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 */ 21ad135b5dSChristopher Siden 22fa9e4066Sahrens /* 233f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 2486714001SSerapheim Dimitropoulos * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 25663207adSDon Brady * Copyright (c) 2017, Intel Corporation. 26e0f1c0afSOlaf Faaland * Copyright 2019 Joyent, Inc. 27fa9e4066Sahrens */ 28fa9e4066Sahrens 29fa9e4066Sahrens /* 30fa9e4066Sahrens * Virtual Device Labels 31fa9e4066Sahrens * --------------------- 32fa9e4066Sahrens * 33fa9e4066Sahrens * The vdev label serves several distinct purposes: 34fa9e4066Sahrens * 35fa9e4066Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 36fa9e4066Sahrens * identity within the pool. 37fa9e4066Sahrens * 38cfd63e1bSMatthew Ahrens * 2. Verify that all the devices given in a configuration are present 39fa9e4066Sahrens * within the pool. 40fa9e4066Sahrens * 41cfd63e1bSMatthew Ahrens * 3. Determine the uberblock for the pool. 42fa9e4066Sahrens * 43cfd63e1bSMatthew Ahrens * 4. In case of an import operation, determine the configuration of the 44fa9e4066Sahrens * toplevel vdev of which it is a part. 45fa9e4066Sahrens * 46cfd63e1bSMatthew Ahrens * 5. If an import operation cannot find all the devices in the pool, 47fa9e4066Sahrens * provide enough information to the administrator to determine which 48fa9e4066Sahrens * devices are missing. 49fa9e4066Sahrens * 50fa9e4066Sahrens * It is important to note that while the kernel is responsible for writing the 51fa9e4066Sahrens * label, it only consumes the information in the first three cases. The 52fa9e4066Sahrens * latter information is only consumed in userland when determining the 53fa9e4066Sahrens * configuration to import a pool. 54fa9e4066Sahrens * 55fa9e4066Sahrens * 56fa9e4066Sahrens * Label Organization 57fa9e4066Sahrens * ------------------ 58fa9e4066Sahrens * 59fa9e4066Sahrens * Before describing the contents of the label, it's important to understand how 60fa9e4066Sahrens * the labels are written and updated with respect to the uberblock. 61fa9e4066Sahrens * 62fa9e4066Sahrens * When the pool configuration is altered, either because it was newly created 63fa9e4066Sahrens * or a device was added, we want to update all the labels such that we can deal 64fa9e4066Sahrens * with fatal failure at any point. To this end, each disk has two labels which 65fa9e4066Sahrens * are updated before and after the uberblock is synced. Assuming we have 663d7072f8Seschrock * labels and an uberblock with the following transaction groups: 67fa9e4066Sahrens * 68fa9e4066Sahrens * L1 UB L2 69fa9e4066Sahrens * +------+ +------+ +------+ 70fa9e4066Sahrens * | | | | | | 71fa9e4066Sahrens * | t10 | | t10 | | t10 | 72fa9e4066Sahrens * | | | | | | 73fa9e4066Sahrens * +------+ +------+ +------+ 74fa9e4066Sahrens * 75fa9e4066Sahrens * In this stable state, the labels and the uberblock were all updated within 76fa9e4066Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 77fa9e4066Sahrens * that we can detect when we fail partway through writing the label. 78fa9e4066Sahrens * 79fa9e4066Sahrens * In order to identify which labels are valid, the labels are written in the 80fa9e4066Sahrens * following manner: 81fa9e4066Sahrens * 82cfd63e1bSMatthew Ahrens * 1. For each vdev, update 'L1' to the new label 83cfd63e1bSMatthew Ahrens * 2. Update the uberblock 84cfd63e1bSMatthew Ahrens * 3. For each vdev, update 'L2' to the new label 85fa9e4066Sahrens * 86fa9e4066Sahrens * Given arbitrary failure, we can determine the correct label to use based on 87fa9e4066Sahrens * the transaction group. If we fail after updating L1 but before updating the 88fa9e4066Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 89fa9e4066Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 90fa9e4066Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 91fa9e4066Sahrens * therefore L1 is valid. 92fa9e4066Sahrens * 93fa9e4066Sahrens * Another added complexity is that not every label is updated when the config 94fa9e4066Sahrens * is synced. If we add a single device, we do not want to have to re-write 95fa9e4066Sahrens * every label for every device in the pool. This means that both L1 and L2 may 96fa9e4066Sahrens * be older than the pool uberblock, because the necessary information is stored 97fa9e4066Sahrens * on another vdev. 98fa9e4066Sahrens * 99fa9e4066Sahrens * 100fa9e4066Sahrens * On-disk Format 101fa9e4066Sahrens * -------------- 102fa9e4066Sahrens * 103fa9e4066Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 104fa9e4066Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 105fa9e4066Sahrens * VTOC disk labels, but is otherwise ignored. 106fa9e4066Sahrens * 107fa9e4066Sahrens * The first half of the label is a packed nvlist which contains pool wide 108fa9e4066Sahrens * properties, per-vdev properties, and configuration information. It is 109fa9e4066Sahrens * described in more detail below. 110fa9e4066Sahrens * 111fa9e4066Sahrens * The latter half of the label consists of a redundant array of uberblocks. 112fa9e4066Sahrens * These uberblocks are updated whenever a transaction group is committed, 113fa9e4066Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 114fa9e4066Sahrens * vdev for the 'best' uberblock. 115fa9e4066Sahrens * 116fa9e4066Sahrens * 117fa9e4066Sahrens * Configuration Information 118fa9e4066Sahrens * ------------------------- 119fa9e4066Sahrens * 120fa9e4066Sahrens * The nvlist describing the pool and vdev contains the following elements: 121fa9e4066Sahrens * 122cfd63e1bSMatthew Ahrens * version ZFS on-disk version 123cfd63e1bSMatthew Ahrens * name Pool name 124cfd63e1bSMatthew Ahrens * state Pool state 125cfd63e1bSMatthew Ahrens * txg Transaction group in which this label was written 126cfd63e1bSMatthew Ahrens * pool_guid Unique identifier for this pool 127cfd63e1bSMatthew Ahrens * vdev_tree An nvlist describing vdev tree. 128ad135b5dSChristopher Siden * features_for_read 129ad135b5dSChristopher Siden * An nvlist of the features necessary for reading the MOS. 130fa9e4066Sahrens * 131fa9e4066Sahrens * Each leaf device label also contains the following: 132fa9e4066Sahrens * 133cfd63e1bSMatthew Ahrens * top_guid Unique ID for top-level vdev in which this is contained 134cfd63e1bSMatthew Ahrens * guid Unique ID for the leaf vdev 135fa9e4066Sahrens * 136fa9e4066Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 137fa9e4066Sahrens */ 138fa9e4066Sahrens 139fa9e4066Sahrens #include <sys/zfs_context.h> 140fa9e4066Sahrens #include <sys/spa.h> 141fa9e4066Sahrens #include <sys/spa_impl.h> 142fa9e4066Sahrens #include <sys/dmu.h> 143fa9e4066Sahrens #include <sys/zap.h> 144fa9e4066Sahrens #include <sys/vdev.h> 145fa9e4066Sahrens #include <sys/vdev_impl.h> 146fa9e4066Sahrens #include <sys/uberblock_impl.h> 147fa9e4066Sahrens #include <sys/metaslab.h> 1485cabbc6bSPrashanth Sreenivasa #include <sys/metaslab_impl.h> 149fa9e4066Sahrens #include <sys/zio.h> 1503f9d6ad7SLin Ling #include <sys/dsl_scan.h> 151770499e1SDan Kimmel #include <sys/abd.h> 152fa9e4066Sahrens #include <sys/fs/zfs.h> 153fa9e4066Sahrens 154fa9e4066Sahrens /* 155fa9e4066Sahrens * Basic routines to read and write from a vdev label. 156fa9e4066Sahrens * Used throughout the rest of this file. 157fa9e4066Sahrens */ 158fa9e4066Sahrens uint64_t 159fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 160fa9e4066Sahrens { 161ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 162e7437265Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 163ecc2d604Sbonwick 164fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 165fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 166fa9e4066Sahrens } 167fa9e4066Sahrens 16821bf64a7Sgw /* 16921bf64a7Sgw * Returns back the vdev label associated with the passed in offset. 17021bf64a7Sgw */ 17121bf64a7Sgw int 17221bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset) 17321bf64a7Sgw { 17421bf64a7Sgw int l; 17521bf64a7Sgw 17621bf64a7Sgw if (offset >= psize - VDEV_LABEL_END_SIZE) { 17721bf64a7Sgw offset -= psize - VDEV_LABEL_END_SIZE; 17821bf64a7Sgw offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 17921bf64a7Sgw } 18021bf64a7Sgw l = offset / sizeof (vdev_label_t); 18121bf64a7Sgw return (l < VDEV_LABELS ? l : -1); 18221bf64a7Sgw } 18321bf64a7Sgw 184fa9e4066Sahrens static void 185770499e1SDan Kimmel vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 1869a686fbcSPaul Dagnelie uint64_t size, zio_done_func_t *done, void *private, int flags) 187fa9e4066Sahrens { 18858447f68SOlaf Faaland ASSERT( 18958447f68SOlaf Faaland spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 19058447f68SOlaf Faaland spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 191e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 192fa9e4066Sahrens 193fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 194fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 195fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 196e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 197fa9e4066Sahrens } 198fa9e4066Sahrens 199e0f1c0afSOlaf Faaland void 200770499e1SDan Kimmel vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset, 2019a686fbcSPaul Dagnelie uint64_t size, zio_done_func_t *done, void *private, int flags) 202fa9e4066Sahrens { 20358447f68SOlaf Faaland ASSERT( 20458447f68SOlaf Faaland spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE || 20558447f68SOlaf Faaland spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE); 206e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 207fa9e4066Sahrens 208fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 209fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 210fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 21117f17c2dSbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 212fa9e4066Sahrens } 213fa9e4066Sahrens 21486714001SSerapheim Dimitropoulos static void 21586714001SSerapheim Dimitropoulos root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl) 21686714001SSerapheim Dimitropoulos { 21786714001SSerapheim Dimitropoulos spa_t *spa = vd->vdev_spa; 21886714001SSerapheim Dimitropoulos 21986714001SSerapheim Dimitropoulos if (vd != spa->spa_root_vdev) 22086714001SSerapheim Dimitropoulos return; 22186714001SSerapheim Dimitropoulos 22286714001SSerapheim Dimitropoulos /* provide either current or previous scan information */ 22386714001SSerapheim Dimitropoulos pool_scan_stat_t ps; 22486714001SSerapheim Dimitropoulos if (spa_scan_get_stats(spa, &ps) == 0) { 22586714001SSerapheim Dimitropoulos fnvlist_add_uint64_array(nvl, 22686714001SSerapheim Dimitropoulos ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 22786714001SSerapheim Dimitropoulos sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 22886714001SSerapheim Dimitropoulos } 22986714001SSerapheim Dimitropoulos 23086714001SSerapheim Dimitropoulos pool_removal_stat_t prs; 23186714001SSerapheim Dimitropoulos if (spa_removal_get_stats(spa, &prs) == 0) { 23286714001SSerapheim Dimitropoulos fnvlist_add_uint64_array(nvl, 23386714001SSerapheim Dimitropoulos ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs, 23486714001SSerapheim Dimitropoulos sizeof (prs) / sizeof (uint64_t)); 23586714001SSerapheim Dimitropoulos } 23686714001SSerapheim Dimitropoulos 23786714001SSerapheim Dimitropoulos pool_checkpoint_stat_t pcs; 23886714001SSerapheim Dimitropoulos if (spa_checkpoint_get_stats(spa, &pcs) == 0) { 23986714001SSerapheim Dimitropoulos fnvlist_add_uint64_array(nvl, 24086714001SSerapheim Dimitropoulos ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs, 24186714001SSerapheim Dimitropoulos sizeof (pcs) / sizeof (uint64_t)); 24286714001SSerapheim Dimitropoulos } 24386714001SSerapheim Dimitropoulos } 24486714001SSerapheim Dimitropoulos 245fa9e4066Sahrens /* 246fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 247fa9e4066Sahrens */ 248fa9e4066Sahrens nvlist_t * 24999653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 2503f9d6ad7SLin Ling vdev_config_flag_t flags) 251fa9e4066Sahrens { 252fa9e4066Sahrens nvlist_t *nv = NULL; 2535cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 254fa9e4066Sahrens 255b4952e17SGeorge Wilson nv = fnvlist_alloc(); 256fa9e4066Sahrens 257b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 2583f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 259b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 260b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 261fa9e4066Sahrens 262fa9e4066Sahrens if (vd->vdev_path != NULL) 263b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 264fa9e4066Sahrens 265fa9e4066Sahrens if (vd->vdev_devid != NULL) 266b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 267fa9e4066Sahrens 2683d7072f8Seschrock if (vd->vdev_physpath != NULL) 269b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 270b4952e17SGeorge Wilson vd->vdev_physpath); 2713d7072f8Seschrock 2726809eb4eSEric Schrock if (vd->vdev_fru != NULL) 273b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 2746809eb4eSEric Schrock 27599653d4eSeschrock if (vd->vdev_nparity != 0) { 27699653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 27799653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 27899653d4eSeschrock 27999653d4eSeschrock /* 28099653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 28199653d4eSeschrock * into a crufty old storage pool. 28299653d4eSeschrock */ 28399653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 284f94275ceSAdam Leventhal (vd->vdev_nparity <= 2 && 285f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ2) || 286f94275ceSAdam Leventhal (vd->vdev_nparity <= 3 && 287f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ3)); 28899653d4eSeschrock 28999653d4eSeschrock /* 29099653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 29199653d4eSeschrock * that only support a single parity device -- older software 29299653d4eSeschrock * will just ignore it. 29399653d4eSeschrock */ 294b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 29599653d4eSeschrock } 29699653d4eSeschrock 297afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 298b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 299b4952e17SGeorge Wilson vd->vdev_wholedisk); 300afefbcddSeschrock 3016f793812SPavel Zakharov if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING)) 302b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 303ea8dc4b6Seschrock 30499653d4eSeschrock if (vd->vdev_isspare) 305b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 30699653d4eSeschrock 3073f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 3083f9d6ad7SLin Ling vd == vd->vdev_top) { 309b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 310b4952e17SGeorge Wilson vd->vdev_ms_array); 311b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 312b4952e17SGeorge Wilson vd->vdev_ms_shift); 313b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 314b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 315b4952e17SGeorge Wilson vd->vdev_asize); 316b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 3175cabbc6bSPrashanth Sreenivasa if (vd->vdev_removing) { 318b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 319b4952e17SGeorge Wilson vd->vdev_removing); 3205cabbc6bSPrashanth Sreenivasa } 321663207adSDon Brady 322663207adSDon Brady /* zpool command expects alloc class data */ 323663207adSDon Brady if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) { 324663207adSDon Brady const char *bias = NULL; 325663207adSDon Brady 326663207adSDon Brady switch (vd->vdev_alloc_bias) { 327663207adSDon Brady case VDEV_BIAS_LOG: 328663207adSDon Brady bias = VDEV_ALLOC_BIAS_LOG; 329663207adSDon Brady break; 330663207adSDon Brady case VDEV_BIAS_SPECIAL: 331663207adSDon Brady bias = VDEV_ALLOC_BIAS_SPECIAL; 332663207adSDon Brady break; 333663207adSDon Brady case VDEV_BIAS_DEDUP: 334663207adSDon Brady bias = VDEV_ALLOC_BIAS_DEDUP; 335663207adSDon Brady break; 336663207adSDon Brady default: 337663207adSDon Brady ASSERT3U(vd->vdev_alloc_bias, ==, 338663207adSDon Brady VDEV_BIAS_NONE); 339663207adSDon Brady } 340663207adSDon Brady fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, 341663207adSDon Brady bias); 342663207adSDon Brady } 343fa9e4066Sahrens } 344fa9e4066Sahrens 3450713e232SGeorge Wilson if (vd->vdev_dtl_sm != NULL) { 346b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 3470713e232SGeorge Wilson space_map_object(vd->vdev_dtl_sm)); 3480713e232SGeorge Wilson } 349fa9e4066Sahrens 3505cabbc6bSPrashanth Sreenivasa if (vic->vic_mapping_object != 0) { 3515cabbc6bSPrashanth Sreenivasa fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 3525cabbc6bSPrashanth Sreenivasa vic->vic_mapping_object); 3535cabbc6bSPrashanth Sreenivasa } 3545cabbc6bSPrashanth Sreenivasa 3555cabbc6bSPrashanth Sreenivasa if (vic->vic_births_object != 0) { 3565cabbc6bSPrashanth Sreenivasa fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 3575cabbc6bSPrashanth Sreenivasa vic->vic_births_object); 3585cabbc6bSPrashanth Sreenivasa } 3595cabbc6bSPrashanth Sreenivasa 3605cabbc6bSPrashanth Sreenivasa if (vic->vic_prev_indirect_vdev != UINT64_MAX) { 3615cabbc6bSPrashanth Sreenivasa fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 3625cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev); 3635cabbc6bSPrashanth Sreenivasa } 3645cabbc6bSPrashanth Sreenivasa 36588ecc943SGeorge Wilson if (vd->vdev_crtxg) 366b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 36788ecc943SGeorge Wilson 368215198a6SJoe Stein if (flags & VDEV_CONFIG_MOS) { 369215198a6SJoe Stein if (vd->vdev_leaf_zap != 0) { 370215198a6SJoe Stein ASSERT(vd->vdev_ops->vdev_op_leaf); 371215198a6SJoe Stein fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 372215198a6SJoe Stein vd->vdev_leaf_zap); 373215198a6SJoe Stein } 374215198a6SJoe Stein 375215198a6SJoe Stein if (vd->vdev_top_zap != 0) { 376215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 377215198a6SJoe Stein fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 378215198a6SJoe Stein vd->vdev_top_zap); 379215198a6SJoe Stein } 380*e4c795beSTom Caputi 381*e4c795beSTom Caputi if (vd->vdev_resilver_deferred) { 382*e4c795beSTom Caputi ASSERT(vd->vdev_ops->vdev_op_leaf); 383*e4c795beSTom Caputi ASSERT(spa->spa_resilver_deferred); 384*e4c795beSTom Caputi fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER); 385*e4c795beSTom Caputi } 386215198a6SJoe Stein } 387215198a6SJoe Stein 388fa9e4066Sahrens if (getstats) { 389fa9e4066Sahrens vdev_stat_t vs; 3903f9d6ad7SLin Ling 391fa9e4066Sahrens vdev_get_stats(vd, &vs); 392b4952e17SGeorge Wilson fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 393b4952e17SGeorge Wilson (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 3943f9d6ad7SLin Ling 39586714001SSerapheim Dimitropoulos root_vdev_actions_getprogress(vd, nv); 3965cabbc6bSPrashanth Sreenivasa 3975cabbc6bSPrashanth Sreenivasa /* 3985cabbc6bSPrashanth Sreenivasa * Note: this can be called from open context 3995cabbc6bSPrashanth Sreenivasa * (spa_get_stats()), so we need the rwlock to prevent 4005cabbc6bSPrashanth Sreenivasa * the mapping from being changed by condensing. 4015cabbc6bSPrashanth Sreenivasa */ 4025cabbc6bSPrashanth Sreenivasa rw_enter(&vd->vdev_indirect_rwlock, RW_READER); 4035cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_mapping != NULL) { 4045cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_indirect_births != NULL); 4055cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_t *vim = 4065cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping; 4075cabbc6bSPrashanth Sreenivasa fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 4085cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_size(vim)); 4095cabbc6bSPrashanth Sreenivasa } 4105cabbc6bSPrashanth Sreenivasa rw_exit(&vd->vdev_indirect_rwlock); 4115cabbc6bSPrashanth Sreenivasa if (vd->vdev_mg != NULL && 4125cabbc6bSPrashanth Sreenivasa vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) { 4135cabbc6bSPrashanth Sreenivasa /* 4145cabbc6bSPrashanth Sreenivasa * Compute approximately how much memory would be used 4155cabbc6bSPrashanth Sreenivasa * for the indirect mapping if this device were to 4165cabbc6bSPrashanth Sreenivasa * be removed. 4175cabbc6bSPrashanth Sreenivasa * 4185cabbc6bSPrashanth Sreenivasa * Note: If the frag metric is invalid, then not 4195cabbc6bSPrashanth Sreenivasa * enough metaslabs have been converted to have 4205cabbc6bSPrashanth Sreenivasa * histograms. 4215cabbc6bSPrashanth Sreenivasa */ 4225cabbc6bSPrashanth Sreenivasa uint64_t seg_count = 0; 423cfd63e1bSMatthew Ahrens uint64_t to_alloc = vd->vdev_stat.vs_alloc; 4245cabbc6bSPrashanth Sreenivasa 4255cabbc6bSPrashanth Sreenivasa /* 4265cabbc6bSPrashanth Sreenivasa * There are the same number of allocated segments 4275cabbc6bSPrashanth Sreenivasa * as free segments, so we will have at least one 428cfd63e1bSMatthew Ahrens * entry per free segment. However, small free 429cfd63e1bSMatthew Ahrens * segments (smaller than vdev_removal_max_span) 430cfd63e1bSMatthew Ahrens * will be combined with adjacent allocated segments 431cfd63e1bSMatthew Ahrens * as a single mapping. 4325cabbc6bSPrashanth Sreenivasa */ 4335cabbc6bSPrashanth Sreenivasa for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { 434cfd63e1bSMatthew Ahrens if (1ULL << (i + 1) < vdev_removal_max_span) { 435cfd63e1bSMatthew Ahrens to_alloc += 436cfd63e1bSMatthew Ahrens vd->vdev_mg->mg_histogram[i] << 437cfd63e1bSMatthew Ahrens i + 1; 438cfd63e1bSMatthew Ahrens } else { 439cfd63e1bSMatthew Ahrens seg_count += 440cfd63e1bSMatthew Ahrens vd->vdev_mg->mg_histogram[i]; 441cfd63e1bSMatthew Ahrens } 4425cabbc6bSPrashanth Sreenivasa } 4435cabbc6bSPrashanth Sreenivasa 4445cabbc6bSPrashanth Sreenivasa /* 445cfd63e1bSMatthew Ahrens * The maximum length of a mapping is 446cfd63e1bSMatthew Ahrens * zfs_remove_max_segment, so we need at least one entry 447cfd63e1bSMatthew Ahrens * per zfs_remove_max_segment of allocated data. 4485cabbc6bSPrashanth Sreenivasa */ 449cfd63e1bSMatthew Ahrens seg_count += to_alloc / zfs_remove_max_segment; 4505cabbc6bSPrashanth Sreenivasa 4515cabbc6bSPrashanth Sreenivasa fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE, 4525cabbc6bSPrashanth Sreenivasa seg_count * 4535cabbc6bSPrashanth Sreenivasa sizeof (vdev_indirect_mapping_entry_phys_t)); 4545cabbc6bSPrashanth Sreenivasa } 455fa9e4066Sahrens } 456fa9e4066Sahrens 457fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 458fa9e4066Sahrens nvlist_t **child; 4593f9d6ad7SLin Ling int c, idx; 460fa9e4066Sahrens 46188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 46288ecc943SGeorge Wilson 463fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 464fa9e4066Sahrens KM_SLEEP); 465fa9e4066Sahrens 4663f9d6ad7SLin Ling for (c = 0, idx = 0; c < vd->vdev_children; c++) { 4673f9d6ad7SLin Ling vdev_t *cvd = vd->vdev_child[c]; 4683f9d6ad7SLin Ling 4693f9d6ad7SLin Ling /* 4703f9d6ad7SLin Ling * If we're generating an nvlist of removing 4713f9d6ad7SLin Ling * vdevs then skip over any device which is 4723f9d6ad7SLin Ling * not being removed. 4733f9d6ad7SLin Ling */ 4743f9d6ad7SLin Ling if ((flags & VDEV_CONFIG_REMOVING) && 4753f9d6ad7SLin Ling !cvd->vdev_removing) 4763f9d6ad7SLin Ling continue; 477fa9e4066Sahrens 4783f9d6ad7SLin Ling child[idx++] = vdev_config_generate(spa, cvd, 4793f9d6ad7SLin Ling getstats, flags); 4803f9d6ad7SLin Ling } 4813f9d6ad7SLin Ling 4823f9d6ad7SLin Ling if (idx) { 483b4952e17SGeorge Wilson fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 484b4952e17SGeorge Wilson child, idx); 4853f9d6ad7SLin Ling } 486fa9e4066Sahrens 4873f9d6ad7SLin Ling for (c = 0; c < idx; c++) 488fa9e4066Sahrens nvlist_free(child[c]); 489fa9e4066Sahrens 490fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 491441d80aaSlling 492441d80aaSlling } else { 493069f55e2SEric Schrock const char *aux = NULL; 494069f55e2SEric Schrock 495ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 496b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 497b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0) 498b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 499b4952e17SGeorge Wilson vd->vdev_resilver_txg); 5003d7072f8Seschrock if (vd->vdev_faulted) 501b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 5023d7072f8Seschrock if (vd->vdev_degraded) 503b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 5043d7072f8Seschrock if (vd->vdev_removed) 505b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 5063d7072f8Seschrock if (vd->vdev_unspare) 507b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 50888ecc943SGeorge Wilson if (vd->vdev_ishole) 509b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 510069f55e2SEric Schrock 511069f55e2SEric Schrock switch (vd->vdev_stat.vs_aux) { 512069f55e2SEric Schrock case VDEV_AUX_ERR_EXCEEDED: 513069f55e2SEric Schrock aux = "err_exceeded"; 514069f55e2SEric Schrock break; 515069f55e2SEric Schrock 516069f55e2SEric Schrock case VDEV_AUX_EXTERNAL: 517069f55e2SEric Schrock aux = "external"; 518069f55e2SEric Schrock break; 519069f55e2SEric Schrock } 520069f55e2SEric Schrock 521069f55e2SEric Schrock if (aux != NULL) 522b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 5231195e687SMark J Musante 5241195e687SMark J Musante if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 525b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 526b4952e17SGeorge Wilson vd->vdev_orig_guid); 5271195e687SMark J Musante } 528fa9e4066Sahrens } 529fa9e4066Sahrens 530fa9e4066Sahrens return (nv); 531fa9e4066Sahrens } 532fa9e4066Sahrens 53388ecc943SGeorge Wilson /* 53488ecc943SGeorge Wilson * Generate a view of the top-level vdevs. If we currently have holes 53588ecc943SGeorge Wilson * in the namespace, then generate an array which contains a list of holey 53688ecc943SGeorge Wilson * vdevs. Additionally, add the number of top-level children that currently 53788ecc943SGeorge Wilson * exist. 53888ecc943SGeorge Wilson */ 53988ecc943SGeorge Wilson void 54088ecc943SGeorge Wilson vdev_top_config_generate(spa_t *spa, nvlist_t *config) 54188ecc943SGeorge Wilson { 54288ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 54388ecc943SGeorge Wilson uint64_t *array; 5443f9d6ad7SLin Ling uint_t c, idx; 54588ecc943SGeorge Wilson 54688ecc943SGeorge Wilson array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 54788ecc943SGeorge Wilson 5483f9d6ad7SLin Ling for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 54988ecc943SGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 55088ecc943SGeorge Wilson 5515cabbc6bSPrashanth Sreenivasa if (tvd->vdev_ishole) { 55288ecc943SGeorge Wilson array[idx++] = c; 5535cabbc6bSPrashanth Sreenivasa } 55488ecc943SGeorge Wilson } 55588ecc943SGeorge Wilson 556312c6e15SGeorge Wilson if (idx) { 557312c6e15SGeorge Wilson VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 558312c6e15SGeorge Wilson array, idx) == 0); 559312c6e15SGeorge Wilson } 560312c6e15SGeorge Wilson 56188ecc943SGeorge Wilson VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 56288ecc943SGeorge Wilson rvd->vdev_children) == 0); 56388ecc943SGeorge Wilson 56488ecc943SGeorge Wilson kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 56588ecc943SGeorge Wilson } 56688ecc943SGeorge Wilson 567ad135b5dSChristopher Siden /* 568dfbb9432SGeorge Wilson * Returns the configuration from the label of the given vdev. For vdevs 569dfbb9432SGeorge Wilson * which don't have a txg value stored on their label (i.e. spares/cache) 570dfbb9432SGeorge Wilson * or have not been completely initialized (txg = 0) just return 571dfbb9432SGeorge Wilson * the configuration from the first valid label we find. Otherwise, 572dfbb9432SGeorge Wilson * find the most up-to-date label that does not exceed the specified 573dfbb9432SGeorge Wilson * 'txg' value. 574ad135b5dSChristopher Siden */ 575fa9e4066Sahrens nvlist_t * 576dfbb9432SGeorge Wilson vdev_label_read_config(vdev_t *vd, uint64_t txg) 577fa9e4066Sahrens { 5780373e76bSbonwick spa_t *spa = vd->vdev_spa; 579fa9e4066Sahrens nvlist_t *config = NULL; 580fa9e4066Sahrens vdev_phys_t *vp; 581770499e1SDan Kimmel abd_t *vp_abd; 582fa9e4066Sahrens zio_t *zio; 583dfbb9432SGeorge Wilson uint64_t best_txg = 0; 584b6bf6e15SPavel Zakharov uint64_t label_txg = 0; 585dfbb9432SGeorge Wilson int error = 0; 5868956713aSEric Schrock int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 5878956713aSEric Schrock ZIO_FLAG_SPECULATIVE; 588fa9e4066Sahrens 589e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 5900373e76bSbonwick 5910a4e9518Sgw if (!vdev_readable(vd)) 592fa9e4066Sahrens return (NULL); 593fa9e4066Sahrens 594770499e1SDan Kimmel vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 595770499e1SDan Kimmel vp = abd_to_buf(vp_abd); 596fa9e4066Sahrens 5978956713aSEric Schrock retry: 598e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 599dfbb9432SGeorge Wilson nvlist_t *label = NULL; 600fa9e4066Sahrens 601e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 602fa9e4066Sahrens 603770499e1SDan Kimmel vdev_label_read(zio, vd, l, vp_abd, 604fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 605e14bb325SJeff Bonwick sizeof (vdev_phys_t), NULL, NULL, flags); 606fa9e4066Sahrens 607fa9e4066Sahrens if (zio_wait(zio) == 0 && 608fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 609dfbb9432SGeorge Wilson &label, 0) == 0) { 610dfbb9432SGeorge Wilson /* 611dfbb9432SGeorge Wilson * Auxiliary vdevs won't have txg values in their 612dfbb9432SGeorge Wilson * labels and newly added vdevs may not have been 613dfbb9432SGeorge Wilson * completely initialized so just return the 614dfbb9432SGeorge Wilson * configuration from the first valid label we 615dfbb9432SGeorge Wilson * encounter. 616dfbb9432SGeorge Wilson */ 617dfbb9432SGeorge Wilson error = nvlist_lookup_uint64(label, 618dfbb9432SGeorge Wilson ZPOOL_CONFIG_POOL_TXG, &label_txg); 619dfbb9432SGeorge Wilson if ((error || label_txg == 0) && !config) { 620dfbb9432SGeorge Wilson config = label; 621dfbb9432SGeorge Wilson break; 622dfbb9432SGeorge Wilson } else if (label_txg <= txg && label_txg > best_txg) { 623dfbb9432SGeorge Wilson best_txg = label_txg; 624dfbb9432SGeorge Wilson nvlist_free(config); 625dfbb9432SGeorge Wilson config = fnvlist_dup(label); 626dfbb9432SGeorge Wilson } 627dfbb9432SGeorge Wilson } 628fa9e4066Sahrens 629dfbb9432SGeorge Wilson if (label != NULL) { 630dfbb9432SGeorge Wilson nvlist_free(label); 631dfbb9432SGeorge Wilson label = NULL; 632fa9e4066Sahrens } 633fa9e4066Sahrens } 634fa9e4066Sahrens 6358956713aSEric Schrock if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 6368956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 6378956713aSEric Schrock goto retry; 6388956713aSEric Schrock } 6398956713aSEric Schrock 640b6bf6e15SPavel Zakharov /* 641b6bf6e15SPavel Zakharov * We found a valid label but it didn't pass txg restrictions. 642b6bf6e15SPavel Zakharov */ 643b6bf6e15SPavel Zakharov if (config == NULL && label_txg != 0) { 644b6bf6e15SPavel Zakharov vdev_dbgmsg(vd, "label discarded as txg is too large " 645b6bf6e15SPavel Zakharov "(%llu > %llu)", (u_longlong_t)label_txg, 646b6bf6e15SPavel Zakharov (u_longlong_t)txg); 647b6bf6e15SPavel Zakharov } 648b6bf6e15SPavel Zakharov 649770499e1SDan Kimmel abd_free(vp_abd); 650fa9e4066Sahrens 651fa9e4066Sahrens return (config); 652fa9e4066Sahrens } 653fa9e4066Sahrens 65439c23413Seschrock /* 65539c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 65639c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 65739c23413Seschrock */ 65839c23413Seschrock static boolean_t 65939c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 660fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 66139c23413Seschrock { 66239c23413Seschrock spa_t *spa = vd->vdev_spa; 66339c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 66439c23413Seschrock uint64_t vdtxg = 0; 66539c23413Seschrock nvlist_t *label; 66639c23413Seschrock 66739c23413Seschrock if (spare_guid) 66839c23413Seschrock *spare_guid = 0ULL; 669fa94a07fSbrendan if (l2cache_guid) 670fa94a07fSbrendan *l2cache_guid = 0ULL; 67139c23413Seschrock 67239c23413Seschrock /* 67339c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 67439c23413Seschrock */ 675dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 67639c23413Seschrock return (B_FALSE); 67739c23413Seschrock 67839c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 67939c23413Seschrock &vdtxg); 68039c23413Seschrock 68139c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 68239c23413Seschrock &state) != 0 || 68339c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 68439c23413Seschrock &device_guid) != 0) { 68539c23413Seschrock nvlist_free(label); 68639c23413Seschrock return (B_FALSE); 68739c23413Seschrock } 68839c23413Seschrock 689fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 69039c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 69139c23413Seschrock &pool_guid) != 0 || 69239c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 69339c23413Seschrock &txg) != 0)) { 69439c23413Seschrock nvlist_free(label); 69539c23413Seschrock return (B_FALSE); 69639c23413Seschrock } 69739c23413Seschrock 69839c23413Seschrock nvlist_free(label); 69939c23413Seschrock 70039c23413Seschrock /* 70139c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 70239c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 70339c23413Seschrock * spare (which we check for later on). 70439c23413Seschrock */ 705fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 70639c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 70789a89ebfSlling !spa_spare_exists(device_guid, NULL, NULL) && 708fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 70939c23413Seschrock return (B_FALSE); 71039c23413Seschrock 71139c23413Seschrock /* 71239c23413Seschrock * If the transaction group is zero, then this an initialized (but 71339c23413Seschrock * unused) label. This is only an error if the create transaction 71439c23413Seschrock * on-disk is the same as the one we're using now, in which case the 71539c23413Seschrock * user has attempted to add the same vdev multiple times in the same 71639c23413Seschrock * transaction. 71739c23413Seschrock */ 718fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 719fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 72039c23413Seschrock return (B_TRUE); 72139c23413Seschrock 72239c23413Seschrock /* 72339c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 72439c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 725fa94a07fSbrendan * to add. We also check if it is an l2cache device. 72639c23413Seschrock */ 72789a89ebfSlling if (spa_spare_exists(device_guid, &spare_pool, NULL) || 72839c23413Seschrock spa_has_spare(spa, device_guid)) { 72939c23413Seschrock if (spare_guid) 73039c23413Seschrock *spare_guid = device_guid; 73139c23413Seschrock 73239c23413Seschrock switch (reason) { 73339c23413Seschrock case VDEV_LABEL_CREATE: 734fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 73539c23413Seschrock return (B_TRUE); 73639c23413Seschrock 73739c23413Seschrock case VDEV_LABEL_REPLACE: 73839c23413Seschrock return (!spa_has_spare(spa, device_guid) || 73939c23413Seschrock spare_pool != 0ULL); 74039c23413Seschrock 74139c23413Seschrock case VDEV_LABEL_SPARE: 74239c23413Seschrock return (spa_has_spare(spa, device_guid)); 74339c23413Seschrock } 74439c23413Seschrock } 74539c23413Seschrock 746fa94a07fSbrendan /* 747fa94a07fSbrendan * Check to see if this is an l2cache device. 748fa94a07fSbrendan */ 749fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 750fa94a07fSbrendan return (B_TRUE); 751fa94a07fSbrendan 752f9af39baSGeorge Wilson /* 753f9af39baSGeorge Wilson * We can't rely on a pool's state if it's been imported 754f9af39baSGeorge Wilson * read-only. Instead we look to see if the pools is marked 755f9af39baSGeorge Wilson * read-only in the namespace and set the state to active. 756f9af39baSGeorge Wilson */ 7575bdd995dSRichard Yao if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 7585bdd995dSRichard Yao (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 759f9af39baSGeorge Wilson spa_mode(spa) == FREAD) 760f9af39baSGeorge Wilson state = POOL_STATE_ACTIVE; 761f9af39baSGeorge Wilson 76239c23413Seschrock /* 76339c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 76439c23413Seschrock * pool on the system. 76539c23413Seschrock */ 76639c23413Seschrock return (state == POOL_STATE_ACTIVE); 76739c23413Seschrock } 76839c23413Seschrock 76939c23413Seschrock /* 77039c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 77139c23413Seschrock * use, and writable. We put down an initial label which we will later 77239c23413Seschrock * overwrite with a complete label. Note that it's important to do this 77339c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 77439c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 77539c23413Seschrock * itself. 77639c23413Seschrock */ 77739c23413Seschrock int 77839c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 779fa9e4066Sahrens { 780fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 781fa9e4066Sahrens nvlist_t *label; 782fa9e4066Sahrens vdev_phys_t *vp; 783770499e1SDan Kimmel abd_t *vp_abd; 784770499e1SDan Kimmel abd_t *pad2; 785ecc2d604Sbonwick uberblock_t *ub; 786770499e1SDan Kimmel abd_t *ub_abd; 787fa9e4066Sahrens zio_t *zio; 788fa9e4066Sahrens char *buf; 789fa9e4066Sahrens size_t buflen; 790fa9e4066Sahrens int error; 791fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 792e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 793fa9e4066Sahrens 794e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 7950373e76bSbonwick 796e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 79739c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 79839c23413Seschrock crtxg, reason)) != 0) 799fa9e4066Sahrens return (error); 800fa9e4066Sahrens 80188ecc943SGeorge Wilson /* Track the creation time for this vdev */ 80288ecc943SGeorge Wilson vd->vdev_crtxg = crtxg; 80388ecc943SGeorge Wilson 804973c78e9SGeorge Wilson if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 805fa9e4066Sahrens return (0); 806fa9e4066Sahrens 807fa9e4066Sahrens /* 80839c23413Seschrock * Dead vdevs cannot be initialized. 809fa9e4066Sahrens */ 810fa9e4066Sahrens if (vdev_is_dead(vd)) 811be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 812fa9e4066Sahrens 813fa9e4066Sahrens /* 81439c23413Seschrock * Determine if the vdev is in use. 815fa9e4066Sahrens */ 8161195e687SMark J Musante if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 817fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 818be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 81939c23413Seschrock 82039c23413Seschrock /* 821fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 822fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 823fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 824fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 82539c23413Seschrock */ 826fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 827fa94a07fSbrendan spare_guid != 0ULL) { 82831157203SJeff Bonwick uint64_t guid_delta = spare_guid - vd->vdev_guid; 82999653d4eSeschrock 83031157203SJeff Bonwick vd->vdev_guid += guid_delta; 83131157203SJeff Bonwick 83231157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 83331157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 83439c23413Seschrock 83599653d4eSeschrock /* 83639c23413Seschrock * If this is a replacement, then we want to fallthrough to the 83739c23413Seschrock * rest of the code. If we're adding a spare, then it's already 8383d7072f8Seschrock * labeled appropriately and we can just return. 83999653d4eSeschrock */ 84039c23413Seschrock if (reason == VDEV_LABEL_SPARE) 84139c23413Seschrock return (0); 8421195e687SMark J Musante ASSERT(reason == VDEV_LABEL_REPLACE || 8431195e687SMark J Musante reason == VDEV_LABEL_SPLIT); 844fa9e4066Sahrens } 845fa9e4066Sahrens 846fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 847fa94a07fSbrendan l2cache_guid != 0ULL) { 84831157203SJeff Bonwick uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 84931157203SJeff Bonwick 85031157203SJeff Bonwick vd->vdev_guid += guid_delta; 851fa94a07fSbrendan 85231157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 85331157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 854fa94a07fSbrendan 855fa94a07fSbrendan /* 856fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 857fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 858fa94a07fSbrendan * already labeled appropriately and we can just return. 859fa94a07fSbrendan */ 860fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 861fa94a07fSbrendan return (0); 862fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 863fa94a07fSbrendan } 864fa94a07fSbrendan 865fa9e4066Sahrens /* 86639c23413Seschrock * Initialize its label. 867fa9e4066Sahrens */ 868770499e1SDan Kimmel vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 869770499e1SDan Kimmel abd_zero(vp_abd, sizeof (vdev_phys_t)); 870770499e1SDan Kimmel vp = abd_to_buf(vp_abd); 871fa9e4066Sahrens 872fa9e4066Sahrens /* 873fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 874fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 875fa9e4066Sahrens * really part of an active pool just yet. The labels will 876fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 877fa9e4066Sahrens */ 87839c23413Seschrock if (reason == VDEV_LABEL_SPARE || 87939c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 88039c23413Seschrock /* 88139c23413Seschrock * For inactive hot spares, we generate a special label that 88239c23413Seschrock * identifies as a mutually shared hot spare. We write the 88339c23413Seschrock * label if we are adding a hot spare, or if we are removing an 88439c23413Seschrock * active hot spare (in which case we want to revert the 88539c23413Seschrock * labels). 88639c23413Seschrock */ 88799653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 88899653d4eSeschrock 88999653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 89099653d4eSeschrock spa_version(spa)) == 0); 89199653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 89299653d4eSeschrock POOL_STATE_SPARE) == 0); 89399653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 89499653d4eSeschrock vd->vdev_guid) == 0); 895fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 896fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 897fa94a07fSbrendan /* 898fa94a07fSbrendan * For level 2 ARC devices, add a special label. 899fa94a07fSbrendan */ 900fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 901fa94a07fSbrendan 902fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 903fa94a07fSbrendan spa_version(spa)) == 0); 904fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 905fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 906fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 907fa94a07fSbrendan vd->vdev_guid) == 0); 90899653d4eSeschrock } else { 9091195e687SMark J Musante uint64_t txg = 0ULL; 9101195e687SMark J Musante 9111195e687SMark J Musante if (reason == VDEV_LABEL_SPLIT) 9121195e687SMark J Musante txg = spa->spa_uberblock.ub_txg; 9131195e687SMark J Musante label = spa_config_generate(spa, vd, txg, B_FALSE); 91499653d4eSeschrock 91599653d4eSeschrock /* 91699653d4eSeschrock * Add our creation time. This allows us to detect multiple 91799653d4eSeschrock * vdev uses as described above, and automatically expires if we 91899653d4eSeschrock * fail. 91999653d4eSeschrock */ 92099653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 92199653d4eSeschrock crtxg) == 0); 92299653d4eSeschrock } 923fa9e4066Sahrens 924fa9e4066Sahrens buf = vp->vp_nvlist; 925fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 926fa9e4066Sahrens 927a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 928a75573b6Smmusante if (error != 0) { 929fa9e4066Sahrens nvlist_free(label); 930770499e1SDan Kimmel abd_free(vp_abd); 931a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 932a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 933fa9e4066Sahrens } 934fa9e4066Sahrens 935fa9e4066Sahrens /* 936fa9e4066Sahrens * Initialize uberblock template. 937fa9e4066Sahrens */ 938770499e1SDan Kimmel ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE); 939770499e1SDan Kimmel abd_zero(ub_abd, VDEV_UBERBLOCK_RING); 940770499e1SDan Kimmel abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t)); 941770499e1SDan Kimmel ub = abd_to_buf(ub_abd); 942ecc2d604Sbonwick ub->ub_txg = 0; 943fa9e4066Sahrens 944f83ffe1aSLin Ling /* Initialize the 2nd padding area. */ 945770499e1SDan Kimmel pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE); 946770499e1SDan Kimmel abd_zero(pad2, VDEV_PAD_SIZE); 947f83ffe1aSLin Ling 948fa9e4066Sahrens /* 949fa9e4066Sahrens * Write everything in parallel. 950fa9e4066Sahrens */ 9518956713aSEric Schrock retry: 95217f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 953fa9e4066Sahrens 954e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 955fa9e4066Sahrens 956770499e1SDan Kimmel vdev_label_write(zio, vd, l, vp_abd, 957fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 95817f17c2dSbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 959fa9e4066Sahrens 960f83ffe1aSLin Ling /* 961f83ffe1aSLin Ling * Skip the 1st padding area. 962f83ffe1aSLin Ling * Zero out the 2nd padding area where it might have 963f83ffe1aSLin Ling * left over data from previous filesystem format. 964f83ffe1aSLin Ling */ 965f83ffe1aSLin Ling vdev_label_write(zio, vd, l, pad2, 966f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2), 967f83ffe1aSLin Ling VDEV_PAD_SIZE, NULL, NULL, flags); 968f83ffe1aSLin Ling 969770499e1SDan Kimmel vdev_label_write(zio, vd, l, ub_abd, 970f64c0e34SEric Taylor offsetof(vdev_label_t, vl_uberblock), 971f64c0e34SEric Taylor VDEV_UBERBLOCK_RING, NULL, NULL, flags); 972fa9e4066Sahrens } 973fa9e4066Sahrens 974fa9e4066Sahrens error = zio_wait(zio); 975fa9e4066Sahrens 9768956713aSEric Schrock if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 9778956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 9788956713aSEric Schrock goto retry; 9798956713aSEric Schrock } 9808956713aSEric Schrock 981fa9e4066Sahrens nvlist_free(label); 982770499e1SDan Kimmel abd_free(pad2); 983770499e1SDan Kimmel abd_free(ub_abd); 984770499e1SDan Kimmel abd_free(vp_abd); 985fa9e4066Sahrens 98639c23413Seschrock /* 98739c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 9883d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 989fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 990fa94a07fSbrendan * level 2 ARC devices. 99139c23413Seschrock */ 99239c23413Seschrock if (error == 0 && !vd->vdev_isspare && 99339c23413Seschrock (reason == VDEV_LABEL_SPARE || 99489a89ebfSlling spa_spare_exists(vd->vdev_guid, NULL, NULL))) 99539c23413Seschrock spa_spare_add(vd); 99699653d4eSeschrock 997fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 998fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 999fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 1000fa94a07fSbrendan spa_l2cache_add(vd); 1001fa94a07fSbrendan 100239c23413Seschrock return (error); 100399653d4eSeschrock } 100499653d4eSeschrock 1005fa9e4066Sahrens /* 1006fa9e4066Sahrens * ========================================================================== 1007fa9e4066Sahrens * uberblock load/sync 1008fa9e4066Sahrens * ========================================================================== 1009fa9e4066Sahrens */ 1010fa9e4066Sahrens 1011fa9e4066Sahrens /* 1012fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 1013fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 1014fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 1015fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 1016fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 1017ad135b5dSChristopher Siden * the missing replica comes back, then for a few seconds we'll have two 1018fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 1019fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 1020fa9e4066Sahrens */ 1021fa9e4066Sahrens static int 1022c4ab0d3fSGvozden Neskovic vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2) 1023fa9e4066Sahrens { 1024c4ab0d3fSGvozden Neskovic int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg); 1025c4ab0d3fSGvozden Neskovic if (likely(cmp)) 1026c4ab0d3fSGvozden Neskovic return (cmp); 1027fa9e4066Sahrens 1028c4ab0d3fSGvozden Neskovic return (AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp)); 1029fa9e4066Sahrens } 1030fa9e4066Sahrens 1031ad135b5dSChristopher Siden struct ubl_cbdata { 1032ad135b5dSChristopher Siden uberblock_t *ubl_ubbest; /* Best uberblock */ 1033ad135b5dSChristopher Siden vdev_t *ubl_vd; /* vdev associated with the above */ 1034ad135b5dSChristopher Siden }; 1035ad135b5dSChristopher Siden 1036fa9e4066Sahrens static void 1037fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 1038fa9e4066Sahrens { 1039ad135b5dSChristopher Siden vdev_t *vd = zio->io_vd; 1040468c413aSTim Haley spa_t *spa = zio->io_spa; 1041e14bb325SJeff Bonwick zio_t *rio = zio->io_private; 1042770499e1SDan Kimmel uberblock_t *ub = abd_to_buf(zio->io_abd); 1043ad135b5dSChristopher Siden struct ubl_cbdata *cbp = rio->io_private; 1044fa9e4066Sahrens 1045ad135b5dSChristopher Siden ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 1046fa9e4066Sahrens 1047ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 1048e14bb325SJeff Bonwick mutex_enter(&rio->io_lock); 1049468c413aSTim Haley if (ub->ub_txg <= spa->spa_load_max_txg && 1050ad135b5dSChristopher Siden vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 1051ad135b5dSChristopher Siden /* 1052dfbb9432SGeorge Wilson * Keep track of the vdev in which this uberblock 1053dfbb9432SGeorge Wilson * was found. We will use this information later 1054dfbb9432SGeorge Wilson * to obtain the config nvlist associated with 1055ad135b5dSChristopher Siden * this uberblock. 1056ad135b5dSChristopher Siden */ 1057ad135b5dSChristopher Siden *cbp->ubl_ubbest = *ub; 1058ad135b5dSChristopher Siden cbp->ubl_vd = vd; 1059ad135b5dSChristopher Siden } 1060e14bb325SJeff Bonwick mutex_exit(&rio->io_lock); 1061fa9e4066Sahrens } 1062fa9e4066Sahrens 1063770499e1SDan Kimmel abd_free(zio->io_abd); 1064fa9e4066Sahrens } 1065fa9e4066Sahrens 1066ad135b5dSChristopher Siden static void 1067ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 1068ad135b5dSChristopher Siden struct ubl_cbdata *cbp) 1069fa9e4066Sahrens { 1070e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1071ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 1072fa9e4066Sahrens 1073e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1074e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 1075e14bb325SJeff Bonwick for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 1076e14bb325SJeff Bonwick vdev_label_read(zio, vd, l, 1077770499e1SDan Kimmel abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), 1078770499e1SDan Kimmel B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n), 1079e14bb325SJeff Bonwick VDEV_UBERBLOCK_SIZE(vd), 1080e14bb325SJeff Bonwick vdev_uberblock_load_done, zio, flags); 1081e14bb325SJeff Bonwick } 1082fa9e4066Sahrens } 1083fa9e4066Sahrens } 1084ad135b5dSChristopher Siden } 1085e14bb325SJeff Bonwick 1086ad135b5dSChristopher Siden /* 1087ad135b5dSChristopher Siden * Reads the 'best' uberblock from disk along with its associated 1088ad135b5dSChristopher Siden * configuration. First, we read the uberblock array of each label of each 1089ad135b5dSChristopher Siden * vdev, keeping track of the uberblock with the highest txg in each array. 1090dfbb9432SGeorge Wilson * Then, we read the configuration from the same vdev as the best uberblock. 1091ad135b5dSChristopher Siden */ 1092ad135b5dSChristopher Siden void 1093ad135b5dSChristopher Siden vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 1094ad135b5dSChristopher Siden { 1095ad135b5dSChristopher Siden zio_t *zio; 1096ad135b5dSChristopher Siden spa_t *spa = rvd->vdev_spa; 1097ad135b5dSChristopher Siden struct ubl_cbdata cb; 1098ad135b5dSChristopher Siden int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 1099ad135b5dSChristopher Siden ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 1100ad135b5dSChristopher Siden 1101ad135b5dSChristopher Siden ASSERT(ub); 1102ad135b5dSChristopher Siden ASSERT(config); 1103ad135b5dSChristopher Siden 1104ad135b5dSChristopher Siden bzero(ub, sizeof (uberblock_t)); 1105ad135b5dSChristopher Siden *config = NULL; 1106ad135b5dSChristopher Siden 1107ad135b5dSChristopher Siden cb.ubl_ubbest = ub; 1108ad135b5dSChristopher Siden cb.ubl_vd = NULL; 1109ad135b5dSChristopher Siden 1110ad135b5dSChristopher Siden spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1111ad135b5dSChristopher Siden zio = zio_root(spa, NULL, &cb, flags); 1112ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio, rvd, flags, &cb); 1113ad135b5dSChristopher Siden (void) zio_wait(zio); 1114dfbb9432SGeorge Wilson 1115dfbb9432SGeorge Wilson /* 1116dfbb9432SGeorge Wilson * It's possible that the best uberblock was discovered on a label 1117dfbb9432SGeorge Wilson * that has a configuration which was written in a future txg. 1118dfbb9432SGeorge Wilson * Search all labels on this vdev to find the configuration that 1119dfbb9432SGeorge Wilson * matches the txg for our uberblock. 1120dfbb9432SGeorge Wilson */ 11213ee8c80cSPavel Zakharov if (cb.ubl_vd != NULL) { 11223ee8c80cSPavel Zakharov vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. " 11233ee8c80cSPavel Zakharov "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg); 11243ee8c80cSPavel Zakharov 1125dfbb9432SGeorge Wilson *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 11266f793812SPavel Zakharov if (*config == NULL && spa->spa_extreme_rewind) { 11276f793812SPavel Zakharov vdev_dbgmsg(cb.ubl_vd, "failed to read label config. " 11286f793812SPavel Zakharov "Trying again without txg restrictions."); 11296f793812SPavel Zakharov *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX); 11306f793812SPavel Zakharov } 11313ee8c80cSPavel Zakharov if (*config == NULL) { 11323ee8c80cSPavel Zakharov vdev_dbgmsg(cb.ubl_vd, "failed to read label config"); 11333ee8c80cSPavel Zakharov } 11343ee8c80cSPavel Zakharov } 1135ad135b5dSChristopher Siden spa_config_exit(spa, SCL_ALL, FTAG); 1136fa9e4066Sahrens } 1137fa9e4066Sahrens 1138fa9e4066Sahrens /* 113917f17c2dSbonwick * On success, increment root zio's count of good writes. 11400373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1141fa9e4066Sahrens */ 1142fa9e4066Sahrens static void 1143fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 1144fa9e4066Sahrens { 114517f17c2dSbonwick uint64_t *good_writes = zio->io_private; 1146fa9e4066Sahrens 11470373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 11481a5e258fSJosef 'Jeff' Sipek atomic_inc_64(good_writes); 1149fa9e4066Sahrens } 1150fa9e4066Sahrens 115117f17c2dSbonwick /* 115217f17c2dSbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 115317f17c2dSbonwick */ 1154fa9e4066Sahrens static void 1155a3b55830SMatthew Ahrens vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes, 1156a3b55830SMatthew Ahrens uberblock_t *ub, vdev_t *vd, int flags) 1157fa9e4066Sahrens { 1158a3b55830SMatthew Ahrens for (uint64_t c = 0; c < vd->vdev_children; c++) { 1159a3b55830SMatthew Ahrens vdev_uberblock_sync(zio, good_writes, 1160a3b55830SMatthew Ahrens ub, vd->vdev_child[c], flags); 1161a3b55830SMatthew Ahrens } 1162fa9e4066Sahrens 1163fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 1164fa9e4066Sahrens return; 1165fa9e4066Sahrens 1166e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 1167fa9e4066Sahrens return; 1168fa9e4066Sahrens 1169e0f1c0afSOlaf Faaland int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0; 1170e0f1c0afSOlaf Faaland int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m); 1171fa9e4066Sahrens 1172770499e1SDan Kimmel /* Copy the uberblock_t into the ABD */ 1173770499e1SDan Kimmel abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE); 1174770499e1SDan Kimmel abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd)); 1175770499e1SDan Kimmel abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t)); 1176fa9e4066Sahrens 1177e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) 1178770499e1SDan Kimmel vdev_label_write(zio, vd, l, ub_abd, 1179e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1180a3b55830SMatthew Ahrens vdev_uberblock_sync_done, good_writes, 1181e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 1182fa9e4066Sahrens 1183770499e1SDan Kimmel abd_free(ub_abd); 1184fa9e4066Sahrens } 1185fa9e4066Sahrens 11863e30c24aSWill Andrews /* Sync the uberblocks to all vdevs in svd[] */ 118717f17c2dSbonwick int 118817f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1189fa9e4066Sahrens { 119017f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1191e14bb325SJeff Bonwick zio_t *zio; 119217f17c2dSbonwick uint64_t good_writes = 0; 1193fa9e4066Sahrens 1194a3b55830SMatthew Ahrens zio = zio_root(spa, NULL, NULL, flags); 1195e14bb325SJeff Bonwick 1196e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 1197a3b55830SMatthew Ahrens vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags); 1198fa9e4066Sahrens 119917f17c2dSbonwick (void) zio_wait(zio); 1200fa9e4066Sahrens 1201fa9e4066Sahrens /* 120217f17c2dSbonwick * Flush the uberblocks to disk. This ensures that the odd labels 120317f17c2dSbonwick * are no longer needed (because the new uberblocks and the even 120417f17c2dSbonwick * labels are safely on disk), so it is safe to overwrite them. 1205fa9e4066Sahrens */ 120617f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 1207fa9e4066Sahrens 12085cabbc6bSPrashanth Sreenivasa for (int v = 0; v < svdcount; v++) { 12095cabbc6bSPrashanth Sreenivasa if (vdev_writeable(svd[v])) { 12105cabbc6bSPrashanth Sreenivasa zio_flush(zio, svd[v]); 12115cabbc6bSPrashanth Sreenivasa } 12125cabbc6bSPrashanth Sreenivasa } 1213fa9e4066Sahrens 121417f17c2dSbonwick (void) zio_wait(zio); 121517f17c2dSbonwick 121617f17c2dSbonwick return (good_writes >= 1 ? 0 : EIO); 1217fa9e4066Sahrens } 1218fa9e4066Sahrens 1219fa9e4066Sahrens /* 122017f17c2dSbonwick * On success, increment the count of good writes for our top-level vdev. 1221fa9e4066Sahrens */ 1222fa9e4066Sahrens static void 122317f17c2dSbonwick vdev_label_sync_done(zio_t *zio) 1224fa9e4066Sahrens { 122517f17c2dSbonwick uint64_t *good_writes = zio->io_private; 1226fa9e4066Sahrens 1227fa9e4066Sahrens if (zio->io_error == 0) 12281a5e258fSJosef 'Jeff' Sipek atomic_inc_64(good_writes); 1229fa9e4066Sahrens } 1230fa9e4066Sahrens 123117f17c2dSbonwick /* 123217f17c2dSbonwick * If there weren't enough good writes, indicate failure to the parent. 123317f17c2dSbonwick */ 1234fa9e4066Sahrens static void 123517f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio) 123617f17c2dSbonwick { 123717f17c2dSbonwick uint64_t *good_writes = zio->io_private; 123817f17c2dSbonwick 123917f17c2dSbonwick if (*good_writes == 0) 1240be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(EIO); 124117f17c2dSbonwick 124217f17c2dSbonwick kmem_free(good_writes, sizeof (uint64_t)); 124317f17c2dSbonwick } 124417f17c2dSbonwick 124551ece835Seschrock /* 12460430f8daSeschrock * We ignore errors for log and cache devices, simply free the private data. 124751ece835Seschrock */ 124851ece835Seschrock static void 12490430f8daSeschrock vdev_label_sync_ignore_done(zio_t *zio) 125051ece835Seschrock { 125151ece835Seschrock kmem_free(zio->io_private, sizeof (uint64_t)); 125251ece835Seschrock } 125351ece835Seschrock 125417f17c2dSbonwick /* 125517f17c2dSbonwick * Write all even or odd labels to all leaves of the specified vdev. 125617f17c2dSbonwick */ 125717f17c2dSbonwick static void 1258a3b55830SMatthew Ahrens vdev_label_sync(zio_t *zio, uint64_t *good_writes, 1259a3b55830SMatthew Ahrens vdev_t *vd, int l, uint64_t txg, int flags) 1260fa9e4066Sahrens { 1261fa9e4066Sahrens nvlist_t *label; 1262fa9e4066Sahrens vdev_phys_t *vp; 1263770499e1SDan Kimmel abd_t *vp_abd; 1264fa9e4066Sahrens char *buf; 1265fa9e4066Sahrens size_t buflen; 1266fa9e4066Sahrens 1267a3b55830SMatthew Ahrens for (int c = 0; c < vd->vdev_children; c++) { 1268a3b55830SMatthew Ahrens vdev_label_sync(zio, good_writes, 1269a3b55830SMatthew Ahrens vd->vdev_child[c], l, txg, flags); 1270a3b55830SMatthew Ahrens } 1271fa9e4066Sahrens 1272fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 1273fa9e4066Sahrens return; 1274fa9e4066Sahrens 1275e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 1276fa9e4066Sahrens return; 1277fa9e4066Sahrens 1278fa9e4066Sahrens /* 1279fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 1280fa9e4066Sahrens */ 12810373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1282fa9e4066Sahrens 1283770499e1SDan Kimmel vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE); 1284770499e1SDan Kimmel abd_zero(vp_abd, sizeof (vdev_phys_t)); 1285770499e1SDan Kimmel vp = abd_to_buf(vp_abd); 1286fa9e4066Sahrens 1287fa9e4066Sahrens buf = vp->vp_nvlist; 1288fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 1289fa9e4066Sahrens 129017f17c2dSbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 129117f17c2dSbonwick for (; l < VDEV_LABELS; l += 2) { 1292770499e1SDan Kimmel vdev_label_write(zio, vd, l, vp_abd, 129317f17c2dSbonwick offsetof(vdev_label_t, vl_vdev_phys), 129417f17c2dSbonwick sizeof (vdev_phys_t), 1295a3b55830SMatthew Ahrens vdev_label_sync_done, good_writes, 1296e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 129717f17c2dSbonwick } 129817f17c2dSbonwick } 1299fa9e4066Sahrens 1300770499e1SDan Kimmel abd_free(vp_abd); 1301fa9e4066Sahrens nvlist_free(label); 1302fa9e4066Sahrens } 1303fa9e4066Sahrens 130417f17c2dSbonwick int 1305e14bb325SJeff Bonwick vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1306fa9e4066Sahrens { 1307e14bb325SJeff Bonwick list_t *dl = &spa->spa_config_dirty_list; 130817f17c2dSbonwick vdev_t *vd; 1309e14bb325SJeff Bonwick zio_t *zio; 1310fa9e4066Sahrens int error; 1311fa9e4066Sahrens 1312fa9e4066Sahrens /* 1313e14bb325SJeff Bonwick * Write the new labels to disk. 1314fa9e4066Sahrens */ 1315e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 1316fa9e4066Sahrens 131717f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 131817f17c2dSbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 131917f17c2dSbonwick KM_SLEEP); 132088ecc943SGeorge Wilson 132188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 132288ecc943SGeorge Wilson 1323a3f829aeSBill Moore zio_t *vio = zio_null(zio, spa, NULL, 13240430f8daSeschrock (vd->vdev_islog || vd->vdev_aux != NULL) ? 13250430f8daSeschrock vdev_label_sync_ignore_done : vdev_label_sync_top_done, 132617f17c2dSbonwick good_writes, flags); 1327a3b55830SMatthew Ahrens vdev_label_sync(vio, good_writes, vd, l, txg, flags); 132817f17c2dSbonwick zio_nowait(vio); 1329fa9e4066Sahrens } 1330e14bb325SJeff Bonwick 1331e14bb325SJeff Bonwick error = zio_wait(zio); 1332fa9e4066Sahrens 13338654d025Sperrin /* 133417f17c2dSbonwick * Flush the new labels to disk. 13358654d025Sperrin */ 133617f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 13378654d025Sperrin 133817f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 133917f17c2dSbonwick zio_flush(zio, vd); 134017f17c2dSbonwick 134117f17c2dSbonwick (void) zio_wait(zio); 1342fa9e4066Sahrens 1343fa9e4066Sahrens return (error); 1344fa9e4066Sahrens } 1345fa9e4066Sahrens 1346fa9e4066Sahrens /* 134717f17c2dSbonwick * Sync the uberblock and any changes to the vdev configuration. 1348fa9e4066Sahrens * 1349fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 1350fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 1351fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 1352fa9e4066Sahrens * describe the failure semantics at each stage. 1353fa9e4066Sahrens * 135417f17c2dSbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1355fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 1356fa9e4066Sahrens */ 1357e14bb325SJeff Bonwick int 1358eb5bb584SWill Andrews vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1359fa9e4066Sahrens { 136017f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1361fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 1362eb5bb584SWill Andrews int error = 0; 1363e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1364fa9e4066Sahrens 136586714001SSerapheim Dimitropoulos ASSERT(svdcount != 0); 1366eb5bb584SWill Andrews retry: 13678956713aSEric Schrock /* 13688956713aSEric Schrock * Normally, we don't want to try too hard to write every label and 13698956713aSEric Schrock * uberblock. If there is a flaky disk, we don't want the rest of the 13708956713aSEric Schrock * sync process to block while we retry. But if we can't write a 13718956713aSEric Schrock * single label out, we should retry with ZIO_FLAG_TRYHARD before 13728956713aSEric Schrock * bailing out and declaring the pool faulted. 13738956713aSEric Schrock */ 1374eb5bb584SWill Andrews if (error != 0) { 1375eb5bb584SWill Andrews if ((flags & ZIO_FLAG_TRYHARD) != 0) 1376eb5bb584SWill Andrews return (error); 13778956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 1378eb5bb584SWill Andrews } 13798956713aSEric Schrock 1380fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 1381fa9e4066Sahrens 1382fa9e4066Sahrens /* 138317f17c2dSbonwick * If this isn't a resync due to I/O errors, 138417f17c2dSbonwick * and nothing changed in this transaction group, 138517f17c2dSbonwick * and the vdev configuration hasn't changed, 13860373e76bSbonwick * then there's nothing to do. 1387fa9e4066Sahrens */ 1388e0f1c0afSOlaf Faaland if (ub->ub_txg < txg) { 1389e0f1c0afSOlaf Faaland boolean_t changed = uberblock_update(ub, spa->spa_root_vdev, 1390e0f1c0afSOlaf Faaland txg, spa->spa_mmp.mmp_delay); 1391e0f1c0afSOlaf Faaland 1392e0f1c0afSOlaf Faaland if (!changed && list_is_empty(&spa->spa_config_dirty_list)) 1393e0f1c0afSOlaf Faaland return (0); 1394e0f1c0afSOlaf Faaland } 1395fa9e4066Sahrens 1396fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 1397e14bb325SJeff Bonwick return (0); 1398fa9e4066Sahrens 13990373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 14000373e76bSbonwick 1401fa9e4066Sahrens /* 1402fa9e4066Sahrens * Flush the write cache of every disk that's been written to 1403fa9e4066Sahrens * in this transaction group. This ensures that all blocks 1404fa9e4066Sahrens * written in this txg will be committed to stable storage 1405fa9e4066Sahrens * before any uberblock that references them. 1406fa9e4066Sahrens */ 140786714001SSerapheim Dimitropoulos zio_t *zio = zio_root(spa, NULL, NULL, flags); 140817f17c2dSbonwick 140986714001SSerapheim Dimitropoulos for (vdev_t *vd = 141086714001SSerapheim Dimitropoulos txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL; 141117f17c2dSbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 141217f17c2dSbonwick zio_flush(zio, vd); 141317f17c2dSbonwick 1414fa9e4066Sahrens (void) zio_wait(zio); 1415fa9e4066Sahrens 1416fa9e4066Sahrens /* 1417fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1418fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1419fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1420fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 142117f17c2dSbonwick * which have not yet been touched, will still be valid. We flush 142217f17c2dSbonwick * the new labels to disk to ensure that all even-label updates 142317f17c2dSbonwick * are committed to stable storage before the uberblock update. 1424fa9e4066Sahrens */ 142586714001SSerapheim Dimitropoulos if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) { 142686714001SSerapheim Dimitropoulos if ((flags & ZIO_FLAG_TRYHARD) != 0) { 142786714001SSerapheim Dimitropoulos zfs_dbgmsg("vdev_label_sync_list() returned error %d " 142886714001SSerapheim Dimitropoulos "for pool '%s' when syncing out the even labels " 142986714001SSerapheim Dimitropoulos "of dirty vdevs", error, spa_name(spa)); 143086714001SSerapheim Dimitropoulos } 1431eb5bb584SWill Andrews goto retry; 143286714001SSerapheim Dimitropoulos } 1433fa9e4066Sahrens 1434fa9e4066Sahrens /* 1435e14bb325SJeff Bonwick * Sync the uberblocks to all vdevs in svd[]. 14360373e76bSbonwick * If the system dies in the middle of this step, there are two cases 14370373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1438fa9e4066Sahrens * 1439fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1440fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1441fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1442fa9e4066Sahrens * to that uberblock. 1443fa9e4066Sahrens * 1444fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1445fa9e4066Sahrens * will be the newest, and the even labels (which had all 1446fa9e4066Sahrens * been successfully committed) will be valid with respect 1447fa9e4066Sahrens * to the new uberblocks. 1448fa9e4066Sahrens */ 144986714001SSerapheim Dimitropoulos if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) { 145086714001SSerapheim Dimitropoulos if ((flags & ZIO_FLAG_TRYHARD) != 0) { 145186714001SSerapheim Dimitropoulos zfs_dbgmsg("vdev_uberblock_sync_list() returned error " 145286714001SSerapheim Dimitropoulos "%d for pool '%s'", error, spa_name(spa)); 145386714001SSerapheim Dimitropoulos } 1454eb5bb584SWill Andrews goto retry; 145586714001SSerapheim Dimitropoulos } 1456fa9e4066Sahrens 1457e0f1c0afSOlaf Faaland if (spa_multihost(spa)) 1458e0f1c0afSOlaf Faaland mmp_update_uberblock(spa, ub); 1459e0f1c0afSOlaf Faaland 1460fa9e4066Sahrens /* 1461fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1462fa9e4066Sahrens * in the middle of this process, the even labels and the new 1463fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1464fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1465fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 146617f17c2dSbonwick * all labels will be brought up to date. We flush the new labels 146717f17c2dSbonwick * to disk to ensure that all odd-label updates are committed to 146817f17c2dSbonwick * stable storage before the next transaction group begins. 1469fa9e4066Sahrens */ 147086714001SSerapheim Dimitropoulos if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) { 147186714001SSerapheim Dimitropoulos if ((flags & ZIO_FLAG_TRYHARD) != 0) { 147286714001SSerapheim Dimitropoulos zfs_dbgmsg("vdev_label_sync_list() returned error %d " 147386714001SSerapheim Dimitropoulos "for pool '%s' when syncing out the odd labels of " 147486714001SSerapheim Dimitropoulos "dirty vdevs", error, spa_name(spa)); 147586714001SSerapheim Dimitropoulos } 1476eb5bb584SWill Andrews goto retry; 147786714001SSerapheim Dimitropoulos } 1478eb5bb584SWill Andrews 1479eb5bb584SWill Andrews return (0); 1480fa9e4066Sahrens } 1481