitropoulos } 362786714001SSerapheim Dimitropoulos 3628fa9e4066Sahrens /* 3629fa9e4066Sahrens * Get statistics for the given vdev. 3630fa9e4066Sahrens */ 3631fa9e4066Sahrens void 3632fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3633fa9e4066Sahrens { 36342e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 36352e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 3636c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 36372e4c9986SGeorge Wilson 36382e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3639fa9e4066Sahrens 3640fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 3641fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3642fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3643fa9e4066Sahrens vs->vs_state = vd->vdev_state; 3644573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 3645094e47e9SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) { 3646573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3647094e47e9SGeorge Wilson /* 3648094e47e9SGeorge Wilson * Report intializing progress. Since we don't have the 3649094e47e9SGeorge Wilson * initializing locks held, this is only an estimate (although a 3650094e47e9SGeorge Wilson * fairly accurate one). 3651094e47e9SGeorge Wilson */ 3652094e47e9SGeorge Wilson vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done; 3653094e47e9SGeorge Wilson vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est; 3654094e47e9SGeorge Wilson vs->vs_initialize_state = vd->vdev_initialize_state; 3655094e47e9SGeorge Wilson vs->vs_initialize_action_time = vd->vdev_initialize_action_time; 3656094e47e9SGeorge Wilson } 3657*084fd14fSBrian Behlendorf 3658*084fd14fSBrian Behlendorf /* 3659*084fd14fSBrian Behlendorf * Report manual TRIM progress. Since we don't have 3660*084fd14fSBrian Behlendorf * the manual TRIM locks held, this is only an 3661*084fd14fSBrian Behlendorf * estimate (although fairly accurate one). 3662*084fd14fSBrian Behlendorf */ 3663*084fd14fSBrian Behlendorf vs->vs_trim_notsup = !vd->vdev_has_trim; 3664*084fd14fSBrian Behlendorf vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 3665*084fd14fSBrian Behlendorf vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 3666*084fd14fSBrian Behlendorf vs->vs_trim_state = vd->vdev_trim_state; 3667*084fd14fSBrian Behlendorf vs->vs_trim_action_time = vd->vdev_trim_action_time; 3668*084fd14fSBrian Behlendorf 3669c39a2aaeSGeorge Wilson /* 3670c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 3671c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 3672c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 3673c39a2aaeSGeorge Wilson */ 3674c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 36757855d95bSToomas Soome vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 36767855d95bSToomas Soome spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3677c39a2aaeSGeorge Wilson } 36785cabbc6bSPrashanth Sreenivasa if (vd->vdev_aux == NULL && vd == vd->vdev_top && 36795cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3680663207adSDon Brady vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3681663207adSDon Brady vd->vdev_mg->mg_fragmentation : 0; 36822986efa8SAlex Reece } 3683e4c795beSTom Caputi if (vd->vdev_ops->vdev_op_leaf) 3684e4c795beSTom Caputi vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 3685fa9e4066Sahrens 3686fa9e4066Sahrens /* 3687fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 3688fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 3689fa9e4066Sahrens */ 3690fa9e4066Sahrens if (vd == rvd) { 3691e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 3692fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 3693fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 3694fa9e4066Sahrens 3695*084fd14fSBrian Behlendorf for (int t = 0; t < VS_ZIO_TYPES; t++) { 3696fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 3697fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 3698fa9e4066Sahrens } 36993f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 3700fa9e4066Sahrens } 3701fa9e4066Sahrens } 37022e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3703fa9e4066Sahrens } 3704fa9e4066Sahrens 3705fa94a07fSbrendan void 3706fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3707fa94a07fSbrendan { 3708fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3709fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3710fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3711fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3712fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3713fa94a07fSbrendan } 3714fa94a07fSbrendan 37153f9d6ad7SLin Ling void 37163f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 37173f9d6ad7SLin Ling { 37183f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 37193f9d6ad7SLin Ling 37203f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 37213f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 37223f9d6ad7SLin Ling 37233f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 37243f9d6ad7SLin Ling vs->vs_scan_processed = 0; 37253f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 37263f9d6ad7SLin Ling } 37273f9d6ad7SLin Ling 3728fa9e4066Sahrens void 3729e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3730fa9e4066Sahrens { 37318ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 37328ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3733e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3734fa9e4066Sahrens vdev_t *pvd; 3735fa9e4066Sahrens uint64_t txg = zio->io_txg; 3736fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3737fa9e4066Sahrens zio_type_t type = zio->io_type; 3738fa9e4066Sahrens int flags = zio->io_flags; 3739fa9e4066Sahrens 3740e14bb325SJeff Bonwick /* 3741e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3742e14bb325SJeff Bonwick */ 3743e14bb325SJeff Bonwick if (zio->io_gang_tree) 3744e14bb325SJeff Bonwick return; 3745e14bb325SJeff Bonwick 3746fa9e4066Sahrens if (zio->io_error == 0) { 3747e14bb325SJeff Bonwick /* 3748e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3749e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3750e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3751e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3752e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3753e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3754e14bb325SJeff Bonwick * remain zero, which is what we want.) 3755e14bb325SJeff Bonwick * 3756e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3757e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3758e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3759e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3760e14bb325SJeff Bonwick */ 3761e14bb325SJeff Bonwick if (vd == rvd) 3762e14bb325SJeff Bonwick return; 3763e14bb325SJeff Bonwick 3764e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 37658ad4d6ddSJeff Bonwick 37668ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 37678ad4d6ddSJeff Bonwick return; 37688ad4d6ddSJeff Bonwick 37698ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 37708ad4d6ddSJeff Bonwick 3771e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 377244ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 37733f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 37743f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 37753f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 37763f9d6ad7SLin Ling 37773f9d6ad7SLin Ling /* XXX cleanup? */ 37783f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 37793f9d6ad7SLin Ling atomic_add_64(processed, psize); 37803f9d6ad7SLin Ling vs->vs_scan_processed += psize; 37813f9d6ad7SLin Ling } 37823f9d6ad7SLin Ling 37838ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3784e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3785fa9e4066Sahrens } 37868ad4d6ddSJeff Bonwick 3787*084fd14fSBrian Behlendorf zio_type_t vs_type = type; 3788*084fd14fSBrian Behlendorf 3789*084fd14fSBrian Behlendorf /* 3790*084fd14fSBrian Behlendorf * TRIM ops and bytes are reported to user space as 3791*084fd14fSBrian Behlendorf * ZIO_TYPE_IOCTL. This is done to preserve the 3792*084fd14fSBrian Behlendorf * vdev_stat_t structure layout for user space. 3793*084fd14fSBrian Behlendorf */ 3794*084fd14fSBrian Behlendorf if (type == ZIO_TYPE_TRIM) 3795*084fd14fSBrian Behlendorf vs_type = ZIO_TYPE_IOCTL; 3796*084fd14fSBrian Behlendorf 3797*084fd14fSBrian Behlendorf vs->vs_ops[vs_type]++; 3798*084fd14fSBrian Behlendorf vs->vs_bytes[vs_type] += psize; 37998ad4d6ddSJeff Bonwick 38008ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3801fa9e4066Sahrens return; 3802fa9e4066Sahrens } 3803fa9e4066Sahrens 3804fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3805fa9e4066Sahrens return; 3806fa9e4066Sahrens 38078956713aSEric Schrock /* 38088956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 38098956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 38108956713aSEric Schrock * hard errors, when in reality they can happen for any number of 38118956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 38128956713aSEric Schrock */ 38138956713aSEric Schrock if (zio->io_error == EIO && 38148956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 38158956713aSEric Schrock return; 38168956713aSEric Schrock 38178f18d1faSGeorge Wilson /* 38188f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 38198f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 38208f18d1faSGeorge Wilson * errors. 38218f18d1faSGeorge Wilson */ 38228f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 38238f18d1faSGeorge Wilson return; 38248f18d1faSGeorge Wilson 3825e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3826b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3827e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3828e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3829e14bb325SJeff Bonwick else 3830e14bb325SJeff Bonwick vs->vs_read_errors++; 3831fa9e4066Sahrens } 3832b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3833e14bb325SJeff Bonwick vs->vs_write_errors++; 3834e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3835fa9e4066Sahrens 38365cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 38375cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 38388ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 383944ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3840b24ab676SJeff Bonwick spa->spa_claiming)) { 38418ad4d6ddSJeff Bonwick /* 3842b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 3843b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 3844b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 3845b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 3846b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 3847b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 3848b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 3849b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 38508ad4d6ddSJeff Bonwick * 38518ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 38528ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 38538ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 38548ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 38558ad4d6ddSJeff Bonwick */ 38568ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 38578ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 385844ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 38598ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 38608ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 38618ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3862b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 3863b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 3864b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3865b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 38668ad4d6ddSJeff Bonwick } 3867b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 38688ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3869fa9e4066Sahrens return; 38708ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 38718ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 38728ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3873fa9e4066Sahrens } 38748ad4d6ddSJeff Bonwick if (vd != rvd) 38758ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3876fa9e4066Sahrens } 3877fa9e4066Sahrens } 3878fa9e4066Sahrens 3879663207adSDon Brady int64_t 3880663207adSDon Brady vdev_deflated_space(vdev_t *vd, int64_t space) 3881663207adSDon Brady { 3882663207adSDon Brady ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 3883663207adSDon Brady ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 3884663207adSDon Brady 3885663207adSDon Brady return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 3886663207adSDon Brady } 3887663207adSDon Brady 3888fa9e4066Sahrens /* 3889*084fd14fSBrian Behlendorf * Update the in-core space usage stats for this vdev, its metaslab class, 3890*084fd14fSBrian Behlendorf * and the root vdev. 3891fa9e4066Sahrens */ 3892fa9e4066Sahrens void 3893b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3894b24ab676SJeff Bonwick int64_t space_delta) 3895fa9e4066Sahrens { 3896663207adSDon Brady int64_t dspace_delta; 38978654d025Sperrin spa_t *spa = vd->vdev_spa; 38988654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 3899fa9e4066Sahrens 39008654d025Sperrin ASSERT(vd == vd->vdev_top); 390199653d4eSeschrock 39028654d025Sperrin /* 39038654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 39048654d025Sperrin * factor. We must calculate this here and not at the root vdev 39058654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 39068654d025Sperrin * childrens', thus not accurate enough for us. 39078654d025Sperrin */ 3908663207adSDon Brady dspace_delta = vdev_deflated_space(vd, space_delta); 39098654d025Sperrin 39108654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 39119740f25fSSerapheim Dimitropoulos /* ensure we won't underflow */ 39129740f25fSSerapheim Dimitropoulos if (alloc_delta < 0) { 39139740f25fSSerapheim Dimitropoulos ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 39149740f25fSSerapheim Dimitropoulos } 39159740f25fSSerapheim Dimitropoulos 39168654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 3917b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 39188654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 39198654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 39208654d025Sperrin 3921663207adSDon Brady /* every class but log contributes to root space stats */ 3922663207adSDon Brady if (vd->vdev_mg != NULL && !vd->vdev_islog) { 39239740f25fSSerapheim Dimitropoulos ASSERT(!vd->vdev_isl2cache); 3924fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 3925fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 3926b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 3927fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 3928fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 3929fa94a07fSbrendan } 3930663207adSDon Brady /* Note: metaslab_class_space_update moved to metaslab_space_update */ 3931fa9e4066Sahrens } 3932fa9e4066Sahrens 3933fa9e4066Sahrens /* 3934fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 3935fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 3936fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3937fa9e4066Sahrens */ 3938fa9e4066Sahrens void 3939fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3940fa9e4066Sahrens { 3941fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3942fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3943fa9e4066Sahrens int c; 3944fa9e4066Sahrens 3945f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3946f9af39baSGeorge Wilson 3947c5904d13Seschrock /* 39486809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 39496809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3950c5904d13Seschrock */ 3951c5904d13Seschrock if (vd->vdev_aux != NULL) { 3952c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3953c5904d13Seschrock nvlist_t **aux; 3954c5904d13Seschrock uint_t naux; 3955c5904d13Seschrock 3956c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3957c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3958c5904d13Seschrock break; 3959c5904d13Seschrock } 3960c5904d13Seschrock 3961e14bb325SJeff Bonwick if (c == sav->sav_count) { 3962e14bb325SJeff Bonwick /* 3963e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3964e14bb325SJeff Bonwick */ 3965e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3966e14bb325SJeff Bonwick return; 3967e14bb325SJeff Bonwick } 3968e14bb325SJeff Bonwick 3969c5904d13Seschrock sav->sav_sync = B_TRUE; 3970c5904d13Seschrock 39716809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 39726809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 39736809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 39746809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 39756809eb4eSEric Schrock } 3976c5904d13Seschrock 3977c5904d13Seschrock ASSERT(c < naux); 3978c5904d13Seschrock 3979c5904d13Seschrock /* 3980c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3981c5904d13Seschrock * sketchy, but it will work. 3982c5904d13Seschrock */ 3983c5904d13Seschrock nvlist_free(aux[c]); 39843f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3985c5904d13Seschrock 3986c5904d13Seschrock return; 3987c5904d13Seschrock } 3988c5904d13Seschrock 39895dabedeeSbonwick /* 3990e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3991e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3992e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 39935dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 39945dabedeeSbonwick */ 3995e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3996e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3997e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 39985dabedeeSbonwick 3999fa9e4066Sahrens if (vd == rvd) { 4000fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 4001fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 4002fa9e4066Sahrens } else { 4003fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 4004fa9e4066Sahrens 400588ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 40065cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 4007e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 40085cabbc6bSPrashanth Sreenivasa } 4009fa9e4066Sahrens } 4010fa9e4066Sahrens } 4011fa9e4066Sahrens 4012fa9e4066Sahrens void 4013fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 4014fa9e4066Sahrens { 40155dabedeeSbonwick spa_t *spa = vd->vdev_spa; 40165dabedeeSbonwick 4017e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4018e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4019e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 40205dabedeeSbonwick 4021e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4022e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 4023e14bb325SJeff Bonwick } 4024e14bb325SJeff Bonwick 4025e14bb325SJeff Bonwick /* 4026e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 4027e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4028e14bb325SJeff Bonwick * the state changes from larger config changes because they require 4029e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 4030e14bb325SJeff Bonwick */ 4031e14bb325SJeff Bonwick void 4032e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 4033e14bb325SJeff Bonwick { 4034e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4035e14bb325SJeff Bonwick 4036f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4037e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 4038e14bb325SJeff Bonwick 4039e14bb325SJeff Bonwick /* 4040e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 4041e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 4042e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 4043e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 4044e14bb325SJeff Bonwick */ 4045e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4046e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4047e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4048e14bb325SJeff Bonwick 40495cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 40505cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 4051e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 4052e14bb325SJeff Bonwick } 4053e14bb325SJeff Bonwick 4054e14bb325SJeff Bonwick void 4055e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 4056e14bb325SJeff Bonwick { 4057e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4058e14bb325SJeff Bonwick 4059e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4060e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4061e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4062e14bb325SJeff Bonwick 4063e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 4064e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 4065fa9e4066Sahrens } 4066fa9e4066Sahrens 406732b87932Sek /* 406832b87932Sek * Propagate vdev state up from children to parent. 406932b87932Sek */ 407044cd46caSbillm void 407144cd46caSbillm vdev_propagate_state(vdev_t *vd) 407244cd46caSbillm { 40738ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 40748ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 407544cd46caSbillm int degraded = 0, faulted = 0; 407644cd46caSbillm int corrupted = 0; 407744cd46caSbillm vdev_t *child; 407844cd46caSbillm 40793d7072f8Seschrock if (vd->vdev_children > 0) { 4080573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 40813d7072f8Seschrock child = vd->vdev_child[c]; 408251ece835Seschrock 408388ecc943SGeorge Wilson /* 40845cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 40855cabbc6bSPrashanth Sreenivasa * decision. 408688ecc943SGeorge Wilson */ 40875cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 408888ecc943SGeorge Wilson continue; 408988ecc943SGeorge Wilson 4090e14bb325SJeff Bonwick if (!vdev_readable(child) || 40918ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 409251ece835Seschrock /* 409351ece835Seschrock * Root special: if there is a top-level log 409451ece835Seschrock * device, treat the root vdev as if it were 409551ece835Seschrock * degraded. 409651ece835Seschrock */ 409751ece835Seschrock if (child->vdev_islog && vd == rvd) 409851ece835Seschrock degraded++; 409951ece835Seschrock else 410051ece835Seschrock faulted++; 410151ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 41023d7072f8Seschrock degraded++; 410351ece835Seschrock } 410444cd46caSbillm 41053d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 41063d7072f8Seschrock corrupted++; 41073d7072f8Seschrock } 410844cd46caSbillm 41093d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 41103d7072f8Seschrock 41113d7072f8Seschrock /* 4112e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 41133d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 41143d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 41153d7072f8Seschrock * replicas'. 41163d7072f8Seschrock */ 41173d7072f8Seschrock if (corrupted && vd == rvd && 41183d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 41193d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 41203d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 41213d7072f8Seschrock } 41223d7072f8Seschrock 412351ece835Seschrock if (vd->vdev_parent) 41243d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 412544cd46caSbillm } 412644cd46caSbillm 4127fa9e4066Sahrens /* 4128ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 4129ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 4130ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 4131ea8dc4b6Seschrock * 4132ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 4133ea8dc4b6Seschrock * generated. 4134fa9e4066Sahrens */ 4135fa9e4066Sahrens void 4136ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4137fa9e4066Sahrens { 4138560e6e96Seschrock uint64_t save_state; 4139c5904d13Seschrock spa_t *spa = vd->vdev_spa; 4140ea8dc4b6Seschrock 4141ea8dc4b6Seschrock if (state == vd->vdev_state) { 4142ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 4143fa9e4066Sahrens return; 4144ea8dc4b6Seschrock } 4145ea8dc4b6Seschrock 4146560e6e96Seschrock save_state = vd->vdev_state; 4147fa9e4066Sahrens 4148fa9e4066Sahrens vd->vdev_state = state; 4149fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 4150fa9e4066Sahrens 41513d7072f8Seschrock /* 41523d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 415398d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 415498d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 415598d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 415698d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 415798d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 415898d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 415998d1cbfeSGeorge Wilson * children. 416098d1cbfeSGeorge Wilson */ 416198d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 416298d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 41633d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 41643d7072f8Seschrock 4165069f55e2SEric Schrock /* 4166069f55e2SEric Schrock * If we have brought this vdev back into service, we need 4167069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 4168069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 4169069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 4170069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 4171069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 4172069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 4173069f55e2SEric Schrock */ 4174069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4175069f55e2SEric Schrock vd->vdev_prevstate != state) 4176069f55e2SEric Schrock zfs_post_state_change(spa, vd); 4177069f55e2SEric Schrock 41783d7072f8Seschrock if (vd->vdev_removed && 41793d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 41803d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 41813d7072f8Seschrock /* 41823d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 41833d7072f8Seschrock * device was previously marked removed and someone attempted to 41843d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 41853d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 41863d7072f8Seschrock * it is one of our special test online cases, which is only 41873d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 41883d7072f8Seschrock * fault. 41893d7072f8Seschrock */ 41903d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 41913d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 41923d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 41933d7072f8Seschrock vd->vdev_removed = B_TRUE; 41943d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 4195ea8dc4b6Seschrock /* 4196cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 4197cb04b873SMark J Musante * mark it as "not available", which signifies that it was 4198cb04b873SMark J Musante * never there to begin with. Failure to open such a device 4199cb04b873SMark J Musante * is not considered an error. 4200ea8dc4b6Seschrock */ 4201cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4202cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 4203560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 4204560e6e96Seschrock vd->vdev_not_present = 1; 4205560e6e96Seschrock 4206560e6e96Seschrock /* 4207560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 4208560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 4209560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 4210560e6e96Seschrock * want to post the ereport if the device was already in the 4211560e6e96Seschrock * CANT_OPEN state beforehand. 42123d7072f8Seschrock * 42133d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 42143d7072f8Seschrock * online the device in response to an insertion event. If we 42153d7072f8Seschrock * hit this case, then we have detected an insertion event for a 42163d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 42173d7072f8Seschrock * In this scenario, we don't post an ereport because we are 42183d7072f8Seschrock * about to replace the device, or attempt an online with 42193d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 4220560e6e96Seschrock */ 42213d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 42223d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 4223c5904d13Seschrock vd != spa->spa_root_vdev) { 4224ea8dc4b6Seschrock const char *class; 4225ea8dc4b6Seschrock 4226ea8dc4b6Seschrock switch (aux) { 4227ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 4228ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4229ea8dc4b6Seschrock break; 4230ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 4231ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4232ea8dc4b6Seschrock break; 4233ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 4234ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4235ea8dc4b6Seschrock break; 4236ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 4237ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4238ea8dc4b6Seschrock break; 4239ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 4240ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4241ea8dc4b6Seschrock break; 4242ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 4243ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4244ea8dc4b6Seschrock break; 42455711d393Sloli case VDEV_AUX_BAD_ASHIFT: 42465711d393Sloli class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 42475711d393Sloli break; 4248ea8dc4b6Seschrock default: 4249ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4250ea8dc4b6Seschrock } 4251ea8dc4b6Seschrock 4252eb633035STom Caputi zfs_ereport_post(class, spa, vd, NULL, NULL, 4253eb633035STom Caputi save_state, 0); 4254ea8dc4b6Seschrock } 4255ea8dc4b6Seschrock 42563d7072f8Seschrock /* Erase any notion of persistent removed state */ 42573d7072f8Seschrock vd->vdev_removed = B_FALSE; 42583d7072f8Seschrock } else { 42593d7072f8Seschrock vd->vdev_removed = B_FALSE; 42603d7072f8Seschrock } 4261ea8dc4b6Seschrock 42628b33d774STim Haley if (!isopen && vd->vdev_parent) 42638b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 4264fa9e4066Sahrens } 426515e6edf1Sgw 42666f793812SPavel Zakharov boolean_t 42676f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 42686f793812SPavel Zakharov { 42696f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 42706f793812SPavel Zakharov 42716f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 42726f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 42736f793812SPavel Zakharov return (B_FALSE); 42746f793812SPavel Zakharov } 42756f793812SPavel Zakharov 42766f793812SPavel Zakharov return (B_TRUE); 42776f793812SPavel Zakharov } 42786f793812SPavel Zakharov 427915e6edf1Sgw /* 428015e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 4281c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 4282c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 428315e6edf1Sgw */ 428415e6edf1Sgw boolean_t 428515e6edf1Sgw vdev_is_bootable(vdev_t *vd) 428615e6edf1Sgw { 428715e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 428815e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 428915e6edf1Sgw 429015e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 429115e6edf1Sgw vd->vdev_children > 1) { 429221c878feSToomas Soome int non_indirect = 0; 429321c878feSToomas Soome 429421c878feSToomas Soome for (int c = 0; c < vd->vdev_children; c++) { 429521c878feSToomas Soome vdev_type = 429621c878feSToomas Soome vd->vdev_child[c]->vdev_ops->vdev_op_type; 429721c878feSToomas Soome if (strcmp(vdev_type, VDEV_TYPE_INDIRECT) != 0) 429821c878feSToomas Soome non_indirect++; 429921c878feSToomas Soome } 430021c878feSToomas Soome /* 430121c878feSToomas Soome * non_indirect > 1 means we have more than one 430221c878feSToomas Soome * top-level vdev, so we stop here. 430321c878feSToomas Soome */ 430421c878feSToomas Soome if (non_indirect > 1) 430521c878feSToomas Soome return (B_FALSE); 430621c878feSToomas Soome } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 430715e6edf1Sgw return (B_FALSE); 430815e6edf1Sgw } 430915e6edf1Sgw } 431015e6edf1Sgw 4311573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 431215e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 431315e6edf1Sgw return (B_FALSE); 431415e6edf1Sgw } 431515e6edf1Sgw return (B_TRUE); 431615e6edf1Sgw } 4317e6ca193dSGeorge Wilson 43185cabbc6bSPrashanth Sreenivasa boolean_t 43195cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 43205cabbc6bSPrashanth Sreenivasa { 43215cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 43225cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 43235cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 43245cabbc6bSPrashanth Sreenivasa return (B_FALSE); 43255cabbc6bSPrashanth Sreenivasa } else { 43265cabbc6bSPrashanth Sreenivasa return (B_TRUE); 43275cabbc6bSPrashanth Sreenivasa } 43285cabbc6bSPrashanth Sreenivasa } 43295cabbc6bSPrashanth Sreenivasa 43304b964adaSGeorge Wilson /* 43314b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 43324b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 43334b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 43344b964adaSGeorge Wilson */ 43354b964adaSGeorge Wilson boolean_t 43364b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 43374b964adaSGeorge Wilson { 43384b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 43394b964adaSGeorge Wilson !vd->vdev_removed) 43404b964adaSGeorge Wilson return (B_TRUE); 43414b964adaSGeorge Wilson 43424b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 43434b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 43444b964adaSGeorge Wilson return (B_TRUE); 43454b964adaSGeorge Wilson 43464b964adaSGeorge Wilson return (B_FALSE); 43474b964adaSGeorge Wilson } 43484b964adaSGeorge Wilson 4349573ca77eSGeorge Wilson /* 4350573ca77eSGeorge Wilson * Expand a vdev if possible. 4351573ca77eSGeorge Wilson */ 4352573ca77eSGeorge Wilson void 4353573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 4354573ca77eSGeorge Wilson { 4355573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 4356573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 435711f6a968SSerapheim Dimitropoulos ASSERT(vdev_is_concrete(vd)); 4358573ca77eSGeorge Wilson 43595cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 43605cabbc6bSPrashanth Sreenivasa 4361663207adSDon Brady if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4362663207adSDon Brady vdev_is_concrete(vd)) { 4363663207adSDon Brady vdev_metaslab_group_create(vd); 4364573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 4365573ca77eSGeorge Wilson vdev_config_dirty(vd); 4366573ca77eSGeorge Wilson } 4367573ca77eSGeorge Wilson } 43681195e687SMark J Musante 43691195e687SMark J Musante /* 43701195e687SMark J Musante * Split a vdev. 43711195e687SMark J Musante */ 43721195e687SMark J Musante void 43731195e687SMark J Musante vdev_split(vdev_t *vd) 43741195e687SMark J Musante { 43751195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 43761195e687SMark J Musante 43771195e687SMark J Musante vdev_remove_child(pvd, vd); 43781195e687SMark J Musante vdev_compact_children(pvd); 43791195e687SMark J Musante 43801195e687SMark J Musante cvd = pvd->vdev_child[0]; 43811195e687SMark J Musante if (pvd->vdev_children == 1) { 43821195e687SMark J Musante vdev_remove_parent(cvd); 43831195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 43841195e687SMark J Musante } 43851195e687SMark J Musante vdev_propagate_state(cvd); 43861195e687SMark J Musante } 4387283b8460SGeorge.Wilson 4388283b8460SGeorge.Wilson void 4389283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 4390283b8460SGeorge.Wilson { 4391283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 4392283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 4393283b8460SGeorge.Wilson 4394283b8460SGeorge.Wilson vdev_deadman(cvd); 4395283b8460SGeorge.Wilson } 4396283b8460SGeorge.Wilson 4397283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 4398283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 4399283b8460SGeorge.Wilson 4400283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 440169962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 4402283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 4403283b8460SGeorge.Wilson zio_t *fio; 4404283b8460SGeorge.Wilson uint64_t delta; 4405283b8460SGeorge.Wilson 4406283b8460SGeorge.Wilson /* 4407283b8460SGeorge.Wilson * Look at the head of all the pending queues, 4408283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 4409283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 4410283b8460SGeorge.Wilson */ 441169962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 4412c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 4413c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 44143ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 44153ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 44163ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 4417283b8460SGeorge.Wilson vq->vq_io_complete_ts); 4418283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 4419283b8460SGeorge.Wilson "hung.", spa_name(spa)); 4420283b8460SGeorge.Wilson } 4421283b8460SGeorge.Wilson } 4422283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 4423283b8460SGeorge.Wilson } 4424283b8460SGeorge.Wilson } 4425e4c795beSTom Caputi 4426e4c795beSTom Caputi void 4427e4c795beSTom Caputi vdev_set_deferred_resilver(spa_t *spa, vdev_t *vd) 4428e4c795beSTom Caputi { 4429e4c795beSTom Caputi for (uint64_t i = 0; i < vd->vdev_children; i++) 4430e4c795beSTom Caputi vdev_set_deferred_resilver(spa, vd->vdev_child[i]); 4431e4c795beSTom Caputi 4432e4c795beSTom Caputi if (!vd->vdev_ops->vdev_op_leaf || !vdev_writeable(vd) || 4433e4c795beSTom Caputi range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 4434e4c795beSTom Caputi return; 4435e4c795beSTom Caputi } 4436e4c795beSTom Caputi 4437e4c795beSTom Caputi vd->vdev_resilver_deferred = B_TRUE; 4438e4c795beSTom Caputi spa->spa_resilver_deferred = B_TRUE; 4439e4c795beSTom Caputi } 4440*084fd14fSBrian Behlendorf 4441*084fd14fSBrian Behlendorf /* 4442*084fd14fSBrian Behlendorf * Translate a logical range to the physical range for the specified vdev_t. 4443*084fd14fSBrian Behlendorf * This function is initially called with a leaf vdev and will walk each 4444*084fd14fSBrian Behlendorf * parent vdev until it reaches a top-level vdev. Once the top-level is 4445*084fd14fSBrian Behlendorf * reached the physical range is initialized and the recursive function 4446*084fd14fSBrian Behlendorf * begins to unwind. As it unwinds it calls the parent's vdev specific 4447*084fd14fSBrian Behlendorf * translation function to do the real conversion. 4448*084fd14fSBrian Behlendorf */ 4449*084fd14fSBrian Behlendorf void 4450*084fd14fSBrian Behlendorf vdev_xlate(vdev_t *vd, const range_seg_t *logical_rs, range_seg_t *physical_rs) 4451*084fd14fSBrian Behlendorf { 4452*084fd14fSBrian Behlendorf /* 4453*084fd14fSBrian Behlendorf * Walk up the vdev tree 4454*084fd14fSBrian Behlendorf */ 4455*084fd14fSBrian Behlendorf if (vd != vd->vdev_top) { 4456*084fd14fSBrian Behlendorf vdev_xlate(vd->vdev_parent, logical_rs, physical_rs); 4457*084fd14fSBrian Behlendorf } else { 4458*084fd14fSBrian Behlendorf /* 4459*084fd14fSBrian Behlendorf * We've reached the top-level vdev, initialize the 4460*084fd14fSBrian Behlendorf * physical range to the logical range and start to 4461*084fd14fSBrian Behlendorf * unwind. 4462*084fd14fSBrian Behlendorf */ 4463*084fd14fSBrian Behlendorf physical_rs->rs_start = logical_rs->rs_start; 4464*084fd14fSBrian Behlendorf physical_rs->rs_end = logical_rs->rs_end; 4465*084fd14fSBrian Behlendorf return; 4466*084fd14fSBrian Behlendorf } 4467*084fd14fSBrian Behlendorf 4468*084fd14fSBrian Behlendorf vdev_t *pvd = vd->vdev_parent; 4469*084fd14fSBrian Behlendorf ASSERT3P(pvd, !=, NULL); 4470*084fd14fSBrian Behlendorf ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 4471*084fd14fSBrian Behlendorf 4472*084fd14fSBrian Behlendorf /* 4473*084fd14fSBrian Behlendorf * As this recursive function unwinds, translate the logical 4474*084fd14fSBrian Behlendorf * range into its physical components by calling the 4475*084fd14fSBrian Behlendorf * vdev specific translate function. 4476*084fd14fSBrian Behlendorf */ 4477*084fd14fSBrian Behlendorf range_seg_t intermediate = { { { 0, 0 } } }; 4478*084fd14fSBrian Behlendorf pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate); 4479*084fd14fSBrian Behlendorf 4480*084fd14fSBrian Behlendorf physical_rs->rs_start = intermediate.rs_start; 4481*084fd14fSBrian Behlendorf physical_rs->rs_end = intermediate.rs_end; 4482*084fd14fSBrian Behlendorf } 4483