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. 300c06d385Sjwpoduska * 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 101*165c5c6fSJohn Poduska /* 102*165c5c6fSJohn Poduska * Ignore errors during scrub/resilver. Allows to work around resilver 103*165c5c6fSJohn Poduska * upon import when there are pool errors. 104*165c5c6fSJohn Poduska */ 105*165c5c6fSJohn Poduska int zfs_scan_ignore_errors = 0; 106*165c5c6fSJohn Poduska 10786714001SSerapheim Dimitropoulos /* 10886714001SSerapheim Dimitropoulos * vdev-wide space maps that have lots of entries written to them at 10986714001SSerapheim Dimitropoulos * the end of each transaction can benefit from a higher I/O bandwidth 11086714001SSerapheim Dimitropoulos * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 11186714001SSerapheim Dimitropoulos */ 112814dcd43SSerapheim Dimitropoulos int zfs_vdev_standard_sm_blksz = (1 << 17); 1136f793812SPavel Zakharov 11493a1902eSMatthew Ahrens int zfs_ashift_min; 11593a1902eSMatthew Ahrens 1163ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 1173ee8c80cSPavel Zakharov void 1183ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 1193ee8c80cSPavel Zakharov { 1203ee8c80cSPavel Zakharov va_list adx; 1213ee8c80cSPavel Zakharov char buf[256]; 1223ee8c80cSPavel Zakharov 1233ee8c80cSPavel Zakharov va_start(adx, fmt); 1243ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 1253ee8c80cSPavel Zakharov va_end(adx); 1263ee8c80cSPavel Zakharov 1273ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 1283ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 1293ee8c80cSPavel Zakharov vd->vdev_path, buf); 1303ee8c80cSPavel Zakharov } else { 1313ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 1323ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 1333ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1343ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1353ee8c80cSPavel Zakharov } 1363ee8c80cSPavel Zakharov } 1373ee8c80cSPavel Zakharov 1386f793812SPavel Zakharov void 1396f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 1406f793812SPavel Zakharov { 1416f793812SPavel Zakharov char state[20]; 1426f793812SPavel Zakharov 1436f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 1446f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 1456f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 1466f793812SPavel Zakharov return; 1476f793812SPavel Zakharov } 1486f793812SPavel Zakharov 1496f793812SPavel Zakharov switch (vd->vdev_state) { 1506f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 1516f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 1526f793812SPavel Zakharov break; 1536f793812SPavel Zakharov case VDEV_STATE_CLOSED: 1546f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 1556f793812SPavel Zakharov break; 1566f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 1576f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 1586f793812SPavel Zakharov break; 1596f793812SPavel Zakharov case VDEV_STATE_REMOVED: 1606f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 1616f793812SPavel Zakharov break; 1626f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 1636f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 1646f793812SPavel Zakharov break; 1656f793812SPavel Zakharov case VDEV_STATE_FAULTED: 1666f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 1676f793812SPavel Zakharov break; 1686f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 1696f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 1706f793812SPavel Zakharov break; 1716f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 1726f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 1736f793812SPavel Zakharov break; 1746f793812SPavel Zakharov default: 1756f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 1766f793812SPavel Zakharov (uint_t)vd->vdev_state); 1776f793812SPavel Zakharov } 1786f793812SPavel Zakharov 1796f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 180c7a7b2faSAndriy Gapon "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 1816f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 1826f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1836f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 1846f793812SPavel Zakharov 1856f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 1866f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 1876f793812SPavel Zakharov } 1886f793812SPavel Zakharov 189fa9e4066Sahrens /* 190fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 191fa9e4066Sahrens */ 192fa9e4066Sahrens static vdev_ops_t * 193fa9e4066Sahrens vdev_getops(const char *type) 194fa9e4066Sahrens { 195fa9e4066Sahrens vdev_ops_t *ops, **opspp; 196fa9e4066Sahrens 197fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 198fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 199fa9e4066Sahrens break; 200fa9e4066Sahrens 201fa9e4066Sahrens return (ops); 202fa9e4066Sahrens } 203fa9e4066Sahrens 204663207adSDon Brady /* 205663207adSDon Brady * Derive the enumerated alloction bias from string input. 206663207adSDon Brady * String origin is either the per-vdev zap or zpool(1M). 207663207adSDon Brady */ 208663207adSDon Brady static vdev_alloc_bias_t 209663207adSDon Brady vdev_derive_alloc_bias(const char *bias) 210663207adSDon Brady { 211663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 212663207adSDon Brady 213663207adSDon Brady if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) 214663207adSDon Brady alloc_bias = VDEV_BIAS_LOG; 215663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) 216663207adSDon Brady alloc_bias = VDEV_BIAS_SPECIAL; 217663207adSDon Brady else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) 218663207adSDon Brady alloc_bias = VDEV_BIAS_DEDUP; 219663207adSDon Brady 220663207adSDon Brady return (alloc_bias); 221663207adSDon Brady } 222663207adSDon Brady 223094e47e9SGeorge Wilson /* ARGSUSED */ 224094e47e9SGeorge Wilson void 2254d7988d6SPaul Dagnelie vdev_default_xlate(vdev_t *vd, const range_seg64_t *in, range_seg64_t *res) 226094e47e9SGeorge Wilson { 227094e47e9SGeorge Wilson res->rs_start = in->rs_start; 228094e47e9SGeorge Wilson res->rs_end = in->rs_end; 229094e47e9SGeorge Wilson } 230094e47e9SGeorge Wilson 231fa9e4066Sahrens /* 232fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 233fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 234fa9e4066Sahrens */ 235fa9e4066Sahrens uint64_t 236fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 237fa9e4066Sahrens { 238ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 239fa9e4066Sahrens uint64_t csize; 240fa9e4066Sahrens 241573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 242fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 243fa9e4066Sahrens asize = MAX(asize, csize); 244fa9e4066Sahrens } 245fa9e4066Sahrens 246fa9e4066Sahrens return (asize); 247fa9e4066Sahrens } 248fa9e4066Sahrens 2492a79c5feSlling /* 250573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 251573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 252573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 253573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 2542a79c5feSlling */ 2552a79c5feSlling uint64_t 256573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 2572a79c5feSlling { 258573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 2592a79c5feSlling 260573ca77eSGeorge Wilson /* 2614263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 262573ca77eSGeorge Wilson * just return our own asize. 263573ca77eSGeorge Wilson */ 264573ca77eSGeorge Wilson if (pvd == NULL) 265573ca77eSGeorge Wilson return (vd->vdev_asize); 2662a79c5feSlling 2672a79c5feSlling /* 268573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 269573ca77eSGeorge Wilson * to the nearest metaslab. 2702a79c5feSlling */ 271573ca77eSGeorge Wilson if (vd == vd->vdev_top) 272573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2732a79c5feSlling 274573ca77eSGeorge Wilson /* 275573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 276573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 277573ca77eSGeorge Wilson */ 278573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 279c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 280c040c10cSSteven Hartland pvd->vdev_children); 2812a79c5feSlling 282573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 283573ca77eSGeorge Wilson } 2842a79c5feSlling 285573ca77eSGeorge Wilson void 286573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 287573ca77eSGeorge Wilson { 288573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 289573ca77eSGeorge Wilson 290573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 291573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2922a79c5feSlling } 2932a79c5feSlling 294fa9e4066Sahrens vdev_t * 295fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 296fa9e4066Sahrens { 297fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 298fa9e4066Sahrens 299e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 300e05725b1Sbonwick 301088f3894Sahrens if (vdev < rvd->vdev_children) { 302088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 303fa9e4066Sahrens return (rvd->vdev_child[vdev]); 304088f3894Sahrens } 305fa9e4066Sahrens 306fa9e4066Sahrens return (NULL); 307fa9e4066Sahrens } 308fa9e4066Sahrens 309fa9e4066Sahrens vdev_t * 310fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 311fa9e4066Sahrens { 312fa9e4066Sahrens vdev_t *mvd; 313fa9e4066Sahrens 3140e34b6a7Sbonwick if (vd->vdev_guid == guid) 315fa9e4066Sahrens return (vd); 316fa9e4066Sahrens 317573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 318fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 319fa9e4066Sahrens NULL) 320fa9e4066Sahrens return (mvd); 321fa9e4066Sahrens 322fa9e4066Sahrens return (NULL); 323fa9e4066Sahrens } 324fa9e4066Sahrens 32512380e1eSArne Jansen static int 32612380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 32712380e1eSArne Jansen { 32812380e1eSArne Jansen int n = 0; 32912380e1eSArne Jansen 33012380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 33112380e1eSArne Jansen return (1); 33212380e1eSArne Jansen 33312380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 33412380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 33512380e1eSArne Jansen 33612380e1eSArne Jansen return (n); 33712380e1eSArne Jansen } 33812380e1eSArne Jansen 33912380e1eSArne Jansen int 34012380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 34112380e1eSArne Jansen { 34212380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 34312380e1eSArne Jansen } 34412380e1eSArne Jansen 345fa9e4066Sahrens void 346fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 347fa9e4066Sahrens { 348fa9e4066Sahrens size_t oldsize, newsize; 349fa9e4066Sahrens uint64_t id = cvd->vdev_id; 350fa9e4066Sahrens vdev_t **newchild; 35181cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 352fa9e4066Sahrens 35381cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 354fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 355fa9e4066Sahrens 356fa9e4066Sahrens cvd->vdev_parent = pvd; 357fa9e4066Sahrens 358fa9e4066Sahrens if (pvd == NULL) 359fa9e4066Sahrens return; 360fa9e4066Sahrens 361fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 362fa9e4066Sahrens 363fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 364fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 365fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 366fa9e4066Sahrens 367fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 368fa9e4066Sahrens if (pvd->vdev_child != NULL) { 369fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 370fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 371fa9e4066Sahrens } 372fa9e4066Sahrens 373fa9e4066Sahrens pvd->vdev_child = newchild; 374fa9e4066Sahrens pvd->vdev_child[id] = cvd; 375fa9e4066Sahrens 376fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 377fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 378fa9e4066Sahrens 379fa9e4066Sahrens /* 380fa9e4066Sahrens * Walk up all ancestors to update guid sum. 381fa9e4066Sahrens */ 382fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 383fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 384e0f1c0afSOlaf Faaland 385e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 386e0f1c0afSOlaf Faaland list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 387e0f1c0afSOlaf Faaland cvd->vdev_spa->spa_leaf_list_gen++; 388e0f1c0afSOlaf Faaland } 389fa9e4066Sahrens } 390fa9e4066Sahrens 391fa9e4066Sahrens void 392fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 393fa9e4066Sahrens { 394fa9e4066Sahrens int c; 395fa9e4066Sahrens uint_t id = cvd->vdev_id; 396fa9e4066Sahrens 397fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 398fa9e4066Sahrens 399fa9e4066Sahrens if (pvd == NULL) 400fa9e4066Sahrens return; 401fa9e4066Sahrens 402fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 403fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 404fa9e4066Sahrens 405fa9e4066Sahrens pvd->vdev_child[id] = NULL; 406fa9e4066Sahrens cvd->vdev_parent = NULL; 407fa9e4066Sahrens 408fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 409fa9e4066Sahrens if (pvd->vdev_child[c]) 410fa9e4066Sahrens break; 411fa9e4066Sahrens 412fa9e4066Sahrens if (c == pvd->vdev_children) { 413fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 414fa9e4066Sahrens pvd->vdev_child = NULL; 415fa9e4066Sahrens pvd->vdev_children = 0; 416fa9e4066Sahrens } 417fa9e4066Sahrens 418e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 419e0f1c0afSOlaf Faaland spa_t *spa = cvd->vdev_spa; 420e0f1c0afSOlaf Faaland list_remove(&spa->spa_leaf_list, cvd); 421e0f1c0afSOlaf Faaland spa->spa_leaf_list_gen++; 422e0f1c0afSOlaf Faaland } 423e0f1c0afSOlaf Faaland 424fa9e4066Sahrens /* 425fa9e4066Sahrens * Walk up all ancestors to update guid sum. 426fa9e4066Sahrens */ 427fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 428fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 429fa9e4066Sahrens } 430fa9e4066Sahrens 431fa9e4066Sahrens /* 432fa9e4066Sahrens * Remove any holes in the child array. 433fa9e4066Sahrens */ 434fa9e4066Sahrens void 435fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 436fa9e4066Sahrens { 437fa9e4066Sahrens vdev_t **newchild, *cvd; 438fa9e4066Sahrens int oldc = pvd->vdev_children; 439573ca77eSGeorge Wilson int newc; 440fa9e4066Sahrens 441e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 442fa9e4066Sahrens 443573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 444fa9e4066Sahrens if (pvd->vdev_child[c]) 445fa9e4066Sahrens newc++; 446fa9e4066Sahrens 447fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 448fa9e4066Sahrens 449573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 450fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 451fa9e4066Sahrens newchild[newc] = cvd; 452fa9e4066Sahrens cvd->vdev_id = newc++; 453fa9e4066Sahrens } 454fa9e4066Sahrens } 455fa9e4066Sahrens 456fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 457fa9e4066Sahrens pvd->vdev_child = newchild; 458fa9e4066Sahrens pvd->vdev_children = newc; 459fa9e4066Sahrens } 460fa9e4066Sahrens 461fa9e4066Sahrens /* 462fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 463fa9e4066Sahrens */ 46488ecc943SGeorge Wilson vdev_t * 465fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 466fa9e4066Sahrens { 467fa9e4066Sahrens vdev_t *vd; 4685cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 469fa9e4066Sahrens 470fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 4715cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 472fa9e4066Sahrens 4730e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4740e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4750e34b6a7Sbonwick spa->spa_root_vdev = vd; 476e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4770e34b6a7Sbonwick } 4780e34b6a7Sbonwick 47988ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4800e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4810e34b6a7Sbonwick /* 4820e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4830e34b6a7Sbonwick * which must be unique among all pools. 4840e34b6a7Sbonwick */ 4851195e687SMark J Musante guid = spa_generate_guid(NULL); 4860e34b6a7Sbonwick } else { 4870e34b6a7Sbonwick /* 4880e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4890e34b6a7Sbonwick */ 4901195e687SMark J Musante guid = spa_generate_guid(spa); 4910e34b6a7Sbonwick } 4920e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4930e34b6a7Sbonwick } 4940e34b6a7Sbonwick 495fa9e4066Sahrens vd->vdev_spa = spa; 496fa9e4066Sahrens vd->vdev_id = id; 497fa9e4066Sahrens vd->vdev_guid = guid; 498fa9e4066Sahrens vd->vdev_guid_sum = guid; 499fa9e4066Sahrens vd->vdev_ops = ops; 500fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 50188ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 5025cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 5035cabbc6bSPrashanth Sreenivasa 5045cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 5055cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 5064d7988d6SPaul Dagnelie vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 5074d7988d6SPaul Dagnelie 0, 0); 508fa9e4066Sahrens 509084fd14fSBrian Behlendorf list_link_init(&vd->vdev_initialize_node); 510e0f1c0afSOlaf Faaland list_link_init(&vd->vdev_leaf_node); 511084fd14fSBrian Behlendorf list_link_init(&vd->vdev_trim_node); 512fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 5135ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 514e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 515a3874b8bSToomas Soome mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); 516094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 517094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 518094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 519094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 520084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); 521084fd14fSBrian Behlendorf mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); 522084fd14fSBrian Behlendorf mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); 523084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); 524084fd14fSBrian Behlendorf cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); 525084fd14fSBrian Behlendorf cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); 526094e47e9SGeorge Wilson 5278ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 5284d7988d6SPaul Dagnelie vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 5294d7988d6SPaul Dagnelie 0); 5308ad4d6ddSJeff Bonwick } 531b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 532fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 533b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 534fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 535fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 5363d7072f8Seschrock vdev_queue_init(vd); 5373d7072f8Seschrock vdev_cache_init(vd); 538fa9e4066Sahrens 539fa9e4066Sahrens return (vd); 540fa9e4066Sahrens } 541fa9e4066Sahrens 542fa9e4066Sahrens /* 543fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 544fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 545fa9e4066Sahrens * different for each case. 546fa9e4066Sahrens */ 54799653d4eSeschrock int 54899653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 54999653d4eSeschrock int alloctype) 550fa9e4066Sahrens { 551fa9e4066Sahrens vdev_ops_t *ops; 552fa9e4066Sahrens char *type; 5538654d025Sperrin uint64_t guid = 0, islog, nparity; 554fa9e4066Sahrens vdev_t *vd; 5555cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 556663207adSDon Brady vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; 557663207adSDon Brady boolean_t top_level = (parent && !parent->vdev_parent); 558fa9e4066Sahrens 559e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 560fa9e4066Sahrens 561fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 562be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 563fa9e4066Sahrens 564fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 565be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 566fa9e4066Sahrens 567fa9e4066Sahrens /* 568fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 569fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 570fa9e4066Sahrens */ 571fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 572fa9e4066Sahrens uint64_t label_id; 573fa9e4066Sahrens 574fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 575fa9e4066Sahrens label_id != id) 576be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 577fa9e4066Sahrens 578fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 579be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 58099653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 58199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 582be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 583fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 584fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 585be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 58621ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 58721ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 588be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 589fa9e4066Sahrens } 590fa9e4066Sahrens 59199653d4eSeschrock /* 59299653d4eSeschrock * The first allocated vdev must be of type 'root'. 59399653d4eSeschrock */ 59499653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 595be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 59699653d4eSeschrock 5978654d025Sperrin /* 5988654d025Sperrin * Determine whether we're a log vdev. 5998654d025Sperrin */ 6008654d025Sperrin islog = 0; 6018654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 602990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 603be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 604fa9e4066Sahrens 60588ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 606be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 60788ecc943SGeorge Wilson 60899653d4eSeschrock /* 6098654d025Sperrin * Set the nparity property for RAID-Z vdevs. 61099653d4eSeschrock */ 6118654d025Sperrin nparity = -1ULL; 61299653d4eSeschrock if (ops == &vdev_raidz_ops) { 61399653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 6148654d025Sperrin &nparity) == 0) { 615b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 616be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 61799653d4eSeschrock /* 618f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 619f94275ceSAdam Leventhal * device. 62099653d4eSeschrock */ 621f94275ceSAdam Leventhal if (nparity > 1 && 622f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 623be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 624f94275ceSAdam Leventhal if (nparity > 2 && 625f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 626be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 62799653d4eSeschrock } else { 62899653d4eSeschrock /* 62999653d4eSeschrock * We require the parity to be specified for SPAs that 63099653d4eSeschrock * support multiple parity levels. 63199653d4eSeschrock */ 632f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 633be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 63499653d4eSeschrock /* 63599653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 63699653d4eSeschrock */ 6378654d025Sperrin nparity = 1; 63899653d4eSeschrock } 63999653d4eSeschrock } else { 6408654d025Sperrin nparity = 0; 64199653d4eSeschrock } 6428654d025Sperrin ASSERT(nparity != -1ULL); 6438654d025Sperrin 644663207adSDon Brady /* 645663207adSDon Brady * If creating a top-level vdev, check for allocation classes input 646663207adSDon Brady */ 647663207adSDon Brady if (top_level && alloctype == VDEV_ALLOC_ADD) { 648663207adSDon Brady char *bias; 649663207adSDon Brady 650663207adSDon Brady if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 651663207adSDon Brady &bias) == 0) { 652663207adSDon Brady alloc_bias = vdev_derive_alloc_bias(bias); 653663207adSDon Brady 654663207adSDon Brady /* spa_vdev_add() expects feature to be enabled */ 655c1064fd7SJerry Jelinek if (alloc_bias != VDEV_BIAS_LOG && 656c1064fd7SJerry Jelinek spa->spa_load_state != SPA_LOAD_CREATE && 657663207adSDon Brady !spa_feature_is_enabled(spa, 658663207adSDon Brady SPA_FEATURE_ALLOCATION_CLASSES)) { 659663207adSDon Brady return (SET_ERROR(ENOTSUP)); 660663207adSDon Brady } 661663207adSDon Brady } 662663207adSDon Brady } 663663207adSDon Brady 6648654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 6655cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 6668654d025Sperrin 6678654d025Sperrin vd->vdev_islog = islog; 6688654d025Sperrin vd->vdev_nparity = nparity; 669663207adSDon Brady if (top_level && alloc_bias != VDEV_BIAS_NONE) 670663207adSDon Brady vd->vdev_alloc_bias = alloc_bias; 6718654d025Sperrin 6728654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 6738654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 6748654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 6758654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 6768654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 6778654d025Sperrin &vd->vdev_physpath) == 0) 6788654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 6796809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 6806809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 68199653d4eSeschrock 682afefbcddSeschrock /* 683afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 684afefbcddSeschrock * as -1. 685afefbcddSeschrock */ 686afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 687afefbcddSeschrock &vd->vdev_wholedisk) != 0) 688afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 689afefbcddSeschrock 6905cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 6915cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 6925cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 6935cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 6945cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 6955cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 6965cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 6975cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 6985cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 6995cabbc6bSPrashanth Sreenivasa 700ea8dc4b6Seschrock /* 701ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 702ea8dc4b6Seschrock * was not present at the time of import. 703ea8dc4b6Seschrock */ 7046809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 7056809eb4eSEric Schrock &vd->vdev_not_present); 706ea8dc4b6Seschrock 707ecc2d604Sbonwick /* 708ecc2d604Sbonwick * Get the alignment requirement. 709ecc2d604Sbonwick */ 710ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 711ecc2d604Sbonwick 71288ecc943SGeorge Wilson /* 71388ecc943SGeorge Wilson * Retrieve the vdev creation time. 71488ecc943SGeorge Wilson */ 71588ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 71688ecc943SGeorge Wilson &vd->vdev_crtxg); 71788ecc943SGeorge Wilson 718fa9e4066Sahrens /* 719fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 720fa9e4066Sahrens */ 721663207adSDon Brady if (top_level && 7221195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 723fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 724fa9e4066Sahrens &vd->vdev_ms_array); 725fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 726fa9e4066Sahrens &vd->vdev_ms_shift); 727fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 728fa9e4066Sahrens &vd->vdev_asize); 7293f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 7303f9d6ad7SLin Ling &vd->vdev_removing); 731215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 732215198a6SJoe Stein &vd->vdev_top_zap); 733215198a6SJoe Stein } else { 734215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 735fa9e4066Sahrens } 736fa9e4066Sahrens 737663207adSDon Brady if (top_level && alloctype != VDEV_ALLOC_ATTACH) { 738a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 7399f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 7401195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 7419f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 742663207adSDon Brady /* Note: metaslab_group_create() is now deferred */ 743a1521560SJeff Bonwick } 744a1521560SJeff Bonwick 745215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 746215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 747215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 748215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 749215198a6SJoe Stein } else { 750215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 751215198a6SJoe Stein } 752215198a6SJoe Stein 753fa9e4066Sahrens /* 7543d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 755fa9e4066Sahrens */ 756215198a6SJoe Stein 757c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 75821ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 75921ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 760c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 761c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 7620713e232SGeorge Wilson &vd->vdev_dtl_object); 763c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 764c5904d13Seschrock &vd->vdev_unspare); 765c5904d13Seschrock } 76621ecdf64SLin Ling 76721ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 76821ecdf64SLin Ling uint64_t spare = 0; 76921ecdf64SLin Ling 77021ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 77121ecdf64SLin Ling &spare) == 0 && spare) 77221ecdf64SLin Ling spa_spare_add(vd); 77321ecdf64SLin Ling } 77421ecdf64SLin Ling 775ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 776ecc2d604Sbonwick &vd->vdev_offline); 777c5904d13Seschrock 778b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 779b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 780cb04b873SMark J Musante 781e4c795beSTom Caputi if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) 7820c06d385Sjwpoduska vdev_defer_resilver(vd); 783e4c795beSTom Caputi 7843d7072f8Seschrock /* 7853d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 7863d7072f8Seschrock * state, as the diagnosis made on another system may not be 787069f55e2SEric Schrock * valid in the current context. Local vdevs will 788069f55e2SEric Schrock * remain in the faulted state. 7893d7072f8Seschrock */ 790b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 7913d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 7923d7072f8Seschrock &vd->vdev_faulted); 7933d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 7943d7072f8Seschrock &vd->vdev_degraded); 7953d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 7963d7072f8Seschrock &vd->vdev_removed); 797069f55e2SEric Schrock 798069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 799069f55e2SEric Schrock char *aux; 800069f55e2SEric Schrock 801069f55e2SEric Schrock vd->vdev_label_aux = 802069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 803069f55e2SEric Schrock if (nvlist_lookup_string(nv, 804069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 805069f55e2SEric Schrock strcmp(aux, "external") == 0) 806069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 807069f55e2SEric Schrock } 8083d7072f8Seschrock } 809fa9e4066Sahrens } 810fa9e4066Sahrens 811fa9e4066Sahrens /* 812fa9e4066Sahrens * Add ourselves to the parent's list of children. 813fa9e4066Sahrens */ 814fa9e4066Sahrens vdev_add_child(parent, vd); 815fa9e4066Sahrens 81699653d4eSeschrock *vdp = vd; 81799653d4eSeschrock 81899653d4eSeschrock return (0); 819fa9e4066Sahrens } 820fa9e4066Sahrens 821fa9e4066Sahrens void 822fa9e4066Sahrens vdev_free(vdev_t *vd) 823fa9e4066Sahrens { 8243d7072f8Seschrock spa_t *spa = vd->vdev_spa; 825084fd14fSBrian Behlendorf 826094e47e9SGeorge Wilson ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 827084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_trim_thread, ==, NULL); 828084fd14fSBrian Behlendorf ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); 829fa9e4066Sahrens 830a3874b8bSToomas Soome /* 831a3874b8bSToomas Soome * Scan queues are normally destroyed at the end of a scan. If the 832a3874b8bSToomas Soome * queue exists here, that implies the vdev is being removed while 833a3874b8bSToomas Soome * the scan is still running. 834a3874b8bSToomas Soome */ 835a3874b8bSToomas Soome if (vd->vdev_scan_io_queue != NULL) { 836a3874b8bSToomas Soome mutex_enter(&vd->vdev_scan_io_queue_lock); 837a3874b8bSToomas Soome dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); 838a3874b8bSToomas Soome vd->vdev_scan_io_queue = NULL; 839a3874b8bSToomas Soome mutex_exit(&vd->vdev_scan_io_queue_lock); 840a3874b8bSToomas Soome } 841a3874b8bSToomas Soome 842fa9e4066Sahrens /* 843fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 844fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 845fa9e4066Sahrens */ 846fa9e4066Sahrens vdev_close(vd); 847fa9e4066Sahrens 848e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 849b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 850fa9e4066Sahrens 851fa9e4066Sahrens /* 852fa9e4066Sahrens * Free all children. 853fa9e4066Sahrens */ 854573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 855fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 856fa9e4066Sahrens 857fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 858fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 859fa9e4066Sahrens 860fa9e4066Sahrens /* 861fa9e4066Sahrens * Discard allocation state. 862fa9e4066Sahrens */ 863a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 864fa9e4066Sahrens vdev_metaslab_fini(vd); 865a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 866814dcd43SSerapheim Dimitropoulos vd->vdev_mg = NULL; 867a1521560SJeff Bonwick } 868fa9e4066Sahrens 869fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 870fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 871fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 872fa9e4066Sahrens 873fa9e4066Sahrens /* 874fa9e4066Sahrens * Remove this vdev from its parent's child list. 875fa9e4066Sahrens */ 876fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 877fa9e4066Sahrens 878fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 879e0f1c0afSOlaf Faaland ASSERT(!list_link_active(&vd->vdev_leaf_node)); 880fa9e4066Sahrens 8813d7072f8Seschrock /* 8823d7072f8Seschrock * Clean up vdev structure. 8833d7072f8Seschrock */ 8843d7072f8Seschrock vdev_queue_fini(vd); 8853d7072f8Seschrock vdev_cache_fini(vd); 8863d7072f8Seschrock 8873d7072f8Seschrock if (vd->vdev_path) 8883d7072f8Seschrock spa_strfree(vd->vdev_path); 8893d7072f8Seschrock if (vd->vdev_devid) 8903d7072f8Seschrock spa_strfree(vd->vdev_devid); 8913d7072f8Seschrock if (vd->vdev_physpath) 8923d7072f8Seschrock spa_strfree(vd->vdev_physpath); 8936809eb4eSEric Schrock if (vd->vdev_fru) 8946809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 8953d7072f8Seschrock 8963d7072f8Seschrock if (vd->vdev_isspare) 8973d7072f8Seschrock spa_spare_remove(vd); 898fa94a07fSbrendan if (vd->vdev_isl2cache) 899fa94a07fSbrendan spa_l2cache_remove(vd); 9003d7072f8Seschrock 9013d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 9023d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 9038ad4d6ddSJeff Bonwick 9043d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 9050713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 9068ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 9070713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 9080713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 9098ad4d6ddSJeff Bonwick } 9103d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 9118ad4d6ddSJeff Bonwick 9125cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 9135cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 9145cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 9155cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 9165cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 9175cabbc6bSPrashanth Sreenivasa } 9185cabbc6bSPrashanth Sreenivasa 9195cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 9205cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 9215cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 9225cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 9235cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 9245cabbc6bSPrashanth Sreenivasa } 9255cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 9265cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 9275cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 9285cabbc6bSPrashanth Sreenivasa 9293d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 9303d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 931e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 932a3874b8bSToomas Soome mutex_destroy(&vd->vdev_scan_io_queue_lock); 933094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_lock); 934094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_io_lock); 935094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_io_cv); 936094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_cv); 937084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_lock); 938084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_autotrim_lock); 939084fd14fSBrian Behlendorf mutex_destroy(&vd->vdev_trim_io_lock); 940084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_cv); 941084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_autotrim_cv); 942084fd14fSBrian Behlendorf cv_destroy(&vd->vdev_trim_io_cv); 9433d7072f8Seschrock 9443d7072f8Seschrock if (vd == spa->spa_root_vdev) 9453d7072f8Seschrock spa->spa_root_vdev = NULL; 9463d7072f8Seschrock 9473d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 948fa9e4066Sahrens } 949fa9e4066Sahrens 950fa9e4066Sahrens /* 951fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 952fa9e4066Sahrens */ 953fa9e4066Sahrens static void 954fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 955fa9e4066Sahrens { 956fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 957fa9e4066Sahrens metaslab_t *msp; 958fa9e4066Sahrens vdev_t *vd; 959fa9e4066Sahrens int t; 960fa9e4066Sahrens 961fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 962fa9e4066Sahrens 963fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 964fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 965fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 966215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 967fa9e4066Sahrens 968fa9e4066Sahrens svd->vdev_ms_array = 0; 969fa9e4066Sahrens svd->vdev_ms_shift = 0; 970fa9e4066Sahrens svd->vdev_ms_count = 0; 971215198a6SJoe Stein svd->vdev_top_zap = 0; 972fa9e4066Sahrens 973cd0837ccSGeorge Wilson if (tvd->vdev_mg) 974cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 975fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 976fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 977fa9e4066Sahrens 978fa9e4066Sahrens svd->vdev_mg = NULL; 979fa9e4066Sahrens svd->vdev_ms = NULL; 980ecc2d604Sbonwick 981ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 982ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 983fa9e4066Sahrens 98486714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 98586714001SSerapheim Dimitropoulos svd->vdev_checkpoint_sm = NULL; 98686714001SSerapheim Dimitropoulos 987663207adSDon Brady tvd->vdev_alloc_bias = svd->vdev_alloc_bias; 988663207adSDon Brady svd->vdev_alloc_bias = VDEV_BIAS_NONE; 989663207adSDon Brady 990fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 991fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 99299653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 993fa9e4066Sahrens 994fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 995fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 99699653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 997fa9e4066Sahrens 9983a4b1be9SMatthew Ahrens /* 9993a4b1be9SMatthew Ahrens * State which may be set on a top-level vdev that's in the 10003a4b1be9SMatthew Ahrens * process of being removed. 10013a4b1be9SMatthew Ahrens */ 10023a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_births_object); 10033a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 10043a4b1be9SMatthew Ahrens ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 10053a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 10063a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 10073a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 10083a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_removing); 10093a4b1be9SMatthew Ahrens tvd->vdev_removing = svd->vdev_removing; 10103a4b1be9SMatthew Ahrens tvd->vdev_indirect_config = svd->vdev_indirect_config; 10113a4b1be9SMatthew Ahrens tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 10123a4b1be9SMatthew Ahrens tvd->vdev_indirect_births = svd->vdev_indirect_births; 10133a4b1be9SMatthew Ahrens range_tree_swap(&svd->vdev_obsolete_segments, 10143a4b1be9SMatthew Ahrens &tvd->vdev_obsolete_segments); 10153a4b1be9SMatthew Ahrens tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 10163a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_mapping_object = 0; 10173a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_births_object = 0; 10183a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 10193a4b1be9SMatthew Ahrens svd->vdev_indirect_mapping = NULL; 10203a4b1be9SMatthew Ahrens svd->vdev_indirect_births = NULL; 10213a4b1be9SMatthew Ahrens svd->vdev_obsolete_sm = NULL; 10223a4b1be9SMatthew Ahrens svd->vdev_removing = 0; 10233a4b1be9SMatthew Ahrens 1024fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 1025fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 1026fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 1027fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 1028fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 1029fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 1030fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 1031fa9e4066Sahrens } 1032fa9e4066Sahrens 1033e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 1034fa9e4066Sahrens vdev_config_clean(svd); 1035fa9e4066Sahrens vdev_config_dirty(tvd); 1036fa9e4066Sahrens } 1037fa9e4066Sahrens 1038e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 1039e14bb325SJeff Bonwick vdev_state_clean(svd); 1040e14bb325SJeff Bonwick vdev_state_dirty(tvd); 1041e14bb325SJeff Bonwick } 1042e14bb325SJeff Bonwick 104399653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 104499653d4eSeschrock svd->vdev_deflate_ratio = 0; 10458654d025Sperrin 10468654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 10478654d025Sperrin svd->vdev_islog = 0; 1048a3874b8bSToomas Soome 1049a3874b8bSToomas Soome dsl_scan_io_queue_vdev_xfer(svd, tvd); 1050fa9e4066Sahrens } 1051fa9e4066Sahrens 1052fa9e4066Sahrens static void 1053fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 1054fa9e4066Sahrens { 1055fa9e4066Sahrens if (vd == NULL) 1056fa9e4066Sahrens return; 1057fa9e4066Sahrens 1058fa9e4066Sahrens vd->vdev_top = tvd; 1059fa9e4066Sahrens 1060573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1061fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 1062fa9e4066Sahrens } 1063fa9e4066Sahrens 1064fa9e4066Sahrens /* 1065fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 1066fa9e4066Sahrens */ 1067fa9e4066Sahrens vdev_t * 1068fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 1069fa9e4066Sahrens { 1070fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 1071fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 1072fa9e4066Sahrens vdev_t *mvd; 1073fa9e4066Sahrens 1074e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1075fa9e4066Sahrens 1076fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 1077ecc2d604Sbonwick 1078ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 1079573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 10804263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 10815cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 1082ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 1083ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 108488ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 1085ecc2d604Sbonwick 1086fa9e4066Sahrens vdev_remove_child(pvd, cvd); 1087fa9e4066Sahrens vdev_add_child(pvd, mvd); 1088fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 1089fa9e4066Sahrens vdev_add_child(mvd, cvd); 1090fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1091fa9e4066Sahrens 1092fa9e4066Sahrens if (mvd == mvd->vdev_top) 1093fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 1094fa9e4066Sahrens 1095fa9e4066Sahrens return (mvd); 1096fa9e4066Sahrens } 1097fa9e4066Sahrens 1098fa9e4066Sahrens /* 1099fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 1100fa9e4066Sahrens */ 1101fa9e4066Sahrens void 1102fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 1103fa9e4066Sahrens { 1104fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 1105fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 1106fa9e4066Sahrens 1107e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1108fa9e4066Sahrens 1109fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 1110fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 111199653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 111299653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 1113ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 1114fa9e4066Sahrens 1115fa9e4066Sahrens vdev_remove_child(mvd, cvd); 1116fa9e4066Sahrens vdev_remove_child(pvd, mvd); 11178ad4d6ddSJeff Bonwick 111899653d4eSeschrock /* 1119e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1120e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 1121e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 1122e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 112399653d4eSeschrock */ 11248ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 11258ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 11261195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 11278ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 11288ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 11298ad4d6ddSJeff Bonwick } 1130e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 1131e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 1132fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1133fa9e4066Sahrens 1134fa9e4066Sahrens if (cvd == cvd->vdev_top) 1135fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 1136fa9e4066Sahrens 1137fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 1138fa9e4066Sahrens vdev_free(mvd); 1139fa9e4066Sahrens } 1140fa9e4066Sahrens 1141663207adSDon Brady static void 1142663207adSDon Brady vdev_metaslab_group_create(vdev_t *vd) 1143663207adSDon Brady { 1144663207adSDon Brady spa_t *spa = vd->vdev_spa; 1145663207adSDon Brady 1146663207adSDon Brady /* 1147663207adSDon Brady * metaslab_group_create was delayed until allocation bias was available 1148663207adSDon Brady */ 1149663207adSDon Brady if (vd->vdev_mg == NULL) { 1150663207adSDon Brady metaslab_class_t *mc; 1151663207adSDon Brady 1152663207adSDon Brady if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) 1153663207adSDon Brady vd->vdev_alloc_bias = VDEV_BIAS_LOG; 1154663207adSDon Brady 1155663207adSDon Brady ASSERT3U(vd->vdev_islog, ==, 1156663207adSDon Brady (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); 1157663207adSDon Brady 1158663207adSDon Brady switch (vd->vdev_alloc_bias) { 1159663207adSDon Brady case VDEV_BIAS_LOG: 1160663207adSDon Brady mc = spa_log_class(spa); 1161663207adSDon Brady break; 1162663207adSDon Brady case VDEV_BIAS_SPECIAL: 1163663207adSDon Brady mc = spa_special_class(spa); 1164663207adSDon Brady break; 1165663207adSDon Brady case VDEV_BIAS_DEDUP: 1166663207adSDon Brady mc = spa_dedup_class(spa); 1167663207adSDon Brady break; 1168663207adSDon Brady default: 1169663207adSDon Brady mc = spa_normal_class(spa); 1170663207adSDon Brady } 1171663207adSDon Brady 1172663207adSDon Brady vd->vdev_mg = metaslab_group_create(mc, vd, 1173663207adSDon Brady spa->spa_alloc_count); 1174663207adSDon Brady 1175663207adSDon Brady /* 1176663207adSDon Brady * The spa ashift values currently only reflect the 1177663207adSDon Brady * general vdev classes. Class destination is late 1178663207adSDon Brady * binding so ashift checking had to wait until now 1179663207adSDon Brady */ 1180663207adSDon Brady if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1181663207adSDon Brady mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { 1182663207adSDon Brady if (vd->vdev_ashift > spa->spa_max_ashift) 1183663207adSDon Brady spa->spa_max_ashift = vd->vdev_ashift; 1184663207adSDon Brady if (vd->vdev_ashift < spa->spa_min_ashift) 1185663207adSDon Brady spa->spa_min_ashift = vd->vdev_ashift; 1186663207adSDon Brady } 1187663207adSDon Brady } 1188663207adSDon Brady } 1189663207adSDon Brady 1190ea8dc4b6Seschrock int 1191fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1192fa9e4066Sahrens { 1193fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1194ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1195ecc2d604Sbonwick uint64_t m; 1196fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 1197fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1198ecc2d604Sbonwick metaslab_t **mspp; 1199ecc2d604Sbonwick int error; 1200663207adSDon Brady boolean_t expanding = (oldc != 0); 1201fa9e4066Sahrens 1202a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1203a1521560SJeff Bonwick 120488ecc943SGeorge Wilson /* 120588ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 120688ecc943SGeorge Wilson */ 120788ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 12080e34b6a7Sbonwick return (0); 12090e34b6a7Sbonwick 121088ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 121188ecc943SGeorge Wilson 1212fa9e4066Sahrens ASSERT(oldc <= newc); 1213fa9e4066Sahrens 1214ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1215fa9e4066Sahrens 1216663207adSDon Brady if (expanding) { 1217ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1218ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1219ecc2d604Sbonwick } 1220fa9e4066Sahrens 1221ecc2d604Sbonwick vd->vdev_ms = mspp; 1222ecc2d604Sbonwick vd->vdev_ms_count = newc; 1223ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 12240713e232SGeorge Wilson uint64_t object = 0; 12250713e232SGeorge Wilson 12265cabbc6bSPrashanth Sreenivasa /* 12275cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 12285cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 12295cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 12305cabbc6bSPrashanth Sreenivasa */ 12315cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1232ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 12337bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 12347bfdf011SNeil Perrin DMU_READ_PREFETCH); 12353ee8c80cSPavel Zakharov if (error != 0) { 12363ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 12373ee8c80cSPavel Zakharov "array [error=%d]", error); 1238ecc2d604Sbonwick return (error); 12393ee8c80cSPavel Zakharov } 1240fa9e4066Sahrens } 12411e9bd7ecSPrakash Surya 1242663207adSDon Brady #ifndef _KERNEL 1243663207adSDon Brady /* 1244663207adSDon Brady * To accomodate zdb_leak_init() fake indirect 1245663207adSDon Brady * metaslabs, we allocate a metaslab group for 1246663207adSDon Brady * indirect vdevs which normally don't have one. 1247663207adSDon Brady */ 1248663207adSDon Brady if (vd->vdev_mg == NULL) { 1249663207adSDon Brady ASSERT0(vdev_is_concrete(vd)); 1250663207adSDon Brady vdev_metaslab_group_create(vd); 1251663207adSDon Brady } 1252663207adSDon Brady #endif 12531e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 12541e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 12553ee8c80cSPavel Zakharov if (error != 0) { 12563ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 12573ee8c80cSPavel Zakharov error); 12581e9bd7ecSPrakash Surya return (error); 12593ee8c80cSPavel Zakharov } 1260fa9e4066Sahrens } 1261fa9e4066Sahrens 1262a1521560SJeff Bonwick if (txg == 0) 1263a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1264a1521560SJeff Bonwick 12653f9d6ad7SLin Ling /* 12663f9d6ad7SLin Ling * If the vdev is being removed we don't activate 12673f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 12683f9d6ad7SLin Ling * allocations are performed on this device. 12693f9d6ad7SLin Ling */ 1270663207adSDon Brady if (!expanding && !vd->vdev_removing) { 1271a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1272663207adSDon Brady } 1273a1521560SJeff Bonwick 1274a1521560SJeff Bonwick if (txg == 0) 1275a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1276a1521560SJeff Bonwick 1277814dcd43SSerapheim Dimitropoulos /* 1278814dcd43SSerapheim Dimitropoulos * Regardless whether this vdev was just added or it is being 1279814dcd43SSerapheim Dimitropoulos * expanded, the metaslab count has changed. Recalculate the 1280814dcd43SSerapheim Dimitropoulos * block limit. 1281814dcd43SSerapheim Dimitropoulos */ 1282814dcd43SSerapheim Dimitropoulos spa_log_sm_set_blocklimit(spa); 1283814dcd43SSerapheim Dimitropoulos 1284ea8dc4b6Seschrock return (0); 1285fa9e4066Sahrens } 1286fa9e4066Sahrens 1287fa9e4066Sahrens void 1288fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1289fa9e4066Sahrens { 129086714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 129186714001SSerapheim Dimitropoulos ASSERT(spa_feature_is_active(vd->vdev_spa, 129286714001SSerapheim Dimitropoulos SPA_FEATURE_POOL_CHECKPOINT)); 129386714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 129486714001SSerapheim Dimitropoulos /* 129586714001SSerapheim Dimitropoulos * Even though we close the space map, we need to set its 129686714001SSerapheim Dimitropoulos * pointer to NULL. The reason is that vdev_metaslab_fini() 129786714001SSerapheim Dimitropoulos * may be called multiple times for certain operations 129886714001SSerapheim Dimitropoulos * (i.e. when destroying a pool) so we need to ensure that 129986714001SSerapheim Dimitropoulos * this clause never executes twice. This logic is similar 130086714001SSerapheim Dimitropoulos * to the one used for the vdev_ms clause below. 130186714001SSerapheim Dimitropoulos */ 130286714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 130386714001SSerapheim Dimitropoulos } 130486714001SSerapheim Dimitropoulos 1305fa9e4066Sahrens if (vd->vdev_ms != NULL) { 1306555d674dSSerapheim Dimitropoulos metaslab_group_t *mg = vd->vdev_mg; 1307555d674dSSerapheim Dimitropoulos metaslab_group_passivate(mg); 13085cabbc6bSPrashanth Sreenivasa 1309555d674dSSerapheim Dimitropoulos uint64_t count = vd->vdev_ms_count; 13105cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 13110713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 13120713e232SGeorge Wilson if (msp != NULL) 13130713e232SGeorge Wilson metaslab_fini(msp); 13140713e232SGeorge Wilson } 1315fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1316fa9e4066Sahrens vd->vdev_ms = NULL; 13175cabbc6bSPrashanth Sreenivasa 13185cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1319555d674dSSerapheim Dimitropoulos 1320555d674dSSerapheim Dimitropoulos for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 1321555d674dSSerapheim Dimitropoulos ASSERT0(mg->mg_histogram[i]); 1322fa9e4066Sahrens } 13235cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1324fa9e4066Sahrens } 1325fa9e4066Sahrens 1326e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1327e14bb325SJeff Bonwick boolean_t vps_readable; 1328e14bb325SJeff Bonwick boolean_t vps_writeable; 1329e14bb325SJeff Bonwick int vps_flags; 1330e14bb325SJeff Bonwick } vdev_probe_stats_t; 1331e14bb325SJeff Bonwick 1332e14bb325SJeff Bonwick static void 1333e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 13340a4e9518Sgw { 13358ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1336a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1337e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1338a3f829aeSBill Moore 1339a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1340e14bb325SJeff Bonwick 1341e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1342e14bb325SJeff Bonwick if (zio->io_error == 0) 1343e14bb325SJeff Bonwick vps->vps_readable = 1; 13448ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1345a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1346770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1347e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1348e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1349e14bb325SJeff Bonwick } else { 1350770499e1SDan Kimmel abd_free(zio->io_abd); 1351e14bb325SJeff Bonwick } 1352e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1353e14bb325SJeff Bonwick if (zio->io_error == 0) 1354e14bb325SJeff Bonwick vps->vps_writeable = 1; 1355770499e1SDan Kimmel abd_free(zio->io_abd); 1356e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1357a3f829aeSBill Moore zio_t *pio; 1358e14bb325SJeff Bonwick 1359e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1360e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1361e14bb325SJeff Bonwick 1362e14bb325SJeff Bonwick if (vdev_readable(vd) && 13638ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1364e14bb325SJeff Bonwick zio->io_error = 0; 1365e14bb325SJeff Bonwick } else { 1366e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 13673ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1368e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1369eb633035STom Caputi spa, vd, NULL, NULL, 0, 0); 1370be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1371e14bb325SJeff Bonwick } 1372a3f829aeSBill Moore 1373a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1374a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1375a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1376a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1377a3f829aeSBill Moore 13780f7643c7SGeorge Wilson zio_link_t *zl = NULL; 13790f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1380a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1381be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1382a3f829aeSBill Moore 1383e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1384e14bb325SJeff Bonwick } 1385e14bb325SJeff Bonwick } 13860a4e9518Sgw 1387e14bb325SJeff Bonwick /* 1388f7170741SWill Andrews * Determine whether this device is accessible. 1389f7170741SWill Andrews * 1390f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1391f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1392f7170741SWill Andrews * a VTOC. 1393e14bb325SJeff Bonwick */ 1394e14bb325SJeff Bonwick zio_t * 1395a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1396e14bb325SJeff Bonwick { 1397e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1398a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1399a3f829aeSBill Moore zio_t *pio; 1400a3f829aeSBill Moore 1401a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 14020a4e9518Sgw 1403a3f829aeSBill Moore /* 1404a3f829aeSBill Moore * Don't probe the probe. 1405a3f829aeSBill Moore */ 1406a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1407a3f829aeSBill Moore return (NULL); 1408e14bb325SJeff Bonwick 1409a3f829aeSBill Moore /* 1410a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1411a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1412a3f829aeSBill Moore * this vdev will become parents of the probe io. 1413a3f829aeSBill Moore */ 1414a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1415e14bb325SJeff Bonwick 1416a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1417a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1418a3f829aeSBill Moore 1419a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1420a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 14218956713aSEric Schrock ZIO_FLAG_TRYHARD; 1422a3f829aeSBill Moore 1423a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1424a3f829aeSBill Moore /* 1425a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1426a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1427a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1428a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1429a3f829aeSBill Moore * any zio looking at these values can assume that 1430a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1431a3f829aeSBill Moore * important because when a device has intermittent 1432a3f829aeSBill Moore * connectivity problems, we want to ensure that 1433a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1434a3f829aeSBill Moore * the zio (EIO). 1435a3f829aeSBill Moore * 1436a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1437a3f829aeSBill Moore * values so the probe can reevaluate from first 1438a3f829aeSBill Moore * principles. 1439a3f829aeSBill Moore */ 1440a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1441a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1442a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1443a3f829aeSBill Moore } 1444a3f829aeSBill Moore 1445a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1446a3f829aeSBill Moore vdev_probe_done, vps, 1447a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1448a3f829aeSBill Moore 144998d1cbfeSGeorge Wilson /* 145098d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 145198d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 145298d1cbfeSGeorge Wilson */ 1453a3f829aeSBill Moore if (zio != NULL) { 1454a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1455a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1456a3f829aeSBill Moore } 1457e14bb325SJeff Bonwick } 1458e14bb325SJeff Bonwick 1459a3f829aeSBill Moore if (zio != NULL) 1460a3f829aeSBill Moore zio_add_child(zio, pio); 1461e14bb325SJeff Bonwick 1462a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1463e14bb325SJeff Bonwick 1464a3f829aeSBill Moore if (vps == NULL) { 1465a3f829aeSBill Moore ASSERT(zio != NULL); 1466a3f829aeSBill Moore return (NULL); 1467a3f829aeSBill Moore } 1468e14bb325SJeff Bonwick 1469e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1470a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1471e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1472770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1473770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1474e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1475e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1476e14bb325SJeff Bonwick } 1477e14bb325SJeff Bonwick 1478a3f829aeSBill Moore if (zio == NULL) 1479a3f829aeSBill Moore return (pio); 1480a3f829aeSBill Moore 1481a3f829aeSBill Moore zio_nowait(pio); 1482a3f829aeSBill Moore return (NULL); 14830a4e9518Sgw } 14840a4e9518Sgw 1485f64c0e34SEric Taylor static void 1486f64c0e34SEric Taylor vdev_open_child(void *arg) 1487f64c0e34SEric Taylor { 1488f64c0e34SEric Taylor vdev_t *vd = arg; 1489f64c0e34SEric Taylor 1490f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1491f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1492f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1493f64c0e34SEric Taylor } 1494f64c0e34SEric Taylor 1495681d9761SEric Taylor boolean_t 1496681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1497681d9761SEric Taylor { 1498681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1499681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1500681d9761SEric Taylor return (B_TRUE); 1501681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1502681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1503681d9761SEric Taylor return (B_TRUE); 1504681d9761SEric Taylor return (B_FALSE); 1505681d9761SEric Taylor } 1506681d9761SEric Taylor 1507f64c0e34SEric Taylor void 1508f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1509f64c0e34SEric Taylor { 1510f64c0e34SEric Taylor taskq_t *tq; 1511f64c0e34SEric Taylor int children = vd->vdev_children; 1512f64c0e34SEric Taylor 1513681d9761SEric Taylor /* 1514681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1515681d9761SEric Taylor * in a single thread so that the same thread holds the 1516681d9761SEric Taylor * spa_namespace_lock 1517681d9761SEric Taylor */ 1518681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 151912a8814cSTom Caputi retry_sync: 1520681d9761SEric Taylor for (int c = 0; c < children; c++) 1521681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1522681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 152312a8814cSTom Caputi } else { 152412a8814cSTom Caputi tq = taskq_create("vdev_open", children, minclsyspri, 152512a8814cSTom Caputi children, children, TASKQ_PREPOPULATE); 152612a8814cSTom Caputi if (tq == NULL) 152712a8814cSTom Caputi goto retry_sync; 152812a8814cSTom Caputi 152912a8814cSTom Caputi for (int c = 0; c < children; c++) 153012a8814cSTom Caputi VERIFY(taskq_dispatch(tq, vdev_open_child, 153112a8814cSTom Caputi vd->vdev_child[c], TQ_SLEEP) != TASKQID_INVALID); 153212a8814cSTom Caputi 153312a8814cSTom Caputi taskq_destroy(tq); 1534681d9761SEric Taylor } 1535f64c0e34SEric Taylor 153612a8814cSTom Caputi vd->vdev_nonrot = B_TRUE; 1537f64c0e34SEric Taylor 153812a8814cSTom Caputi for (int c = 0; c < children; c++) 153912a8814cSTom Caputi vd->vdev_nonrot &= vd->vdev_child[c]->vdev_nonrot; 1540f64c0e34SEric Taylor } 1541f64c0e34SEric Taylor 15425cabbc6bSPrashanth Sreenivasa /* 15435cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 15445cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 15455cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 15465cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 15475cabbc6bSPrashanth Sreenivasa */ 15485cabbc6bSPrashanth Sreenivasa static void 15495cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 15505cabbc6bSPrashanth Sreenivasa { 15515cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 15525cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 15535cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 15545cabbc6bSPrashanth Sreenivasa } 15555cabbc6bSPrashanth Sreenivasa } 15565cabbc6bSPrashanth Sreenivasa 1557fa9e4066Sahrens /* 1558fa9e4066Sahrens * Prepare a virtual device for access. 1559fa9e4066Sahrens */ 1560fa9e4066Sahrens int 1561fa9e4066Sahrens vdev_open(vdev_t *vd) 1562fa9e4066Sahrens { 15638ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1564fa9e4066Sahrens int error; 1565fa9e4066Sahrens uint64_t osize = 0; 15664263d13fSGeorge Wilson uint64_t max_osize = 0; 15674263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1568ecc2d604Sbonwick uint64_t ashift = 0; 1569fa9e4066Sahrens 1570f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1571f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1572fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1573fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1574fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1575fa9e4066Sahrens 1576fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1577e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1578e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1579573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1580fa9e4066Sahrens 1581069f55e2SEric Schrock /* 1582069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1583069f55e2SEric Schrock * faulted, bail out of the open. 1584069f55e2SEric Schrock */ 15853d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 15863d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1587069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1588069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 15893d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1590069f55e2SEric Schrock vd->vdev_label_aux); 1591be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 15923d7072f8Seschrock } else if (vd->vdev_offline) { 1593fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1594ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1595be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1596fa9e4066Sahrens } 1597fa9e4066Sahrens 15984263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1599fa9e4066Sahrens 1600095bcd66SGeorge Wilson /* 1601095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1602095bcd66SGeorge Wilson * the vdev on error. 1603095bcd66SGeorge Wilson */ 1604095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1605ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 16068956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1607ea8dc4b6Seschrock 1608fa9e4066Sahrens if (error) { 16093d7072f8Seschrock if (vd->vdev_removed && 16103d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 16113d7072f8Seschrock vd->vdev_removed = B_FALSE; 16123d7072f8Seschrock 16136f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 16146f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 16156f793812SPavel Zakharov vd->vdev_stat.vs_aux); 16166f793812SPavel Zakharov } else { 16176f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 16186f793812SPavel Zakharov vd->vdev_stat.vs_aux); 16196f793812SPavel Zakharov } 1620fa9e4066Sahrens return (error); 1621fa9e4066Sahrens } 1622fa9e4066Sahrens 16233d7072f8Seschrock vd->vdev_removed = B_FALSE; 16243d7072f8Seschrock 1625096d22d4SEric Schrock /* 1626096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1627096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1628096d22d4SEric Schrock */ 1629096d22d4SEric Schrock if (vd->vdev_faulted) { 1630096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1631096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1632096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1633096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1634096d22d4SEric Schrock vd->vdev_label_aux); 1635be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1636096d22d4SEric Schrock } 1637096d22d4SEric Schrock 16383d7072f8Seschrock if (vd->vdev_degraded) { 16393d7072f8Seschrock ASSERT(vd->vdev_children == 0); 16403d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 16413d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 16423d7072f8Seschrock } else { 1643069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 16443d7072f8Seschrock } 1645fa9e4066Sahrens 164688ecc943SGeorge Wilson /* 164788ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 164888ecc943SGeorge Wilson */ 164988ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 165088ecc943SGeorge Wilson return (0); 165188ecc943SGeorge Wilson 1652573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1653ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1654ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1655ea8dc4b6Seschrock VDEV_AUX_NONE); 1656ea8dc4b6Seschrock break; 1657ea8dc4b6Seschrock } 1658573ca77eSGeorge Wilson } 1659fa9e4066Sahrens 1660fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 16614263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1662fa9e4066Sahrens 1663fa9e4066Sahrens if (vd->vdev_children == 0) { 1664fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1665ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1666ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1667be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1668fa9e4066Sahrens } 1669fa9e4066Sahrens psize = osize; 1670fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 16714263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 16724263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1673fa9e4066Sahrens } else { 1674ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1675fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1676ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1677ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1678be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1679fa9e4066Sahrens } 1680fa9e4066Sahrens psize = 0; 1681fa9e4066Sahrens asize = osize; 16824263d13fSGeorge Wilson max_asize = max_osize; 1683fa9e4066Sahrens } 1684fa9e4066Sahrens 1685fa9e4066Sahrens vd->vdev_psize = psize; 1686fa9e4066Sahrens 1687573ca77eSGeorge Wilson /* 1688c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1689573ca77eSGeorge Wilson */ 1690573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1691573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1692573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1693be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1694573ca77eSGeorge Wilson } 1695573ca77eSGeorge Wilson 1696fa9e4066Sahrens if (vd->vdev_asize == 0) { 1697fa9e4066Sahrens /* 1698fa9e4066Sahrens * This is the first-ever open, so use the computed values. 16995711d393Sloli * For compatibility, a different ashift can be requested. 1700fa9e4066Sahrens */ 1701fa9e4066Sahrens vd->vdev_asize = asize; 17024263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 17035711d393Sloli if (vd->vdev_ashift == 0) { 17045711d393Sloli vd->vdev_ashift = ashift; /* use detected value */ 17055711d393Sloli } 17065711d393Sloli if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || 17075711d393Sloli vd->vdev_ashift > ASHIFT_MAX)) { 17085711d393Sloli vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 17095711d393Sloli VDEV_AUX_BAD_ASHIFT); 17105711d393Sloli return (SET_ERROR(EDOM)); 17115711d393Sloli } 1712fa9e4066Sahrens } else { 1713fa9e4066Sahrens /* 17142384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 17152384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 17165711d393Sloli * post an event instead. 1717fa9e4066Sahrens */ 17182384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 17192384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 17205711d393Sloli zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, 17215711d393Sloli spa, vd, NULL, NULL, 0, 0); 1722fa9e4066Sahrens } 17235711d393Sloli 17244263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1725573ca77eSGeorge Wilson } 1726fa9e4066Sahrens 1727573ca77eSGeorge Wilson /* 1728c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1729c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1730c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1731c040c10cSSteven Hartland * making the additional space available. 1732c040c10cSSteven Hartland * 1733c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1734c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1735c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1736c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1737c040c10cSSteven Hartland * vdev_min_asize. 1738573ca77eSGeorge Wilson */ 1739c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1740c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1741c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1742c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1743573ca77eSGeorge Wilson vd->vdev_asize = asize; 1744fa9e4066Sahrens 1745573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1746fa9e4066Sahrens 17470a4e9518Sgw /* 17480a4e9518Sgw * Ensure we can issue some IO before declaring the 17490a4e9518Sgw * vdev open for business. 17500a4e9518Sgw */ 1751e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1752e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 175398d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 175498d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 17550a4e9518Sgw return (error); 17560a4e9518Sgw } 17570a4e9518Sgw 175881cd5c55SMatthew Ahrens /* 175981cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 1760663207adSDon Brady * 1761663207adSDon Brady * DJB - TBD these should perhaps be tracked per allocation class 1762663207adSDon Brady * (e.g. spa_min_ashift is used to round up post compression buffers) 176381cd5c55SMatthew Ahrens */ 176481cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 1765663207adSDon Brady vd->vdev_alloc_bias == VDEV_BIAS_NONE && 1766663207adSDon Brady vd->vdev_aux == NULL) { 176781cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 176881cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 176981cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 177081cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 177181cd5c55SMatthew Ahrens } 177281cd5c55SMatthew Ahrens 1773088f3894Sahrens /* 17740c06d385Sjwpoduska * If this is a leaf vdev, assess whether a resilver is needed. 17750c06d385Sjwpoduska * But don't do this if we are doing a reopen for a scrub, since 17760c06d385Sjwpoduska * this would just restart the scrub we are already doing. 1777088f3894Sahrens */ 17780c06d385Sjwpoduska if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) 17790c06d385Sjwpoduska dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); 1780088f3894Sahrens 1781fa9e4066Sahrens return (0); 1782fa9e4066Sahrens } 1783fa9e4066Sahrens 1784560e6e96Seschrock /* 1785560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 17866f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 17873d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1788560e6e96Seschrock * 1789560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1790560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1791560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1792560e6e96Seschrock * will be updated but the function will return 0. 1793560e6e96Seschrock */ 1794560e6e96Seschrock int 17956f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1796560e6e96Seschrock { 1797560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1798560e6e96Seschrock nvlist_t *label; 17996f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1800560e6e96Seschrock uint64_t state; 18016f793812SPavel Zakharov nvlist_t *nvl; 18026f793812SPavel Zakharov uint64_t txg; 1803560e6e96Seschrock 18046f793812SPavel Zakharov if (vdev_validate_skip) 18056f793812SPavel Zakharov return (0); 18066f793812SPavel Zakharov 18076f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 18086f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1809be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1810560e6e96Seschrock 1811b5989ec7Seschrock /* 1812b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1813b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1814b5989ec7Seschrock * overwrite the previous state. 1815b5989ec7Seschrock */ 18166f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 18176f793812SPavel Zakharov return (0); 1818560e6e96Seschrock 18196f793812SPavel Zakharov /* 18206f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 18216f793812SPavel Zakharov * was modified at a point after the current txg. 1822d1de72cfSPavel Zakharov * If config lock is not held do not check for the txg. spa_sync could 1823d1de72cfSPavel Zakharov * be updating the vdev's label before updating spa_last_synced_txg. 18246f793812SPavel Zakharov */ 1825d1de72cfSPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1826d1de72cfSPavel Zakharov spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 18276f793812SPavel Zakharov txg = UINT64_MAX; 18286f793812SPavel Zakharov else 18296f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1830560e6e96Seschrock 18316f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 18326f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 18336f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 1834b6bf6e15SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1835b6bf6e15SPavel Zakharov "txg %llu", (u_longlong_t)txg); 18366f793812SPavel Zakharov return (0); 18376f793812SPavel Zakharov } 18381195e687SMark J Musante 18396f793812SPavel Zakharov /* 18406f793812SPavel Zakharov * Determine if this vdev has been split off into another 18416f793812SPavel Zakharov * pool. If so, then refuse to open it. 18426f793812SPavel Zakharov */ 18436f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 18446f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 18456f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18466f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 18476f793812SPavel Zakharov nvlist_free(label); 18486f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 18496f793812SPavel Zakharov return (0); 18506f793812SPavel Zakharov } 1851560e6e96Seschrock 18526f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 18536f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18546f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18556f793812SPavel Zakharov nvlist_free(label); 18566f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18576f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 18586f793812SPavel Zakharov return (0); 18596f793812SPavel Zakharov } 18601195e687SMark J Musante 18616f793812SPavel Zakharov /* 18626f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 18636f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 18646f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 18656f793812SPavel Zakharov * cached config. The check will be performed again once we have the 18666f793812SPavel Zakharov * trusted config from the MOS. 18676f793812SPavel Zakharov */ 18686f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 18696f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18706f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18716f793812SPavel Zakharov nvlist_free(label); 18726f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 18736f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 18746f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 18756f793812SPavel Zakharov return (0); 18766f793812SPavel Zakharov } 18776f793812SPavel Zakharov 18786f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 18796f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 18806f793812SPavel Zakharov &aux_guid) != 0) 18816f793812SPavel Zakharov aux_guid = 0; 18826f793812SPavel Zakharov 18836f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 18846f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18856f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18866f793812SPavel Zakharov nvlist_free(label); 18876f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18886f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 18896f793812SPavel Zakharov return (0); 18906f793812SPavel Zakharov } 18916f793812SPavel Zakharov 18926f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 18936f793812SPavel Zakharov != 0) { 18946f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 18956f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 18966f793812SPavel Zakharov nvlist_free(label); 18976f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 18986f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 18996f793812SPavel Zakharov return (0); 19006f793812SPavel Zakharov } 19016f793812SPavel Zakharov 19026f793812SPavel Zakharov /* 19036f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 19046f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 19056f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 19066f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 19076f793812SPavel Zakharov * vdev, we can safely compare to that instead. 19086f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 19096f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 19106f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 19116f793812SPavel Zakharov * extreme rewind. 19126f793812SPavel Zakharov * 19136f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 19146f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 19156f793812SPavel Zakharov * corrupt if it is partway through a split operation. 19166f793812SPavel Zakharov */ 19176f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 19186f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 19196f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 19206f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 19216f793812SPavel Zakharov mismatch = B_TRUE; 19226f793812SPavel Zakharov } else { 19236f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 19246f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 19256f793812SPavel Zakharov mismatch = B_TRUE; 1926560e6e96Seschrock } 1927560e6e96Seschrock 19286f793812SPavel Zakharov if (mismatch) { 1929560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1930560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1931560e6e96Seschrock nvlist_free(label); 19326f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 19336f793812SPavel Zakharov "doesn't match label guid"); 19346f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 19356f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 19366f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 19376f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 19386f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 19396f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1940560e6e96Seschrock return (0); 1941560e6e96Seschrock } 19426f793812SPavel Zakharov } 1943560e6e96Seschrock 19446f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 19456f793812SPavel Zakharov &state) != 0) { 19466f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 19476f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1948560e6e96Seschrock nvlist_free(label); 19496f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 19506f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 19516f793812SPavel Zakharov return (0); 19526f793812SPavel Zakharov } 1953560e6e96Seschrock 19546f793812SPavel Zakharov nvlist_free(label); 19556f793812SPavel Zakharov 19566f793812SPavel Zakharov /* 19576f793812SPavel Zakharov * If this is a verbatim import, no need to check the 19586f793812SPavel Zakharov * state of the pool. 19596f793812SPavel Zakharov */ 19606f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 19616f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 19626f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 19636f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 19646f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 19656f793812SPavel Zakharov return (SET_ERROR(EBADF)); 19666f793812SPavel Zakharov } 19676f793812SPavel Zakharov 19686f793812SPavel Zakharov /* 19696f793812SPavel Zakharov * If we were able to open and validate a vdev that was 19706f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 19716f793812SPavel Zakharov * now. 19726f793812SPavel Zakharov */ 19736f793812SPavel Zakharov if (vd->vdev_not_present) 19746f793812SPavel Zakharov vd->vdev_not_present = 0; 19756f793812SPavel Zakharov 19766f793812SPavel Zakharov return (0); 19776f793812SPavel Zakharov } 19786f793812SPavel Zakharov 19796f793812SPavel Zakharov static void 19806f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 19816f793812SPavel Zakharov { 19826f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 19836f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 19846f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 19856f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 19866f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 19876f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 19886f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19893ee8c80cSPavel Zakharov } 19906f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 19916f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 19926f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 19936f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 19946f793812SPavel Zakharov } 19956f793812SPavel Zakharov } 1996560e6e96Seschrock 19976f793812SPavel Zakharov /* 19986f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 19996f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 20006f793812SPavel Zakharov * paths from userland config into MOS config. 20016f793812SPavel Zakharov */ 20026f793812SPavel Zakharov int 20036f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 20046f793812SPavel Zakharov { 20056f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 20066f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 20076f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 20086f793812SPavel Zakharov return (0); 20096f793812SPavel Zakharov 20106f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 20116f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 20126f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 20136f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20146f793812SPavel Zakharov } 20156f793812SPavel Zakharov 20166f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 20176f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 20186f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 20196f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 20206f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 20216f793812SPavel Zakharov } 20226f793812SPavel Zakharov 20236f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 20246f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 20256f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 20266f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 20276f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 202851ece835Seschrock } 2029560e6e96Seschrock 20306f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 20316f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 20326f793812SPavel Zakharov dvd->vdev_child[i]); 20336f793812SPavel Zakharov if (error != 0) 20346f793812SPavel Zakharov return (error); 20356f793812SPavel Zakharov } 20366f793812SPavel Zakharov 20376f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 20386f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 20396f793812SPavel Zakharov 2040560e6e96Seschrock return (0); 2041560e6e96Seschrock } 2042560e6e96Seschrock 20436f793812SPavel Zakharov static void 20446f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 20456f793812SPavel Zakharov { 20466f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 20476f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 20486f793812SPavel Zakharov 20496f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 20506f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 20516f793812SPavel Zakharov } 20526f793812SPavel Zakharov 20536f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 20546f793812SPavel Zakharov return; 20556f793812SPavel Zakharov 20566f793812SPavel Zakharov /* 20576f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 20586f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 20596f793812SPavel Zakharov * step outside of it. 20606f793812SPavel Zakharov */ 20616f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 20626f793812SPavel Zakharov 20636f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 20646f793812SPavel Zakharov return; 20656f793812SPavel Zakharov 20666f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 20676f793812SPavel Zakharov 20686f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 20696f793812SPavel Zakharov } 20706f793812SPavel Zakharov 20716f793812SPavel Zakharov /* 20726f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 20736f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 20746f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 20756f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 20766f793812SPavel Zakharov */ 20776f793812SPavel Zakharov void 20786f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 20796f793812SPavel Zakharov { 20806f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 20816f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 20826f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 20836f793812SPavel Zakharov 20846f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 20856f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 20866f793812SPavel Zakharov drvd->vdev_child[i]); 20876f793812SPavel Zakharov } 20886f793812SPavel Zakharov } 20896f793812SPavel Zakharov 2090fa9e4066Sahrens /* 2091fa9e4066Sahrens * Close a virtual device. 2092fa9e4066Sahrens */ 2093fa9e4066Sahrens void 2094fa9e4066Sahrens vdev_close(vdev_t *vd) 2095fa9e4066Sahrens { 20968ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 2097095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 20988ad4d6ddSJeff Bonwick 20998ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 21008ad4d6ddSJeff Bonwick 21011195e687SMark J Musante /* 21021195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 21031195e687SMark J Musante * going offline. 21041195e687SMark J Musante */ 2105095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 21061195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 2107095bcd66SGeorge Wilson 2108fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 2109fa9e4066Sahrens 21103d7072f8Seschrock vdev_cache_purge(vd); 2111fa9e4066Sahrens 2112560e6e96Seschrock /* 2113573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 2114560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 2115560e6e96Seschrock * it's still faulted. 2116560e6e96Seschrock */ 2117560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 2118560e6e96Seschrock 2119fa9e4066Sahrens if (vd->vdev_offline) 2120fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 2121fa9e4066Sahrens else 2122fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 2123ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2124fa9e4066Sahrens } 2125fa9e4066Sahrens 2126dcba9f3fSGeorge Wilson void 2127dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 2128dcba9f3fSGeorge Wilson { 2129dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2130dcba9f3fSGeorge Wilson 2131dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2132dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 2133dcba9f3fSGeorge Wilson return; 2134dcba9f3fSGeorge Wilson 2135dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2136dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 2137dcba9f3fSGeorge Wilson 2138dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2139dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 2140dcba9f3fSGeorge Wilson } 2141dcba9f3fSGeorge Wilson 2142dcba9f3fSGeorge Wilson void 2143dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 2144dcba9f3fSGeorge Wilson { 2145dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 2146dcba9f3fSGeorge Wilson 2147dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 2148dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 2149dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 2150dcba9f3fSGeorge Wilson 2151dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 2152dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 2153dcba9f3fSGeorge Wilson } 2154dcba9f3fSGeorge Wilson 2155095bcd66SGeorge Wilson /* 2156095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 2157095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 2158095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 2159095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 2160095bcd66SGeorge Wilson */ 2161fa9e4066Sahrens void 2162ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 2163fa9e4066Sahrens { 2164ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 2165fa9e4066Sahrens 2166e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2167ea8dc4b6Seschrock 21681195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 21691195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 2170fa9e4066Sahrens vdev_close(vd); 2171fa9e4066Sahrens (void) vdev_open(vd); 2172fa9e4066Sahrens 217339c23413Seschrock /* 217439c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 217539c23413Seschrock * Otherwise, a device with an invalid label could be successfully 217639c23413Seschrock * opened in response to vdev_reopen(). 217739c23413Seschrock */ 2178c5904d13Seschrock if (vd->vdev_aux) { 2179c5904d13Seschrock (void) vdev_validate_aux(vd); 2180e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 21816809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 2182573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 2183573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 2184c5904d13Seschrock } else { 21856f793812SPavel Zakharov (void) vdev_validate(vd); 2186c5904d13Seschrock } 218739c23413Seschrock 2188fa9e4066Sahrens /* 21893d7072f8Seschrock * Reassess parent vdev's health. 2190fa9e4066Sahrens */ 21913d7072f8Seschrock vdev_propagate_state(vd); 2192fa9e4066Sahrens } 2193fa9e4066Sahrens 2194fa9e4066Sahrens int 219599653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2196fa9e4066Sahrens { 2197fa9e4066Sahrens int error; 2198fa9e4066Sahrens 2199fa9e4066Sahrens /* 2200fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 2201fa9e4066Sahrens * For a create, however, we want to fail the request if 2202fa9e4066Sahrens * there are any components we can't open. 2203fa9e4066Sahrens */ 2204fa9e4066Sahrens error = vdev_open(vd); 2205fa9e4066Sahrens 2206fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2207fa9e4066Sahrens vdev_close(vd); 2208fa9e4066Sahrens return (error ? error : ENXIO); 2209fa9e4066Sahrens } 2210fa9e4066Sahrens 2211fa9e4066Sahrens /* 22120713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 2213fa9e4066Sahrens */ 22140713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 22150713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 221639c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2217fa9e4066Sahrens vdev_close(vd); 2218fa9e4066Sahrens return (error); 2219fa9e4066Sahrens } 2220fa9e4066Sahrens 2221fa9e4066Sahrens return (0); 2222fa9e4066Sahrens } 2223fa9e4066Sahrens 22240e34b6a7Sbonwick void 2225573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 2226fa9e4066Sahrens { 222786714001SSerapheim Dimitropoulos uint64_t asize = vd->vdev_asize; 2228a0b03b16SSerapheim Dimitropoulos uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2229b4bf0cf0SDon Brady uint64_t ms_shift; 223086714001SSerapheim Dimitropoulos 2231a3874b8bSToomas Soome /* BEGIN CSTYLED */ 2232fa9e4066Sahrens /* 2233b4bf0cf0SDon Brady * There are two dimensions to the metaslab sizing calculation: 2234b4bf0cf0SDon Brady * the size of the metaslab and the count of metaslabs per vdev. 2235b4bf0cf0SDon Brady * 2236a0b03b16SSerapheim Dimitropoulos * The default values used below are a good balance between memory 2237a0b03b16SSerapheim Dimitropoulos * usage (larger metaslab size means more memory needed for loaded 2238a0b03b16SSerapheim Dimitropoulos * metaslabs; more metaslabs means more memory needed for the 2239a0b03b16SSerapheim Dimitropoulos * metaslab_t structs), metaslab load time (larger metaslabs take 2240a0b03b16SSerapheim Dimitropoulos * longer to load), and metaslab sync time (more metaslabs means 2241a0b03b16SSerapheim Dimitropoulos * more time spent syncing all of them). 2242a0b03b16SSerapheim Dimitropoulos * 2243a0b03b16SSerapheim Dimitropoulos * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2244a0b03b16SSerapheim Dimitropoulos * The range of the dimensions are as follows: 2245a0b03b16SSerapheim Dimitropoulos * 2246a0b03b16SSerapheim Dimitropoulos * 2^29 <= ms_size <= 2^34 2247b4bf0cf0SDon Brady * 16 <= ms_count <= 131,072 2248b4bf0cf0SDon Brady * 2249b4bf0cf0SDon Brady * On the lower end of vdev sizes, we aim for metaslabs sizes of 2250b4bf0cf0SDon Brady * at least 512MB (2^29) to minimize fragmentation effects when 2251b4bf0cf0SDon Brady * testing with smaller devices. However, the count constraint 2252b4bf0cf0SDon Brady * of at least 16 metaslabs will override this minimum size goal. 2253b4bf0cf0SDon Brady * 2254b4bf0cf0SDon Brady * On the upper end of vdev sizes, we aim for a maximum metaslab 2255a0b03b16SSerapheim Dimitropoulos * size of 16GB. However, we will cap the total count to 2^17 2256a0b03b16SSerapheim Dimitropoulos * metaslabs to keep our memory footprint in check and let the 2257a0b03b16SSerapheim Dimitropoulos * metaslab size grow from there if that limit is hit. 2258b4bf0cf0SDon Brady * 2259b4bf0cf0SDon Brady * The net effect of applying above constrains is summarized below. 2260b4bf0cf0SDon Brady * 2261663207adSDon Brady * vdev size metaslab count 2262a0b03b16SSerapheim Dimitropoulos * --------------|----------------- 2263663207adSDon Brady * < 8GB ~16 2264663207adSDon Brady * 8GB - 100GB one per 512MB 2265663207adSDon Brady * 100GB - 3TB ~200 2266663207adSDon Brady * 3TB - 2PB one per 16GB 2267663207adSDon Brady * > 2PB ~131,072 2268a0b03b16SSerapheim Dimitropoulos * -------------------------------- 2269a0b03b16SSerapheim Dimitropoulos * 2270a0b03b16SSerapheim Dimitropoulos * Finally, note that all of the above calculate the initial 2271a0b03b16SSerapheim Dimitropoulos * number of metaslabs. Expanding a top-level vdev will result 2272a0b03b16SSerapheim Dimitropoulos * in additional metaslabs being allocated making it possible 2273a0b03b16SSerapheim Dimitropoulos * to exceed the zfs_vdev_ms_count_limit. 2274a0b03b16SSerapheim Dimitropoulos */ 2275a3874b8bSToomas Soome /* END CSTYLED */ 2276a0b03b16SSerapheim Dimitropoulos 2277a0b03b16SSerapheim Dimitropoulos if (ms_count < zfs_vdev_min_ms_count) 2278a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2279a0b03b16SSerapheim Dimitropoulos else if (ms_count > zfs_vdev_default_ms_count) 2280a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2281b4bf0cf0SDon Brady else 2282a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_default_ms_shift; 2283b4bf0cf0SDon Brady 2284b4bf0cf0SDon Brady if (ms_shift < SPA_MAXBLOCKSHIFT) { 2285b4bf0cf0SDon Brady ms_shift = SPA_MAXBLOCKSHIFT; 2286a0b03b16SSerapheim Dimitropoulos } else if (ms_shift > zfs_vdev_max_ms_shift) { 2287a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_max_ms_shift; 2288b4bf0cf0SDon Brady /* cap the total count to constrain memory footprint */ 2289a0b03b16SSerapheim Dimitropoulos if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2290a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 229186714001SSerapheim Dimitropoulos } 229286714001SSerapheim Dimitropoulos 229386714001SSerapheim Dimitropoulos vd->vdev_ms_shift = ms_shift; 229486714001SSerapheim Dimitropoulos ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2295fa9e4066Sahrens } 2296fa9e4066Sahrens 2297fa9e4066Sahrens void 2298ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2299fa9e4066Sahrens { 2300ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 23015cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 23025cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 2303ecc2d604Sbonwick ASSERT(ISP2(flags)); 2304f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 2305fa9e4066Sahrens 2306ecc2d604Sbonwick if (flags & VDD_METASLAB) 2307ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2308ecc2d604Sbonwick 2309ecc2d604Sbonwick if (flags & VDD_DTL) 2310ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2311ecc2d604Sbonwick 2312ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2313fa9e4066Sahrens } 2314fa9e4066Sahrens 23150713e232SGeorge Wilson void 23160713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 23170713e232SGeorge Wilson { 23180713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 23190713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 23200713e232SGeorge Wilson 23210713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 23220713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 23230713e232SGeorge Wilson } 23240713e232SGeorge Wilson 23258ad4d6ddSJeff Bonwick /* 23268ad4d6ddSJeff Bonwick * DTLs. 23278ad4d6ddSJeff Bonwick * 23288ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 23299fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 23308ad4d6ddSJeff Bonwick * 23318ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 23328ad4d6ddSJeff Bonwick * 23338ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 23348ad4d6ddSJeff Bonwick * 23358ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 23368ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 23378ad4d6ddSJeff Bonwick * txgs that was scrubbed. 23388ad4d6ddSJeff Bonwick * 23398ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 23408ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 23418ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 23428ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 23438ad4d6ddSJeff Bonwick * determine whether a device can be detached. 23448ad4d6ddSJeff Bonwick * 23458ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 23468ad4d6ddSJeff Bonwick * either has the data or it doesn't. 23478ad4d6ddSJeff Bonwick * 23488ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 23498ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 23508ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 23518ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 23528ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 23538ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 23548ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 23558ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 23568ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 23578ad4d6ddSJeff Bonwick * 23588ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 23598ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 23608ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 23618ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 23628ad4d6ddSJeff Bonwick */ 2363fa9e4066Sahrens void 23648ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2365fa9e4066Sahrens { 23660713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23678ad4d6ddSJeff Bonwick 23688ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23698ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2370f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 23718ad4d6ddSJeff Bonwick 23725cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 23730713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 23740713e232SGeorge Wilson range_tree_add(rt, txg, size); 23755cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2376fa9e4066Sahrens } 2377fa9e4066Sahrens 23788ad4d6ddSJeff Bonwick boolean_t 23798ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2380fa9e4066Sahrens { 23810713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 23828ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2383fa9e4066Sahrens 23848ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 23858ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2386fa9e4066Sahrens 23875cabbc6bSPrashanth Sreenivasa /* 23885cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 23895cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 23905cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 23915cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 23925cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 23935cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 23945cabbc6bSPrashanth Sreenivasa */ 23955cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 23965cabbc6bSPrashanth Sreenivasa return (B_FALSE); 23975cabbc6bSPrashanth Sreenivasa 23985cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 239986714001SSerapheim Dimitropoulos if (!range_tree_is_empty(rt)) 24000713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 24015cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2402fa9e4066Sahrens 2403fa9e4066Sahrens return (dirty); 2404fa9e4066Sahrens } 2405fa9e4066Sahrens 24068ad4d6ddSJeff Bonwick boolean_t 24078ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 24088ad4d6ddSJeff Bonwick { 24090713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 24108ad4d6ddSJeff Bonwick boolean_t empty; 24118ad4d6ddSJeff Bonwick 24125cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 241386714001SSerapheim Dimitropoulos empty = range_tree_is_empty(rt); 24145cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 24158ad4d6ddSJeff Bonwick 24168ad4d6ddSJeff Bonwick return (empty); 24178ad4d6ddSJeff Bonwick } 24188ad4d6ddSJeff Bonwick 2419a3874b8bSToomas Soome /* 2420a3874b8bSToomas Soome * Returns B_TRUE if vdev determines offset needs to be resilvered. 2421a3874b8bSToomas Soome */ 2422a3874b8bSToomas Soome boolean_t 2423a3874b8bSToomas Soome vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) 2424a3874b8bSToomas Soome { 2425a3874b8bSToomas Soome ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2426a3874b8bSToomas Soome 2427a3874b8bSToomas Soome if (vd->vdev_ops->vdev_op_need_resilver == NULL || 2428a3874b8bSToomas Soome vd->vdev_ops->vdev_op_leaf) 2429a3874b8bSToomas Soome return (B_TRUE); 2430a3874b8bSToomas Soome 2431a3874b8bSToomas Soome return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); 2432a3874b8bSToomas Soome } 2433a3874b8bSToomas Soome 2434b4952e17SGeorge Wilson /* 2435b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2436b4952e17SGeorge Wilson */ 2437b4952e17SGeorge Wilson static uint64_t 2438b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2439b4952e17SGeorge Wilson { 2440b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 24410713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2442b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2443b4952e17SGeorge Wilson 24444d7988d6SPaul Dagnelie return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); 2445b4952e17SGeorge Wilson } 2446b4952e17SGeorge Wilson 2447b4952e17SGeorge Wilson /* 2448b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2449b4952e17SGeorge Wilson */ 2450b4952e17SGeorge Wilson static uint64_t 2451b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 2452b4952e17SGeorge Wilson { 2453b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 24540713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2455b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2456b4952e17SGeorge Wilson 24574d7988d6SPaul Dagnelie return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); 2458b4952e17SGeorge Wilson } 2459b4952e17SGeorge Wilson 2460b4952e17SGeorge Wilson /* 2461b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2462b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2463b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2464b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2465b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2466b4952e17SGeorge Wilson * excise the DTLs. 2467b4952e17SGeorge Wilson */ 2468b4952e17SGeorge Wilson static boolean_t 2469b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2470b4952e17SGeorge Wilson { 2471b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2472b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2473b4952e17SGeorge Wilson 2474b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2475b4952e17SGeorge Wilson 24762d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 24772d2f193aSMatthew Ahrens return (B_FALSE); 24782d2f193aSMatthew Ahrens 2479e4c795beSTom Caputi if (vd->vdev_resilver_deferred) 2480e4c795beSTom Caputi return (B_FALSE); 2481e4c795beSTom Caputi 2482b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 248386714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2484b4952e17SGeorge Wilson return (B_TRUE); 2485b4952e17SGeorge Wilson 2486b4952e17SGeorge Wilson /* 2487b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2488b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2489b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2490b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2491b4952e17SGeorge Wilson * for excision. 2492b4952e17SGeorge Wilson */ 2493b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2494b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2495b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2496b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2497b4952e17SGeorge Wilson return (B_TRUE); 2498b4952e17SGeorge Wilson } 2499b4952e17SGeorge Wilson return (B_FALSE); 2500b4952e17SGeorge Wilson } 2501b4952e17SGeorge Wilson 2502fa9e4066Sahrens /* 2503fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2504fa9e4066Sahrens */ 2505fa9e4066Sahrens void 2506fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2507fa9e4066Sahrens { 2508ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 25098ad4d6ddSJeff Bonwick avl_tree_t reftree; 25108ad4d6ddSJeff Bonwick int minref; 2511fa9e4066Sahrens 25128ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2513fa9e4066Sahrens 25148ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 25158ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 25168ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 25178ad4d6ddSJeff Bonwick 25185cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 25198ad4d6ddSJeff Bonwick return; 25208ad4d6ddSJeff Bonwick 25218ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 25223f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2523*165c5c6fSJohn Poduska boolean_t wasempty = B_TRUE; 25243f9d6ad7SLin Ling 2525fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2526b4952e17SGeorge Wilson 2527*165c5c6fSJohn Poduska /* 2528*165c5c6fSJohn Poduska * If requested, pretend the scan completed cleanly. 2529*165c5c6fSJohn Poduska */ 2530*165c5c6fSJohn Poduska if (zfs_scan_ignore_errors && scn) 2531*165c5c6fSJohn Poduska scn->scn_phys.scn_errors = 0; 2532*165c5c6fSJohn Poduska 2533*165c5c6fSJohn Poduska if (scrub_txg != 0 && 2534*165c5c6fSJohn Poduska !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 2535*165c5c6fSJohn Poduska wasempty = B_FALSE; 2536*165c5c6fSJohn Poduska zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d " 2537*165c5c6fSJohn Poduska "dtl:%llu/%llu errors:%llu", 2538*165c5c6fSJohn Poduska (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg, 2539*165c5c6fSJohn Poduska (u_longlong_t)scrub_txg, spa->spa_scrub_started, 2540*165c5c6fSJohn Poduska (u_longlong_t)vdev_dtl_min(vd), 2541*165c5c6fSJohn Poduska (u_longlong_t)vdev_dtl_max(vd), 2542*165c5c6fSJohn Poduska (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0)); 2543*165c5c6fSJohn Poduska } 2544*165c5c6fSJohn Poduska 2545b4952e17SGeorge Wilson /* 2546b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2547b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2548b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2549b4952e17SGeorge Wilson * the entire duration of this scan. 2550b4952e17SGeorge Wilson */ 2551088f3894Sahrens if (scrub_txg != 0 && 25523f9d6ad7SLin Ling (spa->spa_scrub_started || 2553b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2554b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2555088f3894Sahrens /* 2556088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2557088f3894Sahrens * did it without rebooting, then the scrub dtl 2558088f3894Sahrens * will be valid, so excise the old region and 2559088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2560088f3894Sahrens * dtl as-is if there was an error. 25618ad4d6ddSJeff Bonwick * 25628ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 25638ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 25648ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 25658ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 25668ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 25678ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 25688ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 25698ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 25708ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2571088f3894Sahrens */ 25720713e232SGeorge Wilson space_reftree_create(&reftree); 25730713e232SGeorge Wilson space_reftree_add_map(&reftree, 25740713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25750713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 25760713e232SGeorge Wilson space_reftree_add_map(&reftree, 25770713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 25780713e232SGeorge Wilson space_reftree_generate_map(&reftree, 25790713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 25800713e232SGeorge Wilson space_reftree_destroy(&reftree); 2581*165c5c6fSJohn Poduska 2582*165c5c6fSJohn Poduska if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { 2583*165c5c6fSJohn Poduska zfs_dbgmsg("update DTL_MISSING:%llu/%llu", 2584*165c5c6fSJohn Poduska (u_longlong_t)vdev_dtl_min(vd), 2585*165c5c6fSJohn Poduska (u_longlong_t)vdev_dtl_max(vd)); 2586*165c5c6fSJohn Poduska } else if (!wasempty) { 2587*165c5c6fSJohn Poduska zfs_dbgmsg("DTL_MISSING is now empty"); 2588*165c5c6fSJohn Poduska } 2589fa9e4066Sahrens } 25900713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 25910713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 25920713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2593fa9e4066Sahrens if (scrub_done) 25940713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 25950713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 25968ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 25970713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 25988ad4d6ddSJeff Bonwick else 25990713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 26000713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2601b4952e17SGeorge Wilson 2602b4952e17SGeorge Wilson /* 2603b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2604b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2605b4952e17SGeorge Wilson */ 2606b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 260786714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 260886714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2609b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2610b4952e17SGeorge Wilson 2611fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2612088f3894Sahrens 2613ecc2d604Sbonwick if (txg != 0) 2614ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2615fa9e4066Sahrens return; 2616fa9e4066Sahrens } 2617fa9e4066Sahrens 2618fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 26198ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 262099bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 262199bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 26228ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 26238ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 26248ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 26258ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 26268ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 26278ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 26288ad4d6ddSJeff Bonwick else 26298ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 26300713e232SGeorge Wilson space_reftree_create(&reftree); 26318ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 26328ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 26338ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 26340713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 26358ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 26368ad4d6ddSJeff Bonwick } 26370713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 26380713e232SGeorge Wilson space_reftree_destroy(&reftree); 2639fa9e4066Sahrens } 26408ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2641fa9e4066Sahrens } 2642fa9e4066Sahrens 26430713e232SGeorge Wilson int 2644fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2645fa9e4066Sahrens { 2646fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2647ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 26480713e232SGeorge Wilson int error = 0; 2649fa9e4066Sahrens 26500713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 26515cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2652fa9e4066Sahrens 26530713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 26545cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 26550713e232SGeorge Wilson if (error) 26560713e232SGeorge Wilson return (error); 26570713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2658fa9e4066Sahrens 26590713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 26600713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 26610713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 26620713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2663fa9e4066Sahrens 26640713e232SGeorge Wilson return (error); 26650713e232SGeorge Wilson } 26660713e232SGeorge Wilson 26670713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 26680713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 26690713e232SGeorge Wilson if (error != 0) 26700713e232SGeorge Wilson break; 26710713e232SGeorge Wilson } 2672fa9e4066Sahrens 2673fa9e4066Sahrens return (error); 2674fa9e4066Sahrens } 2675fa9e4066Sahrens 2676663207adSDon Brady static void 2677663207adSDon Brady vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) 2678663207adSDon Brady { 2679663207adSDon Brady spa_t *spa = vd->vdev_spa; 2680663207adSDon Brady objset_t *mos = spa->spa_meta_objset; 2681663207adSDon Brady vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 2682663207adSDon Brady const char *string; 2683663207adSDon Brady 2684663207adSDon Brady ASSERT(alloc_bias != VDEV_BIAS_NONE); 2685663207adSDon Brady 2686663207adSDon Brady string = 2687663207adSDon Brady (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : 2688663207adSDon Brady (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : 2689663207adSDon Brady (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; 2690663207adSDon Brady 2691663207adSDon Brady ASSERT(string != NULL); 2692663207adSDon Brady VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 2693663207adSDon Brady 1, strlen(string) + 1, string, tx)); 2694663207adSDon Brady 2695663207adSDon Brady if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { 2696663207adSDon Brady spa_activate_allocation_classes(spa, tx); 2697663207adSDon Brady } 2698663207adSDon Brady } 2699663207adSDon Brady 2700215198a6SJoe Stein void 2701215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2702215198a6SJoe Stein { 2703215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2704215198a6SJoe Stein 2705215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2706215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2707215198a6SJoe Stein zapobj, tx)); 2708215198a6SJoe Stein } 2709215198a6SJoe Stein 2710215198a6SJoe Stein uint64_t 2711215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2712215198a6SJoe Stein { 2713215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2714215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2715215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2716215198a6SJoe Stein 2717215198a6SJoe Stein ASSERT(zap != 0); 2718215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2719215198a6SJoe Stein zap, tx)); 2720215198a6SJoe Stein 2721215198a6SJoe Stein return (zap); 2722215198a6SJoe Stein } 2723215198a6SJoe Stein 2724215198a6SJoe Stein void 2725215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2726215198a6SJoe Stein { 2727215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2728215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2729215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2730215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2731215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2732215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2733215198a6SJoe Stein } 2734215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2735215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2736663207adSDon Brady if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) 2737663207adSDon Brady vdev_zap_allocation_data(vd, tx); 2738215198a6SJoe Stein } 2739215198a6SJoe Stein } 2740663207adSDon Brady 2741215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2742215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2743215198a6SJoe Stein } 2744215198a6SJoe Stein } 2745215198a6SJoe Stein 2746fa9e4066Sahrens void 2747fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2748fa9e4066Sahrens { 2749fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 27500713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2751ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 27520713e232SGeorge Wilson range_tree_t *rtsync; 2753fa9e4066Sahrens dmu_tx_t *tx; 27540713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2755fa9e4066Sahrens 27565cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 27570713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 275888ecc943SGeorge Wilson 2759fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2760fa9e4066Sahrens 27610713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 27620713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 27630713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 27640713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 27650713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 27660713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2767215198a6SJoe Stein 2768215198a6SJoe Stein /* 2769215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2770215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2771215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2772215198a6SJoe Stein */ 2773215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2774215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2775215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2776215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2777215198a6SJoe Stein } 2778215198a6SJoe Stein 2779fa9e4066Sahrens dmu_tx_commit(tx); 2780fa9e4066Sahrens return; 2781fa9e4066Sahrens } 2782fa9e4066Sahrens 27830713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 27840713e232SGeorge Wilson uint64_t new_object; 27850713e232SGeorge Wilson 2786814dcd43SSerapheim Dimitropoulos new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); 27870713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 27880713e232SGeorge Wilson 27890713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 27905cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 27910713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2792fa9e4066Sahrens } 2793fa9e4066Sahrens 27944d7988d6SPaul Dagnelie rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); 2795fa9e4066Sahrens 2796fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 27970713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2798fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2799fa9e4066Sahrens 2800814dcd43SSerapheim Dimitropoulos space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); 280117f11284SSerapheim Dimitropoulos space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 28020713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2803fa9e4066Sahrens 28040713e232SGeorge Wilson range_tree_destroy(rtsync); 2805fa9e4066Sahrens 28060713e232SGeorge Wilson /* 28070713e232SGeorge Wilson * If the object for the space map has changed then dirty 28080713e232SGeorge Wilson * the top level so that we update the config. 28090713e232SGeorge Wilson */ 28100713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 28113ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 28123ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 28133ee8c80cSPavel Zakharov (u_longlong_t)object, 28143ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 28150713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 28160713e232SGeorge Wilson } 2817fa9e4066Sahrens 2818fa9e4066Sahrens dmu_tx_commit(tx); 2819fa9e4066Sahrens } 2820fa9e4066Sahrens 28218ad4d6ddSJeff Bonwick /* 28228ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 28238ad4d6ddSJeff Bonwick * without losing data. 28248ad4d6ddSJeff Bonwick */ 28258ad4d6ddSJeff Bonwick boolean_t 28268ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 28278ad4d6ddSJeff Bonwick { 28288ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 28298ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 28308ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 28318ad4d6ddSJeff Bonwick boolean_t required; 28328ad4d6ddSJeff Bonwick 28338ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 28348ad4d6ddSJeff Bonwick 28358ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 28368ad4d6ddSJeff Bonwick return (B_TRUE); 28378ad4d6ddSJeff Bonwick 28388ad4d6ddSJeff Bonwick /* 28398ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 28408ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 28418ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 28428ad4d6ddSJeff Bonwick */ 28438ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 28448ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28458ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 28468ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 28478ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 28488ad4d6ddSJeff Bonwick 2849cb04b873SMark J Musante if (!required && zio_injection_enabled) 2850cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2851cb04b873SMark J Musante 28528ad4d6ddSJeff Bonwick return (required); 28538ad4d6ddSJeff Bonwick } 28548ad4d6ddSJeff Bonwick 2855088f3894Sahrens /* 2856088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2857088f3894Sahrens */ 2858088f3894Sahrens boolean_t 2859088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2860088f3894Sahrens { 2861088f3894Sahrens boolean_t needed = B_FALSE; 2862088f3894Sahrens uint64_t thismin = UINT64_MAX; 2863088f3894Sahrens uint64_t thismax = 0; 2864088f3894Sahrens 2865088f3894Sahrens if (vd->vdev_children == 0) { 2866088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 286786714001SSerapheim Dimitropoulos if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 28688ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2869088f3894Sahrens 2870b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2871b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2872088f3894Sahrens needed = B_TRUE; 2873088f3894Sahrens } 2874088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2875088f3894Sahrens } else { 28768ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2877088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2878088f3894Sahrens uint64_t cmin, cmax; 2879088f3894Sahrens 2880088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2881088f3894Sahrens thismin = MIN(thismin, cmin); 2882088f3894Sahrens thismax = MAX(thismax, cmax); 2883088f3894Sahrens needed = B_TRUE; 2884088f3894Sahrens } 2885088f3894Sahrens } 2886088f3894Sahrens } 2887088f3894Sahrens 2888088f3894Sahrens if (needed && minp) { 2889088f3894Sahrens *minp = thismin; 2890088f3894Sahrens *maxp = thismax; 2891088f3894Sahrens } 2892088f3894Sahrens return (needed); 2893088f3894Sahrens } 2894088f3894Sahrens 289586714001SSerapheim Dimitropoulos /* 289686714001SSerapheim Dimitropoulos * Gets the checkpoint space map object from the vdev's ZAP. 289786714001SSerapheim Dimitropoulos * Returns the spacemap object, or 0 if it wasn't in the ZAP 289886714001SSerapheim Dimitropoulos * or the ZAP doesn't exist yet. 289986714001SSerapheim Dimitropoulos */ 290086714001SSerapheim Dimitropoulos int 290186714001SSerapheim Dimitropoulos vdev_checkpoint_sm_object(vdev_t *vd) 290286714001SSerapheim Dimitropoulos { 290386714001SSerapheim Dimitropoulos ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 290486714001SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) { 290586714001SSerapheim Dimitropoulos return (0); 290686714001SSerapheim Dimitropoulos } 290786714001SSerapheim Dimitropoulos 290886714001SSerapheim Dimitropoulos uint64_t sm_obj = 0; 290986714001SSerapheim Dimitropoulos int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 291086714001SSerapheim Dimitropoulos VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 291186714001SSerapheim Dimitropoulos 291286714001SSerapheim Dimitropoulos ASSERT(err == 0 || err == ENOENT); 291386714001SSerapheim Dimitropoulos 291486714001SSerapheim Dimitropoulos return (sm_obj); 291586714001SSerapheim Dimitropoulos } 291686714001SSerapheim Dimitropoulos 29175cabbc6bSPrashanth Sreenivasa int 2918ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2919fa9e4066Sahrens { 29205cabbc6bSPrashanth Sreenivasa int error = 0; 2921fa9e4066Sahrens /* 2922fa9e4066Sahrens * Recursively load all children. 2923fa9e4066Sahrens */ 29245cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 29255cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 29265cabbc6bSPrashanth Sreenivasa if (error != 0) { 29275cabbc6bSPrashanth Sreenivasa return (error); 29285cabbc6bSPrashanth Sreenivasa } 29295cabbc6bSPrashanth Sreenivasa } 29305cabbc6bSPrashanth Sreenivasa 29315cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2932fa9e4066Sahrens 2933663207adSDon Brady /* 2934663207adSDon Brady * On spa_load path, grab the allocation bias from our zap 2935663207adSDon Brady */ 2936663207adSDon Brady if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { 2937663207adSDon Brady spa_t *spa = vd->vdev_spa; 2938663207adSDon Brady char bias_str[64]; 2939663207adSDon Brady 2940663207adSDon Brady if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, 2941663207adSDon Brady VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), 2942663207adSDon Brady bias_str) == 0) { 2943663207adSDon Brady ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); 2944663207adSDon Brady vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); 2945663207adSDon Brady } 2946663207adSDon Brady } 2947663207adSDon Brady 2948fa9e4066Sahrens /* 29490e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2950fa9e4066Sahrens */ 29515cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 2952663207adSDon Brady vdev_metaslab_group_create(vd); 2953663207adSDon Brady 29545cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 29555cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29565cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29573ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 29583ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 29593ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 29605cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 2961555d674dSSerapheim Dimitropoulos } 2962555d674dSSerapheim Dimitropoulos 2963555d674dSSerapheim Dimitropoulos error = vdev_metaslab_init(vd, 0); 2964555d674dSSerapheim Dimitropoulos if (error != 0) { 29653ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 29663ee8c80cSPavel Zakharov "[error=%d]", error); 29675cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 29685cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 29695cabbc6bSPrashanth Sreenivasa return (error); 29705cabbc6bSPrashanth Sreenivasa } 297186714001SSerapheim Dimitropoulos 297286714001SSerapheim Dimitropoulos uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 297386714001SSerapheim Dimitropoulos if (checkpoint_sm_obj != 0) { 297486714001SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 297586714001SSerapheim Dimitropoulos ASSERT(vd->vdev_asize != 0); 297686714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 297786714001SSerapheim Dimitropoulos 2978555d674dSSerapheim Dimitropoulos error = space_map_open(&vd->vdev_checkpoint_sm, 297986714001SSerapheim Dimitropoulos mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2980555d674dSSerapheim Dimitropoulos vd->vdev_ashift); 2981555d674dSSerapheim Dimitropoulos if (error != 0) { 298286714001SSerapheim Dimitropoulos vdev_dbgmsg(vd, "vdev_load: space_map_open " 298386714001SSerapheim Dimitropoulos "failed for checkpoint spacemap (obj %llu) " 298486714001SSerapheim Dimitropoulos "[error=%d]", 298586714001SSerapheim Dimitropoulos (u_longlong_t)checkpoint_sm_obj, error); 298686714001SSerapheim Dimitropoulos return (error); 298786714001SSerapheim Dimitropoulos } 298886714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 298986714001SSerapheim Dimitropoulos 299086714001SSerapheim Dimitropoulos /* 299186714001SSerapheim Dimitropoulos * Since the checkpoint_sm contains free entries 2992555d674dSSerapheim Dimitropoulos * exclusively we can use space_map_allocated() to 2993555d674dSSerapheim Dimitropoulos * indicate the cumulative checkpointed space that 2994555d674dSSerapheim Dimitropoulos * has been freed. 299586714001SSerapheim Dimitropoulos */ 299686714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space = 2997555d674dSSerapheim Dimitropoulos -space_map_allocated(vd->vdev_checkpoint_sm); 299886714001SSerapheim Dimitropoulos vd->vdev_spa->spa_checkpoint_info.sci_dspace += 299986714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space; 300086714001SSerapheim Dimitropoulos } 30015cabbc6bSPrashanth Sreenivasa } 3002fa9e4066Sahrens 3003fa9e4066Sahrens /* 3004fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 3005fa9e4066Sahrens */ 30065cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 3007560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 3008560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 30093ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 30103ee8c80cSPavel Zakharov "[error=%d]", error); 30115cabbc6bSPrashanth Sreenivasa return (error); 30125cabbc6bSPrashanth Sreenivasa } 30135cabbc6bSPrashanth Sreenivasa 30145cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 30155cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 30165cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 30175cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 301886714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 30195cabbc6bSPrashanth Sreenivasa 30205cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 30215cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 30225cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 30235cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 30243ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 30253ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 30263ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 30275cabbc6bSPrashanth Sreenivasa return (error); 30285cabbc6bSPrashanth Sreenivasa } 30295cabbc6bSPrashanth Sreenivasa } 30305cabbc6bSPrashanth Sreenivasa 30315cabbc6bSPrashanth Sreenivasa return (0); 3032fa9e4066Sahrens } 3033fa9e4066Sahrens 303499653d4eSeschrock /* 3035fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 3036fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 3037fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 3038fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 3039fa94a07fSbrendan * repurposed to another pool. 304099653d4eSeschrock */ 304199653d4eSeschrock int 3042fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 304399653d4eSeschrock { 304499653d4eSeschrock nvlist_t *label; 304599653d4eSeschrock uint64_t guid, version; 304699653d4eSeschrock uint64_t state; 304799653d4eSeschrock 3048e14bb325SJeff Bonwick if (!vdev_readable(vd)) 3049c5904d13Seschrock return (0); 3050c5904d13Seschrock 3051dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 305299653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 305399653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 305499653d4eSeschrock return (-1); 305599653d4eSeschrock } 305699653d4eSeschrock 305799653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 3058ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 305999653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 306099653d4eSeschrock guid != vd->vdev_guid || 306199653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 306299653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 306399653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 306499653d4eSeschrock nvlist_free(label); 306599653d4eSeschrock return (-1); 306699653d4eSeschrock } 306799653d4eSeschrock 306899653d4eSeschrock /* 306999653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 307099653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 307199653d4eSeschrock */ 307299653d4eSeschrock nvlist_free(label); 307399653d4eSeschrock return (0); 307499653d4eSeschrock } 307599653d4eSeschrock 3076814dcd43SSerapheim Dimitropoulos static void 3077814dcd43SSerapheim Dimitropoulos vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) 3078814dcd43SSerapheim Dimitropoulos { 3079814dcd43SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 3080814dcd43SSerapheim Dimitropoulos 3081814dcd43SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) 3082814dcd43SSerapheim Dimitropoulos return; 3083814dcd43SSerapheim Dimitropoulos 3084814dcd43SSerapheim Dimitropoulos uint64_t object = 0; 3085814dcd43SSerapheim Dimitropoulos int err = zap_lookup(mos, vd->vdev_top_zap, 3086814dcd43SSerapheim Dimitropoulos VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); 3087814dcd43SSerapheim Dimitropoulos if (err == ENOENT) 3088814dcd43SSerapheim Dimitropoulos return; 3089814dcd43SSerapheim Dimitropoulos 3090814dcd43SSerapheim Dimitropoulos VERIFY0(dmu_object_free(mos, object, tx)); 3091814dcd43SSerapheim Dimitropoulos VERIFY0(zap_remove(mos, vd->vdev_top_zap, 3092814dcd43SSerapheim Dimitropoulos VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); 3093814dcd43SSerapheim Dimitropoulos } 3094814dcd43SSerapheim Dimitropoulos 30955cabbc6bSPrashanth Sreenivasa /* 30965cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 30975cabbc6bSPrashanth Sreenivasa * that points to them. 30985cabbc6bSPrashanth Sreenivasa */ 309988ecc943SGeorge Wilson void 31005cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 31015cabbc6bSPrashanth Sreenivasa { 31025cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 31035cabbc6bSPrashanth Sreenivasa return; 31045cabbc6bSPrashanth Sreenivasa 31055cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 31065cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 31075cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 31085cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 31095cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 31105cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 31115cabbc6bSPrashanth Sreenivasa 31125cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 31135cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 31145cabbc6bSPrashanth Sreenivasa if (smobj == 0) 31155cabbc6bSPrashanth Sreenivasa continue; 31165cabbc6bSPrashanth Sreenivasa 31175cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 31185cabbc6bSPrashanth Sreenivasa } 31195cabbc6bSPrashanth Sreenivasa 31205cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 31215cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 3122814dcd43SSerapheim Dimitropoulos vdev_destroy_ms_flush_data(vd, tx); 31235cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 31245cabbc6bSPrashanth Sreenivasa } 31255cabbc6bSPrashanth Sreenivasa 31265cabbc6bSPrashanth Sreenivasa static void 31274e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 312888ecc943SGeorge Wilson { 312988ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 313088ecc943SGeorge Wilson 31314e75ba68SSerapheim Dimitropoulos ASSERT(vd->vdev_islog); 3132215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 3133215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 313488ecc943SGeorge Wilson 31354e75ba68SSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 3136215198a6SJoe Stein 31374e75ba68SSerapheim Dimitropoulos vdev_destroy_spacemaps(vd, tx); 31384e75ba68SSerapheim Dimitropoulos if (vd->vdev_top_zap != 0) { 3139215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 3140215198a6SJoe Stein vd->vdev_top_zap = 0; 3141215198a6SJoe Stein } 31424e75ba68SSerapheim Dimitropoulos 314388ecc943SGeorge Wilson dmu_tx_commit(tx); 314488ecc943SGeorge Wilson } 314588ecc943SGeorge Wilson 3146fa9e4066Sahrens void 3147fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 3148fa9e4066Sahrens { 3149fa9e4066Sahrens metaslab_t *msp; 315080eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 3151fa9e4066Sahrens 31525cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 315388ecc943SGeorge Wilson 3154094e47e9SGeorge Wilson while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 3155094e47e9SGeorge Wilson != NULL) 3156fa9e4066Sahrens metaslab_sync_done(msp, txg); 315780eb36f2SGeorge Wilson 315880eb36f2SGeorge Wilson if (reassess) 315980eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 3160fa9e4066Sahrens } 3161fa9e4066Sahrens 3162fa9e4066Sahrens void 3163fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 3164fa9e4066Sahrens { 3165fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3166fa9e4066Sahrens vdev_t *lvd; 3167fa9e4066Sahrens metaslab_t *msp; 3168fa9e4066Sahrens 3169555d674dSSerapheim Dimitropoulos ASSERT3U(txg, ==, spa->spa_syncing_txg); 3170555d674dSSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 31715cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 31725cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 31735cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 317488ecc943SGeorge Wilson 31755cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 31765cabbc6bSPrashanth Sreenivasa 31775cabbc6bSPrashanth Sreenivasa /* 31785cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 31795cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 31805cabbc6bSPrashanth Sreenivasa */ 31815cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 31825cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 31835cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 3184555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 31855cabbc6bSPrashanth Sreenivasa return; 31865cabbc6bSPrashanth Sreenivasa } 31875cabbc6bSPrashanth Sreenivasa } 31885cabbc6bSPrashanth Sreenivasa 31895cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 31905cabbc6bSPrashanth Sreenivasa 31915cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 31925cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 3193ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 31945cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 3195ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 3196ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 3197ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 3198ecc2d604Sbonwick vdev_config_dirty(vd); 3199ecc2d604Sbonwick } 3200fa9e4066Sahrens 3201ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 3202fa9e4066Sahrens metaslab_sync(msp, txg); 3203ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 3204ecc2d604Sbonwick } 3205fa9e4066Sahrens 3206fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 3207fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 3208fa9e4066Sahrens 32095cabbc6bSPrashanth Sreenivasa /* 32104e75ba68SSerapheim Dimitropoulos * If this is an empty log device being removed, destroy the 32114e75ba68SSerapheim Dimitropoulos * metadata associated with it. 32125cabbc6bSPrashanth Sreenivasa */ 32134e75ba68SSerapheim Dimitropoulos if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 32144e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vd, txg); 32155cabbc6bSPrashanth Sreenivasa 3216fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 3217555d674dSSerapheim Dimitropoulos dmu_tx_commit(tx); 3218fa9e4066Sahrens } 3219fa9e4066Sahrens 3220fa9e4066Sahrens uint64_t 3221fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 3222fa9e4066Sahrens { 3223fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 3224fa9e4066Sahrens } 3225fa9e4066Sahrens 32263d7072f8Seschrock /* 32273d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 32283d7072f8Seschrock * not be opened, and no I/O is attempted. 32293d7072f8Seschrock */ 3230fa9e4066Sahrens int 3231069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3232fa9e4066Sahrens { 32334b964adaSGeorge Wilson vdev_t *vd, *tvd; 3234fa9e4066Sahrens 32358f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 3236fa9e4066Sahrens 3237c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3238e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3239e14bb325SJeff Bonwick 32403d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3241e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3242fa9e4066Sahrens 32434b964adaSGeorge Wilson tvd = vd->vdev_top; 32444b964adaSGeorge Wilson 3245069f55e2SEric Schrock /* 3246069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 3247069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 3248069f55e2SEric Schrock * were faulted. 3249069f55e2SEric Schrock */ 3250069f55e2SEric Schrock vd->vdev_label_aux = aux; 3251069f55e2SEric Schrock 32523d7072f8Seschrock /* 32533d7072f8Seschrock * Faulted state takes precedence over degraded. 32543d7072f8Seschrock */ 325598d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 32563d7072f8Seschrock vd->vdev_faulted = 1ULL; 32573d7072f8Seschrock vd->vdev_degraded = 0ULL; 3258069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 32593d7072f8Seschrock 32603d7072f8Seschrock /* 3261c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 3262c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 32633d7072f8Seschrock */ 32644b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 32653d7072f8Seschrock vd->vdev_degraded = 1ULL; 32663d7072f8Seschrock vd->vdev_faulted = 0ULL; 32673d7072f8Seschrock 32683d7072f8Seschrock /* 32693d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 32703d7072f8Seschrock * mark it degraded. 32713d7072f8Seschrock */ 32724b964adaSGeorge Wilson vdev_reopen(tvd); 32733d7072f8Seschrock 3274069f55e2SEric Schrock if (vdev_readable(vd)) 3275069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 32763d7072f8Seschrock } 32773d7072f8Seschrock 3278e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 32793d7072f8Seschrock } 32803d7072f8Seschrock 32813d7072f8Seschrock /* 32823d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 32833d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 32843d7072f8Seschrock * as I/O is concerned. 32853d7072f8Seschrock */ 32863d7072f8Seschrock int 3287069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 32883d7072f8Seschrock { 3289c5904d13Seschrock vdev_t *vd; 32900a4e9518Sgw 32918f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 32923d7072f8Seschrock 3293c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3294e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3295e14bb325SJeff Bonwick 32960e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3297e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 32980e34b6a7Sbonwick 32993d7072f8Seschrock /* 33003d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 33013d7072f8Seschrock */ 3302e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 3303e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 33043d7072f8Seschrock 33053d7072f8Seschrock vd->vdev_degraded = 1ULL; 33063d7072f8Seschrock if (!vdev_is_dead(vd)) 33073d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3308069f55e2SEric Schrock aux); 33093d7072f8Seschrock 3310e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 33113d7072f8Seschrock } 33123d7072f8Seschrock 33133d7072f8Seschrock /* 3314f7170741SWill Andrews * Online the given vdev. 3315f7170741SWill Andrews * 3316f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3317f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 3318f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 3319f7170741SWill Andrews * events are generated if the device fails to open. 33203d7072f8Seschrock */ 33213d7072f8Seschrock int 3322e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 33233d7072f8Seschrock { 3324573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 33255f368aefSYuri Pankov boolean_t wasoffline; 33265f368aefSYuri Pankov vdev_state_t oldstate; 33273d7072f8Seschrock 33288f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 33293d7072f8Seschrock 3330c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3331e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 33323d7072f8Seschrock 33333d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3334e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3335fa9e4066Sahrens 33365f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 33375f368aefSYuri Pankov oldstate = vd->vdev_state; 333814372834SHans Rosenfeld 3339573ca77eSGeorge Wilson tvd = vd->vdev_top; 3340fa9e4066Sahrens vd->vdev_offline = B_FALSE; 3341441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 3342e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3343e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3344573ca77eSGeorge Wilson 3345573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3346573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3347573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3348573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3349573ca77eSGeorge Wilson } 3350573ca77eSGeorge Wilson 3351573ca77eSGeorge Wilson vdev_reopen(tvd); 33523d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 33533d7072f8Seschrock 3354573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3355573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3356573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 3357573ca77eSGeorge Wilson } 3358573ca77eSGeorge Wilson 33593d7072f8Seschrock if (newstate) 33603d7072f8Seschrock *newstate = vd->vdev_state; 33613d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 33623d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 33633d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 33643d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 33653d7072f8Seschrock vd->vdev_unspare = B_TRUE; 3366fa9e4066Sahrens 3367573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3368573ca77eSGeorge Wilson 3369573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3370573ca77eSGeorge Wilson if (vd->vdev_aux) 3371573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3372573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3373573ca77eSGeorge Wilson } 337414372834SHans Rosenfeld 3375094e47e9SGeorge Wilson /* Restart initializing if necessary */ 3376094e47e9SGeorge Wilson mutex_enter(&vd->vdev_initialize_lock); 3377094e47e9SGeorge Wilson if (vdev_writeable(vd) && 3378094e47e9SGeorge Wilson vd->vdev_initialize_thread == NULL && 3379094e47e9SGeorge Wilson vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3380094e47e9SGeorge Wilson (void) vdev_initialize(vd); 3381094e47e9SGeorge Wilson } 3382094e47e9SGeorge Wilson mutex_exit(&vd->vdev_initialize_lock); 3383094e47e9SGeorge Wilson 3384084fd14fSBrian Behlendorf /* Restart trimming if necessary */ 3385084fd14fSBrian Behlendorf mutex_enter(&vd->vdev_trim_lock); 3386084fd14fSBrian Behlendorf if (vdev_writeable(vd) && 3387084fd14fSBrian Behlendorf vd->vdev_trim_thread == NULL && 3388084fd14fSBrian Behlendorf vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { 3389084fd14fSBrian Behlendorf (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, 3390084fd14fSBrian Behlendorf vd->vdev_trim_secure); 3391084fd14fSBrian Behlendorf } 3392084fd14fSBrian Behlendorf mutex_exit(&vd->vdev_trim_lock); 3393084fd14fSBrian Behlendorf 33945f368aefSYuri Pankov if (wasoffline || 33955f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 33965f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 3397ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 339814372834SHans Rosenfeld 33998ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 3400fa9e4066Sahrens } 3401fa9e4066Sahrens 3402a1521560SJeff Bonwick static int 3403a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3404fa9e4066Sahrens { 3405e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 34068f18d1faSGeorge Wilson int error = 0; 34078f18d1faSGeorge Wilson uint64_t generation; 34088f18d1faSGeorge Wilson metaslab_group_t *mg; 34090a4e9518Sgw 34108f18d1faSGeorge Wilson top: 34118f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 3412fa9e4066Sahrens 3413c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3414e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3415fa9e4066Sahrens 34160e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3417e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 34180e34b6a7Sbonwick 3419e6ca193dSGeorge Wilson tvd = vd->vdev_top; 34208f18d1faSGeorge Wilson mg = tvd->vdev_mg; 34218f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 3422e6ca193dSGeorge Wilson 3423fa9e4066Sahrens /* 3424ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 3425fa9e4066Sahrens */ 3426ecc2d604Sbonwick if (!vd->vdev_offline) { 3427ecc2d604Sbonwick /* 34288ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 3429e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 3430e6ca193dSGeorge Wilson * expendable. 3431ecc2d604Sbonwick */ 3432e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3433e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 3434e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3435fa9e4066Sahrens 34368f18d1faSGeorge Wilson /* 3437b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 3438b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 3439b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 3440b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 34418f18d1faSGeorge Wilson */ 34428f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 34438f18d1faSGeorge Wilson /* 34448f18d1faSGeorge Wilson * Prevent any future allocations. 34458f18d1faSGeorge Wilson */ 3446a1521560SJeff Bonwick metaslab_group_passivate(mg); 34478f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 34488f18d1faSGeorge Wilson 34495cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 34508f18d1faSGeorge Wilson 345186714001SSerapheim Dimitropoulos /* 345286714001SSerapheim Dimitropoulos * If the log device was successfully reset but has 345386714001SSerapheim Dimitropoulos * checkpointed data, do not offline it. 345486714001SSerapheim Dimitropoulos */ 345586714001SSerapheim Dimitropoulos if (error == 0 && 345686714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm != NULL) { 345786714001SSerapheim Dimitropoulos error = ZFS_ERR_CHECKPOINT_EXISTS; 345886714001SSerapheim Dimitropoulos } 345986714001SSerapheim Dimitropoulos 34608f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 34618f18d1faSGeorge Wilson 34628f18d1faSGeorge Wilson /* 34638f18d1faSGeorge Wilson * Check to see if the config has changed. 34648f18d1faSGeorge Wilson */ 34658f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 3466a1521560SJeff Bonwick metaslab_group_activate(mg); 34678f18d1faSGeorge Wilson if (error) 34688f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 34698f18d1faSGeorge Wilson vd, error)); 34708f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 34718f18d1faSGeorge Wilson goto top; 34728f18d1faSGeorge Wilson } 3473fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 34748f18d1faSGeorge Wilson } 34758f18d1faSGeorge Wilson 3476ecc2d604Sbonwick /* 3477ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3478e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3479e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3480e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3481ecc2d604Sbonwick */ 3482ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3483e6ca193dSGeorge Wilson vdev_reopen(tvd); 3484e6ca193dSGeorge Wilson 3485e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3486e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3487ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3488e6ca193dSGeorge Wilson vdev_reopen(tvd); 3489e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3490ecc2d604Sbonwick } 34918f18d1faSGeorge Wilson 34928f18d1faSGeorge Wilson /* 34938f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 34948f18d1faSGeorge Wilson * once we online the device it's open for business. 34958f18d1faSGeorge Wilson */ 34968f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3497a1521560SJeff Bonwick metaslab_group_activate(mg); 3498fa9e4066Sahrens } 3499fa9e4066Sahrens 3500e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3501ecc2d604Sbonwick 35028f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3503fa9e4066Sahrens } 3504fa9e4066Sahrens 3505a1521560SJeff Bonwick int 3506a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3507a1521560SJeff Bonwick { 3508a1521560SJeff Bonwick int error; 3509a1521560SJeff Bonwick 3510a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3511a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3512a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3513a1521560SJeff Bonwick 3514a1521560SJeff Bonwick return (error); 3515a1521560SJeff Bonwick } 3516a1521560SJeff Bonwick 3517ea8dc4b6Seschrock /* 3518ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3519ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3520ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3521ea8dc4b6Seschrock */ 3522ea8dc4b6Seschrock void 3523e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3524fa9e4066Sahrens { 3525e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3526e14bb325SJeff Bonwick 3527e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3528fa9e4066Sahrens 3529ea8dc4b6Seschrock if (vd == NULL) 3530e14bb325SJeff Bonwick vd = rvd; 3531fa9e4066Sahrens 3532ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3533ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3534ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3535dd50e0ccSTony Hutter vd->vdev_stat.vs_slow_ios = 0; 3536fa9e4066Sahrens 3537e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3538e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 35393d7072f8Seschrock 35405cabbc6bSPrashanth Sreenivasa /* 35415cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 35425cabbc6bSPrashanth Sreenivasa */ 35435cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 35445cabbc6bSPrashanth Sreenivasa return; 35455cabbc6bSPrashanth Sreenivasa 35463d7072f8Seschrock /* 35478a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 35488a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 35498a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 35508a79c1b5Sek * written out to disk. 35513d7072f8Seschrock */ 3552e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3553e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 35548a79c1b5Sek 3555096d22d4SEric Schrock /* 3556096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3557096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3558096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3559096d22d4SEric Schrock */ 3560096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3561096d22d4SEric Schrock 35624b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3563e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3564e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3565e14bb325SJeff Bonwick 3566f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 35673d7072f8Seschrock 3568096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3569096d22d4SEric Schrock 3570f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3571e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3572e14bb325SJeff Bonwick 35730c06d385Sjwpoduska /* If a resilver isn't required, check if vdevs can be culled */ 35740c06d385Sjwpoduska if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && 35750c06d385Sjwpoduska !dsl_scan_resilvering(spa->spa_dsl_pool) && 35760c06d385Sjwpoduska !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) 35770c06d385Sjwpoduska spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 35783d7072f8Seschrock 3579ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 35803d7072f8Seschrock } 3581096d22d4SEric Schrock 3582096d22d4SEric Schrock /* 3583096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3584096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3585096d22d4SEric Schrock * done in response to the FMA fault. 3586096d22d4SEric Schrock */ 3587096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3588096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3589096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3590096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3591fa9e4066Sahrens } 3592fa9e4066Sahrens 3593e14bb325SJeff Bonwick boolean_t 3594e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 35950a4e9518Sgw { 359688ecc943SGeorge Wilson /* 359788ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 359888ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 359988ecc943SGeorge Wilson * these types of devices in the various code paths. 360088ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 360188ecc943SGeorge Wilson * before issuing I/O to them. 360288ecc943SGeorge Wilson */ 36035cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 36045cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 360588ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 36060a4e9518Sgw } 36070a4e9518Sgw 3608e14bb325SJeff Bonwick boolean_t 3609e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 36100a4e9518Sgw { 3611e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 36120a4e9518Sgw } 36130a4e9518Sgw 3614e14bb325SJeff Bonwick boolean_t 3615e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3616fa9e4066Sahrens { 36175cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 36185cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3619fa9e4066Sahrens } 3620fa9e4066Sahrens 3621a31e6787SGeorge Wilson boolean_t 3622a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3623a31e6787SGeorge Wilson { 36248ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 36258ad4d6ddSJeff Bonwick 3626a31e6787SGeorge Wilson /* 36278ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3628a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3629a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 36308ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 36318ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 36328ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3633a31e6787SGeorge Wilson */ 36348ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 36355cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 36360f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3637a31e6787SGeorge Wilson } 3638a31e6787SGeorge Wilson 3639e14bb325SJeff Bonwick boolean_t 3640e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3641fa9e4066Sahrens { 3642e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3643fa9e4066Sahrens 3644e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3645e14bb325SJeff Bonwick return (B_FALSE); 3646fa9e4066Sahrens 3647e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3648e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3649fa9e4066Sahrens 3650e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3651e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3652fa9e4066Sahrens 3653e14bb325SJeff Bonwick return (B_TRUE); 3654fa9e4066Sahrens } 3655fa9e4066Sahrens 3656dd50e0ccSTony Hutter static void 3657dd50e0ccSTony Hutter vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) 3658dd50e0ccSTony Hutter { 3659dd50e0ccSTony Hutter for (int t = 0; t < VS_ZIO_TYPES; t++) { 3660dd50e0ccSTony Hutter vs->vs_ops[t] += cvs->vs_ops[t]; 3661dd50e0ccSTony Hutter vs->vs_bytes[t] += cvs->vs_bytes[t]; 3662dd50e0ccSTony Hutter } 3663dd50e0ccSTony Hutter 3664dd50e0ccSTony Hutter cvs->vs_scan_removing = cvd->vdev_removing; 3665dd50e0ccSTony Hutter } 3666dd50e0ccSTony Hutter 3667dd50e0ccSTony Hutter /* 3668dd50e0ccSTony Hutter * Get extended stats 3669dd50e0ccSTony Hutter */ 3670dd50e0ccSTony Hutter static void 3671dd50e0ccSTony Hutter vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) 3672dd50e0ccSTony Hutter { 3673dd50e0ccSTony Hutter int t, b; 3674dd50e0ccSTony Hutter for (t = 0; t < ZIO_TYPES; t++) { 3675dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) 3676dd50e0ccSTony Hutter vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; 3677dd50e0ccSTony Hutter 3678dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { 3679dd50e0ccSTony Hutter vsx->vsx_total_histo[t][b] += 3680dd50e0ccSTony Hutter cvsx->vsx_total_histo[t][b]; 3681dd50e0ccSTony Hutter } 3682dd50e0ccSTony Hutter } 3683dd50e0ccSTony Hutter 3684dd50e0ccSTony Hutter for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { 3685dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { 3686dd50e0ccSTony Hutter vsx->vsx_queue_histo[t][b] += 3687dd50e0ccSTony Hutter cvsx->vsx_queue_histo[t][b]; 3688dd50e0ccSTony Hutter } 3689dd50e0ccSTony Hutter vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; 3690dd50e0ccSTony Hutter vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; 3691dd50e0ccSTony Hutter 3692dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) 3693dd50e0ccSTony Hutter vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; 3694dd50e0ccSTony Hutter 3695dd50e0ccSTony Hutter for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) 3696dd50e0ccSTony Hutter vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; 3697dd50e0ccSTony Hutter } 3698dd50e0ccSTony Hutter 3699dd50e0ccSTony Hutter } 3700dd50e0ccSTony Hutter 370186714001SSerapheim Dimitropoulos boolean_t 370286714001SSerapheim Dimitropoulos vdev_is_spacemap_addressable(vdev_t *vd) 370386714001SSerapheim Dimitropoulos { 3704a0b03b16SSerapheim Dimitropoulos if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3705a0b03b16SSerapheim Dimitropoulos return (B_TRUE); 3706a0b03b16SSerapheim Dimitropoulos 370786714001SSerapheim Dimitropoulos /* 3708a0b03b16SSerapheim Dimitropoulos * If double-word space map entries are not enabled we assume 3709a0b03b16SSerapheim Dimitropoulos * 47 bits of the space map entry are dedicated to the entry's 3710a0b03b16SSerapheim Dimitropoulos * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3711a0b03b16SSerapheim Dimitropoulos * to calculate the maximum address that can be described by a 3712a0b03b16SSerapheim Dimitropoulos * space map entry for the given device. 371386714001SSerapheim Dimitropoulos */ 3714a0b03b16SSerapheim Dimitropoulos uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 371586714001SSerapheim Dimitropoulos 371686714001SSerapheim Dimitropoulos if (shift >= 63) /* detect potential overflow */ 371786714001SSerapheim Dimitropoulos return (B_TRUE); 371886714001SSerapheim Dimitropoulos 371986714001SSerapheim Dimitropoulos return (vd->vdev_asize < (1ULL << shift)); 372086714001SSerapheim Dimitropoulos } 372186714001SSerapheim Dimitropoulos 3722fa9e4066Sahrens /* 3723fa9e4066Sahrens * Get statistics for the given vdev. 3724fa9e4066Sahrens */ 3725dd50e0ccSTony Hutter static void 3726dd50e0ccSTony Hutter vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3727fa9e4066Sahrens { 3728dd50e0ccSTony Hutter int t; 3729dd50e0ccSTony Hutter /* 3730dd50e0ccSTony Hutter * If we're getting stats on the root vdev, aggregate the I/O counts 3731dd50e0ccSTony Hutter * over all top-level vdevs (i.e. the direct children of the root). 3732dd50e0ccSTony Hutter */ 3733dd50e0ccSTony Hutter if (!vd->vdev_ops->vdev_op_leaf) { 3734dd50e0ccSTony Hutter if (vs) { 3735dd50e0ccSTony Hutter memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); 3736dd50e0ccSTony Hutter memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); 3737dd50e0ccSTony Hutter } 3738dd50e0ccSTony Hutter if (vsx) 3739dd50e0ccSTony Hutter memset(vsx, 0, sizeof (*vsx)); 37402e4c9986SGeorge Wilson 3741dd50e0ccSTony Hutter for (int c = 0; c < vd->vdev_children; c++) { 3742dd50e0ccSTony Hutter vdev_t *cvd = vd->vdev_child[c]; 3743dd50e0ccSTony Hutter vdev_stat_t *cvs = &cvd->vdev_stat; 3744dd50e0ccSTony Hutter vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; 3745fa9e4066Sahrens 3746dd50e0ccSTony Hutter vdev_get_stats_ex_impl(cvd, cvs, cvsx); 3747dd50e0ccSTony Hutter if (vs) 3748dd50e0ccSTony Hutter vdev_get_child_stat(cvd, vs, cvs); 3749dd50e0ccSTony Hutter if (vsx) 3750dd50e0ccSTony Hutter vdev_get_child_stat_ex(cvd, vsx, cvsx); 3751dd50e0ccSTony Hutter 3752dd50e0ccSTony Hutter } 3753dd50e0ccSTony Hutter } else { 3754094e47e9SGeorge Wilson /* 3755dd50e0ccSTony Hutter * We're a leaf. Just copy our ZIO active queue stats in. The 3756dd50e0ccSTony Hutter * other leaf stats are updated in vdev_stat_update(). 3757094e47e9SGeorge Wilson */ 3758dd50e0ccSTony Hutter if (!vsx) 3759dd50e0ccSTony Hutter return; 3760084fd14fSBrian Behlendorf 3761dd50e0ccSTony Hutter memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); 3762084fd14fSBrian Behlendorf 3763dd50e0ccSTony Hutter for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { 3764dd50e0ccSTony Hutter vsx->vsx_active_queue[t] = 3765dd50e0ccSTony Hutter vd->vdev_queue.vq_class[t].vqc_active; 3766dd50e0ccSTony Hutter vsx->vsx_pend_queue[t] = avl_numnodes( 3767dd50e0ccSTony Hutter &vd->vdev_queue.vq_class[t].vqc_queued_tree); 3768dd50e0ccSTony Hutter } 37692986efa8SAlex Reece } 3770dd50e0ccSTony Hutter } 3771fa9e4066Sahrens 3772dd50e0ccSTony Hutter void 3773dd50e0ccSTony Hutter vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) 3774dd50e0ccSTony Hutter { 3775dd50e0ccSTony Hutter vdev_t *tvd = vd->vdev_top; 3776dd50e0ccSTony Hutter mutex_enter(&vd->vdev_stat_lock); 3777dd50e0ccSTony Hutter if (vs) { 3778dd50e0ccSTony Hutter bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3779dd50e0ccSTony Hutter vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3780dd50e0ccSTony Hutter vs->vs_state = vd->vdev_state; 3781dd50e0ccSTony Hutter vs->vs_rsize = vdev_get_min_asize(vd); 3782dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf) { 3783dd50e0ccSTony Hutter vs->vs_rsize += VDEV_LABEL_START_SIZE + 3784dd50e0ccSTony Hutter VDEV_LABEL_END_SIZE; 3785dd50e0ccSTony Hutter /* 3786dd50e0ccSTony Hutter * Report initializing progress. Since we don't 3787dd50e0ccSTony Hutter * have the initializing locks held, this is only 3788dd50e0ccSTony Hutter * an estimate (although a fairly accurate one). 3789dd50e0ccSTony Hutter */ 3790dd50e0ccSTony Hutter vs->vs_initialize_bytes_done = 3791dd50e0ccSTony Hutter vd->vdev_initialize_bytes_done; 3792dd50e0ccSTony Hutter vs->vs_initialize_bytes_est = 3793dd50e0ccSTony Hutter vd->vdev_initialize_bytes_est; 3794dd50e0ccSTony Hutter vs->vs_initialize_state = vd->vdev_initialize_state; 3795dd50e0ccSTony Hutter vs->vs_initialize_action_time = 3796dd50e0ccSTony Hutter vd->vdev_initialize_action_time; 3797fa9e4066Sahrens 3798dd50e0ccSTony Hutter /* 3799dd50e0ccSTony Hutter * Report manual TRIM progress. Since we don't have 3800dd50e0ccSTony Hutter * the manual TRIM locks held, this is only an 3801dd50e0ccSTony Hutter * estimate (although fairly accurate one). 3802dd50e0ccSTony Hutter */ 3803dd50e0ccSTony Hutter vs->vs_trim_notsup = !vd->vdev_has_trim; 3804dd50e0ccSTony Hutter vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; 3805dd50e0ccSTony Hutter vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; 3806dd50e0ccSTony Hutter vs->vs_trim_state = vd->vdev_trim_state; 3807dd50e0ccSTony Hutter vs->vs_trim_action_time = vd->vdev_trim_action_time; 3808dd50e0ccSTony Hutter } 3809dd50e0ccSTony Hutter /* 3810dd50e0ccSTony Hutter * Report expandable space on top-level, non-auxiliary devices 3811dd50e0ccSTony Hutter * only. The expandable space is reported in terms of metaslab 3812dd50e0ccSTony Hutter * sized units since that determines how much space the pool 3813dd50e0ccSTony Hutter * can expand. 3814dd50e0ccSTony Hutter */ 3815dd50e0ccSTony Hutter if (vd->vdev_aux == NULL && tvd != NULL) { 3816dd50e0ccSTony Hutter vs->vs_esize = P2ALIGN( 3817dd50e0ccSTony Hutter vd->vdev_max_asize - vd->vdev_asize, 3818dd50e0ccSTony Hutter 1ULL << tvd->vdev_ms_shift); 3819dd50e0ccSTony Hutter } 3820dd50e0ccSTony Hutter if (vd->vdev_aux == NULL && vd == vd->vdev_top && 3821dd50e0ccSTony Hutter vdev_is_concrete(vd)) { 3822dd50e0ccSTony Hutter vs->vs_fragmentation = (vd->vdev_mg != NULL) ? 3823dd50e0ccSTony Hutter vd->vdev_mg->mg_fragmentation : 0; 3824fa9e4066Sahrens } 3825dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf) 3826dd50e0ccSTony Hutter vs->vs_resilver_deferred = vd->vdev_resilver_deferred; 3827fa9e4066Sahrens } 3828dd50e0ccSTony Hutter 3829dd50e0ccSTony Hutter vdev_get_stats_ex_impl(vd, vs, vsx); 38302e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3831fa9e4066Sahrens } 3832fa9e4066Sahrens 3833dd50e0ccSTony Hutter void 3834dd50e0ccSTony Hutter vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3835dd50e0ccSTony Hutter { 3836dd50e0ccSTony Hutter return (vdev_get_stats_ex(vd, vs, NULL)); 3837dd50e0ccSTony Hutter } 3838dd50e0ccSTony Hutter 3839fa94a07fSbrendan void 3840fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3841fa94a07fSbrendan { 3842fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3843fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3844fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3845fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3846fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3847fa94a07fSbrendan } 3848fa94a07fSbrendan 38493f9d6ad7SLin Ling void 38503f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 38513f9d6ad7SLin Ling { 38523f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 38533f9d6ad7SLin Ling 38543f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 38553f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 38563f9d6ad7SLin Ling 38573f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 38583f9d6ad7SLin Ling vs->vs_scan_processed = 0; 38593f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 38603f9d6ad7SLin Ling } 38613f9d6ad7SLin Ling 3862fa9e4066Sahrens void 3863e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3864fa9e4066Sahrens { 38658ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 38668ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3867e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3868fa9e4066Sahrens vdev_t *pvd; 3869fa9e4066Sahrens uint64_t txg = zio->io_txg; 3870fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3871dd50e0ccSTony Hutter vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; 3872fa9e4066Sahrens zio_type_t type = zio->io_type; 3873fa9e4066Sahrens int flags = zio->io_flags; 3874fa9e4066Sahrens 3875e14bb325SJeff Bonwick /* 3876e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3877e14bb325SJeff Bonwick */ 3878e14bb325SJeff Bonwick if (zio->io_gang_tree) 3879e14bb325SJeff Bonwick return; 3880e14bb325SJeff Bonwick 3881fa9e4066Sahrens if (zio->io_error == 0) { 3882e14bb325SJeff Bonwick /* 3883e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3884e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3885e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3886e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3887e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3888e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3889e14bb325SJeff Bonwick * remain zero, which is what we want.) 3890e14bb325SJeff Bonwick * 3891e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3892e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3893e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3894e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3895e14bb325SJeff Bonwick */ 3896e14bb325SJeff Bonwick if (vd == rvd) 3897e14bb325SJeff Bonwick return; 3898e14bb325SJeff Bonwick 3899e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 39008ad4d6ddSJeff Bonwick 39018ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 39028ad4d6ddSJeff Bonwick return; 39038ad4d6ddSJeff Bonwick 39048ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 39058ad4d6ddSJeff Bonwick 3906e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 390744ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 39083f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 39093f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 39103f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 39113f9d6ad7SLin Ling 39123f9d6ad7SLin Ling /* XXX cleanup? */ 39133f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 39143f9d6ad7SLin Ling atomic_add_64(processed, psize); 39153f9d6ad7SLin Ling vs->vs_scan_processed += psize; 39163f9d6ad7SLin Ling } 39173f9d6ad7SLin Ling 39188ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3919e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3920fa9e4066Sahrens } 39218ad4d6ddSJeff Bonwick 3922084fd14fSBrian Behlendorf /* 3923dd50e0ccSTony Hutter * The bytes/ops/histograms are recorded at the leaf level and 3924dd50e0ccSTony Hutter * aggregated into the higher level vdevs in vdev_get_stats(). 3925084fd14fSBrian Behlendorf */ 3926dd50e0ccSTony Hutter if (vd->vdev_ops->vdev_op_leaf && 3927dd50e0ccSTony Hutter (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { 3928dd50e0ccSTony Hutter zio_type_t vs_type = type; 3929dd50e0ccSTony Hutter 3930dd50e0ccSTony Hutter /* 3931dd50e0ccSTony Hutter * TRIM ops and bytes are reported to user space as 3932dd50e0ccSTony Hutter * ZIO_TYPE_IOCTL. This is done to preserve the 3933dd50e0ccSTony Hutter * vdev_stat_t structure layout for user space. 3934dd50e0ccSTony Hutter */ 3935dd50e0ccSTony Hutter if (type == ZIO_TYPE_TRIM) 3936dd50e0ccSTony Hutter vs_type = ZIO_TYPE_IOCTL; 3937dd50e0ccSTony Hutter 3938dd50e0ccSTony Hutter vs->vs_ops[vs_type]++; 3939dd50e0ccSTony Hutter vs->vs_bytes[vs_type] += psize; 3940dd50e0ccSTony Hutter 3941dd50e0ccSTony Hutter if (flags & ZIO_FLAG_DELEGATED) { 3942dd50e0ccSTony Hutter vsx->vsx_agg_histo[zio->io_priority] 3943dd50e0ccSTony Hutter [RQ_HISTO(zio->io_size)]++; 3944dd50e0ccSTony Hutter } else { 3945dd50e0ccSTony Hutter vsx->vsx_ind_histo[zio->io_priority] 3946dd50e0ccSTony Hutter [RQ_HISTO(zio->io_size)]++; 3947dd50e0ccSTony Hutter } 3948084fd14fSBrian Behlendorf 3949dd50e0ccSTony Hutter if (zio->io_delta && zio->io_delay) { 3950dd50e0ccSTony Hutter vsx->vsx_queue_histo[zio->io_priority] 3951dd50e0ccSTony Hutter [L_HISTO(zio->io_delta - zio->io_delay)]++; 3952dd50e0ccSTony Hutter vsx->vsx_disk_histo[type] 3953dd50e0ccSTony Hutter [L_HISTO(zio->io_delay)]++; 3954dd50e0ccSTony Hutter vsx->vsx_total_histo[type] 3955dd50e0ccSTony Hutter [L_HISTO(zio->io_delta)]++; 3956dd50e0ccSTony Hutter } 3957dd50e0ccSTony Hutter } 39588ad4d6ddSJeff Bonwick 39598ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3960fa9e4066Sahrens return; 3961fa9e4066Sahrens } 3962fa9e4066Sahrens 3963fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3964fa9e4066Sahrens return; 3965fa9e4066Sahrens 39668956713aSEric Schrock /* 39678956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 39688956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 39698956713aSEric Schrock * hard errors, when in reality they can happen for any number of 39708956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 39718956713aSEric Schrock */ 39728956713aSEric Schrock if (zio->io_error == EIO && 39738956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 39748956713aSEric Schrock return; 39758956713aSEric Schrock 39768f18d1faSGeorge Wilson /* 39778f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 39788f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 39798f18d1faSGeorge Wilson * errors. 39808f18d1faSGeorge Wilson */ 39818f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 39828f18d1faSGeorge Wilson return; 39838f18d1faSGeorge Wilson 3984e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3985b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3986e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3987e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3988e14bb325SJeff Bonwick else 3989e14bb325SJeff Bonwick vs->vs_read_errors++; 3990fa9e4066Sahrens } 3991b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3992e14bb325SJeff Bonwick vs->vs_write_errors++; 3993e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3994fa9e4066Sahrens 39955cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 39965cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 39978ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 399844ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3999b24ab676SJeff Bonwick spa->spa_claiming)) { 40008ad4d6ddSJeff Bonwick /* 4001b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 4002b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 4003b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 4004b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 4005b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 4006b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 4007b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 4008b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 40098ad4d6ddSJeff Bonwick * 40108ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 40118ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 40128ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 40138ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 40148ad4d6ddSJeff Bonwick */ 40158ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 40168ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 401744ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 40188ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 40198ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 40208ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 4021b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 4022b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 4023b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 4024b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 40258ad4d6ddSJeff Bonwick } 4026b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 40278ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 4028fa9e4066Sahrens return; 40298ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 40308ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 40318ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 4032fa9e4066Sahrens } 40338ad4d6ddSJeff Bonwick if (vd != rvd) 40348ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 4035fa9e4066Sahrens } 4036fa9e4066Sahrens } 4037fa9e4066Sahrens 4038663207adSDon Brady int64_t 4039663207adSDon Brady vdev_deflated_space(vdev_t *vd, int64_t space) 4040663207adSDon Brady { 4041663207adSDon Brady ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); 4042663207adSDon Brady ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 4043663207adSDon Brady 4044663207adSDon Brady return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); 4045663207adSDon Brady } 4046663207adSDon Brady 4047fa9e4066Sahrens /* 4048084fd14fSBrian Behlendorf * Update the in-core space usage stats for this vdev, its metaslab class, 4049084fd14fSBrian Behlendorf * and the root vdev. 4050fa9e4066Sahrens */ 4051fa9e4066Sahrens void 4052b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 4053b24ab676SJeff Bonwick int64_t space_delta) 4054fa9e4066Sahrens { 4055663207adSDon Brady int64_t dspace_delta; 40568654d025Sperrin spa_t *spa = vd->vdev_spa; 40578654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 4058fa9e4066Sahrens 40598654d025Sperrin ASSERT(vd == vd->vdev_top); 406099653d4eSeschrock 40618654d025Sperrin /* 40628654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 40638654d025Sperrin * factor. We must calculate this here and not at the root vdev 40648654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 40658654d025Sperrin * childrens', thus not accurate enough for us. 40668654d025Sperrin */ 4067663207adSDon Brady dspace_delta = vdev_deflated_space(vd, space_delta); 40688654d025Sperrin 40698654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 40709740f25fSSerapheim Dimitropoulos /* ensure we won't underflow */ 40719740f25fSSerapheim Dimitropoulos if (alloc_delta < 0) { 40729740f25fSSerapheim Dimitropoulos ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); 40739740f25fSSerapheim Dimitropoulos } 40749740f25fSSerapheim Dimitropoulos 40758654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 4076b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 40778654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 40788654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 40798654d025Sperrin 4080663207adSDon Brady /* every class but log contributes to root space stats */ 4081663207adSDon Brady if (vd->vdev_mg != NULL && !vd->vdev_islog) { 40829740f25fSSerapheim Dimitropoulos ASSERT(!vd->vdev_isl2cache); 4083fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 4084fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 4085b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 4086fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 4087fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 4088fa94a07fSbrendan } 4089663207adSDon Brady /* Note: metaslab_class_space_update moved to metaslab_space_update */ 4090fa9e4066Sahrens } 4091fa9e4066Sahrens 4092fa9e4066Sahrens /* 4093fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 4094fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 4095fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 4096fa9e4066Sahrens */ 4097fa9e4066Sahrens void 4098fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 4099fa9e4066Sahrens { 4100fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 4101fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 4102fa9e4066Sahrens int c; 4103fa9e4066Sahrens 4104f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4105f9af39baSGeorge Wilson 4106c5904d13Seschrock /* 41076809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 41086809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 4109c5904d13Seschrock */ 4110c5904d13Seschrock if (vd->vdev_aux != NULL) { 4111c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 4112c5904d13Seschrock nvlist_t **aux; 4113c5904d13Seschrock uint_t naux; 4114c5904d13Seschrock 4115c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 4116c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 4117c5904d13Seschrock break; 4118c5904d13Seschrock } 4119c5904d13Seschrock 4120e14bb325SJeff Bonwick if (c == sav->sav_count) { 4121e14bb325SJeff Bonwick /* 4122e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 4123e14bb325SJeff Bonwick */ 4124e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 4125e14bb325SJeff Bonwick return; 4126e14bb325SJeff Bonwick } 4127e14bb325SJeff Bonwick 4128c5904d13Seschrock sav->sav_sync = B_TRUE; 4129c5904d13Seschrock 41306809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 41316809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 41326809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 41336809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 41346809eb4eSEric Schrock } 4135c5904d13Seschrock 4136c5904d13Seschrock ASSERT(c < naux); 4137c5904d13Seschrock 4138c5904d13Seschrock /* 4139c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 4140c5904d13Seschrock * sketchy, but it will work. 4141c5904d13Seschrock */ 4142c5904d13Seschrock nvlist_free(aux[c]); 41433f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 4144c5904d13Seschrock 4145c5904d13Seschrock return; 4146c5904d13Seschrock } 4147c5904d13Seschrock 41485dabedeeSbonwick /* 4149e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 4150e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 4151e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 41525dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 41535dabedeeSbonwick */ 4154e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4155e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4156e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 41575dabedeeSbonwick 4158fa9e4066Sahrens if (vd == rvd) { 4159fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 4160fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 4161fa9e4066Sahrens } else { 4162fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 4163fa9e4066Sahrens 416488ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 41655cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 4166e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 41675cabbc6bSPrashanth Sreenivasa } 4168fa9e4066Sahrens } 4169fa9e4066Sahrens } 4170fa9e4066Sahrens 4171fa9e4066Sahrens void 4172fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 4173fa9e4066Sahrens { 41745dabedeeSbonwick spa_t *spa = vd->vdev_spa; 41755dabedeeSbonwick 4176e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 4177e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4178e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 41795dabedeeSbonwick 4180e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 4181e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 4182e14bb325SJeff Bonwick } 4183e14bb325SJeff Bonwick 4184e14bb325SJeff Bonwick /* 4185e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 4186e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 4187e14bb325SJeff Bonwick * the state changes from larger config changes because they require 4188e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 4189e14bb325SJeff Bonwick */ 4190e14bb325SJeff Bonwick void 4191e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 4192e14bb325SJeff Bonwick { 4193e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4194e14bb325SJeff Bonwick 4195f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 4196e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 4197e14bb325SJeff Bonwick 4198e14bb325SJeff Bonwick /* 4199e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 4200e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 4201e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 4202e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 4203e14bb325SJeff Bonwick */ 4204e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4205e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4206e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4207e14bb325SJeff Bonwick 42085cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 42095cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 4210e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 4211e14bb325SJeff Bonwick } 4212e14bb325SJeff Bonwick 4213e14bb325SJeff Bonwick void 4214e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 4215e14bb325SJeff Bonwick { 4216e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 4217e14bb325SJeff Bonwick 4218e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 4219e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 4220e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 4221e14bb325SJeff Bonwick 4222e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 4223e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 4224fa9e4066Sahrens } 4225fa9e4066Sahrens 422632b87932Sek /* 422732b87932Sek * Propagate vdev state up from children to parent. 422832b87932Sek */ 422944cd46caSbillm void 423044cd46caSbillm vdev_propagate_state(vdev_t *vd) 423144cd46caSbillm { 42328ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 42338ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 423444cd46caSbillm int degraded = 0, faulted = 0; 423544cd46caSbillm int corrupted = 0; 423644cd46caSbillm vdev_t *child; 423744cd46caSbillm 42383d7072f8Seschrock if (vd->vdev_children > 0) { 4239573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 42403d7072f8Seschrock child = vd->vdev_child[c]; 424151ece835Seschrock 424288ecc943SGeorge Wilson /* 42435cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 42445cabbc6bSPrashanth Sreenivasa * decision. 424588ecc943SGeorge Wilson */ 42465cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 424788ecc943SGeorge Wilson continue; 424888ecc943SGeorge Wilson 4249e14bb325SJeff Bonwick if (!vdev_readable(child) || 42508ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 425151ece835Seschrock /* 425251ece835Seschrock * Root special: if there is a top-level log 425351ece835Seschrock * device, treat the root vdev as if it were 425451ece835Seschrock * degraded. 425551ece835Seschrock */ 425651ece835Seschrock if (child->vdev_islog && vd == rvd) 425751ece835Seschrock degraded++; 425851ece835Seschrock else 425951ece835Seschrock faulted++; 426051ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 42613d7072f8Seschrock degraded++; 426251ece835Seschrock } 426344cd46caSbillm 42643d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 42653d7072f8Seschrock corrupted++; 42663d7072f8Seschrock } 426744cd46caSbillm 42683d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 42693d7072f8Seschrock 42703d7072f8Seschrock /* 4271e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 42723d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 42733d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 42743d7072f8Seschrock * replicas'. 42753d7072f8Seschrock */ 42763d7072f8Seschrock if (corrupted && vd == rvd && 42773d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 42783d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 42793d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 42803d7072f8Seschrock } 42813d7072f8Seschrock 428251ece835Seschrock if (vd->vdev_parent) 42833d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 428444cd46caSbillm } 428544cd46caSbillm 4286fa9e4066Sahrens /* 4287ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 4288ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 4289ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 4290ea8dc4b6Seschrock * 4291ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 4292ea8dc4b6Seschrock * generated. 4293fa9e4066Sahrens */ 4294fa9e4066Sahrens void 4295ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 4296fa9e4066Sahrens { 4297560e6e96Seschrock uint64_t save_state; 4298c5904d13Seschrock spa_t *spa = vd->vdev_spa; 4299ea8dc4b6Seschrock 4300ea8dc4b6Seschrock if (state == vd->vdev_state) { 4301ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 4302fa9e4066Sahrens return; 4303ea8dc4b6Seschrock } 4304ea8dc4b6Seschrock 4305560e6e96Seschrock save_state = vd->vdev_state; 4306fa9e4066Sahrens 4307fa9e4066Sahrens vd->vdev_state = state; 4308fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 4309fa9e4066Sahrens 43103d7072f8Seschrock /* 43113d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 431298d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 431398d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 431498d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 431598d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 431698d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 431798d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 431898d1cbfeSGeorge Wilson * children. 431998d1cbfeSGeorge Wilson */ 432098d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 432198d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 43223d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 43233d7072f8Seschrock 4324069f55e2SEric Schrock /* 4325069f55e2SEric Schrock * If we have brought this vdev back into service, we need 4326069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 4327069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 4328069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 4329069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 4330069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 4331069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 4332069f55e2SEric Schrock */ 4333069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 4334069f55e2SEric Schrock vd->vdev_prevstate != state) 4335069f55e2SEric Schrock zfs_post_state_change(spa, vd); 4336069f55e2SEric Schrock 43373d7072f8Seschrock if (vd->vdev_removed && 43383d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 43393d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 43403d7072f8Seschrock /* 43413d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 43423d7072f8Seschrock * device was previously marked removed and someone attempted to 43433d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 43443d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 43453d7072f8Seschrock * it is one of our special test online cases, which is only 43463d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 43473d7072f8Seschrock * fault. 43483d7072f8Seschrock */ 43493d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 43503d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 43513d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 43523d7072f8Seschrock vd->vdev_removed = B_TRUE; 43533d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 4354ea8dc4b6Seschrock /* 4355cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 4356cb04b873SMark J Musante * mark it as "not available", which signifies that it was 4357cb04b873SMark J Musante * never there to begin with. Failure to open such a device 4358cb04b873SMark J Musante * is not considered an error. 4359ea8dc4b6Seschrock */ 4360cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 4361cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 4362560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 4363560e6e96Seschrock vd->vdev_not_present = 1; 4364560e6e96Seschrock 4365560e6e96Seschrock /* 4366560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 4367560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 4368560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 4369560e6e96Seschrock * want to post the ereport if the device was already in the 4370560e6e96Seschrock * CANT_OPEN state beforehand. 43713d7072f8Seschrock * 43723d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 43733d7072f8Seschrock * online the device in response to an insertion event. If we 43743d7072f8Seschrock * hit this case, then we have detected an insertion event for a 43753d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 43763d7072f8Seschrock * In this scenario, we don't post an ereport because we are 43773d7072f8Seschrock * about to replace the device, or attempt an online with 43783d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 4379560e6e96Seschrock */ 43803d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 43813d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 4382c5904d13Seschrock vd != spa->spa_root_vdev) { 4383ea8dc4b6Seschrock const char *class; 4384ea8dc4b6Seschrock 4385ea8dc4b6Seschrock switch (aux) { 4386ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 4387ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4388ea8dc4b6Seschrock break; 4389ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 4390ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4391ea8dc4b6Seschrock break; 4392ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 4393ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4394ea8dc4b6Seschrock break; 4395ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 4396ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4397ea8dc4b6Seschrock break; 4398ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 4399ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4400ea8dc4b6Seschrock break; 4401ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 4402ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4403ea8dc4b6Seschrock break; 44045711d393Sloli case VDEV_AUX_BAD_ASHIFT: 44055711d393Sloli class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; 44065711d393Sloli break; 4407ea8dc4b6Seschrock default: 4408ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4409ea8dc4b6Seschrock } 4410ea8dc4b6Seschrock 4411eb633035STom Caputi zfs_ereport_post(class, spa, vd, NULL, NULL, 4412eb633035STom Caputi save_state, 0); 4413ea8dc4b6Seschrock } 4414ea8dc4b6Seschrock 44153d7072f8Seschrock /* Erase any notion of persistent removed state */ 44163d7072f8Seschrock vd->vdev_removed = B_FALSE; 44173d7072f8Seschrock } else { 44183d7072f8Seschrock vd->vdev_removed = B_FALSE; 44193d7072f8Seschrock } 4420ea8dc4b6Seschrock 44218b33d774STim Haley if (!isopen && vd->vdev_parent) 44228b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 4423fa9e4066Sahrens } 442415e6edf1Sgw 44256f793812SPavel Zakharov boolean_t 44266f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 44276f793812SPavel Zakharov { 44286f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 44296f793812SPavel Zakharov 44306f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 44316f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 44326f793812SPavel Zakharov return (B_FALSE); 44336f793812SPavel Zakharov } 44346f793812SPavel Zakharov 44356f793812SPavel Zakharov return (B_TRUE); 44366f793812SPavel Zakharov } 44376f793812SPavel Zakharov 443815e6edf1Sgw /* 443915e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 4440c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 4441c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 444215e6edf1Sgw */ 444315e6edf1Sgw boolean_t 444415e6edf1Sgw vdev_is_bootable(vdev_t *vd) 444515e6edf1Sgw { 444615e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 444715e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 444815e6edf1Sgw 444915e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 445015e6edf1Sgw vd->vdev_children > 1) { 445121c878feSToomas Soome int non_indirect = 0; 445221c878feSToomas Soome 445321c878feSToomas Soome for (int c = 0; c < vd->vdev_children; c++) { 445421c878feSToomas Soome vdev_type = 445521c878feSToomas Soome vd->vdev_child[c]->vdev_ops->vdev_op_type; 445621c878feSToomas Soome if (strcmp(vdev_type, VDEV_TYPE_INDIRECT) != 0) 445721c878feSToomas Soome non_indirect++; 445821c878feSToomas Soome } 445921c878feSToomas Soome /* 446021c878feSToomas Soome * non_indirect > 1 means we have more than one 446121c878feSToomas Soome * top-level vdev, so we stop here. 446221c878feSToomas Soome */ 446321c878feSToomas Soome if (non_indirect > 1) 446421c878feSToomas Soome return (B_FALSE); 446521c878feSToomas Soome } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 446615e6edf1Sgw return (B_FALSE); 446715e6edf1Sgw } 446815e6edf1Sgw } 446915e6edf1Sgw 4470573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 447115e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 447215e6edf1Sgw return (B_FALSE); 447315e6edf1Sgw } 447415e6edf1Sgw return (B_TRUE); 447515e6edf1Sgw } 4476e6ca193dSGeorge Wilson 44775cabbc6bSPrashanth Sreenivasa boolean_t 44785cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 44795cabbc6bSPrashanth Sreenivasa { 44805cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 44815cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 44825cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 44835cabbc6bSPrashanth Sreenivasa return (B_FALSE); 44845cabbc6bSPrashanth Sreenivasa } else { 44855cabbc6bSPrashanth Sreenivasa return (B_TRUE); 44865cabbc6bSPrashanth Sreenivasa } 44875cabbc6bSPrashanth Sreenivasa } 44885cabbc6bSPrashanth Sreenivasa 44894b964adaSGeorge Wilson /* 44904b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 44914b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 44924b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 44934b964adaSGeorge Wilson */ 44944b964adaSGeorge Wilson boolean_t 44954b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 44964b964adaSGeorge Wilson { 44974b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 44984b964adaSGeorge Wilson !vd->vdev_removed) 44994b964adaSGeorge Wilson return (B_TRUE); 45004b964adaSGeorge Wilson 45014b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 45024b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 45034b964adaSGeorge Wilson return (B_TRUE); 45044b964adaSGeorge Wilson 45054b964adaSGeorge Wilson return (B_FALSE); 45064b964adaSGeorge Wilson } 45074b964adaSGeorge Wilson 4508573ca77eSGeorge Wilson /* 4509573ca77eSGeorge Wilson * Expand a vdev if possible. 4510573ca77eSGeorge Wilson */ 4511573ca77eSGeorge Wilson void 4512573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 4513573ca77eSGeorge Wilson { 4514573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 4515573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 451611f6a968SSerapheim Dimitropoulos ASSERT(vdev_is_concrete(vd)); 4517573ca77eSGeorge Wilson 45185cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 45195cabbc6bSPrashanth Sreenivasa 4520663207adSDon Brady if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 4521663207adSDon Brady vdev_is_concrete(vd)) { 4522663207adSDon Brady vdev_metaslab_group_create(vd); 4523573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 4524573ca77eSGeorge Wilson vdev_config_dirty(vd); 4525573ca77eSGeorge Wilson } 4526573ca77eSGeorge Wilson } 45271195e687SMark J Musante 45281195e687SMark J Musante /* 45291195e687SMark J Musante * Split a vdev. 45301195e687SMark J Musante */ 45311195e687SMark J Musante void 45321195e687SMark J Musante vdev_split(vdev_t *vd) 45331195e687SMark J Musante { 45341195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 45351195e687SMark J Musante 45361195e687SMark J Musante vdev_remove_child(pvd, vd); 45371195e687SMark J Musante vdev_compact_children(pvd); 45381195e687SMark J Musante 45391195e687SMark J Musante cvd = pvd->vdev_child[0]; 45401195e687SMark J Musante if (pvd->vdev_children == 1) { 45411195e687SMark J Musante vdev_remove_parent(cvd); 45421195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 45431195e687SMark J Musante } 45441195e687SMark J Musante vdev_propagate_state(cvd); 45451195e687SMark J Musante } 4546283b8460SGeorge.Wilson 4547283b8460SGeorge.Wilson void 4548283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 4549283b8460SGeorge.Wilson { 4550283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 4551283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 4552283b8460SGeorge.Wilson 4553283b8460SGeorge.Wilson vdev_deadman(cvd); 4554283b8460SGeorge.Wilson } 4555283b8460SGeorge.Wilson 4556283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 4557283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 4558283b8460SGeorge.Wilson 4559283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 456069962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 4561283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 4562283b8460SGeorge.Wilson zio_t *fio; 4563283b8460SGeorge.Wilson uint64_t delta; 4564283b8460SGeorge.Wilson 4565283b8460SGeorge.Wilson /* 4566283b8460SGeorge.Wilson * Look at the head of all the pending queues, 4567283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 4568283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 4569283b8460SGeorge.Wilson */ 457069962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 4571c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 4572c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 45733ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 45743ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 45753ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 4576283b8460SGeorge.Wilson vq->vq_io_complete_ts); 4577283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 4578283b8460SGeorge.Wilson "hung.", spa_name(spa)); 4579283b8460SGeorge.Wilson } 4580283b8460SGeorge.Wilson } 4581283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 4582283b8460SGeorge.Wilson } 4583283b8460SGeorge.Wilson } 4584e4c795beSTom Caputi 4585e4c795beSTom Caputi void 45860c06d385Sjwpoduska vdev_defer_resilver(vdev_t *vd) 4587e4c795beSTom Caputi { 45880c06d385Sjwpoduska ASSERT(vd->vdev_ops->vdev_op_leaf); 4589e4c795beSTom Caputi 45900c06d385Sjwpoduska vd->vdev_resilver_deferred = B_TRUE; 45910c06d385Sjwpoduska vd->vdev_spa->spa_resilver_deferred = B_TRUE; 45920c06d385Sjwpoduska } 45930c06d385Sjwpoduska 45940c06d385Sjwpoduska /* 45950c06d385Sjwpoduska * Clears the resilver deferred flag on all leaf devs under vd. Returns 45960c06d385Sjwpoduska * B_TRUE if we have devices that need to be resilvered and are available to 45970c06d385Sjwpoduska * accept resilver I/Os. 45980c06d385Sjwpoduska */ 45990c06d385Sjwpoduska boolean_t 46000c06d385Sjwpoduska vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) 46010c06d385Sjwpoduska { 46020c06d385Sjwpoduska boolean_t resilver_needed = B_FALSE; 46030c06d385Sjwpoduska spa_t *spa = vd->vdev_spa; 46040c06d385Sjwpoduska 46050c06d385Sjwpoduska for (int c = 0; c < vd->vdev_children; c++) { 46060c06d385Sjwpoduska vdev_t *cvd = vd->vdev_child[c]; 46070c06d385Sjwpoduska resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); 4608e4c795beSTom Caputi } 4609e4c795beSTom Caputi 46100c06d385Sjwpoduska if (vd == spa->spa_root_vdev && 46110c06d385Sjwpoduska spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { 46120c06d385Sjwpoduska spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); 46130c06d385Sjwpoduska vdev_config_dirty(vd); 46140c06d385Sjwpoduska spa->spa_resilver_deferred = B_FALSE; 46150c06d385Sjwpoduska return (resilver_needed); 46160c06d385Sjwpoduska } 46170c06d385Sjwpoduska 46180c06d385Sjwpoduska if (!vdev_is_concrete(vd) || vd->vdev_aux || 46190c06d385Sjwpoduska !vd->vdev_ops->vdev_op_leaf) 46200c06d385Sjwpoduska return (resilver_needed); 46210c06d385Sjwpoduska 46220c06d385Sjwpoduska vd->vdev_resilver_deferred = B_FALSE; 46230c06d385Sjwpoduska 46240c06d385Sjwpoduska return (!vdev_is_dead(vd) && !vd->vdev_offline && 46250c06d385Sjwpoduska vdev_resilver_needed(vd, NULL, NULL)); 4626e4c795beSTom Caputi } 4627084fd14fSBrian Behlendorf 4628084fd14fSBrian Behlendorf /* 4629084fd14fSBrian Behlendorf * Translate a logical range to the physical range for the specified vdev_t. 4630084fd14fSBrian Behlendorf * This function is initially called with a leaf vdev and will walk each 4631084fd14fSBrian Behlendorf * parent vdev until it reaches a top-level vdev. Once the top-level is 4632084fd14fSBrian Behlendorf * reached the physical range is initialized and the recursive function 4633084fd14fSBrian Behlendorf * begins to unwind. As it unwinds it calls the parent's vdev specific 4634084fd14fSBrian Behlendorf * translation function to do the real conversion. 4635084fd14fSBrian Behlendorf */ 4636084fd14fSBrian Behlendorf void 46374d7988d6SPaul Dagnelie vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, 46384d7988d6SPaul Dagnelie range_seg64_t *physical_rs) 4639084fd14fSBrian Behlendorf { 4640084fd14fSBrian Behlendorf /* 4641084fd14fSBrian Behlendorf * Walk up the vdev tree 4642084fd14fSBrian Behlendorf */ 4643084fd14fSBrian Behlendorf if (vd != vd->vdev_top) { 4644084fd14fSBrian Behlendorf vdev_xlate(vd->vdev_parent, logical_rs, physical_rs); 4645084fd14fSBrian Behlendorf } else { 4646084fd14fSBrian Behlendorf /* 4647084fd14fSBrian Behlendorf * We've reached the top-level vdev, initialize the 4648084fd14fSBrian Behlendorf * physical range to the logical range and start to 4649084fd14fSBrian Behlendorf * unwind. 4650084fd14fSBrian Behlendorf */ 4651084fd14fSBrian Behlendorf physical_rs->rs_start = logical_rs->rs_start; 4652084fd14fSBrian Behlendorf physical_rs->rs_end = logical_rs->rs_end; 4653084fd14fSBrian Behlendorf return; 4654084fd14fSBrian Behlendorf } 4655084fd14fSBrian Behlendorf 4656084fd14fSBrian Behlendorf vdev_t *pvd = vd->vdev_parent; 4657084fd14fSBrian Behlendorf ASSERT3P(pvd, !=, NULL); 4658084fd14fSBrian Behlendorf ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); 4659084fd14fSBrian Behlendorf 4660084fd14fSBrian Behlendorf /* 4661084fd14fSBrian Behlendorf * As this recursive function unwinds, translate the logical 4662084fd14fSBrian Behlendorf * range into its physical components by calling the 4663084fd14fSBrian Behlendorf * vdev specific translate function. 4664084fd14fSBrian Behlendorf */ 46654d7988d6SPaul Dagnelie range_seg64_t intermediate = { 0 }; 4666084fd14fSBrian Behlendorf pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate); 4667084fd14fSBrian Behlendorf 4668084fd14fSBrian Behlendorf physical_rs->rs_start = intermediate.rs_start; 4669084fd14fSBrian Behlendorf physical_rs->rs_end = intermediate.rs_end; 4670084fd14fSBrian Behlendorf } 4671