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 /* 22*21bf64a7Sgw * Copyright 2008 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 162*21bf64a7Sgw /* 163*21bf64a7Sgw * Returns back the vdev label associated with the passed in offset. 164*21bf64a7Sgw */ 165*21bf64a7Sgw int 166*21bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset) 167*21bf64a7Sgw { 168*21bf64a7Sgw int l; 169*21bf64a7Sgw 170*21bf64a7Sgw if (offset >= psize - VDEV_LABEL_END_SIZE) { 171*21bf64a7Sgw offset -= psize - VDEV_LABEL_END_SIZE; 172*21bf64a7Sgw offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 173*21bf64a7Sgw } 174*21bf64a7Sgw l = offset / sizeof (vdev_label_t); 175*21bf64a7Sgw return (l < VDEV_LABELS ? l : -1); 176*21bf64a7Sgw } 177*21bf64a7Sgw 178fa9e4066Sahrens static void 179fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 180fa9e4066Sahrens uint64_t size, zio_done_func_t *done, void *private) 181fa9e4066Sahrens { 182fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 183fa9e4066Sahrens 184fa9e4066Sahrens zio_nowait(zio_read_phys(zio, vd, 185fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 186fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 187ea8dc4b6Seschrock ZIO_PRIORITY_SYNC_READ, 188fa94a07fSbrendan ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 189fa94a07fSbrendan B_TRUE)); 190fa9e4066Sahrens } 191fa9e4066Sahrens 192fa9e4066Sahrens static void 193fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 19417f17c2dSbonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 195fa9e4066Sahrens { 196fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 197fa9e4066Sahrens 198fa9e4066Sahrens zio_nowait(zio_write_phys(zio, vd, 199fa9e4066Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 200fa9e4066Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 20117f17c2dSbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 202fa9e4066Sahrens } 203fa9e4066Sahrens 204fa9e4066Sahrens /* 205fa9e4066Sahrens * Generate the nvlist representing this vdev's config. 206fa9e4066Sahrens */ 207fa9e4066Sahrens nvlist_t * 20899653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 209fa94a07fSbrendan boolean_t isspare, boolean_t isl2cache) 210fa9e4066Sahrens { 211fa9e4066Sahrens nvlist_t *nv = NULL; 212fa9e4066Sahrens 213ea8dc4b6Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 214fa9e4066Sahrens 215fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 216fa9e4066Sahrens vd->vdev_ops->vdev_op_type) == 0); 217fa94a07fSbrendan if (!isspare && !isl2cache) 21899653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 21999653d4eSeschrock == 0); 220fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 221fa9e4066Sahrens 222fa9e4066Sahrens if (vd->vdev_path != NULL) 223fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 224fa9e4066Sahrens vd->vdev_path) == 0); 225fa9e4066Sahrens 226fa9e4066Sahrens if (vd->vdev_devid != NULL) 227fa9e4066Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 228fa9e4066Sahrens vd->vdev_devid) == 0); 229fa9e4066Sahrens 2303d7072f8Seschrock if (vd->vdev_physpath != NULL) 2313d7072f8Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 2323d7072f8Seschrock vd->vdev_physpath) == 0); 2333d7072f8Seschrock 23499653d4eSeschrock if (vd->vdev_nparity != 0) { 23599653d4eSeschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 23699653d4eSeschrock VDEV_TYPE_RAIDZ) == 0); 23799653d4eSeschrock 23899653d4eSeschrock /* 23999653d4eSeschrock * Make sure someone hasn't managed to sneak a fancy new vdev 24099653d4eSeschrock * into a crufty old storage pool. 24199653d4eSeschrock */ 24299653d4eSeschrock ASSERT(vd->vdev_nparity == 1 || 24399653d4eSeschrock (vd->vdev_nparity == 2 && 244e7437265Sahrens spa_version(spa) >= SPA_VERSION_RAID6)); 24599653d4eSeschrock 24699653d4eSeschrock /* 24799653d4eSeschrock * Note that we'll add the nparity tag even on storage pools 24899653d4eSeschrock * that only support a single parity device -- older software 24999653d4eSeschrock * will just ignore it. 25099653d4eSeschrock */ 25199653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 25299653d4eSeschrock vd->vdev_nparity) == 0); 25399653d4eSeschrock } 25499653d4eSeschrock 255afefbcddSeschrock if (vd->vdev_wholedisk != -1ULL) 256afefbcddSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 257afefbcddSeschrock vd->vdev_wholedisk) == 0); 258afefbcddSeschrock 259ea8dc4b6Seschrock if (vd->vdev_not_present) 260ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 261ea8dc4b6Seschrock 26299653d4eSeschrock if (vd->vdev_isspare) 26399653d4eSeschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 26499653d4eSeschrock 265fa94a07fSbrendan if (!isspare && !isl2cache && vd == vd->vdev_top) { 266fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 267fa9e4066Sahrens vd->vdev_ms_array) == 0); 268fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 269fa9e4066Sahrens vd->vdev_ms_shift) == 0); 270fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 271fa9e4066Sahrens vd->vdev_ashift) == 0); 272fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 273fa9e4066Sahrens vd->vdev_asize) == 0); 2748654d025Sperrin VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 2758654d025Sperrin vd->vdev_islog) == 0); 276fa9e4066Sahrens } 277fa9e4066Sahrens 278fa9e4066Sahrens if (vd->vdev_dtl.smo_object != 0) 279fa9e4066Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 280fa9e4066Sahrens vd->vdev_dtl.smo_object) == 0); 281fa9e4066Sahrens 282fa9e4066Sahrens if (getstats) { 283fa9e4066Sahrens vdev_stat_t vs; 284fa9e4066Sahrens vdev_get_stats(vd, &vs); 285fa9e4066Sahrens VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS, 286fa9e4066Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 287fa9e4066Sahrens } 288fa9e4066Sahrens 289fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 290fa9e4066Sahrens nvlist_t **child; 291fa9e4066Sahrens int c; 292fa9e4066Sahrens 293fa9e4066Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 294fa9e4066Sahrens KM_SLEEP); 295fa9e4066Sahrens 296fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 29799653d4eSeschrock child[c] = vdev_config_generate(spa, vd->vdev_child[c], 298fa94a07fSbrendan getstats, isspare, isl2cache); 299fa9e4066Sahrens 300fa9e4066Sahrens VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 301fa9e4066Sahrens child, vd->vdev_children) == 0); 302fa9e4066Sahrens 303fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 304fa9e4066Sahrens nvlist_free(child[c]); 305fa9e4066Sahrens 306fa9e4066Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 307441d80aaSlling 308441d80aaSlling } else { 309ecc2d604Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 310441d80aaSlling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 311ecc2d604Sbonwick B_TRUE) == 0); 3123d7072f8Seschrock if (vd->vdev_faulted) 3133d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 3143d7072f8Seschrock B_TRUE) == 0); 3153d7072f8Seschrock if (vd->vdev_degraded) 3163d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 3173d7072f8Seschrock B_TRUE) == 0); 3183d7072f8Seschrock if (vd->vdev_removed) 3193d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 3203d7072f8Seschrock B_TRUE) == 0); 3213d7072f8Seschrock if (vd->vdev_unspare) 3223d7072f8Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 3233d7072f8Seschrock B_TRUE) == 0); 324fa9e4066Sahrens } 325fa9e4066Sahrens 326fa9e4066Sahrens return (nv); 327fa9e4066Sahrens } 328fa9e4066Sahrens 329fa9e4066Sahrens nvlist_t * 330fa9e4066Sahrens vdev_label_read_config(vdev_t *vd) 331fa9e4066Sahrens { 3320373e76bSbonwick spa_t *spa = vd->vdev_spa; 333fa9e4066Sahrens nvlist_t *config = NULL; 334fa9e4066Sahrens vdev_phys_t *vp; 335fa9e4066Sahrens zio_t *zio; 336fa9e4066Sahrens int l; 337fa9e4066Sahrens 33891ebeef5Sahrens ASSERT(spa_config_held(spa, RW_READER) || 33991ebeef5Sahrens spa_config_held(spa, RW_WRITER)); 3400373e76bSbonwick 3410a4e9518Sgw if (!vdev_readable(vd)) 342fa9e4066Sahrens return (NULL); 343fa9e4066Sahrens 344fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 345fa9e4066Sahrens 346fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 347fa9e4066Sahrens 3480373e76bSbonwick zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | 349ea8dc4b6Seschrock ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD); 350fa9e4066Sahrens 351fa9e4066Sahrens vdev_label_read(zio, vd, l, vp, 352fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 353fa9e4066Sahrens sizeof (vdev_phys_t), NULL, NULL); 354fa9e4066Sahrens 355fa9e4066Sahrens if (zio_wait(zio) == 0 && 356fa9e4066Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 357ea8dc4b6Seschrock &config, 0) == 0) 358fa9e4066Sahrens break; 359fa9e4066Sahrens 360fa9e4066Sahrens if (config != NULL) { 361fa9e4066Sahrens nvlist_free(config); 362fa9e4066Sahrens config = NULL; 363fa9e4066Sahrens } 364fa9e4066Sahrens } 365fa9e4066Sahrens 366fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 367fa9e4066Sahrens 368fa9e4066Sahrens return (config); 369fa9e4066Sahrens } 370fa9e4066Sahrens 37139c23413Seschrock /* 37239c23413Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 37339c23413Seschrock * in with the device guid if this spare is active elsewhere on the system. 37439c23413Seschrock */ 37539c23413Seschrock static boolean_t 37639c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 377fa94a07fSbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 37839c23413Seschrock { 37939c23413Seschrock spa_t *spa = vd->vdev_spa; 38039c23413Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 38139c23413Seschrock uint64_t vdtxg = 0; 38239c23413Seschrock nvlist_t *label; 38339c23413Seschrock 38439c23413Seschrock if (spare_guid) 38539c23413Seschrock *spare_guid = 0ULL; 386fa94a07fSbrendan if (l2cache_guid) 387fa94a07fSbrendan *l2cache_guid = 0ULL; 38839c23413Seschrock 38939c23413Seschrock /* 39039c23413Seschrock * Read the label, if any, and perform some basic sanity checks. 39139c23413Seschrock */ 39239c23413Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 39339c23413Seschrock return (B_FALSE); 39439c23413Seschrock 39539c23413Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 39639c23413Seschrock &vdtxg); 39739c23413Seschrock 39839c23413Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 39939c23413Seschrock &state) != 0 || 40039c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 40139c23413Seschrock &device_guid) != 0) { 40239c23413Seschrock nvlist_free(label); 40339c23413Seschrock return (B_FALSE); 40439c23413Seschrock } 40539c23413Seschrock 406fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 40739c23413Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 40839c23413Seschrock &pool_guid) != 0 || 40939c23413Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 41039c23413Seschrock &txg) != 0)) { 41139c23413Seschrock nvlist_free(label); 41239c23413Seschrock return (B_FALSE); 41339c23413Seschrock } 41439c23413Seschrock 41539c23413Seschrock nvlist_free(label); 41639c23413Seschrock 41739c23413Seschrock /* 41839c23413Seschrock * Check to see if this device indeed belongs to the pool it claims to 41939c23413Seschrock * be a part of. The only way this is allowed is if the device is a hot 42039c23413Seschrock * spare (which we check for later on). 42139c23413Seschrock */ 422fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 42339c23413Seschrock !spa_guid_exists(pool_guid, device_guid) && 424fa94a07fSbrendan !spa_spare_exists(device_guid, NULL) && 425fa94a07fSbrendan !spa_l2cache_exists(device_guid, NULL)) 42639c23413Seschrock return (B_FALSE); 42739c23413Seschrock 42839c23413Seschrock /* 42939c23413Seschrock * If the transaction group is zero, then this an initialized (but 43039c23413Seschrock * unused) label. This is only an error if the create transaction 43139c23413Seschrock * on-disk is the same as the one we're using now, in which case the 43239c23413Seschrock * user has attempted to add the same vdev multiple times in the same 43339c23413Seschrock * transaction. 43439c23413Seschrock */ 435fa94a07fSbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 436fa94a07fSbrendan txg == 0 && vdtxg == crtxg) 43739c23413Seschrock return (B_TRUE); 43839c23413Seschrock 43939c23413Seschrock /* 44039c23413Seschrock * Check to see if this is a spare device. We do an explicit check for 44139c23413Seschrock * spa_has_spare() here because it may be on our pending list of spares 442fa94a07fSbrendan * to add. We also check if it is an l2cache device. 44339c23413Seschrock */ 44439c23413Seschrock if (spa_spare_exists(device_guid, &spare_pool) || 44539c23413Seschrock spa_has_spare(spa, device_guid)) { 44639c23413Seschrock if (spare_guid) 44739c23413Seschrock *spare_guid = device_guid; 44839c23413Seschrock 44939c23413Seschrock switch (reason) { 45039c23413Seschrock case VDEV_LABEL_CREATE: 451fa94a07fSbrendan case VDEV_LABEL_L2CACHE: 45239c23413Seschrock return (B_TRUE); 45339c23413Seschrock 45439c23413Seschrock case VDEV_LABEL_REPLACE: 45539c23413Seschrock return (!spa_has_spare(spa, device_guid) || 45639c23413Seschrock spare_pool != 0ULL); 45739c23413Seschrock 45839c23413Seschrock case VDEV_LABEL_SPARE: 45939c23413Seschrock return (spa_has_spare(spa, device_guid)); 46039c23413Seschrock } 46139c23413Seschrock } 46239c23413Seschrock 463fa94a07fSbrendan /* 464fa94a07fSbrendan * Check to see if this is an l2cache device. 465fa94a07fSbrendan */ 466fa94a07fSbrendan if (spa_l2cache_exists(device_guid, NULL)) 467fa94a07fSbrendan return (B_TRUE); 468fa94a07fSbrendan 46939c23413Seschrock /* 47039c23413Seschrock * If the device is marked ACTIVE, then this device is in use by another 47139c23413Seschrock * pool on the system. 47239c23413Seschrock */ 47339c23413Seschrock return (state == POOL_STATE_ACTIVE); 47439c23413Seschrock } 47539c23413Seschrock 47639c23413Seschrock /* 47739c23413Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 47839c23413Seschrock * use, and writable. We put down an initial label which we will later 47939c23413Seschrock * overwrite with a complete label. Note that it's important to do this 48039c23413Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 48139c23413Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 48239c23413Seschrock * itself. 48339c23413Seschrock */ 48439c23413Seschrock int 48539c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 486fa9e4066Sahrens { 487fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 488fa9e4066Sahrens nvlist_t *label; 489fa9e4066Sahrens vdev_phys_t *vp; 490fa9e4066Sahrens vdev_boot_header_t *vb; 491ecc2d604Sbonwick uberblock_t *ub; 492fa9e4066Sahrens zio_t *zio; 493fa9e4066Sahrens int l, c, n; 494fa9e4066Sahrens char *buf; 495fa9e4066Sahrens size_t buflen; 496fa9e4066Sahrens int error; 497fa94a07fSbrendan uint64_t spare_guid, l2cache_guid; 49817f17c2dSbonwick int flags = ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL; 499fa9e4066Sahrens 5000373e76bSbonwick ASSERT(spa_config_held(spa, RW_WRITER)); 5010373e76bSbonwick 502fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 50339c23413Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 50439c23413Seschrock crtxg, reason)) != 0) 505fa9e4066Sahrens return (error); 506fa9e4066Sahrens 507fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 508fa9e4066Sahrens return (0); 509fa9e4066Sahrens 510fa9e4066Sahrens /* 51139c23413Seschrock * Dead vdevs cannot be initialized. 512fa9e4066Sahrens */ 513fa9e4066Sahrens if (vdev_is_dead(vd)) 514fa9e4066Sahrens return (EIO); 515fa9e4066Sahrens 516fa9e4066Sahrens /* 51739c23413Seschrock * Determine if the vdev is in use. 518fa9e4066Sahrens */ 51939c23413Seschrock if (reason != VDEV_LABEL_REMOVE && 520fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 52139c23413Seschrock return (EBUSY); 52239c23413Seschrock 52339c23413Seschrock ASSERT(reason != VDEV_LABEL_REMOVE || 524fa94a07fSbrendan vdev_inuse(vd, crtxg, reason, NULL, NULL)); 52539c23413Seschrock 52639c23413Seschrock /* 527fa94a07fSbrendan * If this is a request to add or replace a spare or l2cache device 528fa94a07fSbrendan * that is in use elsewhere on the system, then we must update the 529fa94a07fSbrendan * guid (which was initialized to a random value) to reflect the 530fa94a07fSbrendan * actual GUID (which is shared between multiple pools). 53139c23413Seschrock */ 532fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 533fa94a07fSbrendan spare_guid != 0ULL) { 53439c23413Seschrock vdev_t *pvd = vd->vdev_parent; 53539c23413Seschrock 53639c23413Seschrock for (; pvd != NULL; pvd = pvd->vdev_parent) { 53739c23413Seschrock pvd->vdev_guid_sum -= vd->vdev_guid; 53839c23413Seschrock pvd->vdev_guid_sum += spare_guid; 539fa9e4066Sahrens } 54099653d4eSeschrock 54139c23413Seschrock vd->vdev_guid = vd->vdev_guid_sum = spare_guid; 54239c23413Seschrock 54399653d4eSeschrock /* 54439c23413Seschrock * If this is a replacement, then we want to fallthrough to the 54539c23413Seschrock * rest of the code. If we're adding a spare, then it's already 5463d7072f8Seschrock * labeled appropriately and we can just return. 54799653d4eSeschrock */ 54839c23413Seschrock if (reason == VDEV_LABEL_SPARE) 54939c23413Seschrock return (0); 55039c23413Seschrock ASSERT(reason == VDEV_LABEL_REPLACE); 551fa9e4066Sahrens } 552fa9e4066Sahrens 553fa94a07fSbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 554fa94a07fSbrendan l2cache_guid != 0ULL) { 555fa94a07fSbrendan vdev_t *pvd = vd->vdev_parent; 556fa94a07fSbrendan 557fa94a07fSbrendan for (; pvd != NULL; pvd = pvd->vdev_parent) { 558fa94a07fSbrendan pvd->vdev_guid_sum -= vd->vdev_guid; 559fa94a07fSbrendan pvd->vdev_guid_sum += l2cache_guid; 560fa94a07fSbrendan } 561fa94a07fSbrendan 562fa94a07fSbrendan vd->vdev_guid = vd->vdev_guid_sum = l2cache_guid; 563fa94a07fSbrendan 564fa94a07fSbrendan /* 565fa94a07fSbrendan * If this is a replacement, then we want to fallthrough to the 566fa94a07fSbrendan * rest of the code. If we're adding an l2cache, then it's 567fa94a07fSbrendan * already labeled appropriately and we can just return. 568fa94a07fSbrendan */ 569fa94a07fSbrendan if (reason == VDEV_LABEL_L2CACHE) 570fa94a07fSbrendan return (0); 571fa94a07fSbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 572fa94a07fSbrendan } 573fa94a07fSbrendan 574fa9e4066Sahrens /* 57539c23413Seschrock * Initialize its label. 576fa9e4066Sahrens */ 577fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 578fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 579fa9e4066Sahrens 580fa9e4066Sahrens /* 581fa9e4066Sahrens * Generate a label describing the pool and our top-level vdev. 582fa9e4066Sahrens * We mark it as being from txg 0 to indicate that it's not 583fa9e4066Sahrens * really part of an active pool just yet. The labels will 584fa9e4066Sahrens * be written again with a meaningful txg by spa_sync(). 585fa9e4066Sahrens */ 58639c23413Seschrock if (reason == VDEV_LABEL_SPARE || 58739c23413Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 58839c23413Seschrock /* 58939c23413Seschrock * For inactive hot spares, we generate a special label that 59039c23413Seschrock * identifies as a mutually shared hot spare. We write the 59139c23413Seschrock * label if we are adding a hot spare, or if we are removing an 59239c23413Seschrock * active hot spare (in which case we want to revert the 59339c23413Seschrock * labels). 59439c23413Seschrock */ 59599653d4eSeschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 59699653d4eSeschrock 59799653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 59899653d4eSeschrock spa_version(spa)) == 0); 59999653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 60099653d4eSeschrock POOL_STATE_SPARE) == 0); 60199653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 60299653d4eSeschrock vd->vdev_guid) == 0); 603fa94a07fSbrendan } else if (reason == VDEV_LABEL_L2CACHE || 604fa94a07fSbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 605fa94a07fSbrendan /* 606fa94a07fSbrendan * For level 2 ARC devices, add a special label. 607fa94a07fSbrendan */ 608fa94a07fSbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 609fa94a07fSbrendan 610fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 611fa94a07fSbrendan spa_version(spa)) == 0); 612fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 613fa94a07fSbrendan POOL_STATE_L2CACHE) == 0); 614fa94a07fSbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 615fa94a07fSbrendan vd->vdev_guid) == 0); 61699653d4eSeschrock } else { 61799653d4eSeschrock label = spa_config_generate(spa, vd, 0ULL, B_FALSE); 61899653d4eSeschrock 61999653d4eSeschrock /* 62099653d4eSeschrock * Add our creation time. This allows us to detect multiple 62199653d4eSeschrock * vdev uses as described above, and automatically expires if we 62299653d4eSeschrock * fail. 62399653d4eSeschrock */ 62499653d4eSeschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 62599653d4eSeschrock crtxg) == 0); 62699653d4eSeschrock } 627fa9e4066Sahrens 628fa9e4066Sahrens buf = vp->vp_nvlist; 629fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 630fa9e4066Sahrens 631a75573b6Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 632a75573b6Smmusante if (error != 0) { 633fa9e4066Sahrens nvlist_free(label); 634fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 635a75573b6Smmusante /* EFAULT means nvlist_pack ran out of room */ 636a75573b6Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 637fa9e4066Sahrens } 638fa9e4066Sahrens 639fa9e4066Sahrens /* 640fa9e4066Sahrens * Initialize boot block header. 641fa9e4066Sahrens */ 642fa9e4066Sahrens vb = zio_buf_alloc(sizeof (vdev_boot_header_t)); 643fa9e4066Sahrens bzero(vb, sizeof (vdev_boot_header_t)); 644fa9e4066Sahrens vb->vb_magic = VDEV_BOOT_MAGIC; 645fa9e4066Sahrens vb->vb_version = VDEV_BOOT_VERSION; 646fa9e4066Sahrens vb->vb_offset = VDEV_BOOT_OFFSET; 647fa9e4066Sahrens vb->vb_size = VDEV_BOOT_SIZE; 648fa9e4066Sahrens 649fa9e4066Sahrens /* 650fa9e4066Sahrens * Initialize uberblock template. 651fa9e4066Sahrens */ 652ecc2d604Sbonwick ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 653ecc2d604Sbonwick bzero(ub, VDEV_UBERBLOCK_SIZE(vd)); 654ecc2d604Sbonwick *ub = spa->spa_uberblock; 655ecc2d604Sbonwick ub->ub_txg = 0; 656fa9e4066Sahrens 657fa9e4066Sahrens /* 658fa9e4066Sahrens * Write everything in parallel. 659fa9e4066Sahrens */ 66017f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 661fa9e4066Sahrens 662fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 663fa9e4066Sahrens 664fa9e4066Sahrens vdev_label_write(zio, vd, l, vp, 665fa9e4066Sahrens offsetof(vdev_label_t, vl_vdev_phys), 66617f17c2dSbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 667fa9e4066Sahrens 668fa9e4066Sahrens vdev_label_write(zio, vd, l, vb, 669fa9e4066Sahrens offsetof(vdev_label_t, vl_boot_header), 67017f17c2dSbonwick sizeof (vdev_boot_header_t), NULL, NULL, flags); 671fa9e4066Sahrens 672ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 673ecc2d604Sbonwick vdev_label_write(zio, vd, l, ub, 674ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 67517f17c2dSbonwick VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags); 676fa9e4066Sahrens } 677fa9e4066Sahrens } 678fa9e4066Sahrens 679fa9e4066Sahrens error = zio_wait(zio); 680fa9e4066Sahrens 681fa9e4066Sahrens nvlist_free(label); 682ecc2d604Sbonwick zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd)); 683fa9e4066Sahrens zio_buf_free(vb, sizeof (vdev_boot_header_t)); 684fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 685fa9e4066Sahrens 68639c23413Seschrock /* 68739c23413Seschrock * If this vdev hasn't been previously identified as a spare, then we 6883d7072f8Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 689fa94a07fSbrendan * exists as a spare elsewhere in the system. Do the same for 690fa94a07fSbrendan * level 2 ARC devices. 69139c23413Seschrock */ 69239c23413Seschrock if (error == 0 && !vd->vdev_isspare && 69339c23413Seschrock (reason == VDEV_LABEL_SPARE || 69439c23413Seschrock spa_spare_exists(vd->vdev_guid, NULL))) 69539c23413Seschrock spa_spare_add(vd); 69699653d4eSeschrock 697fa94a07fSbrendan if (error == 0 && !vd->vdev_isl2cache && 698fa94a07fSbrendan (reason == VDEV_LABEL_L2CACHE || 699fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 700fa94a07fSbrendan spa_l2cache_add(vd); 701fa94a07fSbrendan 70239c23413Seschrock return (error); 70399653d4eSeschrock } 70499653d4eSeschrock 705fa9e4066Sahrens /* 706fa9e4066Sahrens * ========================================================================== 707fa9e4066Sahrens * uberblock load/sync 708fa9e4066Sahrens * ========================================================================== 709fa9e4066Sahrens */ 710fa9e4066Sahrens 711fa9e4066Sahrens /* 712fa9e4066Sahrens * Consider the following situation: txg is safely synced to disk. We've 713fa9e4066Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 714fa9e4066Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 715fa9e4066Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 716fa9e4066Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 717fa9e4066Sahrens * the missing replica comes back, then for a new seconds we'll have two 718fa9e4066Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 719fa9e4066Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 720fa9e4066Sahrens */ 721fa9e4066Sahrens static int 722fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 723fa9e4066Sahrens { 724fa9e4066Sahrens if (ub1->ub_txg < ub2->ub_txg) 725fa9e4066Sahrens return (-1); 726fa9e4066Sahrens if (ub1->ub_txg > ub2->ub_txg) 727fa9e4066Sahrens return (1); 728fa9e4066Sahrens 729fa9e4066Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 730fa9e4066Sahrens return (-1); 731fa9e4066Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 732fa9e4066Sahrens return (1); 733fa9e4066Sahrens 734fa9e4066Sahrens return (0); 735fa9e4066Sahrens } 736fa9e4066Sahrens 737fa9e4066Sahrens static void 738fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio) 739fa9e4066Sahrens { 740ecc2d604Sbonwick uberblock_t *ub = zio->io_data; 741fa9e4066Sahrens uberblock_t *ubbest = zio->io_private; 742fa9e4066Sahrens spa_t *spa = zio->io_spa; 743fa9e4066Sahrens 744ecc2d604Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 745fa9e4066Sahrens 746ea8dc4b6Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 747fa9e4066Sahrens mutex_enter(&spa->spa_uberblock_lock); 748fa9e4066Sahrens if (vdev_uberblock_compare(ub, ubbest) > 0) 749fa9e4066Sahrens *ubbest = *ub; 750fa9e4066Sahrens mutex_exit(&spa->spa_uberblock_lock); 751fa9e4066Sahrens } 752fa9e4066Sahrens 753fa9e4066Sahrens zio_buf_free(zio->io_data, zio->io_size); 754fa9e4066Sahrens } 755fa9e4066Sahrens 756fa9e4066Sahrens void 757fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 758fa9e4066Sahrens { 759fa9e4066Sahrens int l, c, n; 760fa9e4066Sahrens 761fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 762fa9e4066Sahrens vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 763fa9e4066Sahrens 764fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 765fa9e4066Sahrens return; 766fa9e4066Sahrens 767fa9e4066Sahrens if (vdev_is_dead(vd)) 768fa9e4066Sahrens return; 769fa9e4066Sahrens 770fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) { 771ecc2d604Sbonwick for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 772fa9e4066Sahrens vdev_label_read(zio, vd, l, 773ecc2d604Sbonwick zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 774ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 775ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 776fa9e4066Sahrens vdev_uberblock_load_done, ubbest); 777fa9e4066Sahrens } 778fa9e4066Sahrens } 779fa9e4066Sahrens } 780fa9e4066Sahrens 781fa9e4066Sahrens /* 78217f17c2dSbonwick * On success, increment root zio's count of good writes. 7830373e76bSbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 784fa9e4066Sahrens */ 785fa9e4066Sahrens static void 786fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio) 787fa9e4066Sahrens { 78817f17c2dSbonwick uint64_t *good_writes = zio->io_private; 789fa9e4066Sahrens 7900373e76bSbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 791fa9e4066Sahrens atomic_add_64(good_writes, 1); 792fa9e4066Sahrens } 793fa9e4066Sahrens 79417f17c2dSbonwick /* 79517f17c2dSbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 79617f17c2dSbonwick */ 797fa9e4066Sahrens static void 79817f17c2dSbonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd) 799fa9e4066Sahrens { 800fa9e4066Sahrens int l, c, n; 80117f17c2dSbonwick uberblock_t *ubbuf; 802fa9e4066Sahrens 803fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 80417f17c2dSbonwick vdev_uberblock_sync(zio, ub, vd->vdev_child[c]); 805fa9e4066Sahrens 806fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 807fa9e4066Sahrens return; 808fa9e4066Sahrens 809fa9e4066Sahrens if (vdev_is_dead(vd)) 810fa9e4066Sahrens return; 811fa9e4066Sahrens 81217f17c2dSbonwick n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 813fa9e4066Sahrens 81417f17c2dSbonwick ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 81517f17c2dSbonwick bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 81617f17c2dSbonwick *ubbuf = *ub; 817fa9e4066Sahrens 818fa9e4066Sahrens for (l = 0; l < VDEV_LABELS; l++) 81917f17c2dSbonwick vdev_label_write(zio, vd, l, ubbuf, 820ecc2d604Sbonwick VDEV_UBERBLOCK_OFFSET(vd, n), 821ecc2d604Sbonwick VDEV_UBERBLOCK_SIZE(vd), 82217f17c2dSbonwick vdev_uberblock_sync_done, zio->io_private, 82317f17c2dSbonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE); 824fa9e4066Sahrens 82517f17c2dSbonwick zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 826fa9e4066Sahrens } 827fa9e4066Sahrens 828*21bf64a7Sgw static void 829*21bf64a7Sgw vdev_uberblock_sync_list_done(zio_t *zio) 830*21bf64a7Sgw { 831*21bf64a7Sgw uint64_t *good_writes = zio->io_private; 832*21bf64a7Sgw 833*21bf64a7Sgw if (*good_writes == 0) 834*21bf64a7Sgw zio->io_error = EIO; 835*21bf64a7Sgw } 836*21bf64a7Sgw 83717f17c2dSbonwick int 83817f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 839fa9e4066Sahrens { 84017f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 84117f17c2dSbonwick int v; 842*21bf64a7Sgw zio_t *zio, *nio; 84317f17c2dSbonwick uint64_t good_writes = 0; 844*21bf64a7Sgw int io_flags = flags; 845fa9e4066Sahrens 846*21bf64a7Sgw /* 847*21bf64a7Sgw * If we've been asked to update all the vdevs then we change 848*21bf64a7Sgw * our flags to ZIO_FLAG_MUSTSUCCEED so that the pipeline can 849*21bf64a7Sgw * handle error should all update fail. 850*21bf64a7Sgw */ 851*21bf64a7Sgw if (svdcount == spa->spa_root_vdev->vdev_children) 852*21bf64a7Sgw io_flags &= ~ZIO_FLAG_CANFAIL; 853fa9e4066Sahrens 854*21bf64a7Sgw /* 855*21bf64a7Sgw * We rely on the value of good_writes and the root I/O to determine 856*21bf64a7Sgw * how a complete failure is handled. In the event that the root is a 857*21bf64a7Sgw * ZIO_FLAG_MUSTSUCCED, then the pipeline will block this I/O if we 858*21bf64a7Sgw * were unable to update any uberblock. Once the I/O is blocked the 859*21bf64a7Sgw * pipeline will retry it when the error is cleared. Unfortunately, 860*21bf64a7Sgw * the pipeline does not have the complete I/O tree so it will be 861*21bf64a7Sgw * unable to retry the actual uberblock update. Instead we rely on 862*21bf64a7Sgw * the value of good_writes to return the failed status to the caller 863*21bf64a7Sgw * which will retry on error and thus resubmit the complete I/O 864*21bf64a7Sgw * tree. 865*21bf64a7Sgw */ 866*21bf64a7Sgw zio = zio_root(spa, NULL, NULL, io_flags); 867*21bf64a7Sgw nio = zio_null(zio, spa, vdev_uberblock_sync_list_done, &good_writes, 868*21bf64a7Sgw flags); 86917f17c2dSbonwick for (v = 0; v < svdcount; v++) 870*21bf64a7Sgw vdev_uberblock_sync(nio, ub, svd[v]); 871*21bf64a7Sgw zio_nowait(nio); 87217f17c2dSbonwick (void) zio_wait(zio); 873fa9e4066Sahrens 874fa9e4066Sahrens /* 87517f17c2dSbonwick * Flush the uberblocks to disk. This ensures that the odd labels 87617f17c2dSbonwick * are no longer needed (because the new uberblocks and the even 87717f17c2dSbonwick * labels are safely on disk), so it is safe to overwrite them. 878fa9e4066Sahrens */ 87917f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 880fa9e4066Sahrens 88117f17c2dSbonwick for (v = 0; v < svdcount; v++) 88217f17c2dSbonwick zio_flush(zio, svd[v]); 883fa9e4066Sahrens 88417f17c2dSbonwick (void) zio_wait(zio); 88517f17c2dSbonwick 88617f17c2dSbonwick return (good_writes >= 1 ? 0 : EIO); 887fa9e4066Sahrens } 888fa9e4066Sahrens 889fa9e4066Sahrens /* 89017f17c2dSbonwick * On success, increment the count of good writes for our top-level vdev. 891fa9e4066Sahrens */ 892fa9e4066Sahrens static void 89317f17c2dSbonwick vdev_label_sync_done(zio_t *zio) 894fa9e4066Sahrens { 89517f17c2dSbonwick uint64_t *good_writes = zio->io_private; 896fa9e4066Sahrens 897fa9e4066Sahrens if (zio->io_error == 0) 898fa9e4066Sahrens atomic_add_64(good_writes, 1); 899fa9e4066Sahrens } 900fa9e4066Sahrens 90117f17c2dSbonwick /* 90217f17c2dSbonwick * If there weren't enough good writes, indicate failure to the parent. 90317f17c2dSbonwick */ 904fa9e4066Sahrens static void 90517f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio) 90617f17c2dSbonwick { 90717f17c2dSbonwick uint64_t *good_writes = zio->io_private; 90817f17c2dSbonwick 90917f17c2dSbonwick if (*good_writes == 0) 91017f17c2dSbonwick zio->io_error = EIO; 91117f17c2dSbonwick 91217f17c2dSbonwick kmem_free(good_writes, sizeof (uint64_t)); 91317f17c2dSbonwick } 91417f17c2dSbonwick 91517f17c2dSbonwick /* 91617f17c2dSbonwick * Write all even or odd labels to all leaves of the specified vdev. 91717f17c2dSbonwick */ 91817f17c2dSbonwick static void 91917f17c2dSbonwick vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg) 920fa9e4066Sahrens { 921fa9e4066Sahrens nvlist_t *label; 922fa9e4066Sahrens vdev_phys_t *vp; 923fa9e4066Sahrens char *buf; 924fa9e4066Sahrens size_t buflen; 925fa9e4066Sahrens int c; 926fa9e4066Sahrens 927fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 92817f17c2dSbonwick vdev_label_sync(zio, vd->vdev_child[c], l, txg); 929fa9e4066Sahrens 930fa9e4066Sahrens if (!vd->vdev_ops->vdev_op_leaf) 931fa9e4066Sahrens return; 932fa9e4066Sahrens 933fa9e4066Sahrens if (vdev_is_dead(vd)) 934fa9e4066Sahrens return; 935fa9e4066Sahrens 936fa9e4066Sahrens /* 937fa9e4066Sahrens * Generate a label describing the top-level config to which we belong. 938fa9e4066Sahrens */ 9390373e76bSbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 940fa9e4066Sahrens 941fa9e4066Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 942fa9e4066Sahrens bzero(vp, sizeof (vdev_phys_t)); 943fa9e4066Sahrens 944fa9e4066Sahrens buf = vp->vp_nvlist; 945fa9e4066Sahrens buflen = sizeof (vp->vp_nvlist); 946fa9e4066Sahrens 94717f17c2dSbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 94817f17c2dSbonwick for (; l < VDEV_LABELS; l += 2) { 94917f17c2dSbonwick vdev_label_write(zio, vd, l, vp, 95017f17c2dSbonwick offsetof(vdev_label_t, vl_vdev_phys), 95117f17c2dSbonwick sizeof (vdev_phys_t), 95217f17c2dSbonwick vdev_label_sync_done, zio->io_private, 95317f17c2dSbonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE); 95417f17c2dSbonwick } 95517f17c2dSbonwick } 956fa9e4066Sahrens 957fa9e4066Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 958fa9e4066Sahrens nvlist_free(label); 959fa9e4066Sahrens } 960fa9e4066Sahrens 96117f17c2dSbonwick int 96217f17c2dSbonwick vdev_label_sync_list(spa_t *spa, int l, int flags, uint64_t txg) 963fa9e4066Sahrens { 96417f17c2dSbonwick list_t *dl = &spa->spa_dirty_list; 96517f17c2dSbonwick vdev_t *vd; 966*21bf64a7Sgw zio_t *zio, *nio; 967fa9e4066Sahrens int error; 968*21bf64a7Sgw int io_flags = flags & ~ZIO_FLAG_CANFAIL; 969fa9e4066Sahrens 970fa9e4066Sahrens /* 971*21bf64a7Sgw * The root I/O for all label updates must succeed and we track 972*21bf64a7Sgw * the error returned back from the null I/O to determine if we 973*21bf64a7Sgw * need to reissue the I/O tree from scratch. If we are unable 974*21bf64a7Sgw * to update any leaf vdev associated with a dirty top-level vdev, 975*21bf64a7Sgw * then the pipeline will either suspend or panic when the root I/O 976*21bf64a7Sgw * is issued. If the error is cleared, then the pipleine will retry 977*21bf64a7Sgw * the root I/O. Unfortunately we've lost the entire I/O tree so we 978*21bf64a7Sgw * return back the original error to the caller and allow the caller 979*21bf64a7Sgw * to call use again so that we can build the I/O tree from scratch. 980fa9e4066Sahrens */ 981*21bf64a7Sgw zio = zio_root(spa, NULL, NULL, io_flags); 982*21bf64a7Sgw nio = zio_null(zio, spa, NULL, NULL, flags); 983fa9e4066Sahrens 98417f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 98517f17c2dSbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 98617f17c2dSbonwick KM_SLEEP); 987*21bf64a7Sgw zio_t *vio = zio_null(nio, spa, vdev_label_sync_top_done, 98817f17c2dSbonwick good_writes, flags); 98917f17c2dSbonwick vdev_label_sync(vio, vd, l, txg); 99017f17c2dSbonwick zio_nowait(vio); 991fa9e4066Sahrens } 992*21bf64a7Sgw error = zio_wait(nio); 993*21bf64a7Sgw (void) zio_wait(zio); 994fa9e4066Sahrens 9958654d025Sperrin /* 99617f17c2dSbonwick * Flush the new labels to disk. 9978654d025Sperrin */ 99817f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 9998654d025Sperrin 100017f17c2dSbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 100117f17c2dSbonwick zio_flush(zio, vd); 100217f17c2dSbonwick 100317f17c2dSbonwick (void) zio_wait(zio); 1004fa9e4066Sahrens 1005fa9e4066Sahrens return (error); 1006fa9e4066Sahrens } 1007fa9e4066Sahrens 1008fa9e4066Sahrens /* 100917f17c2dSbonwick * Sync the uberblock and any changes to the vdev configuration. 1010fa9e4066Sahrens * 1011fa9e4066Sahrens * The order of operations is carefully crafted to ensure that 1012fa9e4066Sahrens * if the system panics or loses power at any time, the state on disk 1013fa9e4066Sahrens * is still transactionally consistent. The in-line comments below 1014fa9e4066Sahrens * describe the failure semantics at each stage. 1015fa9e4066Sahrens * 101617f17c2dSbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1017fa9e4066Sahrens * at any time, you can just call it again, and it will resume its work. 1018fa9e4066Sahrens */ 1019*21bf64a7Sgw void 102017f17c2dSbonwick vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg) 1021fa9e4066Sahrens { 102217f17c2dSbonwick spa_t *spa = svd[0]->vdev_spa; 1023fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 10240373e76bSbonwick vdev_t *vd; 1025fa9e4066Sahrens zio_t *zio; 102617f17c2dSbonwick int flags = ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL; 1027fa9e4066Sahrens 1028fa9e4066Sahrens ASSERT(ub->ub_txg <= txg); 1029fa9e4066Sahrens 1030fa9e4066Sahrens /* 103117f17c2dSbonwick * If this isn't a resync due to I/O errors, 103217f17c2dSbonwick * and nothing changed in this transaction group, 103317f17c2dSbonwick * and the vdev configuration hasn't changed, 10340373e76bSbonwick * then there's nothing to do. 1035fa9e4066Sahrens */ 103617f17c2dSbonwick if (ub->ub_txg < txg && 103717f17c2dSbonwick uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 103817f17c2dSbonwick list_is_empty(&spa->spa_dirty_list)) 1039*21bf64a7Sgw return; 1040fa9e4066Sahrens 1041fa9e4066Sahrens if (txg > spa_freeze_txg(spa)) 1042*21bf64a7Sgw return; 1043fa9e4066Sahrens 10440373e76bSbonwick ASSERT(txg <= spa->spa_final_txg); 10450373e76bSbonwick 1046fa9e4066Sahrens /* 1047fa9e4066Sahrens * Flush the write cache of every disk that's been written to 1048fa9e4066Sahrens * in this transaction group. This ensures that all blocks 1049fa9e4066Sahrens * written in this txg will be committed to stable storage 1050fa9e4066Sahrens * before any uberblock that references them. 1051fa9e4066Sahrens */ 105217f17c2dSbonwick zio = zio_root(spa, NULL, NULL, flags); 105317f17c2dSbonwick 1054fa9e4066Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 105517f17c2dSbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 105617f17c2dSbonwick zio_flush(zio, vd); 105717f17c2dSbonwick 1058fa9e4066Sahrens (void) zio_wait(zio); 1059fa9e4066Sahrens 1060fa9e4066Sahrens /* 1061fa9e4066Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1062fa9e4066Sahrens * system dies in the middle of this process, that's OK: all of the 1063fa9e4066Sahrens * even labels that made it to disk will be newer than any uberblock, 1064fa9e4066Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 106517f17c2dSbonwick * which have not yet been touched, will still be valid. We flush 106617f17c2dSbonwick * the new labels to disk to ensure that all even-label updates 106717f17c2dSbonwick * are committed to stable storage before the uberblock update. 1068*21bf64a7Sgw * Failure to update any of the labels will invoke the 'failmode' 1069*21bf64a7Sgw * code path. Thus we must retry the entire I/O tree once the error 1070*21bf64a7Sgw * is cleared and we ar resumed. 1071fa9e4066Sahrens */ 1072*21bf64a7Sgw while (vdev_label_sync_list(spa, 0, flags, txg) != 0) 1073*21bf64a7Sgw ; 1074fa9e4066Sahrens 1075fa9e4066Sahrens /* 1076*21bf64a7Sgw * Sync the uberblocks to all vdevs in svd[]. If we are unable 1077*21bf64a7Sgw * to do so, then we attempt to sync out to all top-level vdevs. 10780373e76bSbonwick * If the system dies in the middle of this step, there are two cases 10790373e76bSbonwick * to consider, and the on-disk state is consistent either way: 1080fa9e4066Sahrens * 1081fa9e4066Sahrens * (1) If none of the new uberblocks made it to disk, then the 1082fa9e4066Sahrens * previous uberblock will be the newest, and the odd labels 1083fa9e4066Sahrens * (which had not yet been touched) will be valid with respect 1084fa9e4066Sahrens * to that uberblock. 1085fa9e4066Sahrens * 1086fa9e4066Sahrens * (2) If one or more new uberblocks made it to disk, then they 1087fa9e4066Sahrens * will be the newest, and the even labels (which had all 1088fa9e4066Sahrens * been successfully committed) will be valid with respect 1089fa9e4066Sahrens * to the new uberblocks. 1090*21bf64a7Sgw * 1091*21bf64a7Sgw * In addition, if we have failed to update all the uberblocks then 1092*21bf64a7Sgw * we will follow the 'failmode' code path. We must retry the entire 1093*21bf64a7Sgw * I/O tree if we are resumed. 1094fa9e4066Sahrens */ 1095*21bf64a7Sgw if (vdev_uberblock_sync_list(svd, svdcount, ub, flags) != 0) { 1096*21bf64a7Sgw vdev_t *rvd = spa->spa_root_vdev; 1097*21bf64a7Sgw 1098*21bf64a7Sgw while (vdev_uberblock_sync_list(rvd->vdev_child, 1099*21bf64a7Sgw rvd->vdev_children, ub, flags)) 1100*21bf64a7Sgw ; 1101*21bf64a7Sgw } 1102fa9e4066Sahrens 1103fa9e4066Sahrens /* 1104fa9e4066Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1105fa9e4066Sahrens * in the middle of this process, the even labels and the new 1106fa9e4066Sahrens * uberblocks will suffice to open the pool. The next time 1107fa9e4066Sahrens * the pool is opened, the first thing we'll do -- before any 1108fa9e4066Sahrens * user data is modified -- is mark every vdev dirty so that 110917f17c2dSbonwick * all labels will be brought up to date. We flush the new labels 111017f17c2dSbonwick * to disk to ensure that all odd-label updates are committed to 111117f17c2dSbonwick * stable storage before the next transaction group begins. 1112*21bf64a7Sgw * Failure to update any of the labels will invoke the 'failmode' 1113*21bf64a7Sgw * code path. Thus we must retry the entire I/O tree once the error 1114*21bf64a7Sgw * is cleared and we are resumed. 1115fa9e4066Sahrens */ 1116*21bf64a7Sgw while (vdev_label_sync_list(spa, 1, flags, txg) != 0) 1117*21bf64a7Sgw ; 1118fa9e4066Sahrens } 1119