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. 24be6fd75aSMatthew Ahrens * Copyright (c) 2013 by Delphix. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens /* 28fa9e4066Sahrens * Virtual Device Labels 29fa9e4066Sahrens * --------------------- 30fa9e4066Sahrens * 31fa9e4066Sahrens * The vdev label serves several distinct purposes: 32fa9e4066Sahrens * 33fa9e4066Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 34fa9e4066Sahrens * identity within the pool. 35fa9e4066Sahrens * 36fa9e4066Sahrens * 2. Verify that all the devices given in a configuration are present 37fa9e4066Sahrens * within the pool. 38fa9e4066Sahrens * 39fa9e4066Sahrens * 3. Determine the uberblock for the pool. 40fa9e4066Sahrens * 41fa9e4066Sahrens * 4. In case of an import operation, determine the configuration of the 42fa9e4066Sahrens * toplevel vdev of which it is a part. 43fa9e4066Sahrens * 44fa9e4066Sahrens * 5. If an import operation cannot find all the devices in the pool, 45fa9e4066Sahrens * provide enough information to the administrator to determine which 46fa9e4066Sahrens * devices are missing. 47fa9e4066Sahrens * 48fa9e4066Sahrens * It is important to note that while the kernel is responsible for writing the 49fa9e4066Sahrens * label, it only consumes the information in the first three cases. The 50fa9e4066Sahrens * latter information is only consumed in userland when determining the 51fa9e4066Sahrens * configuration to import a pool. 52fa9e4066Sahrens * 53fa9e4066Sahrens * 54fa9e4066Sahrens * Label Organization 55fa9e4066Sahrens * ------------------ 56fa9e4066Sahrens * 57fa9e4066Sahrens * Before describing the contents of the label, it's important to understand how 58fa9e4066Sahrens * the labels are written and updated with respect to the uberblock. 59fa9e4066Sahrens * 60fa9e4066Sahrens * When the pool configuration is altered, either because it was newly created 61fa9e4066Sahrens * or a device was added, we want to update all the labels such that we can deal 62fa9e4066Sahrens * with fatal failure at any point. To this end, each disk has two labels which 63fa9e4066Sahrens * are updated before and after the uberblock is synced. Assuming we have 643d7072f8Seschrock * labels and an uberblock with the following transaction groups: 65fa9e4066Sahrens * 66fa9e4066Sahrens * L1 UB L2 67fa9e4066Sahrens * +------+ +------+ +------+ 68fa9e4066Sahrens * | | | | | | 69fa9e4066Sahrens * | t10 | | t10 | | t10 | 70fa9e4066Sahrens * | | | | | | 71fa9e4066Sahrens * +------+ +------+ +------+ 72fa9e4066Sahrens * 73fa9e4066Sahrens * In this stable state, the labels and the uberblock were all updated within 74fa9e4066Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 75fa9e4066Sahrens * that we can detect when we fail partway through writing the label. 76fa9e4066Sahrens * 77fa9e4066Sahrens * In order to identify which labels are valid, the labels are written in the 78fa9e4066Sahrens * following manner: 79fa9e4066Sahrens * 80fa9e4066Sahrens * 1. For each vdev, update 'L1' to the new label 81fa9e4066Sahrens * 2. Update the uberblock 82fa9e4066Sahrens * 3. For each vdev, update 'L2' to the new label 83fa9e4066Sahrens * 84fa9e4066Sahrens * Given arbitrary failure, we can determine the correct label to use based on 85fa9e4066Sahrens * the transaction group. If we fail after updating L1 but before updating the 86fa9e4066Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 87fa9e4066Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 88fa9e4066Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 89fa9e4066Sahrens * therefore L1 is valid. 90fa9e4066Sahrens * 91fa9e4066Sahrens * Another added complexity is that not every label is updated when the config 92fa9e4066Sahrens * is synced. If we add a single device, we do not want to have to re-write 93fa9e4066Sahrens * every label for every device in the pool. This means that both L1 and L2 may 94fa9e4066Sahrens * be older than the pool uberblock, because the necessary information is stored 95fa9e4066Sahrens * on another vdev. 96fa9e4066Sahrens * 97fa9e4066Sahrens * 98fa9e4066Sahrens * On-disk Format 99fa9e4066Sahrens * -------------- 100fa9e4066Sahrens * 101fa9e4066Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 102fa9e4066Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 103fa9e4066Sahrens * VTOC disk labels, but is otherwise ignored. 104fa9e4066Sahrens * 105fa9e4066Sahrens * The first half of the label is a packed nvlist which contains pool wide 106fa9e4066Sahrens * properties, per-vdev properties, and configuration information. It is 107fa9e4066Sahrens * described in more detail below. 108fa9e4066Sahrens * 109fa9e4066Sahrens * The latter half of the label consists of a redundant array of uberblocks. 110fa9e4066Sahrens * These uberblocks are updated whenever a transaction group is committed, 111fa9e4066Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 112fa9e4066Sahrens * vdev for the 'best' uberblock. 113fa9e4066Sahrens * 114fa9e4066Sahrens * 115fa9e4066Sahrens * Configuration Information 116fa9e4066Sahrens * ------------------------- 117fa9e4066Sahrens * 118fa9e4066Sahrens * The nvlist describing the pool and vdev contains the following elements: 119fa9e4066Sahrens * 120fa9e4066Sahrens * version ZFS on-disk version 121fa9e4066Sahrens * name Pool name 122fa9e4066Sahrens * state Pool state 123fa9e4066Sahrens * txg Transaction group in which this label was written 124fa9e4066Sahrens * pool_guid Unique identifier for this pool 125fa9e4066Sahrens * vdev_tree An nvlist describing vdev tree. 126ad135b5dSChristopher Siden * features_for_read 127ad135b5dSChristopher Siden * An nvlist of the features necessary for reading the MOS. 128fa9e4066Sahrens * 129fa9e4066Sahrens * Each leaf device label also contains the following: 130fa9e4066Sahrens * 131fa9e4066Sahrens * top_guid Unique ID for top-level vdev in which this is contained 132fa9e4066Sahrens * guid Unique ID for the leaf vdev 133fa9e4066Sahrens * 134fa9e4066Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 135fa9e4066Sahrens */ 136fa9e4066Sahrens 137fa9e4066Sahrens #include <sys/zfs_context.h> 138fa9e4066Sahrens #include <sys/spa.h> 139fa9e4066Sahrens #include <sys/spa_impl.h> 140fa9e4066Sahrens #include <sys/dmu.h> 141fa9e4066Sahrens #include <sys/zap.h> 142fa9e4066Sahrens #include <sys/vdev.h> 143fa9e4066Sahrens #include <sys/vdev_impl.h> 144fa9e4066Sahrens #include <sys/uberblock_impl.h> 145fa9e4066Sahrens #include <sys/metaslab.h> 146fa9e4066Sahrens #include <sys/zio.h> 1473f9d6ad7SLin Ling #include <sys/dsl_scan.h> 148fa9e4066Sahrens #include <sys/fs/zfs.h> 149fa9e4066Sahrens 150fa9e4066Sahrens /* 151fa9e4066Sahrens * Basic routines to read and write from a vdev label. 152fa9e4066Sahrens * Used throughout the rest of this file. 153fa9e4066Sahrens */ 154fa9e4066Sahrens uint64_t 155fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 156fa9e4066Sahrens { 157ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 158e7437265Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 159ecc2d604Sbonwick 160fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 161fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 162fa9e4066Sahrens } 163fa9e4066Sahrens 16421bf64a7Sgw /* 16521bf64a7Sgw * Returns back the vdev label associated with the passed in offset. 16621bf64a7Sgw */ 16721bf64a7Sgw int 16821bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset) 16921bf64a7Sgw { 17021bf64a7Sgw int l; 17121bf64a7Sgw 17221bf64a7Sgw if (offset >= psize - VDEV_LABEL_END_SIZE) { 17321bf64a7Sgw offset -= psize - VDEV_LABEL_END_SIZE; 17421bf64a7Sgw offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 17521bf64a7Sgw } 17621bf64a7Sgw l = offset / sizeof (vdev_label_t); 17721bf64a7Sgw return (l < VDEV_LABELS ? l : -1); 17821bf64a7Sgw } 17921bf64a7Sgw 180fa9e4066Sahrens static void 181fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 182e14bb325SJeff Bonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 183fa9e4066Sahrens { 184e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 185e14bb325SJeff Bonwick SCL_STATE_ALL); 186e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 187fa9e4066Sahrens 188fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 189fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 190fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 191e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 192fa9e4066Sahrens } 193fa9e4066Sahrens 194fa9e4066Sahrens static void 195fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 19617f17c2dSbonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 197fa9e4066Sahrens { 198e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 199e14bb325SJeff Bonwick (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 200e14bb325SJeff Bonwick (SCL_CONFIG | SCL_STATE) && 201e14bb325SJeff Bonwick dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 202e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 203fa9e4066Sahrens 204fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 205fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 206fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 20717f17c2dSbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 208fa9e4066Sahrens } 209fa9e4066Sahrens 210fa9e4066Sahrens /* 211fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 212fa9e4066Sahrens */ 213fa9e4066Sahrens nvlist_t * 21499653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 2153f9d6ad7SLin Ling vdev_config_flag_t flags) 216fa9e4066Sahrens { 217fa9e4066Sahrens nvlist_t *nv = NULL; 218fa9e4066Sahrens 219*b4952e17SGeorge Wilson nv = fnvlist_alloc(); 220fa9e4066Sahrens 221*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 2223f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 223*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 224*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 225fa9e4066Sahrens 226fa9e4066Sahrens if (vd->vdev_path != NULL) 227*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 228fa9e4066Sahrens 229fa9e4066Sahrens if (vd->vdev_devid != NULL) 230*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 231fa9e4066Sahrens 2323d7072f8Seschrock if (vd->vdev_physpath != NULL) 233*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 234*b4952e17SGeorge Wilson vd->vdev_physpath); 2353d7072f8Seschrock 2366809eb4eSEric Schrock if (vd->vdev_fru != NULL) 237*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 2386809eb4eSEric Schrock 23999653d4eSeschrock if (vd->vdev_nparity != 0) { 24099653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 24199653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 24299653d4eSeschrock 24399653d4eSeschrock /* 24499653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 24599653d4eSeschrock * into a crufty old storage pool. 24699653d4eSeschrock */ 24799653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 248f94275ceSAdam Leventhal (vd->vdev_nparity <= 2 && 249f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ2) || 250f94275ceSAdam Leventhal (vd->vdev_nparity <= 3 && 251f94275ceSAdam Leventhal spa_version(spa) >= SPA_VERSION_RAIDZ3)); 25299653d4eSeschrock 25399653d4eSeschrock /* 25499653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 25599653d4eSeschrock * that only support a single parity device -- older software 25699653d4eSeschrock * will just ignore it. 25799653d4eSeschrock */ 258*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 25999653d4eSeschrock } 26099653d4eSeschrock 261afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 262*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 263*b4952e17SGeorge Wilson vd->vdev_wholedisk); 264afefbcddSeschrock 265ea8dc4b6Seschrock if (vd->vdev_not_present) 266*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 267ea8dc4b6Seschrock 26899653d4eSeschrock if (vd->vdev_isspare) 269*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 27099653d4eSeschrock 2713f9d6ad7SLin Ling if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 2723f9d6ad7SLin Ling vd == vd->vdev_top) { 273*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 274*b4952e17SGeorge Wilson vd->vdev_ms_array); 275*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 276*b4952e17SGeorge Wilson vd->vdev_ms_shift); 277*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 278*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 279*b4952e17SGeorge Wilson vd->vdev_asize); 280*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 2813f9d6ad7SLin Ling if (vd->vdev_removing) 282*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 283*b4952e17SGeorge Wilson vd->vdev_removing); 284fa9e4066Sahrens } 285fa9e4066Sahrens 2868ad4d6ddSJeff Bonwick if (vd->vdev_dtl_smo.smo_object != 0) 287*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 288*b4952e17SGeorge Wilson vd->vdev_dtl_smo.smo_object); 289fa9e4066Sahrens 29088ecc943SGeorge Wilson if (vd->vdev_crtxg) 291*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 29288ecc943SGeorge Wilson 293fa9e4066Sahrens if (getstats) { 294fa9e4066Sahrens vdev_stat_t vs; 2953f9d6ad7SLin Ling pool_scan_stat_t ps; 2963f9d6ad7SLin Ling 297fa9e4066Sahrens vdev_get_stats(vd, &vs); 298*b4952e17SGeorge Wilson fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 299*b4952e17SGeorge Wilson (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 3003f9d6ad7SLin Ling 3013f9d6ad7SLin Ling /* provide either current or previous scan information */ 3023f9d6ad7SLin Ling if (spa_scan_get_stats(spa, &ps) == 0) { 303*b4952e17SGeorge Wilson fnvlist_add_uint64_array(nv, 3043f9d6ad7SLin Ling ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 305*b4952e17SGeorge Wilson sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 3063f9d6ad7SLin Ling } 307fa9e4066Sahrens } 308fa9e4066Sahrens 309fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 310fa9e4066Sahrens nvlist_t **child; 3113f9d6ad7SLin Ling int c, idx; 312fa9e4066Sahrens 31388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 31488ecc943SGeorge Wilson 315fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 316fa9e4066Sahrens KM_SLEEP); 317fa9e4066Sahrens 3183f9d6ad7SLin Ling for (c = 0, idx = 0; c < vd->vdev_children; c++) { 3193f9d6ad7SLin Ling vdev_t *cvd = vd->vdev_child[c]; 3203f9d6ad7SLin Ling 3213f9d6ad7SLin Ling /* 3223f9d6ad7SLin Ling * If we're generating an nvlist of removing 3233f9d6ad7SLin Ling * vdevs then skip over any device which is 3243f9d6ad7SLin Ling * not being removed. 3253f9d6ad7SLin Ling */ 3263f9d6ad7SLin Ling if ((flags & VDEV_CONFIG_REMOVING) && 3273f9d6ad7SLin Ling !cvd->vdev_removing) 3283f9d6ad7SLin Ling continue; 329fa9e4066Sahrens 3303f9d6ad7SLin Ling child[idx++] = vdev_config_generate(spa, cvd, 3313f9d6ad7SLin Ling getstats, flags); 3323f9d6ad7SLin Ling } 3333f9d6ad7SLin Ling 3343f9d6ad7SLin Ling if (idx) { 335*b4952e17SGeorge Wilson fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 336*b4952e17SGeorge Wilson child, idx); 3373f9d6ad7SLin Ling } 338fa9e4066Sahrens 3393f9d6ad7SLin Ling for (c = 0; c < idx; c++) 340fa9e4066Sahrens nvlist_free(child[c]); 341fa9e4066Sahrens 342fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 343441d80aaSlling 344441d80aaSlling } else { 345069f55e2SEric Schrock const char *aux = NULL; 346069f55e2SEric Schrock 347ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 348*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 349*b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0) 350*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 351*b4952e17SGeorge Wilson vd->vdev_resilver_txg); 3523d7072f8Seschrock if (vd->vdev_faulted) 353*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 3543d7072f8Seschrock if (vd->vdev_degraded) 355*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 3563d7072f8Seschrock if (vd->vdev_removed) 357*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 3583d7072f8Seschrock if (vd->vdev_unspare) 359*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 36088ecc943SGeorge Wilson if (vd->vdev_ishole) 361*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 362069f55e2SEric Schrock 363069f55e2SEric Schrock switch (vd->vdev_stat.vs_aux) { 364069f55e2SEric Schrock case VDEV_AUX_ERR_EXCEEDED: 365069f55e2SEric Schrock aux = "err_exceeded"; 366069f55e2SEric Schrock break; 367069f55e2SEric Schrock 368069f55e2SEric Schrock case VDEV_AUX_EXTERNAL: 369069f55e2SEric Schrock aux = "external"; 370069f55e2SEric Schrock break; 371069f55e2SEric Schrock } 372069f55e2SEric Schrock 373069f55e2SEric Schrock if (aux != NULL) 374*b4952e17SGeorge Wilson fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 3751195e687SMark J Musante 3761195e687SMark J Musante if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 377*b4952e17SGeorge Wilson fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 378*b4952e17SGeorge Wilson vd->vdev_orig_guid); 3791195e687SMark J Musante } 380fa9e4066Sahrens } 381fa9e4066Sahrens 382fa9e4066Sahrens return (nv); 383fa9e4066Sahrens } 384fa9e4066Sahrens 38588ecc943SGeorge Wilson /* 38688ecc943SGeorge Wilson * Generate a view of the top-level vdevs. If we currently have holes 38788ecc943SGeorge Wilson * in the namespace, then generate an array which contains a list of holey 38888ecc943SGeorge Wilson * vdevs. Additionally, add the number of top-level children that currently 38988ecc943SGeorge Wilson * exist. 39088ecc943SGeorge Wilson */ 39188ecc943SGeorge Wilson void 39288ecc943SGeorge Wilson vdev_top_config_generate(spa_t *spa, nvlist_t *config) 39388ecc943SGeorge Wilson { 39488ecc943SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 39588ecc943SGeorge Wilson uint64_t *array; 3963f9d6ad7SLin Ling uint_t c, idx; 39788ecc943SGeorge Wilson 39888ecc943SGeorge Wilson array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 39988ecc943SGeorge Wilson 4003f9d6ad7SLin Ling for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 40188ecc943SGeorge Wilson vdev_t *tvd = rvd->vdev_child[c]; 40288ecc943SGeorge Wilson 40388ecc943SGeorge Wilson if (tvd->vdev_ishole) 40488ecc943SGeorge Wilson array[idx++] = c; 40588ecc943SGeorge Wilson } 40688ecc943SGeorge Wilson 407312c6e15SGeorge Wilson if (idx) { 408312c6e15SGeorge Wilson VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 409312c6e15SGeorge Wilson array, idx) == 0); 410312c6e15SGeorge Wilson } 411312c6e15SGeorge Wilson 41288ecc943SGeorge Wilson VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 41388ecc943SGeorge Wilson rvd->vdev_children) == 0); 41488ecc943SGeorge Wilson 41588ecc943SGeorge Wilson kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 41688ecc943SGeorge Wilson } 41788ecc943SGeorge Wilson 418ad135b5dSChristopher Siden /* 419dfbb9432SGeorge Wilson * Returns the configuration from the label of the given vdev. For vdevs 420dfbb9432SGeorge Wilson * which don't have a txg value stored on their label (i.e. spares/cache) 421dfbb9432SGeorge Wilson * or have not been completely initialized (txg = 0) just return 422dfbb9432SGeorge Wilson * the configuration from the first valid label we find. Otherwise, 423dfbb9432SGeorge Wilson * find the most up-to-date label that does not exceed the specified 424dfbb9432SGeorge Wilson * 'txg' value. 425ad135b5dSChristopher Siden */ 426fa9e4066Sahrens nvlist_t * 427dfbb9432SGeorge Wilson vdev_label_read_config(vdev_t *vd, uint64_t txg) 428fa9e4066Sahrens { 4290373e76bSbonwick spa_t *spa = vd->vdev_spa; 430fa9e4066Sahrens nvlist_t *config = NULL; 431fa9e4066Sahrens vdev_phys_t *vp; 432fa9e4066Sahrens zio_t *zio; 433dfbb9432SGeorge Wilson uint64_t best_txg = 0; 434dfbb9432SGeorge Wilson int error = 0; 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++) { 447dfbb9432SGeorge Wilson nvlist_t *label = NULL; 448fa9e4066Sahrens 449e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 450fa9e4066Sahrens 451fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 452fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 453e14bb325SJeff Bonwick sizeof (vdev_phys_t), NULL, NULL, flags); 454fa9e4066Sahrens 455fa9e4066Sahrens if (zio_wait(zio) == 0 && 456fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 457dfbb9432SGeorge Wilson &label, 0) == 0) { 458dfbb9432SGeorge Wilson uint64_t label_txg = 0; 459dfbb9432SGeorge Wilson 460dfbb9432SGeorge Wilson /* 461dfbb9432SGeorge Wilson * Auxiliary vdevs won't have txg values in their 462dfbb9432SGeorge Wilson * labels and newly added vdevs may not have been 463dfbb9432SGeorge Wilson * completely initialized so just return the 464dfbb9432SGeorge Wilson * configuration from the first valid label we 465dfbb9432SGeorge Wilson * encounter. 466dfbb9432SGeorge Wilson */ 467dfbb9432SGeorge Wilson error = nvlist_lookup_uint64(label, 468dfbb9432SGeorge Wilson ZPOOL_CONFIG_POOL_TXG, &label_txg); 469dfbb9432SGeorge Wilson if ((error || label_txg == 0) && !config) { 470dfbb9432SGeorge Wilson config = label; 471dfbb9432SGeorge Wilson break; 472dfbb9432SGeorge Wilson } else if (label_txg <= txg && label_txg > best_txg) { 473dfbb9432SGeorge Wilson best_txg = label_txg; 474dfbb9432SGeorge Wilson nvlist_free(config); 475dfbb9432SGeorge Wilson config = fnvlist_dup(label); 476dfbb9432SGeorge Wilson } 477dfbb9432SGeorge Wilson } 478fa9e4066Sahrens 479dfbb9432SGeorge Wilson if (label != NULL) { 480dfbb9432SGeorge Wilson nvlist_free(label); 481dfbb9432SGeorge Wilson label = NULL; 482fa9e4066Sahrens } 483fa9e4066Sahrens } 484fa9e4066Sahrens 4858956713aSEric Schrock if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 4868956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 4878956713aSEric Schrock goto retry; 4888956713aSEric Schrock } 4898956713aSEric Schrock 490fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 491fa9e4066Sahrens 492fa9e4066Sahrens return (config); 493fa9e4066Sahrens } 494fa9e4066Sahrens 49539c23413Seschrock /* 49639c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 49739c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 49839c23413Seschrock */ 49939c23413Seschrock static boolean_t 50039c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 501fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 50239c23413Seschrock { 50339c23413Seschrock spa_t *spa = vd->vdev_spa; 50439c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 50539c23413Seschrock uint64_t vdtxg = 0; 50639c23413Seschrock nvlist_t *label; 50739c23413Seschrock 50839c23413Seschrock if (spare_guid) 50939c23413Seschrock *spare_guid = 0ULL; 510fa94a07fSbrendan if (l2cache_guid) 511fa94a07fSbrendan *l2cache_guid = 0ULL; 51239c23413Seschrock 51339c23413Seschrock /* 51439c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 51539c23413Seschrock */ 516dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 51739c23413Seschrock return (B_FALSE); 51839c23413Seschrock 51939c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 52039c23413Seschrock &vdtxg); 52139c23413Seschrock 52239c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 52339c23413Seschrock &state) != 0 || 52439c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 52539c23413Seschrock &device_guid) != 0) { 52639c23413Seschrock nvlist_free(label); 52739c23413Seschrock return (B_FALSE); 52839c23413Seschrock } 52939c23413Seschrock 530fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 53139c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 53239c23413Seschrock &pool_guid) != 0 || 53339c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 53439c23413Seschrock &txg) != 0)) { 53539c23413Seschrock nvlist_free(label); 53639c23413Seschrock return (B_FALSE); 53739c23413Seschrock } 53839c23413Seschrock 53939c23413Seschrock nvlist_free(label); 54039c23413Seschrock 54139c23413Seschrock /* 54239c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 54339c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 54439c23413Seschrock * spare (which we check for later on). 54539c23413Seschrock */ 546fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 54739c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 54889a89ebfSlling !spa_spare_exists(device_guid, NULL, NULL) && 549fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 55039c23413Seschrock return (B_FALSE); 55139c23413Seschrock 55239c23413Seschrock /* 55339c23413Seschrock * If the transaction group is zero, then this an initialized (but 55439c23413Seschrock * unused) label. This is only an error if the create transaction 55539c23413Seschrock * on-disk is the same as the one we're using now, in which case the 55639c23413Seschrock * user has attempted to add the same vdev multiple times in the same 55739c23413Seschrock * transaction. 55839c23413Seschrock */ 559fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 560fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 56139c23413Seschrock return (B_TRUE); 56239c23413Seschrock 56339c23413Seschrock /* 56439c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 56539c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 566fa94a07fSbrendan * to add. We also check if it is an l2cache device. 56739c23413Seschrock */ 56889a89ebfSlling if (spa_spare_exists(device_guid, &spare_pool, NULL) || 56939c23413Seschrock spa_has_spare(spa, device_guid)) { 57039c23413Seschrock if (spare_guid) 57139c23413Seschrock *spare_guid = device_guid; 57239c23413Seschrock 57339c23413Seschrock switch (reason) { 57439c23413Seschrock case VDEV_LABEL_CREATE: 575fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 57639c23413Seschrock return (B_TRUE); 57739c23413Seschrock 57839c23413Seschrock case VDEV_LABEL_REPLACE: 57939c23413Seschrock return (!spa_has_spare(spa, device_guid) || 58039c23413Seschrock spare_pool != 0ULL); 58139c23413Seschrock 58239c23413Seschrock case VDEV_LABEL_SPARE: 58339c23413Seschrock return (spa_has_spare(spa, device_guid)); 58439c23413Seschrock } 58539c23413Seschrock } 58639c23413Seschrock 587fa94a07fSbrendan /* 588fa94a07fSbrendan * Check to see if this is an l2cache device. 589fa94a07fSbrendan */ 590fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 591fa94a07fSbrendan return (B_TRUE); 592fa94a07fSbrendan 593f9af39baSGeorge Wilson /* 594f9af39baSGeorge Wilson * We can't rely on a pool's state if it's been imported 595f9af39baSGeorge Wilson * read-only. Instead we look to see if the pools is marked 596f9af39baSGeorge Wilson * read-only in the namespace and set the state to active. 597f9af39baSGeorge Wilson */ 598f9af39baSGeorge Wilson if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL && 599f9af39baSGeorge Wilson spa_mode(spa) == FREAD) 600f9af39baSGeorge Wilson state = POOL_STATE_ACTIVE; 601f9af39baSGeorge Wilson 60239c23413Seschrock /* 60339c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 60439c23413Seschrock * pool on the system. 60539c23413Seschrock */ 60639c23413Seschrock return (state == POOL_STATE_ACTIVE); 60739c23413Seschrock } 60839c23413Seschrock 60939c23413Seschrock /* 61039c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 61139c23413Seschrock * use, and writable. We put down an initial label which we will later 61239c23413Seschrock * overwrite with a complete label. Note that it's important to do this 61339c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 61439c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 61539c23413Seschrock * itself. 61639c23413Seschrock */ 61739c23413Seschrock int 61839c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 619fa9e4066Sahrens { 620fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 621fa9e4066Sahrens nvlist_t *label; 622fa9e4066Sahrens vdev_phys_t *vp; 623f83ffe1aSLin Ling char *pad2; 624ecc2d604Sbonwick uberblock_t *ub; 625fa9e4066Sahrens zio_t *zio; 626fa9e4066Sahrens char *buf; 627fa9e4066Sahrens size_t buflen; 628fa9e4066Sahrens int error; 629fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 630e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 631fa9e4066Sahrens 632e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 6330373e76bSbonwick 634e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 63539c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 63639c23413Seschrock crtxg, reason)) != 0) 637fa9e4066Sahrens return (error); 638fa9e4066Sahrens 63988ecc943SGeorge Wilson /* Track the creation time for this vdev */ 64088ecc943SGeorge Wilson vd->vdev_crtxg = crtxg; 64188ecc943SGeorge Wilson 642fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 643fa9e4066Sahrens return (0); 644fa9e4066Sahrens 645fa9e4066Sahrens /* 64639c23413Seschrock * Dead vdevs cannot be initialized. 647fa9e4066Sahrens */ 648fa9e4066Sahrens if (vdev_is_dead(vd)) 649be6fd75aSMatthew Ahrens return (SET_ERROR(EIO)); 650fa9e4066Sahrens 651fa9e4066Sahrens /* 65239c23413Seschrock * Determine if the vdev is in use. 653fa9e4066Sahrens */ 6541195e687SMark J Musante if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 655fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 656be6fd75aSMatthew Ahrens return (SET_ERROR(EBUSY)); 65739c23413Seschrock 65839c23413Seschrock /* 659fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 660fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 661fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 662fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 66339c23413Seschrock */ 664fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 665fa94a07fSbrendan spare_guid != 0ULL) { 66631157203SJeff Bonwick uint64_t guid_delta = spare_guid - vd->vdev_guid; 66799653d4eSeschrock 66831157203SJeff Bonwick vd->vdev_guid += guid_delta; 66931157203SJeff Bonwick 67031157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 67131157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 67239c23413Seschrock 67399653d4eSeschrock /* 67439c23413Seschrock * If this is a replacement, then we want to fallthrough to the 67539c23413Seschrock * rest of the code. If we're adding a spare, then it's already 6763d7072f8Seschrock * labeled appropriately and we can just return. 67799653d4eSeschrock */ 67839c23413Seschrock if (reason == VDEV_LABEL_SPARE) 67939c23413Seschrock return (0); 6801195e687SMark J Musante ASSERT(reason == VDEV_LABEL_REPLACE || 6811195e687SMark J Musante reason == VDEV_LABEL_SPLIT); 682fa9e4066Sahrens } 683fa9e4066Sahrens 684fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 685fa94a07fSbrendan l2cache_guid != 0ULL) { 68631157203SJeff Bonwick uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 68731157203SJeff Bonwick 68831157203SJeff Bonwick vd->vdev_guid += guid_delta; 689fa94a07fSbrendan 69031157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 69131157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 692fa94a07fSbrendan 693fa94a07fSbrendan /* 694fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 695fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 696fa94a07fSbrendan * already labeled appropriately and we can just return. 697fa94a07fSbrendan */ 698fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 699fa94a07fSbrendan return (0); 700fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 701fa94a07fSbrendan } 702fa94a07fSbrendan 703fa9e4066Sahrens /* 70439c23413Seschrock * Initialize its label. 705fa9e4066Sahrens */ 706fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 707fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 708fa9e4066Sahrens 709fa9e4066Sahrens /* 710fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 711fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 712fa9e4066Sahrens * really part of an active pool just yet. The labels will 713fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 714fa9e4066Sahrens */ 71539c23413Seschrock if (reason == VDEV_LABEL_SPARE || 71639c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 71739c23413Seschrock /* 71839c23413Seschrock * For inactive hot spares, we generate a special label that 71939c23413Seschrock * identifies as a mutually shared hot spare. We write the 72039c23413Seschrock * label if we are adding a hot spare, or if we are removing an 72139c23413Seschrock * active hot spare (in which case we want to revert the 72239c23413Seschrock * labels). 72339c23413Seschrock */ 72499653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 72599653d4eSeschrock 72699653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 72799653d4eSeschrock spa_version(spa)) == 0); 72899653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 72999653d4eSeschrock POOL_STATE_SPARE) == 0); 73099653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 73199653d4eSeschrock vd->vdev_guid) == 0); 732fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 733fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 734fa94a07fSbrendan /* 735fa94a07fSbrendan * For level 2 ARC devices, add a special label. 736fa94a07fSbrendan */ 737fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 738fa94a07fSbrendan 739fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 740fa94a07fSbrendan spa_version(spa)) == 0); 741fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 742fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 743fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 744fa94a07fSbrendan vd->vdev_guid) == 0); 74599653d4eSeschrock } else { 7461195e687SMark J Musante uint64_t txg = 0ULL; 7471195e687SMark J Musante 7481195e687SMark J Musante if (reason == VDEV_LABEL_SPLIT) 7491195e687SMark J Musante txg = spa->spa_uberblock.ub_txg; 7501195e687SMark J Musante label = spa_config_generate(spa, vd, txg, B_FALSE); 75199653d4eSeschrock 75299653d4eSeschrock /* 75399653d4eSeschrock * Add our creation time. This allows us to detect multiple 75499653d4eSeschrock * vdev uses as described above, and automatically expires if we 75599653d4eSeschrock * fail. 75699653d4eSeschrock */ 75799653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 75899653d4eSeschrock crtxg) == 0); 75999653d4eSeschrock } 760fa9e4066Sahrens 761fa9e4066Sahrens buf = vp->vp_nvlist; 762fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 763fa9e4066Sahrens 764a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 765a75573b6Smmusante if (error != 0) { 766fa9e4066Sahrens nvlist_free(label); 767fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 768a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 769a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 770fa9e4066Sahrens } 771fa9e4066Sahrens 772fa9e4066Sahrens /* 773fa9e4066Sahrens * Initialize uberblock template. 774fa9e4066Sahrens */ 775f64c0e34SEric Taylor ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 776f64c0e34SEric Taylor bzero(ub, VDEV_UBERBLOCK_RING); 777ecc2d604Sbonwick *ub = spa->spa_uberblock; 778ecc2d604Sbonwick ub->ub_txg = 0; 779fa9e4066Sahrens 780f83ffe1aSLin Ling /* Initialize the 2nd padding area. */ 781f83ffe1aSLin Ling pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 782f83ffe1aSLin Ling bzero(pad2, VDEV_PAD_SIZE); 783f83ffe1aSLin Ling 784fa9e4066Sahrens /* 785fa9e4066Sahrens * Write everything in parallel. 786fa9e4066Sahrens */ 7878956713aSEric Schrock retry: 78817f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 789fa9e4066Sahrens 790e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 791fa9e4066Sahrens 792fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 793fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 79417f17c2dSbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 795fa9e4066Sahrens 796f83ffe1aSLin Ling /* 797f83ffe1aSLin Ling * Skip the 1st padding area. 798f83ffe1aSLin Ling * Zero out the 2nd padding area where it might have 799f83ffe1aSLin Ling * left over data from previous filesystem format. 800f83ffe1aSLin Ling */ 801f83ffe1aSLin Ling vdev_label_write(zio, vd, l, pad2, 802f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2), 803f83ffe1aSLin Ling VDEV_PAD_SIZE, NULL, NULL, flags); 804f83ffe1aSLin Ling 805f64c0e34SEric Taylor vdev_label_write(zio, vd, l, ub, 806f64c0e34SEric Taylor offsetof(vdev_label_t, vl_uberblock), 807f64c0e34SEric Taylor VDEV_UBERBLOCK_RING, NULL, NULL, flags); 808fa9e4066Sahrens } 809fa9e4066Sahrens 810fa9e4066Sahrens error = zio_wait(zio); 811fa9e4066Sahrens 8128956713aSEric Schrock if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 8138956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 8148956713aSEric Schrock goto retry; 8158956713aSEric Schrock } 8168956713aSEric Schrock 817fa9e4066Sahrens nvlist_free(label); 818f83ffe1aSLin Ling zio_buf_free(pad2, VDEV_PAD_SIZE); 819f64c0e34SEric Taylor zio_buf_free(ub, VDEV_UBERBLOCK_RING); 820fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 821fa9e4066Sahrens 82239c23413Seschrock /* 82339c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 8243d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 825fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 826fa94a07fSbrendan * level 2 ARC devices. 82739c23413Seschrock */ 82839c23413Seschrock if (error == 0 && !vd->vdev_isspare && 82939c23413Seschrock (reason == VDEV_LABEL_SPARE || 83089a89ebfSlling spa_spare_exists(vd->vdev_guid, NULL, NULL))) 83139c23413Seschrock spa_spare_add(vd); 83299653d4eSeschrock 833fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 834fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 835fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 836fa94a07fSbrendan spa_l2cache_add(vd); 837fa94a07fSbrendan 83839c23413Seschrock return (error); 83999653d4eSeschrock } 84099653d4eSeschrock 841fa9e4066Sahrens /* 842fa9e4066Sahrens * ========================================================================== 843fa9e4066Sahrens * uberblock load/sync 844fa9e4066Sahrens * ========================================================================== 845fa9e4066Sahrens */ 846fa9e4066Sahrens 847fa9e4066Sahrens /* 848fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 849fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 850fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 851fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 852fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 853ad135b5dSChristopher Siden * the missing replica comes back, then for a few seconds we'll have two 854fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 855fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 856fa9e4066Sahrens */ 857fa9e4066Sahrens static int 858fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 859fa9e4066Sahrens { 860fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 861fa9e4066Sahrens return (-1); 862fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 863fa9e4066Sahrens return (1); 864fa9e4066Sahrens 865fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 866fa9e4066Sahrens return (-1); 867fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 868fa9e4066Sahrens return (1); 869fa9e4066Sahrens 870fa9e4066Sahrens return (0); 871fa9e4066Sahrens } 872fa9e4066Sahrens 873ad135b5dSChristopher Siden struct ubl_cbdata { 874ad135b5dSChristopher Siden uberblock_t *ubl_ubbest; /* Best uberblock */ 875ad135b5dSChristopher Siden vdev_t *ubl_vd; /* vdev associated with the above */ 876ad135b5dSChristopher Siden }; 877ad135b5dSChristopher Siden 878fa9e4066Sahrens static void 879fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 880fa9e4066Sahrens { 881ad135b5dSChristopher Siden vdev_t *vd = zio->io_vd; 882468c413aSTim Haley spa_t *spa = zio->io_spa; 883e14bb325SJeff Bonwick zio_t *rio = zio->io_private; 884ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 885ad135b5dSChristopher Siden struct ubl_cbdata *cbp = rio->io_private; 886fa9e4066Sahrens 887ad135b5dSChristopher Siden ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 888fa9e4066Sahrens 889ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 890e14bb325SJeff Bonwick mutex_enter(&rio->io_lock); 891468c413aSTim Haley if (ub->ub_txg <= spa->spa_load_max_txg && 892ad135b5dSChristopher Siden vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 893ad135b5dSChristopher Siden /* 894dfbb9432SGeorge Wilson * Keep track of the vdev in which this uberblock 895dfbb9432SGeorge Wilson * was found. We will use this information later 896dfbb9432SGeorge Wilson * to obtain the config nvlist associated with 897ad135b5dSChristopher Siden * this uberblock. 898ad135b5dSChristopher Siden */ 899ad135b5dSChristopher Siden *cbp->ubl_ubbest = *ub; 900ad135b5dSChristopher Siden cbp->ubl_vd = vd; 901ad135b5dSChristopher Siden } 902e14bb325SJeff Bonwick mutex_exit(&rio->io_lock); 903fa9e4066Sahrens } 904fa9e4066Sahrens 905fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 906fa9e4066Sahrens } 907fa9e4066Sahrens 908ad135b5dSChristopher Siden static void 909ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 910ad135b5dSChristopher Siden struct ubl_cbdata *cbp) 911fa9e4066Sahrens { 912e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 913ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 914fa9e4066Sahrens 915e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 916e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 917e14bb325SJeff Bonwick for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 918e14bb325SJeff Bonwick vdev_label_read(zio, vd, l, 919e14bb325SJeff Bonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 920e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), 921e14bb325SJeff Bonwick VDEV_UBERBLOCK_SIZE(vd), 922e14bb325SJeff Bonwick vdev_uberblock_load_done, zio, flags); 923e14bb325SJeff Bonwick } 924fa9e4066Sahrens } 925fa9e4066Sahrens } 926ad135b5dSChristopher Siden } 927e14bb325SJeff Bonwick 928ad135b5dSChristopher Siden /* 929ad135b5dSChristopher Siden * Reads the 'best' uberblock from disk along with its associated 930ad135b5dSChristopher Siden * configuration. First, we read the uberblock array of each label of each 931ad135b5dSChristopher Siden * vdev, keeping track of the uberblock with the highest txg in each array. 932dfbb9432SGeorge Wilson * Then, we read the configuration from the same vdev as the best uberblock. 933ad135b5dSChristopher Siden */ 934ad135b5dSChristopher Siden void 935ad135b5dSChristopher Siden vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 936ad135b5dSChristopher Siden { 937ad135b5dSChristopher Siden zio_t *zio; 938ad135b5dSChristopher Siden spa_t *spa = rvd->vdev_spa; 939ad135b5dSChristopher Siden struct ubl_cbdata cb; 940ad135b5dSChristopher Siden int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 941ad135b5dSChristopher Siden ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 942ad135b5dSChristopher Siden 943ad135b5dSChristopher Siden ASSERT(ub); 944ad135b5dSChristopher Siden ASSERT(config); 945ad135b5dSChristopher Siden 946ad135b5dSChristopher Siden bzero(ub, sizeof (uberblock_t)); 947ad135b5dSChristopher Siden *config = NULL; 948ad135b5dSChristopher Siden 949ad135b5dSChristopher Siden cb.ubl_ubbest = ub; 950ad135b5dSChristopher Siden cb.ubl_vd = NULL; 951ad135b5dSChristopher Siden 952ad135b5dSChristopher Siden spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 953ad135b5dSChristopher Siden zio = zio_root(spa, NULL, &cb, flags); 954ad135b5dSChristopher Siden vdev_uberblock_load_impl(zio, rvd, flags, &cb); 955ad135b5dSChristopher Siden (void) zio_wait(zio); 956dfbb9432SGeorge Wilson 957dfbb9432SGeorge Wilson /* 958dfbb9432SGeorge Wilson * It's possible that the best uberblock was discovered on a label 959dfbb9432SGeorge Wilson * that has a configuration which was written in a future txg. 960dfbb9432SGeorge Wilson * Search all labels on this vdev to find the configuration that 961dfbb9432SGeorge Wilson * matches the txg for our uberblock. 962dfbb9432SGeorge Wilson */ 963dfbb9432SGeorge Wilson if (cb.ubl_vd != NULL) 964dfbb9432SGeorge Wilson *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 965ad135b5dSChristopher Siden spa_config_exit(spa, SCL_ALL, FTAG); 966fa9e4066Sahrens } 967fa9e4066Sahrens 968fa9e4066Sahrens /* 96917f17c2dSbonwick * On success, increment root zio's count of good writes. 9700373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 971fa9e4066Sahrens */ 972fa9e4066Sahrens static void 973fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 974fa9e4066Sahrens { 97517f17c2dSbonwick uint64_t *good_writes = zio->io_private; 976fa9e4066Sahrens 9770373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 978fa9e4066Sahrens atomic_add_64(good_writes, 1); 979fa9e4066Sahrens } 980fa9e4066Sahrens 98117f17c2dSbonwick /* 98217f17c2dSbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 98317f17c2dSbonwick */ 984fa9e4066Sahrens static void 985e14bb325SJeff Bonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 986fa9e4066Sahrens { 98717f17c2dSbonwick uberblock_t *ubbuf; 988e14bb325SJeff Bonwick int n; 989fa9e4066Sahrens 990e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 991e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 992fa9e4066Sahrens 993fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 994fa9e4066Sahrens return; 995fa9e4066Sahrens 996e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 997fa9e4066Sahrens return; 998fa9e4066Sahrens 99917f17c2dSbonwick n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 1000fa9e4066Sahrens 100117f17c2dSbonwick ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 100217f17c2dSbonwick bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 100317f17c2dSbonwick *ubbuf = *ub; 1004fa9e4066Sahrens 1005e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) 100617f17c2dSbonwick vdev_label_write(zio, vd, l, ubbuf, 1007e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 100817f17c2dSbonwick vdev_uberblock_sync_done, zio->io_private, 1009e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 1010fa9e4066Sahrens 101117f17c2dSbonwick zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 1012fa9e4066Sahrens } 1013fa9e4066Sahrens 10143e30c24aSWill Andrews /* Sync the uberblocks to all vdevs in svd[] */ 101517f17c2dSbonwick int 101617f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1017fa9e4066Sahrens { 101817f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1019e14bb325SJeff Bonwick zio_t *zio; 102017f17c2dSbonwick uint64_t good_writes = 0; 1021fa9e4066Sahrens 1022e14bb325SJeff Bonwick zio = zio_root(spa, NULL, &good_writes, flags); 1023e14bb325SJeff Bonwick 1024e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 1025e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, svd[v], flags); 1026fa9e4066Sahrens 102717f17c2dSbonwick (void) zio_wait(zio); 1028fa9e4066Sahrens 1029fa9e4066Sahrens /* 103017f17c2dSbonwick * Flush the uberblocks to disk. This ensures that the odd labels 103117f17c2dSbonwick * are no longer needed (because the new uberblocks and the even 103217f17c2dSbonwick * labels are safely on disk), so it is safe to overwrite them. 1033fa9e4066Sahrens */ 103417f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 1035fa9e4066Sahrens 1036e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 103717f17c2dSbonwick zio_flush(zio, svd[v]); 1038fa9e4066Sahrens 103917f17c2dSbonwick (void) zio_wait(zio); 104017f17c2dSbonwick 104117f17c2dSbonwick return (good_writes >= 1 ? 0 : EIO); 1042fa9e4066Sahrens } 1043fa9e4066Sahrens 1044fa9e4066Sahrens /* 104517f17c2dSbonwick * On success, increment the count of good writes for our top-level vdev. 1046fa9e4066Sahrens */ 1047fa9e4066Sahrens static void 104817f17c2dSbonwick vdev_label_sync_done(zio_t *zio) 1049fa9e4066Sahrens { 105017f17c2dSbonwick uint64_t *good_writes = zio->io_private; 1051fa9e4066Sahrens 1052fa9e4066Sahrens if (zio->io_error == 0) 1053fa9e4066Sahrens atomic_add_64(good_writes, 1); 1054fa9e4066Sahrens } 1055fa9e4066Sahrens 105617f17c2dSbonwick /* 105717f17c2dSbonwick * If there weren't enough good writes, indicate failure to the parent. 105817f17c2dSbonwick */ 1059fa9e4066Sahrens static void 106017f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio) 106117f17c2dSbonwick { 106217f17c2dSbonwick uint64_t *good_writes = zio->io_private; 106317f17c2dSbonwick 106417f17c2dSbonwick if (*good_writes == 0) 1065be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(EIO); 106617f17c2dSbonwick 106717f17c2dSbonwick kmem_free(good_writes, sizeof (uint64_t)); 106817f17c2dSbonwick } 106917f17c2dSbonwick 107051ece835Seschrock /* 10710430f8daSeschrock * We ignore errors for log and cache devices, simply free the private data. 107251ece835Seschrock */ 107351ece835Seschrock static void 10740430f8daSeschrock vdev_label_sync_ignore_done(zio_t *zio) 107551ece835Seschrock { 107651ece835Seschrock kmem_free(zio->io_private, sizeof (uint64_t)); 107751ece835Seschrock } 107851ece835Seschrock 107917f17c2dSbonwick /* 108017f17c2dSbonwick * Write all even or odd labels to all leaves of the specified vdev. 108117f17c2dSbonwick */ 108217f17c2dSbonwick static void 1083e14bb325SJeff Bonwick vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1084fa9e4066Sahrens { 1085fa9e4066Sahrens nvlist_t *label; 1086fa9e4066Sahrens vdev_phys_t *vp; 1087fa9e4066Sahrens char *buf; 1088fa9e4066Sahrens size_t buflen; 1089fa9e4066Sahrens 1090e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 1091e14bb325SJeff Bonwick vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1092fa9e4066Sahrens 1093fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 1094fa9e4066Sahrens return; 1095fa9e4066Sahrens 1096e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 1097fa9e4066Sahrens return; 1098fa9e4066Sahrens 1099fa9e4066Sahrens /* 1100fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 1101fa9e4066Sahrens */ 11020373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1103fa9e4066Sahrens 1104fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1105fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 1106fa9e4066Sahrens 1107fa9e4066Sahrens buf = vp->vp_nvlist; 1108fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 1109fa9e4066Sahrens 111017f17c2dSbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 111117f17c2dSbonwick for (; l < VDEV_LABELS; l += 2) { 111217f17c2dSbonwick vdev_label_write(zio, vd, l, vp, 111317f17c2dSbonwick offsetof(vdev_label_t, vl_vdev_phys), 111417f17c2dSbonwick sizeof (vdev_phys_t), 111517f17c2dSbonwick vdev_label_sync_done, zio->io_private, 1116e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 111717f17c2dSbonwick } 111817f17c2dSbonwick } 1119fa9e4066Sahrens 1120fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 1121fa9e4066Sahrens nvlist_free(label); 1122fa9e4066Sahrens } 1123fa9e4066Sahrens 112417f17c2dSbonwick int 1125e14bb325SJeff Bonwick vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1126fa9e4066Sahrens { 1127e14bb325SJeff Bonwick list_t *dl = &spa->spa_config_dirty_list; 112817f17c2dSbonwick vdev_t *vd; 1129e14bb325SJeff Bonwick zio_t *zio; 1130fa9e4066Sahrens int error; 1131fa9e4066Sahrens 1132fa9e4066Sahrens /* 1133e14bb325SJeff Bonwick * Write the new labels to disk. 1134fa9e4066Sahrens */ 1135e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 1136fa9e4066Sahrens 113717f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 113817f17c2dSbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 113917f17c2dSbonwick KM_SLEEP); 114088ecc943SGeorge Wilson 114188ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 114288ecc943SGeorge Wilson 1143a3f829aeSBill Moore zio_t *vio = zio_null(zio, spa, NULL, 11440430f8daSeschrock (vd->vdev_islog || vd->vdev_aux != NULL) ? 11450430f8daSeschrock vdev_label_sync_ignore_done : vdev_label_sync_top_done, 114617f17c2dSbonwick good_writes, flags); 1147e14bb325SJeff Bonwick vdev_label_sync(vio, vd, l, txg, flags); 114817f17c2dSbonwick zio_nowait(vio); 1149fa9e4066Sahrens } 1150e14bb325SJeff Bonwick 1151e14bb325SJeff Bonwick error = zio_wait(zio); 1152fa9e4066Sahrens 11538654d025Sperrin /* 115417f17c2dSbonwick * Flush the new labels to disk. 11558654d025Sperrin */ 115617f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 11578654d025Sperrin 115817f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 115917f17c2dSbonwick zio_flush(zio, vd); 116017f17c2dSbonwick 116117f17c2dSbonwick (void) zio_wait(zio); 1162fa9e4066Sahrens 1163fa9e4066Sahrens return (error); 1164fa9e4066Sahrens } 1165fa9e4066Sahrens 1166fa9e4066Sahrens /* 116717f17c2dSbonwick * Sync the uberblock and any changes to the vdev configuration. 1168fa9e4066Sahrens * 1169fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 1170fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 1171fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 1172fa9e4066Sahrens * describe the failure semantics at each stage. 1173fa9e4066Sahrens * 117417f17c2dSbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1175fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 1176fa9e4066Sahrens */ 1177e14bb325SJeff Bonwick int 11788956713aSEric Schrock vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1179fa9e4066Sahrens { 118017f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1181fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 11820373e76bSbonwick vdev_t *vd; 1183fa9e4066Sahrens zio_t *zio; 1184e14bb325SJeff Bonwick int error; 1185e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1186fa9e4066Sahrens 11878956713aSEric Schrock /* 11888956713aSEric Schrock * Normally, we don't want to try too hard to write every label and 11898956713aSEric Schrock * uberblock. If there is a flaky disk, we don't want the rest of the 11908956713aSEric Schrock * sync process to block while we retry. But if we can't write a 11918956713aSEric Schrock * single label out, we should retry with ZIO_FLAG_TRYHARD before 11928956713aSEric Schrock * bailing out and declaring the pool faulted. 11938956713aSEric Schrock */ 11948956713aSEric Schrock if (tryhard) 11958956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 11968956713aSEric Schrock 1197fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 1198fa9e4066Sahrens 1199fa9e4066Sahrens /* 120017f17c2dSbonwick * If this isn't a resync due to I/O errors, 120117f17c2dSbonwick * and nothing changed in this transaction group, 120217f17c2dSbonwick * and the vdev configuration hasn't changed, 12030373e76bSbonwick * then there's nothing to do. 1204fa9e4066Sahrens */ 120517f17c2dSbonwick if (ub->ub_txg < txg && 120617f17c2dSbonwick uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1207e14bb325SJeff Bonwick list_is_empty(&spa->spa_config_dirty_list)) 1208e14bb325SJeff Bonwick return (0); 1209fa9e4066Sahrens 1210fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 1211e14bb325SJeff Bonwick return (0); 1212fa9e4066Sahrens 12130373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 12140373e76bSbonwick 1215fa9e4066Sahrens /* 1216fa9e4066Sahrens * Flush the write cache of every disk that's been written to 1217fa9e4066Sahrens * in this transaction group. This ensures that all blocks 1218fa9e4066Sahrens * written in this txg will be committed to stable storage 1219fa9e4066Sahrens * before any uberblock that references them. 1220fa9e4066Sahrens */ 122117f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 122217f17c2dSbonwick 1223fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 122417f17c2dSbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 122517f17c2dSbonwick zio_flush(zio, vd); 122617f17c2dSbonwick 1227fa9e4066Sahrens (void) zio_wait(zio); 1228fa9e4066Sahrens 1229fa9e4066Sahrens /* 1230fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1231fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1232fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1233fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 123417f17c2dSbonwick * which have not yet been touched, will still be valid. We flush 123517f17c2dSbonwick * the new labels to disk to ensure that all even-label updates 123617f17c2dSbonwick * are committed to stable storage before the uberblock update. 1237fa9e4066Sahrens */ 1238e14bb325SJeff Bonwick if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1239e14bb325SJeff Bonwick return (error); 1240fa9e4066Sahrens 1241fa9e4066Sahrens /* 1242e14bb325SJeff Bonwick * Sync the uberblocks to all vdevs in svd[]. 12430373e76bSbonwick * If the system dies in the middle of this step, there are two cases 12440373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1245fa9e4066Sahrens * 1246fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1247fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1248fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1249fa9e4066Sahrens * to that uberblock. 1250fa9e4066Sahrens * 1251fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1252fa9e4066Sahrens * will be the newest, and the even labels (which had all 1253fa9e4066Sahrens * been successfully committed) will be valid with respect 1254fa9e4066Sahrens * to the new uberblocks. 1255fa9e4066Sahrens */ 1256e14bb325SJeff Bonwick if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1257e14bb325SJeff Bonwick return (error); 1258fa9e4066Sahrens 1259fa9e4066Sahrens /* 1260fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1261fa9e4066Sahrens * in the middle of this process, the even labels and the new 1262fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1263fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1264fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 126517f17c2dSbonwick * all labels will be brought up to date. We flush the new labels 126617f17c2dSbonwick * to disk to ensure that all odd-label updates are committed to 126717f17c2dSbonwick * stable storage before the next transaction group begins. 1268fa9e4066Sahrens */ 1269e14bb325SJeff Bonwick return (vdev_label_sync_list(spa, 1, txg, flags)); 1270fa9e4066Sahrens } 1271