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 /* 22441d80aaSlling * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23fa9e4066Sahrens * Use is subject to license terms. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #pragma ident "%Z%%M% %I% %E% SMI" 27fa9e4066Sahrens 28fa9e4066Sahrens /* 29fa9e4066Sahrens * Virtual Device Labels 30fa9e4066Sahrens * --------------------- 31fa9e4066Sahrens * 32fa9e4066Sahrens * The vdev label serves several distinct purposes: 33fa9e4066Sahrens * 34fa9e4066Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 35fa9e4066Sahrens * identity within the pool. 36fa9e4066Sahrens * 37fa9e4066Sahrens * 2. Verify that all the devices given in a configuration are present 38fa9e4066Sahrens * within the pool. 39fa9e4066Sahrens * 40fa9e4066Sahrens * 3. Determine the uberblock for the pool. 41fa9e4066Sahrens * 42fa9e4066Sahrens * 4. In case of an import operation, determine the configuration of the 43fa9e4066Sahrens * toplevel vdev of which it is a part. 44fa9e4066Sahrens * 45fa9e4066Sahrens * 5. If an import operation cannot find all the devices in the pool, 46fa9e4066Sahrens * provide enough information to the administrator to determine which 47fa9e4066Sahrens * devices are missing. 48fa9e4066Sahrens * 49fa9e4066Sahrens * It is important to note that while the kernel is responsible for writing the 50fa9e4066Sahrens * label, it only consumes the information in the first three cases. The 51fa9e4066Sahrens * latter information is only consumed in userland when determining the 52fa9e4066Sahrens * configuration to import a pool. 53fa9e4066Sahrens * 54fa9e4066Sahrens * 55fa9e4066Sahrens * Label Organization 56fa9e4066Sahrens * ------------------ 57fa9e4066Sahrens * 58fa9e4066Sahrens * Before describing the contents of the label, it's important to understand how 59fa9e4066Sahrens * the labels are written and updated with respect to the uberblock. 60fa9e4066Sahrens * 61fa9e4066Sahrens * When the pool configuration is altered, either because it was newly created 62fa9e4066Sahrens * or a device was added, we want to update all the labels such that we can deal 63fa9e4066Sahrens * with fatal failure at any point. To this end, each disk has two labels which 64fa9e4066Sahrens * are updated before and after the uberblock is synced. Assuming we have 65fa9e4066Sahrens * labels and an uberblock with the following transacation groups: 66fa9e4066Sahrens * 67fa9e4066Sahrens * L1 UB L2 68fa9e4066Sahrens * +------+ +------+ +------+ 69fa9e4066Sahrens * | | | | | | 70fa9e4066Sahrens * | t10 | | t10 | | t10 | 71fa9e4066Sahrens * | | | | | | 72fa9e4066Sahrens * +------+ +------+ +------+ 73fa9e4066Sahrens * 74fa9e4066Sahrens * In this stable state, the labels and the uberblock were all updated within 75fa9e4066Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 76fa9e4066Sahrens * that we can detect when we fail partway through writing the label. 77fa9e4066Sahrens * 78fa9e4066Sahrens * In order to identify which labels are valid, the labels are written in the 79fa9e4066Sahrens * following manner: 80fa9e4066Sahrens * 81fa9e4066Sahrens * 1. For each vdev, update 'L1' to the new label 82fa9e4066Sahrens * 2. Update the uberblock 83fa9e4066Sahrens * 3. For each vdev, update 'L2' to the new label 84fa9e4066Sahrens * 85fa9e4066Sahrens * Given arbitrary failure, we can determine the correct label to use based on 86fa9e4066Sahrens * the transaction group. If we fail after updating L1 but before updating the 87fa9e4066Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 88fa9e4066Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 89fa9e4066Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 90fa9e4066Sahrens * therefore L1 is valid. 91fa9e4066Sahrens * 92fa9e4066Sahrens * Another added complexity is that not every label is updated when the config 93fa9e4066Sahrens * is synced. If we add a single device, we do not want to have to re-write 94fa9e4066Sahrens * every label for every device in the pool. This means that both L1 and L2 may 95fa9e4066Sahrens * be older than the pool uberblock, because the necessary information is stored 96fa9e4066Sahrens * on another vdev. 97fa9e4066Sahrens * 98fa9e4066Sahrens * 99fa9e4066Sahrens * On-disk Format 100fa9e4066Sahrens * -------------- 101fa9e4066Sahrens * 102fa9e4066Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 103fa9e4066Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 104fa9e4066Sahrens * VTOC disk labels, but is otherwise ignored. 105fa9e4066Sahrens * 106fa9e4066Sahrens * The first half of the label is a packed nvlist which contains pool wide 107fa9e4066Sahrens * properties, per-vdev properties, and configuration information. It is 108fa9e4066Sahrens * described in more detail below. 109fa9e4066Sahrens * 110fa9e4066Sahrens * The latter half of the label consists of a redundant array of uberblocks. 111fa9e4066Sahrens * These uberblocks are updated whenever a transaction group is committed, 112fa9e4066Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 113fa9e4066Sahrens * vdev for the 'best' uberblock. 114fa9e4066Sahrens * 115fa9e4066Sahrens * 116fa9e4066Sahrens * Configuration Information 117fa9e4066Sahrens * ------------------------- 118fa9e4066Sahrens * 119fa9e4066Sahrens * The nvlist describing the pool and vdev contains the following elements: 120fa9e4066Sahrens * 121fa9e4066Sahrens * version ZFS on-disk version 122fa9e4066Sahrens * name Pool name 123fa9e4066Sahrens * state Pool state 124fa9e4066Sahrens * txg Transaction group in which this label was written 125fa9e4066Sahrens * pool_guid Unique identifier for this pool 126fa9e4066Sahrens * vdev_tree An nvlist describing vdev tree. 127fa9e4066Sahrens * 128fa9e4066Sahrens * Each leaf device label also contains the following: 129fa9e4066Sahrens * 130fa9e4066Sahrens * top_guid Unique ID for top-level vdev in which this is contained 131fa9e4066Sahrens * guid Unique ID for the leaf vdev 132fa9e4066Sahrens * 133fa9e4066Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 134fa9e4066Sahrens */ 135fa9e4066Sahrens 136fa9e4066Sahrens #include <sys/zfs_context.h> 137fa9e4066Sahrens #include <sys/spa.h> 138fa9e4066Sahrens #include <sys/spa_impl.h> 139fa9e4066Sahrens #include <sys/dmu.h> 140fa9e4066Sahrens #include <sys/zap.h> 141fa9e4066Sahrens #include <sys/vdev.h> 142fa9e4066Sahrens #include <sys/vdev_impl.h> 143fa9e4066Sahrens #include <sys/uberblock_impl.h> 144fa9e4066Sahrens #include <sys/metaslab.h> 145fa9e4066Sahrens #include <sys/zio.h> 146fa9e4066Sahrens #include <sys/fs/zfs.h> 147fa9e4066Sahrens 148fa9e4066Sahrens /* 149fa9e4066Sahrens * Basic routines to read and write from a vdev label. 150fa9e4066Sahrens * Used throughout the rest of this file. 151fa9e4066Sahrens */ 152fa9e4066Sahrens uint64_t 153fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 154fa9e4066Sahrens { 155*ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 156*ecc2d604Sbonwick 157fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 158fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 159fa9e4066Sahrens } 160fa9e4066Sahrens 161fa9e4066Sahrens static void 162fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 163fa9e4066Sahrens uint64_t size, zio_done_func_t *done, void *private) 164fa9e4066Sahrens { 165fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 166fa9e4066Sahrens 167fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 168fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 169fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 170ea8dc4b6Seschrock ZIO_PRIORITY_SYNC_READ, 171ea8dc4b6Seschrock ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE)); 172fa9e4066Sahrens } 173fa9e4066Sahrens 174fa9e4066Sahrens static void 175fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 176fa9e4066Sahrens uint64_t size, zio_done_func_t *done, void *private) 177fa9e4066Sahrens { 178fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 179fa9e4066Sahrens 180fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 181fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 182fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 183ea8dc4b6Seschrock ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL)); 184fa9e4066Sahrens } 185fa9e4066Sahrens 186fa9e4066Sahrens /* 187fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 188fa9e4066Sahrens */ 189fa9e4066Sahrens nvlist_t * 190fa9e4066Sahrens vdev_config_generate(vdev_t *vd, int getstats) 191fa9e4066Sahrens { 192fa9e4066Sahrens nvlist_t *nv = NULL; 193fa9e4066Sahrens 194ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 195fa9e4066Sahrens 196fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 197fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 198fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) == 0); 199fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 200fa9e4066Sahrens 201fa9e4066Sahrens if (vd->vdev_path != NULL) 202fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 203fa9e4066Sahrens vd->vdev_path) == 0); 204fa9e4066Sahrens 205fa9e4066Sahrens if (vd->vdev_devid != NULL) 206fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 207fa9e4066Sahrens vd->vdev_devid) == 0); 208fa9e4066Sahrens 209afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 210afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 211afefbcddSeschrock vd->vdev_wholedisk) == 0); 212afefbcddSeschrock 213ea8dc4b6Seschrock if (vd->vdev_not_present) 214ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 215ea8dc4b6Seschrock 216fa9e4066Sahrens if (vd == vd->vdev_top) { 217fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 218fa9e4066Sahrens vd->vdev_ms_array) == 0); 219fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 220fa9e4066Sahrens vd->vdev_ms_shift) == 0); 221fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 222fa9e4066Sahrens vd->vdev_ashift) == 0); 223fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 224fa9e4066Sahrens vd->vdev_asize) == 0); 225fa9e4066Sahrens } 226fa9e4066Sahrens 227fa9e4066Sahrens if (vd->vdev_dtl.smo_object != 0) 228fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 229fa9e4066Sahrens vd->vdev_dtl.smo_object) == 0); 230fa9e4066Sahrens 231fa9e4066Sahrens if (getstats) { 232fa9e4066Sahrens vdev_stat_t vs; 233fa9e4066Sahrens vdev_get_stats(vd, &vs); 234fa9e4066Sahrens VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 235fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 236fa9e4066Sahrens } 237fa9e4066Sahrens 238fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 239fa9e4066Sahrens nvlist_t **child; 240fa9e4066Sahrens int c; 241fa9e4066Sahrens 242fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 243fa9e4066Sahrens KM_SLEEP); 244fa9e4066Sahrens 245fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 246fa9e4066Sahrens child[c] = vdev_config_generate(vd->vdev_child[c], 247fa9e4066Sahrens getstats); 248fa9e4066Sahrens 249fa9e4066Sahrens VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 250fa9e4066Sahrens child, vd->vdev_children) == 0); 251fa9e4066Sahrens 252fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 253fa9e4066Sahrens nvlist_free(child[c]); 254fa9e4066Sahrens 255fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 256441d80aaSlling 257441d80aaSlling } else { 258*ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 259441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 260*ecc2d604Sbonwick B_TRUE) == 0); 261*ecc2d604Sbonwick else 262441d80aaSlling (void) nvlist_remove(nv, ZPOOL_CONFIG_OFFLINE, 263*ecc2d604Sbonwick DATA_TYPE_UINT64); 264fa9e4066Sahrens } 265fa9e4066Sahrens 266fa9e4066Sahrens return (nv); 267fa9e4066Sahrens } 268fa9e4066Sahrens 269fa9e4066Sahrens nvlist_t * 270fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 271fa9e4066Sahrens { 2720373e76bSbonwick spa_t *spa = vd->vdev_spa; 273fa9e4066Sahrens nvlist_t *config = NULL; 274fa9e4066Sahrens vdev_phys_t *vp; 275fa9e4066Sahrens zio_t *zio; 276fa9e4066Sahrens int l; 277fa9e4066Sahrens 2780373e76bSbonwick ASSERT(spa_config_held(spa, RW_READER)); 2790373e76bSbonwick 280fa9e4066Sahrens if (vdev_is_dead(vd)) 281fa9e4066Sahrens return (NULL); 282fa9e4066Sahrens 283fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 284fa9e4066Sahrens 285fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 286fa9e4066Sahrens 2870373e76bSbonwick zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | 288ea8dc4b6Seschrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD); 289fa9e4066Sahrens 290fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 291fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 292fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 293fa9e4066Sahrens 294fa9e4066Sahrens if (zio_wait(zio) == 0 && 295fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 296ea8dc4b6Seschrock &config, 0) == 0) 297fa9e4066Sahrens break; 298fa9e4066Sahrens 299fa9e4066Sahrens if (config != NULL) { 300fa9e4066Sahrens nvlist_free(config); 301fa9e4066Sahrens config = NULL; 302fa9e4066Sahrens } 303fa9e4066Sahrens } 304fa9e4066Sahrens 305fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 306fa9e4066Sahrens 307fa9e4066Sahrens return (config); 308fa9e4066Sahrens } 309fa9e4066Sahrens 310fa9e4066Sahrens int 311fa9e4066Sahrens vdev_label_init(vdev_t *vd, uint64_t crtxg) 312fa9e4066Sahrens { 313fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 314fa9e4066Sahrens nvlist_t *label; 315fa9e4066Sahrens vdev_phys_t *vp; 316fa9e4066Sahrens vdev_boot_header_t *vb; 317*ecc2d604Sbonwick uberblock_t *ub; 318fa9e4066Sahrens zio_t *zio; 319fa9e4066Sahrens int l, c, n; 320fa9e4066Sahrens char *buf; 321fa9e4066Sahrens size_t buflen; 322fa9e4066Sahrens int error; 323fa9e4066Sahrens 3240373e76bSbonwick ASSERT(spa_config_held(spa, RW_WRITER)); 3250373e76bSbonwick 326fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 327fa9e4066Sahrens if ((error = vdev_label_init(vd->vdev_child[c], crtxg)) != 0) 328fa9e4066Sahrens return (error); 329fa9e4066Sahrens 330fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 331fa9e4066Sahrens return (0); 332fa9e4066Sahrens 333fa9e4066Sahrens /* 334fa9e4066Sahrens * Make sure each leaf device is writable, and zero its initial content. 335fa9e4066Sahrens * Along the way, also make sure that no leaf is already in use. 336fa9e4066Sahrens * Note that it's important to do this sequentially, not in parallel, 337fa9e4066Sahrens * so that we catch cases of multiple use of the same leaf vdev in 338fa9e4066Sahrens * the vdev we're creating -- e.g. mirroring a disk with itself. 339fa9e4066Sahrens */ 340fa9e4066Sahrens if (vdev_is_dead(vd)) 341fa9e4066Sahrens return (EIO); 342fa9e4066Sahrens 343fa9e4066Sahrens /* 344fa9e4066Sahrens * Check whether this device is already in use. 345fa9e4066Sahrens * Ignore the check if crtxg == 0, which we use for device removal. 346fa9e4066Sahrens */ 347ea8dc4b6Seschrock if (crtxg != 0 && 348ea8dc4b6Seschrock (label = vdev_label_read_config(vd)) != NULL) { 349ea8dc4b6Seschrock uint64_t state, pool_guid, device_guid, txg; 350fa9e4066Sahrens uint64_t mycrtxg = 0; 351fa9e4066Sahrens 352fa9e4066Sahrens (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 353fa9e4066Sahrens &mycrtxg); 354fa9e4066Sahrens 355ea8dc4b6Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 356fa9e4066Sahrens &state) == 0 && state == POOL_STATE_ACTIVE && 357fa9e4066Sahrens nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 358fa9e4066Sahrens &pool_guid) == 0 && 359fa9e4066Sahrens nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 360fa9e4066Sahrens &device_guid) == 0 && 361fa9e4066Sahrens spa_guid_exists(pool_guid, device_guid) && 362fa9e4066Sahrens nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 363fa9e4066Sahrens &txg) == 0 && (txg != 0 || mycrtxg == crtxg)) { 364fa9e4066Sahrens dprintf("vdev %s in use, pool_state %d\n", 365fa9e4066Sahrens vdev_description(vd), state); 366fa9e4066Sahrens nvlist_free(label); 367fa9e4066Sahrens return (EBUSY); 368fa9e4066Sahrens } 369fa9e4066Sahrens nvlist_free(label); 370fa9e4066Sahrens } 371fa9e4066Sahrens 372fa9e4066Sahrens /* 373fa9e4066Sahrens * The device isn't in use, so initialize its label. 374fa9e4066Sahrens */ 375fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 376fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 377fa9e4066Sahrens 378fa9e4066Sahrens /* 379fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 380fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 381fa9e4066Sahrens * really part of an active pool just yet. The labels will 382fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 383fa9e4066Sahrens */ 3840373e76bSbonwick label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 385fa9e4066Sahrens 386fa9e4066Sahrens /* 387fa9e4066Sahrens * Add our creation time. This allows us to detect multiple vdev 388fa9e4066Sahrens * uses as described above, and automatically expires if we fail. 389fa9e4066Sahrens */ 390fa9e4066Sahrens VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, crtxg) == 0); 391fa9e4066Sahrens 392fa9e4066Sahrens buf = vp->vp_nvlist; 393fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 394fa9e4066Sahrens 395ea8dc4b6Seschrock if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) != 0) { 396fa9e4066Sahrens nvlist_free(label); 397fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 398fa9e4066Sahrens return (EINVAL); 399fa9e4066Sahrens } 400fa9e4066Sahrens 401fa9e4066Sahrens /* 402fa9e4066Sahrens * Initialize boot block header. 403fa9e4066Sahrens */ 404fa9e4066Sahrens vb = zio_buf_alloc(sizeof (vdev_boot_header_t)); 405fa9e4066Sahrens bzero(vb, sizeof (vdev_boot_header_t)); 406fa9e4066Sahrens vb->vb_magic = VDEV_BOOT_MAGIC; 407fa9e4066Sahrens vb->vb_version = VDEV_BOOT_VERSION; 408fa9e4066Sahrens vb->vb_offset = VDEV_BOOT_OFFSET; 409fa9e4066Sahrens vb->vb_size = VDEV_BOOT_SIZE; 410fa9e4066Sahrens 411fa9e4066Sahrens /* 412fa9e4066Sahrens * Initialize uberblock template. 413fa9e4066Sahrens */ 414*ecc2d604Sbonwick ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 415*ecc2d604Sbonwick bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 416*ecc2d604Sbonwick *ub = spa->spa_uberblock; 417*ecc2d604Sbonwick ub->ub_txg = 0; 418fa9e4066Sahrens 419fa9e4066Sahrens /* 420fa9e4066Sahrens * Write everything in parallel. 421fa9e4066Sahrens */ 422fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 423fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 424fa9e4066Sahrens 425fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 426fa9e4066Sahrens 427fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 428fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 429fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 430fa9e4066Sahrens 431fa9e4066Sahrens vdev_label_write(zio, vd, l, vb, 432fa9e4066Sahrens offsetof(vdev_label_t, vl_boot_header), 433fa9e4066Sahrens sizeof (vdev_boot_header_t), NULL, NULL); 434fa9e4066Sahrens 435*ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 436*ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 437*ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 438*ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), NULL, NULL); 439fa9e4066Sahrens } 440fa9e4066Sahrens } 441fa9e4066Sahrens 442fa9e4066Sahrens error = zio_wait(zio); 443fa9e4066Sahrens 444fa9e4066Sahrens nvlist_free(label); 445*ecc2d604Sbonwick zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 446fa9e4066Sahrens zio_buf_free(vb, sizeof (vdev_boot_header_t)); 447fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 448fa9e4066Sahrens 449fa9e4066Sahrens return (error); 450fa9e4066Sahrens } 451fa9e4066Sahrens 452fa9e4066Sahrens /* 453fa9e4066Sahrens * ========================================================================== 454fa9e4066Sahrens * uberblock load/sync 455fa9e4066Sahrens * ========================================================================== 456fa9e4066Sahrens */ 457fa9e4066Sahrens 458fa9e4066Sahrens /* 459fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 460fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 461fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 462fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 463fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 464fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 465fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 466fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 467fa9e4066Sahrens */ 468fa9e4066Sahrens static int 469fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 470fa9e4066Sahrens { 471fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 472fa9e4066Sahrens return (-1); 473fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 474fa9e4066Sahrens return (1); 475fa9e4066Sahrens 476fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 477fa9e4066Sahrens return (-1); 478fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 479fa9e4066Sahrens return (1); 480fa9e4066Sahrens 481fa9e4066Sahrens return (0); 482fa9e4066Sahrens } 483fa9e4066Sahrens 484fa9e4066Sahrens static void 485fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 486fa9e4066Sahrens { 487*ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 488fa9e4066Sahrens uberblock_t *ubbest = zio->io_private; 489fa9e4066Sahrens spa_t *spa = zio->io_spa; 490fa9e4066Sahrens 491*ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 492fa9e4066Sahrens 493ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 494fa9e4066Sahrens mutex_enter(&spa->spa_uberblock_lock); 495fa9e4066Sahrens if (vdev_uberblock_compare(ub, ubbest) > 0) 496fa9e4066Sahrens *ubbest = *ub; 497fa9e4066Sahrens mutex_exit(&spa->spa_uberblock_lock); 498fa9e4066Sahrens } 499fa9e4066Sahrens 500fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 501fa9e4066Sahrens } 502fa9e4066Sahrens 503fa9e4066Sahrens void 504fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 505fa9e4066Sahrens { 506fa9e4066Sahrens int l, c, n; 507fa9e4066Sahrens 508fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 509fa9e4066Sahrens vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 510fa9e4066Sahrens 511fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 512fa9e4066Sahrens return; 513fa9e4066Sahrens 514fa9e4066Sahrens if (vdev_is_dead(vd)) 515fa9e4066Sahrens return; 516fa9e4066Sahrens 517fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 518*ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 519fa9e4066Sahrens vdev_label_read(zio, vd, l, 520*ecc2d604Sbonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 521*ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 522*ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 523fa9e4066Sahrens vdev_uberblock_load_done, ubbest); 524fa9e4066Sahrens } 525fa9e4066Sahrens } 526fa9e4066Sahrens } 527fa9e4066Sahrens 528fa9e4066Sahrens /* 529fa9e4066Sahrens * Write the uberblock to both labels of all leaves of the specified vdev. 5300373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 531fa9e4066Sahrens */ 532fa9e4066Sahrens static void 533fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 534fa9e4066Sahrens { 535fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 536fa9e4066Sahrens 5370373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 538fa9e4066Sahrens atomic_add_64(good_writes, 1); 539fa9e4066Sahrens } 540fa9e4066Sahrens 541fa9e4066Sahrens static void 542*ecc2d604Sbonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, uint64_t txg) 543fa9e4066Sahrens { 544fa9e4066Sahrens int l, c, n; 545fa9e4066Sahrens 546fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 547*ecc2d604Sbonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], txg); 548fa9e4066Sahrens 549fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 550fa9e4066Sahrens return; 551fa9e4066Sahrens 552fa9e4066Sahrens if (vdev_is_dead(vd)) 553fa9e4066Sahrens return; 554fa9e4066Sahrens 555*ecc2d604Sbonwick n = txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 556fa9e4066Sahrens 557*ecc2d604Sbonwick ASSERT(ub->ub_txg == txg); 558fa9e4066Sahrens 559fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) 560*ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 561*ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 562*ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 563*ecc2d604Sbonwick vdev_uberblock_sync_done, NULL); 564fa9e4066Sahrens 565fa9e4066Sahrens dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg); 566fa9e4066Sahrens } 567fa9e4066Sahrens 568fa9e4066Sahrens static int 569*ecc2d604Sbonwick vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *vd, uint64_t txg) 570fa9e4066Sahrens { 571*ecc2d604Sbonwick uberblock_t *ubbuf; 572*ecc2d604Sbonwick size_t size = vd->vdev_top ? VDEV_UBERBLOCK_SIZE(vd) : SPA_MAXBLOCKSIZE; 573fa9e4066Sahrens uint64_t *good_writes; 574fa9e4066Sahrens zio_t *zio; 575fa9e4066Sahrens int error; 576fa9e4066Sahrens 577*ecc2d604Sbonwick ubbuf = zio_buf_alloc(size); 578*ecc2d604Sbonwick bzero(ubbuf, size); 579*ecc2d604Sbonwick *ubbuf = *ub; 580fa9e4066Sahrens 581fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 582fa9e4066Sahrens 583fa9e4066Sahrens zio = zio_root(spa, NULL, good_writes, 584fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 585fa9e4066Sahrens 586*ecc2d604Sbonwick vdev_uberblock_sync(zio, ubbuf, vd, txg); 587fa9e4066Sahrens 588fa9e4066Sahrens error = zio_wait(zio); 589fa9e4066Sahrens 590fa9e4066Sahrens if (error && *good_writes != 0) { 591fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 592fa9e4066Sahrens error = 0; 593fa9e4066Sahrens } 594fa9e4066Sahrens 595fa9e4066Sahrens /* 596fa9e4066Sahrens * It's possible to have no good writes and no error if every vdev is in 597fa9e4066Sahrens * the CANT_OPEN state. 598fa9e4066Sahrens */ 599fa9e4066Sahrens if (*good_writes == 0 && error == 0) 600fa9e4066Sahrens error = EIO; 601fa9e4066Sahrens 602fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 603*ecc2d604Sbonwick zio_buf_free(ubbuf, size); 604fa9e4066Sahrens 605fa9e4066Sahrens return (error); 606fa9e4066Sahrens } 607fa9e4066Sahrens 608fa9e4066Sahrens /* 609fa9e4066Sahrens * Sync out an individual vdev. 610fa9e4066Sahrens */ 611fa9e4066Sahrens static void 612fa9e4066Sahrens vdev_sync_label_done(zio_t *zio) 613fa9e4066Sahrens { 614fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 615fa9e4066Sahrens 616fa9e4066Sahrens if (zio->io_error == 0) 617fa9e4066Sahrens atomic_add_64(good_writes, 1); 618fa9e4066Sahrens } 619fa9e4066Sahrens 620fa9e4066Sahrens static void 621fa9e4066Sahrens vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg) 622fa9e4066Sahrens { 623fa9e4066Sahrens nvlist_t *label; 624fa9e4066Sahrens vdev_phys_t *vp; 625fa9e4066Sahrens char *buf; 626fa9e4066Sahrens size_t buflen; 627fa9e4066Sahrens int c; 628fa9e4066Sahrens 629fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 630fa9e4066Sahrens vdev_sync_label(zio, vd->vdev_child[c], l, txg); 631fa9e4066Sahrens 632fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 633fa9e4066Sahrens return; 634fa9e4066Sahrens 635fa9e4066Sahrens if (vdev_is_dead(vd)) 636fa9e4066Sahrens return; 637fa9e4066Sahrens 638fa9e4066Sahrens /* 639fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 640fa9e4066Sahrens */ 6410373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 642fa9e4066Sahrens 643fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 644fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 645fa9e4066Sahrens 646fa9e4066Sahrens buf = vp->vp_nvlist; 647fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 648fa9e4066Sahrens 649ea8dc4b6Seschrock if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) 650fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 651fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 652fa9e4066Sahrens vdev_sync_label_done, NULL); 653fa9e4066Sahrens 654fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 655fa9e4066Sahrens nvlist_free(label); 656fa9e4066Sahrens 657fa9e4066Sahrens dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg); 658fa9e4066Sahrens } 659fa9e4066Sahrens 660fa9e4066Sahrens static int 661fa9e4066Sahrens vdev_sync_labels(vdev_t *vd, int l, uint64_t txg) 662fa9e4066Sahrens { 663fa9e4066Sahrens uint64_t *good_writes; 664fa9e4066Sahrens zio_t *zio; 665fa9e4066Sahrens int error; 666fa9e4066Sahrens 667fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 668fa9e4066Sahrens 669fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 670fa9e4066Sahrens 671fa9e4066Sahrens zio = zio_root(vd->vdev_spa, NULL, good_writes, 672fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 673fa9e4066Sahrens 674fa9e4066Sahrens /* 675fa9e4066Sahrens * Recursively kick off writes to all labels. 676fa9e4066Sahrens */ 677fa9e4066Sahrens vdev_sync_label(zio, vd, l, txg); 678fa9e4066Sahrens 679fa9e4066Sahrens error = zio_wait(zio); 680fa9e4066Sahrens 681fa9e4066Sahrens if (error && *good_writes != 0) { 682fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 683fa9e4066Sahrens error = 0; 684fa9e4066Sahrens } 685fa9e4066Sahrens 686fa9e4066Sahrens if (*good_writes == 0 && error == 0) 687fa9e4066Sahrens error = ENODEV; 688fa9e4066Sahrens 689fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 690fa9e4066Sahrens 691fa9e4066Sahrens return (error); 692fa9e4066Sahrens } 693fa9e4066Sahrens 694fa9e4066Sahrens /* 695fa9e4066Sahrens * Sync the entire vdev configuration. 696fa9e4066Sahrens * 697fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 698fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 699fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 700fa9e4066Sahrens * describe the failure semantics at each stage. 701fa9e4066Sahrens * 702fa9e4066Sahrens * Moreover, it is designed to be idempotent: if spa_sync_labels() fails 703fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 704fa9e4066Sahrens */ 705fa9e4066Sahrens int 7060373e76bSbonwick vdev_config_sync(vdev_t *uvd, uint64_t txg) 707fa9e4066Sahrens { 7080373e76bSbonwick spa_t *spa = uvd->vdev_spa; 709fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 710fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 7110373e76bSbonwick vdev_t *vd; 712fa9e4066Sahrens zio_t *zio; 713f65ea9b9Sbonwick int l, error; 714fa9e4066Sahrens 715fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 716fa9e4066Sahrens 717fa9e4066Sahrens /* 718fa9e4066Sahrens * If this isn't a resync due to I/O errors, and nothing changed 719fa9e4066Sahrens * in this transaction group, and the vdev configuration hasn't changed, 7200373e76bSbonwick * then there's nothing to do. 721fa9e4066Sahrens */ 722fa9e4066Sahrens if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE && 723fa9e4066Sahrens list_is_empty(&spa->spa_dirty_list)) { 724fa9e4066Sahrens dprintf("nothing to sync in %s in txg %llu\n", 725fa9e4066Sahrens spa_name(spa), txg); 726fa9e4066Sahrens return (0); 727fa9e4066Sahrens } 728fa9e4066Sahrens 729fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 730fa9e4066Sahrens return (0); 731fa9e4066Sahrens 7320373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 7330373e76bSbonwick 734fa9e4066Sahrens dprintf("syncing %s txg %llu\n", spa_name(spa), txg); 735fa9e4066Sahrens 736fa9e4066Sahrens /* 737fa9e4066Sahrens * Flush the write cache of every disk that's been written to 738fa9e4066Sahrens * in this transaction group. This ensures that all blocks 739fa9e4066Sahrens * written in this txg will be committed to stable storage 740fa9e4066Sahrens * before any uberblock that references them. 741fa9e4066Sahrens */ 742fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 743fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 744fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 745fa9e4066Sahrens vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) { 746fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 747fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 748fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 749fa9e4066Sahrens } 750fa9e4066Sahrens (void) zio_wait(zio); 751fa9e4066Sahrens 752fa9e4066Sahrens /* 753fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 754fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 755fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 756fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 757fa9e4066Sahrens * which have not yet been touched, will still be valid. 758fa9e4066Sahrens */ 759fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 760fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 761fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 762fa9e4066Sahrens if (l & 1) 763fa9e4066Sahrens continue; 764fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 765fa9e4066Sahrens return (error); 766fa9e4066Sahrens } 767fa9e4066Sahrens } 768fa9e4066Sahrens 769fa9e4066Sahrens /* 770fa9e4066Sahrens * Flush the new labels to disk. This ensures that all even-label 771fa9e4066Sahrens * updates are committed to stable storage before the uberblock update. 772fa9e4066Sahrens */ 773fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 774fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 775fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 776fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 777fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 778fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 779fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 780fa9e4066Sahrens } 781fa9e4066Sahrens (void) zio_wait(zio); 782fa9e4066Sahrens 783fa9e4066Sahrens /* 7840373e76bSbonwick * Sync the uberblocks to all vdevs in the tree specified by uvd. 7850373e76bSbonwick * If the system dies in the middle of this step, there are two cases 7860373e76bSbonwick * to consider, and the on-disk state is consistent either way: 787fa9e4066Sahrens * 788fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 789fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 790fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 791fa9e4066Sahrens * to that uberblock. 792fa9e4066Sahrens * 793fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 794fa9e4066Sahrens * will be the newest, and the even labels (which had all 795fa9e4066Sahrens * been successfully committed) will be valid with respect 796fa9e4066Sahrens * to the new uberblocks. 797fa9e4066Sahrens */ 798fa9e4066Sahrens if ((error = vdev_uberblock_sync_tree(spa, ub, uvd, txg)) != 0) 799fa9e4066Sahrens return (error); 800fa9e4066Sahrens 801fa9e4066Sahrens /* 802fa9e4066Sahrens * Flush the uberblocks to disk. This ensures that the odd labels 803fa9e4066Sahrens * are no longer needed (because the new uberblocks and the even 804fa9e4066Sahrens * labels are safely on disk), so it is safe to overwrite them. 805fa9e4066Sahrens */ 806fa9e4066Sahrens (void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE, 807fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 808fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 809fa9e4066Sahrens 810fa9e4066Sahrens /* 811fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 812fa9e4066Sahrens * in the middle of this process, the even labels and the new 813fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 814fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 815fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 816fa9e4066Sahrens * all labels will be brought up to date. 817fa9e4066Sahrens */ 818fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 819fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 820fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 821fa9e4066Sahrens if ((l & 1) == 0) 822fa9e4066Sahrens continue; 823fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 824fa9e4066Sahrens return (error); 825fa9e4066Sahrens } 826fa9e4066Sahrens } 827fa9e4066Sahrens 828fa9e4066Sahrens /* 829fa9e4066Sahrens * Flush the new labels to disk. This ensures that all odd-label 830fa9e4066Sahrens * updates are committed to stable storage before the next 831fa9e4066Sahrens * transaction group begins. 832fa9e4066Sahrens */ 833fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 834fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 835fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 836fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 837fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 838fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 839fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 840fa9e4066Sahrens } 841fa9e4066Sahrens (void) zio_wait(zio); 842fa9e4066Sahrens 843fa9e4066Sahrens return (0); 844fa9e4066Sahrens } 845