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. 30*0c06d385Sjwpoduska * Copyright (c) 2019, Datto Inc. All rights reserved. 31fa9e4066Sahrens */ 32fa9e4066Sahrens 33fa9e4066Sahrens #include <sys/zfs_context.h> 34ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 35fa9e4066Sahrens #include <sys/spa.h> 36fa9e4066Sahrens #include <sys/spa_impl.h> 375cabbc6bSPrashanth Sreenivasa #include <sys/bpobj.h> 38fa9e4066Sahrens #include <sys/dmu.h> 39fa9e4066Sahrens #include <sys/dmu_tx.h> 405cabbc6bSPrashanth Sreenivasa #include <sys/dsl_dir.h> 41fa9e4066Sahrens #include <sys/vdev_impl.h> 42fa9e4066Sahrens #include <sys/uberblock_impl.h> 43fa9e4066Sahrens #include <sys/metaslab.h> 44fa9e4066Sahrens #include <sys/metaslab_impl.h> 45fa9e4066Sahrens #include <sys/space_map.h> 460713e232SGeorge Wilson #include <sys/space_reftree.h> 47fa9e4066Sahrens #include <sys/zio.h> 48fa9e4066Sahrens #include <sys/zap.h> 49fa9e4066Sahrens #include <sys/fs/zfs.h> 50c5904d13Seschrock #include <sys/arc.h> 51e6ca193dSGeorge Wilson #include <sys/zil.h> 523f9d6ad7SLin Ling #include <sys/dsl_scan.h> 53770499e1SDan Kimmel #include <sys/abd.h> 54094e47e9SGeorge Wilson #include <sys/vdev_initialize.h> 55084fd14fSBrian Behlendorf #include <sys/vdev_trim.h> 56fa9e4066Sahrens 57fa9e4066Sahrens /* 58fa9e4066Sahrens * Virtual device management. 59fa9e4066Sahrens */ 60fa9e4066Sahrens 61fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 62fa9e4066Sahrens &vdev_root_ops, 63fa9e4066Sahrens &vdev_raidz_ops, 64fa9e4066Sahrens &vdev_mirror_ops, 65fa9e4066Sahrens &vdev_replacing_ops, 6699653d4eSeschrock &vdev_spare_ops, 67fa9e4066Sahrens &vdev_disk_ops, 68fa9e4066Sahrens &vdev_file_ops, 69fa9e4066Sahrens &vdev_missing_ops, 7088ecc943SGeorge Wilson &vdev_hole_ops, 715cabbc6bSPrashanth Sreenivasa &vdev_indirect_ops, 72fa9e4066Sahrens NULL 73fa9e4066Sahrens }; 74fa9e4066Sahrens 75088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 76088f3894Sahrens int zfs_scrub_limit = 10; 7705b2b3b8Smishra 78a0b03b16SSerapheim Dimitropoulos /* default target for number of metaslabs per top-level vdev */ 79a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_count = 200; 8086714001SSerapheim Dimitropoulos 81b4bf0cf0SDon Brady /* minimum number of metaslabs per top-level vdev */ 82a0b03b16SSerapheim Dimitropoulos int zfs_vdev_min_ms_count = 16; 8386714001SSerapheim Dimitropoulos 84b4bf0cf0SDon Brady /* practical upper limit of total metaslabs per top-level vdev */ 85a0b03b16SSerapheim Dimitropoulos int zfs_vdev_ms_count_limit = 1ULL << 17; 86b4bf0cf0SDon Brady 87b4bf0cf0SDon Brady /* lower limit for metaslab size (512M) */ 88a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_shift = 29; 8986714001SSerapheim Dimitropoulos 90a0b03b16SSerapheim Dimitropoulos /* upper limit for metaslab size (16G) */ 91a0b03b16SSerapheim Dimitropoulos int zfs_vdev_max_ms_shift = 34; 92b4bf0cf0SDon Brady 9386714001SSerapheim Dimitropoulos boolean_t vdev_validate_skip = B_FALSE; 9486714001SSerapheim Dimitropoulos 95bf3e216cSMatthew Ahrens /* 9686714001SSerapheim Dimitropoulos * Since the DTL space map of a vdev is not expected to have a lot of 9786714001SSerapheim Dimitropoulos * entries, we default its block size to 4K. 98bf3e216cSMatthew Ahrens */ 99814dcd43SSerapheim Dimitropoulos int zfs_vdev_dtl_sm_blksz = (1 << 12); 100bf3e216cSMatthew Ahrens 10186714001SSerapheim Dimitropoulos /* 10286714001SSerapheim Dimitropoulos * vdev-wide space maps that have lots of entries written to them at 10386714001SSerapheim Dimitropoulos * the end of each transaction can benefit from a higher I/O bandwidth 10486714001SSerapheim Dimitropoulos * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 10586714001SSerapheim Dimitropoulos */ 106814dcd43SSerapheim Dimitropoulos int zfs_vdev_standard_sm_blksz = (1 << 17); 1076f793812SPavel Zakharov 10893a1902eSMatthew Ahrens int zfs_ashift_min; 10993a1902eSMatthew Ahrens 1103ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 1113ee8c80cSPavel Zakharov void 1123ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 1133ee8c80cSPavel Zakharov { 1143ee8c80cSPavel Zakharov va_list adx; 1153ee8c80cSPavel Zakharov char buf[256]; 1163ee8c80cSPavel Zakharov 1173ee8c80cSPavel Zakharov va_start(adx, fmt); 1183ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 1193ee8c80cSPavel Zakharov va_end(adx); 1203ee8c80cSPavel Zakharov 1213ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 1223ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 1233ee8c80cSPavel Zakharov vd->vdev_path, buf); 1243ee8c80cSPavel Zakharov } else { 1253ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 1263ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 1273ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1283ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1293ee8c80cSPavel Zakharov } 1303ee8c80cSPavel Zakharov } 1313ee8c80cSPavel Zakharov 1326f793812SPavel Zakharov void 1336f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 1346f793812SPavel Zakharov { 1356f793812SPavel Zakharov char state[20]; 1366f793812SPavel Zakharov 1376f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 1386f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 1396f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 1406f793812SPavel Zakharov return; 1416f793812SPavel Zakharov } 1426f793812SPavel Zakharov 1436f793812SPavel Zakharov switch (vd->vdev_state) { 1446f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 1456f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 1466f793812SPavel Zakharov break; 1476f793812SPavel Zakharov case VDEV_STATE_CLOSED: 1486f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 1496f793812SPavel Zakharov break; 1506f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 1516f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 1526f793812SPavel Zakharov break; 1536f793812SPavel Zakharov case VDEV_STATE_REMOVED: 1546f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 1556f793812SPavel Zakharov break; 1566f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 1576f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 1586f793812SPavel Zakharov break; 1596f793812SPavel Zakharov case VDEV_STATE_FAULTED: 1606f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 1616f793812SPavel Zakharov break; 1626f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 1636f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 1646f793812SPavel Zakharov break; 1656f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 1666f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 1676f793812SPavel Zakharov break; 1686f793812SPavel Zakharov default: 1696f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 1706f793812SPavel Zakharov (uint_t)vd->vdev_state); 1716f793812SPavel Zakharov } 1726f793812SPavel Zakharov 1736f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 174c7a7b2faSAndriy Gapon "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 1756f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 1766f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1776f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 1786f793812SPavel Zakharov 1796f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 1806f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 1816f793812SPavel Zakharov } 1826f793812SPavel Zakharov 183fa9e4066Sahrens /* 184fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 185fa9e4066Sahrens */ 186fa9e4066Sahrens static vdev_ops_t * 187fa9e4066Sahrens vdev_getops(const char *type) 188fa9e4066Sahrens { 189fa9e4066Sahrens vdev_ops_t *ops, **opspp; 190fa9e4066Sahrens 191fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 192fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 193fa9e4066Sahrens break; 194fa9e4066Sahrens 195fa9e4066Sahrens return (ops); 196fa9e4066Sahrens } 197fa9e4066Sahrens 198663207adSDon Brady /* 199663207adSDon Brady * Derive the enumerated alloction bias from string input. 200663207adSDon Brady * String origin is either the per-vdev zap or zpool(1M). 201663207adSDon Brady */ 202663207adSDon Brady static vdev_alloc_bias_t 203663207adSDon Brady vdev_derive_alloc_bias(const char *bias) 204663207adSDon Brady { 205663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 206663207adSDon Brady 207663207adSDon Brady if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 208663207adSDon Brady alloc_bias = VDEV_BIAS_LOG; 209663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 210663207adSDon Brady alloc_bias = VDEV_BIAS_SPECIAL; 211663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 212663207adSDon Brady alloc_bias = VDEV_BIAS_DEDUP; 213663207adSDon Brady 214663207adSDon Brady return (alloc_bias); 215663207adSDon Brady } 216663207adSDon Brady 217094e47e9SGeorge Wilson /* ARGSUSED */ 218094e47e9SGeorge Wilson void 2194d7988d6SPaul Dagnelie vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res) 220094e47e9SGeorge Wilson { 221094e47e9SGeorge Wilson res->rs_start = in->rs_start; 222094e47e9SGeorge Wilson res->rs_end = in->rs_end; 223094e47e9SGeorge Wilson } 224094e47e9SGeorge Wilson 225fa9e4066Sahrens /* 226fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 227fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 228fa9e4066Sahrens */ 229fa9e4066Sahrens uint64_t 230fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 231fa9e4066Sahrens { 232ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 233fa9e4066Sahrens uint64_t csize; 234fa9e4066Sahrens 235573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 236fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 237fa9e4066Sahrens asize = MAX(asize, csize); 238fa9e4066Sahrens } 239fa9e4066Sahrens 240fa9e4066Sahrens return (asize); 241fa9e4066Sahrens } 242fa9e4066Sahrens 2432a79c5feSlling /* 244573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 245573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 246573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 247573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 2482a79c5feSlling */ 2492a79c5feSlling uint64_t 250573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 2512a79c5feSlling { 252573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 2532a79c5feSlling 254573ca77eSGeorge Wilson /* 2554263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 256573ca77eSGeorge Wilson * just return our own asize. 257573ca77eSGeorge Wilson */ 258573ca77eSGeorge Wilson if (pvd == NULL) 259573ca77eSGeorge Wilson return (vd->vdev_asize); 2602a79c5feSlling 2612a79c5feSlling /* 262573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 263573ca77eSGeorge Wilson * to the nearest metaslab. 2642a79c5feSlling */ 265573ca77eSGeorge Wilson if (vd == vd->vdev_top) 266573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2672a79c5feSlling 268573ca77eSGeorge Wilson /* 269573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 270573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 271573ca77eSGeorge Wilson */ 272573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 273c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 274c040c10cSSteven Hartland pvd->vdev_children); 2752a79c5feSlling 276573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 277573ca77eSGeorge Wilson } 2782a79c5feSlling 279573ca77eSGeorge Wilson void 280573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 281573ca77eSGeorge Wilson { 282573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 283573ca77eSGeorge Wilson 284573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 285573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2862a79c5feSlling } 2872a79c5feSlling 288fa9e4066Sahrens vdev_t * 289fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 290fa9e4066Sahrens { 291fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 292fa9e4066Sahrens 293e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 294e05725b1Sbonwick 295088f3894Sahrens if (vdev < rvd->vdev_children) { 296088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 297fa9e4066Sahrens return (rvd->vdev_child[vdev]); 298088f3894Sahrens } 299fa9e4066Sahrens 300fa9e4066Sahrens return (NULL); 301fa9e4066Sahrens } 302fa9e4066Sahrens 303fa9e4066Sahrens vdev_t * 304fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 305fa9e4066Sahrens { 306fa9e4066Sahrens vdev_t *mvd; 307fa9e4066Sahrens 3080e34b6a7Sbonwick if (vd->vdev_guid == guid) 309fa9e4066Sahrens return (vd); 310fa9e4066Sahrens 311573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 312fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 313fa9e4066Sahrens NULL) 314fa9e4066Sahrens return (mvd); 315fa9e4066Sahrens 316fa9e4066Sahrens return (NULL); 317fa9e4066Sahrens } 318fa9e4066Sahrens 31912380e1eSArne Jansen static int 32012380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 32112380e1eSArne Jansen { 32212380e1eSArne Jansen int n = 0; 32312380e1eSArne Jansen 32412380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 32512380e1eSArne Jansen return (1); 32612380e1eSArne Jansen 32712380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 32812380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 32912380e1eSArne Jansen 33012380e1eSArne Jansen return (n); 33112380e1eSArne Jansen } 33212380e1eSArne Jansen 33312380e1eSArne Jansen int 33412380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 33512380e1eSArne Jansen { 33612380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 33712380e1eSArne Jansen } 33812380e1eSArne Jansen 339fa9e4066Sahrens void 340fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 341fa9e4066Sahrens { 342fa9e4066Sahrens size_t oldsize, newsize; 343fa9e4066Sahrens uint64_t id = cvd->vdev_id; 344fa9e4066Sahrens vdev_t **newchild; 34581cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 346fa9e4066Sahrens 34781cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 348fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 349fa9e4066Sahrens 350fa9e4066Sahrens cvd->vdev_parent = pvd; 351fa9e4066Sahrens 352fa9e4066Sahrens if (pvd == NULL) 353fa9e4066Sahrens return; 354fa9e4066Sahrens 355fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 356fa9e4066Sahrens 357fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 358fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 359fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 360fa9e4066Sahrens 361fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 362fa9e4066Sahrens if (pvd->vdev_child != NULL) { 363fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 364fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 365fa9e4066Sahrens } 366fa9e4066Sahrens 367fa9e4066Sahrens pvd->vdev_child = newchild; 368fa9e4066Sahrens pvd->vdev_child[id] = cvd; 369fa9e4066Sahrens 370fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 371fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 372fa9e4066Sahrens 373fa9e4066Sahrens /* 374fa9e4066Sahrens * Walk up all ancestors to update guid sum. 375fa9e4066Sahrens */ 376fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 377fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 378e0f1c0afSOlaf Faaland 379e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 380e0f1c0afSOlaf Faaland list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 381e0f1c0afSOlaf Faaland cvd->vdev_spa->spa_leaf_list_gen++; 382e0f1c0afSOlaf Faaland } 383fa9e4066Sahrens } 384fa9e4066Sahrens 385fa9e4066Sahrens void 386fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 387fa9e4066Sahrens { 388fa9e4066Sahrens int c; 389fa9e4066Sahrens uint_t id = cvd->vdev_id; 390fa9e4066Sahrens 391fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 392fa9e4066Sahrens 393fa9e4066Sahrens if (pvd == NULL) 394fa9e4066Sahrens return; 395fa9e4066Sahrens 396fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 397fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 398fa9e4066Sahrens 399fa9e4066Sahrens pvd->vdev_child[id] = NULL; 400fa9e4066Sahrens cvd->vdev_parent = NULL; 401fa9e4066Sahrens 402fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 403fa9e4066Sahrens if (pvd->vdev_child[c]) 404fa9e4066Sahrens break; 405fa9e4066Sahrens 406fa9e4066Sahrens if (c == pvd->vdev_children) { 407fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 408fa9e4066Sahrens pvd->vdev_child = NULL; 409fa9e4066Sahrens pvd->vdev_children = 0; 410fa9e4066Sahrens } 411fa9e4066Sahrens 412e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 413e0f1c0afSOlaf Faaland spa_t *spa = cvd->vdev_spa; 414e0f1c0afSOlaf Faaland list_remove(&spa->spa_leaf_list, cvd); 415e0f1c0afSOlaf Faaland spa->spa_leaf_list_gen++; 416e0f1c0afSOlaf Faaland } 417e0f1c0afSOlaf Faaland 418fa9e4066Sahrens /* 419fa9e4066Sahrens * Walk up all ancestors to update guid sum. 420fa9e4066Sahrens */ 421fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 422fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 423fa9e4066Sahrens } 424fa9e4066Sahrens 425fa9e4066Sahrens /* 426fa9e4066Sahrens * Remove any holes in the child array. 427fa9e4066Sahrens */ 428fa9e4066Sahrens void 429fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 430fa9e4066Sahrens { 431fa9e4066Sahrens vdev_t **newchild, *cvd; 432fa9e4066Sahrens int oldc = pvd->vdev_children; 433573ca77eSGeorge Wilson int newc; 434fa9e4066Sahrens 435e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 436fa9e4066Sahrens 437573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 438fa9e4066Sahrens if (pvd->vdev_child[c]) 439fa9e4066Sahrens newc++; 440fa9e4066Sahrens 441fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 442fa9e4066Sahrens 443573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 444fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 445fa9e4066Sahrens newchild[newc] = cvd; 446fa9e4066Sahrens cvd->vdev_id = newc++; 447fa9e4066Sahrens } 448fa9e4066Sahrens } 449fa9e4066Sahrens 450fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 451fa9e4066Sahrens pvd->vdev_child = newchild; 452fa9e4066Sahrens pvd->vdev_children = newc; 453fa9e4066Sahrens } 454fa9e4066Sahrens 455fa9e4066Sahrens /* 456fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 457fa9e4066Sahrens */ 45888ecc943SGeorge Wilson vdev_t * 459fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 460fa9e4066Sahrens { 461fa9e4066Sahrens vdev_t *vd; 4625cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 463fa9e4066Sahrens 464fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 4655cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 466fa9e4066Sahrens 4670e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4680e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4690e34b6a7Sbonwick spa->spa_root_vdev = vd; 470e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4710e34b6a7Sbonwick } 4720e34b6a7Sbonwick 47388ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4740e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4750e34b6a7Sbonwick /* 4760e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4770e34b6a7Sbonwick * which must be unique among all pools. 4780e34b6a7Sbonwick */ 4791195e687SMark J Musante guid = spa_generate_guid(NULL); 4800e34b6a7Sbonwick } else { 4810e34b6a7Sbonwick /* 4820e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4830e34b6a7Sbonwick */ 4841195e687SMark J Musante guid = spa_generate_guid(spa); 4850e34b6a7Sbonwick } 4860e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4870e34b6a7Sbonwick } 4880e34b6a7Sbonwick 489fa9e4066Sahrens vd->vdev_spa = spa; 490fa9e4066Sahrens vd->vdev_id = id; 491fa9e4066Sahrens vd->vdev_guid = guid; 492fa9e4066Sahrens vd->vdev_guid_sum = guid; 493fa9e4066Sahrens vd->vdev_ops = ops; 494fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 49588ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 4965cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 4975cabbc6bSPrashanth Sreenivasa 4985cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 4995cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 5004d7988d6SPaul Dagnelie vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 5014d7988d6SPaul Dagnelie 0, 0); 502fa9e4066Sahrens 503084fd14fSBrian Behlendorf list_link_init(&vd->vdev_initialize_node); 504e0f1c0afSOlaf Faaland list_link_init(&vd->vdev_leaf_node); 505084fd14fSBrian Behlendorf list_link_init(&vd->vdev_trim_node); 506fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 5075ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 508e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 509a3874b8bSToomas Soome mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 510094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 511094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 512094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 513094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 514084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 515084fd14fSBrian Behlendorf mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 516084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 517084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 518084fd14fSBrian Behlendorf cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 519084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 520094e47e9SGeorge Wilson 5218ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 5224d7988d6SPaul Dagnelie vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 5234d7988d6SPaul Dagnelie 0); 5248ad4d6ddSJeff Bonwick } 525b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 526fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 527b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 528fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 529fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 5303d7072f8Seschrock vdev_queue_init(vd); 5313d7072f8Seschrock vdev_cache_init(vd); 532fa9e4066Sahrens 533fa9e4066Sahrens return (vd); 534fa9e4066Sahrens } 535fa9e4066Sahrens 536fa9e4066Sahrens /* 537fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 538fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 539fa9e4066Sahrens * different for each case. 540fa9e4066Sahrens */ 54199653d4eSeschrock int 54299653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 54399653d4eSeschrock int alloctype) 544fa9e4066Sahrens { 545fa9e4066Sahrens vdev_ops_t *ops; 546fa9e4066Sahrens char *type; 5478654d025Sperrin uint64_t guid = 0, islog, nparity; 548fa9e4066Sahrens vdev_t *vd; 5495cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 550663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 551663207adSDon Brady boolean_t top_level = (parent && !parent->vdev_parent); 552fa9e4066Sahrens 553e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 554fa9e4066Sahrens 555fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 556be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 557fa9e4066Sahrens 558fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 559be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 560fa9e4066Sahrens 561fa9e4066Sahrens /* 562fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 563fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 564fa9e4066Sahrens */ 565fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 566fa9e4066Sahrens uint64_t label_id; 567fa9e4066Sahrens 568fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 569fa9e4066Sahrens label_id != id) 570be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 571fa9e4066Sahrens 572fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 573be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 57499653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 57599653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 576be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 577fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 578fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 579be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 58021ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 58121ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 582be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 583fa9e4066Sahrens } 584fa9e4066Sahrens 58599653d4eSeschrock /* 58699653d4eSeschrock * The first allocated vdev must be of type 'root'. 58799653d4eSeschrock */ 58899653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 589be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 59099653d4eSeschrock 5918654d025Sperrin /* 5928654d025Sperrin * Determine whether we're a log vdev. 5938654d025Sperrin */ 5948654d025Sperrin islog = 0; 5958654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 596990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 597be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 598fa9e4066Sahrens 59988ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 600be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 60188ecc943SGeorge Wilson 60299653d4eSeschrock /* 6038654d025Sperrin * Set the nparity property for RAID-Z vdevs. 60499653d4eSeschrock */ 6058654d025Sperrin nparity = -1ULL; 60699653d4eSeschrock if (ops == &vdev_raidz_ops) { 60799653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 6088654d025Sperrin &nparity) == 0) { 609b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 610be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 61199653d4eSeschrock /* 612f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 613f94275ceSAdam Leventhal * device. 61499653d4eSeschrock */ 615f94275ceSAdam Leventhal if (nparity > 1 && 616f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 617be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 618f94275ceSAdam Leventhal if (nparity > 2 && 619f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 620be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 62199653d4eSeschrock } else { 62299653d4eSeschrock /* 62399653d4eSeschrock * We require the parity to be specified for SPAs that 62499653d4eSeschrock * support multiple parity levels. 62599653d4eSeschrock */ 626f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 627be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 62899653d4eSeschrock /* 62999653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 63099653d4eSeschrock */ 6318654d025Sperrin nparity = 1; 63299653d4eSeschrock } 63399653d4eSeschrock } else { 6348654d025Sperrin nparity = 0; 63599653d4eSeschrock } 6368654d025Sperrin ASSERT(nparity != -1ULL); 6378654d025Sperrin 638663207adSDon Brady /* 639663207adSDon Brady * If creating a top-level vdev, check for allocation classes input 640663207adSDon Brady */ 641663207adSDon Brady if (top_level && alloctype == VDEV_ALLOC_ADD) { 642663207adSDon Brady char *bias; 643663207adSDon Brady 644663207adSDon Brady if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 645663207adSDon Brady &bias) == 0) { 646663207adSDon Brady alloc_bias = vdev_derive_alloc_bias(bias); 647663207adSDon Brady 648663207adSDon Brady /* spa_vdev_add() expects feature to be enabled */ 649c1064fd7SJerry Jelinek if (alloc_bias != VDEV_BIAS_LOG && 650c1064fd7SJerry Jelinek spa->spa_load_state != SPA_LOAD_CREATE && 651663207adSDon Brady !spa_feature_is_enabled(spa, 652663207adSDon Brady SPA_FEATURE_ALLOCATION_CLASSES)) { 653663207adSDon Brady return (SET_ERROR(ENOTSUP)); 654663207adSDon Brady } 655663207adSDon Brady } 656663207adSDon Brady } 657663207adSDon Brady 6588654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 6595cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 6608654d025Sperrin 6618654d025Sperrin vd->vdev_islog = islog; 6628654d025Sperrin vd->vdev_nparity = nparity; 663663207adSDon Brady if (top_level && alloc_bias != VDEV_BIAS_NONE) 664663207adSDon Brady vd->vdev_alloc_bias = alloc_bias; 6658654d025Sperrin 6668654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 6678654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 6688654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 6698654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 6708654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 6718654d025Sperrin &vd->vdev_physpath) == 0) 6728654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 6736809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 6746809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 67599653d4eSeschrock 676afefbcddSeschrock /* 677afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 678afefbcddSeschrock * as -1. 679afefbcddSeschrock */ 680afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 681afefbcddSeschrock &vd->vdev_wholedisk) != 0) 682afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 683afefbcddSeschrock 6845cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 6855cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 6865cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 6875cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 6885cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 6895cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 6905cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 6915cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 6925cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 6935cabbc6bSPrashanth Sreenivasa 694ea8dc4b6Seschrock /* 695ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 696ea8dc4b6Seschrock * was not present at the time of import. 697ea8dc4b6Seschrock */ 6986809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 6996809eb4eSEric Schrock &vd->vdev_not_present); 700ea8dc4b6Seschrock 701ecc2d604Sbonwick /* 702ecc2d604Sbonwick * Get the alignment requirement. 703ecc2d604Sbonwick */ 704ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 705ecc2d604Sbonwick 70688ecc943SGeorge Wilson /* 70788ecc943SGeorge Wilson * Retrieve the vdev creation time. 70888ecc943SGeorge Wilson */ 70988ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 71088ecc943SGeorge Wilson &vd->vdev_crtxg); 71188ecc943SGeorge Wilson 712fa9e4066Sahrens /* 713fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 714fa9e4066Sahrens */ 715663207adSDon Brady if (top_level && 7161195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 717fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 718fa9e4066Sahrens &vd->vdev_ms_array); 719fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 720fa9e4066Sahrens &vd->vdev_ms_shift); 721fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 722fa9e4066Sahrens &vd->vdev_asize); 7233f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 7243f9d6ad7SLin Ling &vd->vdev_removing); 725215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 726215198a6SJoe Stein &vd->vdev_top_zap); 727215198a6SJoe Stein } else { 728215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 729fa9e4066Sahrens } 730fa9e4066Sahrens 731663207adSDon Brady if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 732a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 7339f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 7341195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 7359f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 736663207adSDon Brady /* Note: metaslab_group_create() is now deferred */ 737a1521560SJeff Bonwick } 738a1521560SJeff Bonwick 739215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 740215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 741215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 742215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 743215198a6SJoe Stein } else { 744215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 745215198a6SJoe Stein } 746215198a6SJoe Stein 747fa9e4066Sahrens /* 7483d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 749fa9e4066Sahrens */ 750215198a6SJoe Stein 751c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 75221ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 75321ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 754c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 755c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 7560713e232SGeorge Wilson &vd->vdev_dtl_object); 757c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 758c5904d13Seschrock &vd->vdev_unspare); 759c5904d13Seschrock } 76021ecdf64SLin Ling 76121ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 76221ecdf64SLin Ling uint64_t spare = 0; 76321ecdf64SLin Ling 76421ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 76521ecdf64SLin Ling &spare) == 0 && spare) 76621ecdf64SLin Ling spa_spare_add(vd); 76721ecdf64SLin Ling } 76821ecdf64SLin Ling 769ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 770ecc2d604Sbonwick &vd->vdev_offline); 771c5904d13Seschrock 772b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 773b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 774cb04b873SMark J Musante 775e4c795beSTom Caputi if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 776*0c06d385Sjwpoduska vdev_defer_resilver(vd); 777e4c795beSTom Caputi 7783d7072f8Seschrock /* 7793d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 7803d7072f8Seschrock * state, as the diagnosis made on another system may not be 781069f55e2SEric Schrock * valid in the current context. Local vdevs will 782069f55e2SEric Schrock * remain in the faulted state. 7833d7072f8Seschrock */ 784b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 7853d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 7863d7072f8Seschrock &vd->vdev_faulted); 7873d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 7883d7072f8Seschrock &vd->vdev_degraded); 7893d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 7903d7072f8Seschrock &vd->vdev_removed); 791069f55e2SEric Schrock 792069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 793069f55e2SEric Schrock char *aux; 794069f55e2SEric Schrock 795069f55e2SEric Schrock vd->vdev_label_aux = 796069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 797069f55e2SEric Schrock if (nvlist_lookup_string(nv, 798069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 799069f55e2SEric Schrock strcmp(aux, "external") == 0) 800069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 801069f55e2SEric Schrock } 8023d7072f8Seschrock } 803fa9e4066Sahrens } 804fa9e4066Sahrens 805fa9e4066Sahrens /* 806fa9e4066Sahrens * Add ourselves to the parent's list of children. 807fa9e4066Sahrens */ 808fa9e4066Sahrens vdev_add_child(parent, vd); 809fa9e4066Sahrens 81099653d4eSeschrock *vdp = vd; 81199653d4eSeschrock 81299653d4eSeschrock return (0); 813fa9e4066Sahrens } 814fa9e4066Sahrens 815fa9e4066Sahrens void 816fa9e4066Sahrens vdev_free(vdev_t *vd) 817fa9e4066Sahrens { 8183d7072f8Seschrock spa_t *spa = vd->vdev_spa; 819084fd14fSBrian Behlendorf 820094e47e9SGeorge Wilson ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 821084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_trim_thread, ==, NULL); 822084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 823fa9e4066Sahrens 824a3874b8bSToomas Soome /* 825a3874b8bSToomas Soome * Scan queues are normally destroyed at the end of a scan. If the 826a3874b8bSToomas Soome * queue exists here, that implies the vdev is being removed while 827a3874b8bSToomas Soome * the scan is still running. 828a3874b8bSToomas Soome */ 829a3874b8bSToomas Soome if (vd->vdev_scan_io_queue != NULL) { 830a3874b8bSToomas Soome mutex_enter(&vd->vdev_scan_io_queue_lock); 831a3874b8bSToomas Soome dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 832a3874b8bSToomas Soome vd->vdev_scan_io_queue = NULL; 833a3874b8bSToomas Soome mutex_exit(&vd->vdev_scan_io_queue_lock); 834a3874b8bSToomas Soome } 835a3874b8bSToomas Soome 836fa9e4066Sahrens /* 837fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 838fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 839fa9e4066Sahrens */ 840fa9e4066Sahrens vdev_close(vd); 841fa9e4066Sahrens 842e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 843b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 844fa9e4066Sahrens 845fa9e4066Sahrens /* 846fa9e4066Sahrens * Free all children. 847fa9e4066Sahrens */ 848573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 849fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 850fa9e4066Sahrens 851fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 852fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 853fa9e4066Sahrens 854fa9e4066Sahrens /* 855fa9e4066Sahrens * Discard allocation state. 856fa9e4066Sahrens */ 857a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 858fa9e4066Sahrens vdev_metaslab_fini(vd); 859a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 860814dcd43SSerapheim Dimitropoulos vd->vdev_mg = NULL; 861a1521560SJeff Bonwick } 862fa9e4066Sahrens 863fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 864fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 865fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 866fa9e4066Sahrens 867fa9e4066Sahrens /* 868fa9e4066Sahrens * Remove this vdev from its parent's child list. 869fa9e4066Sahrens */ 870fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 871fa9e4066Sahrens 872fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 873e0f1c0afSOlaf Faaland ASSERT(!list_link_active(&vd->vdev_leaf_node)); 874fa9e4066Sahrens 8753d7072f8Seschrock /* 8763d7072f8Seschrock * Clean up vdev structure. 8773d7072f8Seschrock */ 8783d7072f8Seschrock vdev_queue_fini(vd); 8793d7072f8Seschrock vdev_cache_fini(vd); 8803d7072f8Seschrock 8813d7072f8Seschrock if (vd->vdev_path) 8823d7072f8Seschrock spa_strfree(vd->vdev_path); 8833d7072f8Seschrock if (vd->vdev_devid) 8843d7072f8Seschrock spa_strfree(vd->vdev_devid); 8853d7072f8Seschrock if (vd->vdev_physpath) 8863d7072f8Seschrock spa_strfree(vd->vdev_physpath); 8876809eb4eSEric Schrock if (vd->vdev_fru) 8886809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 8893d7072f8Seschrock 8903d7072f8Seschrock if (vd->vdev_isspare) 8913d7072f8Seschrock spa_spare_remove(vd); 892fa94a07fSbrendan if (vd->vdev_isl2cache) 893fa94a07fSbrendan spa_l2cache_remove(vd); 8943d7072f8Seschrock 8953d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 8963d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 8978ad4d6ddSJeff Bonwick 8983d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 8990713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 9008ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 9010713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 9020713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 9038ad4d6ddSJeff Bonwick } 9043d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 9058ad4d6ddSJeff Bonwick 9065cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 9075cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 9085cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 9095cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 9105cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 9115cabbc6bSPrashanth Sreenivasa } 9125cabbc6bSPrashanth Sreenivasa 9135cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 9145cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 9155cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 9165cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 9175cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 9185cabbc6bSPrashanth Sreenivasa } 9195cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 9205cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 9215cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 9225cabbc6bSPrashanth Sreenivasa 9233d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 9243d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 925e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 926a3874b8bSToomas Soome mutex_destroy(&vd->vdev_scan_io_queue_lock); 927094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_lock); 928094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_io_lock); 929094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_io_cv); 930094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_cv); 931084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_lock); 932084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_autotrim_lock); 933084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_io_lock); 934084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_cv); 935084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_autotrim_cv); 936084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_io_cv); 9373d7072f8Seschrock 9383d7072f8Seschrock if (vd == spa->spa_root_vdev) 9393d7072f8Seschrock spa->spa_root_vdev = NULL; 9403d7072f8Seschrock 9413d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 942fa9e4066Sahrens } 943fa9e4066Sahrens 944fa9e4066Sahrens /* 945fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 946fa9e4066Sahrens */ 947fa9e4066Sahrens static void 948fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 949fa9e4066Sahrens { 950fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 951fa9e4066Sahrens metaslab_t *msp; 952fa9e4066Sahrens vdev_t *vd; 953fa9e4066Sahrens int t; 954fa9e4066Sahrens 955fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 956fa9e4066Sahrens 957fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 958fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 959fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 960215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 961fa9e4066Sahrens 962fa9e4066Sahrens svd->vdev_ms_array = 0; 963fa9e4066Sahrens svd->vdev_ms_shift = 0; 964fa9e4066Sahrens svd->vdev_ms_count = 0; 965215198a6SJoe Stein svd->vdev_top_zap = 0; 966fa9e4066Sahrens 967cd0837ccSGeorge Wilson if (tvd->vdev_mg) 968cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 969fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 970fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 971fa9e4066Sahrens 972fa9e4066Sahrens svd->vdev_mg = NULL; 973fa9e4066Sahrens svd->vdev_ms = NULL; 974ecc2d604Sbonwick 975ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 976ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 977fa9e4066Sahrens 97886714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 97986714001SSerapheim Dimitropoulos svd->vdev_checkpoint_sm = NULL; 98086714001SSerapheim Dimitropoulos 981663207adSDon Brady tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 982663207adSDon Brady svd->vdev_alloc_bias = VDEV_BIAS_NONE; 983663207adSDon Brady 984fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 985fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 98699653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 987fa9e4066Sahrens 988fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 989fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 99099653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 991fa9e4066Sahrens 9923a4b1be9SMatthew Ahrens /* 9933a4b1be9SMatthew Ahrens * State which may be set on a top-level vdev that's in the 9943a4b1be9SMatthew Ahrens * process of being removed. 9953a4b1be9SMatthew Ahrens */ 9963a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_births_object); 9973a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 9983a4b1be9SMatthew Ahrens ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 9993a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 10003a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 10013a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 10023a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_removing); 10033a4b1be9SMatthew Ahrens tvd->vdev_removing = svd->vdev_removing; 10043a4b1be9SMatthew Ahrens tvd->vdev_indirect_config = svd->vdev_indirect_config; 10053a4b1be9SMatthew Ahrens tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 10063a4b1be9SMatthew Ahrens tvd->vdev_indirect_births = svd->vdev_indirect_births; 10073a4b1be9SMatthew Ahrens range_tree_swap(&svd->vdev_obsolete_segments, 10083a4b1be9SMatthew Ahrens &tvd->vdev_obsolete_segments); 10093a4b1be9SMatthew Ahrens tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 10103a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_mapping_object = 0; 10113a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_births_object = 0; 10123a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 10133a4b1be9SMatthew Ahrens svd->vdev_indirect_mapping = NULL; 10143a4b1be9SMatthew Ahrens svd->vdev_indirect_births = NULL; 10153a4b1be9SMatthew Ahrens svd->vdev_obsolete_sm = NULL; 10163a4b1be9SMatthew Ahrens svd->vdev_removing = 0; 10173a4b1be9SMatthew Ahrens 1018fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 1019fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1020fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1021fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1022fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1023fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1024fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1025fa9e4066Sahrens } 1026fa9e4066Sahrens 1027e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 1028fa9e4066Sahrens vdev_config_clean(svd); 1029fa9e4066Sahrens vdev_config_dirty(tvd); 1030fa9e4066Sahrens } 1031fa9e4066Sahrens 1032e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 1033e14bb325SJeff Bonwick vdev_state_clean(svd); 1034e14bb325SJeff Bonwick vdev_state_dirty(tvd); 1035e14bb325SJeff Bonwick } 1036e14bb325SJeff Bonwick 103799653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 103899653d4eSeschrock svd->vdev_deflate_ratio = 0; 10398654d025Sperrin 10408654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 10418654d025Sperrin svd->vdev_islog = 0; 1042a3874b8bSToomas Soome 1043a3874b8bSToomas Soome dsl_scan_io_queue_vdev_xfer(svd, tvd); 1044fa9e4066Sahrens } 1045fa9e4066Sahrens 1046fa9e4066Sahrens static void 1047fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 1048fa9e4066Sahrens { 1049fa9e4066Sahrens if (vd == NULL) 1050fa9e4066Sahrens return; 1051fa9e4066Sahrens 1052fa9e4066Sahrens vd->vdev_top = tvd; 1053fa9e4066Sahrens 1054573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1055fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 1056fa9e4066Sahrens } 1057fa9e4066Sahrens 1058fa9e4066Sahrens /* 1059fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 1060fa9e4066Sahrens */ 1061fa9e4066Sahrens vdev_t * 1062fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1063fa9e4066Sahrens { 1064fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 1065fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 1066fa9e4066Sahrens vdev_t *mvd; 1067fa9e4066Sahrens 1068e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1069fa9e4066Sahrens 1070fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1071ecc2d604Sbonwick 1072ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 1073573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 10744263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 10755cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 1076ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 1077ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 107888ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 1079ecc2d604Sbonwick 1080fa9e4066Sahrens vdev_remove_child(pvd, cvd); 1081fa9e4066Sahrens vdev_add_child(pvd, mvd); 1082fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 1083fa9e4066Sahrens vdev_add_child(mvd, cvd); 1084fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1085fa9e4066Sahrens 1086fa9e4066Sahrens if (mvd == mvd->vdev_top) 1087fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 1088fa9e4066Sahrens 1089fa9e4066Sahrens return (mvd); 1090fa9e4066Sahrens } 1091fa9e4066Sahrens 1092fa9e4066Sahrens /* 1093fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 1094fa9e4066Sahrens */ 1095fa9e4066Sahrens void 1096fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 1097fa9e4066Sahrens { 1098fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 1099fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 1100fa9e4066Sahrens 1101e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1102fa9e4066Sahrens 1103fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 1104fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 110599653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 110699653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 1107ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 1108fa9e4066Sahrens 1109fa9e4066Sahrens vdev_remove_child(mvd, cvd); 1110fa9e4066Sahrens vdev_remove_child(pvd, mvd); 11118ad4d6ddSJeff Bonwick 111299653d4eSeschrock /* 1113e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1114e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 1115e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 1116e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 111799653d4eSeschrock */ 11188ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 11198ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 11201195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 11218ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 11228ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 11238ad4d6ddSJeff Bonwick } 1124e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 1125e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 1126fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1127fa9e4066Sahrens 1128fa9e4066Sahrens if (cvd == cvd->vdev_top) 1129fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 1130fa9e4066Sahrens 1131fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 1132fa9e4066Sahrens vdev_free(mvd); 1133fa9e4066Sahrens } 1134fa9e4066Sahrens 1135663207adSDon Brady static void 1136663207adSDon Brady vdev_metaslab_group_create(vdev_t *vd) 1137663207adSDon Brady { 1138663207adSDon Brady spa_t *spa = vd->vdev_spa; 1139663207adSDon Brady 1140663207adSDon Brady /* 1141663207adSDon Brady * metaslab_group_create was delayed until allocation bias was available 1142663207adSDon Brady */ 1143663207adSDon Brady if (vd->vdev_mg == NULL) { 1144663207adSDon Brady metaslab_class_t *mc; 1145663207adSDon Brady 1146663207adSDon Brady if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1147663207adSDon Brady vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1148663207adSDon Brady 1149663207adSDon Brady ASSERT3U(vd->vdev_islog, ==, 1150663207adSDon Brady (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1151663207adSDon Brady 1152663207adSDon Brady switch (vd->vdev_alloc_bias) { 1153663207adSDon Brady case VDEV_BIAS_LOG: 1154663207adSDon Brady mc = spa_log_class(spa); 1155663207adSDon Brady break; 1156663207adSDon Brady case VDEV_BIAS_SPECIAL: 1157663207adSDon Brady mc = spa_special_class(spa); 1158663207adSDon Brady break; 1159663207adSDon Brady case VDEV_BIAS_DEDUP: 1160663207adSDon Brady mc = spa_dedup_class(spa); 1161663207adSDon Brady break; 1162663207adSDon Brady default: 1163663207adSDon Brady mc = spa_normal_class(spa); 1164663207adSDon Brady } 1165663207adSDon Brady 1166663207adSDon Brady vd->vdev_mg = metaslab_group_create(mc, vd, 1167663207adSDon Brady spa->spa_alloc_count); 1168663207adSDon Brady 1169663207adSDon Brady /* 1170663207adSDon Brady * The spa ashift values currently only reflect the 1171663207adSDon Brady * general vdev classes. Class destination is late 1172663207adSDon Brady * binding so ashift checking had to wait until now 1173663207adSDon Brady */ 1174663207adSDon Brady if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1175663207adSDon Brady mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1176663207adSDon Brady if (vd->vdev_ashift > spa->spa_max_ashift) 1177663207adSDon Brady spa->spa_max_ashift = vd->vdev_ashift; 1178663207adSDon Brady if (vd->vdev_ashift < spa->spa_min_ashift) 1179663207adSDon Brady spa->spa_min_ashift = vd->vdev_ashift; 1180663207adSDon Brady } 1181663207adSDon Brady } 1182663207adSDon Brady } 1183663207adSDon Brady 1184ea8dc4b6Seschrock int 1185fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1186fa9e4066Sahrens { 1187fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1188ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1189ecc2d604Sbonwick uint64_t m; 1190fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 1191fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1192ecc2d604Sbonwick metaslab_t **mspp; 1193ecc2d604Sbonwick int error; 1194663207adSDon Brady boolean_t expanding = (oldc != 0); 1195fa9e4066Sahrens 1196a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1197a1521560SJeff Bonwick 119888ecc943SGeorge Wilson /* 119988ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 120088ecc943SGeorge Wilson */ 120188ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 12020e34b6a7Sbonwick return (0); 12030e34b6a7Sbonwick 120488ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 120588ecc943SGeorge Wilson 1206fa9e4066Sahrens ASSERT(oldc <= newc); 1207fa9e4066Sahrens 1208ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1209fa9e4066Sahrens 1210663207adSDon Brady if (expanding) { 1211ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1212ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1213ecc2d604Sbonwick } 1214fa9e4066Sahrens 1215ecc2d604Sbonwick vd->vdev_ms = mspp; 1216ecc2d604Sbonwick vd->vdev_ms_count = newc; 1217ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 12180713e232SGeorge Wilson uint64_t object = 0; 12190713e232SGeorge Wilson 12205cabbc6bSPrashanth Sreenivasa /* 12215cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 12225cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 12235cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 12245cabbc6bSPrashanth Sreenivasa */ 12255cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1226ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 12277bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 12287bfdf011SNeil Perrin DMU_READ_PREFETCH); 12293ee8c80cSPavel Zakharov if (error != 0) { 12303ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 12313ee8c80cSPavel Zakharov "array [error=%d]", error); 1232ecc2d604Sbonwick return (error); 12333ee8c80cSPavel Zakharov } 1234fa9e4066Sahrens } 12351e9bd7ecSPrakash Surya 1236663207adSDon Brady #ifndef _KERNEL 1237663207adSDon Brady /* 1238663207adSDon Brady * To accomodate zdb_leak_init() fake indirect 1239663207adSDon Brady * metaslabs, we allocate a metaslab group for 1240663207adSDon Brady * indirect vdevs which normally don't have one. 1241663207adSDon Brady */ 1242663207adSDon Brady if (vd->vdev_mg == NULL) { 1243663207adSDon Brady ASSERT0(vdev_is_concrete(vd)); 1244663207adSDon Brady vdev_metaslab_group_create(vd); 1245663207adSDon Brady } 1246663207adSDon Brady #endif 12471e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 12481e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 12493ee8c80cSPavel Zakharov if (error != 0) { 12503ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 12513ee8c80cSPavel Zakharov error); 12521e9bd7ecSPrakash Surya return (error); 12533ee8c80cSPavel Zakharov } 1254fa9e4066Sahrens } 1255fa9e4066Sahrens 1256a1521560SJeff Bonwick if (txg == 0) 1257a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1258a1521560SJeff Bonwick 12593f9d6ad7SLin Ling /* 12603f9d6ad7SLin Ling * If the vdev is being removed we don't activate 12613f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 12623f9d6ad7SLin Ling * allocations are performed on this device. 12633f9d6ad7SLin Ling */ 1264663207adSDon Brady if (!expanding && !vd->vdev_removing) { 1265a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1266663207adSDon Brady } 1267a1521560SJeff Bonwick 1268a1521560SJeff Bonwick if (txg == 0) 1269a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1270a1521560SJeff Bonwick 1271814dcd43SSerapheim Dimitropoulos /* 1272814dcd43SSerapheim Dimitropoulos * Regardless whether this vdev was just added or it is being 1273814dcd43SSerapheim Dimitropoulos * expanded, the metaslab count has changed. Recalculate the 1274814dcd43SSerapheim Dimitropoulos * block limit. 1275814dcd43SSerapheim Dimitropoulos */ 1276814dcd43SSerapheim Dimitropoulos spa_log_sm_set_blocklimit(spa); 1277814dcd43SSerapheim Dimitropoulos 1278ea8dc4b6Seschrock return (0); 1279fa9e4066Sahrens } 1280fa9e4066Sahrens 1281fa9e4066Sahrens void 1282fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1283fa9e4066Sahrens { 128486714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 128586714001SSerapheim Dimitropoulos ASSERT(spa_feature_is_active(vd->vdev_spa, 128686714001SSerapheim Dimitropoulos SPA_FEATURE_POOL_CHECKPOINT)); 128786714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 128886714001SSerapheim Dimitropoulos /* 128986714001SSerapheim Dimitropoulos * Even though we close the space map, we need to set its 129086714001SSerapheim Dimitropoulos * pointer to NULL. The reason is that vdev_metaslab_fini() 129186714001SSerapheim Dimitropoulos * may be called multiple times for certain operations 129286714001SSerapheim Dimitropoulos * (i.e. when destroying a pool) so we need to ensure that 129386714001SSerapheim Dimitropoulos * this clause never executes twice. This logic is similar 129486714001SSerapheim Dimitropoulos * to the one used for the vdev_ms clause below. 129586714001SSerapheim Dimitropoulos */ 129686714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 129786714001SSerapheim Dimitropoulos } 129886714001SSerapheim Dimitropoulos 1299fa9e4066Sahrens if (vd->vdev_ms != NULL) { 1300555d674dSSerapheim Dimitropoulos metaslab_group_t *mg = vd->vdev_mg; 1301555d674dSSerapheim Dimitropoulos metaslab_group_passivate(mg); 13025cabbc6bSPrashanth Sreenivasa 1303555d674dSSerapheim Dimitropoulos uint64_t count = vd->vdev_ms_count; 13045cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 13050713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 13060713e232SGeorge Wilson if (msp != NULL) 13070713e232SGeorge Wilson metaslab_fini(msp); 13080713e232SGeorge Wilson } 1309fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1310fa9e4066Sahrens vd->vdev_ms = NULL; 13115cabbc6bSPrashanth Sreenivasa 13125cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1313555d674dSSerapheim Dimitropoulos 1314555d674dSSerapheim Dimitropoulos for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 1315555d674dSSerapheim Dimitropoulos ASSERT0(mg->mg_histogram[i]); 1316fa9e4066Sahrens } 13175cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1318fa9e4066Sahrens } 1319fa9e4066Sahrens 1320e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1321e14bb325SJeff Bonwick boolean_t vps_readable; 1322e14bb325SJeff Bonwick boolean_t vps_writeable; 1323e14bb325SJeff Bonwick int vps_flags; 1324e14bb325SJeff Bonwick } vdev_probe_stats_t; 1325e14bb325SJeff Bonwick 1326e14bb325SJeff Bonwick static void 1327e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 13280a4e9518Sgw { 13298ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1330a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1331e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1332a3f829aeSBill Moore 1333a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1334e14bb325SJeff Bonwick 1335e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1336e14bb325SJeff Bonwick if (zio->io_error == 0) 1337e14bb325SJeff Bonwick vps->vps_readable = 1; 13388ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1339a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1340770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1341e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1342e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1343e14bb325SJeff Bonwick } else { 1344770499e1SDan Kimmel abd_free(zio->io_abd); 1345e14bb325SJeff Bonwick } 1346e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1347e14bb325SJeff Bonwick if (zio->io_error == 0) 1348e14bb325SJeff Bonwick vps->vps_writeable = 1; 1349770499e1SDan Kimmel abd_free(zio->io_abd); 1350e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1351a3f829aeSBill Moore zio_t *pio; 1352e14bb325SJeff Bonwick 1353e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1354e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1355e14bb325SJeff Bonwick 1356e14bb325SJeff Bonwick if (vdev_readable(vd) && 13578ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1358e14bb325SJeff Bonwick zio->io_error = 0; 1359e14bb325SJeff Bonwick } else { 1360e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 13613ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1362e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1363eb633035STom Caputi spa, vd, NULL, NULL, 0, 0); 1364be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1365e14bb325SJeff Bonwick } 1366a3f829aeSBill Moore 1367a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1368a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1369a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1370a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1371a3f829aeSBill Moore 13720f7643c7SGeorge Wilson zio_link_t *zl = NULL; 13730f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1374a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1375be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1376a3f829aeSBill Moore 1377e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1378e14bb325SJeff Bonwick } 1379e14bb325SJeff Bonwick } 13800a4e9518Sgw 1381e14bb325SJeff Bonwick /* 1382f7170741SWill Andrews * Determine whether this device is accessible. 1383f7170741SWill Andrews * 1384f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1385f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1386f7170741SWill Andrews * a VTOC. 1387e14bb325SJeff Bonwick */ 1388e14bb325SJeff Bonwick zio_t * 1389a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1390e14bb325SJeff Bonwick { 1391e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1392a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1393a3f829aeSBill Moore zio_t *pio; 1394a3f829aeSBill Moore 1395a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 13960a4e9518Sgw 1397a3f829aeSBill Moore /* 1398a3f829aeSBill Moore * Don't probe the probe. 1399a3f829aeSBill Moore */ 1400a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1401a3f829aeSBill Moore return (NULL); 1402e14bb325SJeff Bonwick 1403a3f829aeSBill Moore /* 1404a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1405a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1406a3f829aeSBill Moore * this vdev will become parents of the probe io. 1407a3f829aeSBill Moore */ 1408a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1409e14bb325SJeff Bonwick 1410a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1411a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1412a3f829aeSBill Moore 1413a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1414a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 14158956713aSEric Schrock ZIO_FLAG_TRYHARD; 1416a3f829aeSBill Moore 1417a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1418a3f829aeSBill Moore /* 1419a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1420a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1421a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1422a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1423a3f829aeSBill Moore * any zio looking at these values can assume that 1424a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1425a3f829aeSBill Moore * important because when a device has intermittent 1426a3f829aeSBill Moore * connectivity problems, we want to ensure that 1427a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1428a3f829aeSBill Moore * the zio (EIO). 1429a3f829aeSBill Moore * 1430a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1431a3f829aeSBill Moore * values so the probe can reevaluate from first 1432a3f829aeSBill Moore * principles. 1433a3f829aeSBill Moore */ 1434a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1435a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1436a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1437a3f829aeSBill Moore } 1438a3f829aeSBill Moore 1439a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1440a3f829aeSBill Moore vdev_probe_done, vps, 1441a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1442a3f829aeSBill Moore 144398d1cbfeSGeorge Wilson /* 144498d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 144598d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 144698d1cbfeSGeorge Wilson */ 1447a3f829aeSBill Moore if (zio != NULL) { 1448a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1449a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1450a3f829aeSBill Moore } 1451e14bb325SJeff Bonwick } 1452e14bb325SJeff Bonwick 1453a3f829aeSBill Moore if (zio != NULL) 1454a3f829aeSBill Moore zio_add_child(zio, pio); 1455e14bb325SJeff Bonwick 1456a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1457e14bb325SJeff Bonwick 1458a3f829aeSBill Moore if (vps == NULL) { 1459a3f829aeSBill Moore ASSERT(zio != NULL); 1460a3f829aeSBill Moore return (NULL); 1461a3f829aeSBill Moore } 1462e14bb325SJeff Bonwick 1463e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1464a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1465e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1466770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1467770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1468e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1469e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1470e14bb325SJeff Bonwick } 1471e14bb325SJeff Bonwick 1472a3f829aeSBill Moore if (zio == NULL) 1473a3f829aeSBill Moore return (pio); 1474a3f829aeSBill Moore 1475a3f829aeSBill Moore zio_nowait(pio); 1476a3f829aeSBill Moore return (NULL); 14770a4e9518Sgw } 14780a4e9518Sgw 1479f64c0e34SEric Taylor static void 1480f64c0e34SEric Taylor vdev_open_child(void *arg) 1481f64c0e34SEric Taylor { 1482f64c0e34SEric Taylor vdev_t *vd = arg; 1483f64c0e34SEric Taylor 1484f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1485f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1486f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1487f64c0e34SEric Taylor } 1488f64c0e34SEric Taylor 1489681d9761SEric Taylor boolean_t 1490681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1491681d9761SEric Taylor { 1492681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1493681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1494681d9761SEric Taylor return (B_TRUE); 1495681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1496681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1497681d9761SEric Taylor return (B_TRUE); 1498681d9761SEric Taylor return (B_FALSE); 1499681d9761SEric Taylor } 1500681d9761SEric Taylor 1501f64c0e34SEric Taylor void 1502f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1503f64c0e34SEric Taylor { 1504f64c0e34SEric Taylor taskq_t *tq; 1505f64c0e34SEric Taylor int children = vd->vdev_children; 1506f64c0e34SEric Taylor 1507681d9761SEric Taylor /* 1508681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1509681d9761SEric Taylor * in a single thread so that the same thread holds the 1510681d9761SEric Taylor * spa_namespace_lock 1511681d9761SEric Taylor */ 1512681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 151312a8814cSTom Caputi retry_sync: 1514681d9761SEric Taylor for (int c = 0; c < children; c++) 1515681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1516681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 151712a8814cSTom Caputi } else { 151812a8814cSTom Caputi tq = taskq_create("vdev_open", children, minclsyspri, 151912a8814cSTom Caputi children, children, TASKQ_PREPOPULATE); 152012a8814cSTom Caputi if (tq == NULL) 152112a8814cSTom Caputi goto retry_sync; 152212a8814cSTom Caputi 152312a8814cSTom Caputi for (int c = 0; c < children; c++) 152412a8814cSTom Caputi VERIFY(taskq_dispatch(tq, vdev_open_child, 152512a8814cSTom Caputi vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); 152612a8814cSTom Caputi 152712a8814cSTom Caputi taskq_destroy(tq); 1528681d9761SEric Taylor } 1529f64c0e34SEric Taylor 153012a8814cSTom Caputi vd->vdev_nonrot = B_TRUE; 1531f64c0e34SEric Taylor 153212a8814cSTom Caputi for (int c = 0; c < children; c++) 153312a8814cSTom Caputi vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; 1534f64c0e34SEric Taylor } 1535f64c0e34SEric Taylor 15365cabbc6bSPrashanth Sreenivasa /* 15375cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 15385cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 15395cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 15405cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 15415cabbc6bSPrashanth Sreenivasa */ 15425cabbc6bSPrashanth Sreenivasa static void 15435cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 15445cabbc6bSPrashanth Sreenivasa { 15455cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 15465cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 15475cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 15485cabbc6bSPrashanth Sreenivasa } 15495cabbc6bSPrashanth Sreenivasa } 15505cabbc6bSPrashanth Sreenivasa 1551fa9e4066Sahrens /* 1552fa9e4066Sahrens * Prepare a virtual device for access. 1553fa9e4066Sahrens */ 1554fa9e4066Sahrens int 1555fa9e4066Sahrens vdev_open(vdev_t *vd) 1556fa9e4066Sahrens { 15578ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1558fa9e4066Sahrens int error; 1559fa9e4066Sahrens uint64_t osize = 0; 15604263d13fSGeorge Wilson uint64_t max_osize = 0; 15614263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1562ecc2d604Sbonwick uint64_t ashift = 0; 1563fa9e4066Sahrens 1564f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1565f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1566fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1567fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1568fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1569fa9e4066Sahrens 1570fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1571e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1572e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1573573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1574fa9e4066Sahrens 1575069f55e2SEric Schrock /* 1576069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1577069f55e2SEric Schrock * faulted, bail out of the open. 1578069f55e2SEric Schrock */ 15793d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 15803d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1581069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1582069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 15833d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1584069f55e2SEric Schrock vd->vdev_label_aux); 1585be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 15863d7072f8Seschrock } else if (vd->vdev_offline) { 1587fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1588ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1589be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1590fa9e4066Sahrens } 1591fa9e4066Sahrens 15924263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1593fa9e4066Sahrens 1594095bcd66SGeorge Wilson /* 1595095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1596095bcd66SGeorge Wilson * the vdev on error. 1597095bcd66SGeorge Wilson */ 1598095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1599ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 16008956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1601ea8dc4b6Seschrock 1602fa9e4066Sahrens if (error) { 16033d7072f8Seschrock if (vd->vdev_removed && 16043d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 16053d7072f8Seschrock vd->vdev_removed = B_FALSE; 16063d7072f8Seschrock 16076f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 16086f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 16096f793812SPavel Zakharov vd->vdev_stat.vs_aux); 16106f793812SPavel Zakharov } else { 16116f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 16126f793812SPavel Zakharov vd->vdev_stat.vs_aux); 16136f793812SPavel Zakharov } 1614fa9e4066Sahrens return (error); 1615fa9e4066Sahrens } 1616fa9e4066Sahrens 16173d7072f8Seschrock vd->vdev_removed = B_FALSE; 16183d7072f8Seschrock 1619096d22d4SEric Schrock /* 1620096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1621096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1622096d22d4SEric Schrock */ 1623096d22d4SEric Schrock if (vd->vdev_faulted) { 1624096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1625096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1626096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1627096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1628096d22d4SEric Schrock vd->vdev_label_aux); 1629be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1630096d22d4SEric Schrock } 1631096d22d4SEric Schrock 16323d7072f8Seschrock if (vd->vdev_degraded) { 16333d7072f8Seschrock ASSERT(vd->vdev_children == 0); 16343d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 16353d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 16363d7072f8Seschrock } else { 1637069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 16383d7072f8Seschrock } 1639fa9e4066Sahrens 164088ecc943SGeorge Wilson /* 164188ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 164288ecc943SGeorge Wilson */ 164388ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 164488ecc943SGeorge Wilson return (0); 164588ecc943SGeorge Wilson 1646573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1647ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1648ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1649ea8dc4b6Seschrock VDEV_AUX_NONE); 1650ea8dc4b6Seschrock break; 1651ea8dc4b6Seschrock } 1652573ca77eSGeorge Wilson } 1653fa9e4066Sahrens 1654fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 16554263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1656fa9e4066Sahrens 1657fa9e4066Sahrens if (vd->vdev_children == 0) { 1658fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 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 = osize; 1664fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 16654263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 16664263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1667fa9e4066Sahrens } else { 1668ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1669fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1670ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1671ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1672be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1673fa9e4066Sahrens } 1674fa9e4066Sahrens psize = 0; 1675fa9e4066Sahrens asize = osize; 16764263d13fSGeorge Wilson max_asize = max_osize; 1677fa9e4066Sahrens } 1678fa9e4066Sahrens 1679fa9e4066Sahrens vd->vdev_psize = psize; 1680fa9e4066Sahrens 1681573ca77eSGeorge Wilson /* 1682c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1683573ca77eSGeorge Wilson */ 1684573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1685573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1686573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1687be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1688573ca77eSGeorge Wilson } 1689573ca77eSGeorge Wilson 1690fa9e4066Sahrens if (vd->vdev_asize == 0) { 1691fa9e4066Sahrens /* 1692fa9e4066Sahrens * This is the first-ever open, so use the computed values. 16935711d393Sloli * For compatibility, a different ashift can be requested. 1694fa9e4066Sahrens */ 1695fa9e4066Sahrens vd->vdev_asize = asize; 16964263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 16975711d393Sloli if (vd->vdev_ashift == 0) { 16985711d393Sloli vd->vdev_ashift = ashift; /* use detected value */ 16995711d393Sloli } 17005711d393Sloli if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 17015711d393Sloli vd->vdev_ashift > ASHIFT_MAX)) { 17025711d393Sloli vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 17035711d393Sloli VDEV_AUX_BAD_ASHIFT); 17045711d393Sloli return (SET_ERROR(EDOM)); 17055711d393Sloli } 1706fa9e4066Sahrens } else { 1707fa9e4066Sahrens /* 17082384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 17092384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 17105711d393Sloli * post an event instead. 1711fa9e4066Sahrens */ 17122384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 17132384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 17145711d393Sloli zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 17155711d393Sloli spa, vd, NULL, NULL, 0, 0); 1716fa9e4066Sahrens } 17175711d393Sloli 17184263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1719573ca77eSGeorge Wilson } 1720fa9e4066Sahrens 1721573ca77eSGeorge Wilson /* 1722c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1723c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1724c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1725c040c10cSSteven Hartland * making the additional space available. 1726c040c10cSSteven Hartland * 1727c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1728c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1729c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1730c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1731c040c10cSSteven Hartland * vdev_min_asize. 1732573ca77eSGeorge Wilson */ 1733c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1734c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1735c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1736c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1737573ca77eSGeorge Wilson vd->vdev_asize = asize; 1738fa9e4066Sahrens 1739573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1740fa9e4066Sahrens 17410a4e9518Sgw /* 17420a4e9518Sgw * Ensure we can issue some IO before declaring the 17430a4e9518Sgw * vdev open for business. 17440a4e9518Sgw */ 1745e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1746e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 174798d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 174898d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 17490a4e9518Sgw return (error); 17500a4e9518Sgw } 17510a4e9518Sgw 175281cd5c55SMatthew Ahrens /* 175381cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 1754663207adSDon Brady * 1755663207adSDon Brady * DJB - TBD these should perhaps be tracked per allocation class 1756663207adSDon Brady * (e.g. spa_min_ashift is used to round up post compression buffers) 175781cd5c55SMatthew Ahrens */ 175881cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1759663207adSDon Brady vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1760663207adSDon Brady vd->vdev_aux == NULL) { 176181cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 176281cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 176381cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 176481cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 176581cd5c55SMatthew Ahrens } 176681cd5c55SMatthew Ahrens 1767088f3894Sahrens /* 1768*0c06d385Sjwpoduska * If this is a leaf vdev, assess whether a resilver is needed. 1769*0c06d385Sjwpoduska * But don't do this if we are doing a reopen for a scrub, since 1770*0c06d385Sjwpoduska * this would just restart the scrub we are already doing. 1771088f3894Sahrens */ 1772*0c06d385Sjwpoduska if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) 1773*0c06d385Sjwpoduska dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); 1774088f3894Sahrens 1775fa9e4066Sahrens return (0); 1776fa9e4066Sahrens } 1777fa9e4066Sahrens 1778560e6e96Seschrock /* 1779560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 17806f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 17813d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1782560e6e96Seschrock * 1783560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1784560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1785560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1786560e6e96Seschrock * will be updated but the function will return 0. 1787560e6e96Seschrock */ 1788560e6e96Seschrock int 17896f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1790560e6e96Seschrock { 1791560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1792560e6e96Seschrock nvlist_t *label; 17936f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1794560e6e96Seschrock uint64_t state; 17956f793812SPavel Zakharov nvlist_t *nvl; 17966f793812SPavel Zakharov uint64_t txg; 1797560e6e96Seschrock 17986f793812SPavel Zakharov if (vdev_validate_skip) 17996f793812SPavel Zakharov return (0); 18006f793812SPavel Zakharov 18016f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 18026f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1803be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1804560e6e96Seschrock 1805b5989ec7Seschrock /* 1806b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1807b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1808b5989ec7Seschrock * overwrite the previous state. 1809b5989ec7Seschrock */ 18106f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 18116f793812SPavel Zakharov return (0); 1812560e6e96Seschrock 18136f793812SPavel Zakharov /* 18146f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 18156f793812SPavel Zakharov * was modified at a point after the current txg. 1816d1de72cfSPavel Zakharov * If config lock is not held do not check for the txg. spa_sync could 1817d1de72cfSPavel Zakharov * be updating the vdev's label before updating spa_last_synced_txg. 18186f793812SPavel Zakharov */ 1819d1de72cfSPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1820d1de72cfSPavel Zakharov spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 18216f793812SPavel Zakharov txg = UINT64_MAX; 18226f793812SPavel Zakharov else 18236f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1824560e6e96Seschrock 18256f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 18266f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 18276f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 1828b6bf6e15SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1829b6bf6e15SPavel Zakharov "txg %llu", (u_longlong_t)txg); 18306f793812SPavel Zakharov return (0); 18316f793812SPavel Zakharov } 18321195e687SMark J Musante 18336f793812SPavel Zakharov /* 18346f793812SPavel Zakharov * Determine if this vdev has been split off into another 18356f793812SPavel Zakharov * pool. If so, then refuse to open it. 18366f793812SPavel Zakharov */ 18376f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 18386f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 18396f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18406f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 18416f793812SPavel Zakharov nvlist_free(label); 18426f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 18436f793812SPavel Zakharov return (0); 18446f793812SPavel Zakharov } 1845560e6e96Seschrock 18466f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 18476f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18486f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18496f793812SPavel Zakharov nvlist_free(label); 18506f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18516f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 18526f793812SPavel Zakharov return (0); 18536f793812SPavel Zakharov } 18541195e687SMark J Musante 18556f793812SPavel Zakharov /* 18566f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 18576f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 18586f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 18596f793812SPavel Zakharov * cached config. The check will be performed again once we have the 18606f793812SPavel Zakharov * trusted config from the MOS. 18616f793812SPavel Zakharov */ 18626f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 18636f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18646f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18656f793812SPavel Zakharov nvlist_free(label); 18666f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 18676f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 18686f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 18696f793812SPavel Zakharov return (0); 18706f793812SPavel Zakharov } 18716f793812SPavel Zakharov 18726f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 18736f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 18746f793812SPavel Zakharov &aux_guid) != 0) 18756f793812SPavel Zakharov aux_guid = 0; 18766f793812SPavel Zakharov 18776f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 18786f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18796f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18806f793812SPavel Zakharov nvlist_free(label); 18816f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18826f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 18836f793812SPavel Zakharov return (0); 18846f793812SPavel Zakharov } 18856f793812SPavel Zakharov 18866f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 18876f793812SPavel Zakharov != 0) { 18886f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18896f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18906f793812SPavel Zakharov nvlist_free(label); 18916f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18926f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 18936f793812SPavel Zakharov return (0); 18946f793812SPavel Zakharov } 18956f793812SPavel Zakharov 18966f793812SPavel Zakharov /* 18976f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 18986f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 18996f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 19006f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 19016f793812SPavel Zakharov * vdev, we can safely compare to that instead. 19026f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 19036f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 19046f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 19056f793812SPavel Zakharov * extreme rewind. 19066f793812SPavel Zakharov * 19076f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 19086f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 19096f793812SPavel Zakharov * corrupt if it is partway through a split operation. 19106f793812SPavel Zakharov */ 19116f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 19126f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 19136f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 19146f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 19156f793812SPavel Zakharov mismatch = B_TRUE; 19166f793812SPavel Zakharov } else { 19176f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 19186f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 19196f793812SPavel Zakharov mismatch = B_TRUE; 1920560e6e96Seschrock } 1921560e6e96Seschrock 19226f793812SPavel Zakharov if (mismatch) { 1923560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1924560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1925560e6e96Seschrock nvlist_free(label); 19266f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 19276f793812SPavel Zakharov "doesn't match label guid"); 19286f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 19296f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 19306f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 19316f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 19326f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 19336f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1934560e6e96Seschrock return (0); 1935560e6e96Seschrock } 19366f793812SPavel Zakharov } 1937560e6e96Seschrock 19386f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 19396f793812SPavel Zakharov &state) != 0) { 19406f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 19416f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1942560e6e96Seschrock nvlist_free(label); 19436f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 19446f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 19456f793812SPavel Zakharov return (0); 19466f793812SPavel Zakharov } 1947560e6e96Seschrock 19486f793812SPavel Zakharov nvlist_free(label); 19496f793812SPavel Zakharov 19506f793812SPavel Zakharov /* 19516f793812SPavel Zakharov * If this is a verbatim import, no need to check the 19526f793812SPavel Zakharov * state of the pool. 19536f793812SPavel Zakharov */ 19546f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 19556f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 19566f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 19576f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 19586f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 19596f793812SPavel Zakharov return (SET_ERROR(EBADF)); 19606f793812SPavel Zakharov } 19616f793812SPavel Zakharov 19626f793812SPavel Zakharov /* 19636f793812SPavel Zakharov * If we were able to open and validate a vdev that was 19646f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 19656f793812SPavel Zakharov * now. 19666f793812SPavel Zakharov */ 19676f793812SPavel Zakharov if (vd->vdev_not_present) 19686f793812SPavel Zakharov vd->vdev_not_present = 0; 19696f793812SPavel Zakharov 19706f793812SPavel Zakharov return (0); 19716f793812SPavel Zakharov } 19726f793812SPavel Zakharov 19736f793812SPavel Zakharov static void 19746f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 19756f793812SPavel Zakharov { 19766f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 19776f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 19786f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 19796f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 19806f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 19816f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 19826f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19833ee8c80cSPavel Zakharov } 19846f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 19856f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19866f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 19876f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 19886f793812SPavel Zakharov } 19896f793812SPavel Zakharov } 1990560e6e96Seschrock 19916f793812SPavel Zakharov /* 19926f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 19936f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 19946f793812SPavel Zakharov * paths from userland config into MOS config. 19956f793812SPavel Zakharov */ 19966f793812SPavel Zakharov int 19976f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 19986f793812SPavel Zakharov { 19996f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 20006f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 20016f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 20026f793812SPavel Zakharov return (0); 20036f793812SPavel Zakharov 20046f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 20056f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 20066f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 20076f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20086f793812SPavel Zakharov } 20096f793812SPavel Zakharov 20106f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 20116f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 20126f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 20136f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 20146f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20156f793812SPavel Zakharov } 20166f793812SPavel Zakharov 20176f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 20186f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 20196f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 20206f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 20216f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 202251ece835Seschrock } 2023560e6e96Seschrock 20246f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 20256f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 20266f793812SPavel Zakharov dvd->vdev_child[i]); 20276f793812SPavel Zakharov if (error != 0) 20286f793812SPavel Zakharov return (error); 20296f793812SPavel Zakharov } 20306f793812SPavel Zakharov 20316f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 20326f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 20336f793812SPavel Zakharov 2034560e6e96Seschrock return (0); 2035560e6e96Seschrock } 2036560e6e96Seschrock 20376f793812SPavel Zakharov static void 20386f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 20396f793812SPavel Zakharov { 20406f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 20416f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 20426f793812SPavel Zakharov 20436f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 20446f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 20456f793812SPavel Zakharov } 20466f793812SPavel Zakharov 20476f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 20486f793812SPavel Zakharov return; 20496f793812SPavel Zakharov 20506f793812SPavel Zakharov /* 20516f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 20526f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 20536f793812SPavel Zakharov * step outside of it. 20546f793812SPavel Zakharov */ 20556f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 20566f793812SPavel Zakharov 20576f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 20586f793812SPavel Zakharov return; 20596f793812SPavel Zakharov 20606f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 20616f793812SPavel Zakharov 20626f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 20636f793812SPavel Zakharov } 20646f793812SPavel Zakharov 20656f793812SPavel Zakharov /* 20666f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 20676f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 20686f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 20696f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 20706f793812SPavel Zakharov */ 20716f793812SPavel Zakharov void 20726f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 20736f793812SPavel Zakharov { 20746f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 20756f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 20766f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 20776f793812SPavel Zakharov 20786f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 20796f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 20806f793812SPavel Zakharov drvd->vdev_child[i]); 20816f793812SPavel Zakharov } 20826f793812SPavel Zakharov } 20836f793812SPavel Zakharov 2084fa9e4066Sahrens /* 2085fa9e4066Sahrens * Close a virtual device. 2086fa9e4066Sahrens */ 2087fa9e4066Sahrens void 2088fa9e4066Sahrens vdev_close(vdev_t *vd) 2089fa9e4066Sahrens { 20908ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 2091095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 20928ad4d6ddSJeff Bonwick 20938ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 20948ad4d6ddSJeff Bonwick 20951195e687SMark J Musante /* 20961195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 20971195e687SMark J Musante * going offline. 20981195e687SMark J Musante */ 2099095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 21001195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2101095bcd66SGeorge Wilson 2102fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 2103fa9e4066Sahrens 21043d7072f8Seschrock vdev_cache_purge(vd); 2105fa9e4066Sahrens 2106560e6e96Seschrock /* 2107573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 2108560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 2109560e6e96Seschrock * it's still faulted. 2110560e6e96Seschrock */ 2111560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 2112560e6e96Seschrock 2113fa9e4066Sahrens if (vd->vdev_offline) 2114fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 2115fa9e4066Sahrens else 2116fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 2117ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2118fa9e4066Sahrens } 2119fa9e4066Sahrens 2120dcba9f3fSGeorge Wilson void 2121dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 2122dcba9f3fSGeorge Wilson { 2123dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2124dcba9f3fSGeorge Wilson 2125dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2126dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2127dcba9f3fSGeorge Wilson return; 2128dcba9f3fSGeorge Wilson 2129dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2130dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 2131dcba9f3fSGeorge Wilson 2132dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2133dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 2134dcba9f3fSGeorge Wilson } 2135dcba9f3fSGeorge Wilson 2136dcba9f3fSGeorge Wilson void 2137dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 2138dcba9f3fSGeorge Wilson { 2139dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2140dcba9f3fSGeorge Wilson 2141dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2142dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2143dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 2144dcba9f3fSGeorge Wilson 2145dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2146dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 2147dcba9f3fSGeorge Wilson } 2148dcba9f3fSGeorge Wilson 2149095bcd66SGeorge Wilson /* 2150095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 2151095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 2152095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2153095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 2154095bcd66SGeorge Wilson */ 2155fa9e4066Sahrens void 2156ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 2157fa9e4066Sahrens { 2158ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 2159fa9e4066Sahrens 2160e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2161ea8dc4b6Seschrock 21621195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 21631195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 2164fa9e4066Sahrens vdev_close(vd); 2165fa9e4066Sahrens (void) vdev_open(vd); 2166fa9e4066Sahrens 216739c23413Seschrock /* 216839c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 216939c23413Seschrock * Otherwise, a device with an invalid label could be successfully 217039c23413Seschrock * opened in response to vdev_reopen(). 217139c23413Seschrock */ 2172c5904d13Seschrock if (vd->vdev_aux) { 2173c5904d13Seschrock (void) vdev_validate_aux(vd); 2174e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 21756809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 2176573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 2177573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 2178c5904d13Seschrock } else { 21796f793812SPavel Zakharov (void) vdev_validate(vd); 2180c5904d13Seschrock } 218139c23413Seschrock 2182fa9e4066Sahrens /* 21833d7072f8Seschrock * Reassess parent vdev's health. 2184fa9e4066Sahrens */ 21853d7072f8Seschrock vdev_propagate_state(vd); 2186fa9e4066Sahrens } 2187fa9e4066Sahrens 2188fa9e4066Sahrens int 218999653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2190fa9e4066Sahrens { 2191fa9e4066Sahrens int error; 2192fa9e4066Sahrens 2193fa9e4066Sahrens /* 2194fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 2195fa9e4066Sahrens * For a create, however, we want to fail the request if 2196fa9e4066Sahrens * there are any components we can't open. 2197fa9e4066Sahrens */ 2198fa9e4066Sahrens error = vdev_open(vd); 2199fa9e4066Sahrens 2200fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2201fa9e4066Sahrens vdev_close(vd); 2202fa9e4066Sahrens return (error ? error : ENXIO); 2203fa9e4066Sahrens } 2204fa9e4066Sahrens 2205fa9e4066Sahrens /* 22060713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 2207fa9e4066Sahrens */ 22080713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 22090713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 221039c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2211fa9e4066Sahrens vdev_close(vd); 2212fa9e4066Sahrens return (error); 2213fa9e4066Sahrens } 2214fa9e4066Sahrens 2215fa9e4066Sahrens return (0); 2216fa9e4066Sahrens } 2217fa9e4066Sahrens 22180e34b6a7Sbonwick void 2219573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 2220fa9e4066Sahrens { 222186714001SSerapheim Dimitropoulos uint64_t asize = vd->vdev_asize; 2222a0b03b16SSerapheim Dimitropoulos uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2223b4bf0cf0SDon Brady uint64_t ms_shift; 222486714001SSerapheim Dimitropoulos 2225a3874b8bSToomas Soome /* BEGIN CSTYLED */ 2226fa9e4066Sahrens /* 2227b4bf0cf0SDon Brady * There are two dimensions to the metaslab sizing calculation: 2228b4bf0cf0SDon Brady * the size of the metaslab and the count of metaslabs per vdev. 2229b4bf0cf0SDon Brady * 2230a0b03b16SSerapheim Dimitropoulos * The default values used below are a good balance between memory 2231a0b03b16SSerapheim Dimitropoulos * usage (larger metaslab size means more memory needed for loaded 2232a0b03b16SSerapheim Dimitropoulos * metaslabs; more metaslabs means more memory needed for the 2233a0b03b16SSerapheim Dimitropoulos * metaslab_t structs), metaslab load time (larger metaslabs take 2234a0b03b16SSerapheim Dimitropoulos * longer to load), and metaslab sync time (more metaslabs means 2235a0b03b16SSerapheim Dimitropoulos * more time spent syncing all of them). 2236a0b03b16SSerapheim Dimitropoulos * 2237a0b03b16SSerapheim Dimitropoulos * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2238a0b03b16SSerapheim Dimitropoulos * The range of the dimensions are as follows: 2239a0b03b16SSerapheim Dimitropoulos * 2240a0b03b16SSerapheim Dimitropoulos * 2^29 <= ms_size <= 2^34 2241b4bf0cf0SDon Brady * 16 <= ms_count <= 131,072 2242b4bf0cf0SDon Brady * 2243b4bf0cf0SDon Brady * On the lower end of vdev sizes, we aim for metaslabs sizes of 2244b4bf0cf0SDon Brady * at least 512MB (2^29) to minimize fragmentation effects when 2245b4bf0cf0SDon Brady * testing with smaller devices. However, the count constraint 2246b4bf0cf0SDon Brady * of at least 16 metaslabs will override this minimum size goal. 2247b4bf0cf0SDon Brady * 2248b4bf0cf0SDon Brady * On the upper end of vdev sizes, we aim for a maximum metaslab 2249a0b03b16SSerapheim Dimitropoulos * size of 16GB. However, we will cap the total count to 2^17 2250a0b03b16SSerapheim Dimitropoulos * metaslabs to keep our memory footprint in check and let the 2251a0b03b16SSerapheim Dimitropoulos * metaslab size grow from there if that limit is hit. 2252b4bf0cf0SDon Brady * 2253b4bf0cf0SDon Brady * The net effect of applying above constrains is summarized below. 2254b4bf0cf0SDon Brady * 2255663207adSDon Brady * vdev size metaslab count 2256a0b03b16SSerapheim Dimitropoulos * --------------|----------------- 2257663207adSDon Brady * < 8GB ~16 2258663207adSDon Brady * 8GB - 100GB one per 512MB 2259663207adSDon Brady * 100GB - 3TB ~200 2260663207adSDon Brady * 3TB - 2PB one per 16GB 2261663207adSDon Brady * > 2PB ~131,072 2262a0b03b16SSerapheim Dimitropoulos * -------------------------------- 2263a0b03b16SSerapheim Dimitropoulos * 2264a0b03b16SSerapheim Dimitropoulos * Finally, note that all of the above calculate the initial 2265a0b03b16SSerapheim Dimitropoulos * number of metaslabs. Expanding a top-level vdev will result 2266a0b03b16SSerapheim Dimitropoulos * in additional metaslabs being allocated making it possible 2267a0b03b16SSerapheim Dimitropoulos * to exceed the zfs_vdev_ms_count_limit. 2268a0b03b16SSerapheim Dimitropoulos */ 2269a3874b8bSToomas Soome /* END CSTYLED */ 2270a0b03b16SSerapheim Dimitropoulos 2271a0b03b16SSerapheim Dimitropoulos if (ms_count < zfs_vdev_min_ms_count) 2272a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2273a0b03b16SSerapheim Dimitropoulos else if (ms_count > zfs_vdev_default_ms_count) 2274a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2275b4bf0cf0SDon Brady else 2276a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_default_ms_shift; 2277b4bf0cf0SDon Brady 2278b4bf0cf0SDon Brady if (ms_shift < SPA_MAXBLOCKSHIFT) { 2279b4bf0cf0SDon Brady ms_shift = SPA_MAXBLOCKSHIFT; 2280a0b03b16SSerapheim Dimitropoulos } else if (ms_shift > zfs_vdev_max_ms_shift) { 2281a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_max_ms_shift; 2282b4bf0cf0SDon Brady /* cap the total count to constrain memory footprint */ 2283a0b03b16SSerapheim Dimitropoulos if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2284a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 228586714001SSerapheim Dimitropoulos } 228686714001SSerapheim Dimitropoulos 228786714001SSerapheim Dimitropoulos vd->vdev_ms_shift = ms_shift; 228886714001SSerapheim Dimitropoulos ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2289fa9e4066Sahrens } 2290fa9e4066Sahrens 2291fa9e4066Sahrens void 2292ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2293fa9e4066Sahrens { 2294ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 22955cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 22965cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 2297ecc2d604Sbonwick ASSERT(ISP2(flags)); 2298f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 2299fa9e4066Sahrens 2300ecc2d604Sbonwick if (flags & VDD_METASLAB) 2301ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2302ecc2d604Sbonwick 2303ecc2d604Sbonwick if (flags & VDD_DTL) 2304ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2305ecc2d604Sbonwick 2306ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2307fa9e4066Sahrens } 2308fa9e4066Sahrens 23090713e232SGeorge Wilson void 23100713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 23110713e232SGeorge Wilson { 23120713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 23130713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 23140713e232SGeorge Wilson 23150713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 23160713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 23170713e232SGeorge Wilson } 23180713e232SGeorge Wilson 23198ad4d6ddSJeff Bonwick /* 23208ad4d6ddSJeff Bonwick * DTLs. 23218ad4d6ddSJeff Bonwick * 23228ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 23239fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 23248ad4d6ddSJeff Bonwick * 23258ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 23268ad4d6ddSJeff Bonwick * 23278ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 23288ad4d6ddSJeff Bonwick * 23298ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 23308ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 23318ad4d6ddSJeff Bonwick * txgs that was scrubbed. 23328ad4d6ddSJeff Bonwick * 23338ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 23348ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 23358ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 23368ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 23378ad4d6ddSJeff Bonwick * determine whether a device can be detached. 23388ad4d6ddSJeff Bonwick * 23398ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 23408ad4d6ddSJeff Bonwick * either has the data or it doesn't. 23418ad4d6ddSJeff Bonwick * 23428ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 23438ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 23448ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 23458ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 23468ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 23478ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 23488ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 23498ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 23508ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 23518ad4d6ddSJeff Bonwick * 23528ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 23538ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 23548ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 23558ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 23568ad4d6ddSJeff Bonwick */ 2357fa9e4066Sahrens void 23588ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2359fa9e4066Sahrens { 23600713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23618ad4d6ddSJeff Bonwick 23628ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23638ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2364f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 23658ad4d6ddSJeff Bonwick 23665cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 23670713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 23680713e232SGeorge Wilson range_tree_add(rt, txg, size); 23695cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2370fa9e4066Sahrens } 2371fa9e4066Sahrens 23728ad4d6ddSJeff Bonwick boolean_t 23738ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2374fa9e4066Sahrens { 23750713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23768ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2377fa9e4066Sahrens 23788ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23798ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2380fa9e4066Sahrens 23815cabbc6bSPrashanth Sreenivasa /* 23825cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 23835cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 23845cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 23855cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 23865cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 23875cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 23885cabbc6bSPrashanth Sreenivasa */ 23895cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 23905cabbc6bSPrashanth Sreenivasa return (B_FALSE); 23915cabbc6bSPrashanth Sreenivasa 23925cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 239386714001SSerapheim Dimitropoulos if (!range_tree_is_empty(rt)) 23940713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 23955cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2396fa9e4066Sahrens 2397fa9e4066Sahrens return (dirty); 2398fa9e4066Sahrens } 2399fa9e4066Sahrens 24008ad4d6ddSJeff Bonwick boolean_t 24018ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 24028ad4d6ddSJeff Bonwick { 24030713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 24048ad4d6ddSJeff Bonwick boolean_t empty; 24058ad4d6ddSJeff Bonwick 24065cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 240786714001SSerapheim Dimitropoulos empty = range_tree_is_empty(rt); 24085cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 24098ad4d6ddSJeff Bonwick 24108ad4d6ddSJeff Bonwick return (empty); 24118ad4d6ddSJeff Bonwick } 24128ad4d6ddSJeff Bonwick 2413a3874b8bSToomas Soome /* 2414a3874b8bSToomas Soome * Returns B_TRUE if vdev determines offset needs to be resilvered. 2415a3874b8bSToomas Soome */ 2416a3874b8bSToomas Soome boolean_t 2417a3874b8bSToomas Soome vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2418a3874b8bSToomas Soome { 2419a3874b8bSToomas Soome ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2420a3874b8bSToomas Soome 2421a3874b8bSToomas Soome if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2422a3874b8bSToomas Soome vd->vdev_ops->vdev_op_leaf) 2423a3874b8bSToomas Soome return (B_TRUE); 2424a3874b8bSToomas Soome 2425a3874b8bSToomas Soome return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2426a3874b8bSToomas Soome } 2427a3874b8bSToomas Soome 2428b4952e17SGeorge Wilson /* 2429b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2430b4952e17SGeorge Wilson */ 2431b4952e17SGeorge Wilson static uint64_t 2432b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2433b4952e17SGeorge Wilson { 2434b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 24350713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2436b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2437b4952e17SGeorge Wilson 24384d7988d6SPaul Dagnelie return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); 2439b4952e17SGeorge Wilson } 2440b4952e17SGeorge Wilson 2441b4952e17SGeorge Wilson /* 2442b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2443b4952e17SGeorge Wilson */ 2444b4952e17SGeorge Wilson static uint64_t 2445b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 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 24514d7988d6SPaul Dagnelie return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); 2452b4952e17SGeorge Wilson } 2453b4952e17SGeorge Wilson 2454b4952e17SGeorge Wilson /* 2455b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2456b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2457b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2458b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2459b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2460b4952e17SGeorge Wilson * excise the DTLs. 2461b4952e17SGeorge Wilson */ 2462b4952e17SGeorge Wilson static boolean_t 2463b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2464b4952e17SGeorge Wilson { 2465b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2466b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2467b4952e17SGeorge Wilson 2468b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 2469b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2470b4952e17SGeorge Wilson 24712d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 24722d2f193aSMatthew Ahrens return (B_FALSE); 24732d2f193aSMatthew Ahrens 2474e4c795beSTom Caputi if (vd->vdev_resilver_deferred) 2475e4c795beSTom Caputi return (B_FALSE); 2476e4c795beSTom Caputi 2477b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 247886714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2479b4952e17SGeorge Wilson return (B_TRUE); 2480b4952e17SGeorge Wilson 2481b4952e17SGeorge Wilson /* 2482b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2483b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2484b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2485b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2486b4952e17SGeorge Wilson * for excision. 2487b4952e17SGeorge Wilson */ 2488b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2489b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2490b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2491b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2492b4952e17SGeorge Wilson return (B_TRUE); 2493b4952e17SGeorge Wilson } 2494b4952e17SGeorge Wilson return (B_FALSE); 2495b4952e17SGeorge Wilson } 2496b4952e17SGeorge Wilson 2497fa9e4066Sahrens /* 2498fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2499fa9e4066Sahrens */ 2500fa9e4066Sahrens void 2501fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2502fa9e4066Sahrens { 2503ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 25048ad4d6ddSJeff Bonwick avl_tree_t reftree; 25058ad4d6ddSJeff Bonwick int minref; 2506fa9e4066Sahrens 25078ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2508fa9e4066Sahrens 25098ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 25108ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 25118ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 25128ad4d6ddSJeff Bonwick 25135cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 25148ad4d6ddSJeff Bonwick return; 25158ad4d6ddSJeff Bonwick 25168ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 25173f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 25183f9d6ad7SLin Ling 2519fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2520b4952e17SGeorge Wilson 2521b4952e17SGeorge Wilson /* 2522b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2523b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2524b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2525b4952e17SGeorge Wilson * the entire duration of this scan. 2526b4952e17SGeorge Wilson */ 2527088f3894Sahrens if (scrub_txg != 0 && 25283f9d6ad7SLin Ling (spa->spa_scrub_started || 2529b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2530b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2531088f3894Sahrens /* 2532088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2533088f3894Sahrens * did it without rebooting, then the scrub dtl 2534088f3894Sahrens * will be valid, so excise the old region and 2535088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2536088f3894Sahrens * dtl as-is if there was an error. 25378ad4d6ddSJeff Bonwick * 25388ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 25398ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 25408ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 25418ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 25428ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 25438ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 25448ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 25458ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 25468ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2547088f3894Sahrens */ 25480713e232SGeorge Wilson space_reftree_create(&reftree); 25490713e232SGeorge Wilson space_reftree_add_map(&reftree, 25500713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25510713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 25520713e232SGeorge Wilson space_reftree_add_map(&reftree, 25530713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 25540713e232SGeorge Wilson space_reftree_generate_map(&reftree, 25550713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25560713e232SGeorge Wilson space_reftree_destroy(&reftree); 2557fa9e4066Sahrens } 25580713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 25590713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 25600713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2561fa9e4066Sahrens if (scrub_done) 25620713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 25630713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 25648ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 25650713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 25668ad4d6ddSJeff Bonwick else 25670713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 25680713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2569b4952e17SGeorge Wilson 2570b4952e17SGeorge Wilson /* 2571b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2572b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2573b4952e17SGeorge Wilson */ 2574b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 257586714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 257686714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2577b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2578b4952e17SGeorge Wilson 2579fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2580088f3894Sahrens 2581ecc2d604Sbonwick if (txg != 0) 2582ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2583fa9e4066Sahrens return; 2584fa9e4066Sahrens } 2585fa9e4066Sahrens 2586fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 25878ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 258899bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 258999bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 25908ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 25918ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 25928ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 25938ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 25948ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 25958ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 25968ad4d6ddSJeff Bonwick else 25978ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 25980713e232SGeorge Wilson space_reftree_create(&reftree); 25998ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 26008ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 26018ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 26020713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 26038ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 26048ad4d6ddSJeff Bonwick } 26050713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 26060713e232SGeorge Wilson space_reftree_destroy(&reftree); 2607fa9e4066Sahrens } 26088ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2609fa9e4066Sahrens } 2610fa9e4066Sahrens 26110713e232SGeorge Wilson int 2612fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2613fa9e4066Sahrens { 2614fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2615ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 26160713e232SGeorge Wilson int error = 0; 2617fa9e4066Sahrens 26180713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 26195cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2620fa9e4066Sahrens 26210713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 26225cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 26230713e232SGeorge Wilson if (error) 26240713e232SGeorge Wilson return (error); 26250713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2626fa9e4066Sahrens 26270713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 26280713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 26290713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 26300713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2631fa9e4066Sahrens 26320713e232SGeorge Wilson return (error); 26330713e232SGeorge Wilson } 26340713e232SGeorge Wilson 26350713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 26360713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 26370713e232SGeorge Wilson if (error != 0) 26380713e232SGeorge Wilson break; 26390713e232SGeorge Wilson } 2640fa9e4066Sahrens 2641fa9e4066Sahrens return (error); 2642fa9e4066Sahrens } 2643fa9e4066Sahrens 2644663207adSDon Brady static void 2645663207adSDon Brady vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2646663207adSDon Brady { 2647663207adSDon Brady spa_t *spa = vd->vdev_spa; 2648663207adSDon Brady objset_t *mos = spa->spa_meta_objset; 2649663207adSDon Brady vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2650663207adSDon Brady const char *string; 2651663207adSDon Brady 2652663207adSDon Brady ASSERT(alloc_bias != VDEV_BIAS_NONE); 2653663207adSDon Brady 2654663207adSDon Brady string = 2655663207adSDon Brady (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2656663207adSDon Brady (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2657663207adSDon Brady (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2658663207adSDon Brady 2659663207adSDon Brady ASSERT(string != NULL); 2660663207adSDon Brady VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2661663207adSDon Brady 1, strlen(string) + 1, string, tx)); 2662663207adSDon Brady 2663663207adSDon Brady if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2664663207adSDon Brady spa_activate_allocation_classes(spa, tx); 2665663207adSDon Brady } 2666663207adSDon Brady } 2667663207adSDon Brady 2668215198a6SJoe Stein void 2669215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2670215198a6SJoe Stein { 2671215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2672215198a6SJoe Stein 2673215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2674215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2675215198a6SJoe Stein zapobj, tx)); 2676215198a6SJoe Stein } 2677215198a6SJoe Stein 2678215198a6SJoe Stein uint64_t 2679215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2680215198a6SJoe Stein { 2681215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2682215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2683215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2684215198a6SJoe Stein 2685215198a6SJoe Stein ASSERT(zap != 0); 2686215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2687215198a6SJoe Stein zap, tx)); 2688215198a6SJoe Stein 2689215198a6SJoe Stein return (zap); 2690215198a6SJoe Stein } 2691215198a6SJoe Stein 2692215198a6SJoe Stein void 2693215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2694215198a6SJoe Stein { 2695215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2696215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2697215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2698215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2699215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2700215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2701215198a6SJoe Stein } 2702215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2703215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2704663207adSDon Brady if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2705663207adSDon Brady vdev_zap_allocation_data(vd, tx); 2706215198a6SJoe Stein } 2707215198a6SJoe Stein } 2708663207adSDon Brady 2709215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2710215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2711215198a6SJoe Stein } 2712215198a6SJoe Stein } 2713215198a6SJoe Stein 2714fa9e4066Sahrens void 2715fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2716fa9e4066Sahrens { 2717fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 27180713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2719ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 27200713e232SGeorge Wilson range_tree_t *rtsync; 2721fa9e4066Sahrens dmu_tx_t *tx; 27220713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2723fa9e4066Sahrens 27245cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 27250713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 272688ecc943SGeorge Wilson 2727fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2728fa9e4066Sahrens 27290713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 27300713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 27310713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 27320713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 27330713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 27340713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2735215198a6SJoe Stein 2736215198a6SJoe Stein /* 2737215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2738215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2739215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2740215198a6SJoe Stein */ 2741215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2742215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2743215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2744215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2745215198a6SJoe Stein } 2746215198a6SJoe Stein 2747fa9e4066Sahrens dmu_tx_commit(tx); 2748fa9e4066Sahrens return; 2749fa9e4066Sahrens } 2750fa9e4066Sahrens 27510713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 27520713e232SGeorge Wilson uint64_t new_object; 27530713e232SGeorge Wilson 2754814dcd43SSerapheim Dimitropoulos new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); 27550713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 27560713e232SGeorge Wilson 27570713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 27585cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 27590713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2760fa9e4066Sahrens } 2761fa9e4066Sahrens 27624d7988d6SPaul Dagnelie rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 2763fa9e4066Sahrens 2764fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 27650713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2766fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2767fa9e4066Sahrens 2768814dcd43SSerapheim Dimitropoulos space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); 276917f11284SSerapheim Dimitropoulos space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 27700713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2771fa9e4066Sahrens 27720713e232SGeorge Wilson range_tree_destroy(rtsync); 2773fa9e4066Sahrens 27740713e232SGeorge Wilson /* 27750713e232SGeorge Wilson * If the object for the space map has changed then dirty 27760713e232SGeorge Wilson * the top level so that we update the config. 27770713e232SGeorge Wilson */ 27780713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 27793ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 27803ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 27813ee8c80cSPavel Zakharov (u_longlong_t)object, 27823ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 27830713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 27840713e232SGeorge Wilson } 2785fa9e4066Sahrens 2786fa9e4066Sahrens dmu_tx_commit(tx); 2787fa9e4066Sahrens } 2788fa9e4066Sahrens 27898ad4d6ddSJeff Bonwick /* 27908ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 27918ad4d6ddSJeff Bonwick * without losing data. 27928ad4d6ddSJeff Bonwick */ 27938ad4d6ddSJeff Bonwick boolean_t 27948ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 27958ad4d6ddSJeff Bonwick { 27968ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 27978ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 27988ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 27998ad4d6ddSJeff Bonwick boolean_t required; 28008ad4d6ddSJeff Bonwick 28018ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 28028ad4d6ddSJeff Bonwick 28038ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 28048ad4d6ddSJeff Bonwick return (B_TRUE); 28058ad4d6ddSJeff Bonwick 28068ad4d6ddSJeff Bonwick /* 28078ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 28088ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 28098ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 28108ad4d6ddSJeff Bonwick */ 28118ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 28128ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28138ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 28148ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 28158ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28168ad4d6ddSJeff Bonwick 2817cb04b873SMark J Musante if (!required && zio_injection_enabled) 2818cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2819cb04b873SMark J Musante 28208ad4d6ddSJeff Bonwick return (required); 28218ad4d6ddSJeff Bonwick } 28228ad4d6ddSJeff Bonwick 2823088f3894Sahrens /* 2824088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2825088f3894Sahrens */ 2826088f3894Sahrens boolean_t 2827088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2828088f3894Sahrens { 2829088f3894Sahrens boolean_t needed = B_FALSE; 2830088f3894Sahrens uint64_t thismin = UINT64_MAX; 2831088f3894Sahrens uint64_t thismax = 0; 2832088f3894Sahrens 2833088f3894Sahrens if (vd->vdev_children == 0) { 2834088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 283586714001SSerapheim Dimitropoulos if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 28368ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2837088f3894Sahrens 2838b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2839b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2840088f3894Sahrens needed = B_TRUE; 2841088f3894Sahrens } 2842088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2843088f3894Sahrens } else { 28448ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2845088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2846088f3894Sahrens uint64_t cmin, cmax; 2847088f3894Sahrens 2848088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2849088f3894Sahrens thismin = MIN(thismin, cmin); 2850088f3894Sahrens thismax = MAX(thismax, cmax); 2851088f3894Sahrens needed = B_TRUE; 2852088f3894Sahrens } 2853088f3894Sahrens } 2854088f3894Sahrens } 2855088f3894Sahrens 2856088f3894Sahrens if (needed && minp) { 2857088f3894Sahrens *minp = thismin; 2858088f3894Sahrens *maxp = thismax; 2859088f3894Sahrens } 2860088f3894Sahrens return (needed); 2861088f3894Sahrens } 2862088f3894Sahrens 286386714001SSerapheim Dimitropoulos /* 286486714001SSerapheim Dimitropoulos * Gets the checkpoint space map object from the vdev's ZAP. 286586714001SSerapheim Dimitropoulos * Returns the spacemap object, or 0 if it wasn't in the ZAP 286686714001SSerapheim Dimitropoulos * or the ZAP doesn't exist yet. 286786714001SSerapheim Dimitropoulos */ 286886714001SSerapheim Dimitropoulos int 286986714001SSerapheim Dimitropoulos vdev_checkpoint_sm_object(vdev_t *vd) 287086714001SSerapheim Dimitropoulos { 287186714001SSerapheim Dimitropoulos ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 287286714001SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) { 287386714001SSerapheim Dimitropoulos return (0); 287486714001SSerapheim Dimitropoulos } 287586714001SSerapheim Dimitropoulos 287686714001SSerapheim Dimitropoulos uint64_t sm_obj = 0; 287786714001SSerapheim Dimitropoulos int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 287886714001SSerapheim Dimitropoulos VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 287986714001SSerapheim Dimitropoulos 288086714001SSerapheim Dimitropoulos ASSERT(err == 0 || err == ENOENT); 288186714001SSerapheim Dimitropoulos 288286714001SSerapheim Dimitropoulos return (sm_obj); 288386714001SSerapheim Dimitropoulos } 288486714001SSerapheim Dimitropoulos 28855cabbc6bSPrashanth Sreenivasa int 2886ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2887fa9e4066Sahrens { 28885cabbc6bSPrashanth Sreenivasa int error = 0; 2889fa9e4066Sahrens /* 2890fa9e4066Sahrens * Recursively load all children. 2891fa9e4066Sahrens */ 28925cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 28935cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 28945cabbc6bSPrashanth Sreenivasa if (error != 0) { 28955cabbc6bSPrashanth Sreenivasa return (error); 28965cabbc6bSPrashanth Sreenivasa } 28975cabbc6bSPrashanth Sreenivasa } 28985cabbc6bSPrashanth Sreenivasa 28995cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2900fa9e4066Sahrens 2901663207adSDon Brady /* 2902663207adSDon Brady * On spa_load path, grab the allocation bias from our zap 2903663207adSDon Brady */ 2904663207adSDon Brady if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2905663207adSDon Brady spa_t *spa = vd->vdev_spa; 2906663207adSDon Brady char bias_str[64]; 2907663207adSDon Brady 2908663207adSDon Brady if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2909663207adSDon Brady VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2910663207adSDon Brady bias_str) == 0) { 2911663207adSDon Brady ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2912663207adSDon Brady vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2913663207adSDon Brady } 2914663207adSDon Brady } 2915663207adSDon Brady 2916fa9e4066Sahrens /* 29170e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2918fa9e4066Sahrens */ 29195cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2920663207adSDon Brady vdev_metaslab_group_create(vd); 2921663207adSDon Brady 29225cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 29235cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29245cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29253ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 29263ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 29273ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 29285cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 2929555d674dSSerapheim Dimitropoulos } 2930555d674dSSerapheim Dimitropoulos 2931555d674dSSerapheim Dimitropoulos error = vdev_metaslab_init(vd, 0); 2932555d674dSSerapheim Dimitropoulos if (error != 0) { 29333ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 29343ee8c80cSPavel Zakharov "[error=%d]", error); 29355cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29365cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29375cabbc6bSPrashanth Sreenivasa return (error); 29385cabbc6bSPrashanth Sreenivasa } 293986714001SSerapheim Dimitropoulos 294086714001SSerapheim Dimitropoulos uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 294186714001SSerapheim Dimitropoulos if (checkpoint_sm_obj != 0) { 294286714001SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 294386714001SSerapheim Dimitropoulos ASSERT(vd->vdev_asize != 0); 294486714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 294586714001SSerapheim Dimitropoulos 2946555d674dSSerapheim Dimitropoulos error = space_map_open(&vd->vdev_checkpoint_sm, 294786714001SSerapheim Dimitropoulos mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2948555d674dSSerapheim Dimitropoulos vd->vdev_ashift); 2949555d674dSSerapheim Dimitropoulos if (error != 0) { 295086714001SSerapheim Dimitropoulos vdev_dbgmsg(vd, "vdev_load: space_map_open " 295186714001SSerapheim Dimitropoulos "failed for checkpoint spacemap (obj %llu) " 295286714001SSerapheim Dimitropoulos "[error=%d]", 295386714001SSerapheim Dimitropoulos (u_longlong_t)checkpoint_sm_obj, error); 295486714001SSerapheim Dimitropoulos return (error); 295586714001SSerapheim Dimitropoulos } 295686714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 295786714001SSerapheim Dimitropoulos 295886714001SSerapheim Dimitropoulos /* 295986714001SSerapheim Dimitropoulos * Since the checkpoint_sm contains free entries 2960555d674dSSerapheim Dimitropoulos * exclusively we can use space_map_allocated() to 2961555d674dSSerapheim Dimitropoulos * indicate the cumulative checkpointed space that 2962555d674dSSerapheim Dimitropoulos * has been freed. 296386714001SSerapheim Dimitropoulos */ 296486714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space = 2965555d674dSSerapheim Dimitropoulos -space_map_allocated(vd->vdev_checkpoint_sm); 296686714001SSerapheim Dimitropoulos vd->vdev_spa->spa_checkpoint_info.sci_dspace += 296786714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space; 296886714001SSerapheim Dimitropoulos } 29695cabbc6bSPrashanth Sreenivasa } 2970fa9e4066Sahrens 2971fa9e4066Sahrens /* 2972fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2973fa9e4066Sahrens */ 29745cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2975560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2976560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 29773ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 29783ee8c80cSPavel Zakharov "[error=%d]", error); 29795cabbc6bSPrashanth Sreenivasa return (error); 29805cabbc6bSPrashanth Sreenivasa } 29815cabbc6bSPrashanth Sreenivasa 29825cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 29835cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 29845cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 29855cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 298686714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 29875cabbc6bSPrashanth Sreenivasa 29885cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 29895cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 29905cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29915cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29923ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 29933ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 29943ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 29955cabbc6bSPrashanth Sreenivasa return (error); 29965cabbc6bSPrashanth Sreenivasa } 29975cabbc6bSPrashanth Sreenivasa } 29985cabbc6bSPrashanth Sreenivasa 29995cabbc6bSPrashanth Sreenivasa return (0); 3000fa9e4066Sahrens } 3001fa9e4066Sahrens 300299653d4eSeschrock /* 3003fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 3004fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 3005fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 3006fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 3007fa94a07fSbrendan * repurposed to another pool. 300899653d4eSeschrock */ 300999653d4eSeschrock int 3010fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 301199653d4eSeschrock { 301299653d4eSeschrock nvlist_t *label; 301399653d4eSeschrock uint64_t guid, version; 301499653d4eSeschrock uint64_t state; 301599653d4eSeschrock 3016e14bb325SJeff Bonwick if (!vdev_readable(vd)) 3017c5904d13Seschrock return (0); 3018c5904d13Seschrock 3019dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 302099653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 302199653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 302299653d4eSeschrock return (-1); 302399653d4eSeschrock } 302499653d4eSeschrock 302599653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3026ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 302799653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 302899653d4eSeschrock guid != vd->vdev_guid || 302999653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 303099653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 303199653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 303299653d4eSeschrock nvlist_free(label); 303399653d4eSeschrock return (-1); 303499653d4eSeschrock } 303599653d4eSeschrock 303699653d4eSeschrock /* 303799653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 303899653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 303999653d4eSeschrock */ 304099653d4eSeschrock nvlist_free(label); 304199653d4eSeschrock return (0); 304299653d4eSeschrock } 304399653d4eSeschrock 3044814dcd43SSerapheim Dimitropoulos static void 3045814dcd43SSerapheim Dimitropoulos vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) 3046814dcd43SSerapheim Dimitropoulos { 3047814dcd43SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 3048814dcd43SSerapheim Dimitropoulos 3049814dcd43SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) 3050814dcd43SSerapheim Dimitropoulos return; 3051814dcd43SSerapheim Dimitropoulos 3052814dcd43SSerapheim Dimitropoulos uint64_t object = 0; 3053814dcd43SSerapheim Dimitropoulos int err = zap_lookup(mos, vd->vdev_top_zap, 3054814dcd43SSerapheim Dimitropoulos VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); 3055814dcd43SSerapheim Dimitropoulos if (err == ENOENT) 3056814dcd43SSerapheim Dimitropoulos return; 3057814dcd43SSerapheim Dimitropoulos 3058814dcd43SSerapheim Dimitropoulos VERIFY0(dmu_object_free(mos, object, tx)); 3059814dcd43SSerapheim Dimitropoulos VERIFY0(zap_remove(mos, vd->vdev_top_zap, 3060814dcd43SSerapheim Dimitropoulos VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); 3061814dcd43SSerapheim Dimitropoulos } 3062814dcd43SSerapheim Dimitropoulos 30635cabbc6bSPrashanth Sreenivasa /* 30645cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 30655cabbc6bSPrashanth Sreenivasa * that points to them. 30665cabbc6bSPrashanth Sreenivasa */ 306788ecc943SGeorge Wilson void 30685cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 30695cabbc6bSPrashanth Sreenivasa { 30705cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 30715cabbc6bSPrashanth Sreenivasa return; 30725cabbc6bSPrashanth Sreenivasa 30735cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 30745cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 30755cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 30765cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 30775cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 30785cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 30795cabbc6bSPrashanth Sreenivasa 30805cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 30815cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 30825cabbc6bSPrashanth Sreenivasa if (smobj == 0) 30835cabbc6bSPrashanth Sreenivasa continue; 30845cabbc6bSPrashanth Sreenivasa 30855cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 30865cabbc6bSPrashanth Sreenivasa } 30875cabbc6bSPrashanth Sreenivasa 30885cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 30895cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3090814dcd43SSerapheim Dimitropoulos vdev_destroy_ms_flush_data(vd, tx); 30915cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 30925cabbc6bSPrashanth Sreenivasa } 30935cabbc6bSPrashanth Sreenivasa 30945cabbc6bSPrashanth Sreenivasa static void 30954e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 309688ecc943SGeorge Wilson { 309788ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 309888ecc943SGeorge Wilson 30994e75ba68SSerapheim Dimitropoulos ASSERT(vd->vdev_islog); 3100215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 3101215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 310288ecc943SGeorge Wilson 31034e75ba68SSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3104215198a6SJoe Stein 31054e75ba68SSerapheim Dimitropoulos vdev_destroy_spacemaps(vd, tx); 31064e75ba68SSerapheim Dimitropoulos if (vd->vdev_top_zap != 0) { 3107215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3108215198a6SJoe Stein vd->vdev_top_zap = 0; 3109215198a6SJoe Stein } 31104e75ba68SSerapheim Dimitropoulos 311188ecc943SGeorge Wilson dmu_tx_commit(tx); 311288ecc943SGeorge Wilson } 311388ecc943SGeorge Wilson 3114fa9e4066Sahrens void 3115fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 3116fa9e4066Sahrens { 3117fa9e4066Sahrens metaslab_t *msp; 311880eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3119fa9e4066Sahrens 31205cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 312188ecc943SGeorge Wilson 3122094e47e9SGeorge Wilson while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3123094e47e9SGeorge Wilson != NULL) 3124fa9e4066Sahrens metaslab_sync_done(msp, txg); 312580eb36f2SGeorge Wilson 312680eb36f2SGeorge Wilson if (reassess) 312780eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 3128fa9e4066Sahrens } 3129fa9e4066Sahrens 3130fa9e4066Sahrens void 3131fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 3132fa9e4066Sahrens { 3133fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3134fa9e4066Sahrens vdev_t *lvd; 3135fa9e4066Sahrens metaslab_t *msp; 3136fa9e4066Sahrens 3137555d674dSSerapheim Dimitropoulos ASSERT3U(txg, ==, spa->spa_syncing_txg); 3138555d674dSSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 31395cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 31405cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 31415cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 314288ecc943SGeorge Wilson 31435cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 31445cabbc6bSPrashanth Sreenivasa 31455cabbc6bSPrashanth Sreenivasa /* 31465cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 31475cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 31485cabbc6bSPrashanth Sreenivasa */ 31495cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 31505cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 31515cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3152555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 31535cabbc6bSPrashanth Sreenivasa return; 31545cabbc6bSPrashanth Sreenivasa } 31555cabbc6bSPrashanth Sreenivasa } 31565cabbc6bSPrashanth Sreenivasa 31575cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 31585cabbc6bSPrashanth Sreenivasa 31595cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 31605cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 3161ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 31625cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3163ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3164ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3165ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 3166ecc2d604Sbonwick vdev_config_dirty(vd); 3167ecc2d604Sbonwick } 3168fa9e4066Sahrens 3169ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3170fa9e4066Sahrens metaslab_sync(msp, txg); 3171ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3172ecc2d604Sbonwick } 3173fa9e4066Sahrens 3174fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3175fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 3176fa9e4066Sahrens 31775cabbc6bSPrashanth Sreenivasa /* 31784e75ba68SSerapheim Dimitropoulos * If this is an empty log device being removed, destroy the 31794e75ba68SSerapheim Dimitropoulos * metadata associated with it. 31805cabbc6bSPrashanth Sreenivasa */ 31814e75ba68SSerapheim Dimitropoulos if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 31824e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vd, txg); 31835cabbc6bSPrashanth Sreenivasa 3184fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3185555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 3186fa9e4066Sahrens } 3187fa9e4066Sahrens 3188fa9e4066Sahrens uint64_t 3189fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3190fa9e4066Sahrens { 3191fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3192fa9e4066Sahrens } 3193fa9e4066Sahrens 31943d7072f8Seschrock /* 31953d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 31963d7072f8Seschrock * not be opened, and no I/O is attempted. 31973d7072f8Seschrock */ 3198fa9e4066Sahrens int 3199069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3200fa9e4066Sahrens { 32014b964adaSGeorge Wilson vdev_t *vd, *tvd; 3202fa9e4066Sahrens 32038f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 3204fa9e4066Sahrens 3205c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3206e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3207e14bb325SJeff Bonwick 32083d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3209e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3210fa9e4066Sahrens 32114b964adaSGeorge Wilson tvd = vd->vdev_top; 32124b964adaSGeorge Wilson 3213069f55e2SEric Schrock /* 3214069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 3215069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 3216069f55e2SEric Schrock * were faulted. 3217069f55e2SEric Schrock */ 3218069f55e2SEric Schrock vd->vdev_label_aux = aux; 3219069f55e2SEric Schrock 32203d7072f8Seschrock /* 32213d7072f8Seschrock * Faulted state takes precedence over degraded. 32223d7072f8Seschrock */ 322398d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 32243d7072f8Seschrock vd->vdev_faulted = 1ULL; 32253d7072f8Seschrock vd->vdev_degraded = 0ULL; 3226069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 32273d7072f8Seschrock 32283d7072f8Seschrock /* 3229c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 3230c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 32313d7072f8Seschrock */ 32324b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 32333d7072f8Seschrock vd->vdev_degraded = 1ULL; 32343d7072f8Seschrock vd->vdev_faulted = 0ULL; 32353d7072f8Seschrock 32363d7072f8Seschrock /* 32373d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 32383d7072f8Seschrock * mark it degraded. 32393d7072f8Seschrock */ 32404b964adaSGeorge Wilson vdev_reopen(tvd); 32413d7072f8Seschrock 3242069f55e2SEric Schrock if (vdev_readable(vd)) 3243069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 32443d7072f8Seschrock } 32453d7072f8Seschrock 3246e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 32473d7072f8Seschrock } 32483d7072f8Seschrock 32493d7072f8Seschrock /* 32503d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 32513d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 32523d7072f8Seschrock * as I/O is concerned. 32533d7072f8Seschrock */ 32543d7072f8Seschrock int 3255069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 32563d7072f8Seschrock { 3257c5904d13Seschrock vdev_t *vd; 32580a4e9518Sgw 32598f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 32603d7072f8Seschrock 3261c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3262e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3263e14bb325SJeff Bonwick 32640e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3265e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 32660e34b6a7Sbonwick 32673d7072f8Seschrock /* 32683d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 32693d7072f8Seschrock */ 3270e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 3271e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 32723d7072f8Seschrock 32733d7072f8Seschrock vd->vdev_degraded = 1ULL; 32743d7072f8Seschrock if (!vdev_is_dead(vd)) 32753d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3276069f55e2SEric Schrock aux); 32773d7072f8Seschrock 3278e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 32793d7072f8Seschrock } 32803d7072f8Seschrock 32813d7072f8Seschrock /* 3282f7170741SWill Andrews * Online the given vdev. 3283f7170741SWill Andrews * 3284f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3285f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 3286f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 3287f7170741SWill Andrews * events are generated if the device fails to open. 32883d7072f8Seschrock */ 32893d7072f8Seschrock int 3290e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 32913d7072f8Seschrock { 3292573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 32935f368aefSYuri Pankov boolean_t wasoffline; 32945f368aefSYuri Pankov vdev_state_t oldstate; 32953d7072f8Seschrock 32968f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 32973d7072f8Seschrock 3298c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3299e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 33003d7072f8Seschrock 33013d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3302e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3303fa9e4066Sahrens 33045f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 33055f368aefSYuri Pankov oldstate = vd->vdev_state; 330614372834SHans Rosenfeld 3307573ca77eSGeorge Wilson tvd = vd->vdev_top; 3308fa9e4066Sahrens vd->vdev_offline = B_FALSE; 3309441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 3310e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3311e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3312573ca77eSGeorge Wilson 3313573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3314573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3315573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3316573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3317573ca77eSGeorge Wilson } 3318573ca77eSGeorge Wilson 3319573ca77eSGeorge Wilson vdev_reopen(tvd); 33203d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 33213d7072f8Seschrock 3322573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3323573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3324573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 3325573ca77eSGeorge Wilson } 3326573ca77eSGeorge Wilson 33273d7072f8Seschrock if (newstate) 33283d7072f8Seschrock *newstate = vd->vdev_state; 33293d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 33303d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 33313d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 33323d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 33333d7072f8Seschrock vd->vdev_unspare = B_TRUE; 3334fa9e4066Sahrens 3335573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3336573ca77eSGeorge Wilson 3337573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3338573ca77eSGeorge Wilson if (vd->vdev_aux) 3339573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3340573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3341573ca77eSGeorge Wilson } 334214372834SHans Rosenfeld 3343094e47e9SGeorge Wilson /* Restart initializing if necessary */ 3344094e47e9SGeorge Wilson mutex_enter(&vd->vdev_initialize_lock); 3345094e47e9SGeorge Wilson if (vdev_writeable(vd) && 3346094e47e9SGeorge Wilson vd->vdev_initialize_thread == NULL && 3347094e47e9SGeorge Wilson vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3348094e47e9SGeorge Wilson (void) vdev_initialize(vd); 3349094e47e9SGeorge Wilson } 3350094e47e9SGeorge Wilson mutex_exit(&vd->vdev_initialize_lock); 3351094e47e9SGeorge Wilson 3352084fd14fSBrian Behlendorf /* Restart trimming if necessary */ 3353084fd14fSBrian Behlendorf mutex_enter(&vd->vdev_trim_lock); 3354084fd14fSBrian Behlendorf if (vdev_writeable(vd) && 3355084fd14fSBrian Behlendorf vd->vdev_trim_thread == NULL && 3356084fd14fSBrian Behlendorf vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 3357084fd14fSBrian Behlendorf (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 3358084fd14fSBrian Behlendorf vd->vdev_trim_secure); 3359084fd14fSBrian Behlendorf } 3360084fd14fSBrian Behlendorf mutex_exit(&vd->vdev_trim_lock); 3361084fd14fSBrian Behlendorf 33625f368aefSYuri Pankov if (wasoffline || 33635f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 33645f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 3365ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 336614372834SHans Rosenfeld 33678ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 3368fa9e4066Sahrens } 3369fa9e4066Sahrens 3370a1521560SJeff Bonwick static int 3371a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3372fa9e4066Sahrens { 3373e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 33748f18d1faSGeorge Wilson int error = 0; 33758f18d1faSGeorge Wilson uint64_t generation; 33768f18d1faSGeorge Wilson metaslab_group_t *mg; 33770a4e9518Sgw 33788f18d1faSGeorge Wilson top: 33798f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 3380fa9e4066Sahrens 3381c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3382e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3383fa9e4066Sahrens 33840e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3385e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 33860e34b6a7Sbonwick 3387e6ca193dSGeorge Wilson tvd = vd->vdev_top; 33888f18d1faSGeorge Wilson mg = tvd->vdev_mg; 33898f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 3390e6ca193dSGeorge Wilson 3391fa9e4066Sahrens /* 3392ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 3393fa9e4066Sahrens */ 3394ecc2d604Sbonwick if (!vd->vdev_offline) { 3395ecc2d604Sbonwick /* 33968ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 3397e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 3398e6ca193dSGeorge Wilson * expendable. 3399ecc2d604Sbonwick */ 3400e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3401e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 3402e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3403fa9e4066Sahrens 34048f18d1faSGeorge Wilson /* 3405b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 3406b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 3407b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 3408b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 34098f18d1faSGeorge Wilson */ 34108f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 34118f18d1faSGeorge Wilson /* 34128f18d1faSGeorge Wilson * Prevent any future allocations. 34138f18d1faSGeorge Wilson */ 3414a1521560SJeff Bonwick metaslab_group_passivate(mg); 34158f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 34168f18d1faSGeorge Wilson 34175cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 34188f18d1faSGeorge Wilson 341986714001SSerapheim Dimitropoulos /* 342086714001SSerapheim Dimitropoulos * If the log device was successfully reset but has 342186714001SSerapheim Dimitropoulos * checkpointed data, do not offline it. 342286714001SSerapheim Dimitropoulos */ 342386714001SSerapheim Dimitropoulos if (error == 0 && 342486714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm != NULL) { 342586714001SSerapheim Dimitropoulos error = ZFS_ERR_CHECKPOINT_EXISTS; 342686714001SSerapheim Dimitropoulos } 342786714001SSerapheim Dimitropoulos 34288f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 34298f18d1faSGeorge Wilson 34308f18d1faSGeorge Wilson /* 34318f18d1faSGeorge Wilson * Check to see if the config has changed. 34328f18d1faSGeorge Wilson */ 34338f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 3434a1521560SJeff Bonwick metaslab_group_activate(mg); 34358f18d1faSGeorge Wilson if (error) 34368f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 34378f18d1faSGeorge Wilson vd, error)); 34388f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 34398f18d1faSGeorge Wilson goto top; 34408f18d1faSGeorge Wilson } 3441fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 34428f18d1faSGeorge Wilson } 34438f18d1faSGeorge Wilson 3444ecc2d604Sbonwick /* 3445ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3446e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3447e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3448e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3449ecc2d604Sbonwick */ 3450ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3451e6ca193dSGeorge Wilson vdev_reopen(tvd); 3452e6ca193dSGeorge Wilson 3453e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3454e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3455ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3456e6ca193dSGeorge Wilson vdev_reopen(tvd); 3457e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3458ecc2d604Sbonwick } 34598f18d1faSGeorge Wilson 34608f18d1faSGeorge Wilson /* 34618f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 34628f18d1faSGeorge Wilson * once we online the device it's open for business. 34638f18d1faSGeorge Wilson */ 34648f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3465a1521560SJeff Bonwick metaslab_group_activate(mg); 3466fa9e4066Sahrens } 3467fa9e4066Sahrens 3468e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3469ecc2d604Sbonwick 34708f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3471fa9e4066Sahrens } 3472fa9e4066Sahrens 3473a1521560SJeff Bonwick int 3474a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3475a1521560SJeff Bonwick { 3476a1521560SJeff Bonwick int error; 3477a1521560SJeff Bonwick 3478a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3479a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3480a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3481a1521560SJeff Bonwick 3482a1521560SJeff Bonwick return (error); 3483a1521560SJeff Bonwick } 3484a1521560SJeff Bonwick 3485ea8dc4b6Seschrock /* 3486ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3487ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3488ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3489ea8dc4b6Seschrock */ 3490ea8dc4b6Seschrock void 3491e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3492fa9e4066Sahrens { 3493e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3494e14bb325SJeff Bonwick 3495e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3496fa9e4066Sahrens 3497ea8dc4b6Seschrock if (vd == NULL) 3498e14bb325SJeff Bonwick vd = rvd; 3499fa9e4066Sahrens 3500ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3501ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3502ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3503dd50e0ccSTony Hutter vd->vdev_stat.vs_slow_ios = 0; 3504fa9e4066Sahrens 3505e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3506e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 35073d7072f8Seschrock 35085cabbc6bSPrashanth Sreenivasa /* 35095cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 35105cabbc6bSPrashanth Sreenivasa */ 35115cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 35125cabbc6bSPrashanth Sreenivasa return; 35135cabbc6bSPrashanth Sreenivasa 35143d7072f8Seschrock /* 35158a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 35168a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 35178a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 35188a79c1b5Sek * written out to disk. 35193d7072f8Seschrock */ 3520e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3521e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 35228a79c1b5Sek 3523096d22d4SEric Schrock /* 3524096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3525096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3526096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3527096d22d4SEric Schrock */ 3528096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3529096d22d4SEric Schrock 35304b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3531e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3532e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3533e14bb325SJeff Bonwick 3534f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 35353d7072f8Seschrock 3536096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3537096d22d4SEric Schrock 3538f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3539e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3540e14bb325SJeff Bonwick 3541*0c06d385Sjwpoduska /* If a resilver isn't required, check if vdevs can be culled */ 3542*0c06d385Sjwpoduska if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && 3543*0c06d385Sjwpoduska !dsl_scan_resilvering(spa->spa_dsl_pool) && 3544*0c06d385Sjwpoduska !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) 3545*0c06d385Sjwpoduska spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 35463d7072f8Seschrock 3547ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 35483d7072f8Seschrock } 3549096d22d4SEric Schrock 3550096d22d4SEric Schrock /* 3551096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3552096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3553096d22d4SEric Schrock * done in response to the FMA fault. 3554096d22d4SEric Schrock */ 3555096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3556096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3557096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3558096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3559fa9e4066Sahrens } 3560fa9e4066Sahrens 3561e14bb325SJeff Bonwick boolean_t 3562e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 35630a4e9518Sgw { 356488ecc943SGeorge Wilson /* 356588ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 356688ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 356788ecc943SGeorge Wilson * these types of devices in the various code paths. 356888ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 356988ecc943SGeorge Wilson * before issuing I/O to them. 357088ecc943SGeorge Wilson */ 35715cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 35725cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 357388ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 35740a4e9518Sgw } 35750a4e9518Sgw 3576e14bb325SJeff Bonwick boolean_t 3577e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 35780a4e9518Sgw { 3579e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 35800a4e9518Sgw } 35810a4e9518Sgw 3582e14bb325SJeff Bonwick boolean_t 3583e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3584fa9e4066Sahrens { 35855cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 35865cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3587fa9e4066Sahrens } 3588fa9e4066Sahrens 3589a31e6787SGeorge Wilson boolean_t 3590a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3591a31e6787SGeorge Wilson { 35928ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 35938ad4d6ddSJeff Bonwick 3594a31e6787SGeorge Wilson /* 35958ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3596a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3597a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 35988ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 35998ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 36008ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3601a31e6787SGeorge Wilson */ 36028ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 36035cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 36040f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3605a31e6787SGeorge Wilson } 3606a31e6787SGeorge Wilson 3607e14bb325SJeff Bonwick boolean_t 3608e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3609fa9e4066Sahrens { 3610e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3611fa9e4066Sahrens 3612e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3613e14bb325SJeff Bonwick return (B_FALSE); 3614fa9e4066Sahrens 3615e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3616e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3617fa9e4066Sahrens 3618e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3619e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3620fa9e4066Sahrens 3621e14bb325SJeff Bonwick return (B_TRUE); 3622fa9e4066Sahrens } 3623fa9e4066Sahrens 3624dd50e0ccSTony Hutter static void 3625dd50e0ccSTony Hutter vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) 3626dd50e0ccSTony Hutter { 3627dd50e0ccSTony Hutter for (int t = 0; t < VS_ZIO_TYPES; t++) { 3628dd50e0ccSTony Hutter vs->vs_ops[t] += cvs->vs_ops[t]; 3629dd50e0ccSTony Hutter vs->vs_bytes[t] += cvs->vs_bytes[t]; 3630dd50e0ccSTony Hutter } 3631dd50e0ccSTony Hutter 3632dd50e0ccSTony Hutter cvs->vs_scan_removing = cvd->vdev_removing; 3633dd50e0ccSTony Hutter } 3634dd50e0ccSTony Hutter 3635dd50e0ccSTony Hutter /* 3636dd50e0ccSTony Hutter * Get extended stats 3637dd50e0ccSTony Hutter */ 3638dd50e0ccSTony Hutter static void 3639dd50e0ccSTony Hutter vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) 3640dd50e0ccSTony Hutter { 3641dd50e0ccSTony Hutter int t, b; 3642dd50e0ccSTony Hutter for (t = 0; t < ZIO_TYPES; t++) { 3643dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) 3644dd50e0ccSTony Hutter vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; 3645dd50e0ccSTony Hutter 3646dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { 3647dd50e0ccSTony Hutter vsx->vsx_total_histo[t][b] += 3648dd50e0ccSTony Hutter cvsx->vsx_total_histo[t][b]; 3649dd50e0ccSTony Hutter } 3650dd50e0ccSTony Hutter } 3651dd50e0ccSTony Hutter 3652dd50e0ccSTony Hutter for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { 3653dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { 3654dd50e0ccSTony Hutter vsx->vsx_queue_histo[t][b] += 3655dd50e0ccSTony Hutter cvsx->vsx_queue_histo[t][b]; 3656dd50e0ccSTony Hutter } 3657dd50e0ccSTony Hutter vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; 3658dd50e0ccSTony Hutter vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; 3659dd50e0ccSTony Hutter 3660dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) 3661dd50e0ccSTony Hutter vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; 3662dd50e0ccSTony Hutter 3663dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) 3664dd50e0ccSTony Hutter vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; 3665dd50e0ccSTony Hutter } 3666dd50e0ccSTony Hutter 3667dd50e0ccSTony Hutter } 3668dd50e0ccSTony Hutter 366986714001SSerapheim Dimitropoulos boolean_t 367086714001SSerapheim Dimitropoulos vdev_is_spacemap_addressable(vdev_t *vd) 367186714001SSerapheim Dimitropoulos { 3672a0b03b16SSerapheim Dimitropoulos if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3673a0b03b16SSerapheim Dimitropoulos return (B_TRUE); 3674a0b03b16SSerapheim Dimitropoulos 367586714001SSerapheim Dimitropoulos /* 3676a0b03b16SSerapheim Dimitropoulos * If double-word space map entries are not enabled we assume 3677a0b03b16SSerapheim Dimitropoulos * 47 bits of the space map entry are dedicated to the entry's 3678a0b03b16SSerapheim Dimitropoulos * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3679a0b03b16SSerapheim Dimitropoulos * to calculate the maximum address that can be described by a 3680a0b03b16SSerapheim Dimitropoulos * space map entry for the given device. 368186714001SSerapheim Dimitropoulos */ 3682a0b03b16SSerapheim Dimitropoulos uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 368386714001SSerapheim Dimitropoulos 368486714001SSerapheim Dimitropoulos if (shift >= 63) /* detect potential overflow */ 368586714001SSerapheim Dimitropoulos return (B_TRUE); 368686714001SSerapheim Dimitropoulos 368786714001SSerapheim Dimitropoulos return (vd->vdev_asize < (1ULL << shift)); 368886714001SSerapheim Dimitropoulos } 368986714001SSerapheim Dimitropoulos 3690fa9e4066Sahrens /* 3691fa9e4066Sahrens * Get statistics for the given vdev. 3692fa9e4066Sahrens */ 3693dd50e0ccSTony Hutter static void 3694dd50e0ccSTony Hutter vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3695fa9e4066Sahrens { 3696dd50e0ccSTony Hutter int t; 3697dd50e0ccSTony Hutter /* 3698dd50e0ccSTony Hutter * If we're getting stats on the root vdev, aggregate the I/O counts 3699dd50e0ccSTony Hutter * over all top-level vdevs (i.e. the direct children of the root). 3700dd50e0ccSTony Hutter */ 3701dd50e0ccSTony Hutter if (!vd->vdev_ops->vdev_op_leaf) { 3702dd50e0ccSTony Hutter if (vs) { 3703dd50e0ccSTony Hutter memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); 3704dd50e0ccSTony Hutter memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); 3705dd50e0ccSTony Hutter } 3706dd50e0ccSTony Hutter if (vsx) 3707dd50e0ccSTony Hutter memset(vsx, 0, sizeof (*vsx)); 37082e4c9986SGeorge Wilson 3709dd50e0ccSTony Hutter for (int c = 0; c < vd->vdev_children; c++) { 3710dd50e0ccSTony Hutter vdev_t *cvd = vd->vdev_child[c]; 3711dd50e0ccSTony Hutter vdev_stat_t *cvs = &cvd->vdev_stat; 3712dd50e0ccSTony Hutter vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; 3713fa9e4066Sahrens 3714dd50e0ccSTony Hutter vdev_get_stats_ex_impl(cvd, cvs, cvsx); 3715dd50e0ccSTony Hutter if (vs) 3716dd50e0ccSTony Hutter vdev_get_child_stat(cvd, vs, cvs); 3717dd50e0ccSTony Hutter if (vsx) 3718dd50e0ccSTony Hutter vdev_get_child_stat_ex(cvd, vsx, cvsx); 3719dd50e0ccSTony Hutter 3720dd50e0ccSTony Hutter } 3721dd50e0ccSTony Hutter } else { 3722094e47e9SGeorge Wilson /* 3723dd50e0ccSTony Hutter * We're a leaf. Just copy our ZIO active queue stats in. The 3724dd50e0ccSTony Hutter * other leaf stats are updated in vdev_stat_update(). 3725094e47e9SGeorge Wilson */ 3726dd50e0ccSTony Hutter if (!vsx) 3727dd50e0ccSTony Hutter return; 3728084fd14fSBrian Behlendorf 3729dd50e0ccSTony Hutter memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); 3730084fd14fSBrian Behlendorf 3731dd50e0ccSTony Hutter for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { 3732dd50e0ccSTony Hutter vsx->vsx_active_queue[t] = 3733dd50e0ccSTony Hutter vd->vdev_queue.vq_class[t].vqc_active; 3734dd50e0ccSTony Hutter vsx->vsx_pend_queue[t] = avl_numnodes( 3735dd50e0ccSTony Hutter &vd->vdev_queue.vq_class[t].vqc_queued_tree); 3736dd50e0ccSTony Hutter } 37372986efa8SAlex Reece } 3738dd50e0ccSTony Hutter } 3739fa9e4066Sahrens 3740dd50e0ccSTony Hutter void 3741dd50e0ccSTony Hutter vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3742dd50e0ccSTony Hutter { 3743dd50e0ccSTony Hutter vdev_t *tvd = vd->vdev_top; 3744dd50e0ccSTony Hutter mutex_enter(&vd->vdev_stat_lock); 3745dd50e0ccSTony Hutter if (vs) { 3746dd50e0ccSTony Hutter bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3747dd50e0ccSTony Hutter vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3748dd50e0ccSTony Hutter vs->vs_state = vd->vdev_state; 3749dd50e0ccSTony Hutter vs->vs_rsize = vdev_get_min_asize(vd); 3750dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf) { 3751dd50e0ccSTony Hutter vs->vs_rsize += VDEV_LABEL_START_SIZE + 3752dd50e0ccSTony Hutter VDEV_LABEL_END_SIZE; 3753dd50e0ccSTony Hutter /* 3754dd50e0ccSTony Hutter * Report initializing progress. Since we don't 3755dd50e0ccSTony Hutter * have the initializing locks held, this is only 3756dd50e0ccSTony Hutter * an estimate (although a fairly accurate one). 3757dd50e0ccSTony Hutter */ 3758dd50e0ccSTony Hutter vs->vs_initialize_bytes_done = 3759dd50e0ccSTony Hutter vd->vdev_initialize_bytes_done; 3760dd50e0ccSTony Hutter vs->vs_initialize_bytes_est = 3761dd50e0ccSTony Hutter vd->vdev_initialize_bytes_est; 3762dd50e0ccSTony Hutter vs->vs_initialize_state = vd->vdev_initialize_state; 3763dd50e0ccSTony Hutter vs->vs_initialize_action_time = 3764dd50e0ccSTony Hutter vd->vdev_initialize_action_time; 3765fa9e4066Sahrens 3766dd50e0ccSTony Hutter /* 3767dd50e0ccSTony Hutter * Report manual TRIM progress. Since we don't have 3768dd50e0ccSTony Hutter * the manual TRIM locks held, this is only an 3769dd50e0ccSTony Hutter * estimate (although fairly accurate one). 3770dd50e0ccSTony Hutter */ 3771dd50e0ccSTony Hutter vs->vs_trim_notsup = !vd->vdev_has_trim; 3772dd50e0ccSTony Hutter vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 3773dd50e0ccSTony Hutter vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 3774dd50e0ccSTony Hutter vs->vs_trim_state = vd->vdev_trim_state; 3775dd50e0ccSTony Hutter vs->vs_trim_action_time = vd->vdev_trim_action_time; 3776dd50e0ccSTony Hutter } 3777dd50e0ccSTony Hutter /* 3778dd50e0ccSTony Hutter * Report expandable space on top-level, non-auxiliary devices 3779dd50e0ccSTony Hutter * only. The expandable space is reported in terms of metaslab 3780dd50e0ccSTony Hutter * sized units since that determines how much space the pool 3781dd50e0ccSTony Hutter * can expand. 3782dd50e0ccSTony Hutter */ 3783dd50e0ccSTony Hutter if (vd->vdev_aux == NULL && tvd != NULL) { 3784dd50e0ccSTony Hutter vs->vs_esize = P2ALIGN( 3785dd50e0ccSTony Hutter vd->vdev_max_asize - vd->vdev_asize, 3786dd50e0ccSTony Hutter 1ULL << tvd->vdev_ms_shift); 3787dd50e0ccSTony Hutter } 3788dd50e0ccSTony Hutter if (vd->vdev_aux == NULL && vd == vd->vdev_top && 3789dd50e0ccSTony Hutter vdev_is_concrete(vd)) { 3790dd50e0ccSTony Hutter vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3791dd50e0ccSTony Hutter vd->vdev_mg->mg_fragmentation : 0; 3792fa9e4066Sahrens } 3793dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf) 3794dd50e0ccSTony Hutter vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 3795fa9e4066Sahrens } 3796dd50e0ccSTony Hutter 3797dd50e0ccSTony Hutter vdev_get_stats_ex_impl(vd, vs, vsx); 37982e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3799fa9e4066Sahrens } 3800fa9e4066Sahrens 3801dd50e0ccSTony Hutter void 3802dd50e0ccSTony Hutter vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3803dd50e0ccSTony Hutter { 3804dd50e0ccSTony Hutter return (vdev_get_stats_ex(vd, vs, NULL)); 3805dd50e0ccSTony Hutter } 3806dd50e0ccSTony Hutter 3807fa94a07fSbrendan void 3808fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3809fa94a07fSbrendan { 3810fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3811fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3812fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3813fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3814fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3815fa94a07fSbrendan } 3816fa94a07fSbrendan 38173f9d6ad7SLin Ling void 38183f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 38193f9d6ad7SLin Ling { 38203f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 38213f9d6ad7SLin Ling 38223f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 38233f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 38243f9d6ad7SLin Ling 38253f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 38263f9d6ad7SLin Ling vs->vs_scan_processed = 0; 38273f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 38283f9d6ad7SLin Ling } 38293f9d6ad7SLin Ling 3830fa9e4066Sahrens void 3831e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3832fa9e4066Sahrens { 38338ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 38348ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3835e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3836fa9e4066Sahrens vdev_t *pvd; 3837fa9e4066Sahrens uint64_t txg = zio->io_txg; 3838fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3839dd50e0ccSTony Hutter vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; 3840fa9e4066Sahrens zio_type_t type = zio->io_type; 3841fa9e4066Sahrens int flags = zio->io_flags; 3842fa9e4066Sahrens 3843e14bb325SJeff Bonwick /* 3844e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3845e14bb325SJeff Bonwick */ 3846e14bb325SJeff Bonwick if (zio->io_gang_tree) 3847e14bb325SJeff Bonwick return; 3848e14bb325SJeff Bonwick 3849fa9e4066Sahrens if (zio->io_error == 0) { 3850e14bb325SJeff Bonwick /* 3851e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3852e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3853e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3854e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3855e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3856e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3857e14bb325SJeff Bonwick * remain zero, which is what we want.) 3858e14bb325SJeff Bonwick * 3859e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3860e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3861e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3862e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3863e14bb325SJeff Bonwick */ 3864e14bb325SJeff Bonwick if (vd == rvd) 3865e14bb325SJeff Bonwick return; 3866e14bb325SJeff Bonwick 3867e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 38688ad4d6ddSJeff Bonwick 38698ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 38708ad4d6ddSJeff Bonwick return; 38718ad4d6ddSJeff Bonwick 38728ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 38738ad4d6ddSJeff Bonwick 3874e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 387544ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 38763f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 38773f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 38783f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 38793f9d6ad7SLin Ling 38803f9d6ad7SLin Ling /* XXX cleanup? */ 38813f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 38823f9d6ad7SLin Ling atomic_add_64(processed, psize); 38833f9d6ad7SLin Ling vs->vs_scan_processed += psize; 38843f9d6ad7SLin Ling } 38853f9d6ad7SLin Ling 38868ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3887e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3888fa9e4066Sahrens } 38898ad4d6ddSJeff Bonwick 3890084fd14fSBrian Behlendorf /* 3891dd50e0ccSTony Hutter * The bytes/ops/histograms are recorded at the leaf level and 3892dd50e0ccSTony Hutter * aggregated into the higher level vdevs in vdev_get_stats(). 3893084fd14fSBrian Behlendorf */ 3894dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf && 3895dd50e0ccSTony Hutter (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { 3896dd50e0ccSTony Hutter zio_type_t vs_type = type; 3897dd50e0ccSTony Hutter 3898dd50e0ccSTony Hutter /* 3899dd50e0ccSTony Hutter * TRIM ops and bytes are reported to user space as 3900dd50e0ccSTony Hutter * ZIO_TYPE_IOCTL. This is done to preserve the 3901dd50e0ccSTony Hutter * vdev_stat_t structure layout for user space. 3902dd50e0ccSTony Hutter */ 3903dd50e0ccSTony Hutter if (type == ZIO_TYPE_TRIM) 3904dd50e0ccSTony Hutter vs_type = ZIO_TYPE_IOCTL; 3905dd50e0ccSTony Hutter 3906dd50e0ccSTony Hutter vs->vs_ops[vs_type]++; 3907dd50e0ccSTony Hutter vs->vs_bytes[vs_type] += psize; 3908dd50e0ccSTony Hutter 3909dd50e0ccSTony Hutter if (flags & ZIO_FLAG_DELEGATED) { 3910dd50e0ccSTony Hutter vsx->vsx_agg_histo[zio->io_priority] 3911dd50e0ccSTony Hutter [RQ_HISTO(zio->io_size)]++; 3912dd50e0ccSTony Hutter } else { 3913dd50e0ccSTony Hutter vsx->vsx_ind_histo[zio->io_priority] 3914dd50e0ccSTony Hutter [RQ_HISTO(zio->io_size)]++; 3915dd50e0ccSTony Hutter } 3916084fd14fSBrian Behlendorf 3917dd50e0ccSTony Hutter if (zio->io_delta && zio->io_delay) { 3918dd50e0ccSTony Hutter vsx->vsx_queue_histo[zio->io_priority] 3919dd50e0ccSTony Hutter [L_HISTO(zio->io_delta - zio->io_delay)]++; 3920dd50e0ccSTony Hutter vsx->vsx_disk_histo[type] 3921dd50e0ccSTony Hutter [L_HISTO(zio->io_delay)]++; 3922dd50e0ccSTony Hutter vsx->vsx_total_histo[type] 3923dd50e0ccSTony Hutter [L_HISTO(zio->io_delta)]++; 3924dd50e0ccSTony Hutter } 3925dd50e0ccSTony Hutter } 39268ad4d6ddSJeff Bonwick 39278ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3928fa9e4066Sahrens return; 3929fa9e4066Sahrens } 3930fa9e4066Sahrens 3931fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3932fa9e4066Sahrens return; 3933fa9e4066Sahrens 39348956713aSEric Schrock /* 39358956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 39368956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 39378956713aSEric Schrock * hard errors, when in reality they can happen for any number of 39388956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 39398956713aSEric Schrock */ 39408956713aSEric Schrock if (zio->io_error == EIO && 39418956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 39428956713aSEric Schrock return; 39438956713aSEric Schrock 39448f18d1faSGeorge Wilson /* 39458f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 39468f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 39478f18d1faSGeorge Wilson * errors. 39488f18d1faSGeorge Wilson */ 39498f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 39508f18d1faSGeorge Wilson return; 39518f18d1faSGeorge Wilson 3952e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3953b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3954e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3955e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3956e14bb325SJeff Bonwick else 3957e14bb325SJeff Bonwick vs->vs_read_errors++; 3958fa9e4066Sahrens } 3959b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3960e14bb325SJeff Bonwick vs->vs_write_errors++; 3961e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3962fa9e4066Sahrens 39635cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 39645cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 39658ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 396644ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3967b24ab676SJeff Bonwick spa->spa_claiming)) { 39688ad4d6ddSJeff Bonwick /* 3969b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 3970b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 3971b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 3972b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 3973b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 3974b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 3975b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 3976b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 39778ad4d6ddSJeff Bonwick * 39788ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 39798ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 39808ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 39818ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 39828ad4d6ddSJeff Bonwick */ 39838ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 39848ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 398544ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 39868ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 39878ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 39888ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3989b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 3990b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 3991b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3992b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 39938ad4d6ddSJeff Bonwick } 3994b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 39958ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3996fa9e4066Sahrens return; 39978ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 39988ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 39998ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 4000fa9e4066Sahrens } 40018ad4d6ddSJeff Bonwick if (vd != rvd) 40028ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 4003fa9e4066Sahrens } 4004fa9e4066Sahrens } 4005fa9e4066Sahrens 4006663207adSDon Brady int64_t 4007663207adSDon Brady vdev_deflated_space(vdev_t *vd, int64_t space) 4008663207adSDon Brady { 4009663207adSDon Brady ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 4010663207adSDon Brady ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 4011663207adSDon Brady 4012663207adSDon Brady return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 4013663207adSDon Brady } 4014663207adSDon Brady 4015fa9e4066Sahrens /* 4016084fd14fSBrian Behlendorf * Update the in-core space usage stats for this vdev, its metaslab class, 4017084fd14fSBrian Behlendorf * and the root vdev. 4018fa9e4066Sahrens */ 4019fa9e4066Sahrens void 4020b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 4021b24ab676SJeff Bonwick int64_t space_delta) 4022fa9e4066Sahrens { 4023663207adSDon Brady int64_t dspace_delta; 40248654d025Sperrin spa_t *spa = vd->vdev_spa; 40258654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 4026fa9e4066Sahrens 40278654d025Sperrin ASSERT(vd == vd->vdev_top); 402899653d4eSeschrock 40298654d025Sperrin /* 40308654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 40318654d025Sperrin * factor. We must calculate this here and not at the root vdev 40328654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 40338654d025Sperrin * childrens', thus not accurate enough for us. 40348654d025Sperrin */ 4035663207adSDon Brady dspace_delta = vdev_deflated_space(vd, space_delta); 40368654d025Sperrin 40378654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 40389740f25fSSerapheim Dimitropoulos /* ensure we won't underflow */ 40399740f25fSSerapheim Dimitropoulos if (alloc_delta < 0) { 40409740f25fSSerapheim Dimitropoulos ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 40419740f25fSSerapheim Dimitropoulos } 40429740f25fSSerapheim Dimitropoulos 40438654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 4044b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 40458654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 40468654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 40478654d025Sperrin 4048663207adSDon Brady /* every class but log contributes to root space stats */ 4049663207adSDon Brady if (vd->vdev_mg != NULL && !vd->vdev_islog) { 40509740f25fSSerapheim Dimitropoulos ASSERT(!vd->vdev_isl2cache); 4051fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 4052fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 4053b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 4054fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 4055fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 4056fa94a07fSbrendan } 4057663207adSDon Brady /* Note: metaslab_class_space_update moved to metaslab_space_update */ 4058fa9e4066Sahrens } 4059fa9e4066Sahrens 4060fa9e4066Sahrens /* 4061fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 4062fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 4063fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 4064fa9e4066Sahrens */ 4065fa9e4066Sahrens void 4066fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 4067fa9e4066Sahrens { 4068fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 4069fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4070fa9e4066Sahrens int c; 4071fa9e4066Sahrens 4072f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4073f9af39baSGeorge Wilson 4074c5904d13Seschrock /* 40756809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 40766809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 4077c5904d13Seschrock */ 4078c5904d13Seschrock if (vd->vdev_aux != NULL) { 4079c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 4080c5904d13Seschrock nvlist_t **aux; 4081c5904d13Seschrock uint_t naux; 4082c5904d13Seschrock 4083c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 4084c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 4085c5904d13Seschrock break; 4086c5904d13Seschrock } 4087c5904d13Seschrock 4088e14bb325SJeff Bonwick if (c == sav->sav_count) { 4089e14bb325SJeff Bonwick /* 4090e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 4091e14bb325SJeff Bonwick */ 4092e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 4093e14bb325SJeff Bonwick return; 4094e14bb325SJeff Bonwick } 4095e14bb325SJeff Bonwick 4096c5904d13Seschrock sav->sav_sync = B_TRUE; 4097c5904d13Seschrock 40986809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 40996809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 41006809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 41016809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 41026809eb4eSEric Schrock } 4103c5904d13Seschrock 4104c5904d13Seschrock ASSERT(c < naux); 4105c5904d13Seschrock 4106c5904d13Seschrock /* 4107c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 4108c5904d13Seschrock * sketchy, but it will work. 4109c5904d13Seschrock */ 4110c5904d13Seschrock nvlist_free(aux[c]); 41113f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 4112c5904d13Seschrock 4113c5904d13Seschrock return; 4114c5904d13Seschrock } 4115c5904d13Seschrock 41165dabedeeSbonwick /* 4117e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 4118e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 4119e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 41205dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 41215dabedeeSbonwick */ 4122e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4123e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4124e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 41255dabedeeSbonwick 4126fa9e4066Sahrens if (vd == rvd) { 4127fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 4128fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 4129fa9e4066Sahrens } else { 4130fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 4131fa9e4066Sahrens 413288ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 41335cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 4134e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 41355cabbc6bSPrashanth Sreenivasa } 4136fa9e4066Sahrens } 4137fa9e4066Sahrens } 4138fa9e4066Sahrens 4139fa9e4066Sahrens void 4140fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 4141fa9e4066Sahrens { 41425dabedeeSbonwick spa_t *spa = vd->vdev_spa; 41435dabedeeSbonwick 4144e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4145e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4146e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 41475dabedeeSbonwick 4148e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4149e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 4150e14bb325SJeff Bonwick } 4151e14bb325SJeff Bonwick 4152e14bb325SJeff Bonwick /* 4153e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 4154e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4155e14bb325SJeff Bonwick * the state changes from larger config changes because they require 4156e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 4157e14bb325SJeff Bonwick */ 4158e14bb325SJeff Bonwick void 4159e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 4160e14bb325SJeff Bonwick { 4161e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4162e14bb325SJeff Bonwick 4163f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4164e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 4165e14bb325SJeff Bonwick 4166e14bb325SJeff Bonwick /* 4167e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 4168e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 4169e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 4170e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 4171e14bb325SJeff Bonwick */ 4172e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4173e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4174e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4175e14bb325SJeff Bonwick 41765cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 41775cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 4178e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 4179e14bb325SJeff Bonwick } 4180e14bb325SJeff Bonwick 4181e14bb325SJeff Bonwick void 4182e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 4183e14bb325SJeff Bonwick { 4184e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4185e14bb325SJeff Bonwick 4186e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4187e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4188e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4189e14bb325SJeff Bonwick 4190e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 4191e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 4192fa9e4066Sahrens } 4193fa9e4066Sahrens 419432b87932Sek /* 419532b87932Sek * Propagate vdev state up from children to parent. 419632b87932Sek */ 419744cd46caSbillm void 419844cd46caSbillm vdev_propagate_state(vdev_t *vd) 419944cd46caSbillm { 42008ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 42018ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 420244cd46caSbillm int degraded = 0, faulted = 0; 420344cd46caSbillm int corrupted = 0; 420444cd46caSbillm vdev_t *child; 420544cd46caSbillm 42063d7072f8Seschrock if (vd->vdev_children > 0) { 4207573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 42083d7072f8Seschrock child = vd->vdev_child[c]; 420951ece835Seschrock 421088ecc943SGeorge Wilson /* 42115cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 42125cabbc6bSPrashanth Sreenivasa * decision. 421388ecc943SGeorge Wilson */ 42145cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 421588ecc943SGeorge Wilson continue; 421688ecc943SGeorge Wilson 4217e14bb325SJeff Bonwick if (!vdev_readable(child) || 42188ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 421951ece835Seschrock /* 422051ece835Seschrock * Root special: if there is a top-level log 422151ece835Seschrock * device, treat the root vdev as if it were 422251ece835Seschrock * degraded. 422351ece835Seschrock */ 422451ece835Seschrock if (child->vdev_islog && vd == rvd) 422551ece835Seschrock degraded++; 422651ece835Seschrock else 422751ece835Seschrock faulted++; 422851ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 42293d7072f8Seschrock degraded++; 423051ece835Seschrock } 423144cd46caSbillm 42323d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 42333d7072f8Seschrock corrupted++; 42343d7072f8Seschrock } 423544cd46caSbillm 42363d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 42373d7072f8Seschrock 42383d7072f8Seschrock /* 4239e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 42403d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 42413d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 42423d7072f8Seschrock * replicas'. 42433d7072f8Seschrock */ 42443d7072f8Seschrock if (corrupted && vd == rvd && 42453d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 42463d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 42473d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 42483d7072f8Seschrock } 42493d7072f8Seschrock 425051ece835Seschrock if (vd->vdev_parent) 42513d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 425244cd46caSbillm } 425344cd46caSbillm 4254fa9e4066Sahrens /* 4255ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 4256ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 4257ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 4258ea8dc4b6Seschrock * 4259ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 4260ea8dc4b6Seschrock * generated. 4261fa9e4066Sahrens */ 4262fa9e4066Sahrens void 4263ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4264fa9e4066Sahrens { 4265560e6e96Seschrock uint64_t save_state; 4266c5904d13Seschrock spa_t *spa = vd->vdev_spa; 4267ea8dc4b6Seschrock 4268ea8dc4b6Seschrock if (state == vd->vdev_state) { 4269ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 4270fa9e4066Sahrens return; 4271ea8dc4b6Seschrock } 4272ea8dc4b6Seschrock 4273560e6e96Seschrock save_state = vd->vdev_state; 4274fa9e4066Sahrens 4275fa9e4066Sahrens vd->vdev_state = state; 4276fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 4277fa9e4066Sahrens 42783d7072f8Seschrock /* 42793d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 428098d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 428198d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 428298d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 428398d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 428498d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 428598d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 428698d1cbfeSGeorge Wilson * children. 428798d1cbfeSGeorge Wilson */ 428898d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 428998d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 42903d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 42913d7072f8Seschrock 4292069f55e2SEric Schrock /* 4293069f55e2SEric Schrock * If we have brought this vdev back into service, we need 4294069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 4295069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 4296069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 4297069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 4298069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 4299069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 4300069f55e2SEric Schrock */ 4301069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4302069f55e2SEric Schrock vd->vdev_prevstate != state) 4303069f55e2SEric Schrock zfs_post_state_change(spa, vd); 4304069f55e2SEric Schrock 43053d7072f8Seschrock if (vd->vdev_removed && 43063d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 43073d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 43083d7072f8Seschrock /* 43093d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 43103d7072f8Seschrock * device was previously marked removed and someone attempted to 43113d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 43123d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 43133d7072f8Seschrock * it is one of our special test online cases, which is only 43143d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 43153d7072f8Seschrock * fault. 43163d7072f8Seschrock */ 43173d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 43183d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 43193d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 43203d7072f8Seschrock vd->vdev_removed = B_TRUE; 43213d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 4322ea8dc4b6Seschrock /* 4323cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 4324cb04b873SMark J Musante * mark it as "not available", which signifies that it was 4325cb04b873SMark J Musante * never there to begin with. Failure to open such a device 4326cb04b873SMark J Musante * is not considered an error. 4327ea8dc4b6Seschrock */ 4328cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4329cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 4330560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 4331560e6e96Seschrock vd->vdev_not_present = 1; 4332560e6e96Seschrock 4333560e6e96Seschrock /* 4334560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 4335560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 4336560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 4337560e6e96Seschrock * want to post the ereport if the device was already in the 4338560e6e96Seschrock * CANT_OPEN state beforehand. 43393d7072f8Seschrock * 43403d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 43413d7072f8Seschrock * online the device in response to an insertion event. If we 43423d7072f8Seschrock * hit this case, then we have detected an insertion event for a 43433d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 43443d7072f8Seschrock * In this scenario, we don't post an ereport because we are 43453d7072f8Seschrock * about to replace the device, or attempt an online with 43463d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 4347560e6e96Seschrock */ 43483d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 43493d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 4350c5904d13Seschrock vd != spa->spa_root_vdev) { 4351ea8dc4b6Seschrock const char *class; 4352ea8dc4b6Seschrock 4353ea8dc4b6Seschrock switch (aux) { 4354ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 4355ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4356ea8dc4b6Seschrock break; 4357ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 4358ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4359ea8dc4b6Seschrock break; 4360ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 4361ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4362ea8dc4b6Seschrock break; 4363ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 4364ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4365ea8dc4b6Seschrock break; 4366ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 4367ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4368ea8dc4b6Seschrock break; 4369ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 4370ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4371ea8dc4b6Seschrock break; 43725711d393Sloli case VDEV_AUX_BAD_ASHIFT: 43735711d393Sloli class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 43745711d393Sloli break; 4375ea8dc4b6Seschrock default: 4376ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4377ea8dc4b6Seschrock } 4378ea8dc4b6Seschrock 4379eb633035STom Caputi zfs_ereport_post(class, spa, vd, NULL, NULL, 4380eb633035STom Caputi save_state, 0); 4381ea8dc4b6Seschrock } 4382ea8dc4b6Seschrock 43833d7072f8Seschrock /* Erase any notion of persistent removed state */ 43843d7072f8Seschrock vd->vdev_removed = B_FALSE; 43853d7072f8Seschrock } else { 43863d7072f8Seschrock vd->vdev_removed = B_FALSE; 43873d7072f8Seschrock } 4388ea8dc4b6Seschrock 43898b33d774STim Haley if (!isopen && vd->vdev_parent) 43908b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 4391fa9e4066Sahrens } 439215e6edf1Sgw 43936f793812SPavel Zakharov boolean_t 43946f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 43956f793812SPavel Zakharov { 43966f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 43976f793812SPavel Zakharov 43986f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 43996f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 44006f793812SPavel Zakharov return (B_FALSE); 44016f793812SPavel Zakharov } 44026f793812SPavel Zakharov 44036f793812SPavel Zakharov return (B_TRUE); 44046f793812SPavel Zakharov } 44056f793812SPavel Zakharov 440615e6edf1Sgw /* 440715e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 4408c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 4409c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 441015e6edf1Sgw */ 441115e6edf1Sgw boolean_t 441215e6edf1Sgw vdev_is_bootable(vdev_t *vd) 441315e6edf1Sgw { 441415e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 441515e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 441615e6edf1Sgw 441715e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 441815e6edf1Sgw vd->vdev_children > 1) { 441921c878feSToomas Soome int non_indirect = 0; 442021c878feSToomas Soome 442121c878feSToomas Soome for (int c = 0; c < vd->vdev_children; c++) { 442221c878feSToomas Soome vdev_type = 442321c878feSToomas Soome vd->vdev_child[c]->vdev_ops->vdev_op_type; 442421c878feSToomas Soome if (strcmp(vdev_type, VDEV_TYPE_INDIRECT) != 0) 442521c878feSToomas Soome non_indirect++; 442621c878feSToomas Soome } 442721c878feSToomas Soome /* 442821c878feSToomas Soome * non_indirect > 1 means we have more than one 442921c878feSToomas Soome * top-level vdev, so we stop here. 443021c878feSToomas Soome */ 443121c878feSToomas Soome if (non_indirect > 1) 443221c878feSToomas Soome return (B_FALSE); 443321c878feSToomas Soome } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 443415e6edf1Sgw return (B_FALSE); 443515e6edf1Sgw } 443615e6edf1Sgw } 443715e6edf1Sgw 4438573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 443915e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 444015e6edf1Sgw return (B_FALSE); 444115e6edf1Sgw } 444215e6edf1Sgw return (B_TRUE); 444315e6edf1Sgw } 4444e6ca193dSGeorge Wilson 44455cabbc6bSPrashanth Sreenivasa boolean_t 44465cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 44475cabbc6bSPrashanth Sreenivasa { 44485cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 44495cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 44505cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 44515cabbc6bSPrashanth Sreenivasa return (B_FALSE); 44525cabbc6bSPrashanth Sreenivasa } else { 44535cabbc6bSPrashanth Sreenivasa return (B_TRUE); 44545cabbc6bSPrashanth Sreenivasa } 44555cabbc6bSPrashanth Sreenivasa } 44565cabbc6bSPrashanth Sreenivasa 44574b964adaSGeorge Wilson /* 44584b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 44594b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 44604b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 44614b964adaSGeorge Wilson */ 44624b964adaSGeorge Wilson boolean_t 44634b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 44644b964adaSGeorge Wilson { 44654b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 44664b964adaSGeorge Wilson !vd->vdev_removed) 44674b964adaSGeorge Wilson return (B_TRUE); 44684b964adaSGeorge Wilson 44694b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 44704b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 44714b964adaSGeorge Wilson return (B_TRUE); 44724b964adaSGeorge Wilson 44734b964adaSGeorge Wilson return (B_FALSE); 44744b964adaSGeorge Wilson } 44754b964adaSGeorge Wilson 4476573ca77eSGeorge Wilson /* 4477573ca77eSGeorge Wilson * Expand a vdev if possible. 4478573ca77eSGeorge Wilson */ 4479573ca77eSGeorge Wilson void 4480573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 4481573ca77eSGeorge Wilson { 4482573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 4483573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 448411f6a968SSerapheim Dimitropoulos ASSERT(vdev_is_concrete(vd)); 4485573ca77eSGeorge Wilson 44865cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 44875cabbc6bSPrashanth Sreenivasa 4488663207adSDon Brady if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4489663207adSDon Brady vdev_is_concrete(vd)) { 4490663207adSDon Brady vdev_metaslab_group_create(vd); 4491573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 4492573ca77eSGeorge Wilson vdev_config_dirty(vd); 4493573ca77eSGeorge Wilson } 4494573ca77eSGeorge Wilson } 44951195e687SMark J Musante 44961195e687SMark J Musante /* 44971195e687SMark J Musante * Split a vdev. 44981195e687SMark J Musante */ 44991195e687SMark J Musante void 45001195e687SMark J Musante vdev_split(vdev_t *vd) 45011195e687SMark J Musante { 45021195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 45031195e687SMark J Musante 45041195e687SMark J Musante vdev_remove_child(pvd, vd); 45051195e687SMark J Musante vdev_compact_children(pvd); 45061195e687SMark J Musante 45071195e687SMark J Musante cvd = pvd->vdev_child[0]; 45081195e687SMark J Musante if (pvd->vdev_children == 1) { 45091195e687SMark J Musante vdev_remove_parent(cvd); 45101195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 45111195e687SMark J Musante } 45121195e687SMark J Musante vdev_propagate_state(cvd); 45131195e687SMark J Musante } 4514283b8460SGeorge.Wilson 4515283b8460SGeorge.Wilson void 4516283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 4517283b8460SGeorge.Wilson { 4518283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 4519283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 4520283b8460SGeorge.Wilson 4521283b8460SGeorge.Wilson vdev_deadman(cvd); 4522283b8460SGeorge.Wilson } 4523283b8460SGeorge.Wilson 4524283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 4525283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 4526283b8460SGeorge.Wilson 4527283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 452869962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 4529283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 4530283b8460SGeorge.Wilson zio_t *fio; 4531283b8460SGeorge.Wilson uint64_t delta; 4532283b8460SGeorge.Wilson 4533283b8460SGeorge.Wilson /* 4534283b8460SGeorge.Wilson * Look at the head of all the pending queues, 4535283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 4536283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 4537283b8460SGeorge.Wilson */ 453869962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 4539c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 4540c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 45413ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 45423ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 45433ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 4544283b8460SGeorge.Wilson vq->vq_io_complete_ts); 4545283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 4546283b8460SGeorge.Wilson "hung.", spa_name(spa)); 4547283b8460SGeorge.Wilson } 4548283b8460SGeorge.Wilson } 4549283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 4550283b8460SGeorge.Wilson } 4551283b8460SGeorge.Wilson } 4552e4c795beSTom Caputi 4553e4c795beSTom Caputi void 4554*0c06d385Sjwpoduska vdev_defer_resilver(vdev_t *vd) 4555e4c795beSTom Caputi { 4556*0c06d385Sjwpoduska ASSERT(vd->vdev_ops->vdev_op_leaf); 4557e4c795beSTom Caputi 4558*0c06d385Sjwpoduska vd->vdev_resilver_deferred = B_TRUE; 4559*0c06d385Sjwpoduska vd->vdev_spa->spa_resilver_deferred = B_TRUE; 4560*0c06d385Sjwpoduska } 4561*0c06d385Sjwpoduska 4562*0c06d385Sjwpoduska /* 4563*0c06d385Sjwpoduska * Clears the resilver deferred flag on all leaf devs under vd. Returns 4564*0c06d385Sjwpoduska * B_TRUE if we have devices that need to be resilvered and are available to 4565*0c06d385Sjwpoduska * accept resilver I/Os. 4566*0c06d385Sjwpoduska */ 4567*0c06d385Sjwpoduska boolean_t 4568*0c06d385Sjwpoduska vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) 4569*0c06d385Sjwpoduska { 4570*0c06d385Sjwpoduska boolean_t resilver_needed = B_FALSE; 4571*0c06d385Sjwpoduska spa_t *spa = vd->vdev_spa; 4572*0c06d385Sjwpoduska 4573*0c06d385Sjwpoduska for (int c = 0; c < vd->vdev_children; c++) { 4574*0c06d385Sjwpoduska vdev_t *cvd = vd->vdev_child[c]; 4575*0c06d385Sjwpoduska resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); 4576e4c795beSTom Caputi } 4577e4c795beSTom Caputi 4578*0c06d385Sjwpoduska if (vd == spa->spa_root_vdev && 4579*0c06d385Sjwpoduska spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { 4580*0c06d385Sjwpoduska spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); 4581*0c06d385Sjwpoduska vdev_config_dirty(vd); 4582*0c06d385Sjwpoduska spa->spa_resilver_deferred = B_FALSE; 4583*0c06d385Sjwpoduska return (resilver_needed); 4584*0c06d385Sjwpoduska } 4585*0c06d385Sjwpoduska 4586*0c06d385Sjwpoduska if (!vdev_is_concrete(vd) || vd->vdev_aux || 4587*0c06d385Sjwpoduska !vd->vdev_ops->vdev_op_leaf) 4588*0c06d385Sjwpoduska return (resilver_needed); 4589*0c06d385Sjwpoduska 4590*0c06d385Sjwpoduska vd->vdev_resilver_deferred = B_FALSE; 4591*0c06d385Sjwpoduska 4592*0c06d385Sjwpoduska return (!vdev_is_dead(vd) && !vd->vdev_offline && 4593*0c06d385Sjwpoduska vdev_resilver_needed(vd, NULL, NULL)); 4594e4c795beSTom Caputi } 4595084fd14fSBrian Behlendorf 4596084fd14fSBrian Behlendorf /* 4597084fd14fSBrian Behlendorf * Translate a logical range to the physical range for the specified vdev_t. 4598084fd14fSBrian Behlendorf * This function is initially called with a leaf vdev and will walk each 4599084fd14fSBrian Behlendorf * parent vdev until it reaches a top-level vdev. Once the top-level is 4600084fd14fSBrian Behlendorf * reached the physical range is initialized and the recursive function 4601084fd14fSBrian Behlendorf * begins to unwind. As it unwinds it calls the parent's vdev specific 4602084fd14fSBrian Behlendorf * translation function to do the real conversion. 4603084fd14fSBrian Behlendorf */ 4604084fd14fSBrian Behlendorf void 46054d7988d6SPaul Dagnelie vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 46064d7988d6SPaul Dagnelie range_seg64_t *physical_rs) 4607084fd14fSBrian Behlendorf { 4608084fd14fSBrian Behlendorf /* 4609084fd14fSBrian Behlendorf * Walk up the vdev tree 4610084fd14fSBrian Behlendorf */ 4611084fd14fSBrian Behlendorf if (vd != vd->vdev_top) { 4612084fd14fSBrian Behlendorf vdev_xlate(vd->vdev_parent, logical_rs, physical_rs); 4613084fd14fSBrian Behlendorf } else { 4614084fd14fSBrian Behlendorf /* 4615084fd14fSBrian Behlendorf * We've reached the top-level vdev, initialize the 4616084fd14fSBrian Behlendorf * physical range to the logical range and start to 4617084fd14fSBrian Behlendorf * unwind. 4618084fd14fSBrian Behlendorf */ 4619084fd14fSBrian Behlendorf physical_rs->rs_start = logical_rs->rs_start; 4620084fd14fSBrian Behlendorf physical_rs->rs_end = logical_rs->rs_end; 4621084fd14fSBrian Behlendorf return; 4622084fd14fSBrian Behlendorf } 4623084fd14fSBrian Behlendorf 4624084fd14fSBrian Behlendorf vdev_t *pvd = vd->vdev_parent; 4625084fd14fSBrian Behlendorf ASSERT3P(pvd, !=, NULL); 4626084fd14fSBrian Behlendorf ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 4627084fd14fSBrian Behlendorf 4628084fd14fSBrian Behlendorf /* 4629084fd14fSBrian Behlendorf * As this recursive function unwinds, translate the logical 4630084fd14fSBrian Behlendorf * range into its physical components by calling the 4631084fd14fSBrian Behlendorf * vdev specific translate function. 4632084fd14fSBrian Behlendorf */ 46334d7988d6SPaul Dagnelie range_seg64_t intermediate = { 0 }; 4634084fd14fSBrian Behlendorf pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate); 4635084fd14fSBrian Behlendorf 4636084fd14fSBrian Behlendorf physical_rs->rs_start = intermediate.rs_start; 4637084fd14fSBrian Behlendorf physical_rs->rs_end = intermediate.rs_end; 4638084fd14fSBrian Behlendorf } 4639