vdev_label.c revision 249195
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. 24249195Smm * Copyright (c) 2013 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, 188185029Spjd 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, 202185029Spjd 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 225168404Spjd VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 226168404Spjd 227168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 228168404Spjd vd->vdev_ops->vdev_op_type) == 0); 229219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 230168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 231168404Spjd == 0); 232168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 233168404Spjd 234168404Spjd if (vd->vdev_path != NULL) 235168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 236168404Spjd vd->vdev_path) == 0); 237168404Spjd 238168404Spjd if (vd->vdev_devid != NULL) 239168404Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 240168404Spjd vd->vdev_devid) == 0); 241168404Spjd 242185029Spjd if (vd->vdev_physpath != NULL) 243185029Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 244185029Spjd vd->vdev_physpath) == 0); 245185029Spjd 246209962Smm if (vd->vdev_fru != NULL) 247209962Smm VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, 248209962Smm vd->vdev_fru) == 0); 249209962Smm 250168404Spjd if (vd->vdev_nparity != 0) { 251168404Spjd ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 252168404Spjd VDEV_TYPE_RAIDZ) == 0); 253168404Spjd 254168404Spjd /* 255168404Spjd * Make sure someone hasn't managed to sneak a fancy new vdev 256168404Spjd * into a crufty old storage pool. 257168404Spjd */ 258168404Spjd ASSERT(vd->vdev_nparity == 1 || 259219089Spjd (vd->vdev_nparity <= 2 && 260219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ2) || 261219089Spjd (vd->vdev_nparity <= 3 && 262219089Spjd spa_version(spa) >= SPA_VERSION_RAIDZ3)); 263168404Spjd 264168404Spjd /* 265168404Spjd * Note that we'll add the nparity tag even on storage pools 266168404Spjd * that only support a single parity device -- older software 267168404Spjd * will just ignore it. 268168404Spjd */ 269168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 270168404Spjd vd->vdev_nparity) == 0); 271168404Spjd } 272168404Spjd 273168404Spjd if (vd->vdev_wholedisk != -1ULL) 274168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 275168404Spjd vd->vdev_wholedisk) == 0); 276168404Spjd 277168404Spjd if (vd->vdev_not_present) 278168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 279168404Spjd 280168404Spjd if (vd->vdev_isspare) 281168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 282168404Spjd 283219089Spjd if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 284219089Spjd vd == vd->vdev_top) { 285168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 286168404Spjd vd->vdev_ms_array) == 0); 287168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 288168404Spjd vd->vdev_ms_shift) == 0); 289168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 290168404Spjd vd->vdev_ashift) == 0); 291168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 292168404Spjd vd->vdev_asize) == 0); 293185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 294185029Spjd vd->vdev_islog) == 0); 295219089Spjd if (vd->vdev_removing) 296219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 297219089Spjd vd->vdev_removing) == 0); 298168404Spjd } 299168404Spjd 300209962Smm if (vd->vdev_dtl_smo.smo_object != 0) 301168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 302209962Smm vd->vdev_dtl_smo.smo_object) == 0); 303168404Spjd 304219089Spjd if (vd->vdev_crtxg) 305219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 306219089Spjd vd->vdev_crtxg) == 0); 307219089Spjd 308168404Spjd if (getstats) { 309168404Spjd vdev_stat_t vs; 310219089Spjd pool_scan_stat_t ps; 311219089Spjd 312168404Spjd vdev_get_stats(vd, &vs); 313219089Spjd VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 314168404Spjd (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 315219089Spjd 316219089Spjd /* provide either current or previous scan information */ 317219089Spjd if (spa_scan_get_stats(spa, &ps) == 0) { 318219089Spjd VERIFY(nvlist_add_uint64_array(nv, 319219089Spjd ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 320219089Spjd sizeof (pool_scan_stat_t) / sizeof (uint64_t)) 321219089Spjd == 0); 322219089Spjd } 323168404Spjd } 324168404Spjd 325168404Spjd if (!vd->vdev_ops->vdev_op_leaf) { 326168404Spjd nvlist_t **child; 327219089Spjd int c, idx; 328168404Spjd 329219089Spjd ASSERT(!vd->vdev_ishole); 330219089Spjd 331168404Spjd child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 332168404Spjd KM_SLEEP); 333168404Spjd 334219089Spjd for (c = 0, idx = 0; c < vd->vdev_children; c++) { 335219089Spjd vdev_t *cvd = vd->vdev_child[c]; 336168404Spjd 337219089Spjd /* 338219089Spjd * If we're generating an nvlist of removing 339219089Spjd * vdevs then skip over any device which is 340219089Spjd * not being removed. 341219089Spjd */ 342219089Spjd if ((flags & VDEV_CONFIG_REMOVING) && 343219089Spjd !cvd->vdev_removing) 344219089Spjd continue; 345168404Spjd 346219089Spjd child[idx++] = vdev_config_generate(spa, cvd, 347219089Spjd getstats, flags); 348219089Spjd } 349219089Spjd 350219089Spjd if (idx) { 351219089Spjd VERIFY(nvlist_add_nvlist_array(nv, 352219089Spjd ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 353219089Spjd } 354219089Spjd 355219089Spjd for (c = 0; c < idx; c++) 356168404Spjd nvlist_free(child[c]); 357168404Spjd 358168404Spjd kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 359168404Spjd 360168404Spjd } else { 361219089Spjd const char *aux = NULL; 362219089Spjd 363168404Spjd if (vd->vdev_offline && !vd->vdev_tmpoffline) 364168404Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 365168404Spjd B_TRUE) == 0); 366219089Spjd if (vd->vdev_resilvering) 367219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING, 368219089Spjd B_TRUE) == 0); 369185029Spjd if (vd->vdev_faulted) 370185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 371185029Spjd B_TRUE) == 0); 372185029Spjd if (vd->vdev_degraded) 373185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 374185029Spjd B_TRUE) == 0); 375185029Spjd if (vd->vdev_removed) 376185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 377185029Spjd B_TRUE) == 0); 378185029Spjd if (vd->vdev_unspare) 379185029Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 380185029Spjd B_TRUE) == 0); 381219089Spjd if (vd->vdev_ishole) 382219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, 383219089Spjd B_TRUE) == 0); 384219089Spjd 385219089Spjd switch (vd->vdev_stat.vs_aux) { 386219089Spjd case VDEV_AUX_ERR_EXCEEDED: 387219089Spjd aux = "err_exceeded"; 388219089Spjd break; 389219089Spjd 390219089Spjd case VDEV_AUX_EXTERNAL: 391219089Spjd aux = "external"; 392219089Spjd break; 393219089Spjd } 394219089Spjd 395219089Spjd if (aux != NULL) 396219089Spjd VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, 397219089Spjd aux) == 0); 398219089Spjd 399219089Spjd if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 400219089Spjd VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 401219089Spjd vd->vdev_orig_guid) == 0); 402219089Spjd } 403168404Spjd } 404168404Spjd 405168404Spjd return (nv); 406168404Spjd} 407168404Spjd 408219089Spjd/* 409219089Spjd * Generate a view of the top-level vdevs. If we currently have holes 410219089Spjd * in the namespace, then generate an array which contains a list of holey 411219089Spjd * vdevs. Additionally, add the number of top-level children that currently 412219089Spjd * exist. 413219089Spjd */ 414219089Spjdvoid 415219089Spjdvdev_top_config_generate(spa_t *spa, nvlist_t *config) 416219089Spjd{ 417219089Spjd vdev_t *rvd = spa->spa_root_vdev; 418219089Spjd uint64_t *array; 419219089Spjd uint_t c, idx; 420219089Spjd 421219089Spjd array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 422219089Spjd 423219089Spjd for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 424219089Spjd vdev_t *tvd = rvd->vdev_child[c]; 425219089Spjd 426219089Spjd if (tvd->vdev_ishole) 427219089Spjd array[idx++] = c; 428219089Spjd } 429219089Spjd 430219089Spjd if (idx) { 431219089Spjd VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 432219089Spjd array, idx) == 0); 433219089Spjd } 434219089Spjd 435219089Spjd VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 436219089Spjd rvd->vdev_children) == 0); 437219089Spjd 438219089Spjd kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 439219089Spjd} 440219089Spjd 441236884Smm/* 442239620Smm * Returns the configuration from the label of the given vdev. For vdevs 443239620Smm * which don't have a txg value stored on their label (i.e. spares/cache) 444239620Smm * or have not been completely initialized (txg = 0) just return 445239620Smm * the configuration from the first valid label we find. Otherwise, 446239620Smm * find the most up-to-date label that does not exceed the specified 447239620Smm * 'txg' value. 448236884Smm */ 449168404Spjdnvlist_t * 450239620Smmvdev_label_read_config(vdev_t *vd, uint64_t txg) 451168404Spjd{ 452168404Spjd spa_t *spa = vd->vdev_spa; 453168404Spjd nvlist_t *config = NULL; 454168404Spjd vdev_phys_t *vp; 455168404Spjd zio_t *zio; 456239620Smm uint64_t best_txg = 0; 457239620Smm int error = 0; 458213198Smm int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 459213198Smm ZIO_FLAG_SPECULATIVE; 460168404Spjd 461185029Spjd ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 462168404Spjd 463185029Spjd if (!vdev_readable(vd)) 464168404Spjd return (NULL); 465168404Spjd 466168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 467168404Spjd 468213198Smmretry: 469185029Spjd for (int l = 0; l < VDEV_LABELS; l++) { 470239620Smm nvlist_t *label = NULL; 471168404Spjd 472185029Spjd zio = zio_root(spa, NULL, NULL, flags); 473168404Spjd 474168404Spjd vdev_label_read(zio, vd, l, vp, 475168404Spjd offsetof(vdev_label_t, vl_vdev_phys), 476185029Spjd sizeof (vdev_phys_t), NULL, NULL, flags); 477168404Spjd 478168404Spjd if (zio_wait(zio) == 0 && 479168404Spjd nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 480239620Smm &label, 0) == 0) { 481239620Smm uint64_t label_txg = 0; 482168404Spjd 483239620Smm /* 484239620Smm * Auxiliary vdevs won't have txg values in their 485239620Smm * labels and newly added vdevs may not have been 486239620Smm * completely initialized so just return the 487239620Smm * configuration from the first valid label we 488239620Smm * encounter. 489239620Smm */ 490239620Smm error = nvlist_lookup_uint64(label, 491239620Smm ZPOOL_CONFIG_POOL_TXG, &label_txg); 492239620Smm if ((error || label_txg == 0) && !config) { 493239620Smm config = label; 494239620Smm break; 495239620Smm } else if (label_txg <= txg && label_txg > best_txg) { 496239620Smm best_txg = label_txg; 497239620Smm nvlist_free(config); 498239620Smm config = fnvlist_dup(label); 499239620Smm } 500168404Spjd } 501239620Smm 502239620Smm if (label != NULL) { 503239620Smm nvlist_free(label); 504239620Smm label = NULL; 505239620Smm } 506168404Spjd } 507168404Spjd 508213198Smm if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 509213198Smm flags |= ZIO_FLAG_TRYHARD; 510213198Smm goto retry; 511213198Smm } 512213198Smm 513168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 514168404Spjd 515168404Spjd return (config); 516168404Spjd} 517168404Spjd 518168404Spjd/* 519168404Spjd * Determine if a device is in use. The 'spare_guid' parameter will be filled 520168404Spjd * in with the device guid if this spare is active elsewhere on the system. 521168404Spjd */ 522168404Spjdstatic boolean_t 523168404Spjdvdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 524185029Spjd uint64_t *spare_guid, uint64_t *l2cache_guid) 525168404Spjd{ 526168404Spjd spa_t *spa = vd->vdev_spa; 527168404Spjd uint64_t state, pool_guid, device_guid, txg, spare_pool; 528168404Spjd uint64_t vdtxg = 0; 529168404Spjd nvlist_t *label; 530168404Spjd 531168404Spjd if (spare_guid) 532168404Spjd *spare_guid = 0ULL; 533185029Spjd if (l2cache_guid) 534185029Spjd *l2cache_guid = 0ULL; 535168404Spjd 536168404Spjd /* 537168404Spjd * Read the label, if any, and perform some basic sanity checks. 538168404Spjd */ 539239620Smm if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) 540168404Spjd return (B_FALSE); 541168404Spjd 542168404Spjd (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 543168404Spjd &vdtxg); 544168404Spjd 545168404Spjd if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 546168404Spjd &state) != 0 || 547168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 548168404Spjd &device_guid) != 0) { 549168404Spjd nvlist_free(label); 550168404Spjd return (B_FALSE); 551168404Spjd } 552168404Spjd 553185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 554168404Spjd (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 555168404Spjd &pool_guid) != 0 || 556168404Spjd nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 557168404Spjd &txg) != 0)) { 558168404Spjd nvlist_free(label); 559168404Spjd return (B_FALSE); 560168404Spjd } 561168404Spjd 562168404Spjd nvlist_free(label); 563168404Spjd 564168404Spjd /* 565168404Spjd * Check to see if this device indeed belongs to the pool it claims to 566168404Spjd * be a part of. The only way this is allowed is if the device is a hot 567168404Spjd * spare (which we check for later on). 568168404Spjd */ 569185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 570168404Spjd !spa_guid_exists(pool_guid, device_guid) && 571185029Spjd !spa_spare_exists(device_guid, NULL, NULL) && 572185029Spjd !spa_l2cache_exists(device_guid, NULL)) 573168404Spjd return (B_FALSE); 574168404Spjd 575168404Spjd /* 576168404Spjd * If the transaction group is zero, then this an initialized (but 577168404Spjd * unused) label. This is only an error if the create transaction 578168404Spjd * on-disk is the same as the one we're using now, in which case the 579168404Spjd * user has attempted to add the same vdev multiple times in the same 580168404Spjd * transaction. 581168404Spjd */ 582185029Spjd if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 583185029Spjd txg == 0 && vdtxg == crtxg) 584168404Spjd return (B_TRUE); 585168404Spjd 586168404Spjd /* 587168404Spjd * Check to see if this is a spare device. We do an explicit check for 588168404Spjd * spa_has_spare() here because it may be on our pending list of spares 589185029Spjd * to add. We also check if it is an l2cache device. 590168404Spjd */ 591185029Spjd if (spa_spare_exists(device_guid, &spare_pool, NULL) || 592168404Spjd spa_has_spare(spa, device_guid)) { 593168404Spjd if (spare_guid) 594168404Spjd *spare_guid = device_guid; 595168404Spjd 596168404Spjd switch (reason) { 597168404Spjd case VDEV_LABEL_CREATE: 598185029Spjd case VDEV_LABEL_L2CACHE: 599168404Spjd return (B_TRUE); 600168404Spjd 601168404Spjd case VDEV_LABEL_REPLACE: 602168404Spjd return (!spa_has_spare(spa, device_guid) || 603168404Spjd spare_pool != 0ULL); 604168404Spjd 605168404Spjd case VDEV_LABEL_SPARE: 606168404Spjd return (spa_has_spare(spa, device_guid)); 607168404Spjd } 608168404Spjd } 609168404Spjd 610168404Spjd /* 611185029Spjd * Check to see if this is an l2cache device. 612185029Spjd */ 613185029Spjd if (spa_l2cache_exists(device_guid, NULL)) 614185029Spjd return (B_TRUE); 615185029Spjd 616185029Spjd /* 617219089Spjd * We can't rely on a pool's state if it's been imported 618219089Spjd * read-only. Instead we look to see if the pools is marked 619219089Spjd * read-only in the namespace and set the state to active. 620219089Spjd */ 621219089Spjd if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL && 622219089Spjd spa_mode(spa) == FREAD) 623219089Spjd state = POOL_STATE_ACTIVE; 624219089Spjd 625219089Spjd /* 626168404Spjd * If the device is marked ACTIVE, then this device is in use by another 627168404Spjd * pool on the system. 628168404Spjd */ 629168404Spjd return (state == POOL_STATE_ACTIVE); 630168404Spjd} 631168404Spjd 632168404Spjd/* 633168404Spjd * Initialize a vdev label. We check to make sure each leaf device is not in 634168404Spjd * use, and writable. We put down an initial label which we will later 635168404Spjd * overwrite with a complete label. Note that it's important to do this 636168404Spjd * sequentially, not in parallel, so that we catch cases of multiple use of the 637168404Spjd * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 638168404Spjd * itself. 639168404Spjd */ 640168404Spjdint 641168404Spjdvdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 642168404Spjd{ 643168404Spjd spa_t *spa = vd->vdev_spa; 644168404Spjd nvlist_t *label; 645168404Spjd vdev_phys_t *vp; 646209962Smm char *pad2; 647168404Spjd uberblock_t *ub; 648168404Spjd zio_t *zio; 649168404Spjd char *buf; 650168404Spjd size_t buflen; 651168404Spjd int error; 652185029Spjd uint64_t spare_guid, l2cache_guid; 653185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 654168404Spjd 655185029Spjd ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 656168404Spjd 657185029Spjd for (int c = 0; c < vd->vdev_children; c++) 658168404Spjd if ((error = vdev_label_init(vd->vdev_child[c], 659168404Spjd crtxg, reason)) != 0) 660168404Spjd return (error); 661168404Spjd 662219089Spjd /* Track the creation time for this vdev */ 663219089Spjd vd->vdev_crtxg = crtxg; 664219089Spjd 665168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 666168404Spjd return (0); 667168404Spjd 668168404Spjd /* 669168404Spjd * Dead vdevs cannot be initialized. 670168404Spjd */ 671168404Spjd if (vdev_is_dead(vd)) 672249195Smm return (SET_ERROR(EIO)); 673168404Spjd 674168404Spjd /* 675168404Spjd * Determine if the vdev is in use. 676168404Spjd */ 677219089Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 678185029Spjd vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 679249195Smm return (SET_ERROR(EBUSY)); 680168404Spjd 681168404Spjd /* 682185029Spjd * If this is a request to add or replace a spare or l2cache device 683185029Spjd * that is in use elsewhere on the system, then we must update the 684185029Spjd * guid (which was initialized to a random value) to reflect the 685185029Spjd * actual GUID (which is shared between multiple pools). 686168404Spjd */ 687185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 688185029Spjd spare_guid != 0ULL) { 689185029Spjd uint64_t guid_delta = spare_guid - vd->vdev_guid; 690168404Spjd 691185029Spjd vd->vdev_guid += guid_delta; 692168404Spjd 693185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 694185029Spjd pvd->vdev_guid_sum += guid_delta; 695168404Spjd 696168404Spjd /* 697168404Spjd * If this is a replacement, then we want to fallthrough to the 698168404Spjd * rest of the code. If we're adding a spare, then it's already 699185029Spjd * labeled appropriately and we can just return. 700168404Spjd */ 701168404Spjd if (reason == VDEV_LABEL_SPARE) 702168404Spjd return (0); 703219089Spjd ASSERT(reason == VDEV_LABEL_REPLACE || 704219089Spjd reason == VDEV_LABEL_SPLIT); 705168404Spjd } 706168404Spjd 707185029Spjd if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 708185029Spjd l2cache_guid != 0ULL) { 709185029Spjd uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 710185029Spjd 711185029Spjd vd->vdev_guid += guid_delta; 712185029Spjd 713185029Spjd for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 714185029Spjd pvd->vdev_guid_sum += guid_delta; 715185029Spjd 716185029Spjd /* 717185029Spjd * If this is a replacement, then we want to fallthrough to the 718185029Spjd * rest of the code. If we're adding an l2cache, then it's 719185029Spjd * already labeled appropriately and we can just return. 720185029Spjd */ 721185029Spjd if (reason == VDEV_LABEL_L2CACHE) 722185029Spjd return (0); 723185029Spjd ASSERT(reason == VDEV_LABEL_REPLACE); 724185029Spjd } 725185029Spjd 726168404Spjd /* 727240868Spjd * TRIM the whole thing so that we start with a clean slate. 728240868Spjd * It's just an optimization, so we don't care if it fails. 729240868Spjd * Don't TRIM if removing so that we don't interfere with zpool 730240868Spjd * disaster recovery. 731240868Spjd */ 732244188Ssmh if (!zfs_notrim && vdev_trim_on_init && (reason == VDEV_LABEL_CREATE || 733240868Spjd reason == VDEV_LABEL_SPARE || 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) 1011168404Spjd atomic_add_64(good_writes, 1); 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 1047185029Spjdint 1048185029Spjdvdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 1049168404Spjd{ 1050185029Spjd spa_t *spa = svd[0]->vdev_spa; 1051168404Spjd zio_t *zio; 1052185029Spjd uint64_t good_writes = 0; 1053168404Spjd 1054185029Spjd zio = zio_root(spa, NULL, &good_writes, flags); 1055168404Spjd 1056185029Spjd for (int v = 0; v < svdcount; v++) 1057185029Spjd vdev_uberblock_sync(zio, ub, svd[v], flags); 1058168404Spjd 1059185029Spjd (void) zio_wait(zio); 1060168404Spjd 1061168404Spjd /* 1062185029Spjd * Flush the uberblocks to disk. This ensures that the odd labels 1063185029Spjd * are no longer needed (because the new uberblocks and the even 1064185029Spjd * labels are safely on disk), so it is safe to overwrite them. 1065168404Spjd */ 1066185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1067168404Spjd 1068185029Spjd for (int v = 0; v < svdcount; v++) 1069185029Spjd zio_flush(zio, svd[v]); 1070168404Spjd 1071185029Spjd (void) zio_wait(zio); 1072185029Spjd 1073185029Spjd return (good_writes >= 1 ? 0 : EIO); 1074168404Spjd} 1075168404Spjd 1076168404Spjd/* 1077185029Spjd * On success, increment the count of good writes for our top-level vdev. 1078168404Spjd */ 1079168404Spjdstatic void 1080185029Spjdvdev_label_sync_done(zio_t *zio) 1081168404Spjd{ 1082185029Spjd uint64_t *good_writes = zio->io_private; 1083168404Spjd 1084168404Spjd if (zio->io_error == 0) 1085168404Spjd atomic_add_64(good_writes, 1); 1086168404Spjd} 1087168404Spjd 1088185029Spjd/* 1089185029Spjd * If there weren't enough good writes, indicate failure to the parent. 1090185029Spjd */ 1091168404Spjdstatic void 1092185029Spjdvdev_label_sync_top_done(zio_t *zio) 1093168404Spjd{ 1094185029Spjd uint64_t *good_writes = zio->io_private; 1095185029Spjd 1096185029Spjd if (*good_writes == 0) 1097249195Smm zio->io_error = SET_ERROR(EIO); 1098185029Spjd 1099185029Spjd kmem_free(good_writes, sizeof (uint64_t)); 1100185029Spjd} 1101185029Spjd 1102185029Spjd/* 1103185029Spjd * We ignore errors for log and cache devices, simply free the private data. 1104185029Spjd */ 1105185029Spjdstatic void 1106185029Spjdvdev_label_sync_ignore_done(zio_t *zio) 1107185029Spjd{ 1108185029Spjd kmem_free(zio->io_private, sizeof (uint64_t)); 1109185029Spjd} 1110185029Spjd 1111185029Spjd/* 1112185029Spjd * Write all even or odd labels to all leaves of the specified vdev. 1113185029Spjd */ 1114185029Spjdstatic void 1115185029Spjdvdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1116185029Spjd{ 1117168404Spjd nvlist_t *label; 1118168404Spjd vdev_phys_t *vp; 1119168404Spjd char *buf; 1120168404Spjd size_t buflen; 1121168404Spjd 1122185029Spjd for (int c = 0; c < vd->vdev_children; c++) 1123185029Spjd vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1124168404Spjd 1125168404Spjd if (!vd->vdev_ops->vdev_op_leaf) 1126168404Spjd return; 1127168404Spjd 1128185029Spjd if (!vdev_writeable(vd)) 1129168404Spjd return; 1130168404Spjd 1131168404Spjd /* 1132168404Spjd * Generate a label describing the top-level config to which we belong. 1133168404Spjd */ 1134168404Spjd label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1135168404Spjd 1136168404Spjd vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1137168404Spjd bzero(vp, sizeof (vdev_phys_t)); 1138168404Spjd 1139168404Spjd buf = vp->vp_nvlist; 1140168404Spjd buflen = sizeof (vp->vp_nvlist); 1141168404Spjd 1142185029Spjd if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 1143185029Spjd for (; l < VDEV_LABELS; l += 2) { 1144185029Spjd vdev_label_write(zio, vd, l, vp, 1145185029Spjd offsetof(vdev_label_t, vl_vdev_phys), 1146185029Spjd sizeof (vdev_phys_t), 1147185029Spjd vdev_label_sync_done, zio->io_private, 1148185029Spjd flags | ZIO_FLAG_DONT_PROPAGATE); 1149185029Spjd } 1150185029Spjd } 1151168404Spjd 1152168404Spjd zio_buf_free(vp, sizeof (vdev_phys_t)); 1153168404Spjd nvlist_free(label); 1154168404Spjd} 1155168404Spjd 1156185029Spjdint 1157185029Spjdvdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1158168404Spjd{ 1159185029Spjd list_t *dl = &spa->spa_config_dirty_list; 1160185029Spjd vdev_t *vd; 1161168404Spjd zio_t *zio; 1162168404Spjd int error; 1163168404Spjd 1164185029Spjd /* 1165185029Spjd * Write the new labels to disk. 1166185029Spjd */ 1167185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1168168404Spjd 1169185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 1170185029Spjd uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 1171185029Spjd KM_SLEEP); 1172219089Spjd 1173219089Spjd ASSERT(!vd->vdev_ishole); 1174219089Spjd 1175209962Smm zio_t *vio = zio_null(zio, spa, NULL, 1176185029Spjd (vd->vdev_islog || vd->vdev_aux != NULL) ? 1177185029Spjd vdev_label_sync_ignore_done : vdev_label_sync_top_done, 1178185029Spjd good_writes, flags); 1179185029Spjd vdev_label_sync(vio, vd, l, txg, flags); 1180185029Spjd zio_nowait(vio); 1181185029Spjd } 1182168404Spjd 1183185029Spjd error = zio_wait(zio); 1184168404Spjd 1185168404Spjd /* 1186185029Spjd * Flush the new labels to disk. 1187168404Spjd */ 1188185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1189168404Spjd 1190185029Spjd for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 1191185029Spjd zio_flush(zio, vd); 1192168404Spjd 1193185029Spjd (void) zio_wait(zio); 1194168404Spjd 1195168404Spjd return (error); 1196168404Spjd} 1197168404Spjd 1198168404Spjd/* 1199185029Spjd * Sync the uberblock and any changes to the vdev configuration. 1200168404Spjd * 1201168404Spjd * The order of operations is carefully crafted to ensure that 1202168404Spjd * if the system panics or loses power at any time, the state on disk 1203168404Spjd * is still transactionally consistent. The in-line comments below 1204168404Spjd * describe the failure semantics at each stage. 1205168404Spjd * 1206185029Spjd * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1207168404Spjd * at any time, you can just call it again, and it will resume its work. 1208168404Spjd */ 1209168404Spjdint 1210213198Smmvdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1211168404Spjd{ 1212185029Spjd spa_t *spa = svd[0]->vdev_spa; 1213168404Spjd uberblock_t *ub = &spa->spa_uberblock; 1214168404Spjd vdev_t *vd; 1215168404Spjd zio_t *zio; 1216185029Spjd int error; 1217185029Spjd int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1218168404Spjd 1219213198Smm /* 1220213198Smm * Normally, we don't want to try too hard to write every label and 1221213198Smm * uberblock. If there is a flaky disk, we don't want the rest of the 1222213198Smm * sync process to block while we retry. But if we can't write a 1223213198Smm * single label out, we should retry with ZIO_FLAG_TRYHARD before 1224213198Smm * bailing out and declaring the pool faulted. 1225213198Smm */ 1226213198Smm if (tryhard) 1227213198Smm flags |= ZIO_FLAG_TRYHARD; 1228213198Smm 1229168404Spjd ASSERT(ub->ub_txg <= txg); 1230168404Spjd 1231168404Spjd /* 1232185029Spjd * If this isn't a resync due to I/O errors, 1233185029Spjd * and nothing changed in this transaction group, 1234185029Spjd * and the vdev configuration hasn't changed, 1235168404Spjd * then there's nothing to do. 1236168404Spjd */ 1237185029Spjd if (ub->ub_txg < txg && 1238185029Spjd uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1239185029Spjd list_is_empty(&spa->spa_config_dirty_list)) 1240168404Spjd return (0); 1241168404Spjd 1242168404Spjd if (txg > spa_freeze_txg(spa)) 1243168404Spjd return (0); 1244168404Spjd 1245168404Spjd ASSERT(txg <= spa->spa_final_txg); 1246168404Spjd 1247168404Spjd /* 1248168404Spjd * Flush the write cache of every disk that's been written to 1249168404Spjd * in this transaction group. This ensures that all blocks 1250168404Spjd * written in this txg will be committed to stable storage 1251168404Spjd * before any uberblock that references them. 1252168404Spjd */ 1253185029Spjd zio = zio_root(spa, NULL, NULL, flags); 1254185029Spjd 1255168404Spjd for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 1256185029Spjd vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 1257185029Spjd zio_flush(zio, vd); 1258185029Spjd 1259168404Spjd (void) zio_wait(zio); 1260168404Spjd 1261168404Spjd /* 1262168404Spjd * Sync out the even labels (L0, L2) for every dirty vdev. If the 1263168404Spjd * system dies in the middle of this process, that's OK: all of the 1264168404Spjd * even labels that made it to disk will be newer than any uberblock, 1265168404Spjd * and will therefore be considered invalid. The odd labels (L1, L3), 1266185029Spjd * which have not yet been touched, will still be valid. We flush 1267185029Spjd * the new labels to disk to ensure that all even-label updates 1268185029Spjd * are committed to stable storage before the uberblock update. 1269168404Spjd */ 1270185029Spjd if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1271185029Spjd return (error); 1272168404Spjd 1273168404Spjd /* 1274185029Spjd * Sync the uberblocks to all vdevs in svd[]. 1275168404Spjd * If the system dies in the middle of this step, there are two cases 1276168404Spjd * to consider, and the on-disk state is consistent either way: 1277168404Spjd * 1278168404Spjd * (1) If none of the new uberblocks made it to disk, then the 1279168404Spjd * previous uberblock will be the newest, and the odd labels 1280168404Spjd * (which had not yet been touched) will be valid with respect 1281168404Spjd * to that uberblock. 1282168404Spjd * 1283168404Spjd * (2) If one or more new uberblocks made it to disk, then they 1284168404Spjd * will be the newest, and the even labels (which had all 1285168404Spjd * been successfully committed) will be valid with respect 1286168404Spjd * to the new uberblocks. 1287168404Spjd */ 1288185029Spjd if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1289168404Spjd return (error); 1290168404Spjd 1291168404Spjd /* 1292168404Spjd * Sync out odd labels for every dirty vdev. If the system dies 1293168404Spjd * in the middle of this process, the even labels and the new 1294168404Spjd * uberblocks will suffice to open the pool. The next time 1295168404Spjd * the pool is opened, the first thing we'll do -- before any 1296168404Spjd * user data is modified -- is mark every vdev dirty so that 1297185029Spjd * all labels will be brought up to date. We flush the new labels 1298185029Spjd * to disk to ensure that all odd-label updates are committed to 1299185029Spjd * stable storage before the next transaction group begins. 1300168404Spjd */ 1301240868Spjd if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) 1302240868Spjd return (error); 1303240868Spjd 1304240868Spjd trim_thread_wakeup(spa); 1305240868Spjd 1306240868Spjd return (0); 1307168404Spjd} 1308