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 /* 2239c23413Seschrock * Copyright 2007 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 653d7072f8Seschrock * labels and an uberblock with the following transaction 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 { 155ecc2d604Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 156e7437265Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 157ecc2d604Sbonwick 158fa9e4066Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 159fa9e4066Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 160fa9e4066Sahrens } 161fa9e4066Sahrens 162fa9e4066Sahrens static void 163fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 164fa9e4066Sahrens uint64_t size, zio_done_func_t *done, void *private) 165fa9e4066Sahrens { 166fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 167fa9e4066Sahrens 168fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 169fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 170fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 171ea8dc4b6Seschrock ZIO_PRIORITY_SYNC_READ, 172*fa94a07fSbrendan ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 173*fa94a07fSbrendan B_TRUE)); 174fa9e4066Sahrens } 175fa9e4066Sahrens 176fa9e4066Sahrens static void 177fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 178fa9e4066Sahrens uint64_t size, zio_done_func_t *done, void *private) 179fa9e4066Sahrens { 180fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 181fa9e4066Sahrens 182fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 183fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 184fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 185*fa94a07fSbrendan ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL, 186*fa94a07fSbrendan B_TRUE)); 187fa9e4066Sahrens } 188fa9e4066Sahrens 189fa9e4066Sahrens /* 190fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 191fa9e4066Sahrens */ 192fa9e4066Sahrens nvlist_t * 19399653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 194*fa94a07fSbrendan boolean_t isspare, boolean_t isl2cache) 195fa9e4066Sahrens { 196fa9e4066Sahrens nvlist_t *nv = NULL; 197fa9e4066Sahrens 198ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 199fa9e4066Sahrens 200fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 201fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 202*fa94a07fSbrendan if (!isspare && !isl2cache) 20399653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 20499653d4eSeschrock == 0); 205fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 206fa9e4066Sahrens 207fa9e4066Sahrens if (vd->vdev_path != NULL) 208fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 209fa9e4066Sahrens vd->vdev_path) == 0); 210fa9e4066Sahrens 211fa9e4066Sahrens if (vd->vdev_devid != NULL) 212fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 213fa9e4066Sahrens vd->vdev_devid) == 0); 214fa9e4066Sahrens 2153d7072f8Seschrock if (vd->vdev_physpath != NULL) 2163d7072f8Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 2173d7072f8Seschrock vd->vdev_physpath) == 0); 2183d7072f8Seschrock 21999653d4eSeschrock if (vd->vdev_nparity != 0) { 22099653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 22199653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 22299653d4eSeschrock 22399653d4eSeschrock /* 22499653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 22599653d4eSeschrock * into a crufty old storage pool. 22699653d4eSeschrock */ 22799653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 22899653d4eSeschrock (vd->vdev_nparity == 2 && 229e7437265Sahrens spa_version(spa) >= SPA_VERSION_RAID6)); 23099653d4eSeschrock 23199653d4eSeschrock /* 23299653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 23399653d4eSeschrock * that only support a single parity device -- older software 23499653d4eSeschrock * will just ignore it. 23599653d4eSeschrock */ 23699653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 23799653d4eSeschrock vd->vdev_nparity) == 0); 23899653d4eSeschrock } 23999653d4eSeschrock 240afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 241afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 242afefbcddSeschrock vd->vdev_wholedisk) == 0); 243afefbcddSeschrock 244ea8dc4b6Seschrock if (vd->vdev_not_present) 245ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 246ea8dc4b6Seschrock 24799653d4eSeschrock if (vd->vdev_isspare) 24899653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 24999653d4eSeschrock 250*fa94a07fSbrendan if (!isspare && !isl2cache && vd == vd->vdev_top) { 251fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 252fa9e4066Sahrens vd->vdev_ms_array) == 0); 253fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 254fa9e4066Sahrens vd->vdev_ms_shift) == 0); 255fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 256fa9e4066Sahrens vd->vdev_ashift) == 0); 257fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 258fa9e4066Sahrens vd->vdev_asize) == 0); 2598654d025Sperrin VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 2608654d025Sperrin vd->vdev_islog) == 0); 261fa9e4066Sahrens } 262fa9e4066Sahrens 263fa9e4066Sahrens if (vd->vdev_dtl.smo_object != 0) 264fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 265fa9e4066Sahrens vd->vdev_dtl.smo_object) == 0); 266fa9e4066Sahrens 267fa9e4066Sahrens if (getstats) { 268fa9e4066Sahrens vdev_stat_t vs; 269fa9e4066Sahrens vdev_get_stats(vd, &vs); 270fa9e4066Sahrens VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 271fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 272fa9e4066Sahrens } 273fa9e4066Sahrens 274fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 275fa9e4066Sahrens nvlist_t **child; 276fa9e4066Sahrens int c; 277fa9e4066Sahrens 278fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 279fa9e4066Sahrens KM_SLEEP); 280fa9e4066Sahrens 281fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 28299653d4eSeschrock child[c] = vdev_config_generate(spa, vd->vdev_child[c], 283*fa94a07fSbrendan getstats, isspare, isl2cache); 284fa9e4066Sahrens 285fa9e4066Sahrens VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 286fa9e4066Sahrens child, vd->vdev_children) == 0); 287fa9e4066Sahrens 288fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 289fa9e4066Sahrens nvlist_free(child[c]); 290fa9e4066Sahrens 291fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 292441d80aaSlling 293441d80aaSlling } else { 294ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 295441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 296ecc2d604Sbonwick B_TRUE) == 0); 2973d7072f8Seschrock if (vd->vdev_faulted) 2983d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 2993d7072f8Seschrock B_TRUE) == 0); 3003d7072f8Seschrock if (vd->vdev_degraded) 3013d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 3023d7072f8Seschrock B_TRUE) == 0); 3033d7072f8Seschrock if (vd->vdev_removed) 3043d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 3053d7072f8Seschrock B_TRUE) == 0); 3063d7072f8Seschrock if (vd->vdev_unspare) 3073d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 3083d7072f8Seschrock B_TRUE) == 0); 309fa9e4066Sahrens } 310fa9e4066Sahrens 311fa9e4066Sahrens return (nv); 312fa9e4066Sahrens } 313fa9e4066Sahrens 314fa9e4066Sahrens nvlist_t * 315fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 316fa9e4066Sahrens { 3170373e76bSbonwick spa_t *spa = vd->vdev_spa; 318fa9e4066Sahrens nvlist_t *config = NULL; 319fa9e4066Sahrens vdev_phys_t *vp; 320fa9e4066Sahrens zio_t *zio; 321fa9e4066Sahrens int l; 322fa9e4066Sahrens 32391ebeef5Sahrens ASSERT(spa_config_held(spa, RW_READER) || 32491ebeef5Sahrens spa_config_held(spa, RW_WRITER)); 3250373e76bSbonwick 3260a4e9518Sgw if (!vdev_readable(vd)) 327fa9e4066Sahrens return (NULL); 328fa9e4066Sahrens 329fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 330fa9e4066Sahrens 331fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 332fa9e4066Sahrens 3330373e76bSbonwick zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | 334ea8dc4b6Seschrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD); 335fa9e4066Sahrens 336fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 337fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 338fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 339fa9e4066Sahrens 340fa9e4066Sahrens if (zio_wait(zio) == 0 && 341fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 342ea8dc4b6Seschrock &config, 0) == 0) 343fa9e4066Sahrens break; 344fa9e4066Sahrens 345fa9e4066Sahrens if (config != NULL) { 346fa9e4066Sahrens nvlist_free(config); 347fa9e4066Sahrens config = NULL; 348fa9e4066Sahrens } 349fa9e4066Sahrens } 350fa9e4066Sahrens 351fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 352fa9e4066Sahrens 353fa9e4066Sahrens return (config); 354fa9e4066Sahrens } 355fa9e4066Sahrens 35639c23413Seschrock /* 35739c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 35839c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 35939c23413Seschrock */ 36039c23413Seschrock static boolean_t 36139c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 362*fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 36339c23413Seschrock { 36439c23413Seschrock spa_t *spa = vd->vdev_spa; 36539c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 36639c23413Seschrock uint64_t vdtxg = 0; 36739c23413Seschrock nvlist_t *label; 36839c23413Seschrock 36939c23413Seschrock if (spare_guid) 37039c23413Seschrock *spare_guid = 0ULL; 371*fa94a07fSbrendan if (l2cache_guid) 372*fa94a07fSbrendan *l2cache_guid = 0ULL; 37339c23413Seschrock 37439c23413Seschrock /* 37539c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 37639c23413Seschrock */ 37739c23413Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 37839c23413Seschrock return (B_FALSE); 37939c23413Seschrock 38039c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 38139c23413Seschrock &vdtxg); 38239c23413Seschrock 38339c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 38439c23413Seschrock &state) != 0 || 38539c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 38639c23413Seschrock &device_guid) != 0) { 38739c23413Seschrock nvlist_free(label); 38839c23413Seschrock return (B_FALSE); 38939c23413Seschrock } 39039c23413Seschrock 391*fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 39239c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 39339c23413Seschrock &pool_guid) != 0 || 39439c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 39539c23413Seschrock &txg) != 0)) { 39639c23413Seschrock nvlist_free(label); 39739c23413Seschrock return (B_FALSE); 39839c23413Seschrock } 39939c23413Seschrock 40039c23413Seschrock nvlist_free(label); 40139c23413Seschrock 40239c23413Seschrock /* 40339c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 40439c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 40539c23413Seschrock * spare (which we check for later on). 40639c23413Seschrock */ 407*fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 40839c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 409*fa94a07fSbrendan !spa_spare_exists(device_guid, NULL) && 410*fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 41139c23413Seschrock return (B_FALSE); 41239c23413Seschrock 41339c23413Seschrock /* 41439c23413Seschrock * If the transaction group is zero, then this an initialized (but 41539c23413Seschrock * unused) label. This is only an error if the create transaction 41639c23413Seschrock * on-disk is the same as the one we're using now, in which case the 41739c23413Seschrock * user has attempted to add the same vdev multiple times in the same 41839c23413Seschrock * transaction. 41939c23413Seschrock */ 420*fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 421*fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 42239c23413Seschrock return (B_TRUE); 42339c23413Seschrock 42439c23413Seschrock /* 42539c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 42639c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 427*fa94a07fSbrendan * to add. We also check if it is an l2cache device. 42839c23413Seschrock */ 42939c23413Seschrock if (spa_spare_exists(device_guid, &spare_pool) || 43039c23413Seschrock spa_has_spare(spa, device_guid)) { 43139c23413Seschrock if (spare_guid) 43239c23413Seschrock *spare_guid = device_guid; 43339c23413Seschrock 43439c23413Seschrock switch (reason) { 43539c23413Seschrock case VDEV_LABEL_CREATE: 436*fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 43739c23413Seschrock return (B_TRUE); 43839c23413Seschrock 43939c23413Seschrock case VDEV_LABEL_REPLACE: 44039c23413Seschrock return (!spa_has_spare(spa, device_guid) || 44139c23413Seschrock spare_pool != 0ULL); 44239c23413Seschrock 44339c23413Seschrock case VDEV_LABEL_SPARE: 44439c23413Seschrock return (spa_has_spare(spa, device_guid)); 44539c23413Seschrock } 44639c23413Seschrock } 44739c23413Seschrock 448*fa94a07fSbrendan /* 449*fa94a07fSbrendan * Check to see if this is an l2cache device. 450*fa94a07fSbrendan */ 451*fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 452*fa94a07fSbrendan return (B_TRUE); 453*fa94a07fSbrendan 45439c23413Seschrock /* 45539c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 45639c23413Seschrock * pool on the system. 45739c23413Seschrock */ 45839c23413Seschrock return (state == POOL_STATE_ACTIVE); 45939c23413Seschrock } 46039c23413Seschrock 46139c23413Seschrock /* 46239c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 46339c23413Seschrock * use, and writable. We put down an initial label which we will later 46439c23413Seschrock * overwrite with a complete label. Note that it's important to do this 46539c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 46639c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 46739c23413Seschrock * itself. 46839c23413Seschrock */ 46939c23413Seschrock int 47039c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 471fa9e4066Sahrens { 472fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 473fa9e4066Sahrens nvlist_t *label; 474fa9e4066Sahrens vdev_phys_t *vp; 475fa9e4066Sahrens vdev_boot_header_t *vb; 476ecc2d604Sbonwick uberblock_t *ub; 477fa9e4066Sahrens zio_t *zio; 478fa9e4066Sahrens int l, c, n; 479fa9e4066Sahrens char *buf; 480fa9e4066Sahrens size_t buflen; 481fa9e4066Sahrens int error; 482*fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 483fa9e4066Sahrens 4840373e76bSbonwick ASSERT(spa_config_held(spa, RW_WRITER)); 4850373e76bSbonwick 486fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 48739c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 48839c23413Seschrock crtxg, reason)) != 0) 489fa9e4066Sahrens return (error); 490fa9e4066Sahrens 491fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 492fa9e4066Sahrens return (0); 493fa9e4066Sahrens 494fa9e4066Sahrens /* 49539c23413Seschrock * Dead vdevs cannot be initialized. 496fa9e4066Sahrens */ 497fa9e4066Sahrens if (vdev_is_dead(vd)) 498fa9e4066Sahrens return (EIO); 499fa9e4066Sahrens 500fa9e4066Sahrens /* 50139c23413Seschrock * Determine if the vdev is in use. 502fa9e4066Sahrens */ 50339c23413Seschrock if (reason != VDEV_LABEL_REMOVE && 504*fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 50539c23413Seschrock return (EBUSY); 50639c23413Seschrock 50739c23413Seschrock ASSERT(reason != VDEV_LABEL_REMOVE || 508*fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, NULL, NULL)); 50939c23413Seschrock 51039c23413Seschrock /* 511*fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 512*fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 513*fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 514*fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 51539c23413Seschrock */ 516*fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 517*fa94a07fSbrendan spare_guid != 0ULL) { 51839c23413Seschrock vdev_t *pvd = vd->vdev_parent; 51939c23413Seschrock 52039c23413Seschrock for (; pvd != NULL; pvd = pvd->vdev_parent) { 52139c23413Seschrock pvd->vdev_guid_sum -= vd->vdev_guid; 52239c23413Seschrock pvd->vdev_guid_sum += spare_guid; 523fa9e4066Sahrens } 52499653d4eSeschrock 52539c23413Seschrock vd->vdev_guid = vd->vdev_guid_sum = spare_guid; 52639c23413Seschrock 52799653d4eSeschrock /* 52839c23413Seschrock * If this is a replacement, then we want to fallthrough to the 52939c23413Seschrock * rest of the code. If we're adding a spare, then it's already 5303d7072f8Seschrock * labeled appropriately and we can just return. 53199653d4eSeschrock */ 53239c23413Seschrock if (reason == VDEV_LABEL_SPARE) 53339c23413Seschrock return (0); 53439c23413Seschrock ASSERT(reason == VDEV_LABEL_REPLACE); 535fa9e4066Sahrens } 536fa9e4066Sahrens 537*fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 538*fa94a07fSbrendan l2cache_guid != 0ULL) { 539*fa94a07fSbrendan vdev_t *pvd = vd->vdev_parent; 540*fa94a07fSbrendan 541*fa94a07fSbrendan for (; pvd != NULL; pvd = pvd->vdev_parent) { 542*fa94a07fSbrendan pvd->vdev_guid_sum -= vd->vdev_guid; 543*fa94a07fSbrendan pvd->vdev_guid_sum += l2cache_guid; 544*fa94a07fSbrendan } 545*fa94a07fSbrendan 546*fa94a07fSbrendan vd->vdev_guid = vd->vdev_guid_sum = l2cache_guid; 547*fa94a07fSbrendan 548*fa94a07fSbrendan /* 549*fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 550*fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 551*fa94a07fSbrendan * already labeled appropriately and we can just return. 552*fa94a07fSbrendan */ 553*fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 554*fa94a07fSbrendan return (0); 555*fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 556*fa94a07fSbrendan } 557*fa94a07fSbrendan 558fa9e4066Sahrens /* 55939c23413Seschrock * Initialize its label. 560fa9e4066Sahrens */ 561fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 562fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 563fa9e4066Sahrens 564fa9e4066Sahrens /* 565fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 566fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 567fa9e4066Sahrens * really part of an active pool just yet. The labels will 568fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 569fa9e4066Sahrens */ 57039c23413Seschrock if (reason == VDEV_LABEL_SPARE || 57139c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 57239c23413Seschrock /* 57339c23413Seschrock * For inactive hot spares, we generate a special label that 57439c23413Seschrock * identifies as a mutually shared hot spare. We write the 57539c23413Seschrock * label if we are adding a hot spare, or if we are removing an 57639c23413Seschrock * active hot spare (in which case we want to revert the 57739c23413Seschrock * labels). 57839c23413Seschrock */ 57999653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 58099653d4eSeschrock 58199653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 58299653d4eSeschrock spa_version(spa)) == 0); 58399653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 58499653d4eSeschrock POOL_STATE_SPARE) == 0); 58599653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 58699653d4eSeschrock vd->vdev_guid) == 0); 587*fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 588*fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 589*fa94a07fSbrendan /* 590*fa94a07fSbrendan * For level 2 ARC devices, add a special label. 591*fa94a07fSbrendan */ 592*fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 593*fa94a07fSbrendan 594*fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 595*fa94a07fSbrendan spa_version(spa)) == 0); 596*fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 597*fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 598*fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 599*fa94a07fSbrendan vd->vdev_guid) == 0); 60099653d4eSeschrock } else { 60199653d4eSeschrock label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 60299653d4eSeschrock 60399653d4eSeschrock /* 60499653d4eSeschrock * Add our creation time. This allows us to detect multiple 60599653d4eSeschrock * vdev uses as described above, and automatically expires if we 60699653d4eSeschrock * fail. 60799653d4eSeschrock */ 60899653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 60999653d4eSeschrock crtxg) == 0); 61099653d4eSeschrock } 611fa9e4066Sahrens 612fa9e4066Sahrens buf = vp->vp_nvlist; 613fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 614fa9e4066Sahrens 615a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 616a75573b6Smmusante if (error != 0) { 617fa9e4066Sahrens nvlist_free(label); 618fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 619a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 620a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 621fa9e4066Sahrens } 622fa9e4066Sahrens 623fa9e4066Sahrens /* 624fa9e4066Sahrens * Initialize boot block header. 625fa9e4066Sahrens */ 626fa9e4066Sahrens vb = zio_buf_alloc(sizeof (vdev_boot_header_t)); 627fa9e4066Sahrens bzero(vb, sizeof (vdev_boot_header_t)); 628fa9e4066Sahrens vb->vb_magic = VDEV_BOOT_MAGIC; 629fa9e4066Sahrens vb->vb_version = VDEV_BOOT_VERSION; 630fa9e4066Sahrens vb->vb_offset = VDEV_BOOT_OFFSET; 631fa9e4066Sahrens vb->vb_size = VDEV_BOOT_SIZE; 632fa9e4066Sahrens 633fa9e4066Sahrens /* 634fa9e4066Sahrens * Initialize uberblock template. 635fa9e4066Sahrens */ 636ecc2d604Sbonwick ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 637ecc2d604Sbonwick bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 638ecc2d604Sbonwick *ub = spa->spa_uberblock; 639ecc2d604Sbonwick ub->ub_txg = 0; 640fa9e4066Sahrens 641fa9e4066Sahrens /* 642fa9e4066Sahrens * Write everything in parallel. 643fa9e4066Sahrens */ 644fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 645fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 646fa9e4066Sahrens 647fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 648fa9e4066Sahrens 649fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 650fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 651fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 652fa9e4066Sahrens 653fa9e4066Sahrens vdev_label_write(zio, vd, l, vb, 654fa9e4066Sahrens offsetof(vdev_label_t, vl_boot_header), 655fa9e4066Sahrens sizeof (vdev_boot_header_t), NULL, NULL); 656fa9e4066Sahrens 657ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 658ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 659ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 660ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), NULL, NULL); 661fa9e4066Sahrens } 662fa9e4066Sahrens } 663fa9e4066Sahrens 664fa9e4066Sahrens error = zio_wait(zio); 665fa9e4066Sahrens 666fa9e4066Sahrens nvlist_free(label); 667ecc2d604Sbonwick zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 668fa9e4066Sahrens zio_buf_free(vb, sizeof (vdev_boot_header_t)); 669fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 670fa9e4066Sahrens 67139c23413Seschrock /* 67239c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 6733d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 674*fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 675*fa94a07fSbrendan * level 2 ARC devices. 67639c23413Seschrock */ 67739c23413Seschrock if (error == 0 && !vd->vdev_isspare && 67839c23413Seschrock (reason == VDEV_LABEL_SPARE || 67939c23413Seschrock spa_spare_exists(vd->vdev_guid, NULL))) 68039c23413Seschrock spa_spare_add(vd); 68199653d4eSeschrock 682*fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 683*fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 684*fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 685*fa94a07fSbrendan spa_l2cache_add(vd); 686*fa94a07fSbrendan 68739c23413Seschrock return (error); 68899653d4eSeschrock } 68999653d4eSeschrock 690fa9e4066Sahrens /* 691fa9e4066Sahrens * ========================================================================== 692fa9e4066Sahrens * uberblock load/sync 693fa9e4066Sahrens * ========================================================================== 694fa9e4066Sahrens */ 695fa9e4066Sahrens 696fa9e4066Sahrens /* 697fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 698fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 699fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 700fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 701fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 702fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 703fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 704fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 705fa9e4066Sahrens */ 706fa9e4066Sahrens static int 707fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 708fa9e4066Sahrens { 709fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 710fa9e4066Sahrens return (-1); 711fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 712fa9e4066Sahrens return (1); 713fa9e4066Sahrens 714fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 715fa9e4066Sahrens return (-1); 716fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 717fa9e4066Sahrens return (1); 718fa9e4066Sahrens 719fa9e4066Sahrens return (0); 720fa9e4066Sahrens } 721fa9e4066Sahrens 722fa9e4066Sahrens static void 723fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 724fa9e4066Sahrens { 725ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 726fa9e4066Sahrens uberblock_t *ubbest = zio->io_private; 727fa9e4066Sahrens spa_t *spa = zio->io_spa; 728fa9e4066Sahrens 729ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 730fa9e4066Sahrens 731ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 732fa9e4066Sahrens mutex_enter(&spa->spa_uberblock_lock); 733fa9e4066Sahrens if (vdev_uberblock_compare(ub, ubbest) > 0) 734fa9e4066Sahrens *ubbest = *ub; 735fa9e4066Sahrens mutex_exit(&spa->spa_uberblock_lock); 736fa9e4066Sahrens } 737fa9e4066Sahrens 738fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 739fa9e4066Sahrens } 740fa9e4066Sahrens 741fa9e4066Sahrens void 742fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 743fa9e4066Sahrens { 744fa9e4066Sahrens int l, c, n; 745fa9e4066Sahrens 746fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 747fa9e4066Sahrens vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 748fa9e4066Sahrens 749fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 750fa9e4066Sahrens return; 751fa9e4066Sahrens 752fa9e4066Sahrens if (vdev_is_dead(vd)) 753fa9e4066Sahrens return; 754fa9e4066Sahrens 755fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 756ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 757fa9e4066Sahrens vdev_label_read(zio, vd, l, 758ecc2d604Sbonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 759ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 760ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 761fa9e4066Sahrens vdev_uberblock_load_done, ubbest); 762fa9e4066Sahrens } 763fa9e4066Sahrens } 764fa9e4066Sahrens } 765fa9e4066Sahrens 766fa9e4066Sahrens /* 767fa9e4066Sahrens * Write the uberblock to both labels of all leaves of the specified vdev. 7680373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 769fa9e4066Sahrens */ 770fa9e4066Sahrens static void 771fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 772fa9e4066Sahrens { 773fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 774fa9e4066Sahrens 7750373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 776fa9e4066Sahrens atomic_add_64(good_writes, 1); 777fa9e4066Sahrens } 778fa9e4066Sahrens 779fa9e4066Sahrens static void 780ecc2d604Sbonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, uint64_t txg) 781fa9e4066Sahrens { 782fa9e4066Sahrens int l, c, n; 783fa9e4066Sahrens 784fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 785ecc2d604Sbonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], txg); 786fa9e4066Sahrens 787fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 788fa9e4066Sahrens return; 789fa9e4066Sahrens 790fa9e4066Sahrens if (vdev_is_dead(vd)) 791fa9e4066Sahrens return; 792fa9e4066Sahrens 793ecc2d604Sbonwick n = txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 794fa9e4066Sahrens 795ecc2d604Sbonwick ASSERT(ub->ub_txg == txg); 796fa9e4066Sahrens 797fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) 798ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 799ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 800ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 801ecc2d604Sbonwick vdev_uberblock_sync_done, NULL); 802fa9e4066Sahrens 803fa9e4066Sahrens dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg); 804fa9e4066Sahrens } 805fa9e4066Sahrens 806fa9e4066Sahrens static int 807ecc2d604Sbonwick vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *vd, uint64_t txg) 808fa9e4066Sahrens { 809ecc2d604Sbonwick uberblock_t *ubbuf; 810ecc2d604Sbonwick size_t size = vd->vdev_top ? VDEV_UBERBLOCK_SIZE(vd) : SPA_MAXBLOCKSIZE; 811fa9e4066Sahrens uint64_t *good_writes; 812fa9e4066Sahrens zio_t *zio; 813fa9e4066Sahrens int error; 814fa9e4066Sahrens 815ecc2d604Sbonwick ubbuf = zio_buf_alloc(size); 816ecc2d604Sbonwick bzero(ubbuf, size); 817ecc2d604Sbonwick *ubbuf = *ub; 818fa9e4066Sahrens 819fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 820fa9e4066Sahrens 821fa9e4066Sahrens zio = zio_root(spa, NULL, good_writes, 822fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 823fa9e4066Sahrens 824ecc2d604Sbonwick vdev_uberblock_sync(zio, ubbuf, vd, txg); 825fa9e4066Sahrens 826fa9e4066Sahrens error = zio_wait(zio); 827fa9e4066Sahrens 828fa9e4066Sahrens if (error && *good_writes != 0) { 829fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 830fa9e4066Sahrens error = 0; 831fa9e4066Sahrens } 832fa9e4066Sahrens 833fa9e4066Sahrens /* 834fa9e4066Sahrens * It's possible to have no good writes and no error if every vdev is in 835fa9e4066Sahrens * the CANT_OPEN state. 836fa9e4066Sahrens */ 837fa9e4066Sahrens if (*good_writes == 0 && error == 0) 838fa9e4066Sahrens error = EIO; 839fa9e4066Sahrens 840fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 841ecc2d604Sbonwick zio_buf_free(ubbuf, size); 842fa9e4066Sahrens 843fa9e4066Sahrens return (error); 844fa9e4066Sahrens } 845fa9e4066Sahrens 846fa9e4066Sahrens /* 847fa9e4066Sahrens * Sync out an individual vdev. 848fa9e4066Sahrens */ 849fa9e4066Sahrens static void 850fa9e4066Sahrens vdev_sync_label_done(zio_t *zio) 851fa9e4066Sahrens { 852fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 853fa9e4066Sahrens 854fa9e4066Sahrens if (zio->io_error == 0) 855fa9e4066Sahrens atomic_add_64(good_writes, 1); 856fa9e4066Sahrens } 857fa9e4066Sahrens 858fa9e4066Sahrens static void 859fa9e4066Sahrens vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg) 860fa9e4066Sahrens { 861fa9e4066Sahrens nvlist_t *label; 862fa9e4066Sahrens vdev_phys_t *vp; 863fa9e4066Sahrens char *buf; 864fa9e4066Sahrens size_t buflen; 865fa9e4066Sahrens int c; 866fa9e4066Sahrens 867fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 868fa9e4066Sahrens vdev_sync_label(zio, vd->vdev_child[c], l, txg); 869fa9e4066Sahrens 870fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 871fa9e4066Sahrens return; 872fa9e4066Sahrens 873fa9e4066Sahrens if (vdev_is_dead(vd)) 874fa9e4066Sahrens return; 875fa9e4066Sahrens 876fa9e4066Sahrens /* 877fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 878fa9e4066Sahrens */ 8790373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 880fa9e4066Sahrens 881fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 882fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 883fa9e4066Sahrens 884fa9e4066Sahrens buf = vp->vp_nvlist; 885fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 886fa9e4066Sahrens 887ea8dc4b6Seschrock if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) 888fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 889fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 890fa9e4066Sahrens vdev_sync_label_done, NULL); 891fa9e4066Sahrens 892fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 893fa9e4066Sahrens nvlist_free(label); 894fa9e4066Sahrens 895fa9e4066Sahrens dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg); 896fa9e4066Sahrens } 897fa9e4066Sahrens 898fa9e4066Sahrens static int 899fa9e4066Sahrens vdev_sync_labels(vdev_t *vd, int l, uint64_t txg) 900fa9e4066Sahrens { 901fa9e4066Sahrens uint64_t *good_writes; 902fa9e4066Sahrens zio_t *zio; 903fa9e4066Sahrens int error; 904fa9e4066Sahrens 905fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 906fa9e4066Sahrens 907fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 908fa9e4066Sahrens 909fa9e4066Sahrens zio = zio_root(vd->vdev_spa, NULL, good_writes, 910fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 911fa9e4066Sahrens 912fa9e4066Sahrens /* 913fa9e4066Sahrens * Recursively kick off writes to all labels. 914fa9e4066Sahrens */ 915fa9e4066Sahrens vdev_sync_label(zio, vd, l, txg); 916fa9e4066Sahrens 917fa9e4066Sahrens error = zio_wait(zio); 918fa9e4066Sahrens 919fa9e4066Sahrens if (error && *good_writes != 0) { 920fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 921fa9e4066Sahrens error = 0; 922fa9e4066Sahrens } 923fa9e4066Sahrens 924fa9e4066Sahrens if (*good_writes == 0 && error == 0) 925fa9e4066Sahrens error = ENODEV; 926fa9e4066Sahrens 9278654d025Sperrin /* 9288654d025Sperrin * Failure to write a label can be fatal for a 9298654d025Sperrin * top level vdev. We don't want this for slogs 9308654d025Sperrin * as we use the main pool if they go away. 9318654d025Sperrin */ 9328654d025Sperrin if (vd->vdev_islog) 9338654d025Sperrin error = 0; 9348654d025Sperrin 935fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 936fa9e4066Sahrens 937fa9e4066Sahrens return (error); 938fa9e4066Sahrens } 939fa9e4066Sahrens 940fa9e4066Sahrens /* 941fa9e4066Sahrens * Sync the entire vdev configuration. 942fa9e4066Sahrens * 943fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 944fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 945fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 946fa9e4066Sahrens * describe the failure semantics at each stage. 947fa9e4066Sahrens * 948fa9e4066Sahrens * Moreover, it is designed to be idempotent: if spa_sync_labels() fails 949fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 950fa9e4066Sahrens */ 951fa9e4066Sahrens int 9520373e76bSbonwick vdev_config_sync(vdev_t *uvd, uint64_t txg) 953fa9e4066Sahrens { 9540373e76bSbonwick spa_t *spa = uvd->vdev_spa; 955fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 956fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 9570373e76bSbonwick vdev_t *vd; 958fa9e4066Sahrens zio_t *zio; 9590a4e9518Sgw int l, last_error = 0, error = 0; 9600a4e9518Sgw uint64_t good_writes = 0; 9610a4e9518Sgw boolean_t retry_avail = B_TRUE; 962fa9e4066Sahrens 963fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 964fa9e4066Sahrens 965fa9e4066Sahrens /* 966fa9e4066Sahrens * If this isn't a resync due to I/O errors, and nothing changed 967fa9e4066Sahrens * in this transaction group, and the vdev configuration hasn't changed, 9680373e76bSbonwick * then there's nothing to do. 969fa9e4066Sahrens */ 970fa9e4066Sahrens if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE && 971fa9e4066Sahrens list_is_empty(&spa->spa_dirty_list)) { 972fa9e4066Sahrens dprintf("nothing to sync in %s in txg %llu\n", 973fa9e4066Sahrens spa_name(spa), txg); 974fa9e4066Sahrens return (0); 975fa9e4066Sahrens } 976fa9e4066Sahrens 977fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 978fa9e4066Sahrens return (0); 979fa9e4066Sahrens 9800373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 9810373e76bSbonwick 982fa9e4066Sahrens dprintf("syncing %s txg %llu\n", spa_name(spa), txg); 983fa9e4066Sahrens 984fa9e4066Sahrens /* 985fa9e4066Sahrens * Flush the write cache of every disk that's been written to 986fa9e4066Sahrens * in this transaction group. This ensures that all blocks 987fa9e4066Sahrens * written in this txg will be committed to stable storage 988fa9e4066Sahrens * before any uberblock that references them. 989fa9e4066Sahrens */ 990fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 991fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 992fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 993fa9e4066Sahrens vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) { 994fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 995fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 996fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 997fa9e4066Sahrens } 998fa9e4066Sahrens (void) zio_wait(zio); 999fa9e4066Sahrens 10000a4e9518Sgw retry: 1001fa9e4066Sahrens /* 1002fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1003fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1004fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1005fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 1006fa9e4066Sahrens * which have not yet been touched, will still be valid. 1007fa9e4066Sahrens */ 1008fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1009fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1010fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 1011fa9e4066Sahrens if (l & 1) 1012fa9e4066Sahrens continue; 1013fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 10140a4e9518Sgw last_error = error; 10150a4e9518Sgw else 10160a4e9518Sgw good_writes++; 1017fa9e4066Sahrens } 1018fa9e4066Sahrens } 1019fa9e4066Sahrens 10200a4e9518Sgw /* 10210a4e9518Sgw * If all the vdevs that are currently dirty have failed or the 10220a4e9518Sgw * spa_dirty_list is empty then we dirty all the vdevs and try again. 10230a4e9518Sgw * This is a last ditch effort to ensure that we get at least one 10240a4e9518Sgw * update before proceeding to the uberblock. 10250a4e9518Sgw */ 10260a4e9518Sgw if (good_writes == 0 && retry_avail) { 10270a4e9518Sgw vdev_config_dirty(rvd); 10280a4e9518Sgw retry_avail = B_FALSE; 10290a4e9518Sgw last_error = 0; 10300a4e9518Sgw goto retry; 10310a4e9518Sgw } 10320a4e9518Sgw 10330a4e9518Sgw if (good_writes == 0) 10340a4e9518Sgw return (last_error); 10350a4e9518Sgw 1036fa9e4066Sahrens /* 1037fa9e4066Sahrens * Flush the new labels to disk. This ensures that all even-label 1038fa9e4066Sahrens * updates are committed to stable storage before the uberblock update. 1039fa9e4066Sahrens */ 1040fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 1041fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 1042fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1043fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1044fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 1045fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 1046fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 1047fa9e4066Sahrens } 1048fa9e4066Sahrens (void) zio_wait(zio); 1049fa9e4066Sahrens 1050fa9e4066Sahrens /* 10510373e76bSbonwick * Sync the uberblocks to all vdevs in the tree specified by uvd. 10520373e76bSbonwick * If the system dies in the middle of this step, there are two cases 10530373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1054fa9e4066Sahrens * 1055fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1056fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1057fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1058fa9e4066Sahrens * to that uberblock. 1059fa9e4066Sahrens * 1060fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1061fa9e4066Sahrens * will be the newest, and the even labels (which had all 1062fa9e4066Sahrens * been successfully committed) will be valid with respect 1063fa9e4066Sahrens * to the new uberblocks. 10640a4e9518Sgw * 10650a4e9518Sgw * NOTE: We retry to an uberblock update on the root if we were 10660a4e9518Sgw * failed our initial update attempt. 1067fa9e4066Sahrens */ 10680a4e9518Sgw error = vdev_uberblock_sync_tree(spa, ub, uvd, txg); 10690a4e9518Sgw if (error && uvd != rvd) 10700a4e9518Sgw error = vdev_uberblock_sync_tree(spa, ub, rvd, txg); 10710a4e9518Sgw 10720a4e9518Sgw if (error) 1073fa9e4066Sahrens return (error); 1074fa9e4066Sahrens 1075fa9e4066Sahrens /* 1076fa9e4066Sahrens * Flush the uberblocks to disk. This ensures that the odd labels 1077fa9e4066Sahrens * are no longer needed (because the new uberblocks and the even 1078fa9e4066Sahrens * labels are safely on disk), so it is safe to overwrite them. 1079fa9e4066Sahrens */ 1080fa9e4066Sahrens (void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE, 1081fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 1082fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 1083fa9e4066Sahrens 10840a4e9518Sgw last_error = 0; 1085fa9e4066Sahrens /* 1086fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1087fa9e4066Sahrens * in the middle of this process, the even labels and the new 1088fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1089fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1090fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 1091fa9e4066Sahrens * all labels will be brought up to date. 1092fa9e4066Sahrens */ 1093fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1094fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1095fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 1096fa9e4066Sahrens if ((l & 1) == 0) 1097fa9e4066Sahrens continue; 1098fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 10990a4e9518Sgw last_error = error; 11000a4e9518Sgw else 11010a4e9518Sgw good_writes++; 1102fa9e4066Sahrens } 1103fa9e4066Sahrens } 1104fa9e4066Sahrens 11050a4e9518Sgw if (good_writes == 0) 11060a4e9518Sgw return (last_error); 11070a4e9518Sgw 1108fa9e4066Sahrens /* 1109fa9e4066Sahrens * Flush the new labels to disk. This ensures that all odd-label 1110fa9e4066Sahrens * updates are committed to stable storage before the next 1111fa9e4066Sahrens * transaction group begins. 1112fa9e4066Sahrens */ 1113fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 1114fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 1115fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1116fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1117fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 1118fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 1119fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 1120fa9e4066Sahrens } 1121fa9e4066Sahrens (void) zio_wait(zio); 1122fa9e4066Sahrens 1123fa9e4066Sahrens return (0); 1124fa9e4066Sahrens } 1125