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> 28ce1577b0SDave Eddy * Copyright 2017 Joyent, Inc. 29fa9e4066Sahrens */ 30fa9e4066Sahrens 31fa9e4066Sahrens #include <sys/zfs_context.h> 32ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 33fa9e4066Sahrens #include <sys/spa.h> 34fa9e4066Sahrens #include <sys/spa_impl.h> 355cabbc6bSPrashanth Sreenivasa #include <sys/bpobj.h> 36fa9e4066Sahrens #include <sys/dmu.h> 37fa9e4066Sahrens #include <sys/dmu_tx.h> 385cabbc6bSPrashanth Sreenivasa #include <sys/dsl_dir.h> 39fa9e4066Sahrens #include <sys/vdev_impl.h> 40fa9e4066Sahrens #include <sys/uberblock_impl.h> 41fa9e4066Sahrens #include <sys/metaslab.h> 42fa9e4066Sahrens #include <sys/metaslab_impl.h> 43fa9e4066Sahrens #include <sys/space_map.h> 440713e232SGeorge Wilson #include <sys/space_reftree.h> 45fa9e4066Sahrens #include <sys/zio.h> 46fa9e4066Sahrens #include <sys/zap.h> 47fa9e4066Sahrens #include <sys/fs/zfs.h> 48c5904d13Seschrock #include <sys/arc.h> 49e6ca193dSGeorge Wilson #include <sys/zil.h> 503f9d6ad7SLin Ling #include <sys/dsl_scan.h> 51770499e1SDan Kimmel #include <sys/abd.h> 52094e47e9SGeorge Wilson #include <sys/vdev_initialize.h> 53fa9e4066Sahrens 54fa9e4066Sahrens /* 55fa9e4066Sahrens * Virtual device management. 56fa9e4066Sahrens */ 57fa9e4066Sahrens 58fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 59fa9e4066Sahrens &vdev_root_ops, 60fa9e4066Sahrens &vdev_raidz_ops, 61fa9e4066Sahrens &vdev_mirror_ops, 62fa9e4066Sahrens &vdev_replacing_ops, 6399653d4eSeschrock &vdev_spare_ops, 64fa9e4066Sahrens &vdev_disk_ops, 65fa9e4066Sahrens &vdev_file_ops, 66fa9e4066Sahrens &vdev_missing_ops, 6788ecc943SGeorge Wilson &vdev_hole_ops, 685cabbc6bSPrashanth Sreenivasa &vdev_indirect_ops, 69fa9e4066Sahrens NULL 70fa9e4066Sahrens }; 71fa9e4066Sahrens 72088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 73088f3894Sahrens int zfs_scrub_limit = 10; 7405b2b3b8Smishra 75a0b03b16SSerapheim Dimitropoulos /* default target for number of metaslabs per top-level vdev */ 76a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_count = 200; 7786714001SSerapheim Dimitropoulos 78b4bf0cf0SDon Brady /* minimum number of metaslabs per top-level vdev */ 79a0b03b16SSerapheim Dimitropoulos int zfs_vdev_min_ms_count = 16; 8086714001SSerapheim Dimitropoulos 81b4bf0cf0SDon Brady /* practical upper limit of total metaslabs per top-level vdev */ 82a0b03b16SSerapheim Dimitropoulos int zfs_vdev_ms_count_limit = 1ULL << 17; 83b4bf0cf0SDon Brady 84b4bf0cf0SDon Brady /* lower limit for metaslab size (512M) */ 85a0b03b16SSerapheim Dimitropoulos int zfs_vdev_default_ms_shift = 29; 8686714001SSerapheim Dimitropoulos 87a0b03b16SSerapheim Dimitropoulos /* upper limit for metaslab size (16G) */ 88a0b03b16SSerapheim Dimitropoulos int zfs_vdev_max_ms_shift = 34; 89b4bf0cf0SDon Brady 9086714001SSerapheim Dimitropoulos boolean_t vdev_validate_skip = B_FALSE; 9186714001SSerapheim Dimitropoulos 92bf3e216cSMatthew Ahrens /* 9386714001SSerapheim Dimitropoulos * Since the DTL space map of a vdev is not expected to have a lot of 9486714001SSerapheim Dimitropoulos * entries, we default its block size to 4K. 95bf3e216cSMatthew Ahrens */ 9686714001SSerapheim Dimitropoulos int vdev_dtl_sm_blksz = (1 << 12); 97bf3e216cSMatthew Ahrens 9886714001SSerapheim Dimitropoulos /* 9986714001SSerapheim Dimitropoulos * vdev-wide space maps that have lots of entries written to them at 10086714001SSerapheim Dimitropoulos * the end of each transaction can benefit from a higher I/O bandwidth 10186714001SSerapheim Dimitropoulos * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 10286714001SSerapheim Dimitropoulos */ 10386714001SSerapheim Dimitropoulos int vdev_standard_sm_blksz = (1 << 17); 1046f793812SPavel Zakharov 10593a1902eSMatthew Ahrens int zfs_ashift_min; 10693a1902eSMatthew Ahrens 1073ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 1083ee8c80cSPavel Zakharov void 1093ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 1103ee8c80cSPavel Zakharov { 1113ee8c80cSPavel Zakharov va_list adx; 1123ee8c80cSPavel Zakharov char buf[256]; 1133ee8c80cSPavel Zakharov 1143ee8c80cSPavel Zakharov va_start(adx, fmt); 1153ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 1163ee8c80cSPavel Zakharov va_end(adx); 1173ee8c80cSPavel Zakharov 1183ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 1193ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 1203ee8c80cSPavel Zakharov vd->vdev_path, buf); 1213ee8c80cSPavel Zakharov } else { 1223ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 1233ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 1243ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1253ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1263ee8c80cSPavel Zakharov } 1273ee8c80cSPavel Zakharov } 1283ee8c80cSPavel Zakharov 1296f793812SPavel Zakharov void 1306f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 1316f793812SPavel Zakharov { 1326f793812SPavel Zakharov char state[20]; 1336f793812SPavel Zakharov 1346f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 1356f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 1366f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 1376f793812SPavel Zakharov return; 1386f793812SPavel Zakharov } 1396f793812SPavel Zakharov 1406f793812SPavel Zakharov switch (vd->vdev_state) { 1416f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 1426f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 1436f793812SPavel Zakharov break; 1446f793812SPavel Zakharov case VDEV_STATE_CLOSED: 1456f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 1466f793812SPavel Zakharov break; 1476f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 1486f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 1496f793812SPavel Zakharov break; 1506f793812SPavel Zakharov case VDEV_STATE_REMOVED: 1516f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 1526f793812SPavel Zakharov break; 1536f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 1546f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 1556f793812SPavel Zakharov break; 1566f793812SPavel Zakharov case VDEV_STATE_FAULTED: 1576f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 1586f793812SPavel Zakharov break; 1596f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 1606f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 1616f793812SPavel Zakharov break; 1626f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 1636f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 1646f793812SPavel Zakharov break; 1656f793812SPavel Zakharov default: 1666f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 1676f793812SPavel Zakharov (uint_t)vd->vdev_state); 1686f793812SPavel Zakharov } 1696f793812SPavel Zakharov 1706f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 171c7a7b2faSAndriy Gapon "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, 1726f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 1736f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1746f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 1756f793812SPavel Zakharov 1766f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 1776f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 1786f793812SPavel Zakharov } 1796f793812SPavel Zakharov 180fa9e4066Sahrens /* 181fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 182fa9e4066Sahrens */ 183fa9e4066Sahrens static vdev_ops_t * 184fa9e4066Sahrens vdev_getops(const char *type) 185fa9e4066Sahrens { 186fa9e4066Sahrens vdev_ops_t *ops, **opspp; 187fa9e4066Sahrens 188fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 189fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 190fa9e4066Sahrens break; 191fa9e4066Sahrens 192fa9e4066Sahrens return (ops); 193fa9e4066Sahrens } 194fa9e4066Sahrens 195094e47e9SGeorge Wilson /* ARGSUSED */ 196094e47e9SGeorge Wilson void 197094e47e9SGeorge Wilson vdev_default_xlate(vdev_t *vd, const range_seg_t *in, range_seg_t *res) 198094e47e9SGeorge Wilson { 199094e47e9SGeorge Wilson res->rs_start = in->rs_start; 200094e47e9SGeorge Wilson res->rs_end = in->rs_end; 201094e47e9SGeorge Wilson } 202094e47e9SGeorge Wilson 203fa9e4066Sahrens /* 204fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 205fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 206fa9e4066Sahrens */ 207fa9e4066Sahrens uint64_t 208fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 209fa9e4066Sahrens { 210ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 211fa9e4066Sahrens uint64_t csize; 212fa9e4066Sahrens 213573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 214fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 215fa9e4066Sahrens asize = MAX(asize, csize); 216fa9e4066Sahrens } 217fa9e4066Sahrens 218fa9e4066Sahrens return (asize); 219fa9e4066Sahrens } 220fa9e4066Sahrens 2212a79c5feSlling /* 222573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 223573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 224573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 225573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 2262a79c5feSlling */ 2272a79c5feSlling uint64_t 228573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 2292a79c5feSlling { 230573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 2312a79c5feSlling 232573ca77eSGeorge Wilson /* 2334263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 234573ca77eSGeorge Wilson * just return our own asize. 235573ca77eSGeorge Wilson */ 236573ca77eSGeorge Wilson if (pvd == NULL) 237573ca77eSGeorge Wilson return (vd->vdev_asize); 2382a79c5feSlling 2392a79c5feSlling /* 240573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 241573ca77eSGeorge Wilson * to the nearest metaslab. 2422a79c5feSlling */ 243573ca77eSGeorge Wilson if (vd == vd->vdev_top) 244573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2452a79c5feSlling 246573ca77eSGeorge Wilson /* 247573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 248573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 249573ca77eSGeorge Wilson */ 250573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 251c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 252c040c10cSSteven Hartland pvd->vdev_children); 2532a79c5feSlling 254573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 255573ca77eSGeorge Wilson } 2562a79c5feSlling 257573ca77eSGeorge Wilson void 258573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 259573ca77eSGeorge Wilson { 260573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 261573ca77eSGeorge Wilson 262573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 263573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2642a79c5feSlling } 2652a79c5feSlling 266fa9e4066Sahrens vdev_t * 267fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 268fa9e4066Sahrens { 269fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 270fa9e4066Sahrens 271e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 272e05725b1Sbonwick 273088f3894Sahrens if (vdev < rvd->vdev_children) { 274088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 275fa9e4066Sahrens return (rvd->vdev_child[vdev]); 276088f3894Sahrens } 277fa9e4066Sahrens 278fa9e4066Sahrens return (NULL); 279fa9e4066Sahrens } 280fa9e4066Sahrens 281fa9e4066Sahrens vdev_t * 282fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 283fa9e4066Sahrens { 284fa9e4066Sahrens vdev_t *mvd; 285fa9e4066Sahrens 2860e34b6a7Sbonwick if (vd->vdev_guid == guid) 287fa9e4066Sahrens return (vd); 288fa9e4066Sahrens 289573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 290fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 291fa9e4066Sahrens NULL) 292fa9e4066Sahrens return (mvd); 293fa9e4066Sahrens 294fa9e4066Sahrens return (NULL); 295fa9e4066Sahrens } 296fa9e4066Sahrens 29712380e1eSArne Jansen static int 29812380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 29912380e1eSArne Jansen { 30012380e1eSArne Jansen int n = 0; 30112380e1eSArne Jansen 30212380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 30312380e1eSArne Jansen return (1); 30412380e1eSArne Jansen 30512380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 30612380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 30712380e1eSArne Jansen 30812380e1eSArne Jansen return (n); 30912380e1eSArne Jansen } 31012380e1eSArne Jansen 31112380e1eSArne Jansen int 31212380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 31312380e1eSArne Jansen { 31412380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 31512380e1eSArne Jansen } 31612380e1eSArne Jansen 317fa9e4066Sahrens void 318fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 319fa9e4066Sahrens { 320fa9e4066Sahrens size_t oldsize, newsize; 321fa9e4066Sahrens uint64_t id = cvd->vdev_id; 322fa9e4066Sahrens vdev_t **newchild; 32381cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 324fa9e4066Sahrens 32581cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 326fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 327fa9e4066Sahrens 328fa9e4066Sahrens cvd->vdev_parent = pvd; 329fa9e4066Sahrens 330fa9e4066Sahrens if (pvd == NULL) 331fa9e4066Sahrens return; 332fa9e4066Sahrens 333fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 334fa9e4066Sahrens 335fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 336fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 337fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 338fa9e4066Sahrens 339fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 340fa9e4066Sahrens if (pvd->vdev_child != NULL) { 341fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 342fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 343fa9e4066Sahrens } 344fa9e4066Sahrens 345fa9e4066Sahrens pvd->vdev_child = newchild; 346fa9e4066Sahrens pvd->vdev_child[id] = cvd; 347fa9e4066Sahrens 348fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 349fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 350fa9e4066Sahrens 351fa9e4066Sahrens /* 352fa9e4066Sahrens * Walk up all ancestors to update guid sum. 353fa9e4066Sahrens */ 354fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 355fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 356*e0f1c0afSOlaf Faaland 357*e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 358*e0f1c0afSOlaf Faaland list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); 359*e0f1c0afSOlaf Faaland cvd->vdev_spa->spa_leaf_list_gen++; 360*e0f1c0afSOlaf Faaland } 361fa9e4066Sahrens } 362fa9e4066Sahrens 363fa9e4066Sahrens void 364fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 365fa9e4066Sahrens { 366fa9e4066Sahrens int c; 367fa9e4066Sahrens uint_t id = cvd->vdev_id; 368fa9e4066Sahrens 369fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 370fa9e4066Sahrens 371fa9e4066Sahrens if (pvd == NULL) 372fa9e4066Sahrens return; 373fa9e4066Sahrens 374fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 375fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 376fa9e4066Sahrens 377fa9e4066Sahrens pvd->vdev_child[id] = NULL; 378fa9e4066Sahrens cvd->vdev_parent = NULL; 379fa9e4066Sahrens 380fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 381fa9e4066Sahrens if (pvd->vdev_child[c]) 382fa9e4066Sahrens break; 383fa9e4066Sahrens 384fa9e4066Sahrens if (c == pvd->vdev_children) { 385fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 386fa9e4066Sahrens pvd->vdev_child = NULL; 387fa9e4066Sahrens pvd->vdev_children = 0; 388fa9e4066Sahrens } 389fa9e4066Sahrens 390*e0f1c0afSOlaf Faaland if (cvd->vdev_ops->vdev_op_leaf) { 391*e0f1c0afSOlaf Faaland spa_t *spa = cvd->vdev_spa; 392*e0f1c0afSOlaf Faaland list_remove(&spa->spa_leaf_list, cvd); 393*e0f1c0afSOlaf Faaland spa->spa_leaf_list_gen++; 394*e0f1c0afSOlaf Faaland } 395*e0f1c0afSOlaf Faaland 396fa9e4066Sahrens /* 397fa9e4066Sahrens * Walk up all ancestors to update guid sum. 398fa9e4066Sahrens */ 399fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 400fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 401fa9e4066Sahrens } 402fa9e4066Sahrens 403fa9e4066Sahrens /* 404fa9e4066Sahrens * Remove any holes in the child array. 405fa9e4066Sahrens */ 406fa9e4066Sahrens void 407fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 408fa9e4066Sahrens { 409fa9e4066Sahrens vdev_t **newchild, *cvd; 410fa9e4066Sahrens int oldc = pvd->vdev_children; 411573ca77eSGeorge Wilson int newc; 412fa9e4066Sahrens 413e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 414fa9e4066Sahrens 415573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 416fa9e4066Sahrens if (pvd->vdev_child[c]) 417fa9e4066Sahrens newc++; 418fa9e4066Sahrens 419fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 420fa9e4066Sahrens 421573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 422fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 423fa9e4066Sahrens newchild[newc] = cvd; 424fa9e4066Sahrens cvd->vdev_id = newc++; 425fa9e4066Sahrens } 426fa9e4066Sahrens } 427fa9e4066Sahrens 428fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 429fa9e4066Sahrens pvd->vdev_child = newchild; 430fa9e4066Sahrens pvd->vdev_children = newc; 431fa9e4066Sahrens } 432fa9e4066Sahrens 433fa9e4066Sahrens /* 434fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 435fa9e4066Sahrens */ 43688ecc943SGeorge Wilson vdev_t * 437fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 438fa9e4066Sahrens { 439fa9e4066Sahrens vdev_t *vd; 4405cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 441fa9e4066Sahrens 442fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 4435cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 444fa9e4066Sahrens 4450e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4460e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4470e34b6a7Sbonwick spa->spa_root_vdev = vd; 448e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4490e34b6a7Sbonwick } 4500e34b6a7Sbonwick 45188ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4520e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4530e34b6a7Sbonwick /* 4540e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4550e34b6a7Sbonwick * which must be unique among all pools. 4560e34b6a7Sbonwick */ 4571195e687SMark J Musante guid = spa_generate_guid(NULL); 4580e34b6a7Sbonwick } else { 4590e34b6a7Sbonwick /* 4600e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4610e34b6a7Sbonwick */ 4621195e687SMark J Musante guid = spa_generate_guid(spa); 4630e34b6a7Sbonwick } 4640e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4650e34b6a7Sbonwick } 4660e34b6a7Sbonwick 467fa9e4066Sahrens vd->vdev_spa = spa; 468fa9e4066Sahrens vd->vdev_id = id; 469fa9e4066Sahrens vd->vdev_guid = guid; 470fa9e4066Sahrens vd->vdev_guid_sum = guid; 471fa9e4066Sahrens vd->vdev_ops = ops; 472fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 47388ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 4745cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 4755cabbc6bSPrashanth Sreenivasa 4765cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 4775cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 4785cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); 479fa9e4066Sahrens 480*e0f1c0afSOlaf Faaland list_link_init(&vd->vdev_leaf_node); 481fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 4825ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 483e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 4840f7643c7SGeorge Wilson mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); 485094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); 486094e47e9SGeorge Wilson mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); 487094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); 488094e47e9SGeorge Wilson cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); 489094e47e9SGeorge Wilson 4908ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 4915cabbc6bSPrashanth Sreenivasa vd->vdev_dtl[t] = range_tree_create(NULL, NULL); 4928ad4d6ddSJeff Bonwick } 493b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 494fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 495b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 496fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 497fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 4983d7072f8Seschrock vdev_queue_init(vd); 4993d7072f8Seschrock vdev_cache_init(vd); 500fa9e4066Sahrens 501fa9e4066Sahrens return (vd); 502fa9e4066Sahrens } 503fa9e4066Sahrens 504fa9e4066Sahrens /* 505fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 506fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 507fa9e4066Sahrens * different for each case. 508fa9e4066Sahrens */ 50999653d4eSeschrock int 51099653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 51199653d4eSeschrock int alloctype) 512fa9e4066Sahrens { 513fa9e4066Sahrens vdev_ops_t *ops; 514fa9e4066Sahrens char *type; 5158654d025Sperrin uint64_t guid = 0, islog, nparity; 516fa9e4066Sahrens vdev_t *vd; 5175cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 518fa9e4066Sahrens 519e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 520fa9e4066Sahrens 521fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 522be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 523fa9e4066Sahrens 524fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 525be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 526fa9e4066Sahrens 527fa9e4066Sahrens /* 528fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 529fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 530fa9e4066Sahrens */ 531fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 532fa9e4066Sahrens uint64_t label_id; 533fa9e4066Sahrens 534fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 535fa9e4066Sahrens label_id != id) 536be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 537fa9e4066Sahrens 538fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 539be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 54099653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 54199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 542be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 543fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 544fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 545be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 54621ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 54721ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 548be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 549fa9e4066Sahrens } 550fa9e4066Sahrens 55199653d4eSeschrock /* 55299653d4eSeschrock * The first allocated vdev must be of type 'root'. 55399653d4eSeschrock */ 55499653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 555be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 55699653d4eSeschrock 5578654d025Sperrin /* 5588654d025Sperrin * Determine whether we're a log vdev. 5598654d025Sperrin */ 5608654d025Sperrin islog = 0; 5618654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 562990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 563be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 564fa9e4066Sahrens 56588ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 566be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 56788ecc943SGeorge Wilson 56899653d4eSeschrock /* 5698654d025Sperrin * Set the nparity property for RAID-Z vdevs. 57099653d4eSeschrock */ 5718654d025Sperrin nparity = -1ULL; 57299653d4eSeschrock if (ops == &vdev_raidz_ops) { 57399653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 5748654d025Sperrin &nparity) == 0) { 575b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 576be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 57799653d4eSeschrock /* 578f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 579f94275ceSAdam Leventhal * device. 58099653d4eSeschrock */ 581f94275ceSAdam Leventhal if (nparity > 1 && 582f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 583be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 584f94275ceSAdam Leventhal if (nparity > 2 && 585f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 586be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 58799653d4eSeschrock } else { 58899653d4eSeschrock /* 58999653d4eSeschrock * We require the parity to be specified for SPAs that 59099653d4eSeschrock * support multiple parity levels. 59199653d4eSeschrock */ 592f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 593be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 59499653d4eSeschrock /* 59599653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 59699653d4eSeschrock */ 5978654d025Sperrin nparity = 1; 59899653d4eSeschrock } 59999653d4eSeschrock } else { 6008654d025Sperrin nparity = 0; 60199653d4eSeschrock } 6028654d025Sperrin ASSERT(nparity != -1ULL); 6038654d025Sperrin 6048654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 6055cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 6068654d025Sperrin 6078654d025Sperrin vd->vdev_islog = islog; 6088654d025Sperrin vd->vdev_nparity = nparity; 6098654d025Sperrin 6108654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 6118654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 6128654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 6138654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 6148654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 6158654d025Sperrin &vd->vdev_physpath) == 0) 6168654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 6176809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 6186809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 61999653d4eSeschrock 620afefbcddSeschrock /* 621afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 622afefbcddSeschrock * as -1. 623afefbcddSeschrock */ 624afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 625afefbcddSeschrock &vd->vdev_wholedisk) != 0) 626afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 627afefbcddSeschrock 6285cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 6295cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 6305cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 6315cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 6325cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 6335cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 6345cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 6355cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 6365cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 6375cabbc6bSPrashanth Sreenivasa 638ea8dc4b6Seschrock /* 639ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 640ea8dc4b6Seschrock * was not present at the time of import. 641ea8dc4b6Seschrock */ 6426809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 6436809eb4eSEric Schrock &vd->vdev_not_present); 644ea8dc4b6Seschrock 645ecc2d604Sbonwick /* 646ecc2d604Sbonwick * Get the alignment requirement. 647ecc2d604Sbonwick */ 648ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 649ecc2d604Sbonwick 65088ecc943SGeorge Wilson /* 65188ecc943SGeorge Wilson * Retrieve the vdev creation time. 65288ecc943SGeorge Wilson */ 65388ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 65488ecc943SGeorge Wilson &vd->vdev_crtxg); 65588ecc943SGeorge Wilson 656fa9e4066Sahrens /* 657fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 658fa9e4066Sahrens */ 6591195e687SMark J Musante if (parent && !parent->vdev_parent && 6601195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 661fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 662fa9e4066Sahrens &vd->vdev_ms_array); 663fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 664fa9e4066Sahrens &vd->vdev_ms_shift); 665fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 666fa9e4066Sahrens &vd->vdev_asize); 6673f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 6683f9d6ad7SLin Ling &vd->vdev_removing); 669215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 670215198a6SJoe Stein &vd->vdev_top_zap); 671215198a6SJoe Stein } else { 672215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 673fa9e4066Sahrens } 674fa9e4066Sahrens 675cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 676a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 6779f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 6781195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 6799f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 680a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 681f78cdc34SPaul Dagnelie spa_log_class(spa) : spa_normal_class(spa), vd, 682f78cdc34SPaul Dagnelie spa->spa_alloc_count); 683a1521560SJeff Bonwick } 684a1521560SJeff Bonwick 685215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 686215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 687215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 688215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 689215198a6SJoe Stein } else { 690215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 691215198a6SJoe Stein } 692215198a6SJoe Stein 693fa9e4066Sahrens /* 6943d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 695fa9e4066Sahrens */ 696215198a6SJoe Stein 697c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 69821ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 69921ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 700c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 701c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 7020713e232SGeorge Wilson &vd->vdev_dtl_object); 703c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 704c5904d13Seschrock &vd->vdev_unspare); 705c5904d13Seschrock } 70621ecdf64SLin Ling 70721ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 70821ecdf64SLin Ling uint64_t spare = 0; 70921ecdf64SLin Ling 71021ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 71121ecdf64SLin Ling &spare) == 0 && spare) 71221ecdf64SLin Ling spa_spare_add(vd); 71321ecdf64SLin Ling } 71421ecdf64SLin Ling 715ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 716ecc2d604Sbonwick &vd->vdev_offline); 717c5904d13Seschrock 718b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 719b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 720cb04b873SMark J Musante 7213d7072f8Seschrock /* 7223d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 7233d7072f8Seschrock * state, as the diagnosis made on another system may not be 724069f55e2SEric Schrock * valid in the current context. Local vdevs will 725069f55e2SEric Schrock * remain in the faulted state. 7263d7072f8Seschrock */ 727b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 7283d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 7293d7072f8Seschrock &vd->vdev_faulted); 7303d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 7313d7072f8Seschrock &vd->vdev_degraded); 7323d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 7333d7072f8Seschrock &vd->vdev_removed); 734069f55e2SEric Schrock 735069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 736069f55e2SEric Schrock char *aux; 737069f55e2SEric Schrock 738069f55e2SEric Schrock vd->vdev_label_aux = 739069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 740069f55e2SEric Schrock if (nvlist_lookup_string(nv, 741069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 742069f55e2SEric Schrock strcmp(aux, "external") == 0) 743069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 744069f55e2SEric Schrock } 7453d7072f8Seschrock } 746fa9e4066Sahrens } 747fa9e4066Sahrens 748fa9e4066Sahrens /* 749fa9e4066Sahrens * Add ourselves to the parent's list of children. 750fa9e4066Sahrens */ 751fa9e4066Sahrens vdev_add_child(parent, vd); 752fa9e4066Sahrens 75399653d4eSeschrock *vdp = vd; 75499653d4eSeschrock 75599653d4eSeschrock return (0); 756fa9e4066Sahrens } 757fa9e4066Sahrens 758fa9e4066Sahrens void 759fa9e4066Sahrens vdev_free(vdev_t *vd) 760fa9e4066Sahrens { 7613d7072f8Seschrock spa_t *spa = vd->vdev_spa; 762094e47e9SGeorge Wilson ASSERT3P(vd->vdev_initialize_thread, ==, NULL); 763fa9e4066Sahrens 764fa9e4066Sahrens /* 765fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 766fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 767fa9e4066Sahrens */ 768fa9e4066Sahrens vdev_close(vd); 769fa9e4066Sahrens 770e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 771b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 772fa9e4066Sahrens 773fa9e4066Sahrens /* 774fa9e4066Sahrens * Free all children. 775fa9e4066Sahrens */ 776573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 777fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 778fa9e4066Sahrens 779fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 780fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 781094e47e9SGeorge Wilson ASSERT(vd->vdev_initialize_thread == NULL); 782fa9e4066Sahrens 783fa9e4066Sahrens /* 784fa9e4066Sahrens * Discard allocation state. 785fa9e4066Sahrens */ 786a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 787fa9e4066Sahrens vdev_metaslab_fini(vd); 788a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 789a1521560SJeff Bonwick } 790fa9e4066Sahrens 791fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 792fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 793fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 794fa9e4066Sahrens 795fa9e4066Sahrens /* 796fa9e4066Sahrens * Remove this vdev from its parent's child list. 797fa9e4066Sahrens */ 798fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 799fa9e4066Sahrens 800fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 801*e0f1c0afSOlaf Faaland ASSERT(!list_link_active(&vd->vdev_leaf_node)); 802fa9e4066Sahrens 8033d7072f8Seschrock /* 8043d7072f8Seschrock * Clean up vdev structure. 8053d7072f8Seschrock */ 8063d7072f8Seschrock vdev_queue_fini(vd); 8073d7072f8Seschrock vdev_cache_fini(vd); 8083d7072f8Seschrock 8093d7072f8Seschrock if (vd->vdev_path) 8103d7072f8Seschrock spa_strfree(vd->vdev_path); 8113d7072f8Seschrock if (vd->vdev_devid) 8123d7072f8Seschrock spa_strfree(vd->vdev_devid); 8133d7072f8Seschrock if (vd->vdev_physpath) 8143d7072f8Seschrock spa_strfree(vd->vdev_physpath); 8156809eb4eSEric Schrock if (vd->vdev_fru) 8166809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 8173d7072f8Seschrock 8183d7072f8Seschrock if (vd->vdev_isspare) 8193d7072f8Seschrock spa_spare_remove(vd); 820fa94a07fSbrendan if (vd->vdev_isl2cache) 821fa94a07fSbrendan spa_l2cache_remove(vd); 8223d7072f8Seschrock 8233d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 8243d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 8258ad4d6ddSJeff Bonwick 8263d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 8270713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 8288ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 8290713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 8300713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 8318ad4d6ddSJeff Bonwick } 8323d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 8338ad4d6ddSJeff Bonwick 8345cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 8355cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 8365cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 8375cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 8385cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 8395cabbc6bSPrashanth Sreenivasa } 8405cabbc6bSPrashanth Sreenivasa 8415cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 8425cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 8435cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 8445cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 8455cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 8465cabbc6bSPrashanth Sreenivasa } 8475cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 8485cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 8495cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 8505cabbc6bSPrashanth Sreenivasa 8510f7643c7SGeorge Wilson mutex_destroy(&vd->vdev_queue_lock); 8523d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 8533d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 854e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 855094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_lock); 856094e47e9SGeorge Wilson mutex_destroy(&vd->vdev_initialize_io_lock); 857094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_io_cv); 858094e47e9SGeorge Wilson cv_destroy(&vd->vdev_initialize_cv); 8593d7072f8Seschrock 8603d7072f8Seschrock if (vd == spa->spa_root_vdev) 8613d7072f8Seschrock spa->spa_root_vdev = NULL; 8623d7072f8Seschrock 8633d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 864fa9e4066Sahrens } 865fa9e4066Sahrens 866fa9e4066Sahrens /* 867fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 868fa9e4066Sahrens */ 869fa9e4066Sahrens static void 870fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 871fa9e4066Sahrens { 872fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 873fa9e4066Sahrens metaslab_t *msp; 874fa9e4066Sahrens vdev_t *vd; 875fa9e4066Sahrens int t; 876fa9e4066Sahrens 877fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 878fa9e4066Sahrens 879fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 880fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 881fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 882215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 883fa9e4066Sahrens 884fa9e4066Sahrens svd->vdev_ms_array = 0; 885fa9e4066Sahrens svd->vdev_ms_shift = 0; 886fa9e4066Sahrens svd->vdev_ms_count = 0; 887215198a6SJoe Stein svd->vdev_top_zap = 0; 888fa9e4066Sahrens 889cd0837ccSGeorge Wilson if (tvd->vdev_mg) 890cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 891fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 892fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 893fa9e4066Sahrens 894fa9e4066Sahrens svd->vdev_mg = NULL; 895fa9e4066Sahrens svd->vdev_ms = NULL; 896ecc2d604Sbonwick 897ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 898ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 899fa9e4066Sahrens 90086714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 90186714001SSerapheim Dimitropoulos svd->vdev_checkpoint_sm = NULL; 90286714001SSerapheim Dimitropoulos 903fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 904fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 90599653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 906fa9e4066Sahrens 907fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 908fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 90999653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 910fa9e4066Sahrens 9113a4b1be9SMatthew Ahrens /* 9123a4b1be9SMatthew Ahrens * State which may be set on a top-level vdev that's in the 9133a4b1be9SMatthew Ahrens * process of being removed. 9143a4b1be9SMatthew Ahrens */ 9153a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_births_object); 9163a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); 9173a4b1be9SMatthew Ahrens ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); 9183a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); 9193a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_indirect_births, ==, NULL); 9203a4b1be9SMatthew Ahrens ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); 9213a4b1be9SMatthew Ahrens ASSERT0(tvd->vdev_removing); 9223a4b1be9SMatthew Ahrens tvd->vdev_removing = svd->vdev_removing; 9233a4b1be9SMatthew Ahrens tvd->vdev_indirect_config = svd->vdev_indirect_config; 9243a4b1be9SMatthew Ahrens tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; 9253a4b1be9SMatthew Ahrens tvd->vdev_indirect_births = svd->vdev_indirect_births; 9263a4b1be9SMatthew Ahrens range_tree_swap(&svd->vdev_obsolete_segments, 9273a4b1be9SMatthew Ahrens &tvd->vdev_obsolete_segments); 9283a4b1be9SMatthew Ahrens tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; 9293a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_mapping_object = 0; 9303a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_births_object = 0; 9313a4b1be9SMatthew Ahrens svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; 9323a4b1be9SMatthew Ahrens svd->vdev_indirect_mapping = NULL; 9333a4b1be9SMatthew Ahrens svd->vdev_indirect_births = NULL; 9343a4b1be9SMatthew Ahrens svd->vdev_obsolete_sm = NULL; 9353a4b1be9SMatthew Ahrens svd->vdev_removing = 0; 9363a4b1be9SMatthew Ahrens 937fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 938fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 939fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 940fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 941fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 942fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 943fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 944fa9e4066Sahrens } 945fa9e4066Sahrens 946e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 947fa9e4066Sahrens vdev_config_clean(svd); 948fa9e4066Sahrens vdev_config_dirty(tvd); 949fa9e4066Sahrens } 950fa9e4066Sahrens 951e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 952e14bb325SJeff Bonwick vdev_state_clean(svd); 953e14bb325SJeff Bonwick vdev_state_dirty(tvd); 954e14bb325SJeff Bonwick } 955e14bb325SJeff Bonwick 95699653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 95799653d4eSeschrock svd->vdev_deflate_ratio = 0; 9588654d025Sperrin 9598654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 9608654d025Sperrin svd->vdev_islog = 0; 961fa9e4066Sahrens } 962fa9e4066Sahrens 963fa9e4066Sahrens static void 964fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 965fa9e4066Sahrens { 966fa9e4066Sahrens if (vd == NULL) 967fa9e4066Sahrens return; 968fa9e4066Sahrens 969fa9e4066Sahrens vd->vdev_top = tvd; 970fa9e4066Sahrens 971573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 972fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 973fa9e4066Sahrens } 974fa9e4066Sahrens 975fa9e4066Sahrens /* 976fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 977fa9e4066Sahrens */ 978fa9e4066Sahrens vdev_t * 979fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 980fa9e4066Sahrens { 981fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 982fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 983fa9e4066Sahrens vdev_t *mvd; 984fa9e4066Sahrens 985e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 986fa9e4066Sahrens 987fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 988ecc2d604Sbonwick 989ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 990573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 9914263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 9925cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 993ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 994ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 99588ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 996ecc2d604Sbonwick 997fa9e4066Sahrens vdev_remove_child(pvd, cvd); 998fa9e4066Sahrens vdev_add_child(pvd, mvd); 999fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 1000fa9e4066Sahrens vdev_add_child(mvd, cvd); 1001fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1002fa9e4066Sahrens 1003fa9e4066Sahrens if (mvd == mvd->vdev_top) 1004fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 1005fa9e4066Sahrens 1006fa9e4066Sahrens return (mvd); 1007fa9e4066Sahrens } 1008fa9e4066Sahrens 1009fa9e4066Sahrens /* 1010fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 1011fa9e4066Sahrens */ 1012fa9e4066Sahrens void 1013fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 1014fa9e4066Sahrens { 1015fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 1016fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 1017fa9e4066Sahrens 1018e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1019fa9e4066Sahrens 1020fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 1021fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 102299653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 102399653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 1024ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 1025fa9e4066Sahrens 1026fa9e4066Sahrens vdev_remove_child(mvd, cvd); 1027fa9e4066Sahrens vdev_remove_child(pvd, mvd); 10288ad4d6ddSJeff Bonwick 102999653d4eSeschrock /* 1030e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 1031e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 1032e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 1033e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 103499653d4eSeschrock */ 10358ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 10368ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 10371195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 10388ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 10398ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 10408ad4d6ddSJeff Bonwick } 1041e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 1042e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 1043fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 1044fa9e4066Sahrens 1045fa9e4066Sahrens if (cvd == cvd->vdev_top) 1046fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 1047fa9e4066Sahrens 1048fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 1049fa9e4066Sahrens vdev_free(mvd); 1050fa9e4066Sahrens } 1051fa9e4066Sahrens 1052ea8dc4b6Seschrock int 1053fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 1054fa9e4066Sahrens { 1055fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 1056ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 1057ecc2d604Sbonwick uint64_t m; 1058fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 1059fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 1060ecc2d604Sbonwick metaslab_t **mspp; 1061ecc2d604Sbonwick int error; 1062fa9e4066Sahrens 1063a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1064a1521560SJeff Bonwick 106588ecc943SGeorge Wilson /* 106688ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 106788ecc943SGeorge Wilson */ 106888ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 10690e34b6a7Sbonwick return (0); 10700e34b6a7Sbonwick 107188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 107288ecc943SGeorge Wilson 1073fa9e4066Sahrens ASSERT(oldc <= newc); 1074fa9e4066Sahrens 1075ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1076fa9e4066Sahrens 1077ecc2d604Sbonwick if (oldc != 0) { 1078ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1079ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1080ecc2d604Sbonwick } 1081fa9e4066Sahrens 1082ecc2d604Sbonwick vd->vdev_ms = mspp; 1083ecc2d604Sbonwick vd->vdev_ms_count = newc; 1084ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 10850713e232SGeorge Wilson uint64_t object = 0; 10860713e232SGeorge Wilson 10875cabbc6bSPrashanth Sreenivasa /* 10885cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 10895cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 10905cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 10915cabbc6bSPrashanth Sreenivasa */ 10925cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1093ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 10947bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 10957bfdf011SNeil Perrin DMU_READ_PREFETCH); 10963ee8c80cSPavel Zakharov if (error != 0) { 10973ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 10983ee8c80cSPavel Zakharov "array [error=%d]", error); 1099ecc2d604Sbonwick return (error); 11003ee8c80cSPavel Zakharov } 1101fa9e4066Sahrens } 11021e9bd7ecSPrakash Surya 11031e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 11041e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 11053ee8c80cSPavel Zakharov if (error != 0) { 11063ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 11073ee8c80cSPavel Zakharov error); 11081e9bd7ecSPrakash Surya return (error); 11093ee8c80cSPavel Zakharov } 1110fa9e4066Sahrens } 1111fa9e4066Sahrens 1112a1521560SJeff Bonwick if (txg == 0) 1113a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1114a1521560SJeff Bonwick 11153f9d6ad7SLin Ling /* 11163f9d6ad7SLin Ling * If the vdev is being removed we don't activate 11173f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 11183f9d6ad7SLin Ling * allocations are performed on this device. 11193f9d6ad7SLin Ling */ 11203f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 1121a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1122a1521560SJeff Bonwick 1123a1521560SJeff Bonwick if (txg == 0) 1124a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1125a1521560SJeff Bonwick 1126ea8dc4b6Seschrock return (0); 1127fa9e4066Sahrens } 1128fa9e4066Sahrens 1129fa9e4066Sahrens void 1130fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1131fa9e4066Sahrens { 113286714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 113386714001SSerapheim Dimitropoulos ASSERT(spa_feature_is_active(vd->vdev_spa, 113486714001SSerapheim Dimitropoulos SPA_FEATURE_POOL_CHECKPOINT)); 113586714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 113686714001SSerapheim Dimitropoulos /* 113786714001SSerapheim Dimitropoulos * Even though we close the space map, we need to set its 113886714001SSerapheim Dimitropoulos * pointer to NULL. The reason is that vdev_metaslab_fini() 113986714001SSerapheim Dimitropoulos * may be called multiple times for certain operations 114086714001SSerapheim Dimitropoulos * (i.e. when destroying a pool) so we need to ensure that 114186714001SSerapheim Dimitropoulos * this clause never executes twice. This logic is similar 114286714001SSerapheim Dimitropoulos * to the one used for the vdev_ms clause below. 114386714001SSerapheim Dimitropoulos */ 114486714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 114586714001SSerapheim Dimitropoulos } 114686714001SSerapheim Dimitropoulos 1147fa9e4066Sahrens if (vd->vdev_ms != NULL) { 11485cabbc6bSPrashanth Sreenivasa uint64_t count = vd->vdev_ms_count; 11495cabbc6bSPrashanth Sreenivasa 1150a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 11515cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 11520713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 11530713e232SGeorge Wilson 11540713e232SGeorge Wilson if (msp != NULL) 11550713e232SGeorge Wilson metaslab_fini(msp); 11560713e232SGeorge Wilson } 1157fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1158fa9e4066Sahrens vd->vdev_ms = NULL; 11595cabbc6bSPrashanth Sreenivasa 11605cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1161fa9e4066Sahrens } 11625cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1163fa9e4066Sahrens } 1164fa9e4066Sahrens 1165e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1166e14bb325SJeff Bonwick boolean_t vps_readable; 1167e14bb325SJeff Bonwick boolean_t vps_writeable; 1168e14bb325SJeff Bonwick int vps_flags; 1169e14bb325SJeff Bonwick } vdev_probe_stats_t; 1170e14bb325SJeff Bonwick 1171e14bb325SJeff Bonwick static void 1172e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 11730a4e9518Sgw { 11748ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1175a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1176e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1177a3f829aeSBill Moore 1178a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1179e14bb325SJeff Bonwick 1180e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1181e14bb325SJeff Bonwick if (zio->io_error == 0) 1182e14bb325SJeff Bonwick vps->vps_readable = 1; 11838ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1184a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1185770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1186e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1187e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1188e14bb325SJeff Bonwick } else { 1189770499e1SDan Kimmel abd_free(zio->io_abd); 1190e14bb325SJeff Bonwick } 1191e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1192e14bb325SJeff Bonwick if (zio->io_error == 0) 1193e14bb325SJeff Bonwick vps->vps_writeable = 1; 1194770499e1SDan Kimmel abd_free(zio->io_abd); 1195e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1196a3f829aeSBill Moore zio_t *pio; 1197e14bb325SJeff Bonwick 1198e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1199e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1200e14bb325SJeff Bonwick 1201e14bb325SJeff Bonwick if (vdev_readable(vd) && 12028ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1203e14bb325SJeff Bonwick zio->io_error = 0; 1204e14bb325SJeff Bonwick } else { 1205e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 12063ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1207e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 12088ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 1209be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1210e14bb325SJeff Bonwick } 1211a3f829aeSBill Moore 1212a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1213a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1214a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1215a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1216a3f829aeSBill Moore 12170f7643c7SGeorge Wilson zio_link_t *zl = NULL; 12180f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1219a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1220be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1221a3f829aeSBill Moore 1222e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1223e14bb325SJeff Bonwick } 1224e14bb325SJeff Bonwick } 12250a4e9518Sgw 1226e14bb325SJeff Bonwick /* 1227f7170741SWill Andrews * Determine whether this device is accessible. 1228f7170741SWill Andrews * 1229f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1230f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1231f7170741SWill Andrews * a VTOC. 1232e14bb325SJeff Bonwick */ 1233e14bb325SJeff Bonwick zio_t * 1234a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1235e14bb325SJeff Bonwick { 1236e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1237a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1238a3f829aeSBill Moore zio_t *pio; 1239a3f829aeSBill Moore 1240a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 12410a4e9518Sgw 1242a3f829aeSBill Moore /* 1243a3f829aeSBill Moore * Don't probe the probe. 1244a3f829aeSBill Moore */ 1245a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1246a3f829aeSBill Moore return (NULL); 1247e14bb325SJeff Bonwick 1248a3f829aeSBill Moore /* 1249a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1250a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1251a3f829aeSBill Moore * this vdev will become parents of the probe io. 1252a3f829aeSBill Moore */ 1253a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1254e14bb325SJeff Bonwick 1255a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1256a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1257a3f829aeSBill Moore 1258a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1259a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 12608956713aSEric Schrock ZIO_FLAG_TRYHARD; 1261a3f829aeSBill Moore 1262a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1263a3f829aeSBill Moore /* 1264a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1265a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1266a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1267a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1268a3f829aeSBill Moore * any zio looking at these values can assume that 1269a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1270a3f829aeSBill Moore * important because when a device has intermittent 1271a3f829aeSBill Moore * connectivity problems, we want to ensure that 1272a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1273a3f829aeSBill Moore * the zio (EIO). 1274a3f829aeSBill Moore * 1275a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1276a3f829aeSBill Moore * values so the probe can reevaluate from first 1277a3f829aeSBill Moore * principles. 1278a3f829aeSBill Moore */ 1279a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1280a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1281a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1282a3f829aeSBill Moore } 1283a3f829aeSBill Moore 1284a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1285a3f829aeSBill Moore vdev_probe_done, vps, 1286a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1287a3f829aeSBill Moore 128898d1cbfeSGeorge Wilson /* 128998d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 129098d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 129198d1cbfeSGeorge Wilson */ 1292a3f829aeSBill Moore if (zio != NULL) { 1293a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1294a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1295a3f829aeSBill Moore } 1296e14bb325SJeff Bonwick } 1297e14bb325SJeff Bonwick 1298a3f829aeSBill Moore if (zio != NULL) 1299a3f829aeSBill Moore zio_add_child(zio, pio); 1300e14bb325SJeff Bonwick 1301a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1302e14bb325SJeff Bonwick 1303a3f829aeSBill Moore if (vps == NULL) { 1304a3f829aeSBill Moore ASSERT(zio != NULL); 1305a3f829aeSBill Moore return (NULL); 1306a3f829aeSBill Moore } 1307e14bb325SJeff Bonwick 1308e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1309a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1310e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1311770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1312770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1313e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1314e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1315e14bb325SJeff Bonwick } 1316e14bb325SJeff Bonwick 1317a3f829aeSBill Moore if (zio == NULL) 1318a3f829aeSBill Moore return (pio); 1319a3f829aeSBill Moore 1320a3f829aeSBill Moore zio_nowait(pio); 1321a3f829aeSBill Moore return (NULL); 13220a4e9518Sgw } 13230a4e9518Sgw 1324f64c0e34SEric Taylor static void 1325f64c0e34SEric Taylor vdev_open_child(void *arg) 1326f64c0e34SEric Taylor { 1327f64c0e34SEric Taylor vdev_t *vd = arg; 1328f64c0e34SEric Taylor 1329f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1330f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1331f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1332f64c0e34SEric Taylor } 1333f64c0e34SEric Taylor 1334681d9761SEric Taylor boolean_t 1335681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1336681d9761SEric Taylor { 1337681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1338681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1339681d9761SEric Taylor return (B_TRUE); 1340681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1341681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1342681d9761SEric Taylor return (B_TRUE); 1343681d9761SEric Taylor return (B_FALSE); 1344681d9761SEric Taylor } 1345681d9761SEric Taylor 1346f64c0e34SEric Taylor void 1347f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1348f64c0e34SEric Taylor { 1349f64c0e34SEric Taylor taskq_t *tq; 1350f64c0e34SEric Taylor int children = vd->vdev_children; 1351f64c0e34SEric Taylor 1352681d9761SEric Taylor /* 1353681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1354681d9761SEric Taylor * in a single thread so that the same thread holds the 1355681d9761SEric Taylor * spa_namespace_lock 1356681d9761SEric Taylor */ 1357681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1358681d9761SEric Taylor for (int c = 0; c < children; c++) 1359681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1360681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1361681d9761SEric Taylor return; 1362681d9761SEric Taylor } 1363f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1364f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1365f64c0e34SEric Taylor 1366f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1367f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1368fc8ae2ecSToomas Soome TQ_SLEEP) != TASKQID_INVALID); 1369f64c0e34SEric Taylor 1370f64c0e34SEric Taylor taskq_destroy(tq); 1371f64c0e34SEric Taylor } 1372f64c0e34SEric Taylor 13735cabbc6bSPrashanth Sreenivasa /* 13745cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 13755cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 13765cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 13775cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 13785cabbc6bSPrashanth Sreenivasa */ 13795cabbc6bSPrashanth Sreenivasa static void 13805cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 13815cabbc6bSPrashanth Sreenivasa { 13825cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 13835cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 13845cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 13855cabbc6bSPrashanth Sreenivasa } 13865cabbc6bSPrashanth Sreenivasa } 13875cabbc6bSPrashanth Sreenivasa 1388fa9e4066Sahrens /* 1389fa9e4066Sahrens * Prepare a virtual device for access. 1390fa9e4066Sahrens */ 1391fa9e4066Sahrens int 1392fa9e4066Sahrens vdev_open(vdev_t *vd) 1393fa9e4066Sahrens { 13948ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1395fa9e4066Sahrens int error; 1396fa9e4066Sahrens uint64_t osize = 0; 13974263d13fSGeorge Wilson uint64_t max_osize = 0; 13984263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1399ecc2d604Sbonwick uint64_t ashift = 0; 1400fa9e4066Sahrens 1401f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1402f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1403fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1404fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1405fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1406fa9e4066Sahrens 1407fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1408e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1409e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1410573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1411fa9e4066Sahrens 1412069f55e2SEric Schrock /* 1413069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1414069f55e2SEric Schrock * faulted, bail out of the open. 1415069f55e2SEric Schrock */ 14163d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 14173d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1418069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1419069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 14203d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1421069f55e2SEric Schrock vd->vdev_label_aux); 1422be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 14233d7072f8Seschrock } else if (vd->vdev_offline) { 1424fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1425ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1426be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1427fa9e4066Sahrens } 1428fa9e4066Sahrens 14294263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1430fa9e4066Sahrens 1431095bcd66SGeorge Wilson /* 1432095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1433095bcd66SGeorge Wilson * the vdev on error. 1434095bcd66SGeorge Wilson */ 1435095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1436ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 14378956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1438ea8dc4b6Seschrock 1439fa9e4066Sahrens if (error) { 14403d7072f8Seschrock if (vd->vdev_removed && 14413d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 14423d7072f8Seschrock vd->vdev_removed = B_FALSE; 14433d7072f8Seschrock 14446f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 14456f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 14466f793812SPavel Zakharov vd->vdev_stat.vs_aux); 14476f793812SPavel Zakharov } else { 14486f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 14496f793812SPavel Zakharov vd->vdev_stat.vs_aux); 14506f793812SPavel Zakharov } 1451fa9e4066Sahrens return (error); 1452fa9e4066Sahrens } 1453fa9e4066Sahrens 14543d7072f8Seschrock vd->vdev_removed = B_FALSE; 14553d7072f8Seschrock 1456096d22d4SEric Schrock /* 1457096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1458096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1459096d22d4SEric Schrock */ 1460096d22d4SEric Schrock if (vd->vdev_faulted) { 1461096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1462096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1463096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1464096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1465096d22d4SEric Schrock vd->vdev_label_aux); 1466be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1467096d22d4SEric Schrock } 1468096d22d4SEric Schrock 14693d7072f8Seschrock if (vd->vdev_degraded) { 14703d7072f8Seschrock ASSERT(vd->vdev_children == 0); 14713d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 14723d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 14733d7072f8Seschrock } else { 1474069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 14753d7072f8Seschrock } 1476fa9e4066Sahrens 147788ecc943SGeorge Wilson /* 147888ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 147988ecc943SGeorge Wilson */ 148088ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 148188ecc943SGeorge Wilson return (0); 148288ecc943SGeorge Wilson 1483573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1484ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1485ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1486ea8dc4b6Seschrock VDEV_AUX_NONE); 1487ea8dc4b6Seschrock break; 1488ea8dc4b6Seschrock } 1489573ca77eSGeorge Wilson } 1490fa9e4066Sahrens 1491fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 14924263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1493fa9e4066Sahrens 1494fa9e4066Sahrens if (vd->vdev_children == 0) { 1495fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1496ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1497ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1498be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1499fa9e4066Sahrens } 1500fa9e4066Sahrens psize = osize; 1501fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 15024263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 15034263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1504fa9e4066Sahrens } else { 1505ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1506fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1507ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1508ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1509be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1510fa9e4066Sahrens } 1511fa9e4066Sahrens psize = 0; 1512fa9e4066Sahrens asize = osize; 15134263d13fSGeorge Wilson max_asize = max_osize; 1514fa9e4066Sahrens } 1515fa9e4066Sahrens 1516fa9e4066Sahrens vd->vdev_psize = psize; 1517fa9e4066Sahrens 1518573ca77eSGeorge Wilson /* 1519c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1520573ca77eSGeorge Wilson */ 1521573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1522573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1523573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1524be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1525573ca77eSGeorge Wilson } 1526573ca77eSGeorge Wilson 1527fa9e4066Sahrens if (vd->vdev_asize == 0) { 1528fa9e4066Sahrens /* 1529fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1530ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1531fa9e4066Sahrens */ 1532fa9e4066Sahrens vd->vdev_asize = asize; 15334263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1534ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 153593a1902eSMatthew Ahrens vd->vdev_ashift = MAX(zfs_ashift_min, vd->vdev_ashift); 1536fa9e4066Sahrens } else { 1537fa9e4066Sahrens /* 15382384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 15392384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 15402384d9f8SGeorge Wilson * issue a warning instead. 1541fa9e4066Sahrens */ 15422384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 15432384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 15442384d9f8SGeorge Wilson cmn_err(CE_WARN, 15452384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 15462384d9f8SGeorge Wilson "larger than the pool's alignment\n", 15472384d9f8SGeorge Wilson vd->vdev_path); 1548fa9e4066Sahrens } 15494263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1550573ca77eSGeorge Wilson } 1551fa9e4066Sahrens 1552573ca77eSGeorge Wilson /* 1553c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1554c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1555c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1556c040c10cSSteven Hartland * making the additional space available. 1557c040c10cSSteven Hartland * 1558c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1559c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1560c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1561c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1562c040c10cSSteven Hartland * vdev_min_asize. 1563573ca77eSGeorge Wilson */ 1564c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1565c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1566c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1567c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1568573ca77eSGeorge Wilson vd->vdev_asize = asize; 1569fa9e4066Sahrens 1570573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1571fa9e4066Sahrens 15720a4e9518Sgw /* 15730a4e9518Sgw * Ensure we can issue some IO before declaring the 15740a4e9518Sgw * vdev open for business. 15750a4e9518Sgw */ 1576e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1577e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 157898d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 157998d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 15800a4e9518Sgw return (error); 15810a4e9518Sgw } 15820a4e9518Sgw 158381cd5c55SMatthew Ahrens /* 158481cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 158581cd5c55SMatthew Ahrens */ 158681cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 158781cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 158881cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 158981cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 159081cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 159181cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 159281cd5c55SMatthew Ahrens } 159381cd5c55SMatthew Ahrens 1594088f3894Sahrens /* 1595088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 15968ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 15978ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1598088f3894Sahrens */ 15998ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 16008ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 16018ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1602088f3894Sahrens 1603fa9e4066Sahrens return (0); 1604fa9e4066Sahrens } 1605fa9e4066Sahrens 1606560e6e96Seschrock /* 1607560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 16086f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 16093d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1610560e6e96Seschrock * 1611560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1612560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1613560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1614560e6e96Seschrock * will be updated but the function will return 0. 1615560e6e96Seschrock */ 1616560e6e96Seschrock int 16176f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1618560e6e96Seschrock { 1619560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1620560e6e96Seschrock nvlist_t *label; 16216f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1622560e6e96Seschrock uint64_t state; 16236f793812SPavel Zakharov nvlist_t *nvl; 16246f793812SPavel Zakharov uint64_t txg; 1625560e6e96Seschrock 16266f793812SPavel Zakharov if (vdev_validate_skip) 16276f793812SPavel Zakharov return (0); 16286f793812SPavel Zakharov 16296f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 16306f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1631be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1632560e6e96Seschrock 1633b5989ec7Seschrock /* 1634b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1635b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1636b5989ec7Seschrock * overwrite the previous state. 1637b5989ec7Seschrock */ 16386f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 16396f793812SPavel Zakharov return (0); 1640560e6e96Seschrock 16416f793812SPavel Zakharov /* 16426f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 16436f793812SPavel Zakharov * was modified at a point after the current txg. 1644d1de72cfSPavel Zakharov * If config lock is not held do not check for the txg. spa_sync could 1645d1de72cfSPavel Zakharov * be updating the vdev's label before updating spa_last_synced_txg. 16466f793812SPavel Zakharov */ 1647d1de72cfSPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || 1648d1de72cfSPavel Zakharov spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) 16496f793812SPavel Zakharov txg = UINT64_MAX; 16506f793812SPavel Zakharov else 16516f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1652560e6e96Seschrock 16536f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 16546f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 16556f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 1656b6bf6e15SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config for " 1657b6bf6e15SPavel Zakharov "txg %llu", (u_longlong_t)txg); 16586f793812SPavel Zakharov return (0); 16596f793812SPavel Zakharov } 16601195e687SMark J Musante 16616f793812SPavel Zakharov /* 16626f793812SPavel Zakharov * Determine if this vdev has been split off into another 16636f793812SPavel Zakharov * pool. If so, then refuse to open it. 16646f793812SPavel Zakharov */ 16656f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 16666f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 16676f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16686f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 16696f793812SPavel Zakharov nvlist_free(label); 16706f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 16716f793812SPavel Zakharov return (0); 16726f793812SPavel Zakharov } 1673560e6e96Seschrock 16746f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 16756f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16766f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16776f793812SPavel Zakharov nvlist_free(label); 16786f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 16796f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 16806f793812SPavel Zakharov return (0); 16816f793812SPavel Zakharov } 16821195e687SMark J Musante 16836f793812SPavel Zakharov /* 16846f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 16856f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 16866f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 16876f793812SPavel Zakharov * cached config. The check will be performed again once we have the 16886f793812SPavel Zakharov * trusted config from the MOS. 16896f793812SPavel Zakharov */ 16906f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 16916f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16926f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16936f793812SPavel Zakharov nvlist_free(label); 16946f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 16956f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 16966f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 16976f793812SPavel Zakharov return (0); 16986f793812SPavel Zakharov } 16996f793812SPavel Zakharov 17006f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 17016f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 17026f793812SPavel Zakharov &aux_guid) != 0) 17036f793812SPavel Zakharov aux_guid = 0; 17046f793812SPavel Zakharov 17056f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 17066f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 17076f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 17086f793812SPavel Zakharov nvlist_free(label); 17096f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 17106f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 17116f793812SPavel Zakharov return (0); 17126f793812SPavel Zakharov } 17136f793812SPavel Zakharov 17146f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 17156f793812SPavel Zakharov != 0) { 17166f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 17176f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 17186f793812SPavel Zakharov nvlist_free(label); 17196f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 17206f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 17216f793812SPavel Zakharov return (0); 17226f793812SPavel Zakharov } 17236f793812SPavel Zakharov 17246f793812SPavel Zakharov /* 17256f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 17266f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 17276f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 17286f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 17296f793812SPavel Zakharov * vdev, we can safely compare to that instead. 17306f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 17316f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 17326f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 17336f793812SPavel Zakharov * extreme rewind. 17346f793812SPavel Zakharov * 17356f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 17366f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 17376f793812SPavel Zakharov * corrupt if it is partway through a split operation. 17386f793812SPavel Zakharov */ 17396f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 17406f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 17416f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 17426f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 17436f793812SPavel Zakharov mismatch = B_TRUE; 17446f793812SPavel Zakharov } else { 17456f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 17466f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 17476f793812SPavel Zakharov mismatch = B_TRUE; 1748560e6e96Seschrock } 1749560e6e96Seschrock 17506f793812SPavel Zakharov if (mismatch) { 1751560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1752560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1753560e6e96Seschrock nvlist_free(label); 17546f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 17556f793812SPavel Zakharov "doesn't match label guid"); 17566f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 17576f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 17586f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 17596f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 17606f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 17616f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1762560e6e96Seschrock return (0); 1763560e6e96Seschrock } 17646f793812SPavel Zakharov } 1765560e6e96Seschrock 17666f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 17676f793812SPavel Zakharov &state) != 0) { 17686f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 17696f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1770560e6e96Seschrock nvlist_free(label); 17716f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 17726f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 17736f793812SPavel Zakharov return (0); 17746f793812SPavel Zakharov } 1775560e6e96Seschrock 17766f793812SPavel Zakharov nvlist_free(label); 17776f793812SPavel Zakharov 17786f793812SPavel Zakharov /* 17796f793812SPavel Zakharov * If this is a verbatim import, no need to check the 17806f793812SPavel Zakharov * state of the pool. 17816f793812SPavel Zakharov */ 17826f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 17836f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 17846f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 17856f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 17866f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 17876f793812SPavel Zakharov return (SET_ERROR(EBADF)); 17886f793812SPavel Zakharov } 17896f793812SPavel Zakharov 17906f793812SPavel Zakharov /* 17916f793812SPavel Zakharov * If we were able to open and validate a vdev that was 17926f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 17936f793812SPavel Zakharov * now. 17946f793812SPavel Zakharov */ 17956f793812SPavel Zakharov if (vd->vdev_not_present) 17966f793812SPavel Zakharov vd->vdev_not_present = 0; 17976f793812SPavel Zakharov 17986f793812SPavel Zakharov return (0); 17996f793812SPavel Zakharov } 18006f793812SPavel Zakharov 18016f793812SPavel Zakharov static void 18026f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 18036f793812SPavel Zakharov { 18046f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 18056f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 18066f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 18076f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 18086f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 18096f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 18106f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 18113ee8c80cSPavel Zakharov } 18126f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 18136f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 18146f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 18156f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 18166f793812SPavel Zakharov } 18176f793812SPavel Zakharov } 1818560e6e96Seschrock 18196f793812SPavel Zakharov /* 18206f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 18216f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 18226f793812SPavel Zakharov * paths from userland config into MOS config. 18236f793812SPavel Zakharov */ 18246f793812SPavel Zakharov int 18256f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 18266f793812SPavel Zakharov { 18276f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 18286f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 18296f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 18306f793812SPavel Zakharov return (0); 18316f793812SPavel Zakharov 18326f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 18336f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 18346f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 18356f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 18366f793812SPavel Zakharov } 18376f793812SPavel Zakharov 18386f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 18396f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 18406f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 18416f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 18426f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 18436f793812SPavel Zakharov } 18446f793812SPavel Zakharov 18456f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 18466f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 18476f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 18486f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 18496f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 185051ece835Seschrock } 1851560e6e96Seschrock 18526f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 18536f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 18546f793812SPavel Zakharov dvd->vdev_child[i]); 18556f793812SPavel Zakharov if (error != 0) 18566f793812SPavel Zakharov return (error); 18576f793812SPavel Zakharov } 18586f793812SPavel Zakharov 18596f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 18606f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 18616f793812SPavel Zakharov 1862560e6e96Seschrock return (0); 1863560e6e96Seschrock } 1864560e6e96Seschrock 18656f793812SPavel Zakharov static void 18666f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 18676f793812SPavel Zakharov { 18686f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 18696f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 18706f793812SPavel Zakharov 18716f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 18726f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 18736f793812SPavel Zakharov } 18746f793812SPavel Zakharov 18756f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 18766f793812SPavel Zakharov return; 18776f793812SPavel Zakharov 18786f793812SPavel Zakharov /* 18796f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 18806f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 18816f793812SPavel Zakharov * step outside of it. 18826f793812SPavel Zakharov */ 18836f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 18846f793812SPavel Zakharov 18856f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 18866f793812SPavel Zakharov return; 18876f793812SPavel Zakharov 18886f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 18896f793812SPavel Zakharov 18906f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 18916f793812SPavel Zakharov } 18926f793812SPavel Zakharov 18936f793812SPavel Zakharov /* 18946f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 18956f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 18966f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 18976f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 18986f793812SPavel Zakharov */ 18996f793812SPavel Zakharov void 19006f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 19016f793812SPavel Zakharov { 19026f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 19036f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 19046f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 19056f793812SPavel Zakharov 19066f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 19076f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 19086f793812SPavel Zakharov drvd->vdev_child[i]); 19096f793812SPavel Zakharov } 19106f793812SPavel Zakharov } 19116f793812SPavel Zakharov 1912fa9e4066Sahrens /* 1913fa9e4066Sahrens * Close a virtual device. 1914fa9e4066Sahrens */ 1915fa9e4066Sahrens void 1916fa9e4066Sahrens vdev_close(vdev_t *vd) 1917fa9e4066Sahrens { 19188ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1919095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 19208ad4d6ddSJeff Bonwick 19218ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 19228ad4d6ddSJeff Bonwick 19231195e687SMark J Musante /* 19241195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 19251195e687SMark J Musante * going offline. 19261195e687SMark J Musante */ 1927095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 19281195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1929095bcd66SGeorge Wilson 1930fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1931fa9e4066Sahrens 19323d7072f8Seschrock vdev_cache_purge(vd); 1933fa9e4066Sahrens 1934560e6e96Seschrock /* 1935573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1936560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1937560e6e96Seschrock * it's still faulted. 1938560e6e96Seschrock */ 1939560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1940560e6e96Seschrock 1941fa9e4066Sahrens if (vd->vdev_offline) 1942fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1943fa9e4066Sahrens else 1944fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1945ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1946fa9e4066Sahrens } 1947fa9e4066Sahrens 1948dcba9f3fSGeorge Wilson void 1949dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1950dcba9f3fSGeorge Wilson { 1951dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1952dcba9f3fSGeorge Wilson 1953dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1954dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1955dcba9f3fSGeorge Wilson return; 1956dcba9f3fSGeorge Wilson 1957dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1958dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1959dcba9f3fSGeorge Wilson 1960dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1961dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1962dcba9f3fSGeorge Wilson } 1963dcba9f3fSGeorge Wilson 1964dcba9f3fSGeorge Wilson void 1965dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1966dcba9f3fSGeorge Wilson { 1967dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1968dcba9f3fSGeorge Wilson 1969dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1970dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1971dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1972dcba9f3fSGeorge Wilson 1973dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1974dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1975dcba9f3fSGeorge Wilson } 1976dcba9f3fSGeorge Wilson 1977095bcd66SGeorge Wilson /* 1978095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1979095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1980095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1981095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1982095bcd66SGeorge Wilson */ 1983fa9e4066Sahrens void 1984ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1985fa9e4066Sahrens { 1986ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1987fa9e4066Sahrens 1988e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1989ea8dc4b6Seschrock 19901195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 19911195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1992fa9e4066Sahrens vdev_close(vd); 1993fa9e4066Sahrens (void) vdev_open(vd); 1994fa9e4066Sahrens 199539c23413Seschrock /* 199639c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 199739c23413Seschrock * Otherwise, a device with an invalid label could be successfully 199839c23413Seschrock * opened in response to vdev_reopen(). 199939c23413Seschrock */ 2000c5904d13Seschrock if (vd->vdev_aux) { 2001c5904d13Seschrock (void) vdev_validate_aux(vd); 2002e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 20036809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 2004573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 2005573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 2006c5904d13Seschrock } else { 20076f793812SPavel Zakharov (void) vdev_validate(vd); 2008c5904d13Seschrock } 200939c23413Seschrock 2010fa9e4066Sahrens /* 20113d7072f8Seschrock * Reassess parent vdev's health. 2012fa9e4066Sahrens */ 20133d7072f8Seschrock vdev_propagate_state(vd); 2014fa9e4066Sahrens } 2015fa9e4066Sahrens 2016fa9e4066Sahrens int 201799653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 2018fa9e4066Sahrens { 2019fa9e4066Sahrens int error; 2020fa9e4066Sahrens 2021fa9e4066Sahrens /* 2022fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 2023fa9e4066Sahrens * For a create, however, we want to fail the request if 2024fa9e4066Sahrens * there are any components we can't open. 2025fa9e4066Sahrens */ 2026fa9e4066Sahrens error = vdev_open(vd); 2027fa9e4066Sahrens 2028fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 2029fa9e4066Sahrens vdev_close(vd); 2030fa9e4066Sahrens return (error ? error : ENXIO); 2031fa9e4066Sahrens } 2032fa9e4066Sahrens 2033fa9e4066Sahrens /* 20340713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 2035fa9e4066Sahrens */ 20360713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 20370713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 203839c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 2039fa9e4066Sahrens vdev_close(vd); 2040fa9e4066Sahrens return (error); 2041fa9e4066Sahrens } 2042fa9e4066Sahrens 2043fa9e4066Sahrens return (0); 2044fa9e4066Sahrens } 2045fa9e4066Sahrens 20460e34b6a7Sbonwick void 2047573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 2048fa9e4066Sahrens { 204986714001SSerapheim Dimitropoulos uint64_t asize = vd->vdev_asize; 2050a0b03b16SSerapheim Dimitropoulos uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; 2051b4bf0cf0SDon Brady uint64_t ms_shift; 205286714001SSerapheim Dimitropoulos 2053fa9e4066Sahrens /* 2054b4bf0cf0SDon Brady * There are two dimensions to the metaslab sizing calculation: 2055b4bf0cf0SDon Brady * the size of the metaslab and the count of metaslabs per vdev. 2056b4bf0cf0SDon Brady * 2057a0b03b16SSerapheim Dimitropoulos * The default values used below are a good balance between memory 2058a0b03b16SSerapheim Dimitropoulos * usage (larger metaslab size means more memory needed for loaded 2059a0b03b16SSerapheim Dimitropoulos * metaslabs; more metaslabs means more memory needed for the 2060a0b03b16SSerapheim Dimitropoulos * metaslab_t structs), metaslab load time (larger metaslabs take 2061a0b03b16SSerapheim Dimitropoulos * longer to load), and metaslab sync time (more metaslabs means 2062a0b03b16SSerapheim Dimitropoulos * more time spent syncing all of them). 2063a0b03b16SSerapheim Dimitropoulos * 2064a0b03b16SSerapheim Dimitropoulos * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. 2065a0b03b16SSerapheim Dimitropoulos * The range of the dimensions are as follows: 2066a0b03b16SSerapheim Dimitropoulos * 2067a0b03b16SSerapheim Dimitropoulos * 2^29 <= ms_size <= 2^34 2068b4bf0cf0SDon Brady * 16 <= ms_count <= 131,072 2069b4bf0cf0SDon Brady * 2070b4bf0cf0SDon Brady * On the lower end of vdev sizes, we aim for metaslabs sizes of 2071b4bf0cf0SDon Brady * at least 512MB (2^29) to minimize fragmentation effects when 2072b4bf0cf0SDon Brady * testing with smaller devices. However, the count constraint 2073b4bf0cf0SDon Brady * of at least 16 metaslabs will override this minimum size goal. 2074b4bf0cf0SDon Brady * 2075b4bf0cf0SDon Brady * On the upper end of vdev sizes, we aim for a maximum metaslab 2076a0b03b16SSerapheim Dimitropoulos * size of 16GB. However, we will cap the total count to 2^17 2077a0b03b16SSerapheim Dimitropoulos * metaslabs to keep our memory footprint in check and let the 2078a0b03b16SSerapheim Dimitropoulos * metaslab size grow from there if that limit is hit. 2079b4bf0cf0SDon Brady * 2080b4bf0cf0SDon Brady * The net effect of applying above constrains is summarized below. 2081b4bf0cf0SDon Brady * 2082a0b03b16SSerapheim Dimitropoulos * vdev size metaslab count 2083a0b03b16SSerapheim Dimitropoulos * --------------|----------------- 2084a0b03b16SSerapheim Dimitropoulos * < 8GB ~16 2085a0b03b16SSerapheim Dimitropoulos * 8GB - 100GB one per 512MB 2086a0b03b16SSerapheim Dimitropoulos * 100GB - 3TB ~200 2087a0b03b16SSerapheim Dimitropoulos * 3TB - 2PB one per 16GB 2088a0b03b16SSerapheim Dimitropoulos * > 2PB ~131,072 2089a0b03b16SSerapheim Dimitropoulos * -------------------------------- 2090a0b03b16SSerapheim Dimitropoulos * 2091a0b03b16SSerapheim Dimitropoulos * Finally, note that all of the above calculate the initial 2092a0b03b16SSerapheim Dimitropoulos * number of metaslabs. Expanding a top-level vdev will result 2093a0b03b16SSerapheim Dimitropoulos * in additional metaslabs being allocated making it possible 2094a0b03b16SSerapheim Dimitropoulos * to exceed the zfs_vdev_ms_count_limit. 2095a0b03b16SSerapheim Dimitropoulos */ 2096a0b03b16SSerapheim Dimitropoulos 2097a0b03b16SSerapheim Dimitropoulos if (ms_count < zfs_vdev_min_ms_count) 2098a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_min_ms_count); 2099a0b03b16SSerapheim Dimitropoulos else if (ms_count > zfs_vdev_default_ms_count) 2100a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_default_ms_count); 2101b4bf0cf0SDon Brady else 2102a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_default_ms_shift; 2103b4bf0cf0SDon Brady 2104b4bf0cf0SDon Brady if (ms_shift < SPA_MAXBLOCKSHIFT) { 2105b4bf0cf0SDon Brady ms_shift = SPA_MAXBLOCKSHIFT; 2106a0b03b16SSerapheim Dimitropoulos } else if (ms_shift > zfs_vdev_max_ms_shift) { 2107a0b03b16SSerapheim Dimitropoulos ms_shift = zfs_vdev_max_ms_shift; 2108b4bf0cf0SDon Brady /* cap the total count to constrain memory footprint */ 2109a0b03b16SSerapheim Dimitropoulos if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) 2110a0b03b16SSerapheim Dimitropoulos ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); 211186714001SSerapheim Dimitropoulos } 211286714001SSerapheim Dimitropoulos 211386714001SSerapheim Dimitropoulos vd->vdev_ms_shift = ms_shift; 211486714001SSerapheim Dimitropoulos ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2115fa9e4066Sahrens } 2116fa9e4066Sahrens 2117fa9e4066Sahrens void 2118ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2119fa9e4066Sahrens { 2120ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 21215cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 21225cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 2123ecc2d604Sbonwick ASSERT(ISP2(flags)); 2124f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 2125fa9e4066Sahrens 2126ecc2d604Sbonwick if (flags & VDD_METASLAB) 2127ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2128ecc2d604Sbonwick 2129ecc2d604Sbonwick if (flags & VDD_DTL) 2130ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2131ecc2d604Sbonwick 2132ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2133fa9e4066Sahrens } 2134fa9e4066Sahrens 21350713e232SGeorge Wilson void 21360713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 21370713e232SGeorge Wilson { 21380713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 21390713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 21400713e232SGeorge Wilson 21410713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 21420713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 21430713e232SGeorge Wilson } 21440713e232SGeorge Wilson 21458ad4d6ddSJeff Bonwick /* 21468ad4d6ddSJeff Bonwick * DTLs. 21478ad4d6ddSJeff Bonwick * 21488ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 21499fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 21508ad4d6ddSJeff Bonwick * 21518ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 21528ad4d6ddSJeff Bonwick * 21538ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 21548ad4d6ddSJeff Bonwick * 21558ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 21568ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 21578ad4d6ddSJeff Bonwick * txgs that was scrubbed. 21588ad4d6ddSJeff Bonwick * 21598ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 21608ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 21618ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 21628ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 21638ad4d6ddSJeff Bonwick * determine whether a device can be detached. 21648ad4d6ddSJeff Bonwick * 21658ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 21668ad4d6ddSJeff Bonwick * either has the data or it doesn't. 21678ad4d6ddSJeff Bonwick * 21688ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 21698ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 21708ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 21718ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 21728ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 21738ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 21748ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 21758ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 21768ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 21778ad4d6ddSJeff Bonwick * 21788ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 21798ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 21808ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 21818ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 21828ad4d6ddSJeff Bonwick */ 2183fa9e4066Sahrens void 21848ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2185fa9e4066Sahrens { 21860713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 21878ad4d6ddSJeff Bonwick 21888ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 21898ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2190f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 21918ad4d6ddSJeff Bonwick 21925cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 21930713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 21940713e232SGeorge Wilson range_tree_add(rt, txg, size); 21955cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2196fa9e4066Sahrens } 2197fa9e4066Sahrens 21988ad4d6ddSJeff Bonwick boolean_t 21998ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2200fa9e4066Sahrens { 22010713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 22028ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2203fa9e4066Sahrens 22048ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 22058ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2206fa9e4066Sahrens 22075cabbc6bSPrashanth Sreenivasa /* 22085cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 22095cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 22105cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 22115cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 22125cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 22135cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 22145cabbc6bSPrashanth Sreenivasa */ 22155cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 22165cabbc6bSPrashanth Sreenivasa return (B_FALSE); 22175cabbc6bSPrashanth Sreenivasa 22185cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 221986714001SSerapheim Dimitropoulos if (!range_tree_is_empty(rt)) 22200713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 22215cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2222fa9e4066Sahrens 2223fa9e4066Sahrens return (dirty); 2224fa9e4066Sahrens } 2225fa9e4066Sahrens 22268ad4d6ddSJeff Bonwick boolean_t 22278ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 22288ad4d6ddSJeff Bonwick { 22290713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 22308ad4d6ddSJeff Bonwick boolean_t empty; 22318ad4d6ddSJeff Bonwick 22325cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 223386714001SSerapheim Dimitropoulos empty = range_tree_is_empty(rt); 22345cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 22358ad4d6ddSJeff Bonwick 22368ad4d6ddSJeff Bonwick return (empty); 22378ad4d6ddSJeff Bonwick } 22388ad4d6ddSJeff Bonwick 2239b4952e17SGeorge Wilson /* 2240b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2241b4952e17SGeorge Wilson */ 2242b4952e17SGeorge Wilson static uint64_t 2243b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2244b4952e17SGeorge Wilson { 22450713e232SGeorge Wilson range_seg_t *rs; 2246b4952e17SGeorge Wilson 2247b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 22480713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2249b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2250b4952e17SGeorge Wilson 22510713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 22520713e232SGeorge Wilson return (rs->rs_start - 1); 2253b4952e17SGeorge Wilson } 2254b4952e17SGeorge Wilson 2255b4952e17SGeorge Wilson /* 2256b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2257b4952e17SGeorge Wilson */ 2258b4952e17SGeorge Wilson static uint64_t 2259b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 2260b4952e17SGeorge Wilson { 22610713e232SGeorge Wilson range_seg_t *rs; 2262b4952e17SGeorge Wilson 2263b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 22640713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2265b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2266b4952e17SGeorge Wilson 22670713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 22680713e232SGeorge Wilson return (rs->rs_end); 2269b4952e17SGeorge Wilson } 2270b4952e17SGeorge Wilson 2271b4952e17SGeorge Wilson /* 2272b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2273b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2274b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2275b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2276b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2277b4952e17SGeorge Wilson * excise the DTLs. 2278b4952e17SGeorge Wilson */ 2279b4952e17SGeorge Wilson static boolean_t 2280b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2281b4952e17SGeorge Wilson { 2282b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2283b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2284b4952e17SGeorge Wilson 2285b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 2286b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2287b4952e17SGeorge Wilson 22882d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 22892d2f193aSMatthew Ahrens return (B_FALSE); 22902d2f193aSMatthew Ahrens 2291b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 229286714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2293b4952e17SGeorge Wilson return (B_TRUE); 2294b4952e17SGeorge Wilson 2295b4952e17SGeorge Wilson /* 2296b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2297b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2298b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2299b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2300b4952e17SGeorge Wilson * for excision. 2301b4952e17SGeorge Wilson */ 2302b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2303b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2304b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2305b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2306b4952e17SGeorge Wilson return (B_TRUE); 2307b4952e17SGeorge Wilson } 2308b4952e17SGeorge Wilson return (B_FALSE); 2309b4952e17SGeorge Wilson } 2310b4952e17SGeorge Wilson 2311fa9e4066Sahrens /* 2312fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2313fa9e4066Sahrens */ 2314fa9e4066Sahrens void 2315fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2316fa9e4066Sahrens { 2317ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 23188ad4d6ddSJeff Bonwick avl_tree_t reftree; 23198ad4d6ddSJeff Bonwick int minref; 2320fa9e4066Sahrens 23218ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2322fa9e4066Sahrens 23238ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 23248ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 23258ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 23268ad4d6ddSJeff Bonwick 23275cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 23288ad4d6ddSJeff Bonwick return; 23298ad4d6ddSJeff Bonwick 23308ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 23313f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 23323f9d6ad7SLin Ling 2333fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2334b4952e17SGeorge Wilson 2335b4952e17SGeorge Wilson /* 2336b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2337b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2338b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2339b4952e17SGeorge Wilson * the entire duration of this scan. 2340b4952e17SGeorge Wilson */ 2341088f3894Sahrens if (scrub_txg != 0 && 23423f9d6ad7SLin Ling (spa->spa_scrub_started || 2343b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2344b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2345088f3894Sahrens /* 2346088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2347088f3894Sahrens * did it without rebooting, then the scrub dtl 2348088f3894Sahrens * will be valid, so excise the old region and 2349088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2350088f3894Sahrens * dtl as-is if there was an error. 23518ad4d6ddSJeff Bonwick * 23528ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 23538ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 23548ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 23558ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 23568ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 23578ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 23588ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 23598ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 23608ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2361088f3894Sahrens */ 23620713e232SGeorge Wilson space_reftree_create(&reftree); 23630713e232SGeorge Wilson space_reftree_add_map(&reftree, 23640713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 23650713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 23660713e232SGeorge Wilson space_reftree_add_map(&reftree, 23670713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 23680713e232SGeorge Wilson space_reftree_generate_map(&reftree, 23690713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 23700713e232SGeorge Wilson space_reftree_destroy(&reftree); 2371fa9e4066Sahrens } 23720713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 23730713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 23740713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2375fa9e4066Sahrens if (scrub_done) 23760713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 23770713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 23788ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 23790713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 23808ad4d6ddSJeff Bonwick else 23810713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 23820713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2383b4952e17SGeorge Wilson 2384b4952e17SGeorge Wilson /* 2385b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2386b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2387b4952e17SGeorge Wilson */ 2388b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 238986714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 239086714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2391b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2392b4952e17SGeorge Wilson 2393fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2394088f3894Sahrens 2395ecc2d604Sbonwick if (txg != 0) 2396ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2397fa9e4066Sahrens return; 2398fa9e4066Sahrens } 2399fa9e4066Sahrens 2400fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 24018ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 240299bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 240399bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 24048ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 24058ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 24068ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 24078ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 24088ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 24098ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 24108ad4d6ddSJeff Bonwick else 24118ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 24120713e232SGeorge Wilson space_reftree_create(&reftree); 24138ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 24148ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 24158ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 24160713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 24178ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 24188ad4d6ddSJeff Bonwick } 24190713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 24200713e232SGeorge Wilson space_reftree_destroy(&reftree); 2421fa9e4066Sahrens } 24228ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2423fa9e4066Sahrens } 2424fa9e4066Sahrens 24250713e232SGeorge Wilson int 2426fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2427fa9e4066Sahrens { 2428fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2429ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 24300713e232SGeorge Wilson int error = 0; 2431fa9e4066Sahrens 24320713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 24335cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2434fa9e4066Sahrens 24350713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 24365cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 24370713e232SGeorge Wilson if (error) 24380713e232SGeorge Wilson return (error); 24390713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2440fa9e4066Sahrens 24410713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 244288ecc943SGeorge Wilson 24430713e232SGeorge Wilson /* 24440713e232SGeorge Wilson * Now that we've opened the space_map we need to update 24450713e232SGeorge Wilson * the in-core DTL. 24460713e232SGeorge Wilson */ 24470713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 2448ecc2d604Sbonwick 24490713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 24500713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 24510713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2452fa9e4066Sahrens 24530713e232SGeorge Wilson return (error); 24540713e232SGeorge Wilson } 24550713e232SGeorge Wilson 24560713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 24570713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 24580713e232SGeorge Wilson if (error != 0) 24590713e232SGeorge Wilson break; 24600713e232SGeorge Wilson } 2461fa9e4066Sahrens 2462fa9e4066Sahrens return (error); 2463fa9e4066Sahrens } 2464fa9e4066Sahrens 2465215198a6SJoe Stein void 2466215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2467215198a6SJoe Stein { 2468215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2469215198a6SJoe Stein 2470215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2471215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2472215198a6SJoe Stein zapobj, tx)); 2473215198a6SJoe Stein } 2474215198a6SJoe Stein 2475215198a6SJoe Stein uint64_t 2476215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2477215198a6SJoe Stein { 2478215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2479215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2480215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2481215198a6SJoe Stein 2482215198a6SJoe Stein ASSERT(zap != 0); 2483215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2484215198a6SJoe Stein zap, tx)); 2485215198a6SJoe Stein 2486215198a6SJoe Stein return (zap); 2487215198a6SJoe Stein } 2488215198a6SJoe Stein 2489215198a6SJoe Stein void 2490215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2491215198a6SJoe Stein { 2492215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2493215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2494215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2495215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2496215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2497215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2498215198a6SJoe Stein } 2499215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2500215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2501215198a6SJoe Stein } 2502215198a6SJoe Stein } 2503215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2504215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2505215198a6SJoe Stein } 2506215198a6SJoe Stein } 2507215198a6SJoe Stein 2508fa9e4066Sahrens void 2509fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2510fa9e4066Sahrens { 2511fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 25120713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2513ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 25140713e232SGeorge Wilson range_tree_t *rtsync; 2515fa9e4066Sahrens dmu_tx_t *tx; 25160713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2517fa9e4066Sahrens 25185cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 25190713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 252088ecc943SGeorge Wilson 2521fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2522fa9e4066Sahrens 25230713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 25240713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 25250713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 25260713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 25270713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 25280713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2529215198a6SJoe Stein 2530215198a6SJoe Stein /* 2531215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2532215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2533215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2534215198a6SJoe Stein */ 2535215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2536215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2537215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2538215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2539215198a6SJoe Stein } 2540215198a6SJoe Stein 2541fa9e4066Sahrens dmu_tx_commit(tx); 2542fa9e4066Sahrens return; 2543fa9e4066Sahrens } 2544fa9e4066Sahrens 25450713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 25460713e232SGeorge Wilson uint64_t new_object; 25470713e232SGeorge Wilson 254886714001SSerapheim Dimitropoulos new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); 25490713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 25500713e232SGeorge Wilson 25510713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 25525cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 25530713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2554fa9e4066Sahrens } 2555fa9e4066Sahrens 25565cabbc6bSPrashanth Sreenivasa rtsync = range_tree_create(NULL, NULL); 2557fa9e4066Sahrens 2558fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 25590713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2560fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2561fa9e4066Sahrens 256286714001SSerapheim Dimitropoulos space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); 256317f11284SSerapheim Dimitropoulos space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); 25640713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2565fa9e4066Sahrens 25660713e232SGeorge Wilson range_tree_destroy(rtsync); 2567fa9e4066Sahrens 25680713e232SGeorge Wilson /* 25690713e232SGeorge Wilson * If the object for the space map has changed then dirty 25700713e232SGeorge Wilson * the top level so that we update the config. 25710713e232SGeorge Wilson */ 25720713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 25733ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 25743ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 25753ee8c80cSPavel Zakharov (u_longlong_t)object, 25763ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 25770713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 25780713e232SGeorge Wilson } 2579fa9e4066Sahrens 2580fa9e4066Sahrens dmu_tx_commit(tx); 25810713e232SGeorge Wilson 25820713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 25830713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 25840713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2585fa9e4066Sahrens } 2586fa9e4066Sahrens 25878ad4d6ddSJeff Bonwick /* 25888ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 25898ad4d6ddSJeff Bonwick * without losing data. 25908ad4d6ddSJeff Bonwick */ 25918ad4d6ddSJeff Bonwick boolean_t 25928ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 25938ad4d6ddSJeff Bonwick { 25948ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 25958ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 25968ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 25978ad4d6ddSJeff Bonwick boolean_t required; 25988ad4d6ddSJeff Bonwick 25998ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 26008ad4d6ddSJeff Bonwick 26018ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 26028ad4d6ddSJeff Bonwick return (B_TRUE); 26038ad4d6ddSJeff Bonwick 26048ad4d6ddSJeff Bonwick /* 26058ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 26068ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 26078ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 26088ad4d6ddSJeff Bonwick */ 26098ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 26108ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 26118ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 26128ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 26138ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 26148ad4d6ddSJeff Bonwick 2615cb04b873SMark J Musante if (!required && zio_injection_enabled) 2616cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2617cb04b873SMark J Musante 26188ad4d6ddSJeff Bonwick return (required); 26198ad4d6ddSJeff Bonwick } 26208ad4d6ddSJeff Bonwick 2621088f3894Sahrens /* 2622088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2623088f3894Sahrens */ 2624088f3894Sahrens boolean_t 2625088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2626088f3894Sahrens { 2627088f3894Sahrens boolean_t needed = B_FALSE; 2628088f3894Sahrens uint64_t thismin = UINT64_MAX; 2629088f3894Sahrens uint64_t thismax = 0; 2630088f3894Sahrens 2631088f3894Sahrens if (vd->vdev_children == 0) { 2632088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 263386714001SSerapheim Dimitropoulos if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 26348ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2635088f3894Sahrens 2636b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2637b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2638088f3894Sahrens needed = B_TRUE; 2639088f3894Sahrens } 2640088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2641088f3894Sahrens } else { 26428ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2643088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2644088f3894Sahrens uint64_t cmin, cmax; 2645088f3894Sahrens 2646088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2647088f3894Sahrens thismin = MIN(thismin, cmin); 2648088f3894Sahrens thismax = MAX(thismax, cmax); 2649088f3894Sahrens needed = B_TRUE; 2650088f3894Sahrens } 2651088f3894Sahrens } 2652088f3894Sahrens } 2653088f3894Sahrens 2654088f3894Sahrens if (needed && minp) { 2655088f3894Sahrens *minp = thismin; 2656088f3894Sahrens *maxp = thismax; 2657088f3894Sahrens } 2658088f3894Sahrens return (needed); 2659088f3894Sahrens } 2660088f3894Sahrens 266186714001SSerapheim Dimitropoulos /* 266286714001SSerapheim Dimitropoulos * Gets the checkpoint space map object from the vdev's ZAP. 266386714001SSerapheim Dimitropoulos * Returns the spacemap object, or 0 if it wasn't in the ZAP 266486714001SSerapheim Dimitropoulos * or the ZAP doesn't exist yet. 266586714001SSerapheim Dimitropoulos */ 266686714001SSerapheim Dimitropoulos int 266786714001SSerapheim Dimitropoulos vdev_checkpoint_sm_object(vdev_t *vd) 266886714001SSerapheim Dimitropoulos { 266986714001SSerapheim Dimitropoulos ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 267086714001SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) { 267186714001SSerapheim Dimitropoulos return (0); 267286714001SSerapheim Dimitropoulos } 267386714001SSerapheim Dimitropoulos 267486714001SSerapheim Dimitropoulos uint64_t sm_obj = 0; 267586714001SSerapheim Dimitropoulos int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 267686714001SSerapheim Dimitropoulos VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 267786714001SSerapheim Dimitropoulos 267886714001SSerapheim Dimitropoulos ASSERT(err == 0 || err == ENOENT); 267986714001SSerapheim Dimitropoulos 268086714001SSerapheim Dimitropoulos return (sm_obj); 268186714001SSerapheim Dimitropoulos } 268286714001SSerapheim Dimitropoulos 26835cabbc6bSPrashanth Sreenivasa int 2684ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2685fa9e4066Sahrens { 26865cabbc6bSPrashanth Sreenivasa int error = 0; 2687fa9e4066Sahrens /* 2688fa9e4066Sahrens * Recursively load all children. 2689fa9e4066Sahrens */ 26905cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 26915cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 26925cabbc6bSPrashanth Sreenivasa if (error != 0) { 26935cabbc6bSPrashanth Sreenivasa return (error); 26945cabbc6bSPrashanth Sreenivasa } 26955cabbc6bSPrashanth Sreenivasa } 26965cabbc6bSPrashanth Sreenivasa 26975cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2698fa9e4066Sahrens 2699fa9e4066Sahrens /* 27000e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2701fa9e4066Sahrens */ 27025cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 27035cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 27045cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 27055cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 27063ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 27073ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 27083ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 27095cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 27105cabbc6bSPrashanth Sreenivasa } else if ((error = vdev_metaslab_init(vd, 0)) != 0) { 27113ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 27123ee8c80cSPavel Zakharov "[error=%d]", error); 27135cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 27145cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 27155cabbc6bSPrashanth Sreenivasa return (error); 27165cabbc6bSPrashanth Sreenivasa } 271786714001SSerapheim Dimitropoulos 271886714001SSerapheim Dimitropoulos uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 271986714001SSerapheim Dimitropoulos if (checkpoint_sm_obj != 0) { 272086714001SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 272186714001SSerapheim Dimitropoulos ASSERT(vd->vdev_asize != 0); 272286714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 272386714001SSerapheim Dimitropoulos 272486714001SSerapheim Dimitropoulos if ((error = space_map_open(&vd->vdev_checkpoint_sm, 272586714001SSerapheim Dimitropoulos mos, checkpoint_sm_obj, 0, vd->vdev_asize, 272686714001SSerapheim Dimitropoulos vd->vdev_ashift))) { 272786714001SSerapheim Dimitropoulos vdev_dbgmsg(vd, "vdev_load: space_map_open " 272886714001SSerapheim Dimitropoulos "failed for checkpoint spacemap (obj %llu) " 272986714001SSerapheim Dimitropoulos "[error=%d]", 273086714001SSerapheim Dimitropoulos (u_longlong_t)checkpoint_sm_obj, error); 273186714001SSerapheim Dimitropoulos return (error); 273286714001SSerapheim Dimitropoulos } 273386714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 273486714001SSerapheim Dimitropoulos space_map_update(vd->vdev_checkpoint_sm); 273586714001SSerapheim Dimitropoulos 273686714001SSerapheim Dimitropoulos /* 273786714001SSerapheim Dimitropoulos * Since the checkpoint_sm contains free entries 273886714001SSerapheim Dimitropoulos * exclusively we can use sm_alloc to indicate the 273986714001SSerapheim Dimitropoulos * culmulative checkpointed space that has been freed. 274086714001SSerapheim Dimitropoulos */ 274186714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space = 274286714001SSerapheim Dimitropoulos -vd->vdev_checkpoint_sm->sm_alloc; 274386714001SSerapheim Dimitropoulos vd->vdev_spa->spa_checkpoint_info.sci_dspace += 274486714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space; 274586714001SSerapheim Dimitropoulos } 27465cabbc6bSPrashanth Sreenivasa } 2747fa9e4066Sahrens 2748fa9e4066Sahrens /* 2749fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2750fa9e4066Sahrens */ 27515cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2752560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2753560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 27543ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 27553ee8c80cSPavel Zakharov "[error=%d]", error); 27565cabbc6bSPrashanth Sreenivasa return (error); 27575cabbc6bSPrashanth Sreenivasa } 27585cabbc6bSPrashanth Sreenivasa 27595cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 27605cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 27615cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 27625cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 276386714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 27645cabbc6bSPrashanth Sreenivasa 27655cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 27665cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 27675cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 27685cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 27693ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 27703ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 27713ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 27725cabbc6bSPrashanth Sreenivasa return (error); 27735cabbc6bSPrashanth Sreenivasa } 27745cabbc6bSPrashanth Sreenivasa space_map_update(vd->vdev_obsolete_sm); 27755cabbc6bSPrashanth Sreenivasa } 27765cabbc6bSPrashanth Sreenivasa 27775cabbc6bSPrashanth Sreenivasa return (0); 2778fa9e4066Sahrens } 2779fa9e4066Sahrens 278099653d4eSeschrock /* 2781fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2782fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2783fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2784fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2785fa94a07fSbrendan * repurposed to another pool. 278699653d4eSeschrock */ 278799653d4eSeschrock int 2788fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 278999653d4eSeschrock { 279099653d4eSeschrock nvlist_t *label; 279199653d4eSeschrock uint64_t guid, version; 279299653d4eSeschrock uint64_t state; 279399653d4eSeschrock 2794e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2795c5904d13Seschrock return (0); 2796c5904d13Seschrock 2797dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 279899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 279999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 280099653d4eSeschrock return (-1); 280199653d4eSeschrock } 280299653d4eSeschrock 280399653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2804ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 280599653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 280699653d4eSeschrock guid != vd->vdev_guid || 280799653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 280899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 280999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 281099653d4eSeschrock nvlist_free(label); 281199653d4eSeschrock return (-1); 281299653d4eSeschrock } 281399653d4eSeschrock 281499653d4eSeschrock /* 281599653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 281699653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 281799653d4eSeschrock */ 281899653d4eSeschrock nvlist_free(label); 281999653d4eSeschrock return (0); 282099653d4eSeschrock } 282199653d4eSeschrock 28225cabbc6bSPrashanth Sreenivasa /* 28235cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 28245cabbc6bSPrashanth Sreenivasa * that points to them. 28255cabbc6bSPrashanth Sreenivasa */ 282688ecc943SGeorge Wilson void 28275cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 28285cabbc6bSPrashanth Sreenivasa { 28295cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 28305cabbc6bSPrashanth Sreenivasa return; 28315cabbc6bSPrashanth Sreenivasa 28325cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 28335cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 28345cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 28355cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 28365cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 28375cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 28385cabbc6bSPrashanth Sreenivasa 28395cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 28405cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 28415cabbc6bSPrashanth Sreenivasa if (smobj == 0) 28425cabbc6bSPrashanth Sreenivasa continue; 28435cabbc6bSPrashanth Sreenivasa 28445cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 28455cabbc6bSPrashanth Sreenivasa } 28465cabbc6bSPrashanth Sreenivasa 28475cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 28485cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 28495cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 28505cabbc6bSPrashanth Sreenivasa } 28515cabbc6bSPrashanth Sreenivasa 28525cabbc6bSPrashanth Sreenivasa static void 28534e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vdev_t *vd, uint64_t txg) 285488ecc943SGeorge Wilson { 285588ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 285688ecc943SGeorge Wilson 28574e75ba68SSerapheim Dimitropoulos ASSERT(vd->vdev_islog); 2858215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 2859215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 286088ecc943SGeorge Wilson 286188ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 28622e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 28632e4c9986SGeorge Wilson 28642e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 28652e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 28662e4c9986SGeorge Wilson 286788ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 286888ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 286988ecc943SGeorge Wilson 28700713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 287188ecc943SGeorge Wilson continue; 287288ecc943SGeorge Wilson 28730713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 28742e4c9986SGeorge Wilson /* 28752e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 28762e4c9986SGeorge Wilson * was removed then the histogram accounting may 28772e4c9986SGeorge Wilson * not be accurate. Update the histogram information 28782e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 28792e4c9986SGeorge Wilson * and metaslab class are up-to-date. 28802e4c9986SGeorge Wilson */ 28812e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 28822e4c9986SGeorge Wilson 28830713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 28840713e232SGeorge Wilson space_map_close(msp->ms_sm); 28850713e232SGeorge Wilson msp->ms_sm = NULL; 28860713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 288788ecc943SGeorge Wilson } 28882e4c9986SGeorge Wilson 288986714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 289086714001SSerapheim Dimitropoulos ASSERT(spa_has_checkpoint(spa)); 289186714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 289286714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 289386714001SSerapheim Dimitropoulos } 289486714001SSerapheim Dimitropoulos 28952e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 28962e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 28972e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 28982e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 289988ecc943SGeorge Wilson } 290088ecc943SGeorge Wilson 29014e75ba68SSerapheim Dimitropoulos dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2902215198a6SJoe Stein 29034e75ba68SSerapheim Dimitropoulos vdev_destroy_spacemaps(vd, tx); 29044e75ba68SSerapheim Dimitropoulos if (vd->vdev_top_zap != 0) { 2905215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 2906215198a6SJoe Stein vd->vdev_top_zap = 0; 2907215198a6SJoe Stein } 29084e75ba68SSerapheim Dimitropoulos 290988ecc943SGeorge Wilson dmu_tx_commit(tx); 291088ecc943SGeorge Wilson } 291188ecc943SGeorge Wilson 2912fa9e4066Sahrens void 2913fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2914fa9e4066Sahrens { 2915fa9e4066Sahrens metaslab_t *msp; 291680eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2917fa9e4066Sahrens 29185cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 291988ecc943SGeorge Wilson 2920094e47e9SGeorge Wilson while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2921094e47e9SGeorge Wilson != NULL) 2922fa9e4066Sahrens metaslab_sync_done(msp, txg); 292380eb36f2SGeorge Wilson 292480eb36f2SGeorge Wilson if (reassess) 292580eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2926fa9e4066Sahrens } 2927fa9e4066Sahrens 2928fa9e4066Sahrens void 2929fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2930fa9e4066Sahrens { 2931fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2932fa9e4066Sahrens vdev_t *lvd; 2933fa9e4066Sahrens metaslab_t *msp; 2934ecc2d604Sbonwick dmu_tx_t *tx; 2935fa9e4066Sahrens 29365cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 29375cabbc6bSPrashanth Sreenivasa dmu_tx_t *tx; 29385cabbc6bSPrashanth Sreenivasa 29395cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 29405cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 294188ecc943SGeorge Wilson 29425cabbc6bSPrashanth Sreenivasa tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 29435cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 29445cabbc6bSPrashanth Sreenivasa dmu_tx_commit(tx); 29455cabbc6bSPrashanth Sreenivasa 29465cabbc6bSPrashanth Sreenivasa /* 29475cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 29485cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 29495cabbc6bSPrashanth Sreenivasa */ 29505cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 29515cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 29525cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 29535cabbc6bSPrashanth Sreenivasa return; 29545cabbc6bSPrashanth Sreenivasa } 29555cabbc6bSPrashanth Sreenivasa } 29565cabbc6bSPrashanth Sreenivasa 29575cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 29585cabbc6bSPrashanth Sreenivasa 29595cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 29605cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 2961ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 29625cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 2963ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2964ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2965ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2966ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2967ecc2d604Sbonwick vdev_config_dirty(vd); 2968ecc2d604Sbonwick dmu_tx_commit(tx); 2969ecc2d604Sbonwick } 2970fa9e4066Sahrens 2971ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2972fa9e4066Sahrens metaslab_sync(msp, txg); 2973ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2974ecc2d604Sbonwick } 2975fa9e4066Sahrens 2976fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2977fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2978fa9e4066Sahrens 29795cabbc6bSPrashanth Sreenivasa /* 29804e75ba68SSerapheim Dimitropoulos * If this is an empty log device being removed, destroy the 29814e75ba68SSerapheim Dimitropoulos * metadata associated with it. 29825cabbc6bSPrashanth Sreenivasa */ 29834e75ba68SSerapheim Dimitropoulos if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 29844e75ba68SSerapheim Dimitropoulos vdev_remove_empty_log(vd, txg); 29855cabbc6bSPrashanth Sreenivasa 2986fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2987fa9e4066Sahrens } 2988fa9e4066Sahrens 2989fa9e4066Sahrens uint64_t 2990fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2991fa9e4066Sahrens { 2992fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2993fa9e4066Sahrens } 2994fa9e4066Sahrens 29953d7072f8Seschrock /* 29963d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 29973d7072f8Seschrock * not be opened, and no I/O is attempted. 29983d7072f8Seschrock */ 2999fa9e4066Sahrens int 3000069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 3001fa9e4066Sahrens { 30024b964adaSGeorge Wilson vdev_t *vd, *tvd; 3003fa9e4066Sahrens 30048f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 3005fa9e4066Sahrens 3006c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3007e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3008e14bb325SJeff Bonwick 30093d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3010e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3011fa9e4066Sahrens 30124b964adaSGeorge Wilson tvd = vd->vdev_top; 30134b964adaSGeorge Wilson 3014069f55e2SEric Schrock /* 3015069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 3016069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 3017069f55e2SEric Schrock * were faulted. 3018069f55e2SEric Schrock */ 3019069f55e2SEric Schrock vd->vdev_label_aux = aux; 3020069f55e2SEric Schrock 30213d7072f8Seschrock /* 30223d7072f8Seschrock * Faulted state takes precedence over degraded. 30233d7072f8Seschrock */ 302498d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 30253d7072f8Seschrock vd->vdev_faulted = 1ULL; 30263d7072f8Seschrock vd->vdev_degraded = 0ULL; 3027069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 30283d7072f8Seschrock 30293d7072f8Seschrock /* 3030c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 3031c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 30323d7072f8Seschrock */ 30334b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 30343d7072f8Seschrock vd->vdev_degraded = 1ULL; 30353d7072f8Seschrock vd->vdev_faulted = 0ULL; 30363d7072f8Seschrock 30373d7072f8Seschrock /* 30383d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 30393d7072f8Seschrock * mark it degraded. 30403d7072f8Seschrock */ 30414b964adaSGeorge Wilson vdev_reopen(tvd); 30423d7072f8Seschrock 3043069f55e2SEric Schrock if (vdev_readable(vd)) 3044069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 30453d7072f8Seschrock } 30463d7072f8Seschrock 3047e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 30483d7072f8Seschrock } 30493d7072f8Seschrock 30503d7072f8Seschrock /* 30513d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 30523d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 30533d7072f8Seschrock * as I/O is concerned. 30543d7072f8Seschrock */ 30553d7072f8Seschrock int 3056069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 30573d7072f8Seschrock { 3058c5904d13Seschrock vdev_t *vd; 30590a4e9518Sgw 30608f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 30613d7072f8Seschrock 3062c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3063e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3064e14bb325SJeff Bonwick 30650e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3066e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 30670e34b6a7Sbonwick 30683d7072f8Seschrock /* 30693d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 30703d7072f8Seschrock */ 3071e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 3072e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 30733d7072f8Seschrock 30743d7072f8Seschrock vd->vdev_degraded = 1ULL; 30753d7072f8Seschrock if (!vdev_is_dead(vd)) 30763d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 3077069f55e2SEric Schrock aux); 30783d7072f8Seschrock 3079e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 30803d7072f8Seschrock } 30813d7072f8Seschrock 30823d7072f8Seschrock /* 3083f7170741SWill Andrews * Online the given vdev. 3084f7170741SWill Andrews * 3085f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 3086f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 3087f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 3088f7170741SWill Andrews * events are generated if the device fails to open. 30893d7072f8Seschrock */ 30903d7072f8Seschrock int 3091e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 30923d7072f8Seschrock { 3093573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 30945f368aefSYuri Pankov boolean_t wasoffline; 30955f368aefSYuri Pankov vdev_state_t oldstate; 30963d7072f8Seschrock 30978f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 30983d7072f8Seschrock 3099c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3100e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 31013d7072f8Seschrock 31023d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 3103e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3104fa9e4066Sahrens 31055f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 31065f368aefSYuri Pankov oldstate = vd->vdev_state; 310714372834SHans Rosenfeld 3108573ca77eSGeorge Wilson tvd = vd->vdev_top; 3109fa9e4066Sahrens vd->vdev_offline = B_FALSE; 3110441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 3111e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3112e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3113573ca77eSGeorge Wilson 3114573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3115573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3116573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3117573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3118573ca77eSGeorge Wilson } 3119573ca77eSGeorge Wilson 3120573ca77eSGeorge Wilson vdev_reopen(tvd); 31213d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 31223d7072f8Seschrock 3123573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3124573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3125573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 3126573ca77eSGeorge Wilson } 3127573ca77eSGeorge Wilson 31283d7072f8Seschrock if (newstate) 31293d7072f8Seschrock *newstate = vd->vdev_state; 31303d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 31313d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 31323d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 31333d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 31343d7072f8Seschrock vd->vdev_unspare = B_TRUE; 3135fa9e4066Sahrens 3136573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3137573ca77eSGeorge Wilson 3138573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3139573ca77eSGeorge Wilson if (vd->vdev_aux) 3140573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3141573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3142573ca77eSGeorge Wilson } 314314372834SHans Rosenfeld 3144094e47e9SGeorge Wilson /* Restart initializing if necessary */ 3145094e47e9SGeorge Wilson mutex_enter(&vd->vdev_initialize_lock); 3146094e47e9SGeorge Wilson if (vdev_writeable(vd) && 3147094e47e9SGeorge Wilson vd->vdev_initialize_thread == NULL && 3148094e47e9SGeorge Wilson vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { 3149094e47e9SGeorge Wilson (void) vdev_initialize(vd); 3150094e47e9SGeorge Wilson } 3151094e47e9SGeorge Wilson mutex_exit(&vd->vdev_initialize_lock); 3152094e47e9SGeorge Wilson 31535f368aefSYuri Pankov if (wasoffline || 31545f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 31555f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 3156ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 315714372834SHans Rosenfeld 31588ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 3159fa9e4066Sahrens } 3160fa9e4066Sahrens 3161a1521560SJeff Bonwick static int 3162a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3163fa9e4066Sahrens { 3164e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 31658f18d1faSGeorge Wilson int error = 0; 31668f18d1faSGeorge Wilson uint64_t generation; 31678f18d1faSGeorge Wilson metaslab_group_t *mg; 31680a4e9518Sgw 31698f18d1faSGeorge Wilson top: 31708f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 3171fa9e4066Sahrens 3172c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3173e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3174fa9e4066Sahrens 31750e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3176e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 31770e34b6a7Sbonwick 3178e6ca193dSGeorge Wilson tvd = vd->vdev_top; 31798f18d1faSGeorge Wilson mg = tvd->vdev_mg; 31808f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 3181e6ca193dSGeorge Wilson 3182fa9e4066Sahrens /* 3183ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 3184fa9e4066Sahrens */ 3185ecc2d604Sbonwick if (!vd->vdev_offline) { 3186ecc2d604Sbonwick /* 31878ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 3188e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 3189e6ca193dSGeorge Wilson * expendable. 3190ecc2d604Sbonwick */ 3191e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3192e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 3193e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3194fa9e4066Sahrens 31958f18d1faSGeorge Wilson /* 3196b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 3197b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 3198b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 3199b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 32008f18d1faSGeorge Wilson */ 32018f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 32028f18d1faSGeorge Wilson /* 32038f18d1faSGeorge Wilson * Prevent any future allocations. 32048f18d1faSGeorge Wilson */ 3205a1521560SJeff Bonwick metaslab_group_passivate(mg); 32068f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 32078f18d1faSGeorge Wilson 32085cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 32098f18d1faSGeorge Wilson 321086714001SSerapheim Dimitropoulos /* 321186714001SSerapheim Dimitropoulos * If the log device was successfully reset but has 321286714001SSerapheim Dimitropoulos * checkpointed data, do not offline it. 321386714001SSerapheim Dimitropoulos */ 321486714001SSerapheim Dimitropoulos if (error == 0 && 321586714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm != NULL) { 321686714001SSerapheim Dimitropoulos ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc, 321786714001SSerapheim Dimitropoulos !=, 0); 321886714001SSerapheim Dimitropoulos error = ZFS_ERR_CHECKPOINT_EXISTS; 321986714001SSerapheim Dimitropoulos } 322086714001SSerapheim Dimitropoulos 32218f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 32228f18d1faSGeorge Wilson 32238f18d1faSGeorge Wilson /* 32248f18d1faSGeorge Wilson * Check to see if the config has changed. 32258f18d1faSGeorge Wilson */ 32268f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 3227a1521560SJeff Bonwick metaslab_group_activate(mg); 32288f18d1faSGeorge Wilson if (error) 32298f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 32308f18d1faSGeorge Wilson vd, error)); 32318f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 32328f18d1faSGeorge Wilson goto top; 32338f18d1faSGeorge Wilson } 3234fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 32358f18d1faSGeorge Wilson } 32368f18d1faSGeorge Wilson 3237ecc2d604Sbonwick /* 3238ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3239e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3240e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3241e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3242ecc2d604Sbonwick */ 3243ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3244e6ca193dSGeorge Wilson vdev_reopen(tvd); 3245e6ca193dSGeorge Wilson 3246e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3247e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3248ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3249e6ca193dSGeorge Wilson vdev_reopen(tvd); 3250e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3251ecc2d604Sbonwick } 32528f18d1faSGeorge Wilson 32538f18d1faSGeorge Wilson /* 32548f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 32558f18d1faSGeorge Wilson * once we online the device it's open for business. 32568f18d1faSGeorge Wilson */ 32578f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3258a1521560SJeff Bonwick metaslab_group_activate(mg); 3259fa9e4066Sahrens } 3260fa9e4066Sahrens 3261e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3262ecc2d604Sbonwick 32638f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3264fa9e4066Sahrens } 3265fa9e4066Sahrens 3266a1521560SJeff Bonwick int 3267a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3268a1521560SJeff Bonwick { 3269a1521560SJeff Bonwick int error; 3270a1521560SJeff Bonwick 3271a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3272a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3273a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3274a1521560SJeff Bonwick 3275a1521560SJeff Bonwick return (error); 3276a1521560SJeff Bonwick } 3277a1521560SJeff Bonwick 3278ea8dc4b6Seschrock /* 3279ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3280ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3281ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3282ea8dc4b6Seschrock */ 3283ea8dc4b6Seschrock void 3284e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3285fa9e4066Sahrens { 3286e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3287e14bb325SJeff Bonwick 3288e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3289fa9e4066Sahrens 3290ea8dc4b6Seschrock if (vd == NULL) 3291e14bb325SJeff Bonwick vd = rvd; 3292fa9e4066Sahrens 3293ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3294ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3295ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3296fa9e4066Sahrens 3297e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3298e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 32993d7072f8Seschrock 33005cabbc6bSPrashanth Sreenivasa /* 33015cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 33025cabbc6bSPrashanth Sreenivasa */ 33035cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 33045cabbc6bSPrashanth Sreenivasa return; 33055cabbc6bSPrashanth Sreenivasa 33063d7072f8Seschrock /* 33078a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 33088a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 33098a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 33108a79c1b5Sek * written out to disk. 33113d7072f8Seschrock */ 3312e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3313e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 33148a79c1b5Sek 3315096d22d4SEric Schrock /* 3316096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3317096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3318096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3319096d22d4SEric Schrock */ 3320096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3321096d22d4SEric Schrock 33224b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3323e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3324e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3325e14bb325SJeff Bonwick 3326f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 33273d7072f8Seschrock 3328096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3329096d22d4SEric Schrock 3330f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3331e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3332e14bb325SJeff Bonwick 3333e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 3334bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 33353d7072f8Seschrock 3336ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 33373d7072f8Seschrock } 3338096d22d4SEric Schrock 3339096d22d4SEric Schrock /* 3340096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3341096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3342096d22d4SEric Schrock * done in response to the FMA fault. 3343096d22d4SEric Schrock */ 3344096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3345096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3346096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3347096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3348fa9e4066Sahrens } 3349fa9e4066Sahrens 3350e14bb325SJeff Bonwick boolean_t 3351e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 33520a4e9518Sgw { 335388ecc943SGeorge Wilson /* 335488ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 335588ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 335688ecc943SGeorge Wilson * these types of devices in the various code paths. 335788ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 335888ecc943SGeorge Wilson * before issuing I/O to them. 335988ecc943SGeorge Wilson */ 33605cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 33615cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 336288ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 33630a4e9518Sgw } 33640a4e9518Sgw 3365e14bb325SJeff Bonwick boolean_t 3366e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 33670a4e9518Sgw { 3368e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 33690a4e9518Sgw } 33700a4e9518Sgw 3371e14bb325SJeff Bonwick boolean_t 3372e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3373fa9e4066Sahrens { 33745cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 33755cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3376fa9e4066Sahrens } 3377fa9e4066Sahrens 3378a31e6787SGeorge Wilson boolean_t 3379a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3380a31e6787SGeorge Wilson { 33818ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 33828ad4d6ddSJeff Bonwick 3383a31e6787SGeorge Wilson /* 33848ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3385a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3386a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 33878ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 33888ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 33898ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3390a31e6787SGeorge Wilson */ 33918ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 33925cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 33930f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3394a31e6787SGeorge Wilson } 3395a31e6787SGeorge Wilson 3396e14bb325SJeff Bonwick boolean_t 3397e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3398fa9e4066Sahrens { 3399e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3400fa9e4066Sahrens 3401e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3402e14bb325SJeff Bonwick return (B_FALSE); 3403fa9e4066Sahrens 3404e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3405e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3406fa9e4066Sahrens 3407e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3408e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3409fa9e4066Sahrens 3410e14bb325SJeff Bonwick return (B_TRUE); 3411fa9e4066Sahrens } 3412fa9e4066Sahrens 341386714001SSerapheim Dimitropoulos boolean_t 341486714001SSerapheim Dimitropoulos vdev_is_spacemap_addressable(vdev_t *vd) 341586714001SSerapheim Dimitropoulos { 3416a0b03b16SSerapheim Dimitropoulos if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) 3417a0b03b16SSerapheim Dimitropoulos return (B_TRUE); 3418a0b03b16SSerapheim Dimitropoulos 341986714001SSerapheim Dimitropoulos /* 3420a0b03b16SSerapheim Dimitropoulos * If double-word space map entries are not enabled we assume 3421a0b03b16SSerapheim Dimitropoulos * 47 bits of the space map entry are dedicated to the entry's 3422a0b03b16SSerapheim Dimitropoulos * offset (see SM_OFFSET_BITS in space_map.h). We then use that 3423a0b03b16SSerapheim Dimitropoulos * to calculate the maximum address that can be described by a 3424a0b03b16SSerapheim Dimitropoulos * space map entry for the given device. 342586714001SSerapheim Dimitropoulos */ 3426a0b03b16SSerapheim Dimitropoulos uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; 342786714001SSerapheim Dimitropoulos 342886714001SSerapheim Dimitropoulos if (shift >= 63) /* detect potential overflow */ 342986714001SSerapheim Dimitropoulos return (B_TRUE); 343086714001SSerapheim Dimitropoulos 343186714001SSerapheim Dimitropoulos return (vd->vdev_asize < (1ULL << shift)); 343286714001SSerapheim Dimitropoulos } 343386714001SSerapheim Dimitropoulos 3434fa9e4066Sahrens /* 3435fa9e4066Sahrens * Get statistics for the given vdev. 3436fa9e4066Sahrens */ 3437fa9e4066Sahrens void 3438fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3439fa9e4066Sahrens { 34402e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 34412e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 3442c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 34432e4c9986SGeorge Wilson 34442e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3445fa9e4066Sahrens 3446fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 3447fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3448fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3449fa9e4066Sahrens vs->vs_state = vd->vdev_state; 3450573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 3451094e47e9SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) { 3452573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3453094e47e9SGeorge Wilson /* 3454094e47e9SGeorge Wilson * Report intializing progress. Since we don't have the 3455094e47e9SGeorge Wilson * initializing locks held, this is only an estimate (although a 3456094e47e9SGeorge Wilson * fairly accurate one). 3457094e47e9SGeorge Wilson */ 3458094e47e9SGeorge Wilson vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done; 3459094e47e9SGeorge Wilson vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est; 3460094e47e9SGeorge Wilson vs->vs_initialize_state = vd->vdev_initialize_state; 3461094e47e9SGeorge Wilson vs->vs_initialize_action_time = vd->vdev_initialize_action_time; 3462094e47e9SGeorge Wilson } 3463c39a2aaeSGeorge Wilson /* 3464c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 3465c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 3466c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 3467c39a2aaeSGeorge Wilson */ 3468c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 34697855d95bSToomas Soome vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 34707855d95bSToomas Soome spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3471c39a2aaeSGeorge Wilson } 34725cabbc6bSPrashanth Sreenivasa if (vd->vdev_aux == NULL && vd == vd->vdev_top && 34735cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 34742e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 34752986efa8SAlex Reece } 3476fa9e4066Sahrens 3477fa9e4066Sahrens /* 3478fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 3479fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 3480fa9e4066Sahrens */ 3481fa9e4066Sahrens if (vd == rvd) { 3482e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 3483fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 3484fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 3485fa9e4066Sahrens 3486e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 3487fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 3488fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 3489fa9e4066Sahrens } 34903f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 3491fa9e4066Sahrens } 3492fa9e4066Sahrens } 34932e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3494fa9e4066Sahrens } 3495fa9e4066Sahrens 3496fa94a07fSbrendan void 3497fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3498fa94a07fSbrendan { 3499fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3500fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3501fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3502fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3503fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3504fa94a07fSbrendan } 3505fa94a07fSbrendan 35063f9d6ad7SLin Ling void 35073f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 35083f9d6ad7SLin Ling { 35093f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 35103f9d6ad7SLin Ling 35113f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 35123f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 35133f9d6ad7SLin Ling 35143f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 35153f9d6ad7SLin Ling vs->vs_scan_processed = 0; 35163f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 35173f9d6ad7SLin Ling } 35183f9d6ad7SLin Ling 3519fa9e4066Sahrens void 3520e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3521fa9e4066Sahrens { 35228ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 35238ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3524e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3525fa9e4066Sahrens vdev_t *pvd; 3526fa9e4066Sahrens uint64_t txg = zio->io_txg; 3527fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3528fa9e4066Sahrens zio_type_t type = zio->io_type; 3529fa9e4066Sahrens int flags = zio->io_flags; 3530fa9e4066Sahrens 3531e14bb325SJeff Bonwick /* 3532e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3533e14bb325SJeff Bonwick */ 3534e14bb325SJeff Bonwick if (zio->io_gang_tree) 3535e14bb325SJeff Bonwick return; 3536e14bb325SJeff Bonwick 3537fa9e4066Sahrens if (zio->io_error == 0) { 3538e14bb325SJeff Bonwick /* 3539e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3540e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3541e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3542e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3543e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3544e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3545e14bb325SJeff Bonwick * remain zero, which is what we want.) 3546e14bb325SJeff Bonwick * 3547e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3548e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3549e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3550e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3551e14bb325SJeff Bonwick */ 3552e14bb325SJeff Bonwick if (vd == rvd) 3553e14bb325SJeff Bonwick return; 3554e14bb325SJeff Bonwick 3555e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 35568ad4d6ddSJeff Bonwick 35578ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 35588ad4d6ddSJeff Bonwick return; 35598ad4d6ddSJeff Bonwick 35608ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 35618ad4d6ddSJeff Bonwick 3562e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 356344ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 35643f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 35653f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 35663f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 35673f9d6ad7SLin Ling 35683f9d6ad7SLin Ling /* XXX cleanup? */ 35693f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 35703f9d6ad7SLin Ling atomic_add_64(processed, psize); 35713f9d6ad7SLin Ling vs->vs_scan_processed += psize; 35723f9d6ad7SLin Ling } 35733f9d6ad7SLin Ling 35748ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3575e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3576fa9e4066Sahrens } 35778ad4d6ddSJeff Bonwick 35788ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 35798ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 35808ad4d6ddSJeff Bonwick 35818ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3582fa9e4066Sahrens return; 3583fa9e4066Sahrens } 3584fa9e4066Sahrens 3585fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3586fa9e4066Sahrens return; 3587fa9e4066Sahrens 35888956713aSEric Schrock /* 35898956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 35908956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 35918956713aSEric Schrock * hard errors, when in reality they can happen for any number of 35928956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 35938956713aSEric Schrock */ 35948956713aSEric Schrock if (zio->io_error == EIO && 35958956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 35968956713aSEric Schrock return; 35978956713aSEric Schrock 35988f18d1faSGeorge Wilson /* 35998f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 36008f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 36018f18d1faSGeorge Wilson * errors. 36028f18d1faSGeorge Wilson */ 36038f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 36048f18d1faSGeorge Wilson return; 36058f18d1faSGeorge Wilson 3606e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3607b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3608e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3609e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3610e14bb325SJeff Bonwick else 3611e14bb325SJeff Bonwick vs->vs_read_errors++; 3612fa9e4066Sahrens } 3613b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3614e14bb325SJeff Bonwick vs->vs_write_errors++; 3615e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3616fa9e4066Sahrens 36175cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 36185cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 36198ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 362044ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3621b24ab676SJeff Bonwick spa->spa_claiming)) { 36228ad4d6ddSJeff Bonwick /* 3623b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 3624b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 3625b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 3626b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 3627b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 3628b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 3629b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 3630b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 36318ad4d6ddSJeff Bonwick * 36328ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 36338ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 36348ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 36358ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 36368ad4d6ddSJeff Bonwick */ 36378ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 36388ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 363944ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 36408ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 36418ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 36428ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3643b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 3644b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 3645b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3646b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 36478ad4d6ddSJeff Bonwick } 3648b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 36498ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3650fa9e4066Sahrens return; 36518ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 36528ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 36538ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3654fa9e4066Sahrens } 36558ad4d6ddSJeff Bonwick if (vd != rvd) 36568ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3657fa9e4066Sahrens } 3658fa9e4066Sahrens } 3659fa9e4066Sahrens 3660fa9e4066Sahrens /* 3661b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 3662b24ab676SJeff Bonwick * and the root vdev. 3663fa9e4066Sahrens */ 3664fa9e4066Sahrens void 3665b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3666b24ab676SJeff Bonwick int64_t space_delta) 3667fa9e4066Sahrens { 366899653d4eSeschrock int64_t dspace_delta = space_delta; 36698654d025Sperrin spa_t *spa = vd->vdev_spa; 36708654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 3671b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 3672b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 3673fa9e4066Sahrens 36748654d025Sperrin ASSERT(vd == vd->vdev_top); 367599653d4eSeschrock 36768654d025Sperrin /* 36778654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 36788654d025Sperrin * factor. We must calculate this here and not at the root vdev 36798654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 36808654d025Sperrin * childrens', thus not accurate enough for us. 36818654d025Sperrin */ 36828654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 3683e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 36848654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 36858654d025Sperrin vd->vdev_deflate_ratio; 36868654d025Sperrin 36878654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 36888654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 3689b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 36908654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 36918654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 36928654d025Sperrin 3693b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 3694fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 3695fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 3696b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 3697fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 3698fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 3699fa94a07fSbrendan } 3700b24ab676SJeff Bonwick 3701b24ab676SJeff Bonwick if (mc != NULL) { 3702b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 3703b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 3704b24ab676SJeff Bonwick 3705b24ab676SJeff Bonwick metaslab_class_space_update(mc, 3706b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 3707b24ab676SJeff Bonwick } 3708fa9e4066Sahrens } 3709fa9e4066Sahrens 3710fa9e4066Sahrens /* 3711fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 3712fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 3713fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3714fa9e4066Sahrens */ 3715fa9e4066Sahrens void 3716fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3717fa9e4066Sahrens { 3718fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3719fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3720fa9e4066Sahrens int c; 3721fa9e4066Sahrens 3722f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3723f9af39baSGeorge Wilson 3724c5904d13Seschrock /* 37256809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 37266809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3727c5904d13Seschrock */ 3728c5904d13Seschrock if (vd->vdev_aux != NULL) { 3729c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3730c5904d13Seschrock nvlist_t **aux; 3731c5904d13Seschrock uint_t naux; 3732c5904d13Seschrock 3733c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3734c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3735c5904d13Seschrock break; 3736c5904d13Seschrock } 3737c5904d13Seschrock 3738e14bb325SJeff Bonwick if (c == sav->sav_count) { 3739e14bb325SJeff Bonwick /* 3740e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3741e14bb325SJeff Bonwick */ 3742e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3743e14bb325SJeff Bonwick return; 3744e14bb325SJeff Bonwick } 3745e14bb325SJeff Bonwick 3746c5904d13Seschrock sav->sav_sync = B_TRUE; 3747c5904d13Seschrock 37486809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 37496809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 37506809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 37516809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 37526809eb4eSEric Schrock } 3753c5904d13Seschrock 3754c5904d13Seschrock ASSERT(c < naux); 3755c5904d13Seschrock 3756c5904d13Seschrock /* 3757c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3758c5904d13Seschrock * sketchy, but it will work. 3759c5904d13Seschrock */ 3760c5904d13Seschrock nvlist_free(aux[c]); 37613f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3762c5904d13Seschrock 3763c5904d13Seschrock return; 3764c5904d13Seschrock } 3765c5904d13Seschrock 37665dabedeeSbonwick /* 3767e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3768e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3769e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 37705dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 37715dabedeeSbonwick */ 3772e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3773e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3774e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 37755dabedeeSbonwick 3776fa9e4066Sahrens if (vd == rvd) { 3777fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 3778fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 3779fa9e4066Sahrens } else { 3780fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 3781fa9e4066Sahrens 378288ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 37835cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3784e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 37855cabbc6bSPrashanth Sreenivasa } 3786fa9e4066Sahrens } 3787fa9e4066Sahrens } 3788fa9e4066Sahrens 3789fa9e4066Sahrens void 3790fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 3791fa9e4066Sahrens { 37925dabedeeSbonwick spa_t *spa = vd->vdev_spa; 37935dabedeeSbonwick 3794e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3795e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3796e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 37975dabedeeSbonwick 3798e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3799e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 3800e14bb325SJeff Bonwick } 3801e14bb325SJeff Bonwick 3802e14bb325SJeff Bonwick /* 3803e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 3804e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3805e14bb325SJeff Bonwick * the state changes from larger config changes because they require 3806e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3807e14bb325SJeff Bonwick */ 3808e14bb325SJeff Bonwick void 3809e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3810e14bb325SJeff Bonwick { 3811e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3812e14bb325SJeff Bonwick 3813f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3814e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3815e14bb325SJeff Bonwick 3816e14bb325SJeff Bonwick /* 3817e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3818e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3819e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3820e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3821e14bb325SJeff Bonwick */ 3822e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3823e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3824e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3825e14bb325SJeff Bonwick 38265cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 38275cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 3828e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3829e14bb325SJeff Bonwick } 3830e14bb325SJeff Bonwick 3831e14bb325SJeff Bonwick void 3832e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3833e14bb325SJeff Bonwick { 3834e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3835e14bb325SJeff Bonwick 3836e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3837e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3838e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3839e14bb325SJeff Bonwick 3840e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3841e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3842fa9e4066Sahrens } 3843fa9e4066Sahrens 384432b87932Sek /* 384532b87932Sek * Propagate vdev state up from children to parent. 384632b87932Sek */ 384744cd46caSbillm void 384844cd46caSbillm vdev_propagate_state(vdev_t *vd) 384944cd46caSbillm { 38508ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 38518ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 385244cd46caSbillm int degraded = 0, faulted = 0; 385344cd46caSbillm int corrupted = 0; 385444cd46caSbillm vdev_t *child; 385544cd46caSbillm 38563d7072f8Seschrock if (vd->vdev_children > 0) { 3857573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 38583d7072f8Seschrock child = vd->vdev_child[c]; 385951ece835Seschrock 386088ecc943SGeorge Wilson /* 38615cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 38625cabbc6bSPrashanth Sreenivasa * decision. 386388ecc943SGeorge Wilson */ 38645cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 386588ecc943SGeorge Wilson continue; 386688ecc943SGeorge Wilson 3867e14bb325SJeff Bonwick if (!vdev_readable(child) || 38688ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 386951ece835Seschrock /* 387051ece835Seschrock * Root special: if there is a top-level log 387151ece835Seschrock * device, treat the root vdev as if it were 387251ece835Seschrock * degraded. 387351ece835Seschrock */ 387451ece835Seschrock if (child->vdev_islog && vd == rvd) 387551ece835Seschrock degraded++; 387651ece835Seschrock else 387751ece835Seschrock faulted++; 387851ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 38793d7072f8Seschrock degraded++; 388051ece835Seschrock } 388144cd46caSbillm 38823d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 38833d7072f8Seschrock corrupted++; 38843d7072f8Seschrock } 388544cd46caSbillm 38863d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 38873d7072f8Seschrock 38883d7072f8Seschrock /* 3889e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 38903d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 38913d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 38923d7072f8Seschrock * replicas'. 38933d7072f8Seschrock */ 38943d7072f8Seschrock if (corrupted && vd == rvd && 38953d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 38963d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 38973d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 38983d7072f8Seschrock } 38993d7072f8Seschrock 390051ece835Seschrock if (vd->vdev_parent) 39013d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 390244cd46caSbillm } 390344cd46caSbillm 3904fa9e4066Sahrens /* 3905ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3906ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3907ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3908ea8dc4b6Seschrock * 3909ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3910ea8dc4b6Seschrock * generated. 3911fa9e4066Sahrens */ 3912fa9e4066Sahrens void 3913ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3914fa9e4066Sahrens { 3915560e6e96Seschrock uint64_t save_state; 3916c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3917ea8dc4b6Seschrock 3918ea8dc4b6Seschrock if (state == vd->vdev_state) { 3919ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3920fa9e4066Sahrens return; 3921ea8dc4b6Seschrock } 3922ea8dc4b6Seschrock 3923560e6e96Seschrock save_state = vd->vdev_state; 3924fa9e4066Sahrens 3925fa9e4066Sahrens vd->vdev_state = state; 3926fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3927fa9e4066Sahrens 39283d7072f8Seschrock /* 39293d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 393098d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 393198d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 393298d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 393398d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 393498d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 393598d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 393698d1cbfeSGeorge Wilson * children. 393798d1cbfeSGeorge Wilson */ 393898d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 393998d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 39403d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 39413d7072f8Seschrock 3942069f55e2SEric Schrock /* 3943069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3944069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3945069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3946069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3947069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3948069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3949069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3950069f55e2SEric Schrock */ 3951069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3952069f55e2SEric Schrock vd->vdev_prevstate != state) 3953069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3954069f55e2SEric Schrock 39553d7072f8Seschrock if (vd->vdev_removed && 39563d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 39573d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 39583d7072f8Seschrock /* 39593d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 39603d7072f8Seschrock * device was previously marked removed and someone attempted to 39613d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 39623d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 39633d7072f8Seschrock * it is one of our special test online cases, which is only 39643d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 39653d7072f8Seschrock * fault. 39663d7072f8Seschrock */ 39673d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 39683d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 39693d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 39703d7072f8Seschrock vd->vdev_removed = B_TRUE; 39713d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3972ea8dc4b6Seschrock /* 3973cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3974cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3975cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3976cb04b873SMark J Musante * is not considered an error. 3977ea8dc4b6Seschrock */ 3978cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3979cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3980560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3981560e6e96Seschrock vd->vdev_not_present = 1; 3982560e6e96Seschrock 3983560e6e96Seschrock /* 3984560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3985560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3986560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3987560e6e96Seschrock * want to post the ereport if the device was already in the 3988560e6e96Seschrock * CANT_OPEN state beforehand. 39893d7072f8Seschrock * 39903d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 39913d7072f8Seschrock * online the device in response to an insertion event. If we 39923d7072f8Seschrock * hit this case, then we have detected an insertion event for a 39933d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 39943d7072f8Seschrock * In this scenario, we don't post an ereport because we are 39953d7072f8Seschrock * about to replace the device, or attempt an online with 39963d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3997560e6e96Seschrock */ 39983d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 39993d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 4000c5904d13Seschrock vd != spa->spa_root_vdev) { 4001ea8dc4b6Seschrock const char *class; 4002ea8dc4b6Seschrock 4003ea8dc4b6Seschrock switch (aux) { 4004ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 4005ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 4006ea8dc4b6Seschrock break; 4007ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 4008ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 4009ea8dc4b6Seschrock break; 4010ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 4011ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 4012ea8dc4b6Seschrock break; 4013ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 4014ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 4015ea8dc4b6Seschrock break; 4016ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 4017ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 4018ea8dc4b6Seschrock break; 4019ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 4020ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 4021ea8dc4b6Seschrock break; 4022ea8dc4b6Seschrock default: 4023ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 4024ea8dc4b6Seschrock } 4025ea8dc4b6Seschrock 4026c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 4027ea8dc4b6Seschrock } 4028ea8dc4b6Seschrock 40293d7072f8Seschrock /* Erase any notion of persistent removed state */ 40303d7072f8Seschrock vd->vdev_removed = B_FALSE; 40313d7072f8Seschrock } else { 40323d7072f8Seschrock vd->vdev_removed = B_FALSE; 40333d7072f8Seschrock } 4034ea8dc4b6Seschrock 40358b33d774STim Haley if (!isopen && vd->vdev_parent) 40368b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 4037fa9e4066Sahrens } 403815e6edf1Sgw 40396f793812SPavel Zakharov boolean_t 40406f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 40416f793812SPavel Zakharov { 40426f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 40436f793812SPavel Zakharov 40446f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 40456f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 40466f793812SPavel Zakharov return (B_FALSE); 40476f793812SPavel Zakharov } 40486f793812SPavel Zakharov 40496f793812SPavel Zakharov return (B_TRUE); 40506f793812SPavel Zakharov } 40516f793812SPavel Zakharov 405215e6edf1Sgw /* 405315e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 4054c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 4055c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 405615e6edf1Sgw */ 405715e6edf1Sgw boolean_t 405815e6edf1Sgw vdev_is_bootable(vdev_t *vd) 405915e6edf1Sgw { 406015e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 406115e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 406215e6edf1Sgw 406315e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 406415e6edf1Sgw vd->vdev_children > 1) { 406515e6edf1Sgw return (B_FALSE); 40665cabbc6bSPrashanth Sreenivasa } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || 40675cabbc6bSPrashanth Sreenivasa strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { 406815e6edf1Sgw return (B_FALSE); 406915e6edf1Sgw } 407015e6edf1Sgw } 407115e6edf1Sgw 4072573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 407315e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 407415e6edf1Sgw return (B_FALSE); 407515e6edf1Sgw } 407615e6edf1Sgw return (B_TRUE); 407715e6edf1Sgw } 4078e6ca193dSGeorge Wilson 40795cabbc6bSPrashanth Sreenivasa boolean_t 40805cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 40815cabbc6bSPrashanth Sreenivasa { 40825cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 40835cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 40845cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 40855cabbc6bSPrashanth Sreenivasa return (B_FALSE); 40865cabbc6bSPrashanth Sreenivasa } else { 40875cabbc6bSPrashanth Sreenivasa return (B_TRUE); 40885cabbc6bSPrashanth Sreenivasa } 40895cabbc6bSPrashanth Sreenivasa } 40905cabbc6bSPrashanth Sreenivasa 40914b964adaSGeorge Wilson /* 40924b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 40934b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 40944b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 40954b964adaSGeorge Wilson */ 40964b964adaSGeorge Wilson boolean_t 40974b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 40984b964adaSGeorge Wilson { 40994b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 41004b964adaSGeorge Wilson !vd->vdev_removed) 41014b964adaSGeorge Wilson return (B_TRUE); 41024b964adaSGeorge Wilson 41034b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 41044b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 41054b964adaSGeorge Wilson return (B_TRUE); 41064b964adaSGeorge Wilson 41074b964adaSGeorge Wilson return (B_FALSE); 41084b964adaSGeorge Wilson } 41094b964adaSGeorge Wilson 4110573ca77eSGeorge Wilson /* 4111573ca77eSGeorge Wilson * Expand a vdev if possible. 4112573ca77eSGeorge Wilson */ 4113573ca77eSGeorge Wilson void 4114573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 4115573ca77eSGeorge Wilson { 4116573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 4117573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 411811f6a968SSerapheim Dimitropoulos ASSERT(vdev_is_concrete(vd)); 4119573ca77eSGeorge Wilson 41205cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 41215cabbc6bSPrashanth Sreenivasa 412211f6a968SSerapheim Dimitropoulos if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 4123573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 4124573ca77eSGeorge Wilson vdev_config_dirty(vd); 4125573ca77eSGeorge Wilson } 4126573ca77eSGeorge Wilson } 41271195e687SMark J Musante 41281195e687SMark J Musante /* 41291195e687SMark J Musante * Split a vdev. 41301195e687SMark J Musante */ 41311195e687SMark J Musante void 41321195e687SMark J Musante vdev_split(vdev_t *vd) 41331195e687SMark J Musante { 41341195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 41351195e687SMark J Musante 41361195e687SMark J Musante vdev_remove_child(pvd, vd); 41371195e687SMark J Musante vdev_compact_children(pvd); 41381195e687SMark J Musante 41391195e687SMark J Musante cvd = pvd->vdev_child[0]; 41401195e687SMark J Musante if (pvd->vdev_children == 1) { 41411195e687SMark J Musante vdev_remove_parent(cvd); 41421195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 41431195e687SMark J Musante } 41441195e687SMark J Musante vdev_propagate_state(cvd); 41451195e687SMark J Musante } 4146283b8460SGeorge.Wilson 4147283b8460SGeorge.Wilson void 4148283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 4149283b8460SGeorge.Wilson { 4150283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 4151283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 4152283b8460SGeorge.Wilson 4153283b8460SGeorge.Wilson vdev_deadman(cvd); 4154283b8460SGeorge.Wilson } 4155283b8460SGeorge.Wilson 4156283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 4157283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 4158283b8460SGeorge.Wilson 4159283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 416069962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 4161283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 4162283b8460SGeorge.Wilson zio_t *fio; 4163283b8460SGeorge.Wilson uint64_t delta; 4164283b8460SGeorge.Wilson 4165283b8460SGeorge.Wilson /* 4166283b8460SGeorge.Wilson * Look at the head of all the pending queues, 4167283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 4168283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 4169283b8460SGeorge.Wilson */ 417069962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 4171c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 4172c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 41733ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 41743ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 41753ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 4176283b8460SGeorge.Wilson vq->vq_io_complete_ts); 4177283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 4178283b8460SGeorge.Wilson "hung.", spa_name(spa)); 4179283b8460SGeorge.Wilson } 4180283b8460SGeorge.Wilson } 4181283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 4182283b8460SGeorge.Wilson } 4183283b8460SGeorge.Wilson } 4184