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 */ 21236884Smm 22168404Spjd/* 23219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24289562Smav * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 25168404Spjd */ 26168404Spjd 27168404Spjd/* 28168404Spjd * Virtual Device Labels 29168404Spjd * --------------------- 30168404Spjd * 31168404Spjd * The vdev label serves several distinct purposes: 32168404Spjd * 33168404Spjd * 1. Uniquely identify this device as part of a ZFS pool and confirm its 34168404Spjd * identity within the pool. 35168404Spjd * 36168404Spjd * 2. Verify that all the devices given in a configuration are present 37168404Spjd * within the pool. 38168404Spjd * 39168404Spjd * 3. Determine the uberblock for the pool. 40168404Spjd * 41168404Spjd * 4. In case of an import operation, determine the configuration of the 42168404Spjd * toplevel vdev of which it is a part. 43168404Spjd * 44168404Spjd * 5. If an import operation cannot find all the devices in the pool, 45168404Spjd * provide enough information to the administrator to determine which 46168404Spjd * devices are missing. 47168404Spjd * 48168404Spjd * It is important to note that while the kernel is responsible for writing the 49168404Spjd * label, it only consumes the information in the first three cases. The 50168404Spjd * latter information is only consumed in userland when determining the 51168404Spjd * configuration to import a pool. 52168404Spjd * 53168404Spjd * 54168404Spjd * Label Organization 55168404Spjd * ------------------ 56168404Spjd * 57168404Spjd * Before describing the contents of the label, it's important to understand how 58168404Spjd * the labels are written and updated with respect to the uberblock. 59168404Spjd * 60168404Spjd * When the pool configuration is altered, either because it was newly created 61168404Spjd * or a device was added, we want to update all the labels such that we can deal 62168404Spjd * with fatal failure at any point. To this end, each disk has two labels which 63168404Spjd * are updated before and after the uberblock is synced. Assuming we have 64185029Spjd * labels and an uberblock with the following transaction groups: 65168404Spjd * 66168404Spjd * L1 UB L2 67168404Spjd * +------+ +------+ +------+ 68168404Spjd * | | | | | | 69168404Spjd * | t10 | | t10 | | t10 | 70168404Spjd * | | | | | | 71168404Spjd * +------+ +------+ +------+ 72168404Spjd * 73168404Spjd * In this stable state, the labels and the uberblock were all updated within 74168404Spjd * the same transaction group (10). Each label is mirrored and checksummed, so 75168404Spjd * that we can detect when we fail partway through writing the label. 76168404Spjd * 77168404Spjd * In order to identify which labels are valid, the labels are written in the 78168404Spjd * following manner: 79168404Spjd * 80168404Spjd * 1. For each vdev, update 'L1' to the new label 81168404Spjd * 2. Update the uberblock 82168404Spjd * 3. For each vdev, update 'L2' to the new label 83168404Spjd * 84168404Spjd * Given arbitrary failure, we can determine the correct label to use based on 85168404Spjd * the transaction group. If we fail after updating L1 but before updating the 86168404Spjd * UB, we will notice that L1's transaction group is greater than the uberblock, 87168404Spjd * so L2 must be valid. If we fail after writing the uberblock but before 88168404Spjd * writing L2, we will notice that L2's transaction group is less than L1, and 89168404Spjd * therefore L1 is valid. 90168404Spjd * 91168404Spjd * Another added complexity is that not every label is updated when the config 92168404Spjd * is synced. If we add a single device, we do not want to have to re-write 93168404Spjd * every label for every device in the pool. This means that both L1 and L2 may 94168404Spjd * be older than the pool uberblock, because the necessary information is stored 95168404Spjd * on another vdev. 96168404Spjd * 97168404Spjd * 98168404Spjd * On-disk Format 99168404Spjd * -------------- 100168404Spjd * 101168404Spjd * The vdev label consists of two distinct parts, and is wrapped within the 102168404Spjd * vdev_label_t structure. The label includes 8k of padding to permit legacy 103168404Spjd * VTOC disk labels, but is otherwise ignored. 104168404Spjd * 105168404Spjd * The first half of the label is a packed nvlist which contains pool wide 106168404Spjd * properties, per-vdev properties, and configuration information. It is 107168404Spjd * described in more detail below. 108168404Spjd * 109168404Spjd * The latter half of the label consists of a redundant array of uberblocks. 110168404Spjd * These uberblocks are updated whenever a transaction group is committed, 111168404Spjd * or when the configuration is updated. When a pool is loaded, we scan each 112168404Spjd * vdev for the 'best' uberblock. 113168404Spjd * 114168404Spjd * 115168404Spjd * Configuration Information 116168404Spjd * ------------------------- 117168404Spjd * 118168404Spjd * The nvlist describing the pool and vdev contains the following elements: 119168404Spjd * 120168404Spjd * version ZFS on-disk version 121168404Spjd * name Pool name 122168404Spjd * state Pool state 123168404Spjd * txg Transaction group in which this label was written 124168404Spjd * pool_guid Unique identifier for this pool 125168404Spjd * vdev_tree An nvlist describing vdev tree. 126236884Smm * features_for_read 127236884Smm * An nvlist of the features necessary for reading the MOS. 128168404Spjd * 129168404Spjd * Each leaf device label also contains the following: 130168404Spjd * 131168404Spjd * top_guid Unique ID for top-level vdev in which this is contained 132168404Spjd * guid Unique ID for the leaf vdev 133168404Spjd * 134168404Spjd * The 'vs' configuration follows the format described in 'spa_config.c'. 135168404Spjd */ 136168404Spjd 137168404Spjd#include <sys/zfs_context.h> 138168404Spjd#include <sys/spa.h> 139168404Spjd#include <sys/spa_impl.h> 140168404Spjd#include <sys/dmu.h> 141168404Spjd#include <sys/zap.h> 142168404Spjd#include <sys/vdev.h> 143168404Spjd#include <sys/vdev_impl.h> 144168404Spjd#include <sys/uberblock_impl.h> 145168404Spjd#include <sys/metaslab.h> 146168404Spjd#include <sys/zio.h> 147219089Spjd#include <sys/dsl_scan.h> 148240868Spjd#include <sys/trim_map.h> 149168404Spjd#include <sys/fs/zfs.h> 150168404Spjd 151244188Ssmhstatic boolean_t vdev_trim_on_init = B_TRUE; 152244188SsmhSYSCTL_DECL(_vfs_zfs_vdev); 153244188SsmhSYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW, 154244188Ssmh &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation"); 155244188Ssmh 156168404Spjd/* 157168404Spjd * Basic routines to read and write from a vdev label. 158168404Spjd * Used throughout the rest of this file. 159168404Spjd */ 160168404Spjduint64_t 161168404Spjdvdev_label_offset(uint64_t psize, int l, uint64_t offset) 162168404Spjd{ 163168404Spjd ASSERT(offset < sizeof (vdev_label_t)); 164185029Spjd ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 165168404Spjd 166168404Spjd return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 167168404Spjd 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 168168404Spjd} 169168404Spjd 170185029Spjd/* 171185029Spjd * Returns back the vdev label associated with the passed in offset. 172185029Spjd */ 173185029Spjdint 174185029Spjdvdev_label_number(uint64_t psize, uint64_t offset) 175185029Spjd{ 176185029Spjd int l; 177185029Spjd 178185029Spjd if (offset >= psize - VDEV_LABEL_END_SIZE) { 179185029Spjd offset -= psize - VDEV_LABEL_END_SIZE; 180185029Spjd offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 181185029Spjd } 182185029Spjd l = offset / sizeof (vdev_label_t); 183185029Spjd return (l < VDEV_LABELS ? l : -1); 184185029Spjd} 185185029Spjd 186168404Spjdstatic void 187168404Spjdvdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 188289562Smav uint64_t size, zio_done_func_t *done, void *private, int flags) 189168404Spjd{ 190185029Spjd ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 191185029Spjd SCL_STATE_ALL); 192185029Spjd ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 193168404Spjd 194168404Spjd zio_nowait(zio_read_phys(zio, vd, 195168404Spjd vdev_label_offset(vd->vdev_psize, l, offset), 196168404Spjd size, buf, ZIO_CHECKSUM_LABEL, done, private, 197185029Spjd ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 198168404Spjd} 199168404Spjd 200168404Spjdstatic void 201168404Spjdvdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 202289562Smav uint64_t size, zio_done_func_t *done, void *private, int flags) 203168404Spjd{ 204185029Spjd ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 205185029Spjd (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 206185029Spjd (SCL_CONFIG | SCL_STATE) && 207185029Spjd dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 208185029Spjd ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 209168404Spjd 210168404Spjd zio_nowait(zio_write_phys(zio, vd, 211168404Spjd vdev_label_offset(vd->vdev_psize, l, offset), 212168404Spjd size, buf, ZIO_CHECKSUM_LABEL, done, private, 213185029Spjd ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 214168404Spjd} 215168404Spjd 216168404Spjd/* 217168404Spjd * Generate the nvlist representing this vdev's config. 218168404Spjd */ 219168404Spjdnvlist_t * 220168404Spjdvdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 221219089Spjd vdev_config_flag_t flags) 222168404Spjd{ 223168404Spjd nvlist_t *nv = NULL; 224168404Spjd 225254112Sdelphij nv = fnvlist_alloc(); 226168404Spjd 227254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); 228219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 229254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); 230254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); 231168404Spjd 232168404Spjd if (vd->vdev_path != NULL) 233254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); 234168404Spjd 235168404Spjd if (vd->vdev_devid != NULL) 236254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); 237168404Spjd 238185029Spjd if (vd->vdev_physpath != NULL) 239254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 240254112Sdelphij vd->vdev_physpath); 241185029Spjd 242209962Smm if (vd->vdev_fru != NULL) 243254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); 244209962Smm 245168404Spjd if (vd->vdev_nparity != 0) { 246168404Spjd ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 247168404Spjd VDEV_TYPE_RAIDZ) == 0); 248168404Spjd 249168404Spjd /* 250168404Spjd * Make sure someone hasn't managed to sneak a fancy new vdev 251168404Spjd * into a crufty old storage pool. 252168404Spjd */ 253168404Spjd ASSERT(vd->vdev_nparity == 1 || 254219089Spjd (vd->vdev_nparity <= 2 && 255219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ2) || 256219089Spjd (vd->vdev_nparity <= 3 && 257219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ3)); 258168404Spjd 259168404Spjd /* 260168404Spjd * Note that we'll add the nparity tag even on storage pools 261168404Spjd * that only support a single parity device -- older software 262168404Spjd * will just ignore it. 263168404Spjd */ 264254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); 265168404Spjd } 266168404Spjd 267168404Spjd if (vd->vdev_wholedisk != -1ULL) 268254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 269254112Sdelphij vd->vdev_wholedisk); 270168404Spjd 271168404Spjd if (vd->vdev_not_present) 272254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); 273168404Spjd 274168404Spjd if (vd->vdev_isspare) 275254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); 276168404Spjd 277219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 278219089Spjd vd == vd->vdev_top) { 279254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 280254112Sdelphij vd->vdev_ms_array); 281254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 282254112Sdelphij vd->vdev_ms_shift); 283254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); 284254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 285254112Sdelphij vd->vdev_asize); 286254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); 287219089Spjd if (vd->vdev_removing) 288254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 289254112Sdelphij vd->vdev_removing); 290168404Spjd } 291168404Spjd 292258717Savg if (vd->vdev_dtl_sm != NULL) { 293254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 294258717Savg space_map_object(vd->vdev_dtl_sm)); 295258717Savg } 296168404Spjd 297219089Spjd if (vd->vdev_crtxg) 298254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); 299219089Spjd 300299441Smav if (flags & VDEV_CONFIG_MOS) { 301299441Smav if (vd->vdev_leaf_zap != 0) { 302299441Smav ASSERT(vd->vdev_ops->vdev_op_leaf); 303299441Smav fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, 304299441Smav vd->vdev_leaf_zap); 305299441Smav } 306299441Smav 307299441Smav if (vd->vdev_top_zap != 0) { 308299441Smav ASSERT(vd == vd->vdev_top); 309299441Smav fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 310299441Smav vd->vdev_top_zap); 311299441Smav } 312299441Smav } 313299441Smav 314168404Spjd if (getstats) { 315168404Spjd vdev_stat_t vs; 316219089Spjd pool_scan_stat_t ps; 317219089Spjd 318168404Spjd vdev_get_stats(vd, &vs); 319254112Sdelphij fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 320254112Sdelphij (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); 321219089Spjd 322219089Spjd /* provide either current or previous scan information */ 323219089Spjd if (spa_scan_get_stats(spa, &ps) == 0) { 324254112Sdelphij fnvlist_add_uint64_array(nv, 325219089Spjd ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 326254112Sdelphij sizeof (pool_scan_stat_t) / sizeof (uint64_t)); 327219089Spjd } 328168404Spjd } 329168404Spjd 330168404Spjd if (!vd->vdev_ops->vdev_op_leaf) { 331168404Spjd nvlist_t **child; 332219089Spjd int c, idx; 333168404Spjd 334219089Spjd ASSERT(!vd->vdev_ishole); 335219089Spjd 336168404Spjd child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 337168404Spjd KM_SLEEP); 338168404Spjd 339219089Spjd for (c = 0, idx = 0; c < vd->vdev_children; c++) { 340219089Spjd vdev_t *cvd = vd->vdev_child[c]; 341168404Spjd 342219089Spjd /* 343219089Spjd * If we're generating an nvlist of removing 344219089Spjd * vdevs then skip over any device which is 345219089Spjd * not being removed. 346219089Spjd */ 347219089Spjd if ((flags & VDEV_CONFIG_REMOVING) && 348219089Spjd !cvd->vdev_removing) 349219089Spjd continue; 350168404Spjd 351219089Spjd child[idx++] = vdev_config_generate(spa, cvd, 352219089Spjd getstats, flags); 353219089Spjd } 354219089Spjd 355219089Spjd if (idx) { 356254112Sdelphij fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 357254112Sdelphij child, idx); 358219089Spjd } 359219089Spjd 360219089Spjd for (c = 0; c < idx; c++) 361168404Spjd nvlist_free(child[c]); 362168404Spjd 363168404Spjd kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 364168404Spjd 365168404Spjd } else { 366219089Spjd const char *aux = NULL; 367219089Spjd 368168404Spjd if (vd->vdev_offline && !vd->vdev_tmpoffline) 369254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); 370254112Sdelphij if (vd->vdev_resilver_txg != 0) 371254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 372254112Sdelphij vd->vdev_resilver_txg); 373185029Spjd if (vd->vdev_faulted) 374254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); 375185029Spjd if (vd->vdev_degraded) 376254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); 377185029Spjd if (vd->vdev_removed) 378254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); 379185029Spjd if (vd->vdev_unspare) 380254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); 381219089Spjd if (vd->vdev_ishole) 382254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); 383219089Spjd 384219089Spjd switch (vd->vdev_stat.vs_aux) { 385219089Spjd case VDEV_AUX_ERR_EXCEEDED: 386219089Spjd aux = "err_exceeded"; 387219089Spjd break; 388219089Spjd 389219089Spjd case VDEV_AUX_EXTERNAL: 390219089Spjd aux = "external"; 391219089Spjd break; 392219089Spjd } 393219089Spjd 394219089Spjd if (aux != NULL) 395254112Sdelphij fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); 396219089Spjd 397219089Spjd if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 398254112Sdelphij fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 399254112Sdelphij vd->vdev_orig_guid); 400219089Spjd } 401168404Spjd } 402168404Spjd 403168404Spjd return (nv); 404168404Spjd} 405168404Spjd 406219089Spjd/* 407219089Spjd * Generate a view of the top-level vdevs. If we currently have holes 408219089Spjd * in the namespace, then generate an array which contains a list of holey 409219089Spjd * vdevs. Additionally, add the number of top-level children that currently 410219089Spjd * exist. 411219089Spjd */ 412219089Spjdvoid 413219089Spjdvdev_top_config_generate(spa_t *spa, nvlist_t *config) 414219089Spjd{ 415219089Spjd vdev_t *rvd = spa->spa_root_vdev; 416219089Spjd uint64_t *array; 417219089Spjd uint_t c, idx; 418219089Spjd 419219089Spjd array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 420219089Spjd 421219089Spjd for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 422219089Spjd vdev_t *tvd = rvd->vdev_child[c]; 423219089Spjd 424219089Spjd if (tvd->vdev_ishole) 425219089Spjd array[idx++] = c; 426219089Spjd } 427219089Spjd 428219089Spjd if (idx) { 429219089Spjd VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 430219089Spjd array, idx) == 0); 431219089Spjd } 432219089Spjd 433219089Spjd VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 434219089Spjd rvd->vdev_children) == 0); 435219089Spjd 436219089Spjd kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 437219089Spjd} 438219089Spjd 439236884Smm/* 440239620Smm * Returns the configuration from the label of the given vdev. For vdevs 441239620Smm * which don't have a txg value stored on their label (i.e. spares/cache) 442239620Smm * or have not been completely initialized (txg = 0) just return 443239620Smm * the configuration from the first valid label we find. Otherwise, 444239620Smm * find the most up-to-date label that does not exceed the specified 445239620Smm * 'txg' value. 446236884Smm */ 447168404Spjdnvlist_t * 448239620Smmvdev_label_read_config(vdev_t *vd, uint64_t txg) 449168404Spjd{ 450168404Spjd spa_t *spa = vd->vdev_spa; 451168404Spjd nvlist_t *config = NULL; 452168404Spjd vdev_phys_t *vp; 453168404Spjd zio_t *zio; 454239620Smm uint64_t best_txg = 0; 455239620Smm int error = 0; 456213198Smm int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 457213198Smm ZIO_FLAG_SPECULATIVE; 458168404Spjd 459185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 460168404Spjd 461185029Spjd if (!vdev_readable(vd)) 462168404Spjd return (NULL); 463168404Spjd 464168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 465168404Spjd 466213198Smmretry: 467185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 468239620Smm nvlist_t *label = NULL; 469168404Spjd 470185029Spjd zio = zio_root(spa, NULL, NULL, flags); 471168404Spjd 472168404Spjd vdev_label_read(zio, vd, l, vp, 473168404Spjd offsetof(vdev_label_t, vl_vdev_phys), 474185029Spjd sizeof (vdev_phys_t), NULL, NULL, flags); 475168404Spjd 476168404Spjd if (zio_wait(zio) == 0 && 477168404Spjd nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 478239620Smm &label, 0) == 0) { 479239620Smm uint64_t label_txg = 0; 480168404Spjd 481239620Smm /* 482239620Smm * Auxiliary vdevs won't have txg values in their 483239620Smm * labels and newly added vdevs may not have been 484239620Smm * completely initialized so just return the 485239620Smm * configuration from the first valid label we 486239620Smm * encounter. 487239620Smm */ 488239620Smm error = nvlist_lookup_uint64(label, 489239620Smm ZPOOL_CONFIG_POOL_TXG, &label_txg); 490239620Smm if ((error || label_txg == 0) && !config) { 491239620Smm config = label; 492239620Smm break; 493239620Smm } else if (label_txg <= txg && label_txg > best_txg) { 494239620Smm best_txg = label_txg; 495239620Smm nvlist_free(config); 496239620Smm config = fnvlist_dup(label); 497239620Smm } 498168404Spjd } 499239620Smm 500239620Smm if (label != NULL) { 501239620Smm nvlist_free(label); 502239620Smm label = NULL; 503239620Smm } 504168404Spjd } 505168404Spjd 506213198Smm if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 507213198Smm flags |= ZIO_FLAG_TRYHARD; 508213198Smm goto retry; 509213198Smm } 510213198Smm 511168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 512168404Spjd 513168404Spjd return (config); 514168404Spjd} 515168404Spjd 516168404Spjd/* 517168404Spjd * Determine if a device is in use. The 'spare_guid' parameter will be filled 518168404Spjd * in with the device guid if this spare is active elsewhere on the system. 519168404Spjd */ 520168404Spjdstatic boolean_t 521168404Spjdvdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 522185029Spjd uint64_t *spare_guid, uint64_t *l2cache_guid) 523168404Spjd{ 524168404Spjd spa_t *spa = vd->vdev_spa; 525168404Spjd uint64_t state, pool_guid, device_guid, txg, spare_pool; 526168404Spjd uint64_t vdtxg = 0; 527168404Spjd nvlist_t *label; 528168404Spjd 529168404Spjd if (spare_guid) 530168404Spjd *spare_guid = 0ULL; 531185029Spjd if (l2cache_guid) 532185029Spjd *l2cache_guid = 0ULL; 533168404Spjd 534168404Spjd /* 535168404Spjd * Read the label, if any, and perform some basic sanity checks. 536168404Spjd */ 537239620Smm if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 538168404Spjd return (B_FALSE); 539168404Spjd 540168404Spjd (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 541168404Spjd &vdtxg); 542168404Spjd 543168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 544168404Spjd &state) != 0 || 545168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 546168404Spjd &device_guid) != 0) { 547168404Spjd nvlist_free(label); 548168404Spjd return (B_FALSE); 549168404Spjd } 550168404Spjd 551185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 552168404Spjd (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 553168404Spjd &pool_guid) != 0 || 554168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 555168404Spjd &txg) != 0)) { 556168404Spjd nvlist_free(label); 557168404Spjd return (B_FALSE); 558168404Spjd } 559168404Spjd 560168404Spjd nvlist_free(label); 561168404Spjd 562168404Spjd /* 563168404Spjd * Check to see if this device indeed belongs to the pool it claims to 564168404Spjd * be a part of. The only way this is allowed is if the device is a hot 565168404Spjd * spare (which we check for later on). 566168404Spjd */ 567185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 568168404Spjd !spa_guid_exists(pool_guid, device_guid) && 569185029Spjd !spa_spare_exists(device_guid, NULL, NULL) && 570185029Spjd !spa_l2cache_exists(device_guid, NULL)) 571168404Spjd return (B_FALSE); 572168404Spjd 573168404Spjd /* 574168404Spjd * If the transaction group is zero, then this an initialized (but 575168404Spjd * unused) label. This is only an error if the create transaction 576168404Spjd * on-disk is the same as the one we're using now, in which case the 577168404Spjd * user has attempted to add the same vdev multiple times in the same 578168404Spjd * transaction. 579168404Spjd */ 580185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 581185029Spjd txg == 0 && vdtxg == crtxg) 582168404Spjd return (B_TRUE); 583168404Spjd 584168404Spjd /* 585168404Spjd * Check to see if this is a spare device. We do an explicit check for 586168404Spjd * spa_has_spare() here because it may be on our pending list of spares 587185029Spjd * to add. We also check if it is an l2cache device. 588168404Spjd */ 589185029Spjd if (spa_spare_exists(device_guid, &spare_pool, NULL) || 590168404Spjd spa_has_spare(spa, device_guid)) { 591168404Spjd if (spare_guid) 592168404Spjd *spare_guid = device_guid; 593168404Spjd 594168404Spjd switch (reason) { 595168404Spjd case VDEV_LABEL_CREATE: 596185029Spjd case VDEV_LABEL_L2CACHE: 597168404Spjd return (B_TRUE); 598168404Spjd 599168404Spjd case VDEV_LABEL_REPLACE: 600168404Spjd return (!spa_has_spare(spa, device_guid) || 601168404Spjd spare_pool != 0ULL); 602168404Spjd 603168404Spjd case VDEV_LABEL_SPARE: 604168404Spjd return (spa_has_spare(spa, device_guid)); 605168404Spjd } 606168404Spjd } 607168404Spjd 608168404Spjd /* 609185029Spjd * Check to see if this is an l2cache device. 610185029Spjd */ 611185029Spjd if (spa_l2cache_exists(device_guid, NULL)) 612185029Spjd return (B_TRUE); 613185029Spjd 614185029Spjd /* 615219089Spjd * We can't rely on a pool's state if it's been imported 616219089Spjd * read-only. Instead we look to see if the pools is marked 617219089Spjd * read-only in the namespace and set the state to active. 618219089Spjd */ 619294805Smav if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 620294805Smav (spa = spa_by_guid(pool_guid, device_guid)) != NULL && 621219089Spjd spa_mode(spa) == FREAD) 622219089Spjd state = POOL_STATE_ACTIVE; 623219089Spjd 624219089Spjd /* 625168404Spjd * If the device is marked ACTIVE, then this device is in use by another 626168404Spjd * pool on the system. 627168404Spjd */ 628168404Spjd return (state == POOL_STATE_ACTIVE); 629168404Spjd} 630168404Spjd 631168404Spjd/* 632168404Spjd * Initialize a vdev label. We check to make sure each leaf device is not in 633168404Spjd * use, and writable. We put down an initial label which we will later 634168404Spjd * overwrite with a complete label. Note that it's important to do this 635168404Spjd * sequentially, not in parallel, so that we catch cases of multiple use of the 636168404Spjd * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 637168404Spjd * itself. 638168404Spjd */ 639168404Spjdint 640168404Spjdvdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 641168404Spjd{ 642168404Spjd spa_t *spa = vd->vdev_spa; 643168404Spjd nvlist_t *label; 644168404Spjd vdev_phys_t *vp; 645209962Smm char *pad2; 646168404Spjd uberblock_t *ub; 647168404Spjd zio_t *zio; 648168404Spjd char *buf; 649168404Spjd size_t buflen; 650168404Spjd int error; 651185029Spjd uint64_t spare_guid, l2cache_guid; 652185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 653168404Spjd 654185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 655168404Spjd 656185029Spjd for (int c = 0; c < vd->vdev_children; c++) 657168404Spjd if ((error = vdev_label_init(vd->vdev_child[c], 658168404Spjd crtxg, reason)) != 0) 659168404Spjd return (error); 660168404Spjd 661219089Spjd /* Track the creation time for this vdev */ 662219089Spjd vd->vdev_crtxg = crtxg; 663219089Spjd 664260157Sdelphij if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) 665168404Spjd return (0); 666168404Spjd 667168404Spjd /* 668168404Spjd * Dead vdevs cannot be initialized. 669168404Spjd */ 670168404Spjd if (vdev_is_dead(vd)) 671249195Smm return (SET_ERROR(EIO)); 672168404Spjd 673168404Spjd /* 674168404Spjd * Determine if the vdev is in use. 675168404Spjd */ 676219089Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 677185029Spjd vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 678249195Smm return (SET_ERROR(EBUSY)); 679168404Spjd 680168404Spjd /* 681185029Spjd * If this is a request to add or replace a spare or l2cache device 682185029Spjd * that is in use elsewhere on the system, then we must update the 683185029Spjd * guid (which was initialized to a random value) to reflect the 684185029Spjd * actual GUID (which is shared between multiple pools). 685168404Spjd */ 686185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 687185029Spjd spare_guid != 0ULL) { 688185029Spjd uint64_t guid_delta = spare_guid - vd->vdev_guid; 689168404Spjd 690185029Spjd vd->vdev_guid += guid_delta; 691168404Spjd 692185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 693185029Spjd pvd->vdev_guid_sum += guid_delta; 694168404Spjd 695168404Spjd /* 696168404Spjd * If this is a replacement, then we want to fallthrough to the 697168404Spjd * rest of the code. If we're adding a spare, then it's already 698185029Spjd * labeled appropriately and we can just return. 699168404Spjd */ 700168404Spjd if (reason == VDEV_LABEL_SPARE) 701168404Spjd return (0); 702219089Spjd ASSERT(reason == VDEV_LABEL_REPLACE || 703219089Spjd reason == VDEV_LABEL_SPLIT); 704168404Spjd } 705168404Spjd 706185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 707185029Spjd l2cache_guid != 0ULL) { 708185029Spjd uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 709185029Spjd 710185029Spjd vd->vdev_guid += guid_delta; 711185029Spjd 712185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 713185029Spjd pvd->vdev_guid_sum += guid_delta; 714185029Spjd 715185029Spjd /* 716185029Spjd * If this is a replacement, then we want to fallthrough to the 717185029Spjd * rest of the code. If we're adding an l2cache, then it's 718185029Spjd * already labeled appropriately and we can just return. 719185029Spjd */ 720185029Spjd if (reason == VDEV_LABEL_L2CACHE) 721185029Spjd return (0); 722185029Spjd ASSERT(reason == VDEV_LABEL_REPLACE); 723185029Spjd } 724185029Spjd 725168404Spjd /* 726240868Spjd * TRIM the whole thing so that we start with a clean slate. 727240868Spjd * It's just an optimization, so we don't care if it fails. 728240868Spjd * Don't TRIM if removing so that we don't interfere with zpool 729240868Spjd * disaster recovery. 730240868Spjd */ 731274619Ssmh if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim && 732274619Ssmh (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE || 733274619Ssmh reason == VDEV_LABEL_L2CACHE)) 734240868Spjd zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize)); 735240868Spjd 736240868Spjd /* 737168404Spjd * Initialize its label. 738168404Spjd */ 739168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 740168404Spjd bzero(vp, sizeof (vdev_phys_t)); 741168404Spjd 742168404Spjd /* 743168404Spjd * Generate a label describing the pool and our top-level vdev. 744168404Spjd * We mark it as being from txg 0 to indicate that it's not 745168404Spjd * really part of an active pool just yet. The labels will 746168404Spjd * be written again with a meaningful txg by spa_sync(). 747168404Spjd */ 748168404Spjd if (reason == VDEV_LABEL_SPARE || 749168404Spjd (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 750168404Spjd /* 751168404Spjd * For inactive hot spares, we generate a special label that 752168404Spjd * identifies as a mutually shared hot spare. We write the 753168404Spjd * label if we are adding a hot spare, or if we are removing an 754168404Spjd * active hot spare (in which case we want to revert the 755168404Spjd * labels). 756168404Spjd */ 757168404Spjd VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 758168404Spjd 759168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 760168404Spjd spa_version(spa)) == 0); 761168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 762168404Spjd POOL_STATE_SPARE) == 0); 763168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 764168404Spjd vd->vdev_guid) == 0); 765185029Spjd } else if (reason == VDEV_LABEL_L2CACHE || 766185029Spjd (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 767185029Spjd /* 768185029Spjd * For level 2 ARC devices, add a special label. 769185029Spjd */ 770185029Spjd VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 771185029Spjd 772185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 773185029Spjd spa_version(spa)) == 0); 774185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 775185029Spjd POOL_STATE_L2CACHE) == 0); 776185029Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 777185029Spjd vd->vdev_guid) == 0); 778168404Spjd } else { 779219089Spjd uint64_t txg = 0ULL; 780168404Spjd 781219089Spjd if (reason == VDEV_LABEL_SPLIT) 782219089Spjd txg = spa->spa_uberblock.ub_txg; 783219089Spjd label = spa_config_generate(spa, vd, txg, B_FALSE); 784219089Spjd 785168404Spjd /* 786168404Spjd * Add our creation time. This allows us to detect multiple 787168404Spjd * vdev uses as described above, and automatically expires if we 788168404Spjd * fail. 789168404Spjd */ 790168404Spjd VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 791168404Spjd crtxg) == 0); 792168404Spjd } 793168404Spjd 794168404Spjd buf = vp->vp_nvlist; 795168404Spjd buflen = sizeof (vp->vp_nvlist); 796168404Spjd 797168404Spjd error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 798168404Spjd if (error != 0) { 799168404Spjd nvlist_free(label); 800168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 801168404Spjd /* EFAULT means nvlist_pack ran out of room */ 802168404Spjd return (error == EFAULT ? ENAMETOOLONG : EINVAL); 803168404Spjd } 804168404Spjd 805168404Spjd /* 806168404Spjd * Initialize uberblock template. 807168404Spjd */ 808219089Spjd ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 809219089Spjd bzero(ub, VDEV_UBERBLOCK_RING); 810168404Spjd *ub = spa->spa_uberblock; 811168404Spjd ub->ub_txg = 0; 812168404Spjd 813209962Smm /* Initialize the 2nd padding area. */ 814209962Smm pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 815209962Smm bzero(pad2, VDEV_PAD_SIZE); 816209962Smm 817168404Spjd /* 818168404Spjd * Write everything in parallel. 819168404Spjd */ 820213198Smmretry: 821185029Spjd zio = zio_root(spa, NULL, NULL, flags); 822168404Spjd 823185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 824168404Spjd 825168404Spjd vdev_label_write(zio, vd, l, vp, 826168404Spjd offsetof(vdev_label_t, vl_vdev_phys), 827185029Spjd sizeof (vdev_phys_t), NULL, NULL, flags); 828168404Spjd 829209962Smm /* 830209962Smm * Skip the 1st padding area. 831209962Smm * Zero out the 2nd padding area where it might have 832209962Smm * left over data from previous filesystem format. 833209962Smm */ 834209962Smm vdev_label_write(zio, vd, l, pad2, 835209962Smm offsetof(vdev_label_t, vl_pad2), 836209962Smm VDEV_PAD_SIZE, NULL, NULL, flags); 837168404Spjd 838219089Spjd vdev_label_write(zio, vd, l, ub, 839219089Spjd offsetof(vdev_label_t, vl_uberblock), 840219089Spjd VDEV_UBERBLOCK_RING, NULL, NULL, flags); 841168404Spjd } 842168404Spjd 843168404Spjd error = zio_wait(zio); 844168404Spjd 845213198Smm if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 846213198Smm flags |= ZIO_FLAG_TRYHARD; 847213198Smm goto retry; 848213198Smm } 849213198Smm 850168404Spjd nvlist_free(label); 851209962Smm zio_buf_free(pad2, VDEV_PAD_SIZE); 852219089Spjd zio_buf_free(ub, VDEV_UBERBLOCK_RING); 853168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 854168404Spjd 855168404Spjd /* 856168404Spjd * If this vdev hasn't been previously identified as a spare, then we 857185029Spjd * mark it as such only if a) we are labeling it as a spare, or b) it 858185029Spjd * exists as a spare elsewhere in the system. Do the same for 859185029Spjd * level 2 ARC devices. 860168404Spjd */ 861168404Spjd if (error == 0 && !vd->vdev_isspare && 862168404Spjd (reason == VDEV_LABEL_SPARE || 863185029Spjd spa_spare_exists(vd->vdev_guid, NULL, NULL))) 864168404Spjd spa_spare_add(vd); 865168404Spjd 866185029Spjd if (error == 0 && !vd->vdev_isl2cache && 867185029Spjd (reason == VDEV_LABEL_L2CACHE || 868185029Spjd spa_l2cache_exists(vd->vdev_guid, NULL))) 869185029Spjd spa_l2cache_add(vd); 870185029Spjd 871168404Spjd return (error); 872168404Spjd} 873168404Spjd 874168404Spjd/* 875168404Spjd * ========================================================================== 876168404Spjd * uberblock load/sync 877168404Spjd * ========================================================================== 878168404Spjd */ 879168404Spjd 880168404Spjd/* 881168404Spjd * Consider the following situation: txg is safely synced to disk. We've 882168404Spjd * written the first uberblock for txg + 1, and then we lose power. When we 883168404Spjd * come back up, we fail to see the uberblock for txg + 1 because, say, 884168404Spjd * it was on a mirrored device and the replica to which we wrote txg + 1 885168404Spjd * is now offline. If we then make some changes and sync txg + 1, and then 886236884Smm * the missing replica comes back, then for a few seconds we'll have two 887168404Spjd * conflicting uberblocks on disk with the same txg. The solution is simple: 888168404Spjd * among uberblocks with equal txg, choose the one with the latest timestamp. 889168404Spjd */ 890168404Spjdstatic int 891168404Spjdvdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 892168404Spjd{ 893168404Spjd if (ub1->ub_txg < ub2->ub_txg) 894168404Spjd return (-1); 895168404Spjd if (ub1->ub_txg > ub2->ub_txg) 896168404Spjd return (1); 897168404Spjd 898168404Spjd if (ub1->ub_timestamp < ub2->ub_timestamp) 899168404Spjd return (-1); 900168404Spjd if (ub1->ub_timestamp > ub2->ub_timestamp) 901168404Spjd return (1); 902168404Spjd 903168404Spjd return (0); 904168404Spjd} 905168404Spjd 906236884Smmstruct ubl_cbdata { 907236884Smm uberblock_t *ubl_ubbest; /* Best uberblock */ 908236884Smm vdev_t *ubl_vd; /* vdev associated with the above */ 909236884Smm}; 910236884Smm 911168404Spjdstatic void 912168404Spjdvdev_uberblock_load_done(zio_t *zio) 913168404Spjd{ 914236884Smm vdev_t *vd = zio->io_vd; 915219089Spjd spa_t *spa = zio->io_spa; 916185029Spjd zio_t *rio = zio->io_private; 917168404Spjd uberblock_t *ub = zio->io_data; 918236884Smm struct ubl_cbdata *cbp = rio->io_private; 919168404Spjd 920236884Smm ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); 921168404Spjd 922168404Spjd if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 923185029Spjd mutex_enter(&rio->io_lock); 924219089Spjd if (ub->ub_txg <= spa->spa_load_max_txg && 925236884Smm vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { 926236884Smm /* 927239620Smm * Keep track of the vdev in which this uberblock 928239620Smm * was found. We will use this information later 929239620Smm * to obtain the config nvlist associated with 930236884Smm * this uberblock. 931236884Smm */ 932236884Smm *cbp->ubl_ubbest = *ub; 933236884Smm cbp->ubl_vd = vd; 934236884Smm } 935185029Spjd mutex_exit(&rio->io_lock); 936168404Spjd } 937168404Spjd 938168404Spjd zio_buf_free(zio->io_data, zio->io_size); 939168404Spjd} 940168404Spjd 941236884Smmstatic void 942236884Smmvdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, 943236884Smm struct ubl_cbdata *cbp) 944168404Spjd{ 945185029Spjd for (int c = 0; c < vd->vdev_children; c++) 946236884Smm vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); 947168404Spjd 948185029Spjd if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 949185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 950185029Spjd for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 951185029Spjd vdev_label_read(zio, vd, l, 952185029Spjd zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 953185029Spjd VDEV_UBERBLOCK_OFFSET(vd, n), 954185029Spjd VDEV_UBERBLOCK_SIZE(vd), 955185029Spjd vdev_uberblock_load_done, zio, flags); 956185029Spjd } 957168404Spjd } 958168404Spjd } 959236884Smm} 960185029Spjd 961236884Smm/* 962236884Smm * Reads the 'best' uberblock from disk along with its associated 963236884Smm * configuration. First, we read the uberblock array of each label of each 964236884Smm * vdev, keeping track of the uberblock with the highest txg in each array. 965239620Smm * Then, we read the configuration from the same vdev as the best uberblock. 966236884Smm */ 967236884Smmvoid 968236884Smmvdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) 969236884Smm{ 970236884Smm zio_t *zio; 971236884Smm spa_t *spa = rvd->vdev_spa; 972236884Smm struct ubl_cbdata cb; 973236884Smm int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 974236884Smm ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 975236884Smm 976236884Smm ASSERT(ub); 977236884Smm ASSERT(config); 978236884Smm 979236884Smm bzero(ub, sizeof (uberblock_t)); 980236884Smm *config = NULL; 981236884Smm 982236884Smm cb.ubl_ubbest = ub; 983236884Smm cb.ubl_vd = NULL; 984236884Smm 985236884Smm spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 986236884Smm zio = zio_root(spa, NULL, &cb, flags); 987236884Smm vdev_uberblock_load_impl(zio, rvd, flags, &cb); 988236884Smm (void) zio_wait(zio); 989239620Smm 990239620Smm /* 991239620Smm * It's possible that the best uberblock was discovered on a label 992239620Smm * that has a configuration which was written in a future txg. 993239620Smm * Search all labels on this vdev to find the configuration that 994239620Smm * matches the txg for our uberblock. 995239620Smm */ 996239620Smm if (cb.ubl_vd != NULL) 997239620Smm *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); 998236884Smm spa_config_exit(spa, SCL_ALL, FTAG); 999168404Spjd} 1000168404Spjd 1001168404Spjd/* 1002185029Spjd * On success, increment root zio's count of good writes. 1003168404Spjd * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 1004168404Spjd */ 1005168404Spjdstatic void 1006168404Spjdvdev_uberblock_sync_done(zio_t *zio) 1007168404Spjd{ 1008185029Spjd uint64_t *good_writes = zio->io_private; 1009168404Spjd 1010168404Spjd if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 1011270247Sdelphij atomic_inc_64(good_writes); 1012168404Spjd} 1013168404Spjd 1014185029Spjd/* 1015185029Spjd * Write the uberblock to all labels of all leaves of the specified vdev. 1016185029Spjd */ 1017168404Spjdstatic void 1018185029Spjdvdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 1019168404Spjd{ 1020185029Spjd uberblock_t *ubbuf; 1021185029Spjd int n; 1022168404Spjd 1023185029Spjd for (int c = 0; c < vd->vdev_children; c++) 1024185029Spjd vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 1025168404Spjd 1026168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1027168404Spjd return; 1028168404Spjd 1029185029Spjd if (!vdev_writeable(vd)) 1030168404Spjd return; 1031168404Spjd 1032185029Spjd n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 1033168404Spjd 1034185029Spjd ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 1035185029Spjd bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 1036185029Spjd *ubbuf = *ub; 1037168404Spjd 1038185029Spjd for (int l = 0; l < VDEV_LABELS; l++) 1039185029Spjd vdev_label_write(zio, vd, l, ubbuf, 1040185029Spjd VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 1041185029Spjd vdev_uberblock_sync_done, zio->io_private, 1042185029Spjd flags | ZIO_FLAG_DONT_PROPAGATE); 1043168404Spjd 1044185029Spjd zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 1045168404Spjd} 1046168404Spjd 1047251629Sdelphij/* Sync the uberblocks to all vdevs in svd[] */ 1048185029Spjdint 1049185029Spjdvdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1050168404Spjd{ 1051185029Spjd spa_t *spa = svd[0]->vdev_spa; 1052168404Spjd zio_t *zio; 1053185029Spjd uint64_t good_writes = 0; 1054168404Spjd 1055185029Spjd zio = zio_root(spa, NULL, &good_writes, flags); 1056168404Spjd 1057185029Spjd for (int v = 0; v < svdcount; v++) 1058185029Spjd vdev_uberblock_sync(zio, ub, svd[v], flags); 1059168404Spjd 1060185029Spjd (void) zio_wait(zio); 1061168404Spjd 1062168404Spjd /* 1063185029Spjd * Flush the uberblocks to disk. This ensures that the odd labels 1064185029Spjd * are no longer needed (because the new uberblocks and the even 1065185029Spjd * labels are safely on disk), so it is safe to overwrite them. 1066168404Spjd */ 1067185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1068168404Spjd 1069185029Spjd for (int v = 0; v < svdcount; v++) 1070185029Spjd zio_flush(zio, svd[v]); 1071168404Spjd 1072185029Spjd (void) zio_wait(zio); 1073185029Spjd 1074185029Spjd return (good_writes >= 1 ? 0 : EIO); 1075168404Spjd} 1076168404Spjd 1077168404Spjd/* 1078185029Spjd * On success, increment the count of good writes for our top-level vdev. 1079168404Spjd */ 1080168404Spjdstatic void 1081185029Spjdvdev_label_sync_done(zio_t *zio) 1082168404Spjd{ 1083185029Spjd uint64_t *good_writes = zio->io_private; 1084168404Spjd 1085168404Spjd if (zio->io_error == 0) 1086270247Sdelphij atomic_inc_64(good_writes); 1087168404Spjd} 1088168404Spjd 1089185029Spjd/* 1090185029Spjd * If there weren't enough good writes, indicate failure to the parent. 1091185029Spjd */ 1092168404Spjdstatic void 1093185029Spjdvdev_label_sync_top_done(zio_t *zio) 1094168404Spjd{ 1095185029Spjd uint64_t *good_writes = zio->io_private; 1096185029Spjd 1097185029Spjd if (*good_writes == 0) 1098249195Smm zio->io_error = SET_ERROR(EIO); 1099185029Spjd 1100185029Spjd kmem_free(good_writes, sizeof (uint64_t)); 1101185029Spjd} 1102185029Spjd 1103185029Spjd/* 1104185029Spjd * We ignore errors for log and cache devices, simply free the private data. 1105185029Spjd */ 1106185029Spjdstatic void 1107185029Spjdvdev_label_sync_ignore_done(zio_t *zio) 1108185029Spjd{ 1109185029Spjd kmem_free(zio->io_private, sizeof (uint64_t)); 1110185029Spjd} 1111185029Spjd 1112185029Spjd/* 1113185029Spjd * Write all even or odd labels to all leaves of the specified vdev. 1114185029Spjd */ 1115185029Spjdstatic void 1116185029Spjdvdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1117185029Spjd{ 1118168404Spjd nvlist_t *label; 1119168404Spjd vdev_phys_t *vp; 1120168404Spjd char *buf; 1121168404Spjd size_t buflen; 1122168404Spjd 1123185029Spjd for (int c = 0; c < vd->vdev_children; c++) 1124185029Spjd vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1125168404Spjd 1126168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1127168404Spjd return; 1128168404Spjd 1129185029Spjd if (!vdev_writeable(vd)) 1130168404Spjd return; 1131168404Spjd 1132168404Spjd /* 1133168404Spjd * Generate a label describing the top-level config to which we belong. 1134168404Spjd */ 1135168404Spjd label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1136168404Spjd 1137168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1138168404Spjd bzero(vp, sizeof (vdev_phys_t)); 1139168404Spjd 1140168404Spjd buf = vp->vp_nvlist; 1141168404Spjd buflen = sizeof (vp->vp_nvlist); 1142168404Spjd 1143185029Spjd if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1144185029Spjd for (; l < VDEV_LABELS; l += 2) { 1145185029Spjd vdev_label_write(zio, vd, l, vp, 1146185029Spjd offsetof(vdev_label_t, vl_vdev_phys), 1147185029Spjd sizeof (vdev_phys_t), 1148185029Spjd vdev_label_sync_done, zio->io_private, 1149185029Spjd flags | ZIO_FLAG_DONT_PROPAGATE); 1150185029Spjd } 1151185029Spjd } 1152168404Spjd 1153168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 1154168404Spjd nvlist_free(label); 1155168404Spjd} 1156168404Spjd 1157185029Spjdint 1158185029Spjdvdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1159168404Spjd{ 1160185029Spjd list_t *dl = &spa->spa_config_dirty_list; 1161185029Spjd vdev_t *vd; 1162168404Spjd zio_t *zio; 1163168404Spjd int error; 1164168404Spjd 1165185029Spjd /* 1166185029Spjd * Write the new labels to disk. 1167185029Spjd */ 1168185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1169168404Spjd 1170185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1171185029Spjd uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1172185029Spjd KM_SLEEP); 1173219089Spjd 1174219089Spjd ASSERT(!vd->vdev_ishole); 1175219089Spjd 1176209962Smm zio_t *vio = zio_null(zio, spa, NULL, 1177185029Spjd (vd->vdev_islog || vd->vdev_aux != NULL) ? 1178185029Spjd vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1179185029Spjd good_writes, flags); 1180185029Spjd vdev_label_sync(vio, vd, l, txg, flags); 1181185029Spjd zio_nowait(vio); 1182185029Spjd } 1183168404Spjd 1184185029Spjd error = zio_wait(zio); 1185168404Spjd 1186168404Spjd /* 1187185029Spjd * Flush the new labels to disk. 1188168404Spjd */ 1189185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1190168404Spjd 1191185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1192185029Spjd zio_flush(zio, vd); 1193168404Spjd 1194185029Spjd (void) zio_wait(zio); 1195168404Spjd 1196168404Spjd return (error); 1197168404Spjd} 1198168404Spjd 1199168404Spjd/* 1200185029Spjd * Sync the uberblock and any changes to the vdev configuration. 1201168404Spjd * 1202168404Spjd * The order of operations is carefully crafted to ensure that 1203168404Spjd * if the system panics or loses power at any time, the state on disk 1204168404Spjd * is still transactionally consistent. The in-line comments below 1205168404Spjd * describe the failure semantics at each stage. 1206168404Spjd * 1207185029Spjd * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1208168404Spjd * at any time, you can just call it again, and it will resume its work. 1209168404Spjd */ 1210168404Spjdint 1211294811Smavvdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1212168404Spjd{ 1213185029Spjd spa_t *spa = svd[0]->vdev_spa; 1214168404Spjd uberblock_t *ub = &spa->spa_uberblock; 1215168404Spjd vdev_t *vd; 1216168404Spjd zio_t *zio; 1217294811Smav int error = 0; 1218185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1219168404Spjd 1220294811Smavretry: 1221213198Smm /* 1222213198Smm * Normally, we don't want to try too hard to write every label and 1223213198Smm * uberblock. If there is a flaky disk, we don't want the rest of the 1224213198Smm * sync process to block while we retry. But if we can't write a 1225213198Smm * single label out, we should retry with ZIO_FLAG_TRYHARD before 1226213198Smm * bailing out and declaring the pool faulted. 1227213198Smm */ 1228294811Smav if (error != 0) { 1229294811Smav if ((flags & ZIO_FLAG_TRYHARD) != 0) 1230294811Smav return (error); 1231213198Smm flags |= ZIO_FLAG_TRYHARD; 1232294811Smav } 1233213198Smm 1234168404Spjd ASSERT(ub->ub_txg <= txg); 1235168404Spjd 1236168404Spjd /* 1237185029Spjd * If this isn't a resync due to I/O errors, 1238185029Spjd * and nothing changed in this transaction group, 1239185029Spjd * and the vdev configuration hasn't changed, 1240168404Spjd * then there's nothing to do. 1241168404Spjd */ 1242185029Spjd if (ub->ub_txg < txg && 1243185029Spjd uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1244185029Spjd list_is_empty(&spa->spa_config_dirty_list)) 1245168404Spjd return (0); 1246168404Spjd 1247168404Spjd if (txg > spa_freeze_txg(spa)) 1248168404Spjd return (0); 1249168404Spjd 1250168404Spjd ASSERT(txg <= spa->spa_final_txg); 1251168404Spjd 1252168404Spjd /* 1253168404Spjd * Flush the write cache of every disk that's been written to 1254168404Spjd * in this transaction group. This ensures that all blocks 1255168404Spjd * written in this txg will be committed to stable storage 1256168404Spjd * before any uberblock that references them. 1257168404Spjd */ 1258185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1259185029Spjd 1260168404Spjd for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 1261185029Spjd vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1262185029Spjd zio_flush(zio, vd); 1263185029Spjd 1264168404Spjd (void) zio_wait(zio); 1265168404Spjd 1266168404Spjd /* 1267168404Spjd * Sync out the even labels (L0, L2) for every dirty vdev. If the 1268168404Spjd * system dies in the middle of this process, that's OK: all of the 1269168404Spjd * even labels that made it to disk will be newer than any uberblock, 1270168404Spjd * and will therefore be considered invalid. The odd labels (L1, L3), 1271185029Spjd * which have not yet been touched, will still be valid. We flush 1272185029Spjd * the new labels to disk to ensure that all even-label updates 1273185029Spjd * are committed to stable storage before the uberblock update. 1274168404Spjd */ 1275185029Spjd if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1276294811Smav goto retry; 1277168404Spjd 1278168404Spjd /* 1279185029Spjd * Sync the uberblocks to all vdevs in svd[]. 1280168404Spjd * If the system dies in the middle of this step, there are two cases 1281168404Spjd * to consider, and the on-disk state is consistent either way: 1282168404Spjd * 1283168404Spjd * (1) If none of the new uberblocks made it to disk, then the 1284168404Spjd * previous uberblock will be the newest, and the odd labels 1285168404Spjd * (which had not yet been touched) will be valid with respect 1286168404Spjd * to that uberblock. 1287168404Spjd * 1288168404Spjd * (2) If one or more new uberblocks made it to disk, then they 1289168404Spjd * will be the newest, and the even labels (which had all 1290168404Spjd * been successfully committed) will be valid with respect 1291168404Spjd * to the new uberblocks. 1292168404Spjd */ 1293185029Spjd if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1294294811Smav goto retry; 1295168404Spjd 1296168404Spjd /* 1297168404Spjd * Sync out odd labels for every dirty vdev. If the system dies 1298168404Spjd * in the middle of this process, the even labels and the new 1299168404Spjd * uberblocks will suffice to open the pool. The next time 1300168404Spjd * the pool is opened, the first thing we'll do -- before any 1301168404Spjd * user data is modified -- is mark every vdev dirty so that 1302185029Spjd * all labels will be brought up to date. We flush the new labels 1303185029Spjd * to disk to ensure that all odd-label updates are committed to 1304185029Spjd * stable storage before the next transaction group begins. 1305168404Spjd */ 1306240868Spjd if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) 1307294811Smav goto retry;; 1308240868Spjd 1309240868Spjd trim_thread_wakeup(spa); 1310240868Spjd 1311240868Spjd return (0); 1312168404Spjd} 1313