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 65*3d7072f8Seschrock * 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)); 156ecc2d604Sbonwick 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 * 19099653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 19199653d4eSeschrock boolean_t isspare) 192fa9e4066Sahrens { 193fa9e4066Sahrens nvlist_t *nv = NULL; 194fa9e4066Sahrens 195ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 196fa9e4066Sahrens 197fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 198fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 19999653d4eSeschrock if (!isspare) 20099653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 20199653d4eSeschrock == 0); 202fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 203fa9e4066Sahrens 204fa9e4066Sahrens if (vd->vdev_path != NULL) 205fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 206fa9e4066Sahrens vd->vdev_path) == 0); 207fa9e4066Sahrens 208fa9e4066Sahrens if (vd->vdev_devid != NULL) 209fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 210fa9e4066Sahrens vd->vdev_devid) == 0); 211fa9e4066Sahrens 212*3d7072f8Seschrock if (vd->vdev_physpath != NULL) 213*3d7072f8Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 214*3d7072f8Seschrock vd->vdev_physpath) == 0); 215*3d7072f8Seschrock 21699653d4eSeschrock if (vd->vdev_nparity != 0) { 21799653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 21899653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 21999653d4eSeschrock 22099653d4eSeschrock /* 22199653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 22299653d4eSeschrock * into a crufty old storage pool. 22399653d4eSeschrock */ 22499653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 22599653d4eSeschrock (vd->vdev_nparity == 2 && 22699653d4eSeschrock spa_version(spa) >= ZFS_VERSION_RAID6)); 22799653d4eSeschrock 22899653d4eSeschrock /* 22999653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 23099653d4eSeschrock * that only support a single parity device -- older software 23199653d4eSeschrock * will just ignore it. 23299653d4eSeschrock */ 23399653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 23499653d4eSeschrock vd->vdev_nparity) == 0); 23599653d4eSeschrock } 23699653d4eSeschrock 237afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 238afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 239afefbcddSeschrock vd->vdev_wholedisk) == 0); 240afefbcddSeschrock 241ea8dc4b6Seschrock if (vd->vdev_not_present) 242ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 243ea8dc4b6Seschrock 24499653d4eSeschrock if (vd->vdev_isspare) 24599653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 24699653d4eSeschrock 24799653d4eSeschrock if (!isspare && vd == vd->vdev_top) { 248fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 249fa9e4066Sahrens vd->vdev_ms_array) == 0); 250fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 251fa9e4066Sahrens vd->vdev_ms_shift) == 0); 252fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 253fa9e4066Sahrens vd->vdev_ashift) == 0); 254fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 255fa9e4066Sahrens vd->vdev_asize) == 0); 256fa9e4066Sahrens } 257fa9e4066Sahrens 258fa9e4066Sahrens if (vd->vdev_dtl.smo_object != 0) 259fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 260fa9e4066Sahrens vd->vdev_dtl.smo_object) == 0); 261fa9e4066Sahrens 262fa9e4066Sahrens if (getstats) { 263fa9e4066Sahrens vdev_stat_t vs; 264fa9e4066Sahrens vdev_get_stats(vd, &vs); 265fa9e4066Sahrens VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 266fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 267fa9e4066Sahrens } 268fa9e4066Sahrens 269fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 270fa9e4066Sahrens nvlist_t **child; 271fa9e4066Sahrens int c; 272fa9e4066Sahrens 273fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 274fa9e4066Sahrens KM_SLEEP); 275fa9e4066Sahrens 276fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 27799653d4eSeschrock child[c] = vdev_config_generate(spa, vd->vdev_child[c], 27899653d4eSeschrock getstats, isspare); 279fa9e4066Sahrens 280fa9e4066Sahrens VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 281fa9e4066Sahrens child, vd->vdev_children) == 0); 282fa9e4066Sahrens 283fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 284fa9e4066Sahrens nvlist_free(child[c]); 285fa9e4066Sahrens 286fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 287441d80aaSlling 288441d80aaSlling } else { 289ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 290441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 291ecc2d604Sbonwick B_TRUE) == 0); 292*3d7072f8Seschrock if (vd->vdev_faulted) 293*3d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 294*3d7072f8Seschrock B_TRUE) == 0); 295*3d7072f8Seschrock if (vd->vdev_degraded) 296*3d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 297*3d7072f8Seschrock B_TRUE) == 0); 298*3d7072f8Seschrock if (vd->vdev_removed) 299*3d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 300*3d7072f8Seschrock B_TRUE) == 0); 301*3d7072f8Seschrock if (vd->vdev_unspare) 302*3d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 303*3d7072f8Seschrock B_TRUE) == 0); 304fa9e4066Sahrens } 305fa9e4066Sahrens 306fa9e4066Sahrens return (nv); 307fa9e4066Sahrens } 308fa9e4066Sahrens 309fa9e4066Sahrens nvlist_t * 310fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 311fa9e4066Sahrens { 3120373e76bSbonwick spa_t *spa = vd->vdev_spa; 313fa9e4066Sahrens nvlist_t *config = NULL; 314fa9e4066Sahrens vdev_phys_t *vp; 315fa9e4066Sahrens zio_t *zio; 316fa9e4066Sahrens int l; 317fa9e4066Sahrens 3180373e76bSbonwick ASSERT(spa_config_held(spa, RW_READER)); 3190373e76bSbonwick 320fa9e4066Sahrens if (vdev_is_dead(vd)) 321fa9e4066Sahrens return (NULL); 322fa9e4066Sahrens 323fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 324fa9e4066Sahrens 325fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 326fa9e4066Sahrens 3270373e76bSbonwick zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | 328ea8dc4b6Seschrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD); 329fa9e4066Sahrens 330fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 331fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 332fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 333fa9e4066Sahrens 334fa9e4066Sahrens if (zio_wait(zio) == 0 && 335fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 336ea8dc4b6Seschrock &config, 0) == 0) 337fa9e4066Sahrens break; 338fa9e4066Sahrens 339fa9e4066Sahrens if (config != NULL) { 340fa9e4066Sahrens nvlist_free(config); 341fa9e4066Sahrens config = NULL; 342fa9e4066Sahrens } 343fa9e4066Sahrens } 344fa9e4066Sahrens 345fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 346fa9e4066Sahrens 347fa9e4066Sahrens return (config); 348fa9e4066Sahrens } 349fa9e4066Sahrens 35039c23413Seschrock /* 35139c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 35239c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 35339c23413Seschrock */ 35439c23413Seschrock static boolean_t 35539c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 35639c23413Seschrock uint64_t *spare_guid) 35739c23413Seschrock { 35839c23413Seschrock spa_t *spa = vd->vdev_spa; 35939c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 36039c23413Seschrock uint64_t vdtxg = 0; 36139c23413Seschrock nvlist_t *label; 36239c23413Seschrock 36339c23413Seschrock if (spare_guid) 36439c23413Seschrock *spare_guid = 0ULL; 36539c23413Seschrock 36639c23413Seschrock /* 36739c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 36839c23413Seschrock */ 36939c23413Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 37039c23413Seschrock return (B_FALSE); 37139c23413Seschrock 37239c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 37339c23413Seschrock &vdtxg); 37439c23413Seschrock 37539c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 37639c23413Seschrock &state) != 0 || 37739c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 37839c23413Seschrock &device_guid) != 0) { 37939c23413Seschrock nvlist_free(label); 38039c23413Seschrock return (B_FALSE); 38139c23413Seschrock } 38239c23413Seschrock 38339c23413Seschrock if (state != POOL_STATE_SPARE && 38439c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 38539c23413Seschrock &pool_guid) != 0 || 38639c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 38739c23413Seschrock &txg) != 0)) { 38839c23413Seschrock nvlist_free(label); 38939c23413Seschrock return (B_FALSE); 39039c23413Seschrock } 39139c23413Seschrock 39239c23413Seschrock nvlist_free(label); 39339c23413Seschrock 39439c23413Seschrock /* 39539c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 39639c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 39739c23413Seschrock * spare (which we check for later on). 39839c23413Seschrock */ 39939c23413Seschrock if (state != POOL_STATE_SPARE && 40039c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 40139c23413Seschrock !spa_spare_exists(device_guid, NULL)) 40239c23413Seschrock return (B_FALSE); 40339c23413Seschrock 40439c23413Seschrock /* 40539c23413Seschrock * If the transaction group is zero, then this an initialized (but 40639c23413Seschrock * unused) label. This is only an error if the create transaction 40739c23413Seschrock * on-disk is the same as the one we're using now, in which case the 40839c23413Seschrock * user has attempted to add the same vdev multiple times in the same 40939c23413Seschrock * transaction. 41039c23413Seschrock */ 41139c23413Seschrock if (state != POOL_STATE_SPARE && txg == 0 && vdtxg == crtxg) 41239c23413Seschrock return (B_TRUE); 41339c23413Seschrock 41439c23413Seschrock /* 41539c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 41639c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 41739c23413Seschrock * to add. 41839c23413Seschrock */ 41939c23413Seschrock if (spa_spare_exists(device_guid, &spare_pool) || 42039c23413Seschrock spa_has_spare(spa, device_guid)) { 42139c23413Seschrock if (spare_guid) 42239c23413Seschrock *spare_guid = device_guid; 42339c23413Seschrock 42439c23413Seschrock switch (reason) { 42539c23413Seschrock case VDEV_LABEL_CREATE: 42639c23413Seschrock return (B_TRUE); 42739c23413Seschrock 42839c23413Seschrock case VDEV_LABEL_REPLACE: 42939c23413Seschrock return (!spa_has_spare(spa, device_guid) || 43039c23413Seschrock spare_pool != 0ULL); 43139c23413Seschrock 43239c23413Seschrock case VDEV_LABEL_SPARE: 43339c23413Seschrock return (spa_has_spare(spa, device_guid)); 43439c23413Seschrock } 43539c23413Seschrock } 43639c23413Seschrock 43739c23413Seschrock /* 43839c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 43939c23413Seschrock * pool on the system. 44039c23413Seschrock */ 44139c23413Seschrock return (state == POOL_STATE_ACTIVE); 44239c23413Seschrock } 44339c23413Seschrock 44439c23413Seschrock /* 44539c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 44639c23413Seschrock * use, and writable. We put down an initial label which we will later 44739c23413Seschrock * overwrite with a complete label. Note that it's important to do this 44839c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 44939c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 45039c23413Seschrock * itself. 45139c23413Seschrock */ 45239c23413Seschrock int 45339c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 454fa9e4066Sahrens { 455fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 456fa9e4066Sahrens nvlist_t *label; 457fa9e4066Sahrens vdev_phys_t *vp; 458fa9e4066Sahrens vdev_boot_header_t *vb; 459ecc2d604Sbonwick uberblock_t *ub; 460fa9e4066Sahrens zio_t *zio; 461fa9e4066Sahrens int l, c, n; 462fa9e4066Sahrens char *buf; 463fa9e4066Sahrens size_t buflen; 464fa9e4066Sahrens int error; 46539c23413Seschrock uint64_t spare_guid; 466fa9e4066Sahrens 4670373e76bSbonwick ASSERT(spa_config_held(spa, RW_WRITER)); 4680373e76bSbonwick 469fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 47039c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 47139c23413Seschrock crtxg, reason)) != 0) 472fa9e4066Sahrens return (error); 473fa9e4066Sahrens 474fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 475fa9e4066Sahrens return (0); 476fa9e4066Sahrens 477fa9e4066Sahrens /* 47839c23413Seschrock * Dead vdevs cannot be initialized. 479fa9e4066Sahrens */ 480fa9e4066Sahrens if (vdev_is_dead(vd)) 481fa9e4066Sahrens return (EIO); 482fa9e4066Sahrens 483fa9e4066Sahrens /* 48439c23413Seschrock * Determine if the vdev is in use. 485fa9e4066Sahrens */ 48639c23413Seschrock if (reason != VDEV_LABEL_REMOVE && 48739c23413Seschrock vdev_inuse(vd, crtxg, reason, &spare_guid)) 48839c23413Seschrock return (EBUSY); 48939c23413Seschrock 49039c23413Seschrock ASSERT(reason != VDEV_LABEL_REMOVE || 49139c23413Seschrock vdev_inuse(vd, crtxg, reason, NULL)); 49239c23413Seschrock 49339c23413Seschrock /* 49439c23413Seschrock * If this is a request to add or replace a spare that is in use 49539c23413Seschrock * elsewhere on the system, then we must update the guid (which was 49639c23413Seschrock * initialized to a random value) to reflect the actual GUID (which is 49739c23413Seschrock * shared between multiple pools). 49839c23413Seschrock */ 49939c23413Seschrock if (reason != VDEV_LABEL_REMOVE && spare_guid != 0ULL) { 50039c23413Seschrock vdev_t *pvd = vd->vdev_parent; 50139c23413Seschrock 50239c23413Seschrock for (; pvd != NULL; pvd = pvd->vdev_parent) { 50339c23413Seschrock pvd->vdev_guid_sum -= vd->vdev_guid; 50439c23413Seschrock pvd->vdev_guid_sum += spare_guid; 505fa9e4066Sahrens } 50699653d4eSeschrock 50739c23413Seschrock vd->vdev_guid = vd->vdev_guid_sum = spare_guid; 50839c23413Seschrock 50999653d4eSeschrock /* 51039c23413Seschrock * If this is a replacement, then we want to fallthrough to the 51139c23413Seschrock * rest of the code. If we're adding a spare, then it's already 512*3d7072f8Seschrock * labeled appropriately and we can just return. 51399653d4eSeschrock */ 51439c23413Seschrock if (reason == VDEV_LABEL_SPARE) 51539c23413Seschrock return (0); 51639c23413Seschrock ASSERT(reason == VDEV_LABEL_REPLACE); 517fa9e4066Sahrens } 518fa9e4066Sahrens 519fa9e4066Sahrens /* 52039c23413Seschrock * Initialize its label. 521fa9e4066Sahrens */ 522fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 523fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 524fa9e4066Sahrens 525fa9e4066Sahrens /* 526fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 527fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 528fa9e4066Sahrens * really part of an active pool just yet. The labels will 529fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 530fa9e4066Sahrens */ 53139c23413Seschrock if (reason == VDEV_LABEL_SPARE || 53239c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 53339c23413Seschrock /* 53439c23413Seschrock * For inactive hot spares, we generate a special label that 53539c23413Seschrock * identifies as a mutually shared hot spare. We write the 53639c23413Seschrock * label if we are adding a hot spare, or if we are removing an 53739c23413Seschrock * active hot spare (in which case we want to revert the 53839c23413Seschrock * labels). 53939c23413Seschrock */ 54099653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 54199653d4eSeschrock 54299653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 54399653d4eSeschrock spa_version(spa)) == 0); 54499653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 54599653d4eSeschrock POOL_STATE_SPARE) == 0); 54699653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 54799653d4eSeschrock vd->vdev_guid) == 0); 54899653d4eSeschrock } else { 54999653d4eSeschrock label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 55099653d4eSeschrock 55199653d4eSeschrock /* 55299653d4eSeschrock * Add our creation time. This allows us to detect multiple 55399653d4eSeschrock * vdev uses as described above, and automatically expires if we 55499653d4eSeschrock * fail. 55599653d4eSeschrock */ 55699653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 55799653d4eSeschrock crtxg) == 0); 55899653d4eSeschrock } 559fa9e4066Sahrens 560fa9e4066Sahrens buf = vp->vp_nvlist; 561fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 562fa9e4066Sahrens 563a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 564a75573b6Smmusante if (error != 0) { 565fa9e4066Sahrens nvlist_free(label); 566fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 567a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 568a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 569fa9e4066Sahrens } 570fa9e4066Sahrens 571fa9e4066Sahrens /* 572fa9e4066Sahrens * Initialize boot block header. 573fa9e4066Sahrens */ 574fa9e4066Sahrens vb = zio_buf_alloc(sizeof (vdev_boot_header_t)); 575fa9e4066Sahrens bzero(vb, sizeof (vdev_boot_header_t)); 576fa9e4066Sahrens vb->vb_magic = VDEV_BOOT_MAGIC; 577fa9e4066Sahrens vb->vb_version = VDEV_BOOT_VERSION; 578fa9e4066Sahrens vb->vb_offset = VDEV_BOOT_OFFSET; 579fa9e4066Sahrens vb->vb_size = VDEV_BOOT_SIZE; 580fa9e4066Sahrens 581fa9e4066Sahrens /* 582fa9e4066Sahrens * Initialize uberblock template. 583fa9e4066Sahrens */ 584ecc2d604Sbonwick ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 585ecc2d604Sbonwick bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 586ecc2d604Sbonwick *ub = spa->spa_uberblock; 587ecc2d604Sbonwick ub->ub_txg = 0; 588fa9e4066Sahrens 589fa9e4066Sahrens /* 590fa9e4066Sahrens * Write everything in parallel. 591fa9e4066Sahrens */ 592fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 593fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 594fa9e4066Sahrens 595fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 596fa9e4066Sahrens 597fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 598fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 599fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 600fa9e4066Sahrens 601fa9e4066Sahrens vdev_label_write(zio, vd, l, vb, 602fa9e4066Sahrens offsetof(vdev_label_t, vl_boot_header), 603fa9e4066Sahrens sizeof (vdev_boot_header_t), NULL, NULL); 604fa9e4066Sahrens 605ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 606ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 607ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 608ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), NULL, NULL); 609fa9e4066Sahrens } 610fa9e4066Sahrens } 611fa9e4066Sahrens 612fa9e4066Sahrens error = zio_wait(zio); 613fa9e4066Sahrens 614fa9e4066Sahrens nvlist_free(label); 615ecc2d604Sbonwick zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 616fa9e4066Sahrens zio_buf_free(vb, sizeof (vdev_boot_header_t)); 617fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 618fa9e4066Sahrens 61939c23413Seschrock /* 62039c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 621*3d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 62239c23413Seschrock * exists as a spare elsewhere in the system. 62339c23413Seschrock */ 62439c23413Seschrock if (error == 0 && !vd->vdev_isspare && 62539c23413Seschrock (reason == VDEV_LABEL_SPARE || 62639c23413Seschrock spa_spare_exists(vd->vdev_guid, NULL))) 62739c23413Seschrock spa_spare_add(vd); 62899653d4eSeschrock 62939c23413Seschrock return (error); 63099653d4eSeschrock } 63199653d4eSeschrock 632fa9e4066Sahrens /* 633fa9e4066Sahrens * ========================================================================== 634fa9e4066Sahrens * uberblock load/sync 635fa9e4066Sahrens * ========================================================================== 636fa9e4066Sahrens */ 637fa9e4066Sahrens 638fa9e4066Sahrens /* 639fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 640fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 641fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 642fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 643fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 644fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 645fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 646fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 647fa9e4066Sahrens */ 648fa9e4066Sahrens static int 649fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 650fa9e4066Sahrens { 651fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 652fa9e4066Sahrens return (-1); 653fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 654fa9e4066Sahrens return (1); 655fa9e4066Sahrens 656fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 657fa9e4066Sahrens return (-1); 658fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 659fa9e4066Sahrens return (1); 660fa9e4066Sahrens 661fa9e4066Sahrens return (0); 662fa9e4066Sahrens } 663fa9e4066Sahrens 664fa9e4066Sahrens static void 665fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 666fa9e4066Sahrens { 667ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 668fa9e4066Sahrens uberblock_t *ubbest = zio->io_private; 669fa9e4066Sahrens spa_t *spa = zio->io_spa; 670fa9e4066Sahrens 671ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 672fa9e4066Sahrens 673ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 674fa9e4066Sahrens mutex_enter(&spa->spa_uberblock_lock); 675fa9e4066Sahrens if (vdev_uberblock_compare(ub, ubbest) > 0) 676fa9e4066Sahrens *ubbest = *ub; 677fa9e4066Sahrens mutex_exit(&spa->spa_uberblock_lock); 678fa9e4066Sahrens } 679fa9e4066Sahrens 680fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 681fa9e4066Sahrens } 682fa9e4066Sahrens 683fa9e4066Sahrens void 684fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 685fa9e4066Sahrens { 686fa9e4066Sahrens int l, c, n; 687fa9e4066Sahrens 688fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 689fa9e4066Sahrens vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 690fa9e4066Sahrens 691fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 692fa9e4066Sahrens return; 693fa9e4066Sahrens 694fa9e4066Sahrens if (vdev_is_dead(vd)) 695fa9e4066Sahrens return; 696fa9e4066Sahrens 697fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 698ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 699fa9e4066Sahrens vdev_label_read(zio, vd, l, 700ecc2d604Sbonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 701ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 702ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 703fa9e4066Sahrens vdev_uberblock_load_done, ubbest); 704fa9e4066Sahrens } 705fa9e4066Sahrens } 706fa9e4066Sahrens } 707fa9e4066Sahrens 708fa9e4066Sahrens /* 709fa9e4066Sahrens * Write the uberblock to both labels of all leaves of the specified vdev. 7100373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 711fa9e4066Sahrens */ 712fa9e4066Sahrens static void 713fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 714fa9e4066Sahrens { 715fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 716fa9e4066Sahrens 7170373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 718fa9e4066Sahrens atomic_add_64(good_writes, 1); 719fa9e4066Sahrens } 720fa9e4066Sahrens 721fa9e4066Sahrens static void 722ecc2d604Sbonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, uint64_t txg) 723fa9e4066Sahrens { 724fa9e4066Sahrens int l, c, n; 725fa9e4066Sahrens 726fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 727ecc2d604Sbonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c], txg); 728fa9e4066Sahrens 729fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 730fa9e4066Sahrens return; 731fa9e4066Sahrens 732fa9e4066Sahrens if (vdev_is_dead(vd)) 733fa9e4066Sahrens return; 734fa9e4066Sahrens 735ecc2d604Sbonwick n = txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 736fa9e4066Sahrens 737ecc2d604Sbonwick ASSERT(ub->ub_txg == txg); 738fa9e4066Sahrens 739fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) 740ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 741ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 742ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 743ecc2d604Sbonwick vdev_uberblock_sync_done, NULL); 744fa9e4066Sahrens 745fa9e4066Sahrens dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg); 746fa9e4066Sahrens } 747fa9e4066Sahrens 748fa9e4066Sahrens static int 749ecc2d604Sbonwick vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *vd, uint64_t txg) 750fa9e4066Sahrens { 751ecc2d604Sbonwick uberblock_t *ubbuf; 752ecc2d604Sbonwick size_t size = vd->vdev_top ? VDEV_UBERBLOCK_SIZE(vd) : SPA_MAXBLOCKSIZE; 753fa9e4066Sahrens uint64_t *good_writes; 754fa9e4066Sahrens zio_t *zio; 755fa9e4066Sahrens int error; 756fa9e4066Sahrens 757ecc2d604Sbonwick ubbuf = zio_buf_alloc(size); 758ecc2d604Sbonwick bzero(ubbuf, size); 759ecc2d604Sbonwick *ubbuf = *ub; 760fa9e4066Sahrens 761fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 762fa9e4066Sahrens 763fa9e4066Sahrens zio = zio_root(spa, NULL, good_writes, 764fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 765fa9e4066Sahrens 766ecc2d604Sbonwick vdev_uberblock_sync(zio, ubbuf, vd, txg); 767fa9e4066Sahrens 768fa9e4066Sahrens error = zio_wait(zio); 769fa9e4066Sahrens 770fa9e4066Sahrens if (error && *good_writes != 0) { 771fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 772fa9e4066Sahrens error = 0; 773fa9e4066Sahrens } 774fa9e4066Sahrens 775fa9e4066Sahrens /* 776fa9e4066Sahrens * It's possible to have no good writes and no error if every vdev is in 777fa9e4066Sahrens * the CANT_OPEN state. 778fa9e4066Sahrens */ 779fa9e4066Sahrens if (*good_writes == 0 && error == 0) 780fa9e4066Sahrens error = EIO; 781fa9e4066Sahrens 782fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 783ecc2d604Sbonwick zio_buf_free(ubbuf, size); 784fa9e4066Sahrens 785fa9e4066Sahrens return (error); 786fa9e4066Sahrens } 787fa9e4066Sahrens 788fa9e4066Sahrens /* 789fa9e4066Sahrens * Sync out an individual vdev. 790fa9e4066Sahrens */ 791fa9e4066Sahrens static void 792fa9e4066Sahrens vdev_sync_label_done(zio_t *zio) 793fa9e4066Sahrens { 794fa9e4066Sahrens uint64_t *good_writes = zio->io_root->io_private; 795fa9e4066Sahrens 796fa9e4066Sahrens if (zio->io_error == 0) 797fa9e4066Sahrens atomic_add_64(good_writes, 1); 798fa9e4066Sahrens } 799fa9e4066Sahrens 800fa9e4066Sahrens static void 801fa9e4066Sahrens vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg) 802fa9e4066Sahrens { 803fa9e4066Sahrens nvlist_t *label; 804fa9e4066Sahrens vdev_phys_t *vp; 805fa9e4066Sahrens char *buf; 806fa9e4066Sahrens size_t buflen; 807fa9e4066Sahrens int c; 808fa9e4066Sahrens 809fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 810fa9e4066Sahrens vdev_sync_label(zio, vd->vdev_child[c], l, txg); 811fa9e4066Sahrens 812fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 813fa9e4066Sahrens return; 814fa9e4066Sahrens 815fa9e4066Sahrens if (vdev_is_dead(vd)) 816fa9e4066Sahrens return; 817fa9e4066Sahrens 818fa9e4066Sahrens /* 819fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 820fa9e4066Sahrens */ 8210373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 822fa9e4066Sahrens 823fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 824fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 825fa9e4066Sahrens 826fa9e4066Sahrens buf = vp->vp_nvlist; 827fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 828fa9e4066Sahrens 829ea8dc4b6Seschrock if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) 830fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 831fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), 832fa9e4066Sahrens vdev_sync_label_done, NULL); 833fa9e4066Sahrens 834fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 835fa9e4066Sahrens nvlist_free(label); 836fa9e4066Sahrens 837fa9e4066Sahrens dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg); 838fa9e4066Sahrens } 839fa9e4066Sahrens 840fa9e4066Sahrens static int 841fa9e4066Sahrens vdev_sync_labels(vdev_t *vd, int l, uint64_t txg) 842fa9e4066Sahrens { 843fa9e4066Sahrens uint64_t *good_writes; 844fa9e4066Sahrens zio_t *zio; 845fa9e4066Sahrens int error; 846fa9e4066Sahrens 847fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 848fa9e4066Sahrens 849fa9e4066Sahrens good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); 850fa9e4066Sahrens 851fa9e4066Sahrens zio = zio_root(vd->vdev_spa, NULL, good_writes, 852fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 853fa9e4066Sahrens 854fa9e4066Sahrens /* 855fa9e4066Sahrens * Recursively kick off writes to all labels. 856fa9e4066Sahrens */ 857fa9e4066Sahrens vdev_sync_label(zio, vd, l, txg); 858fa9e4066Sahrens 859fa9e4066Sahrens error = zio_wait(zio); 860fa9e4066Sahrens 861fa9e4066Sahrens if (error && *good_writes != 0) { 862fa9e4066Sahrens dprintf("partial success: good_writes = %llu\n", *good_writes); 863fa9e4066Sahrens error = 0; 864fa9e4066Sahrens } 865fa9e4066Sahrens 866fa9e4066Sahrens if (*good_writes == 0 && error == 0) 867fa9e4066Sahrens error = ENODEV; 868fa9e4066Sahrens 869fa9e4066Sahrens kmem_free(good_writes, sizeof (uint64_t)); 870fa9e4066Sahrens 871fa9e4066Sahrens return (error); 872fa9e4066Sahrens } 873fa9e4066Sahrens 874fa9e4066Sahrens /* 875fa9e4066Sahrens * Sync the entire vdev configuration. 876fa9e4066Sahrens * 877fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 878fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 879fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 880fa9e4066Sahrens * describe the failure semantics at each stage. 881fa9e4066Sahrens * 882fa9e4066Sahrens * Moreover, it is designed to be idempotent: if spa_sync_labels() fails 883fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 884fa9e4066Sahrens */ 885fa9e4066Sahrens int 8860373e76bSbonwick vdev_config_sync(vdev_t *uvd, uint64_t txg) 887fa9e4066Sahrens { 8880373e76bSbonwick spa_t *spa = uvd->vdev_spa; 889fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 890fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 8910373e76bSbonwick vdev_t *vd; 892fa9e4066Sahrens zio_t *zio; 893f65ea9b9Sbonwick int l, error; 894fa9e4066Sahrens 895fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 896fa9e4066Sahrens 897fa9e4066Sahrens /* 898fa9e4066Sahrens * If this isn't a resync due to I/O errors, and nothing changed 899fa9e4066Sahrens * in this transaction group, and the vdev configuration hasn't changed, 9000373e76bSbonwick * then there's nothing to do. 901fa9e4066Sahrens */ 902fa9e4066Sahrens if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE && 903fa9e4066Sahrens list_is_empty(&spa->spa_dirty_list)) { 904fa9e4066Sahrens dprintf("nothing to sync in %s in txg %llu\n", 905fa9e4066Sahrens spa_name(spa), txg); 906fa9e4066Sahrens return (0); 907fa9e4066Sahrens } 908fa9e4066Sahrens 909fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 910fa9e4066Sahrens return (0); 911fa9e4066Sahrens 9120373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 9130373e76bSbonwick 914fa9e4066Sahrens dprintf("syncing %s txg %llu\n", spa_name(spa), txg); 915fa9e4066Sahrens 916fa9e4066Sahrens /* 917fa9e4066Sahrens * Flush the write cache of every disk that's been written to 918fa9e4066Sahrens * in this transaction group. This ensures that all blocks 919fa9e4066Sahrens * written in this txg will be committed to stable storage 920fa9e4066Sahrens * before any uberblock that references them. 921fa9e4066Sahrens */ 922fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 923fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 924fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 925fa9e4066Sahrens vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) { 926fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 927fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 928fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 929fa9e4066Sahrens } 930fa9e4066Sahrens (void) zio_wait(zio); 931fa9e4066Sahrens 932fa9e4066Sahrens /* 933fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 934fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 935fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 936fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 937fa9e4066Sahrens * which have not yet been touched, will still be valid. 938fa9e4066Sahrens */ 939fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 940fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 941fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 942fa9e4066Sahrens if (l & 1) 943fa9e4066Sahrens continue; 944fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 945fa9e4066Sahrens return (error); 946fa9e4066Sahrens } 947fa9e4066Sahrens } 948fa9e4066Sahrens 949fa9e4066Sahrens /* 950fa9e4066Sahrens * Flush the new labels to disk. This ensures that all even-label 951fa9e4066Sahrens * updates are committed to stable storage before the uberblock update. 952fa9e4066Sahrens */ 953fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 954fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 955fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 956fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 957fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 958fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 959fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 960fa9e4066Sahrens } 961fa9e4066Sahrens (void) zio_wait(zio); 962fa9e4066Sahrens 963fa9e4066Sahrens /* 9640373e76bSbonwick * Sync the uberblocks to all vdevs in the tree specified by uvd. 9650373e76bSbonwick * If the system dies in the middle of this step, there are two cases 9660373e76bSbonwick * to consider, and the on-disk state is consistent either way: 967fa9e4066Sahrens * 968fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 969fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 970fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 971fa9e4066Sahrens * to that uberblock. 972fa9e4066Sahrens * 973fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 974fa9e4066Sahrens * will be the newest, and the even labels (which had all 975fa9e4066Sahrens * been successfully committed) will be valid with respect 976fa9e4066Sahrens * to the new uberblocks. 977fa9e4066Sahrens */ 978fa9e4066Sahrens if ((error = vdev_uberblock_sync_tree(spa, ub, uvd, txg)) != 0) 979fa9e4066Sahrens return (error); 980fa9e4066Sahrens 981fa9e4066Sahrens /* 982fa9e4066Sahrens * Flush the uberblocks to disk. This ensures that the odd labels 983fa9e4066Sahrens * are no longer needed (because the new uberblocks and the even 984fa9e4066Sahrens * labels are safely on disk), so it is safe to overwrite them. 985fa9e4066Sahrens */ 986fa9e4066Sahrens (void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE, 987fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 988fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 989fa9e4066Sahrens 990fa9e4066Sahrens /* 991fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 992fa9e4066Sahrens * in the middle of this process, the even labels and the new 993fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 994fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 995fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 996fa9e4066Sahrens * all labels will be brought up to date. 997fa9e4066Sahrens */ 998fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 999fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1000fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 1001fa9e4066Sahrens if ((l & 1) == 0) 1002fa9e4066Sahrens continue; 1003fa9e4066Sahrens if ((error = vdev_sync_labels(vd, l, txg)) != 0) 1004fa9e4066Sahrens return (error); 1005fa9e4066Sahrens } 1006fa9e4066Sahrens } 1007fa9e4066Sahrens 1008fa9e4066Sahrens /* 1009fa9e4066Sahrens * Flush the new labels to disk. This ensures that all odd-label 1010fa9e4066Sahrens * updates are committed to stable storage before the next 1011fa9e4066Sahrens * transaction group begins. 1012fa9e4066Sahrens */ 1013fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 1014fa9e4066Sahrens ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL); 1015fa9e4066Sahrens for (vd = list_head(&spa->spa_dirty_list); vd != NULL; 1016fa9e4066Sahrens vd = list_next(&spa->spa_dirty_list, vd)) { 1017fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 1018fa9e4066Sahrens NULL, NULL, ZIO_PRIORITY_NOW, 1019fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 1020fa9e4066Sahrens } 1021fa9e4066Sahrens (void) zio_wait(zio); 1022fa9e4066Sahrens 1023fa9e4066Sahrens return (0); 1024fa9e4066Sahrens } 1025