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 */ 21fa9e4066Sahrens /* 22*3f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23fa9e4066Sahrens */ 24fa9e4066Sahrens 25fa9e4066Sahrens /* 26fa9e4066Sahrens * Virtual Device Labels 27fa9e4066Sahrens * --------------------- 28fa9e4066Sahrens * 29fa9e4066Sahrens * The vdev label serves several distinct purposes: 30fa9e4066Sahrens * 31fa9e4066Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 32fa9e4066Sahrens * identity within the pool. 33fa9e4066Sahrens * 34fa9e4066Sahrens * 2. Verify that all the devices given in a configuration are present 35fa9e4066Sahrens * within the pool. 36fa9e4066Sahrens * 37fa9e4066Sahrens * 3. Determine the uberblock for the pool. 38fa9e4066Sahrens * 39fa9e4066Sahrens * 4. In case of an import operation, determine the configuration of the 40fa9e4066Sahrens * toplevel vdev of which it is a part. 41fa9e4066Sahrens * 42fa9e4066Sahrens * 5. If an import operation cannot find all the devices in the pool, 43fa9e4066Sahrens * provide enough information to the administrator to determine which 44fa9e4066Sahrens * devices are missing. 45fa9e4066Sahrens * 46fa9e4066Sahrens * It is important to note that while the kernel is responsible for writing the 47fa9e4066Sahrens * label, it only consumes the information in the first three cases. The 48fa9e4066Sahrens * latter information is only consumed in userland when determining the 49fa9e4066Sahrens * configuration to import a pool. 50fa9e4066Sahrens * 51fa9e4066Sahrens * 52fa9e4066Sahrens * Label Organization 53fa9e4066Sahrens * ------------------ 54fa9e4066Sahrens * 55fa9e4066Sahrens * Before describing the contents of the label, it's important to understand how 56fa9e4066Sahrens * the labels are written and updated with respect to the uberblock. 57fa9e4066Sahrens * 58fa9e4066Sahrens * When the pool configuration is altered, either because it was newly created 59fa9e4066Sahrens * or a device was added, we want to update all the labels such that we can deal 60fa9e4066Sahrens * with fatal failure at any point. To this end, each disk has two labels which 61fa9e4066Sahrens * are updated before and after the uberblock is synced. Assuming we have 623d7072f8Seschrock * labels and an uberblock with the following transaction groups: 63fa9e4066Sahrens * 64fa9e4066Sahrens * L1 UB L2 65fa9e4066Sahrens * +------+ +------+ +------+ 66fa9e4066Sahrens * | | | | | | 67fa9e4066Sahrens * | t10 | | t10 | | t10 | 68fa9e4066Sahrens * | | | | | | 69fa9e4066Sahrens * +------+ +------+ +------+ 70fa9e4066Sahrens * 71fa9e4066Sahrens * In this stable state, the labels and the uberblock were all updated within 72fa9e4066Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 73fa9e4066Sahrens * that we can detect when we fail partway through writing the label. 74fa9e4066Sahrens * 75fa9e4066Sahrens * In order to identify which labels are valid, the labels are written in the 76fa9e4066Sahrens * following manner: 77fa9e4066Sahrens * 78fa9e4066Sahrens * 1. For each vdev, update 'L1' to the new label 79fa9e4066Sahrens * 2. Update the uberblock 80fa9e4066Sahrens * 3. For each vdev, update 'L2' to the new label 81fa9e4066Sahrens * 82fa9e4066Sahrens * Given arbitrary failure, we can determine the correct label to use based on 83fa9e4066Sahrens * the transaction group. If we fail after updating L1 but before updating the 84fa9e4066Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 85fa9e4066Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 86fa9e4066Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 87fa9e4066Sahrens * therefore L1 is valid. 88fa9e4066Sahrens * 89fa9e4066Sahrens * Another added complexity is that not every label is updated when the config 90fa9e4066Sahrens * is synced. If we add a single device, we do not want to have to re-write 91fa9e4066Sahrens * every label for every device in the pool. This means that both L1 and L2 may 92fa9e4066Sahrens * be older than the pool uberblock, because the necessary information is stored 93fa9e4066Sahrens * on another vdev. 94fa9e4066Sahrens * 95fa9e4066Sahrens * 96fa9e4066Sahrens * On-disk Format 97fa9e4066Sahrens * -------------- 98fa9e4066Sahrens * 99fa9e4066Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 100fa9e4066Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 101fa9e4066Sahrens * VTOC disk labels, but is otherwise ignored. 102fa9e4066Sahrens * 103fa9e4066Sahrens * The first half of the label is a packed nvlist which contains pool wide 104fa9e4066Sahrens * properties, per-vdev properties, and configuration information. It is 105fa9e4066Sahrens * described in more detail below. 106fa9e4066Sahrens * 107fa9e4066Sahrens * The latter half of the label consists of a redundant array of uberblocks. 108fa9e4066Sahrens * These uberblocks are updated whenever a transaction group is committed, 109fa9e4066Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 110fa9e4066Sahrens * vdev for the 'best' uberblock. 111fa9e4066Sahrens * 112fa9e4066Sahrens * 113fa9e4066Sahrens * Configuration Information 114fa9e4066Sahrens * ------------------------- 115fa9e4066Sahrens * 116fa9e4066Sahrens * The nvlist describing the pool and vdev contains the following elements: 117fa9e4066Sahrens * 118fa9e4066Sahrens * version ZFS on-disk version 119fa9e4066Sahrens * name Pool name 120fa9e4066Sahrens * state Pool state 121fa9e4066Sahrens * txg Transaction group in which this label was written 122fa9e4066Sahrens * pool_guid Unique identifier for this pool 123fa9e4066Sahrens * vdev_tree An nvlist describing vdev tree. 124fa9e4066Sahrens * 125fa9e4066Sahrens * Each leaf device label also contains the following: 126fa9e4066Sahrens * 127fa9e4066Sahrens * top_guid Unique ID for top-level vdev in which this is contained 128fa9e4066Sahrens * guid Unique ID for the leaf vdev 129fa9e4066Sahrens * 130fa9e4066Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 131fa9e4066Sahrens */ 132fa9e4066Sahrens 133fa9e4066Sahrens #include <sys/zfs_context.h> 134fa9e4066Sahrens #include <sys/spa.h> 135fa9e4066Sahrens #include <sys/spa_impl.h> 136fa9e4066Sahrens #include <sys/dmu.h> 137fa9e4066Sahrens #include <sys/zap.h> 138fa9e4066Sahrens #include <sys/vdev.h> 139fa9e4066Sahrens #include <sys/vdev_impl.h> 140fa9e4066Sahrens #include <sys/uberblock_impl.h> 141fa9e4066Sahrens #include <sys/metaslab.h> 142fa9e4066Sahrens #include <sys/zio.h> 143*3f9d6ad7SLin Ling #include <sys/dsl_scan.h> 144fa9e4066Sahrens #include <sys/fs/zfs.h> 145fa9e4066Sahrens 146fa9e4066Sahrens /* 147fa9e4066Sahrens * Basic routines to read and write from a vdev label. 148fa9e4066Sahrens * Used throughout the rest of this file. 149fa9e4066Sahrens */ 150fa9e4066Sahrens uint64_t 151fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 152fa9e4066Sahrens { 153ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 154e7437265Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 155ecc2d604Sbonwick 156fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 157fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 158fa9e4066Sahrens } 159fa9e4066Sahrens 16021bf64a7Sgw /* 16121bf64a7Sgw * Returns back the vdev label associated with the passed in offset. 16221bf64a7Sgw */ 16321bf64a7Sgw int 16421bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset) 16521bf64a7Sgw { 16621bf64a7Sgw int l; 16721bf64a7Sgw 16821bf64a7Sgw if (offset >= psize - VDEV_LABEL_END_SIZE) { 16921bf64a7Sgw offset -= psize - VDEV_LABEL_END_SIZE; 17021bf64a7Sgw offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 17121bf64a7Sgw } 17221bf64a7Sgw l = offset / sizeof (vdev_label_t); 17321bf64a7Sgw return (l < VDEV_LABELS ? l : -1); 17421bf64a7Sgw } 17521bf64a7Sgw 176fa9e4066Sahrens static void 177fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 178e14bb325SJeff Bonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 179fa9e4066Sahrens { 180e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 181e14bb325SJeff Bonwick SCL_STATE_ALL); 182e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 183fa9e4066Sahrens 184fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 185fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 186fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 187e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 188fa9e4066Sahrens } 189fa9e4066Sahrens 190fa9e4066Sahrens static void 191fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 19217f17c2dSbonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 193fa9e4066Sahrens { 194e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 195e14bb325SJeff Bonwick (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 196e14bb325SJeff Bonwick (SCL_CONFIG | SCL_STATE) && 197e14bb325SJeff Bonwick dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 198e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 199fa9e4066Sahrens 200fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 201fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 202fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 20317f17c2dSbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 204fa9e4066Sahrens } 205fa9e4066Sahrens 206fa9e4066Sahrens /* 207fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 208fa9e4066Sahrens */ 209fa9e4066Sahrens nvlist_t * 21099653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 211*3f9d6ad7SLin Ling vdev_config_flag_t flags) 212fa9e4066Sahrens { 213fa9e4066Sahrens nvlist_t *nv = NULL; 214fa9e4066Sahrens 215ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 216fa9e4066Sahrens 217fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 218fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 219*3f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 22099653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 22199653d4eSeschrock == 0); 222fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 223fa9e4066Sahrens 224fa9e4066Sahrens if (vd->vdev_path != NULL) 225fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 226fa9e4066Sahrens vd->vdev_path) == 0); 227fa9e4066Sahrens 228fa9e4066Sahrens if (vd->vdev_devid != NULL) 229fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 230fa9e4066Sahrens vd->vdev_devid) == 0); 231fa9e4066Sahrens 2323d7072f8Seschrock if (vd->vdev_physpath != NULL) 2333d7072f8Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 2343d7072f8Seschrock vd->vdev_physpath) == 0); 2353d7072f8Seschrock 2366809eb4eSEric Schrock if (vd->vdev_fru != NULL) 2376809eb4eSEric Schrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, 2386809eb4eSEric Schrock vd->vdev_fru) == 0); 2396809eb4eSEric Schrock 24099653d4eSeschrock if (vd->vdev_nparity != 0) { 24199653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 24299653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 24399653d4eSeschrock 24499653d4eSeschrock /* 24599653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 24699653d4eSeschrock * into a crufty old storage pool. 24799653d4eSeschrock */ 24899653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 249f94275ceSAdam Leventhal (vd->vdev_nparity <= 2 && 250f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ2) || 251f94275ceSAdam Leventhal (vd->vdev_nparity <= 3 && 252f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ3)); 25399653d4eSeschrock 25499653d4eSeschrock /* 25599653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 25699653d4eSeschrock * that only support a single parity device -- older software 25799653d4eSeschrock * will just ignore it. 25899653d4eSeschrock */ 25999653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 26099653d4eSeschrock vd->vdev_nparity) == 0); 26199653d4eSeschrock } 26299653d4eSeschrock 263afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 264afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 265afefbcddSeschrock vd->vdev_wholedisk) == 0); 266afefbcddSeschrock 267ea8dc4b6Seschrock if (vd->vdev_not_present) 268ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 269ea8dc4b6Seschrock 27099653d4eSeschrock if (vd->vdev_isspare) 27199653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 27299653d4eSeschrock 273*3f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 274*3f9d6ad7SLin Ling vd == vd->vdev_top) { 275fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 276fa9e4066Sahrens vd->vdev_ms_array) == 0); 277fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 278fa9e4066Sahrens vd->vdev_ms_shift) == 0); 279fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 280fa9e4066Sahrens vd->vdev_ashift) == 0); 281fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 282fa9e4066Sahrens vd->vdev_asize) == 0); 2838654d025Sperrin VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 2848654d025Sperrin vd->vdev_islog) == 0); 285*3f9d6ad7SLin Ling if (vd->vdev_removing) 286*3f9d6ad7SLin Ling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 287*3f9d6ad7SLin Ling vd->vdev_removing) == 0); 288fa9e4066Sahrens } 289fa9e4066Sahrens 2908ad4d6ddSJeff Bonwick if (vd->vdev_dtl_smo.smo_object != 0) 291fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 2928ad4d6ddSJeff Bonwick vd->vdev_dtl_smo.smo_object) == 0); 293fa9e4066Sahrens 29488ecc943SGeorge Wilson if (vd->vdev_crtxg) 29588ecc943SGeorge Wilson VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 29688ecc943SGeorge Wilson vd->vdev_crtxg) == 0); 29788ecc943SGeorge Wilson 298fa9e4066Sahrens if (getstats) { 299fa9e4066Sahrens vdev_stat_t vs; 300*3f9d6ad7SLin Ling pool_scan_stat_t ps; 301*3f9d6ad7SLin Ling 302fa9e4066Sahrens vdev_get_stats(vd, &vs); 303*3f9d6ad7SLin Ling VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 304fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 305*3f9d6ad7SLin Ling 306*3f9d6ad7SLin Ling /* provide either current or previous scan information */ 307*3f9d6ad7SLin Ling if (spa_scan_get_stats(spa, &ps) == 0) { 308*3f9d6ad7SLin Ling VERIFY(nvlist_add_uint64_array(nv, 309*3f9d6ad7SLin Ling ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 310*3f9d6ad7SLin Ling sizeof (pool_scan_stat_t) / sizeof (uint64_t)) 311*3f9d6ad7SLin Ling == 0); 312*3f9d6ad7SLin Ling } 313fa9e4066Sahrens } 314fa9e4066Sahrens 315fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 316fa9e4066Sahrens nvlist_t **child; 317*3f9d6ad7SLin Ling int c, idx; 318fa9e4066Sahrens 31988ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 32088ecc943SGeorge Wilson 321fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 322fa9e4066Sahrens KM_SLEEP); 323fa9e4066Sahrens 324*3f9d6ad7SLin Ling for (c = 0, idx = 0; c < vd->vdev_children; c++) { 325*3f9d6ad7SLin Ling vdev_t *cvd = vd->vdev_child[c]; 326*3f9d6ad7SLin Ling 327*3f9d6ad7SLin Ling /* 328*3f9d6ad7SLin Ling * If we're generating an nvlist of removing 329*3f9d6ad7SLin Ling * vdevs then skip over any device which is 330*3f9d6ad7SLin Ling * not being removed. 331*3f9d6ad7SLin Ling */ 332*3f9d6ad7SLin Ling if ((flags & VDEV_CONFIG_REMOVING) && 333*3f9d6ad7SLin Ling !cvd->vdev_removing) 334*3f9d6ad7SLin Ling continue; 335fa9e4066Sahrens 336*3f9d6ad7SLin Ling child[idx++] = vdev_config_generate(spa, cvd, 337*3f9d6ad7SLin Ling getstats, flags); 338*3f9d6ad7SLin Ling } 339*3f9d6ad7SLin Ling 340*3f9d6ad7SLin Ling if (idx) { 341*3f9d6ad7SLin Ling VERIFY(nvlist_add_nvlist_array(nv, 342*3f9d6ad7SLin Ling ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 343*3f9d6ad7SLin Ling } 344fa9e4066Sahrens 345*3f9d6ad7SLin Ling for (c = 0; c < idx; c++) 346fa9e4066Sahrens nvlist_free(child[c]); 347fa9e4066Sahrens 348fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 349441d80aaSlling 350441d80aaSlling } else { 351069f55e2SEric Schrock const char *aux = NULL; 352069f55e2SEric Schrock 353ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 354441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 355ecc2d604Sbonwick B_TRUE) == 0); 3563d7072f8Seschrock if (vd->vdev_faulted) 3573d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 3583d7072f8Seschrock B_TRUE) == 0); 3593d7072f8Seschrock if (vd->vdev_degraded) 3603d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 3613d7072f8Seschrock B_TRUE) == 0); 3623d7072f8Seschrock if (vd->vdev_removed) 3633d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 3643d7072f8Seschrock B_TRUE) == 0); 3653d7072f8Seschrock if (vd->vdev_unspare) 3663d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 3673d7072f8Seschrock B_TRUE) == 0); 36888ecc943SGeorge Wilson if (vd->vdev_ishole) 36988ecc943SGeorge Wilson VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, 37088ecc943SGeorge Wilson B_TRUE) == 0); 371069f55e2SEric Schrock 372069f55e2SEric Schrock switch (vd->vdev_stat.vs_aux) { 373069f55e2SEric Schrock case VDEV_AUX_ERR_EXCEEDED: 374069f55e2SEric Schrock aux = "err_exceeded"; 375069f55e2SEric Schrock break; 376069f55e2SEric Schrock 377069f55e2SEric Schrock case VDEV_AUX_EXTERNAL: 378069f55e2SEric Schrock aux = "external"; 379069f55e2SEric Schrock break; 380069f55e2SEric Schrock } 381069f55e2SEric Schrock 382069f55e2SEric Schrock if (aux != NULL) 383069f55e2SEric Schrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, 384069f55e2SEric Schrock aux) == 0); 3851195e687SMark J Musante 3861195e687SMark J Musante if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 3871195e687SMark J Musante VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 3881195e687SMark J Musante vd->vdev_orig_guid) == 0); 3891195e687SMark J Musante } 390fa9e4066Sahrens } 391fa9e4066Sahrens 392fa9e4066Sahrens return (nv); 393fa9e4066Sahrens } 394fa9e4066Sahrens 39588ecc943SGeorge Wilson /* 39688ecc943SGeorge Wilson * Generate a view of the top-level vdevs. If we currently have holes 39788ecc943SGeorge Wilson * in the namespace, then generate an array which contains a list of holey 39888ecc943SGeorge Wilson * vdevs. Additionally, add the number of top-level children that currently 39988ecc943SGeorge Wilson * exist. 40088ecc943SGeorge Wilson */ 40188ecc943SGeorge Wilson void 40288ecc943SGeorge Wilson vdev_top_config_generate(spa_t *spa, nvlist_t *config) 40388ecc943SGeorge Wilson { 40488ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 40588ecc943SGeorge Wilson uint64_t *array; 406*3f9d6ad7SLin Ling uint_t c, idx; 40788ecc943SGeorge Wilson 40888ecc943SGeorge Wilson array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 40988ecc943SGeorge Wilson 410*3f9d6ad7SLin Ling for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 41188ecc943SGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 41288ecc943SGeorge Wilson 41388ecc943SGeorge Wilson if (tvd->vdev_ishole) 41488ecc943SGeorge Wilson array[idx++] = c; 41588ecc943SGeorge Wilson } 41688ecc943SGeorge Wilson 417312c6e15SGeorge Wilson if (idx) { 418312c6e15SGeorge Wilson VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 419312c6e15SGeorge Wilson array, idx) == 0); 420312c6e15SGeorge Wilson } 421312c6e15SGeorge Wilson 42288ecc943SGeorge Wilson VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 42388ecc943SGeorge Wilson rvd->vdev_children) == 0); 42488ecc943SGeorge Wilson 42588ecc943SGeorge Wilson kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 42688ecc943SGeorge Wilson } 42788ecc943SGeorge Wilson 428fa9e4066Sahrens nvlist_t * 429fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 430fa9e4066Sahrens { 4310373e76bSbonwick spa_t *spa = vd->vdev_spa; 432fa9e4066Sahrens nvlist_t *config = NULL; 433fa9e4066Sahrens vdev_phys_t *vp; 434fa9e4066Sahrens zio_t *zio; 4358956713aSEric Schrock int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 4368956713aSEric Schrock ZIO_FLAG_SPECULATIVE; 437fa9e4066Sahrens 438e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 4390373e76bSbonwick 4400a4e9518Sgw if (!vdev_readable(vd)) 441fa9e4066Sahrens return (NULL); 442fa9e4066Sahrens 443fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 444fa9e4066Sahrens 4458956713aSEric Schrock retry: 446e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 447fa9e4066Sahrens 448e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 449fa9e4066Sahrens 450fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 451fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 452e14bb325SJeff Bonwick sizeof (vdev_phys_t), NULL, NULL, flags); 453fa9e4066Sahrens 454fa9e4066Sahrens if (zio_wait(zio) == 0 && 455fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 456ea8dc4b6Seschrock &config, 0) == 0) 457fa9e4066Sahrens break; 458fa9e4066Sahrens 459fa9e4066Sahrens if (config != NULL) { 460fa9e4066Sahrens nvlist_free(config); 461fa9e4066Sahrens config = NULL; 462fa9e4066Sahrens } 463fa9e4066Sahrens } 464fa9e4066Sahrens 4658956713aSEric Schrock if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 4668956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 4678956713aSEric Schrock goto retry; 4688956713aSEric Schrock } 4698956713aSEric Schrock 470fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 471fa9e4066Sahrens 472fa9e4066Sahrens return (config); 473fa9e4066Sahrens } 474fa9e4066Sahrens 47539c23413Seschrock /* 47639c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 47739c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 47839c23413Seschrock */ 47939c23413Seschrock static boolean_t 48039c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 481fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 48239c23413Seschrock { 48339c23413Seschrock spa_t *spa = vd->vdev_spa; 48439c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 48539c23413Seschrock uint64_t vdtxg = 0; 48639c23413Seschrock nvlist_t *label; 48739c23413Seschrock 48839c23413Seschrock if (spare_guid) 48939c23413Seschrock *spare_guid = 0ULL; 490fa94a07fSbrendan if (l2cache_guid) 491fa94a07fSbrendan *l2cache_guid = 0ULL; 49239c23413Seschrock 49339c23413Seschrock /* 49439c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 49539c23413Seschrock */ 49639c23413Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 49739c23413Seschrock return (B_FALSE); 49839c23413Seschrock 49939c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 50039c23413Seschrock &vdtxg); 50139c23413Seschrock 50239c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 50339c23413Seschrock &state) != 0 || 50439c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 50539c23413Seschrock &device_guid) != 0) { 50639c23413Seschrock nvlist_free(label); 50739c23413Seschrock return (B_FALSE); 50839c23413Seschrock } 50939c23413Seschrock 510fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 51139c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 51239c23413Seschrock &pool_guid) != 0 || 51339c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 51439c23413Seschrock &txg) != 0)) { 51539c23413Seschrock nvlist_free(label); 51639c23413Seschrock return (B_FALSE); 51739c23413Seschrock } 51839c23413Seschrock 51939c23413Seschrock nvlist_free(label); 52039c23413Seschrock 52139c23413Seschrock /* 52239c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 52339c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 52439c23413Seschrock * spare (which we check for later on). 52539c23413Seschrock */ 526fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 52739c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 52889a89ebfSlling !spa_spare_exists(device_guid, NULL, NULL) && 529fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 53039c23413Seschrock return (B_FALSE); 53139c23413Seschrock 53239c23413Seschrock /* 53339c23413Seschrock * If the transaction group is zero, then this an initialized (but 53439c23413Seschrock * unused) label. This is only an error if the create transaction 53539c23413Seschrock * on-disk is the same as the one we're using now, in which case the 53639c23413Seschrock * user has attempted to add the same vdev multiple times in the same 53739c23413Seschrock * transaction. 53839c23413Seschrock */ 539fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 540fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 54139c23413Seschrock return (B_TRUE); 54239c23413Seschrock 54339c23413Seschrock /* 54439c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 54539c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 546fa94a07fSbrendan * to add. We also check if it is an l2cache device. 54739c23413Seschrock */ 54889a89ebfSlling if (spa_spare_exists(device_guid, &spare_pool, NULL) || 54939c23413Seschrock spa_has_spare(spa, device_guid)) { 55039c23413Seschrock if (spare_guid) 55139c23413Seschrock *spare_guid = device_guid; 55239c23413Seschrock 55339c23413Seschrock switch (reason) { 55439c23413Seschrock case VDEV_LABEL_CREATE: 555fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 55639c23413Seschrock return (B_TRUE); 55739c23413Seschrock 55839c23413Seschrock case VDEV_LABEL_REPLACE: 55939c23413Seschrock return (!spa_has_spare(spa, device_guid) || 56039c23413Seschrock spare_pool != 0ULL); 56139c23413Seschrock 56239c23413Seschrock case VDEV_LABEL_SPARE: 56339c23413Seschrock return (spa_has_spare(spa, device_guid)); 56439c23413Seschrock } 56539c23413Seschrock } 56639c23413Seschrock 567fa94a07fSbrendan /* 568fa94a07fSbrendan * Check to see if this is an l2cache device. 569fa94a07fSbrendan */ 570fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 571fa94a07fSbrendan return (B_TRUE); 572fa94a07fSbrendan 57339c23413Seschrock /* 57439c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 57539c23413Seschrock * pool on the system. 57639c23413Seschrock */ 57739c23413Seschrock return (state == POOL_STATE_ACTIVE); 57839c23413Seschrock } 57939c23413Seschrock 58039c23413Seschrock /* 58139c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 58239c23413Seschrock * use, and writable. We put down an initial label which we will later 58339c23413Seschrock * overwrite with a complete label. Note that it's important to do this 58439c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 58539c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 58639c23413Seschrock * itself. 58739c23413Seschrock */ 58839c23413Seschrock int 58939c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 590fa9e4066Sahrens { 591fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 592fa9e4066Sahrens nvlist_t *label; 593fa9e4066Sahrens vdev_phys_t *vp; 594f83ffe1aSLin Ling char *pad2; 595ecc2d604Sbonwick uberblock_t *ub; 596fa9e4066Sahrens zio_t *zio; 597fa9e4066Sahrens char *buf; 598fa9e4066Sahrens size_t buflen; 599fa9e4066Sahrens int error; 600fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 601e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 602fa9e4066Sahrens 603e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 6040373e76bSbonwick 605e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 60639c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 60739c23413Seschrock crtxg, reason)) != 0) 608fa9e4066Sahrens return (error); 609fa9e4066Sahrens 61088ecc943SGeorge Wilson /* Track the creation time for this vdev */ 61188ecc943SGeorge Wilson vd->vdev_crtxg = crtxg; 61288ecc943SGeorge Wilson 613fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 614fa9e4066Sahrens return (0); 615fa9e4066Sahrens 616fa9e4066Sahrens /* 61739c23413Seschrock * Dead vdevs cannot be initialized. 618fa9e4066Sahrens */ 619fa9e4066Sahrens if (vdev_is_dead(vd)) 620fa9e4066Sahrens return (EIO); 621fa9e4066Sahrens 622fa9e4066Sahrens /* 62339c23413Seschrock * Determine if the vdev is in use. 624fa9e4066Sahrens */ 6251195e687SMark J Musante if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 626fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 62739c23413Seschrock return (EBUSY); 62839c23413Seschrock 62939c23413Seschrock /* 630fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 631fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 632fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 633fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 63439c23413Seschrock */ 635fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 636fa94a07fSbrendan spare_guid != 0ULL) { 63731157203SJeff Bonwick uint64_t guid_delta = spare_guid - vd->vdev_guid; 63899653d4eSeschrock 63931157203SJeff Bonwick vd->vdev_guid += guid_delta; 64031157203SJeff Bonwick 64131157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 64231157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 64339c23413Seschrock 64499653d4eSeschrock /* 64539c23413Seschrock * If this is a replacement, then we want to fallthrough to the 64639c23413Seschrock * rest of the code. If we're adding a spare, then it's already 6473d7072f8Seschrock * labeled appropriately and we can just return. 64899653d4eSeschrock */ 64939c23413Seschrock if (reason == VDEV_LABEL_SPARE) 65039c23413Seschrock return (0); 6511195e687SMark J Musante ASSERT(reason == VDEV_LABEL_REPLACE || 6521195e687SMark J Musante reason == VDEV_LABEL_SPLIT); 653fa9e4066Sahrens } 654fa9e4066Sahrens 655fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 656fa94a07fSbrendan l2cache_guid != 0ULL) { 65731157203SJeff Bonwick uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 65831157203SJeff Bonwick 65931157203SJeff Bonwick vd->vdev_guid += guid_delta; 660fa94a07fSbrendan 66131157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 66231157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 663fa94a07fSbrendan 664fa94a07fSbrendan /* 665fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 666fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 667fa94a07fSbrendan * already labeled appropriately and we can just return. 668fa94a07fSbrendan */ 669fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 670fa94a07fSbrendan return (0); 671fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 672fa94a07fSbrendan } 673fa94a07fSbrendan 674fa9e4066Sahrens /* 67539c23413Seschrock * Initialize its label. 676fa9e4066Sahrens */ 677fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 678fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 679fa9e4066Sahrens 680fa9e4066Sahrens /* 681fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 682fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 683fa9e4066Sahrens * really part of an active pool just yet. The labels will 684fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 685fa9e4066Sahrens */ 68639c23413Seschrock if (reason == VDEV_LABEL_SPARE || 68739c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 68839c23413Seschrock /* 68939c23413Seschrock * For inactive hot spares, we generate a special label that 69039c23413Seschrock * identifies as a mutually shared hot spare. We write the 69139c23413Seschrock * label if we are adding a hot spare, or if we are removing an 69239c23413Seschrock * active hot spare (in which case we want to revert the 69339c23413Seschrock * labels). 69439c23413Seschrock */ 69599653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 69699653d4eSeschrock 69799653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 69899653d4eSeschrock spa_version(spa)) == 0); 69999653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 70099653d4eSeschrock POOL_STATE_SPARE) == 0); 70199653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 70299653d4eSeschrock vd->vdev_guid) == 0); 703fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 704fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 705fa94a07fSbrendan /* 706fa94a07fSbrendan * For level 2 ARC devices, add a special label. 707fa94a07fSbrendan */ 708fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 709fa94a07fSbrendan 710fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 711fa94a07fSbrendan spa_version(spa)) == 0); 712fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 713fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 714fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 715fa94a07fSbrendan vd->vdev_guid) == 0); 71699653d4eSeschrock } else { 7171195e687SMark J Musante uint64_t txg = 0ULL; 7181195e687SMark J Musante 7191195e687SMark J Musante if (reason == VDEV_LABEL_SPLIT) 7201195e687SMark J Musante txg = spa->spa_uberblock.ub_txg; 7211195e687SMark J Musante label = spa_config_generate(spa, vd, txg, B_FALSE); 72299653d4eSeschrock 72399653d4eSeschrock /* 72499653d4eSeschrock * Add our creation time. This allows us to detect multiple 72599653d4eSeschrock * vdev uses as described above, and automatically expires if we 72699653d4eSeschrock * fail. 72799653d4eSeschrock */ 72899653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 72999653d4eSeschrock crtxg) == 0); 73099653d4eSeschrock } 731fa9e4066Sahrens 732fa9e4066Sahrens buf = vp->vp_nvlist; 733fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 734fa9e4066Sahrens 735a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 736a75573b6Smmusante if (error != 0) { 737fa9e4066Sahrens nvlist_free(label); 738fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 739a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 740a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 741fa9e4066Sahrens } 742fa9e4066Sahrens 743fa9e4066Sahrens /* 744fa9e4066Sahrens * Initialize uberblock template. 745fa9e4066Sahrens */ 746f64c0e34SEric Taylor ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 747f64c0e34SEric Taylor bzero(ub, VDEV_UBERBLOCK_RING); 748ecc2d604Sbonwick *ub = spa->spa_uberblock; 749ecc2d604Sbonwick ub->ub_txg = 0; 750fa9e4066Sahrens 751f83ffe1aSLin Ling /* Initialize the 2nd padding area. */ 752f83ffe1aSLin Ling pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 753f83ffe1aSLin Ling bzero(pad2, VDEV_PAD_SIZE); 754f83ffe1aSLin Ling 755fa9e4066Sahrens /* 756fa9e4066Sahrens * Write everything in parallel. 757fa9e4066Sahrens */ 7588956713aSEric Schrock retry: 75917f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 760fa9e4066Sahrens 761e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 762fa9e4066Sahrens 763fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 764fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 76517f17c2dSbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 766fa9e4066Sahrens 767f83ffe1aSLin Ling /* 768f83ffe1aSLin Ling * Skip the 1st padding area. 769f83ffe1aSLin Ling * Zero out the 2nd padding area where it might have 770f83ffe1aSLin Ling * left over data from previous filesystem format. 771f83ffe1aSLin Ling */ 772f83ffe1aSLin Ling vdev_label_write(zio, vd, l, pad2, 773f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2), 774f83ffe1aSLin Ling VDEV_PAD_SIZE, NULL, NULL, flags); 775f83ffe1aSLin Ling 776f64c0e34SEric Taylor vdev_label_write(zio, vd, l, ub, 777f64c0e34SEric Taylor offsetof(vdev_label_t, vl_uberblock), 778f64c0e34SEric Taylor VDEV_UBERBLOCK_RING, NULL, NULL, flags); 779fa9e4066Sahrens } 780fa9e4066Sahrens 781fa9e4066Sahrens error = zio_wait(zio); 782fa9e4066Sahrens 7838956713aSEric Schrock if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 7848956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 7858956713aSEric Schrock goto retry; 7868956713aSEric Schrock } 7878956713aSEric Schrock 788fa9e4066Sahrens nvlist_free(label); 789f83ffe1aSLin Ling zio_buf_free(pad2, VDEV_PAD_SIZE); 790f64c0e34SEric Taylor zio_buf_free(ub, VDEV_UBERBLOCK_RING); 791fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 792fa9e4066Sahrens 79339c23413Seschrock /* 79439c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 7953d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 796fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 797fa94a07fSbrendan * level 2 ARC devices. 79839c23413Seschrock */ 79939c23413Seschrock if (error == 0 && !vd->vdev_isspare && 80039c23413Seschrock (reason == VDEV_LABEL_SPARE || 80189a89ebfSlling spa_spare_exists(vd->vdev_guid, NULL, NULL))) 80239c23413Seschrock spa_spare_add(vd); 80399653d4eSeschrock 804fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 805fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 806fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 807fa94a07fSbrendan spa_l2cache_add(vd); 808fa94a07fSbrendan 80939c23413Seschrock return (error); 81099653d4eSeschrock } 81199653d4eSeschrock 812fa9e4066Sahrens /* 813fa9e4066Sahrens * ========================================================================== 814fa9e4066Sahrens * uberblock load/sync 815fa9e4066Sahrens * ========================================================================== 816fa9e4066Sahrens */ 817fa9e4066Sahrens 818fa9e4066Sahrens /* 819fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 820fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 821fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 822fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 823fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 824fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 825fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 826fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 827fa9e4066Sahrens */ 828fa9e4066Sahrens static int 829fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 830fa9e4066Sahrens { 831fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 832fa9e4066Sahrens return (-1); 833fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 834fa9e4066Sahrens return (1); 835fa9e4066Sahrens 836fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 837fa9e4066Sahrens return (-1); 838fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 839fa9e4066Sahrens return (1); 840fa9e4066Sahrens 841fa9e4066Sahrens return (0); 842fa9e4066Sahrens } 843fa9e4066Sahrens 844fa9e4066Sahrens static void 845fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 846fa9e4066Sahrens { 847468c413aSTim Haley spa_t *spa = zio->io_spa; 848e14bb325SJeff Bonwick zio_t *rio = zio->io_private; 849ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 850e14bb325SJeff Bonwick uberblock_t *ubbest = rio->io_private; 851fa9e4066Sahrens 852ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 853fa9e4066Sahrens 854ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 855e14bb325SJeff Bonwick mutex_enter(&rio->io_lock); 856468c413aSTim Haley if (ub->ub_txg <= spa->spa_load_max_txg && 8572e551927SVictor Latushkin vdev_uberblock_compare(ub, ubbest) > 0) 858fa9e4066Sahrens *ubbest = *ub; 859e14bb325SJeff Bonwick mutex_exit(&rio->io_lock); 860fa9e4066Sahrens } 861fa9e4066Sahrens 862fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 863fa9e4066Sahrens } 864fa9e4066Sahrens 865fa9e4066Sahrens void 866fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 867fa9e4066Sahrens { 868e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 869e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 8708956713aSEric Schrock int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 8718956713aSEric Schrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 872e14bb325SJeff Bonwick 873e14bb325SJeff Bonwick if (vd == rvd) { 874e14bb325SJeff Bonwick ASSERT(zio == NULL); 875e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 876e14bb325SJeff Bonwick zio = zio_root(spa, NULL, ubbest, flags); 877e14bb325SJeff Bonwick bzero(ubbest, sizeof (uberblock_t)); 878e14bb325SJeff Bonwick } 879fa9e4066Sahrens 880e14bb325SJeff Bonwick ASSERT(zio != NULL); 881fa9e4066Sahrens 882e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 883e14bb325SJeff Bonwick vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 884fa9e4066Sahrens 885e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 886e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 887e14bb325SJeff Bonwick for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 888e14bb325SJeff Bonwick vdev_label_read(zio, vd, l, 889e14bb325SJeff Bonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 890e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), 891e14bb325SJeff Bonwick VDEV_UBERBLOCK_SIZE(vd), 892e14bb325SJeff Bonwick vdev_uberblock_load_done, zio, flags); 893e14bb325SJeff Bonwick } 894fa9e4066Sahrens } 895fa9e4066Sahrens } 896e14bb325SJeff Bonwick 897e14bb325SJeff Bonwick if (vd == rvd) { 898e14bb325SJeff Bonwick (void) zio_wait(zio); 899e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 900e14bb325SJeff Bonwick } 901fa9e4066Sahrens } 902fa9e4066Sahrens 903fa9e4066Sahrens /* 90417f17c2dSbonwick * On success, increment root zio's count of good writes. 9050373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 906fa9e4066Sahrens */ 907fa9e4066Sahrens static void 908fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 909fa9e4066Sahrens { 91017f17c2dSbonwick uint64_t *good_writes = zio->io_private; 911fa9e4066Sahrens 9120373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 913fa9e4066Sahrens atomic_add_64(good_writes, 1); 914fa9e4066Sahrens } 915fa9e4066Sahrens 91617f17c2dSbonwick /* 91717f17c2dSbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 91817f17c2dSbonwick */ 919fa9e4066Sahrens static void 920e14bb325SJeff Bonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 921fa9e4066Sahrens { 92217f17c2dSbonwick uberblock_t *ubbuf; 923e14bb325SJeff Bonwick int n; 924fa9e4066Sahrens 925e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 926e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 927fa9e4066Sahrens 928fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 929fa9e4066Sahrens return; 930fa9e4066Sahrens 931e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 932fa9e4066Sahrens return; 933fa9e4066Sahrens 93417f17c2dSbonwick n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 935fa9e4066Sahrens 93617f17c2dSbonwick ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 93717f17c2dSbonwick bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 93817f17c2dSbonwick *ubbuf = *ub; 939fa9e4066Sahrens 940e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) 94117f17c2dSbonwick vdev_label_write(zio, vd, l, ubbuf, 942e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 94317f17c2dSbonwick vdev_uberblock_sync_done, zio->io_private, 944e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 945fa9e4066Sahrens 94617f17c2dSbonwick zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 947fa9e4066Sahrens } 948fa9e4066Sahrens 94917f17c2dSbonwick int 95017f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 951fa9e4066Sahrens { 95217f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 953e14bb325SJeff Bonwick zio_t *zio; 95417f17c2dSbonwick uint64_t good_writes = 0; 955fa9e4066Sahrens 956e14bb325SJeff Bonwick zio = zio_root(spa, NULL, &good_writes, flags); 957e14bb325SJeff Bonwick 958e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 959e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, svd[v], flags); 960fa9e4066Sahrens 96117f17c2dSbonwick (void) zio_wait(zio); 962fa9e4066Sahrens 963fa9e4066Sahrens /* 96417f17c2dSbonwick * Flush the uberblocks to disk. This ensures that the odd labels 96517f17c2dSbonwick * are no longer needed (because the new uberblocks and the even 96617f17c2dSbonwick * labels are safely on disk), so it is safe to overwrite them. 967fa9e4066Sahrens */ 96817f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 969fa9e4066Sahrens 970e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 97117f17c2dSbonwick zio_flush(zio, svd[v]); 972fa9e4066Sahrens 97317f17c2dSbonwick (void) zio_wait(zio); 97417f17c2dSbonwick 97517f17c2dSbonwick return (good_writes >= 1 ? 0 : EIO); 976fa9e4066Sahrens } 977fa9e4066Sahrens 978fa9e4066Sahrens /* 97917f17c2dSbonwick * On success, increment the count of good writes for our top-level vdev. 980fa9e4066Sahrens */ 981fa9e4066Sahrens static void 98217f17c2dSbonwick vdev_label_sync_done(zio_t *zio) 983fa9e4066Sahrens { 98417f17c2dSbonwick uint64_t *good_writes = zio->io_private; 985fa9e4066Sahrens 986fa9e4066Sahrens if (zio->io_error == 0) 987fa9e4066Sahrens atomic_add_64(good_writes, 1); 988fa9e4066Sahrens } 989fa9e4066Sahrens 99017f17c2dSbonwick /* 99117f17c2dSbonwick * If there weren't enough good writes, indicate failure to the parent. 99217f17c2dSbonwick */ 993fa9e4066Sahrens static void 99417f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio) 99517f17c2dSbonwick { 99617f17c2dSbonwick uint64_t *good_writes = zio->io_private; 99717f17c2dSbonwick 99817f17c2dSbonwick if (*good_writes == 0) 99917f17c2dSbonwick zio->io_error = EIO; 100017f17c2dSbonwick 100117f17c2dSbonwick kmem_free(good_writes, sizeof (uint64_t)); 100217f17c2dSbonwick } 100317f17c2dSbonwick 100451ece835Seschrock /* 10050430f8daSeschrock * We ignore errors for log and cache devices, simply free the private data. 100651ece835Seschrock */ 100751ece835Seschrock static void 10080430f8daSeschrock vdev_label_sync_ignore_done(zio_t *zio) 100951ece835Seschrock { 101051ece835Seschrock kmem_free(zio->io_private, sizeof (uint64_t)); 101151ece835Seschrock } 101251ece835Seschrock 101317f17c2dSbonwick /* 101417f17c2dSbonwick * Write all even or odd labels to all leaves of the specified vdev. 101517f17c2dSbonwick */ 101617f17c2dSbonwick static void 1017e14bb325SJeff Bonwick vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1018fa9e4066Sahrens { 1019fa9e4066Sahrens nvlist_t *label; 1020fa9e4066Sahrens vdev_phys_t *vp; 1021fa9e4066Sahrens char *buf; 1022fa9e4066Sahrens size_t buflen; 1023fa9e4066Sahrens 1024e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1025e14bb325SJeff Bonwick vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1026fa9e4066Sahrens 1027fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 1028fa9e4066Sahrens return; 1029fa9e4066Sahrens 1030e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 1031fa9e4066Sahrens return; 1032fa9e4066Sahrens 1033fa9e4066Sahrens /* 1034fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 1035fa9e4066Sahrens */ 10360373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1037fa9e4066Sahrens 1038fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1039fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 1040fa9e4066Sahrens 1041fa9e4066Sahrens buf = vp->vp_nvlist; 1042fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 1043fa9e4066Sahrens 104417f17c2dSbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 104517f17c2dSbonwick for (; l < VDEV_LABELS; l += 2) { 104617f17c2dSbonwick vdev_label_write(zio, vd, l, vp, 104717f17c2dSbonwick offsetof(vdev_label_t, vl_vdev_phys), 104817f17c2dSbonwick sizeof (vdev_phys_t), 104917f17c2dSbonwick vdev_label_sync_done, zio->io_private, 1050e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 105117f17c2dSbonwick } 105217f17c2dSbonwick } 1053fa9e4066Sahrens 1054fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 1055fa9e4066Sahrens nvlist_free(label); 1056fa9e4066Sahrens } 1057fa9e4066Sahrens 105817f17c2dSbonwick int 1059e14bb325SJeff Bonwick vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1060fa9e4066Sahrens { 1061e14bb325SJeff Bonwick list_t *dl = &spa->spa_config_dirty_list; 106217f17c2dSbonwick vdev_t *vd; 1063e14bb325SJeff Bonwick zio_t *zio; 1064fa9e4066Sahrens int error; 1065fa9e4066Sahrens 1066fa9e4066Sahrens /* 1067e14bb325SJeff Bonwick * Write the new labels to disk. 1068fa9e4066Sahrens */ 1069e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 1070fa9e4066Sahrens 107117f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 107217f17c2dSbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 107317f17c2dSbonwick KM_SLEEP); 107488ecc943SGeorge Wilson 107588ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 107688ecc943SGeorge Wilson 1077a3f829aeSBill Moore zio_t *vio = zio_null(zio, spa, NULL, 10780430f8daSeschrock (vd->vdev_islog || vd->vdev_aux != NULL) ? 10790430f8daSeschrock vdev_label_sync_ignore_done : vdev_label_sync_top_done, 108017f17c2dSbonwick good_writes, flags); 1081e14bb325SJeff Bonwick vdev_label_sync(vio, vd, l, txg, flags); 108217f17c2dSbonwick zio_nowait(vio); 1083fa9e4066Sahrens } 1084e14bb325SJeff Bonwick 1085e14bb325SJeff Bonwick error = zio_wait(zio); 1086fa9e4066Sahrens 10878654d025Sperrin /* 108817f17c2dSbonwick * Flush the new labels to disk. 10898654d025Sperrin */ 109017f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 10918654d025Sperrin 109217f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 109317f17c2dSbonwick zio_flush(zio, vd); 109417f17c2dSbonwick 109517f17c2dSbonwick (void) zio_wait(zio); 1096fa9e4066Sahrens 1097fa9e4066Sahrens return (error); 1098fa9e4066Sahrens } 1099fa9e4066Sahrens 1100fa9e4066Sahrens /* 110117f17c2dSbonwick * Sync the uberblock and any changes to the vdev configuration. 1102fa9e4066Sahrens * 1103fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 1104fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 1105fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 1106fa9e4066Sahrens * describe the failure semantics at each stage. 1107fa9e4066Sahrens * 110817f17c2dSbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1109fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 1110fa9e4066Sahrens */ 1111e14bb325SJeff Bonwick int 11128956713aSEric Schrock vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1113fa9e4066Sahrens { 111417f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1115fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 11160373e76bSbonwick vdev_t *vd; 1117fa9e4066Sahrens zio_t *zio; 1118e14bb325SJeff Bonwick int error; 1119e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1120fa9e4066Sahrens 11218956713aSEric Schrock /* 11228956713aSEric Schrock * Normally, we don't want to try too hard to write every label and 11238956713aSEric Schrock * uberblock. If there is a flaky disk, we don't want the rest of the 11248956713aSEric Schrock * sync process to block while we retry. But if we can't write a 11258956713aSEric Schrock * single label out, we should retry with ZIO_FLAG_TRYHARD before 11268956713aSEric Schrock * bailing out and declaring the pool faulted. 11278956713aSEric Schrock */ 11288956713aSEric Schrock if (tryhard) 11298956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 11308956713aSEric Schrock 1131fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 1132fa9e4066Sahrens 1133fa9e4066Sahrens /* 113417f17c2dSbonwick * If this isn't a resync due to I/O errors, 113517f17c2dSbonwick * and nothing changed in this transaction group, 113617f17c2dSbonwick * and the vdev configuration hasn't changed, 11370373e76bSbonwick * then there's nothing to do. 1138fa9e4066Sahrens */ 113917f17c2dSbonwick if (ub->ub_txg < txg && 114017f17c2dSbonwick uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1141e14bb325SJeff Bonwick list_is_empty(&spa->spa_config_dirty_list)) 1142e14bb325SJeff Bonwick return (0); 1143fa9e4066Sahrens 1144fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 1145e14bb325SJeff Bonwick return (0); 1146fa9e4066Sahrens 11470373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 11480373e76bSbonwick 1149fa9e4066Sahrens /* 1150fa9e4066Sahrens * Flush the write cache of every disk that's been written to 1151fa9e4066Sahrens * in this transaction group. This ensures that all blocks 1152fa9e4066Sahrens * written in this txg will be committed to stable storage 1153fa9e4066Sahrens * before any uberblock that references them. 1154fa9e4066Sahrens */ 115517f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 115617f17c2dSbonwick 1157fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 115817f17c2dSbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 115917f17c2dSbonwick zio_flush(zio, vd); 116017f17c2dSbonwick 1161fa9e4066Sahrens (void) zio_wait(zio); 1162fa9e4066Sahrens 1163fa9e4066Sahrens /* 1164fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1165fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1166fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1167fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 116817f17c2dSbonwick * which have not yet been touched, will still be valid. We flush 116917f17c2dSbonwick * the new labels to disk to ensure that all even-label updates 117017f17c2dSbonwick * are committed to stable storage before the uberblock update. 1171fa9e4066Sahrens */ 1172e14bb325SJeff Bonwick if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1173e14bb325SJeff Bonwick return (error); 1174fa9e4066Sahrens 1175fa9e4066Sahrens /* 1176e14bb325SJeff Bonwick * Sync the uberblocks to all vdevs in svd[]. 11770373e76bSbonwick * If the system dies in the middle of this step, there are two cases 11780373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1179fa9e4066Sahrens * 1180fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1181fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1182fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1183fa9e4066Sahrens * to that uberblock. 1184fa9e4066Sahrens * 1185fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1186fa9e4066Sahrens * will be the newest, and the even labels (which had all 1187fa9e4066Sahrens * been successfully committed) will be valid with respect 1188fa9e4066Sahrens * to the new uberblocks. 1189fa9e4066Sahrens */ 1190e14bb325SJeff Bonwick if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1191e14bb325SJeff Bonwick return (error); 1192fa9e4066Sahrens 1193fa9e4066Sahrens /* 1194fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1195fa9e4066Sahrens * in the middle of this process, the even labels and the new 1196fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1197fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1198fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 119917f17c2dSbonwick * all labels will be brought up to date. We flush the new labels 120017f17c2dSbonwick * to disk to ensure that all odd-label updates are committed to 120117f17c2dSbonwick * stable storage before the next transaction group begins. 1202fa9e4066Sahrens */ 1203e14bb325SJeff Bonwick return (vdev_label_sync_list(spa, 1, txg, flags)); 1204fa9e4066Sahrens } 1205