vdev_label.c revision 219089
1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd/* 22219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23168404Spjd */ 24168404Spjd 25168404Spjd/* 26168404Spjd * Virtual Device Labels 27168404Spjd * --------------------- 28168404Spjd * 29168404Spjd * The vdev label serves several distinct purposes: 30168404Spjd * 31168404Spjd * 1. Uniquely identify this device as part of a ZFS pool and confirm its 32168404Spjd * identity within the pool. 33168404Spjd * 34168404Spjd * 2. Verify that all the devices given in a configuration are present 35168404Spjd * within the pool. 36168404Spjd * 37168404Spjd * 3. Determine the uberblock for the pool. 38168404Spjd * 39168404Spjd * 4. In case of an import operation, determine the configuration of the 40168404Spjd * toplevel vdev of which it is a part. 41168404Spjd * 42168404Spjd * 5. If an import operation cannot find all the devices in the pool, 43168404Spjd * provide enough information to the administrator to determine which 44168404Spjd * devices are missing. 45168404Spjd * 46168404Spjd * It is important to note that while the kernel is responsible for writing the 47168404Spjd * label, it only consumes the information in the first three cases. The 48168404Spjd * latter information is only consumed in userland when determining the 49168404Spjd * configuration to import a pool. 50168404Spjd * 51168404Spjd * 52168404Spjd * Label Organization 53168404Spjd * ------------------ 54168404Spjd * 55168404Spjd * Before describing the contents of the label, it's important to understand how 56168404Spjd * the labels are written and updated with respect to the uberblock. 57168404Spjd * 58168404Spjd * When the pool configuration is altered, either because it was newly created 59168404Spjd * or a device was added, we want to update all the labels such that we can deal 60168404Spjd * with fatal failure at any point. To this end, each disk has two labels which 61168404Spjd * are updated before and after the uberblock is synced. Assuming we have 62185029Spjd * labels and an uberblock with the following transaction groups: 63168404Spjd * 64168404Spjd * L1 UB L2 65168404Spjd * +------+ +------+ +------+ 66168404Spjd * | | | | | | 67168404Spjd * | t10 | | t10 | | t10 | 68168404Spjd * | | | | | | 69168404Spjd * +------+ +------+ +------+ 70168404Spjd * 71168404Spjd * In this stable state, the labels and the uberblock were all updated within 72168404Spjd * the same transaction group (10). Each label is mirrored and checksummed, so 73168404Spjd * that we can detect when we fail partway through writing the label. 74168404Spjd * 75168404Spjd * In order to identify which labels are valid, the labels are written in the 76168404Spjd * following manner: 77168404Spjd * 78168404Spjd * 1. For each vdev, update 'L1' to the new label 79168404Spjd * 2. Update the uberblock 80168404Spjd * 3. For each vdev, update 'L2' to the new label 81168404Spjd * 82168404Spjd * Given arbitrary failure, we can determine the correct label to use based on 83168404Spjd * the transaction group. If we fail after updating L1 but before updating the 84168404Spjd * UB, we will notice that L1's transaction group is greater than the uberblock, 85168404Spjd * so L2 must be valid. If we fail after writing the uberblock but before 86168404Spjd * writing L2, we will notice that L2's transaction group is less than L1, and 87168404Spjd * therefore L1 is valid. 88168404Spjd * 89168404Spjd * Another added complexity is that not every label is updated when the config 90168404Spjd * is synced. If we add a single device, we do not want to have to re-write 91168404Spjd * every label for every device in the pool. This means that both L1 and L2 may 92168404Spjd * be older than the pool uberblock, because the necessary information is stored 93168404Spjd * on another vdev. 94168404Spjd * 95168404Spjd * 96168404Spjd * On-disk Format 97168404Spjd * -------------- 98168404Spjd * 99168404Spjd * The vdev label consists of two distinct parts, and is wrapped within the 100168404Spjd * vdev_label_t structure. The label includes 8k of padding to permit legacy 101168404Spjd * VTOC disk labels, but is otherwise ignored. 102168404Spjd * 103168404Spjd * The first half of the label is a packed nvlist which contains pool wide 104168404Spjd * properties, per-vdev properties, and configuration information. It is 105168404Spjd * described in more detail below. 106168404Spjd * 107168404Spjd * The latter half of the label consists of a redundant array of uberblocks. 108168404Spjd * These uberblocks are updated whenever a transaction group is committed, 109168404Spjd * or when the configuration is updated. When a pool is loaded, we scan each 110168404Spjd * vdev for the 'best' uberblock. 111168404Spjd * 112168404Spjd * 113168404Spjd * Configuration Information 114168404Spjd * ------------------------- 115168404Spjd * 116168404Spjd * The nvlist describing the pool and vdev contains the following elements: 117168404Spjd * 118168404Spjd * version ZFS on-disk version 119168404Spjd * name Pool name 120168404Spjd * state Pool state 121168404Spjd * txg Transaction group in which this label was written 122168404Spjd * pool_guid Unique identifier for this pool 123168404Spjd * vdev_tree An nvlist describing vdev tree. 124168404Spjd * 125168404Spjd * Each leaf device label also contains the following: 126168404Spjd * 127168404Spjd * top_guid Unique ID for top-level vdev in which this is contained 128168404Spjd * guid Unique ID for the leaf vdev 129168404Spjd * 130168404Spjd * The 'vs' configuration follows the format described in 'spa_config.c'. 131168404Spjd */ 132168404Spjd 133168404Spjd#include <sys/zfs_context.h> 134168404Spjd#include <sys/spa.h> 135168404Spjd#include <sys/spa_impl.h> 136168404Spjd#include <sys/dmu.h> 137168404Spjd#include <sys/zap.h> 138168404Spjd#include <sys/vdev.h> 139168404Spjd#include <sys/vdev_impl.h> 140168404Spjd#include <sys/uberblock_impl.h> 141168404Spjd#include <sys/metaslab.h> 142168404Spjd#include <sys/zio.h> 143219089Spjd#include <sys/dsl_scan.h> 144168404Spjd#include <sys/fs/zfs.h> 145168404Spjd 146168404Spjd/* 147168404Spjd * Basic routines to read and write from a vdev label. 148168404Spjd * Used throughout the rest of this file. 149168404Spjd */ 150168404Spjduint64_t 151168404Spjdvdev_label_offset(uint64_t psize, int l, uint64_t offset) 152168404Spjd{ 153168404Spjd ASSERT(offset < sizeof (vdev_label_t)); 154185029Spjd ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 155168404Spjd 156168404Spjd return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 157168404Spjd 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 158168404Spjd} 159168404Spjd 160185029Spjd/* 161185029Spjd * Returns back the vdev label associated with the passed in offset. 162185029Spjd */ 163185029Spjdint 164185029Spjdvdev_label_number(uint64_t psize, uint64_t offset) 165185029Spjd{ 166185029Spjd int l; 167185029Spjd 168185029Spjd if (offset >= psize - VDEV_LABEL_END_SIZE) { 169185029Spjd offset -= psize - VDEV_LABEL_END_SIZE; 170185029Spjd offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 171185029Spjd } 172185029Spjd l = offset / sizeof (vdev_label_t); 173185029Spjd return (l < VDEV_LABELS ? l : -1); 174185029Spjd} 175185029Spjd 176168404Spjdstatic void 177168404Spjdvdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 178185029Spjd uint64_t size, zio_done_func_t *done, void *private, int flags) 179168404Spjd{ 180185029Spjd ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 181185029Spjd SCL_STATE_ALL); 182185029Spjd ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 183168404Spjd 184168404Spjd zio_nowait(zio_read_phys(zio, vd, 185168404Spjd vdev_label_offset(vd->vdev_psize, l, offset), 186168404Spjd size, buf, ZIO_CHECKSUM_LABEL, done, private, 187185029Spjd ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 188168404Spjd} 189168404Spjd 190168404Spjdstatic void 191168404Spjdvdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 192185029Spjd uint64_t size, zio_done_func_t *done, void *private, int flags) 193168404Spjd{ 194185029Spjd ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 195185029Spjd (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 196185029Spjd (SCL_CONFIG | SCL_STATE) && 197185029Spjd dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 198185029Spjd ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 199168404Spjd 200168404Spjd zio_nowait(zio_write_phys(zio, vd, 201168404Spjd vdev_label_offset(vd->vdev_psize, l, offset), 202168404Spjd size, buf, ZIO_CHECKSUM_LABEL, done, private, 203185029Spjd ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 204168404Spjd} 205168404Spjd 206168404Spjd/* 207168404Spjd * Generate the nvlist representing this vdev's config. 208168404Spjd */ 209168404Spjdnvlist_t * 210168404Spjdvdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 211219089Spjd vdev_config_flag_t flags) 212168404Spjd{ 213168404Spjd nvlist_t *nv = NULL; 214168404Spjd 215168404Spjd VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 216168404Spjd 217168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 218168404Spjd vd->vdev_ops->vdev_op_type) == 0); 219219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 220168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 221168404Spjd == 0); 222168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 223168404Spjd 224168404Spjd if (vd->vdev_path != NULL) 225168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 226168404Spjd vd->vdev_path) == 0); 227168404Spjd 228168404Spjd if (vd->vdev_devid != NULL) 229168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 230168404Spjd vd->vdev_devid) == 0); 231168404Spjd 232185029Spjd if (vd->vdev_physpath != NULL) 233185029Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 234185029Spjd vd->vdev_physpath) == 0); 235185029Spjd 236209962Smm if (vd->vdev_fru != NULL) 237209962Smm VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, 238209962Smm vd->vdev_fru) == 0); 239209962Smm 240168404Spjd if (vd->vdev_nparity != 0) { 241168404Spjd ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 242168404Spjd VDEV_TYPE_RAIDZ) == 0); 243168404Spjd 244168404Spjd /* 245168404Spjd * Make sure someone hasn't managed to sneak a fancy new vdev 246168404Spjd * into a crufty old storage pool. 247168404Spjd */ 248168404Spjd ASSERT(vd->vdev_nparity == 1 || 249219089Spjd (vd->vdev_nparity <= 2 && 250219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ2) || 251219089Spjd (vd->vdev_nparity <= 3 && 252219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ3)); 253168404Spjd 254168404Spjd /* 255168404Spjd * Note that we'll add the nparity tag even on storage pools 256168404Spjd * that only support a single parity device -- older software 257168404Spjd * will just ignore it. 258168404Spjd */ 259168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 260168404Spjd vd->vdev_nparity) == 0); 261168404Spjd } 262168404Spjd 263168404Spjd if (vd->vdev_wholedisk != -1ULL) 264168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 265168404Spjd vd->vdev_wholedisk) == 0); 266168404Spjd 267168404Spjd if (vd->vdev_not_present) 268168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 269168404Spjd 270168404Spjd if (vd->vdev_isspare) 271168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 272168404Spjd 273219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 274219089Spjd vd == vd->vdev_top) { 275168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 276168404Spjd vd->vdev_ms_array) == 0); 277168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 278168404Spjd vd->vdev_ms_shift) == 0); 279168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 280168404Spjd vd->vdev_ashift) == 0); 281168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 282168404Spjd vd->vdev_asize) == 0); 283185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 284185029Spjd vd->vdev_islog) == 0); 285219089Spjd if (vd->vdev_removing) 286219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 287219089Spjd vd->vdev_removing) == 0); 288168404Spjd } 289168404Spjd 290209962Smm if (vd->vdev_dtl_smo.smo_object != 0) 291168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 292209962Smm vd->vdev_dtl_smo.smo_object) == 0); 293168404Spjd 294219089Spjd if (vd->vdev_crtxg) 295219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 296219089Spjd vd->vdev_crtxg) == 0); 297219089Spjd 298168404Spjd if (getstats) { 299168404Spjd vdev_stat_t vs; 300219089Spjd pool_scan_stat_t ps; 301219089Spjd 302168404Spjd vdev_get_stats(vd, &vs); 303219089Spjd VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 304168404Spjd (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 305219089Spjd 306219089Spjd /* provide either current or previous scan information */ 307219089Spjd if (spa_scan_get_stats(spa, &ps) == 0) { 308219089Spjd VERIFY(nvlist_add_uint64_array(nv, 309219089Spjd ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 310219089Spjd sizeof (pool_scan_stat_t) / sizeof (uint64_t)) 311219089Spjd == 0); 312219089Spjd } 313168404Spjd } 314168404Spjd 315168404Spjd if (!vd->vdev_ops->vdev_op_leaf) { 316168404Spjd nvlist_t **child; 317219089Spjd int c, idx; 318168404Spjd 319219089Spjd ASSERT(!vd->vdev_ishole); 320219089Spjd 321168404Spjd child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 322168404Spjd KM_SLEEP); 323168404Spjd 324219089Spjd for (c = 0, idx = 0; c < vd->vdev_children; c++) { 325219089Spjd vdev_t *cvd = vd->vdev_child[c]; 326168404Spjd 327219089Spjd /* 328219089Spjd * If we're generating an nvlist of removing 329219089Spjd * vdevs then skip over any device which is 330219089Spjd * not being removed. 331219089Spjd */ 332219089Spjd if ((flags & VDEV_CONFIG_REMOVING) && 333219089Spjd !cvd->vdev_removing) 334219089Spjd continue; 335168404Spjd 336219089Spjd child[idx++] = vdev_config_generate(spa, cvd, 337219089Spjd getstats, flags); 338219089Spjd } 339219089Spjd 340219089Spjd if (idx) { 341219089Spjd VERIFY(nvlist_add_nvlist_array(nv, 342219089Spjd ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 343219089Spjd } 344219089Spjd 345219089Spjd for (c = 0; c < idx; c++) 346168404Spjd nvlist_free(child[c]); 347168404Spjd 348168404Spjd kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 349168404Spjd 350168404Spjd } else { 351219089Spjd const char *aux = NULL; 352219089Spjd 353168404Spjd if (vd->vdev_offline && !vd->vdev_tmpoffline) 354168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 355168404Spjd B_TRUE) == 0); 356219089Spjd if (vd->vdev_resilvering) 357219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING, 358219089Spjd B_TRUE) == 0); 359185029Spjd if (vd->vdev_faulted) 360185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 361185029Spjd B_TRUE) == 0); 362185029Spjd if (vd->vdev_degraded) 363185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 364185029Spjd B_TRUE) == 0); 365185029Spjd if (vd->vdev_removed) 366185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 367185029Spjd B_TRUE) == 0); 368185029Spjd if (vd->vdev_unspare) 369185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 370185029Spjd B_TRUE) == 0); 371219089Spjd if (vd->vdev_ishole) 372219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, 373219089Spjd B_TRUE) == 0); 374219089Spjd 375219089Spjd switch (vd->vdev_stat.vs_aux) { 376219089Spjd case VDEV_AUX_ERR_EXCEEDED: 377219089Spjd aux = "err_exceeded"; 378219089Spjd break; 379219089Spjd 380219089Spjd case VDEV_AUX_EXTERNAL: 381219089Spjd aux = "external"; 382219089Spjd break; 383219089Spjd } 384219089Spjd 385219089Spjd if (aux != NULL) 386219089Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, 387219089Spjd aux) == 0); 388219089Spjd 389219089Spjd if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 390219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 391219089Spjd vd->vdev_orig_guid) == 0); 392219089Spjd } 393168404Spjd } 394168404Spjd 395168404Spjd return (nv); 396168404Spjd} 397168404Spjd 398219089Spjd/* 399219089Spjd * Generate a view of the top-level vdevs. If we currently have holes 400219089Spjd * in the namespace, then generate an array which contains a list of holey 401219089Spjd * vdevs. Additionally, add the number of top-level children that currently 402219089Spjd * exist. 403219089Spjd */ 404219089Spjdvoid 405219089Spjdvdev_top_config_generate(spa_t *spa, nvlist_t *config) 406219089Spjd{ 407219089Spjd vdev_t *rvd = spa->spa_root_vdev; 408219089Spjd uint64_t *array; 409219089Spjd uint_t c, idx; 410219089Spjd 411219089Spjd array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 412219089Spjd 413219089Spjd for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 414219089Spjd vdev_t *tvd = rvd->vdev_child[c]; 415219089Spjd 416219089Spjd if (tvd->vdev_ishole) 417219089Spjd array[idx++] = c; 418219089Spjd } 419219089Spjd 420219089Spjd if (idx) { 421219089Spjd VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 422219089Spjd array, idx) == 0); 423219089Spjd } 424219089Spjd 425219089Spjd VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 426219089Spjd rvd->vdev_children) == 0); 427219089Spjd 428219089Spjd kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 429219089Spjd} 430219089Spjd 431168404Spjdnvlist_t * 432168404Spjdvdev_label_read_config(vdev_t *vd) 433168404Spjd{ 434168404Spjd spa_t *spa = vd->vdev_spa; 435168404Spjd nvlist_t *config = NULL; 436168404Spjd vdev_phys_t *vp; 437168404Spjd zio_t *zio; 438213198Smm int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 439213198Smm ZIO_FLAG_SPECULATIVE; 440168404Spjd 441185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 442168404Spjd 443185029Spjd if (!vdev_readable(vd)) 444168404Spjd return (NULL); 445168404Spjd 446168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 447168404Spjd 448213198Smmretry: 449185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 450168404Spjd 451185029Spjd zio = zio_root(spa, NULL, NULL, flags); 452168404Spjd 453168404Spjd vdev_label_read(zio, vd, l, vp, 454168404Spjd offsetof(vdev_label_t, vl_vdev_phys), 455185029Spjd sizeof (vdev_phys_t), NULL, NULL, flags); 456168404Spjd 457168404Spjd if (zio_wait(zio) == 0 && 458168404Spjd nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 459168404Spjd &config, 0) == 0) 460168404Spjd break; 461168404Spjd 462168404Spjd if (config != NULL) { 463168404Spjd nvlist_free(config); 464168404Spjd config = NULL; 465168404Spjd } 466168404Spjd } 467168404Spjd 468213198Smm if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 469213198Smm flags |= ZIO_FLAG_TRYHARD; 470213198Smm goto retry; 471213198Smm } 472213198Smm 473168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 474168404Spjd 475168404Spjd return (config); 476168404Spjd} 477168404Spjd 478168404Spjd/* 479168404Spjd * Determine if a device is in use. The 'spare_guid' parameter will be filled 480168404Spjd * in with the device guid if this spare is active elsewhere on the system. 481168404Spjd */ 482168404Spjdstatic boolean_t 483168404Spjdvdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 484185029Spjd uint64_t *spare_guid, uint64_t *l2cache_guid) 485168404Spjd{ 486168404Spjd spa_t *spa = vd->vdev_spa; 487168404Spjd uint64_t state, pool_guid, device_guid, txg, spare_pool; 488168404Spjd uint64_t vdtxg = 0; 489168404Spjd nvlist_t *label; 490168404Spjd 491168404Spjd if (spare_guid) 492168404Spjd *spare_guid = 0ULL; 493185029Spjd if (l2cache_guid) 494185029Spjd *l2cache_guid = 0ULL; 495168404Spjd 496168404Spjd /* 497168404Spjd * Read the label, if any, and perform some basic sanity checks. 498168404Spjd */ 499168404Spjd if ((label = vdev_label_read_config(vd)) == NULL) 500168404Spjd return (B_FALSE); 501168404Spjd 502168404Spjd (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 503168404Spjd &vdtxg); 504168404Spjd 505168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 506168404Spjd &state) != 0 || 507168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 508168404Spjd &device_guid) != 0) { 509168404Spjd nvlist_free(label); 510168404Spjd return (B_FALSE); 511168404Spjd } 512168404Spjd 513185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 514168404Spjd (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 515168404Spjd &pool_guid) != 0 || 516168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 517168404Spjd &txg) != 0)) { 518168404Spjd nvlist_free(label); 519168404Spjd return (B_FALSE); 520168404Spjd } 521168404Spjd 522168404Spjd nvlist_free(label); 523168404Spjd 524168404Spjd /* 525168404Spjd * Check to see if this device indeed belongs to the pool it claims to 526168404Spjd * be a part of. The only way this is allowed is if the device is a hot 527168404Spjd * spare (which we check for later on). 528168404Spjd */ 529185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 530168404Spjd !spa_guid_exists(pool_guid, device_guid) && 531185029Spjd !spa_spare_exists(device_guid, NULL, NULL) && 532185029Spjd !spa_l2cache_exists(device_guid, NULL)) 533168404Spjd return (B_FALSE); 534168404Spjd 535168404Spjd /* 536168404Spjd * If the transaction group is zero, then this an initialized (but 537168404Spjd * unused) label. This is only an error if the create transaction 538168404Spjd * on-disk is the same as the one we're using now, in which case the 539168404Spjd * user has attempted to add the same vdev multiple times in the same 540168404Spjd * transaction. 541168404Spjd */ 542185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 543185029Spjd txg == 0 && vdtxg == crtxg) 544168404Spjd return (B_TRUE); 545168404Spjd 546168404Spjd /* 547168404Spjd * Check to see if this is a spare device. We do an explicit check for 548168404Spjd * spa_has_spare() here because it may be on our pending list of spares 549185029Spjd * to add. We also check if it is an l2cache device. 550168404Spjd */ 551185029Spjd if (spa_spare_exists(device_guid, &spare_pool, NULL) || 552168404Spjd spa_has_spare(spa, device_guid)) { 553168404Spjd if (spare_guid) 554168404Spjd *spare_guid = device_guid; 555168404Spjd 556168404Spjd switch (reason) { 557168404Spjd case VDEV_LABEL_CREATE: 558185029Spjd case VDEV_LABEL_L2CACHE: 559168404Spjd return (B_TRUE); 560168404Spjd 561168404Spjd case VDEV_LABEL_REPLACE: 562168404Spjd return (!spa_has_spare(spa, device_guid) || 563168404Spjd spare_pool != 0ULL); 564168404Spjd 565168404Spjd case VDEV_LABEL_SPARE: 566168404Spjd return (spa_has_spare(spa, device_guid)); 567168404Spjd } 568168404Spjd } 569168404Spjd 570168404Spjd /* 571185029Spjd * Check to see if this is an l2cache device. 572185029Spjd */ 573185029Spjd if (spa_l2cache_exists(device_guid, NULL)) 574185029Spjd return (B_TRUE); 575185029Spjd 576185029Spjd /* 577219089Spjd * We can't rely on a pool's state if it's been imported 578219089Spjd * read-only. Instead we look to see if the pools is marked 579219089Spjd * read-only in the namespace and set the state to active. 580219089Spjd */ 581219089Spjd if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL && 582219089Spjd spa_mode(spa) == FREAD) 583219089Spjd state = POOL_STATE_ACTIVE; 584219089Spjd 585219089Spjd /* 586168404Spjd * If the device is marked ACTIVE, then this device is in use by another 587168404Spjd * pool on the system. 588168404Spjd */ 589168404Spjd return (state == POOL_STATE_ACTIVE); 590168404Spjd} 591168404Spjd 592168404Spjd/* 593168404Spjd * Initialize a vdev label. We check to make sure each leaf device is not in 594168404Spjd * use, and writable. We put down an initial label which we will later 595168404Spjd * overwrite with a complete label. Note that it's important to do this 596168404Spjd * sequentially, not in parallel, so that we catch cases of multiple use of the 597168404Spjd * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 598168404Spjd * itself. 599168404Spjd */ 600168404Spjdint 601168404Spjdvdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 602168404Spjd{ 603168404Spjd spa_t *spa = vd->vdev_spa; 604168404Spjd nvlist_t *label; 605168404Spjd vdev_phys_t *vp; 606209962Smm char *pad2; 607168404Spjd uberblock_t *ub; 608168404Spjd zio_t *zio; 609168404Spjd char *buf; 610168404Spjd size_t buflen; 611168404Spjd int error; 612185029Spjd uint64_t spare_guid, l2cache_guid; 613185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 614168404Spjd 615185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 616168404Spjd 617185029Spjd for (int c = 0; c < vd->vdev_children; c++) 618168404Spjd if ((error = vdev_label_init(vd->vdev_child[c], 619168404Spjd crtxg, reason)) != 0) 620168404Spjd return (error); 621168404Spjd 622219089Spjd /* Track the creation time for this vdev */ 623219089Spjd vd->vdev_crtxg = crtxg; 624219089Spjd 625168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 626168404Spjd return (0); 627168404Spjd 628168404Spjd /* 629168404Spjd * Dead vdevs cannot be initialized. 630168404Spjd */ 631168404Spjd if (vdev_is_dead(vd)) 632168404Spjd return (EIO); 633168404Spjd 634168404Spjd /* 635168404Spjd * Determine if the vdev is in use. 636168404Spjd */ 637219089Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 638185029Spjd vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 639168404Spjd return (EBUSY); 640168404Spjd 641168404Spjd /* 642185029Spjd * If this is a request to add or replace a spare or l2cache device 643185029Spjd * that is in use elsewhere on the system, then we must update the 644185029Spjd * guid (which was initialized to a random value) to reflect the 645185029Spjd * actual GUID (which is shared between multiple pools). 646168404Spjd */ 647185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 648185029Spjd spare_guid != 0ULL) { 649185029Spjd uint64_t guid_delta = spare_guid - vd->vdev_guid; 650168404Spjd 651185029Spjd vd->vdev_guid += guid_delta; 652168404Spjd 653185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 654185029Spjd pvd->vdev_guid_sum += guid_delta; 655168404Spjd 656168404Spjd /* 657168404Spjd * If this is a replacement, then we want to fallthrough to the 658168404Spjd * rest of the code. If we're adding a spare, then it's already 659185029Spjd * labeled appropriately and we can just return. 660168404Spjd */ 661168404Spjd if (reason == VDEV_LABEL_SPARE) 662168404Spjd return (0); 663219089Spjd ASSERT(reason == VDEV_LABEL_REPLACE || 664219089Spjd reason == VDEV_LABEL_SPLIT); 665168404Spjd } 666168404Spjd 667185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 668185029Spjd l2cache_guid != 0ULL) { 669185029Spjd uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 670185029Spjd 671185029Spjd vd->vdev_guid += guid_delta; 672185029Spjd 673185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 674185029Spjd pvd->vdev_guid_sum += guid_delta; 675185029Spjd 676185029Spjd /* 677185029Spjd * If this is a replacement, then we want to fallthrough to the 678185029Spjd * rest of the code. If we're adding an l2cache, then it's 679185029Spjd * already labeled appropriately and we can just return. 680185029Spjd */ 681185029Spjd if (reason == VDEV_LABEL_L2CACHE) 682185029Spjd return (0); 683185029Spjd ASSERT(reason == VDEV_LABEL_REPLACE); 684185029Spjd } 685185029Spjd 686168404Spjd /* 687168404Spjd * Initialize its label. 688168404Spjd */ 689168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 690168404Spjd bzero(vp, sizeof (vdev_phys_t)); 691168404Spjd 692168404Spjd /* 693168404Spjd * Generate a label describing the pool and our top-level vdev. 694168404Spjd * We mark it as being from txg 0 to indicate that it's not 695168404Spjd * really part of an active pool just yet. The labels will 696168404Spjd * be written again with a meaningful txg by spa_sync(). 697168404Spjd */ 698168404Spjd if (reason == VDEV_LABEL_SPARE || 699168404Spjd (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 700168404Spjd /* 701168404Spjd * For inactive hot spares, we generate a special label that 702168404Spjd * identifies as a mutually shared hot spare. We write the 703168404Spjd * label if we are adding a hot spare, or if we are removing an 704168404Spjd * active hot spare (in which case we want to revert the 705168404Spjd * labels). 706168404Spjd */ 707168404Spjd VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 708168404Spjd 709168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 710168404Spjd spa_version(spa)) == 0); 711168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 712168404Spjd POOL_STATE_SPARE) == 0); 713168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 714168404Spjd vd->vdev_guid) == 0); 715185029Spjd } else if (reason == VDEV_LABEL_L2CACHE || 716185029Spjd (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 717185029Spjd /* 718185029Spjd * For level 2 ARC devices, add a special label. 719185029Spjd */ 720185029Spjd VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 721185029Spjd 722185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 723185029Spjd spa_version(spa)) == 0); 724185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 725185029Spjd POOL_STATE_L2CACHE) == 0); 726185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 727185029Spjd vd->vdev_guid) == 0); 728168404Spjd } else { 729219089Spjd uint64_t txg = 0ULL; 730168404Spjd 731219089Spjd if (reason == VDEV_LABEL_SPLIT) 732219089Spjd txg = spa->spa_uberblock.ub_txg; 733219089Spjd label = spa_config_generate(spa, vd, txg, B_FALSE); 734219089Spjd 735168404Spjd /* 736168404Spjd * Add our creation time. This allows us to detect multiple 737168404Spjd * vdev uses as described above, and automatically expires if we 738168404Spjd * fail. 739168404Spjd */ 740168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 741168404Spjd crtxg) == 0); 742168404Spjd } 743168404Spjd 744168404Spjd buf = vp->vp_nvlist; 745168404Spjd buflen = sizeof (vp->vp_nvlist); 746168404Spjd 747168404Spjd error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 748168404Spjd if (error != 0) { 749168404Spjd nvlist_free(label); 750168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 751168404Spjd /* EFAULT means nvlist_pack ran out of room */ 752168404Spjd return (error == EFAULT ? ENAMETOOLONG : EINVAL); 753168404Spjd } 754168404Spjd 755168404Spjd /* 756168404Spjd * Initialize uberblock template. 757168404Spjd */ 758219089Spjd ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 759219089Spjd bzero(ub, VDEV_UBERBLOCK_RING); 760168404Spjd *ub = spa->spa_uberblock; 761168404Spjd ub->ub_txg = 0; 762168404Spjd 763209962Smm /* Initialize the 2nd padding area. */ 764209962Smm pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 765209962Smm bzero(pad2, VDEV_PAD_SIZE); 766209962Smm 767168404Spjd /* 768168404Spjd * Write everything in parallel. 769168404Spjd */ 770213198Smmretry: 771185029Spjd zio = zio_root(spa, NULL, NULL, flags); 772168404Spjd 773185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 774168404Spjd 775168404Spjd vdev_label_write(zio, vd, l, vp, 776168404Spjd offsetof(vdev_label_t, vl_vdev_phys), 777185029Spjd sizeof (vdev_phys_t), NULL, NULL, flags); 778168404Spjd 779209962Smm /* 780209962Smm * Skip the 1st padding area. 781209962Smm * Zero out the 2nd padding area where it might have 782209962Smm * left over data from previous filesystem format. 783209962Smm */ 784209962Smm vdev_label_write(zio, vd, l, pad2, 785209962Smm offsetof(vdev_label_t, vl_pad2), 786209962Smm VDEV_PAD_SIZE, NULL, NULL, flags); 787168404Spjd 788219089Spjd vdev_label_write(zio, vd, l, ub, 789219089Spjd offsetof(vdev_label_t, vl_uberblock), 790219089Spjd VDEV_UBERBLOCK_RING, NULL, NULL, flags); 791168404Spjd } 792168404Spjd 793168404Spjd error = zio_wait(zio); 794168404Spjd 795213198Smm if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 796213198Smm flags |= ZIO_FLAG_TRYHARD; 797213198Smm goto retry; 798213198Smm } 799213198Smm 800168404Spjd nvlist_free(label); 801209962Smm zio_buf_free(pad2, VDEV_PAD_SIZE); 802219089Spjd zio_buf_free(ub, VDEV_UBERBLOCK_RING); 803168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 804168404Spjd 805168404Spjd /* 806168404Spjd * If this vdev hasn't been previously identified as a spare, then we 807185029Spjd * mark it as such only if a) we are labeling it as a spare, or b) it 808185029Spjd * exists as a spare elsewhere in the system. Do the same for 809185029Spjd * level 2 ARC devices. 810168404Spjd */ 811168404Spjd if (error == 0 && !vd->vdev_isspare && 812168404Spjd (reason == VDEV_LABEL_SPARE || 813185029Spjd spa_spare_exists(vd->vdev_guid, NULL, NULL))) 814168404Spjd spa_spare_add(vd); 815168404Spjd 816185029Spjd if (error == 0 && !vd->vdev_isl2cache && 817185029Spjd (reason == VDEV_LABEL_L2CACHE || 818185029Spjd spa_l2cache_exists(vd->vdev_guid, NULL))) 819185029Spjd spa_l2cache_add(vd); 820185029Spjd 821168404Spjd return (error); 822168404Spjd} 823168404Spjd 824168404Spjd/* 825168404Spjd * ========================================================================== 826168404Spjd * uberblock load/sync 827168404Spjd * ========================================================================== 828168404Spjd */ 829168404Spjd 830168404Spjd/* 831168404Spjd * Consider the following situation: txg is safely synced to disk. We've 832168404Spjd * written the first uberblock for txg + 1, and then we lose power. When we 833168404Spjd * come back up, we fail to see the uberblock for txg + 1 because, say, 834168404Spjd * it was on a mirrored device and the replica to which we wrote txg + 1 835168404Spjd * is now offline. If we then make some changes and sync txg + 1, and then 836168404Spjd * the missing replica comes back, then for a new seconds we'll have two 837168404Spjd * conflicting uberblocks on disk with the same txg. The solution is simple: 838168404Spjd * among uberblocks with equal txg, choose the one with the latest timestamp. 839168404Spjd */ 840168404Spjdstatic int 841168404Spjdvdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 842168404Spjd{ 843168404Spjd if (ub1->ub_txg < ub2->ub_txg) 844168404Spjd return (-1); 845168404Spjd if (ub1->ub_txg > ub2->ub_txg) 846168404Spjd return (1); 847168404Spjd 848168404Spjd if (ub1->ub_timestamp < ub2->ub_timestamp) 849168404Spjd return (-1); 850168404Spjd if (ub1->ub_timestamp > ub2->ub_timestamp) 851168404Spjd return (1); 852168404Spjd 853168404Spjd return (0); 854168404Spjd} 855168404Spjd 856168404Spjdstatic void 857168404Spjdvdev_uberblock_load_done(zio_t *zio) 858168404Spjd{ 859219089Spjd spa_t *spa = zio->io_spa; 860185029Spjd zio_t *rio = zio->io_private; 861168404Spjd uberblock_t *ub = zio->io_data; 862185029Spjd uberblock_t *ubbest = rio->io_private; 863168404Spjd 864168404Spjd ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 865168404Spjd 866168404Spjd if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 867185029Spjd mutex_enter(&rio->io_lock); 868219089Spjd if (ub->ub_txg <= spa->spa_load_max_txg && 869209962Smm vdev_uberblock_compare(ub, ubbest) > 0) 870168404Spjd *ubbest = *ub; 871185029Spjd mutex_exit(&rio->io_lock); 872168404Spjd } 873168404Spjd 874168404Spjd zio_buf_free(zio->io_data, zio->io_size); 875168404Spjd} 876168404Spjd 877168404Spjdvoid 878168404Spjdvdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 879168404Spjd{ 880185029Spjd spa_t *spa = vd->vdev_spa; 881185029Spjd vdev_t *rvd = spa->spa_root_vdev; 882213198Smm int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 883213198Smm ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 884168404Spjd 885185029Spjd if (vd == rvd) { 886185029Spjd ASSERT(zio == NULL); 887185029Spjd spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 888185029Spjd zio = zio_root(spa, NULL, ubbest, flags); 889185029Spjd bzero(ubbest, sizeof (uberblock_t)); 890185029Spjd } 891168404Spjd 892185029Spjd ASSERT(zio != NULL); 893168404Spjd 894185029Spjd for (int c = 0; c < vd->vdev_children; c++) 895185029Spjd vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 896168404Spjd 897185029Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 898185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 899185029Spjd for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 900185029Spjd vdev_label_read(zio, vd, l, 901185029Spjd zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 902185029Spjd VDEV_UBERBLOCK_OFFSET(vd, n), 903185029Spjd VDEV_UBERBLOCK_SIZE(vd), 904185029Spjd vdev_uberblock_load_done, zio, flags); 905185029Spjd } 906168404Spjd } 907168404Spjd } 908185029Spjd 909185029Spjd if (vd == rvd) { 910185029Spjd (void) zio_wait(zio); 911185029Spjd spa_config_exit(spa, SCL_ALL, FTAG); 912185029Spjd } 913168404Spjd} 914168404Spjd 915168404Spjd/* 916185029Spjd * On success, increment root zio's count of good writes. 917168404Spjd * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 918168404Spjd */ 919168404Spjdstatic void 920168404Spjdvdev_uberblock_sync_done(zio_t *zio) 921168404Spjd{ 922185029Spjd uint64_t *good_writes = zio->io_private; 923168404Spjd 924168404Spjd if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 925168404Spjd atomic_add_64(good_writes, 1); 926168404Spjd} 927168404Spjd 928185029Spjd/* 929185029Spjd * Write the uberblock to all labels of all leaves of the specified vdev. 930185029Spjd */ 931168404Spjdstatic void 932185029Spjdvdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 933168404Spjd{ 934185029Spjd uberblock_t *ubbuf; 935185029Spjd int n; 936168404Spjd 937185029Spjd for (int c = 0; c < vd->vdev_children; c++) 938185029Spjd vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 939168404Spjd 940168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 941168404Spjd return; 942168404Spjd 943185029Spjd if (!vdev_writeable(vd)) 944168404Spjd return; 945168404Spjd 946185029Spjd n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 947168404Spjd 948185029Spjd ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 949185029Spjd bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 950185029Spjd *ubbuf = *ub; 951168404Spjd 952185029Spjd for (int l = 0; l < VDEV_LABELS; l++) 953185029Spjd vdev_label_write(zio, vd, l, ubbuf, 954185029Spjd VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 955185029Spjd vdev_uberblock_sync_done, zio->io_private, 956185029Spjd flags | ZIO_FLAG_DONT_PROPAGATE); 957168404Spjd 958185029Spjd zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 959168404Spjd} 960168404Spjd 961185029Spjdint 962185029Spjdvdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 963168404Spjd{ 964185029Spjd spa_t *spa = svd[0]->vdev_spa; 965168404Spjd zio_t *zio; 966185029Spjd uint64_t good_writes = 0; 967168404Spjd 968185029Spjd zio = zio_root(spa, NULL, &good_writes, flags); 969168404Spjd 970185029Spjd for (int v = 0; v < svdcount; v++) 971185029Spjd vdev_uberblock_sync(zio, ub, svd[v], flags); 972168404Spjd 973185029Spjd (void) zio_wait(zio); 974168404Spjd 975168404Spjd /* 976185029Spjd * Flush the uberblocks to disk. This ensures that the odd labels 977185029Spjd * are no longer needed (because the new uberblocks and the even 978185029Spjd * labels are safely on disk), so it is safe to overwrite them. 979168404Spjd */ 980185029Spjd zio = zio_root(spa, NULL, NULL, flags); 981168404Spjd 982185029Spjd for (int v = 0; v < svdcount; v++) 983185029Spjd zio_flush(zio, svd[v]); 984168404Spjd 985185029Spjd (void) zio_wait(zio); 986185029Spjd 987185029Spjd return (good_writes >= 1 ? 0 : EIO); 988168404Spjd} 989168404Spjd 990168404Spjd/* 991185029Spjd * On success, increment the count of good writes for our top-level vdev. 992168404Spjd */ 993168404Spjdstatic void 994185029Spjdvdev_label_sync_done(zio_t *zio) 995168404Spjd{ 996185029Spjd uint64_t *good_writes = zio->io_private; 997168404Spjd 998168404Spjd if (zio->io_error == 0) 999168404Spjd atomic_add_64(good_writes, 1); 1000168404Spjd} 1001168404Spjd 1002185029Spjd/* 1003185029Spjd * If there weren't enough good writes, indicate failure to the parent. 1004185029Spjd */ 1005168404Spjdstatic void 1006185029Spjdvdev_label_sync_top_done(zio_t *zio) 1007168404Spjd{ 1008185029Spjd uint64_t *good_writes = zio->io_private; 1009185029Spjd 1010185029Spjd if (*good_writes == 0) 1011185029Spjd zio->io_error = EIO; 1012185029Spjd 1013185029Spjd kmem_free(good_writes, sizeof (uint64_t)); 1014185029Spjd} 1015185029Spjd 1016185029Spjd/* 1017185029Spjd * We ignore errors for log and cache devices, simply free the private data. 1018185029Spjd */ 1019185029Spjdstatic void 1020185029Spjdvdev_label_sync_ignore_done(zio_t *zio) 1021185029Spjd{ 1022185029Spjd kmem_free(zio->io_private, sizeof (uint64_t)); 1023185029Spjd} 1024185029Spjd 1025185029Spjd/* 1026185029Spjd * Write all even or odd labels to all leaves of the specified vdev. 1027185029Spjd */ 1028185029Spjdstatic void 1029185029Spjdvdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1030185029Spjd{ 1031168404Spjd nvlist_t *label; 1032168404Spjd vdev_phys_t *vp; 1033168404Spjd char *buf; 1034168404Spjd size_t buflen; 1035168404Spjd 1036185029Spjd for (int c = 0; c < vd->vdev_children; c++) 1037185029Spjd vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1038168404Spjd 1039168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1040168404Spjd return; 1041168404Spjd 1042185029Spjd if (!vdev_writeable(vd)) 1043168404Spjd return; 1044168404Spjd 1045168404Spjd /* 1046168404Spjd * Generate a label describing the top-level config to which we belong. 1047168404Spjd */ 1048168404Spjd label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1049168404Spjd 1050168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1051168404Spjd bzero(vp, sizeof (vdev_phys_t)); 1052168404Spjd 1053168404Spjd buf = vp->vp_nvlist; 1054168404Spjd buflen = sizeof (vp->vp_nvlist); 1055168404Spjd 1056185029Spjd if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1057185029Spjd for (; l < VDEV_LABELS; l += 2) { 1058185029Spjd vdev_label_write(zio, vd, l, vp, 1059185029Spjd offsetof(vdev_label_t, vl_vdev_phys), 1060185029Spjd sizeof (vdev_phys_t), 1061185029Spjd vdev_label_sync_done, zio->io_private, 1062185029Spjd flags | ZIO_FLAG_DONT_PROPAGATE); 1063185029Spjd } 1064185029Spjd } 1065168404Spjd 1066168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 1067168404Spjd nvlist_free(label); 1068168404Spjd} 1069168404Spjd 1070185029Spjdint 1071185029Spjdvdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1072168404Spjd{ 1073185029Spjd list_t *dl = &spa->spa_config_dirty_list; 1074185029Spjd vdev_t *vd; 1075168404Spjd zio_t *zio; 1076168404Spjd int error; 1077168404Spjd 1078185029Spjd /* 1079185029Spjd * Write the new labels to disk. 1080185029Spjd */ 1081185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1082168404Spjd 1083185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1084185029Spjd uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1085185029Spjd KM_SLEEP); 1086219089Spjd 1087219089Spjd ASSERT(!vd->vdev_ishole); 1088219089Spjd 1089209962Smm zio_t *vio = zio_null(zio, spa, NULL, 1090185029Spjd (vd->vdev_islog || vd->vdev_aux != NULL) ? 1091185029Spjd vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1092185029Spjd good_writes, flags); 1093185029Spjd vdev_label_sync(vio, vd, l, txg, flags); 1094185029Spjd zio_nowait(vio); 1095185029Spjd } 1096168404Spjd 1097185029Spjd error = zio_wait(zio); 1098168404Spjd 1099168404Spjd /* 1100185029Spjd * Flush the new labels to disk. 1101168404Spjd */ 1102185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1103168404Spjd 1104185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1105185029Spjd zio_flush(zio, vd); 1106168404Spjd 1107185029Spjd (void) zio_wait(zio); 1108168404Spjd 1109168404Spjd return (error); 1110168404Spjd} 1111168404Spjd 1112168404Spjd/* 1113185029Spjd * Sync the uberblock and any changes to the vdev configuration. 1114168404Spjd * 1115168404Spjd * The order of operations is carefully crafted to ensure that 1116168404Spjd * if the system panics or loses power at any time, the state on disk 1117168404Spjd * is still transactionally consistent. The in-line comments below 1118168404Spjd * describe the failure semantics at each stage. 1119168404Spjd * 1120185029Spjd * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1121168404Spjd * at any time, you can just call it again, and it will resume its work. 1122168404Spjd */ 1123168404Spjdint 1124213198Smmvdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1125168404Spjd{ 1126185029Spjd spa_t *spa = svd[0]->vdev_spa; 1127168404Spjd uberblock_t *ub = &spa->spa_uberblock; 1128168404Spjd vdev_t *vd; 1129168404Spjd zio_t *zio; 1130185029Spjd int error; 1131185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1132168404Spjd 1133213198Smm /* 1134213198Smm * Normally, we don't want to try too hard to write every label and 1135213198Smm * uberblock. If there is a flaky disk, we don't want the rest of the 1136213198Smm * sync process to block while we retry. But if we can't write a 1137213198Smm * single label out, we should retry with ZIO_FLAG_TRYHARD before 1138213198Smm * bailing out and declaring the pool faulted. 1139213198Smm */ 1140213198Smm if (tryhard) 1141213198Smm flags |= ZIO_FLAG_TRYHARD; 1142213198Smm 1143168404Spjd ASSERT(ub->ub_txg <= txg); 1144168404Spjd 1145168404Spjd /* 1146185029Spjd * If this isn't a resync due to I/O errors, 1147185029Spjd * and nothing changed in this transaction group, 1148185029Spjd * and the vdev configuration hasn't changed, 1149168404Spjd * then there's nothing to do. 1150168404Spjd */ 1151185029Spjd if (ub->ub_txg < txg && 1152185029Spjd uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1153185029Spjd list_is_empty(&spa->spa_config_dirty_list)) 1154168404Spjd return (0); 1155168404Spjd 1156168404Spjd if (txg > spa_freeze_txg(spa)) 1157168404Spjd return (0); 1158168404Spjd 1159168404Spjd ASSERT(txg <= spa->spa_final_txg); 1160168404Spjd 1161168404Spjd /* 1162168404Spjd * Flush the write cache of every disk that's been written to 1163168404Spjd * in this transaction group. This ensures that all blocks 1164168404Spjd * written in this txg will be committed to stable storage 1165168404Spjd * before any uberblock that references them. 1166168404Spjd */ 1167185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1168185029Spjd 1169168404Spjd for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 1170185029Spjd vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1171185029Spjd zio_flush(zio, vd); 1172185029Spjd 1173168404Spjd (void) zio_wait(zio); 1174168404Spjd 1175168404Spjd /* 1176168404Spjd * Sync out the even labels (L0, L2) for every dirty vdev. If the 1177168404Spjd * system dies in the middle of this process, that's OK: all of the 1178168404Spjd * even labels that made it to disk will be newer than any uberblock, 1179168404Spjd * and will therefore be considered invalid. The odd labels (L1, L3), 1180185029Spjd * which have not yet been touched, will still be valid. We flush 1181185029Spjd * the new labels to disk to ensure that all even-label updates 1182185029Spjd * are committed to stable storage before the uberblock update. 1183168404Spjd */ 1184185029Spjd if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1185185029Spjd return (error); 1186168404Spjd 1187168404Spjd /* 1188185029Spjd * Sync the uberblocks to all vdevs in svd[]. 1189168404Spjd * If the system dies in the middle of this step, there are two cases 1190168404Spjd * to consider, and the on-disk state is consistent either way: 1191168404Spjd * 1192168404Spjd * (1) If none of the new uberblocks made it to disk, then the 1193168404Spjd * previous uberblock will be the newest, and the odd labels 1194168404Spjd * (which had not yet been touched) will be valid with respect 1195168404Spjd * to that uberblock. 1196168404Spjd * 1197168404Spjd * (2) If one or more new uberblocks made it to disk, then they 1198168404Spjd * will be the newest, and the even labels (which had all 1199168404Spjd * been successfully committed) will be valid with respect 1200168404Spjd * to the new uberblocks. 1201168404Spjd */ 1202185029Spjd if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1203168404Spjd return (error); 1204168404Spjd 1205168404Spjd /* 1206168404Spjd * Sync out odd labels for every dirty vdev. If the system dies 1207168404Spjd * in the middle of this process, the even labels and the new 1208168404Spjd * uberblocks will suffice to open the pool. The next time 1209168404Spjd * the pool is opened, the first thing we'll do -- before any 1210168404Spjd * user data is modified -- is mark every vdev dirty so that 1211185029Spjd * all labels will be brought up to date. We flush the new labels 1212185029Spjd * to disk to ensure that all odd-label updates are committed to 1213185029Spjd * stable storage before the next transaction group begins. 1214168404Spjd */ 1215185029Spjd return (vdev_label_sync_list(spa, 1, txg, flags)); 1216168404Spjd} 1217