1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 22a3f829aeSBill Moore * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23fa9e4066Sahrens * Use is subject to license terms. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens /* 27fa9e4066Sahrens * Virtual Device Labels 28fa9e4066Sahrens * --------------------- 29fa9e4066Sahrens * 30fa9e4066Sahrens * The vdev label serves several distinct purposes: 31fa9e4066Sahrens * 32fa9e4066Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 33fa9e4066Sahrens * identity within the pool. 34fa9e4066Sahrens * 35fa9e4066Sahrens * 2. Verify that all the devices given in a configuration are present 36fa9e4066Sahrens * within the pool. 37fa9e4066Sahrens * 38fa9e4066Sahrens * 3. Determine the uberblock for the pool. 39fa9e4066Sahrens * 40fa9e4066Sahrens * 4. In case of an import operation, determine the configuration of the 41fa9e4066Sahrens * toplevel vdev of which it is a part. 42fa9e4066Sahrens * 43fa9e4066Sahrens * 5. If an import operation cannot find all the devices in the pool, 44fa9e4066Sahrens * provide enough information to the administrator to determine which 45fa9e4066Sahrens * devices are missing. 46fa9e4066Sahrens * 47fa9e4066Sahrens * It is important to note that while the kernel is responsible for writing the 48fa9e4066Sahrens * label, it only consumes the information in the first three cases. The 49fa9e4066Sahrens * latter information is only consumed in userland when determining the 50fa9e4066Sahrens * configuration to import a pool. 51fa9e4066Sahrens * 52fa9e4066Sahrens * 53fa9e4066Sahrens * Label Organization 54fa9e4066Sahrens * ------------------ 55fa9e4066Sahrens * 56fa9e4066Sahrens * Before describing the contents of the label, it's important to understand how 57fa9e4066Sahrens * the labels are written and updated with respect to the uberblock. 58fa9e4066Sahrens * 59fa9e4066Sahrens * When the pool configuration is altered, either because it was newly created 60fa9e4066Sahrens * or a device was added, we want to update all the labels such that we can deal 61fa9e4066Sahrens * with fatal failure at any point. To this end, each disk has two labels which 62fa9e4066Sahrens * are updated before and after the uberblock is synced. Assuming we have 633d7072f8Seschrock * labels and an uberblock with the following transaction groups: 64fa9e4066Sahrens * 65fa9e4066Sahrens * L1 UB L2 66fa9e4066Sahrens * +------+ +------+ +------+ 67fa9e4066Sahrens * | | | | | | 68fa9e4066Sahrens * | t10 | | t10 | | t10 | 69fa9e4066Sahrens * | | | | | | 70fa9e4066Sahrens * +------+ +------+ +------+ 71fa9e4066Sahrens * 72fa9e4066Sahrens * In this stable state, the labels and the uberblock were all updated within 73fa9e4066Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 74fa9e4066Sahrens * that we can detect when we fail partway through writing the label. 75fa9e4066Sahrens * 76fa9e4066Sahrens * In order to identify which labels are valid, the labels are written in the 77fa9e4066Sahrens * following manner: 78fa9e4066Sahrens * 79fa9e4066Sahrens * 1. For each vdev, update 'L1' to the new label 80fa9e4066Sahrens * 2. Update the uberblock 81fa9e4066Sahrens * 3. For each vdev, update 'L2' to the new label 82fa9e4066Sahrens * 83fa9e4066Sahrens * Given arbitrary failure, we can determine the correct label to use based on 84fa9e4066Sahrens * the transaction group. If we fail after updating L1 but before updating the 85fa9e4066Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 86fa9e4066Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 87fa9e4066Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 88fa9e4066Sahrens * therefore L1 is valid. 89fa9e4066Sahrens * 90fa9e4066Sahrens * Another added complexity is that not every label is updated when the config 91fa9e4066Sahrens * is synced. If we add a single device, we do not want to have to re-write 92fa9e4066Sahrens * every label for every device in the pool. This means that both L1 and L2 may 93fa9e4066Sahrens * be older than the pool uberblock, because the necessary information is stored 94fa9e4066Sahrens * on another vdev. 95fa9e4066Sahrens * 96fa9e4066Sahrens * 97fa9e4066Sahrens * On-disk Format 98fa9e4066Sahrens * -------------- 99fa9e4066Sahrens * 100fa9e4066Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 101fa9e4066Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 102fa9e4066Sahrens * VTOC disk labels, but is otherwise ignored. 103fa9e4066Sahrens * 104fa9e4066Sahrens * The first half of the label is a packed nvlist which contains pool wide 105fa9e4066Sahrens * properties, per-vdev properties, and configuration information. It is 106fa9e4066Sahrens * described in more detail below. 107fa9e4066Sahrens * 108fa9e4066Sahrens * The latter half of the label consists of a redundant array of uberblocks. 109fa9e4066Sahrens * These uberblocks are updated whenever a transaction group is committed, 110fa9e4066Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 111fa9e4066Sahrens * vdev for the 'best' uberblock. 112fa9e4066Sahrens * 113fa9e4066Sahrens * 114fa9e4066Sahrens * Configuration Information 115fa9e4066Sahrens * ------------------------- 116fa9e4066Sahrens * 117fa9e4066Sahrens * The nvlist describing the pool and vdev contains the following elements: 118fa9e4066Sahrens * 119fa9e4066Sahrens * version ZFS on-disk version 120fa9e4066Sahrens * name Pool name 121fa9e4066Sahrens * state Pool state 122fa9e4066Sahrens * txg Transaction group in which this label was written 123fa9e4066Sahrens * pool_guid Unique identifier for this pool 124fa9e4066Sahrens * vdev_tree An nvlist describing vdev tree. 125fa9e4066Sahrens * 126fa9e4066Sahrens * Each leaf device label also contains the following: 127fa9e4066Sahrens * 128fa9e4066Sahrens * top_guid Unique ID for top-level vdev in which this is contained 129fa9e4066Sahrens * guid Unique ID for the leaf vdev 130fa9e4066Sahrens * 131fa9e4066Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 132fa9e4066Sahrens */ 133fa9e4066Sahrens 134fa9e4066Sahrens #include <sys/zfs_context.h> 135fa9e4066Sahrens #include <sys/spa.h> 136fa9e4066Sahrens #include <sys/spa_impl.h> 137fa9e4066Sahrens #include <sys/dmu.h> 138fa9e4066Sahrens #include <sys/zap.h> 139fa9e4066Sahrens #include <sys/vdev.h> 140fa9e4066Sahrens #include <sys/vdev_impl.h> 141fa9e4066Sahrens #include <sys/uberblock_impl.h> 142fa9e4066Sahrens #include <sys/metaslab.h> 143fa9e4066Sahrens #include <sys/zio.h> 144fa9e4066Sahrens #include <sys/fs/zfs.h> 145fa9e4066Sahrens 146fa9e4066Sahrens /* 147fa9e4066Sahrens * Basic routines to read and write from a vdev label. 148fa9e4066Sahrens * Used throughout the rest of this file. 149fa9e4066Sahrens */ 150fa9e4066Sahrens uint64_t 151fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 152fa9e4066Sahrens { 153ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 154e7437265Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 155ecc2d604Sbonwick 156fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 157fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 158fa9e4066Sahrens } 159fa9e4066Sahrens 16021bf64a7Sgw /* 16121bf64a7Sgw * Returns back the vdev label associated with the passed in offset. 16221bf64a7Sgw */ 16321bf64a7Sgw int 16421bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset) 16521bf64a7Sgw { 16621bf64a7Sgw int l; 16721bf64a7Sgw 16821bf64a7Sgw if (offset >= psize - VDEV_LABEL_END_SIZE) { 16921bf64a7Sgw offset -= psize - VDEV_LABEL_END_SIZE; 17021bf64a7Sgw offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 17121bf64a7Sgw } 17221bf64a7Sgw l = offset / sizeof (vdev_label_t); 17321bf64a7Sgw return (l < VDEV_LABELS ? l : -1); 17421bf64a7Sgw } 17521bf64a7Sgw 176fa9e4066Sahrens static void 177fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 178e14bb325SJeff Bonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 179fa9e4066Sahrens { 180e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 181e14bb325SJeff Bonwick SCL_STATE_ALL); 182e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 183fa9e4066Sahrens 184fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 185fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 186fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 187e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 188fa9e4066Sahrens } 189fa9e4066Sahrens 190fa9e4066Sahrens static void 191fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 19217f17c2dSbonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 193fa9e4066Sahrens { 194e14bb325SJeff Bonwick ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 195e14bb325SJeff Bonwick (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 196e14bb325SJeff Bonwick (SCL_CONFIG | SCL_STATE) && 197e14bb325SJeff Bonwick dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 198e14bb325SJeff Bonwick ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 199fa9e4066Sahrens 200fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 201fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 202fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 20317f17c2dSbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 204fa9e4066Sahrens } 205fa9e4066Sahrens 206fa9e4066Sahrens /* 207fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 208fa9e4066Sahrens */ 209fa9e4066Sahrens nvlist_t * 21099653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 211fa94a07fSbrendan boolean_t isspare, boolean_t isl2cache) 212fa9e4066Sahrens { 213fa9e4066Sahrens nvlist_t *nv = NULL; 214fa9e4066Sahrens 215ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 216fa9e4066Sahrens 217fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 218fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 219fa94a07fSbrendan if (!isspare && !isl2cache) 22099653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 22199653d4eSeschrock == 0); 222fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 223fa9e4066Sahrens 224fa9e4066Sahrens if (vd->vdev_path != NULL) 225fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 226fa9e4066Sahrens vd->vdev_path) == 0); 227fa9e4066Sahrens 228fa9e4066Sahrens if (vd->vdev_devid != NULL) 229fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 230fa9e4066Sahrens vd->vdev_devid) == 0); 231fa9e4066Sahrens 2323d7072f8Seschrock if (vd->vdev_physpath != NULL) 2333d7072f8Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 2343d7072f8Seschrock vd->vdev_physpath) == 0); 2353d7072f8Seschrock 2366809eb4eSEric Schrock if (vd->vdev_fru != NULL) 2376809eb4eSEric Schrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, 2386809eb4eSEric Schrock vd->vdev_fru) == 0); 2396809eb4eSEric Schrock 24099653d4eSeschrock if (vd->vdev_nparity != 0) { 24199653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 24299653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 24399653d4eSeschrock 24499653d4eSeschrock /* 24599653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 24699653d4eSeschrock * into a crufty old storage pool. 24799653d4eSeschrock */ 24899653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 24999653d4eSeschrock (vd->vdev_nparity == 2 && 250e7437265Sahrens spa_version(spa) >= SPA_VERSION_RAID6)); 25199653d4eSeschrock 25299653d4eSeschrock /* 25399653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 25499653d4eSeschrock * that only support a single parity device -- older software 25599653d4eSeschrock * will just ignore it. 25699653d4eSeschrock */ 25799653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 25899653d4eSeschrock vd->vdev_nparity) == 0); 25999653d4eSeschrock } 26099653d4eSeschrock 261afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 262afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 263afefbcddSeschrock vd->vdev_wholedisk) == 0); 264afefbcddSeschrock 265ea8dc4b6Seschrock if (vd->vdev_not_present) 266ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 267ea8dc4b6Seschrock 26899653d4eSeschrock if (vd->vdev_isspare) 26999653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 27099653d4eSeschrock 271fa94a07fSbrendan if (!isspare && !isl2cache && vd == vd->vdev_top) { 272fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 273fa9e4066Sahrens vd->vdev_ms_array) == 0); 274fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 275fa9e4066Sahrens vd->vdev_ms_shift) == 0); 276fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 277fa9e4066Sahrens vd->vdev_ashift) == 0); 278fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 279fa9e4066Sahrens vd->vdev_asize) == 0); 2808654d025Sperrin VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 2818654d025Sperrin vd->vdev_islog) == 0); 282fa9e4066Sahrens } 283fa9e4066Sahrens 2848ad4d6ddSJeff Bonwick if (vd->vdev_dtl_smo.smo_object != 0) 285fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 2868ad4d6ddSJeff Bonwick vd->vdev_dtl_smo.smo_object) == 0); 287fa9e4066Sahrens 288fa9e4066Sahrens if (getstats) { 289fa9e4066Sahrens vdev_stat_t vs; 290fa9e4066Sahrens vdev_get_stats(vd, &vs); 291fa9e4066Sahrens VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 292fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 293fa9e4066Sahrens } 294fa9e4066Sahrens 295fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 296fa9e4066Sahrens nvlist_t **child; 297fa9e4066Sahrens int c; 298fa9e4066Sahrens 299fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 300fa9e4066Sahrens KM_SLEEP); 301fa9e4066Sahrens 302fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 30399653d4eSeschrock child[c] = vdev_config_generate(spa, vd->vdev_child[c], 304fa94a07fSbrendan getstats, isspare, isl2cache); 305fa9e4066Sahrens 306fa9e4066Sahrens VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 307fa9e4066Sahrens child, vd->vdev_children) == 0); 308fa9e4066Sahrens 309fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 310fa9e4066Sahrens nvlist_free(child[c]); 311fa9e4066Sahrens 312fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 313441d80aaSlling 314441d80aaSlling } else { 315ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 316441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 317ecc2d604Sbonwick B_TRUE) == 0); 3183d7072f8Seschrock if (vd->vdev_faulted) 3193d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 3203d7072f8Seschrock B_TRUE) == 0); 3213d7072f8Seschrock if (vd->vdev_degraded) 3223d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 3233d7072f8Seschrock B_TRUE) == 0); 3243d7072f8Seschrock if (vd->vdev_removed) 3253d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 3263d7072f8Seschrock B_TRUE) == 0); 3273d7072f8Seschrock if (vd->vdev_unspare) 3283d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 3293d7072f8Seschrock B_TRUE) == 0); 330fa9e4066Sahrens } 331fa9e4066Sahrens 332fa9e4066Sahrens return (nv); 333fa9e4066Sahrens } 334fa9e4066Sahrens 335fa9e4066Sahrens nvlist_t * 336fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 337fa9e4066Sahrens { 3380373e76bSbonwick spa_t *spa = vd->vdev_spa; 339fa9e4066Sahrens nvlist_t *config = NULL; 340fa9e4066Sahrens vdev_phys_t *vp; 341fa9e4066Sahrens zio_t *zio; 342*8956713aSEric Schrock int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 343*8956713aSEric Schrock ZIO_FLAG_SPECULATIVE; 344fa9e4066Sahrens 345e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3460373e76bSbonwick 3470a4e9518Sgw if (!vdev_readable(vd)) 348fa9e4066Sahrens return (NULL); 349fa9e4066Sahrens 350fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 351fa9e4066Sahrens 352*8956713aSEric Schrock retry: 353e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 354fa9e4066Sahrens 355e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 356fa9e4066Sahrens 357fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 358fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 359e14bb325SJeff Bonwick sizeof (vdev_phys_t), NULL, NULL, flags); 360fa9e4066Sahrens 361fa9e4066Sahrens if (zio_wait(zio) == 0 && 362fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 363ea8dc4b6Seschrock &config, 0) == 0) 364fa9e4066Sahrens break; 365fa9e4066Sahrens 366fa9e4066Sahrens if (config != NULL) { 367fa9e4066Sahrens nvlist_free(config); 368fa9e4066Sahrens config = NULL; 369fa9e4066Sahrens } 370fa9e4066Sahrens } 371fa9e4066Sahrens 372*8956713aSEric Schrock if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 373*8956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 374*8956713aSEric Schrock goto retry; 375*8956713aSEric Schrock } 376*8956713aSEric Schrock 377fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 378fa9e4066Sahrens 379fa9e4066Sahrens return (config); 380fa9e4066Sahrens } 381fa9e4066Sahrens 38239c23413Seschrock /* 38339c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 38439c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 38539c23413Seschrock */ 38639c23413Seschrock static boolean_t 38739c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 388fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 38939c23413Seschrock { 39039c23413Seschrock spa_t *spa = vd->vdev_spa; 39139c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 39239c23413Seschrock uint64_t vdtxg = 0; 39339c23413Seschrock nvlist_t *label; 39439c23413Seschrock 39539c23413Seschrock if (spare_guid) 39639c23413Seschrock *spare_guid = 0ULL; 397fa94a07fSbrendan if (l2cache_guid) 398fa94a07fSbrendan *l2cache_guid = 0ULL; 39939c23413Seschrock 40039c23413Seschrock /* 40139c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 40239c23413Seschrock */ 40339c23413Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 40439c23413Seschrock return (B_FALSE); 40539c23413Seschrock 40639c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 40739c23413Seschrock &vdtxg); 40839c23413Seschrock 40939c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 41039c23413Seschrock &state) != 0 || 41139c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 41239c23413Seschrock &device_guid) != 0) { 41339c23413Seschrock nvlist_free(label); 41439c23413Seschrock return (B_FALSE); 41539c23413Seschrock } 41639c23413Seschrock 417fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 41839c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 41939c23413Seschrock &pool_guid) != 0 || 42039c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 42139c23413Seschrock &txg) != 0)) { 42239c23413Seschrock nvlist_free(label); 42339c23413Seschrock return (B_FALSE); 42439c23413Seschrock } 42539c23413Seschrock 42639c23413Seschrock nvlist_free(label); 42739c23413Seschrock 42839c23413Seschrock /* 42939c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 43039c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 43139c23413Seschrock * spare (which we check for later on). 43239c23413Seschrock */ 433fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 43439c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 43589a89ebfSlling !spa_spare_exists(device_guid, NULL, NULL) && 436fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 43739c23413Seschrock return (B_FALSE); 43839c23413Seschrock 43939c23413Seschrock /* 44039c23413Seschrock * If the transaction group is zero, then this an initialized (but 44139c23413Seschrock * unused) label. This is only an error if the create transaction 44239c23413Seschrock * on-disk is the same as the one we're using now, in which case the 44339c23413Seschrock * user has attempted to add the same vdev multiple times in the same 44439c23413Seschrock * transaction. 44539c23413Seschrock */ 446fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 447fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 44839c23413Seschrock return (B_TRUE); 44939c23413Seschrock 45039c23413Seschrock /* 45139c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 45239c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 453fa94a07fSbrendan * to add. We also check if it is an l2cache device. 45439c23413Seschrock */ 45589a89ebfSlling if (spa_spare_exists(device_guid, &spare_pool, NULL) || 45639c23413Seschrock spa_has_spare(spa, device_guid)) { 45739c23413Seschrock if (spare_guid) 45839c23413Seschrock *spare_guid = device_guid; 45939c23413Seschrock 46039c23413Seschrock switch (reason) { 46139c23413Seschrock case VDEV_LABEL_CREATE: 462fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 46339c23413Seschrock return (B_TRUE); 46439c23413Seschrock 46539c23413Seschrock case VDEV_LABEL_REPLACE: 46639c23413Seschrock return (!spa_has_spare(spa, device_guid) || 46739c23413Seschrock spare_pool != 0ULL); 46839c23413Seschrock 46939c23413Seschrock case VDEV_LABEL_SPARE: 47039c23413Seschrock return (spa_has_spare(spa, device_guid)); 47139c23413Seschrock } 47239c23413Seschrock } 47339c23413Seschrock 474fa94a07fSbrendan /* 475fa94a07fSbrendan * Check to see if this is an l2cache device. 476fa94a07fSbrendan */ 477fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 478fa94a07fSbrendan return (B_TRUE); 479fa94a07fSbrendan 48039c23413Seschrock /* 48139c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 48239c23413Seschrock * pool on the system. 48339c23413Seschrock */ 48439c23413Seschrock return (state == POOL_STATE_ACTIVE); 48539c23413Seschrock } 48639c23413Seschrock 48739c23413Seschrock /* 48839c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 48939c23413Seschrock * use, and writable. We put down an initial label which we will later 49039c23413Seschrock * overwrite with a complete label. Note that it's important to do this 49139c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 49239c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 49339c23413Seschrock * itself. 49439c23413Seschrock */ 49539c23413Seschrock int 49639c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 497fa9e4066Sahrens { 498fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 499fa9e4066Sahrens nvlist_t *label; 500fa9e4066Sahrens vdev_phys_t *vp; 501f83ffe1aSLin Ling char *pad2; 502ecc2d604Sbonwick uberblock_t *ub; 503fa9e4066Sahrens zio_t *zio; 504fa9e4066Sahrens char *buf; 505fa9e4066Sahrens size_t buflen; 506fa9e4066Sahrens int error; 507fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 508e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 509fa9e4066Sahrens 510e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5110373e76bSbonwick 512e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 51339c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 51439c23413Seschrock crtxg, reason)) != 0) 515fa9e4066Sahrens return (error); 516fa9e4066Sahrens 517fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 518fa9e4066Sahrens return (0); 519fa9e4066Sahrens 520fa9e4066Sahrens /* 52139c23413Seschrock * Dead vdevs cannot be initialized. 522fa9e4066Sahrens */ 523fa9e4066Sahrens if (vdev_is_dead(vd)) 524fa9e4066Sahrens return (EIO); 525fa9e4066Sahrens 526fa9e4066Sahrens /* 52739c23413Seschrock * Determine if the vdev is in use. 528fa9e4066Sahrens */ 52939c23413Seschrock if (reason != VDEV_LABEL_REMOVE && 530fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 53139c23413Seschrock return (EBUSY); 53239c23413Seschrock 53339c23413Seschrock /* 534fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 535fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 536fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 537fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 53839c23413Seschrock */ 539fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 540fa94a07fSbrendan spare_guid != 0ULL) { 54131157203SJeff Bonwick uint64_t guid_delta = spare_guid - vd->vdev_guid; 54299653d4eSeschrock 54331157203SJeff Bonwick vd->vdev_guid += guid_delta; 54431157203SJeff Bonwick 54531157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 54631157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 54739c23413Seschrock 54899653d4eSeschrock /* 54939c23413Seschrock * If this is a replacement, then we want to fallthrough to the 55039c23413Seschrock * rest of the code. If we're adding a spare, then it's already 5513d7072f8Seschrock * labeled appropriately and we can just return. 55299653d4eSeschrock */ 55339c23413Seschrock if (reason == VDEV_LABEL_SPARE) 55439c23413Seschrock return (0); 55539c23413Seschrock ASSERT(reason == VDEV_LABEL_REPLACE); 556fa9e4066Sahrens } 557fa9e4066Sahrens 558fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 559fa94a07fSbrendan l2cache_guid != 0ULL) { 56031157203SJeff Bonwick uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 56131157203SJeff Bonwick 56231157203SJeff Bonwick vd->vdev_guid += guid_delta; 563fa94a07fSbrendan 56431157203SJeff Bonwick for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 56531157203SJeff Bonwick pvd->vdev_guid_sum += guid_delta; 566fa94a07fSbrendan 567fa94a07fSbrendan /* 568fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 569fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 570fa94a07fSbrendan * already labeled appropriately and we can just return. 571fa94a07fSbrendan */ 572fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 573fa94a07fSbrendan return (0); 574fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 575fa94a07fSbrendan } 576fa94a07fSbrendan 577fa9e4066Sahrens /* 57839c23413Seschrock * Initialize its label. 579fa9e4066Sahrens */ 580fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 581fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 582fa9e4066Sahrens 583fa9e4066Sahrens /* 584fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 585fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 586fa9e4066Sahrens * really part of an active pool just yet. The labels will 587fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 588fa9e4066Sahrens */ 58939c23413Seschrock if (reason == VDEV_LABEL_SPARE || 59039c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 59139c23413Seschrock /* 59239c23413Seschrock * For inactive hot spares, we generate a special label that 59339c23413Seschrock * identifies as a mutually shared hot spare. We write the 59439c23413Seschrock * label if we are adding a hot spare, or if we are removing an 59539c23413Seschrock * active hot spare (in which case we want to revert the 59639c23413Seschrock * labels). 59739c23413Seschrock */ 59899653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 59999653d4eSeschrock 60099653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 60199653d4eSeschrock spa_version(spa)) == 0); 60299653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 60399653d4eSeschrock POOL_STATE_SPARE) == 0); 60499653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 60599653d4eSeschrock vd->vdev_guid) == 0); 606fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 607fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 608fa94a07fSbrendan /* 609fa94a07fSbrendan * For level 2 ARC devices, add a special label. 610fa94a07fSbrendan */ 611fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 612fa94a07fSbrendan 613fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 614fa94a07fSbrendan spa_version(spa)) == 0); 615fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 616fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 617fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 618fa94a07fSbrendan vd->vdev_guid) == 0); 61999653d4eSeschrock } else { 62099653d4eSeschrock label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 62199653d4eSeschrock 62299653d4eSeschrock /* 62399653d4eSeschrock * Add our creation time. This allows us to detect multiple 62499653d4eSeschrock * vdev uses as described above, and automatically expires if we 62599653d4eSeschrock * fail. 62699653d4eSeschrock */ 62799653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 62899653d4eSeschrock crtxg) == 0); 62999653d4eSeschrock } 630fa9e4066Sahrens 631fa9e4066Sahrens buf = vp->vp_nvlist; 632fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 633fa9e4066Sahrens 634a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 635a75573b6Smmusante if (error != 0) { 636fa9e4066Sahrens nvlist_free(label); 637fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 638a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 639a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 640fa9e4066Sahrens } 641fa9e4066Sahrens 642fa9e4066Sahrens /* 643fa9e4066Sahrens * Initialize uberblock template. 644fa9e4066Sahrens */ 645ecc2d604Sbonwick ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 646ecc2d604Sbonwick bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 647ecc2d604Sbonwick *ub = spa->spa_uberblock; 648ecc2d604Sbonwick ub->ub_txg = 0; 649fa9e4066Sahrens 650f83ffe1aSLin Ling /* Initialize the 2nd padding area. */ 651f83ffe1aSLin Ling pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 652f83ffe1aSLin Ling bzero(pad2, VDEV_PAD_SIZE); 653f83ffe1aSLin Ling 654fa9e4066Sahrens /* 655fa9e4066Sahrens * Write everything in parallel. 656fa9e4066Sahrens */ 657*8956713aSEric Schrock retry: 65817f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 659fa9e4066Sahrens 660e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 661fa9e4066Sahrens 662fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 663fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 66417f17c2dSbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 665fa9e4066Sahrens 666f83ffe1aSLin Ling /* 667f83ffe1aSLin Ling * Skip the 1st padding area. 668f83ffe1aSLin Ling * Zero out the 2nd padding area where it might have 669f83ffe1aSLin Ling * left over data from previous filesystem format. 670f83ffe1aSLin Ling */ 671f83ffe1aSLin Ling vdev_label_write(zio, vd, l, pad2, 672f83ffe1aSLin Ling offsetof(vdev_label_t, vl_pad2), 673f83ffe1aSLin Ling VDEV_PAD_SIZE, NULL, NULL, flags); 674f83ffe1aSLin Ling 675e14bb325SJeff Bonwick for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 676ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 677ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 67817f17c2dSbonwick VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags); 679fa9e4066Sahrens } 680fa9e4066Sahrens } 681fa9e4066Sahrens 682fa9e4066Sahrens error = zio_wait(zio); 683fa9e4066Sahrens 684*8956713aSEric Schrock if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 685*8956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 686*8956713aSEric Schrock goto retry; 687*8956713aSEric Schrock } 688*8956713aSEric Schrock 689fa9e4066Sahrens nvlist_free(label); 690f83ffe1aSLin Ling zio_buf_free(pad2, VDEV_PAD_SIZE); 691ecc2d604Sbonwick zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 692fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 693fa9e4066Sahrens 69439c23413Seschrock /* 69539c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 6963d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 697fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 698fa94a07fSbrendan * level 2 ARC devices. 69939c23413Seschrock */ 70039c23413Seschrock if (error == 0 && !vd->vdev_isspare && 70139c23413Seschrock (reason == VDEV_LABEL_SPARE || 70289a89ebfSlling spa_spare_exists(vd->vdev_guid, NULL, NULL))) 70339c23413Seschrock spa_spare_add(vd); 70499653d4eSeschrock 705fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 706fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 707fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 708fa94a07fSbrendan spa_l2cache_add(vd); 709fa94a07fSbrendan 71039c23413Seschrock return (error); 71199653d4eSeschrock } 71299653d4eSeschrock 713fa9e4066Sahrens /* 714fa9e4066Sahrens * ========================================================================== 715fa9e4066Sahrens * uberblock load/sync 716fa9e4066Sahrens * ========================================================================== 717fa9e4066Sahrens */ 718fa9e4066Sahrens 7192e551927SVictor Latushkin /* 7202e551927SVictor Latushkin * For use by zdb and debugging purposes only 7212e551927SVictor Latushkin */ 7222e551927SVictor Latushkin uint64_t ub_max_txg = UINT64_MAX; 7232e551927SVictor Latushkin 724fa9e4066Sahrens /* 725fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 726fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 727fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 728fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 729fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 730fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 731fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 732fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 733fa9e4066Sahrens */ 734fa9e4066Sahrens static int 735fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 736fa9e4066Sahrens { 737fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 738fa9e4066Sahrens return (-1); 739fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 740fa9e4066Sahrens return (1); 741fa9e4066Sahrens 742fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 743fa9e4066Sahrens return (-1); 744fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 745fa9e4066Sahrens return (1); 746fa9e4066Sahrens 747fa9e4066Sahrens return (0); 748fa9e4066Sahrens } 749fa9e4066Sahrens 750fa9e4066Sahrens static void 751fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 752fa9e4066Sahrens { 753e14bb325SJeff Bonwick zio_t *rio = zio->io_private; 754ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 755e14bb325SJeff Bonwick uberblock_t *ubbest = rio->io_private; 756fa9e4066Sahrens 757ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 758fa9e4066Sahrens 759ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 760e14bb325SJeff Bonwick mutex_enter(&rio->io_lock); 7612e551927SVictor Latushkin if (ub->ub_txg <= ub_max_txg && 7622e551927SVictor Latushkin vdev_uberblock_compare(ub, ubbest) > 0) 763fa9e4066Sahrens *ubbest = *ub; 764e14bb325SJeff Bonwick mutex_exit(&rio->io_lock); 765fa9e4066Sahrens } 766fa9e4066Sahrens 767fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 768fa9e4066Sahrens } 769fa9e4066Sahrens 770fa9e4066Sahrens void 771fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 772fa9e4066Sahrens { 773e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 774e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 775*8956713aSEric Schrock int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 776*8956713aSEric Schrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 777e14bb325SJeff Bonwick 778e14bb325SJeff Bonwick if (vd == rvd) { 779e14bb325SJeff Bonwick ASSERT(zio == NULL); 780e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 781e14bb325SJeff Bonwick zio = zio_root(spa, NULL, ubbest, flags); 782e14bb325SJeff Bonwick bzero(ubbest, sizeof (uberblock_t)); 783e14bb325SJeff Bonwick } 784fa9e4066Sahrens 785e14bb325SJeff Bonwick ASSERT(zio != NULL); 786fa9e4066Sahrens 787e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 788e14bb325SJeff Bonwick vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 789fa9e4066Sahrens 790e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 791e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) { 792e14bb325SJeff Bonwick for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 793e14bb325SJeff Bonwick vdev_label_read(zio, vd, l, 794e14bb325SJeff Bonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 795e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), 796e14bb325SJeff Bonwick VDEV_UBERBLOCK_SIZE(vd), 797e14bb325SJeff Bonwick vdev_uberblock_load_done, zio, flags); 798e14bb325SJeff Bonwick } 799fa9e4066Sahrens } 800fa9e4066Sahrens } 801e14bb325SJeff Bonwick 802e14bb325SJeff Bonwick if (vd == rvd) { 803e14bb325SJeff Bonwick (void) zio_wait(zio); 804e14bb325SJeff Bonwick spa_config_exit(spa, SCL_ALL, FTAG); 805e14bb325SJeff Bonwick } 806fa9e4066Sahrens } 807fa9e4066Sahrens 808fa9e4066Sahrens /* 80917f17c2dSbonwick * On success, increment root zio's count of good writes. 8100373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 811fa9e4066Sahrens */ 812fa9e4066Sahrens static void 813fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 814fa9e4066Sahrens { 81517f17c2dSbonwick uint64_t *good_writes = zio->io_private; 816fa9e4066Sahrens 8170373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 818fa9e4066Sahrens atomic_add_64(good_writes, 1); 819fa9e4066Sahrens } 820fa9e4066Sahrens 82117f17c2dSbonwick /* 82217f17c2dSbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 82317f17c2dSbonwick */ 824fa9e4066Sahrens static void 825e14bb325SJeff Bonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 826fa9e4066Sahrens { 82717f17c2dSbonwick uberblock_t *ubbuf; 828e14bb325SJeff Bonwick int n; 829fa9e4066Sahrens 830e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 831e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 832fa9e4066Sahrens 833fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 834fa9e4066Sahrens return; 835fa9e4066Sahrens 836e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 837fa9e4066Sahrens return; 838fa9e4066Sahrens 83917f17c2dSbonwick n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 840fa9e4066Sahrens 84117f17c2dSbonwick ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 84217f17c2dSbonwick bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 84317f17c2dSbonwick *ubbuf = *ub; 844fa9e4066Sahrens 845e14bb325SJeff Bonwick for (int l = 0; l < VDEV_LABELS; l++) 84617f17c2dSbonwick vdev_label_write(zio, vd, l, ubbuf, 847e14bb325SJeff Bonwick VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 84817f17c2dSbonwick vdev_uberblock_sync_done, zio->io_private, 849e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 850fa9e4066Sahrens 85117f17c2dSbonwick zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 852fa9e4066Sahrens } 853fa9e4066Sahrens 85417f17c2dSbonwick int 85517f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 856fa9e4066Sahrens { 85717f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 858e14bb325SJeff Bonwick zio_t *zio; 85917f17c2dSbonwick uint64_t good_writes = 0; 860fa9e4066Sahrens 861e14bb325SJeff Bonwick zio = zio_root(spa, NULL, &good_writes, flags); 862e14bb325SJeff Bonwick 863e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 864e14bb325SJeff Bonwick vdev_uberblock_sync(zio, ub, svd[v], flags); 865fa9e4066Sahrens 86617f17c2dSbonwick (void) zio_wait(zio); 867fa9e4066Sahrens 868fa9e4066Sahrens /* 86917f17c2dSbonwick * Flush the uberblocks to disk. This ensures that the odd labels 87017f17c2dSbonwick * are no longer needed (because the new uberblocks and the even 87117f17c2dSbonwick * labels are safely on disk), so it is safe to overwrite them. 872fa9e4066Sahrens */ 87317f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 874fa9e4066Sahrens 875e14bb325SJeff Bonwick for (int v = 0; v < svdcount; v++) 87617f17c2dSbonwick zio_flush(zio, svd[v]); 877fa9e4066Sahrens 87817f17c2dSbonwick (void) zio_wait(zio); 87917f17c2dSbonwick 88017f17c2dSbonwick return (good_writes >= 1 ? 0 : EIO); 881fa9e4066Sahrens } 882fa9e4066Sahrens 883fa9e4066Sahrens /* 88417f17c2dSbonwick * On success, increment the count of good writes for our top-level vdev. 885fa9e4066Sahrens */ 886fa9e4066Sahrens static void 88717f17c2dSbonwick vdev_label_sync_done(zio_t *zio) 888fa9e4066Sahrens { 88917f17c2dSbonwick uint64_t *good_writes = zio->io_private; 890fa9e4066Sahrens 891fa9e4066Sahrens if (zio->io_error == 0) 892fa9e4066Sahrens atomic_add_64(good_writes, 1); 893fa9e4066Sahrens } 894fa9e4066Sahrens 89517f17c2dSbonwick /* 89617f17c2dSbonwick * If there weren't enough good writes, indicate failure to the parent. 89717f17c2dSbonwick */ 898fa9e4066Sahrens static void 89917f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio) 90017f17c2dSbonwick { 90117f17c2dSbonwick uint64_t *good_writes = zio->io_private; 90217f17c2dSbonwick 90317f17c2dSbonwick if (*good_writes == 0) 90417f17c2dSbonwick zio->io_error = EIO; 90517f17c2dSbonwick 90617f17c2dSbonwick kmem_free(good_writes, sizeof (uint64_t)); 90717f17c2dSbonwick } 90817f17c2dSbonwick 90951ece835Seschrock /* 9100430f8daSeschrock * We ignore errors for log and cache devices, simply free the private data. 91151ece835Seschrock */ 91251ece835Seschrock static void 9130430f8daSeschrock vdev_label_sync_ignore_done(zio_t *zio) 91451ece835Seschrock { 91551ece835Seschrock kmem_free(zio->io_private, sizeof (uint64_t)); 91651ece835Seschrock } 91751ece835Seschrock 91817f17c2dSbonwick /* 91917f17c2dSbonwick * Write all even or odd labels to all leaves of the specified vdev. 92017f17c2dSbonwick */ 92117f17c2dSbonwick static void 922e14bb325SJeff Bonwick vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 923fa9e4066Sahrens { 924fa9e4066Sahrens nvlist_t *label; 925fa9e4066Sahrens vdev_phys_t *vp; 926fa9e4066Sahrens char *buf; 927fa9e4066Sahrens size_t buflen; 928fa9e4066Sahrens 929e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 930e14bb325SJeff Bonwick vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 931fa9e4066Sahrens 932fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 933fa9e4066Sahrens return; 934fa9e4066Sahrens 935e14bb325SJeff Bonwick if (!vdev_writeable(vd)) 936fa9e4066Sahrens return; 937fa9e4066Sahrens 938fa9e4066Sahrens /* 939fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 940fa9e4066Sahrens */ 9410373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 942fa9e4066Sahrens 943fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 944fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 945fa9e4066Sahrens 946fa9e4066Sahrens buf = vp->vp_nvlist; 947fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 948fa9e4066Sahrens 94917f17c2dSbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 95017f17c2dSbonwick for (; l < VDEV_LABELS; l += 2) { 95117f17c2dSbonwick vdev_label_write(zio, vd, l, vp, 95217f17c2dSbonwick offsetof(vdev_label_t, vl_vdev_phys), 95317f17c2dSbonwick sizeof (vdev_phys_t), 95417f17c2dSbonwick vdev_label_sync_done, zio->io_private, 955e14bb325SJeff Bonwick flags | ZIO_FLAG_DONT_PROPAGATE); 95617f17c2dSbonwick } 95717f17c2dSbonwick } 958fa9e4066Sahrens 959fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 960fa9e4066Sahrens nvlist_free(label); 961fa9e4066Sahrens } 962fa9e4066Sahrens 96317f17c2dSbonwick int 964e14bb325SJeff Bonwick vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 965fa9e4066Sahrens { 966e14bb325SJeff Bonwick list_t *dl = &spa->spa_config_dirty_list; 96717f17c2dSbonwick vdev_t *vd; 968e14bb325SJeff Bonwick zio_t *zio; 969fa9e4066Sahrens int error; 970fa9e4066Sahrens 971fa9e4066Sahrens /* 972e14bb325SJeff Bonwick * Write the new labels to disk. 973fa9e4066Sahrens */ 974e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, flags); 975fa9e4066Sahrens 97617f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 97717f17c2dSbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 97817f17c2dSbonwick KM_SLEEP); 979a3f829aeSBill Moore zio_t *vio = zio_null(zio, spa, NULL, 9800430f8daSeschrock (vd->vdev_islog || vd->vdev_aux != NULL) ? 9810430f8daSeschrock vdev_label_sync_ignore_done : vdev_label_sync_top_done, 98217f17c2dSbonwick good_writes, flags); 983e14bb325SJeff Bonwick vdev_label_sync(vio, vd, l, txg, flags); 98417f17c2dSbonwick zio_nowait(vio); 985fa9e4066Sahrens } 986e14bb325SJeff Bonwick 987e14bb325SJeff Bonwick error = zio_wait(zio); 988fa9e4066Sahrens 9898654d025Sperrin /* 99017f17c2dSbonwick * Flush the new labels to disk. 9918654d025Sperrin */ 99217f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 9938654d025Sperrin 99417f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 99517f17c2dSbonwick zio_flush(zio, vd); 99617f17c2dSbonwick 99717f17c2dSbonwick (void) zio_wait(zio); 998fa9e4066Sahrens 999fa9e4066Sahrens return (error); 1000fa9e4066Sahrens } 1001fa9e4066Sahrens 1002fa9e4066Sahrens /* 100317f17c2dSbonwick * Sync the uberblock and any changes to the vdev configuration. 1004fa9e4066Sahrens * 1005fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 1006fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 1007fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 1008fa9e4066Sahrens * describe the failure semantics at each stage. 1009fa9e4066Sahrens * 101017f17c2dSbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1011fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 1012fa9e4066Sahrens */ 1013e14bb325SJeff Bonwick int 1014*8956713aSEric Schrock vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1015fa9e4066Sahrens { 101617f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1017fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 10180373e76bSbonwick vdev_t *vd; 1019fa9e4066Sahrens zio_t *zio; 1020e14bb325SJeff Bonwick int error; 1021e14bb325SJeff Bonwick int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1022fa9e4066Sahrens 1023*8956713aSEric Schrock /* 1024*8956713aSEric Schrock * Normally, we don't want to try too hard to write every label and 1025*8956713aSEric Schrock * uberblock. If there is a flaky disk, we don't want the rest of the 1026*8956713aSEric Schrock * sync process to block while we retry. But if we can't write a 1027*8956713aSEric Schrock * single label out, we should retry with ZIO_FLAG_TRYHARD before 1028*8956713aSEric Schrock * bailing out and declaring the pool faulted. 1029*8956713aSEric Schrock */ 1030*8956713aSEric Schrock if (tryhard) 1031*8956713aSEric Schrock flags |= ZIO_FLAG_TRYHARD; 1032*8956713aSEric Schrock 1033fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 1034fa9e4066Sahrens 1035fa9e4066Sahrens /* 103617f17c2dSbonwick * If this isn't a resync due to I/O errors, 103717f17c2dSbonwick * and nothing changed in this transaction group, 103817f17c2dSbonwick * and the vdev configuration hasn't changed, 10390373e76bSbonwick * then there's nothing to do. 1040fa9e4066Sahrens */ 104117f17c2dSbonwick if (ub->ub_txg < txg && 104217f17c2dSbonwick uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 1043e14bb325SJeff Bonwick list_is_empty(&spa->spa_config_dirty_list)) 1044e14bb325SJeff Bonwick return (0); 1045fa9e4066Sahrens 1046fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 1047e14bb325SJeff Bonwick return (0); 1048fa9e4066Sahrens 10490373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 10500373e76bSbonwick 1051fa9e4066Sahrens /* 1052fa9e4066Sahrens * Flush the write cache of every disk that's been written to 1053fa9e4066Sahrens * in this transaction group. This ensures that all blocks 1054fa9e4066Sahrens * written in this txg will be committed to stable storage 1055fa9e4066Sahrens * before any uberblock that references them. 1056fa9e4066Sahrens */ 105717f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 105817f17c2dSbonwick 1059fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 106017f17c2dSbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 106117f17c2dSbonwick zio_flush(zio, vd); 106217f17c2dSbonwick 1063fa9e4066Sahrens (void) zio_wait(zio); 1064fa9e4066Sahrens 1065fa9e4066Sahrens /* 1066fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1067fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1068fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1069fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 107017f17c2dSbonwick * which have not yet been touched, will still be valid. We flush 107117f17c2dSbonwick * the new labels to disk to ensure that all even-label updates 107217f17c2dSbonwick * are committed to stable storage before the uberblock update. 1073fa9e4066Sahrens */ 1074e14bb325SJeff Bonwick if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 1075e14bb325SJeff Bonwick return (error); 1076fa9e4066Sahrens 1077fa9e4066Sahrens /* 1078e14bb325SJeff Bonwick * Sync the uberblocks to all vdevs in svd[]. 10790373e76bSbonwick * If the system dies in the middle of this step, there are two cases 10800373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1081fa9e4066Sahrens * 1082fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1083fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1084fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1085fa9e4066Sahrens * to that uberblock. 1086fa9e4066Sahrens * 1087fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1088fa9e4066Sahrens * will be the newest, and the even labels (which had all 1089fa9e4066Sahrens * been successfully committed) will be valid with respect 1090fa9e4066Sahrens * to the new uberblocks. 1091fa9e4066Sahrens */ 1092e14bb325SJeff Bonwick if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 1093e14bb325SJeff Bonwick return (error); 1094fa9e4066Sahrens 1095fa9e4066Sahrens /* 1096fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1097fa9e4066Sahrens * in the middle of this process, the even labels and the new 1098fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1099fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1100fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 110117f17c2dSbonwick * all labels will be brought up to date. We flush the new labels 110217f17c2dSbonwick * to disk to ensure that all odd-label updates are committed to 110317f17c2dSbonwick * stable storage before the next transaction group begins. 1104fa9e4066Sahrens */ 1105e14bb325SJeff Bonwick return (vdev_label_sync_list(spa, 1, txg, flags)); 1106fa9e4066Sahrens } 1107