1789Sahrens /* 2789Sahrens * CDDL HEADER START 3789Sahrens * 4789Sahrens * The contents of this file are subject to the terms of the 51485Slling * Common Development and Distribution License (the "License"). 61485Slling * You may not use this file except in compliance with the License. 7789Sahrens * 8789Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9789Sahrens * or http://www.opensolaris.org/os/licensing. 10789Sahrens * See the License for the specific language governing permissions 11789Sahrens * and limitations under the License. 12789Sahrens * 13789Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14789Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15789Sahrens * If applicable, add the following below this CDDL HEADER, with the 16789Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17789Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18789Sahrens * 19789Sahrens * CDDL HEADER END 20789Sahrens */ 21789Sahrens /* 22*12296SLin.Ling@Sun.COM * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23789Sahrens */ 24789Sahrens 25789Sahrens /* 26789Sahrens * Virtual Device Labels 27789Sahrens * --------------------- 28789Sahrens * 29789Sahrens * The vdev label serves several distinct purposes: 30789Sahrens * 31789Sahrens * 1. Uniquely identify this device as part of a ZFS pool and confirm its 32789Sahrens * identity within the pool. 33789Sahrens * 34789Sahrens * 2. Verify that all the devices given in a configuration are present 35789Sahrens * within the pool. 36789Sahrens * 37789Sahrens * 3. Determine the uberblock for the pool. 38789Sahrens * 39789Sahrens * 4. In case of an import operation, determine the configuration of the 40789Sahrens * toplevel vdev of which it is a part. 41789Sahrens * 42789Sahrens * 5. If an import operation cannot find all the devices in the pool, 43789Sahrens * provide enough information to the administrator to determine which 44789Sahrens * devices are missing. 45789Sahrens * 46789Sahrens * It is important to note that while the kernel is responsible for writing the 47789Sahrens * label, it only consumes the information in the first three cases. The 48789Sahrens * latter information is only consumed in userland when determining the 49789Sahrens * configuration to import a pool. 50789Sahrens * 51789Sahrens * 52789Sahrens * Label Organization 53789Sahrens * ------------------ 54789Sahrens * 55789Sahrens * Before describing the contents of the label, it's important to understand how 56789Sahrens * the labels are written and updated with respect to the uberblock. 57789Sahrens * 58789Sahrens * When the pool configuration is altered, either because it was newly created 59789Sahrens * or a device was added, we want to update all the labels such that we can deal 60789Sahrens * with fatal failure at any point. To this end, each disk has two labels which 61789Sahrens * are updated before and after the uberblock is synced. Assuming we have 624451Seschrock * labels and an uberblock with the following transaction groups: 63789Sahrens * 64789Sahrens * L1 UB L2 65789Sahrens * +------+ +------+ +------+ 66789Sahrens * | | | | | | 67789Sahrens * | t10 | | t10 | | t10 | 68789Sahrens * | | | | | | 69789Sahrens * +------+ +------+ +------+ 70789Sahrens * 71789Sahrens * In this stable state, the labels and the uberblock were all updated within 72789Sahrens * the same transaction group (10). Each label is mirrored and checksummed, so 73789Sahrens * that we can detect when we fail partway through writing the label. 74789Sahrens * 75789Sahrens * In order to identify which labels are valid, the labels are written in the 76789Sahrens * following manner: 77789Sahrens * 78789Sahrens * 1. For each vdev, update 'L1' to the new label 79789Sahrens * 2. Update the uberblock 80789Sahrens * 3. For each vdev, update 'L2' to the new label 81789Sahrens * 82789Sahrens * Given arbitrary failure, we can determine the correct label to use based on 83789Sahrens * the transaction group. If we fail after updating L1 but before updating the 84789Sahrens * UB, we will notice that L1's transaction group is greater than the uberblock, 85789Sahrens * so L2 must be valid. If we fail after writing the uberblock but before 86789Sahrens * writing L2, we will notice that L2's transaction group is less than L1, and 87789Sahrens * therefore L1 is valid. 88789Sahrens * 89789Sahrens * Another added complexity is that not every label is updated when the config 90789Sahrens * is synced. If we add a single device, we do not want to have to re-write 91789Sahrens * every label for every device in the pool. This means that both L1 and L2 may 92789Sahrens * be older than the pool uberblock, because the necessary information is stored 93789Sahrens * on another vdev. 94789Sahrens * 95789Sahrens * 96789Sahrens * On-disk Format 97789Sahrens * -------------- 98789Sahrens * 99789Sahrens * The vdev label consists of two distinct parts, and is wrapped within the 100789Sahrens * vdev_label_t structure. The label includes 8k of padding to permit legacy 101789Sahrens * VTOC disk labels, but is otherwise ignored. 102789Sahrens * 103789Sahrens * The first half of the label is a packed nvlist which contains pool wide 104789Sahrens * properties, per-vdev properties, and configuration information. It is 105789Sahrens * described in more detail below. 106789Sahrens * 107789Sahrens * The latter half of the label consists of a redundant array of uberblocks. 108789Sahrens * These uberblocks are updated whenever a transaction group is committed, 109789Sahrens * or when the configuration is updated. When a pool is loaded, we scan each 110789Sahrens * vdev for the 'best' uberblock. 111789Sahrens * 112789Sahrens * 113789Sahrens * Configuration Information 114789Sahrens * ------------------------- 115789Sahrens * 116789Sahrens * The nvlist describing the pool and vdev contains the following elements: 117789Sahrens * 118789Sahrens * version ZFS on-disk version 119789Sahrens * name Pool name 120789Sahrens * state Pool state 121789Sahrens * txg Transaction group in which this label was written 122789Sahrens * pool_guid Unique identifier for this pool 123789Sahrens * vdev_tree An nvlist describing vdev tree. 124789Sahrens * 125789Sahrens * Each leaf device label also contains the following: 126789Sahrens * 127789Sahrens * top_guid Unique ID for top-level vdev in which this is contained 128789Sahrens * guid Unique ID for the leaf vdev 129789Sahrens * 130789Sahrens * The 'vs' configuration follows the format described in 'spa_config.c'. 131789Sahrens */ 132789Sahrens 133789Sahrens #include <sys/zfs_context.h> 134789Sahrens #include <sys/spa.h> 135789Sahrens #include <sys/spa_impl.h> 136789Sahrens #include <sys/dmu.h> 137789Sahrens #include <sys/zap.h> 138789Sahrens #include <sys/vdev.h> 139789Sahrens #include <sys/vdev_impl.h> 140789Sahrens #include <sys/uberblock_impl.h> 141789Sahrens #include <sys/metaslab.h> 142789Sahrens #include <sys/zio.h> 143*12296SLin.Ling@Sun.COM #include <sys/dsl_scan.h> 144789Sahrens #include <sys/fs/zfs.h> 145789Sahrens 146789Sahrens /* 147789Sahrens * Basic routines to read and write from a vdev label. 148789Sahrens * Used throughout the rest of this file. 149789Sahrens */ 150789Sahrens uint64_t 151789Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset) 152789Sahrens { 1531732Sbonwick ASSERT(offset < sizeof (vdev_label_t)); 1544577Sahrens ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); 1551732Sbonwick 156789Sahrens return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 157789Sahrens 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); 158789Sahrens } 159789Sahrens 1606615Sgw25295 /* 1616615Sgw25295 * Returns back the vdev label associated with the passed in offset. 1626615Sgw25295 */ 1636615Sgw25295 int 1646615Sgw25295 vdev_label_number(uint64_t psize, uint64_t offset) 1656615Sgw25295 { 1666615Sgw25295 int l; 1676615Sgw25295 1686615Sgw25295 if (offset >= psize - VDEV_LABEL_END_SIZE) { 1696615Sgw25295 offset -= psize - VDEV_LABEL_END_SIZE; 1706615Sgw25295 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); 1716615Sgw25295 } 1726615Sgw25295 l = offset / sizeof (vdev_label_t); 1736615Sgw25295 return (l < VDEV_LABELS ? l : -1); 1746615Sgw25295 } 1756615Sgw25295 176789Sahrens static void 177789Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 1787754SJeff.Bonwick@Sun.COM uint64_t size, zio_done_func_t *done, void *private, int flags) 179789Sahrens { 1807754SJeff.Bonwick@Sun.COM ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == 1817754SJeff.Bonwick@Sun.COM SCL_STATE_ALL); 1827754SJeff.Bonwick@Sun.COM ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 183789Sahrens 184789Sahrens zio_nowait(zio_read_phys(zio, vd, 185789Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 186789Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 1877754SJeff.Bonwick@Sun.COM ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); 188789Sahrens } 189789Sahrens 190789Sahrens static void 191789Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, 1925688Sbonwick uint64_t size, zio_done_func_t *done, void *private, int flags) 193789Sahrens { 1947754SJeff.Bonwick@Sun.COM ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || 1957754SJeff.Bonwick@Sun.COM (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == 1967754SJeff.Bonwick@Sun.COM (SCL_CONFIG | SCL_STATE) && 1977754SJeff.Bonwick@Sun.COM dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); 1987754SJeff.Bonwick@Sun.COM ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); 199789Sahrens 200789Sahrens zio_nowait(zio_write_phys(zio, vd, 201789Sahrens vdev_label_offset(vd->vdev_psize, l, offset), 202789Sahrens size, buf, ZIO_CHECKSUM_LABEL, done, private, 2035688Sbonwick ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); 204789Sahrens } 205789Sahrens 206789Sahrens /* 207789Sahrens * Generate the nvlist representing this vdev's config. 208789Sahrens */ 209789Sahrens nvlist_t * 2102082Seschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, 211*12296SLin.Ling@Sun.COM vdev_config_flag_t flags) 212789Sahrens { 213789Sahrens nvlist_t *nv = NULL; 214789Sahrens 2151544Seschrock VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 216789Sahrens 217789Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 218789Sahrens vd->vdev_ops->vdev_op_type) == 0); 219*12296SLin.Ling@Sun.COM if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) 2202082Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) 2212082Seschrock == 0); 222789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); 223789Sahrens 224789Sahrens if (vd->vdev_path != NULL) 225789Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, 226789Sahrens vd->vdev_path) == 0); 227789Sahrens 228789Sahrens if (vd->vdev_devid != NULL) 229789Sahrens VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, 230789Sahrens vd->vdev_devid) == 0); 231789Sahrens 2324451Seschrock if (vd->vdev_physpath != NULL) 2334451Seschrock VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, 2344451Seschrock vd->vdev_physpath) == 0); 2354451Seschrock 2369425SEric.Schrock@Sun.COM if (vd->vdev_fru != NULL) 2379425SEric.Schrock@Sun.COM VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU, 2389425SEric.Schrock@Sun.COM vd->vdev_fru) == 0); 2399425SEric.Schrock@Sun.COM 2402082Seschrock if (vd->vdev_nparity != 0) { 2412082Seschrock ASSERT(strcmp(vd->vdev_ops->vdev_op_type, 2422082Seschrock VDEV_TYPE_RAIDZ) == 0); 2432082Seschrock 2442082Seschrock /* 2452082Seschrock * Make sure someone hasn't managed to sneak a fancy new vdev 2462082Seschrock * into a crufty old storage pool. 2472082Seschrock */ 2482082Seschrock ASSERT(vd->vdev_nparity == 1 || 24910105Sadam.leventhal@sun.com (vd->vdev_nparity <= 2 && 25010105Sadam.leventhal@sun.com spa_version(spa) >= SPA_VERSION_RAIDZ2) || 25110105Sadam.leventhal@sun.com (vd->vdev_nparity <= 3 && 25210105Sadam.leventhal@sun.com spa_version(spa) >= SPA_VERSION_RAIDZ3)); 2532082Seschrock 2542082Seschrock /* 2552082Seschrock * Note that we'll add the nparity tag even on storage pools 2562082Seschrock * that only support a single parity device -- older software 2572082Seschrock * will just ignore it. 2582082Seschrock */ 2592082Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, 2602082Seschrock vd->vdev_nparity) == 0); 2612082Seschrock } 2622082Seschrock 2631171Seschrock if (vd->vdev_wholedisk != -1ULL) 2641171Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 2651171Seschrock vd->vdev_wholedisk) == 0); 2661171Seschrock 2671544Seschrock if (vd->vdev_not_present) 2681544Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0); 2691544Seschrock 2702082Seschrock if (vd->vdev_isspare) 2712082Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0); 2722082Seschrock 273*12296SLin.Ling@Sun.COM if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && 274*12296SLin.Ling@Sun.COM vd == vd->vdev_top) { 275789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 276789Sahrens vd->vdev_ms_array) == 0); 277789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 278789Sahrens vd->vdev_ms_shift) == 0); 279789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, 280789Sahrens vd->vdev_ashift) == 0); 281789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, 282789Sahrens vd->vdev_asize) == 0); 2834527Sperrin VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, 2844527Sperrin vd->vdev_islog) == 0); 285*12296SLin.Ling@Sun.COM if (vd->vdev_removing) 286*12296SLin.Ling@Sun.COM VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, 287*12296SLin.Ling@Sun.COM vd->vdev_removing) == 0); 288789Sahrens } 289789Sahrens 2908241SJeff.Bonwick@Sun.COM if (vd->vdev_dtl_smo.smo_object != 0) 291789Sahrens VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, 2928241SJeff.Bonwick@Sun.COM vd->vdev_dtl_smo.smo_object) == 0); 293789Sahrens 29410594SGeorge.Wilson@Sun.COM if (vd->vdev_crtxg) 29510594SGeorge.Wilson@Sun.COM VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 29610594SGeorge.Wilson@Sun.COM vd->vdev_crtxg) == 0); 29710594SGeorge.Wilson@Sun.COM 298789Sahrens if (getstats) { 299789Sahrens vdev_stat_t vs; 300*12296SLin.Ling@Sun.COM pool_scan_stat_t ps; 301*12296SLin.Ling@Sun.COM 302789Sahrens vdev_get_stats(vd, &vs); 303*12296SLin.Ling@Sun.COM VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, 304789Sahrens (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0); 305*12296SLin.Ling@Sun.COM 306*12296SLin.Ling@Sun.COM /* provide either current or previous scan information */ 307*12296SLin.Ling@Sun.COM if (spa_scan_get_stats(spa, &ps) == 0) { 308*12296SLin.Ling@Sun.COM VERIFY(nvlist_add_uint64_array(nv, 309*12296SLin.Ling@Sun.COM ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, 310*12296SLin.Ling@Sun.COM sizeof (pool_scan_stat_t) / sizeof (uint64_t)) 311*12296SLin.Ling@Sun.COM == 0); 312*12296SLin.Ling@Sun.COM } 313789Sahrens } 314789Sahrens 315789Sahrens if (!vd->vdev_ops->vdev_op_leaf) { 316789Sahrens nvlist_t **child; 317*12296SLin.Ling@Sun.COM int c, idx; 318789Sahrens 31910594SGeorge.Wilson@Sun.COM ASSERT(!vd->vdev_ishole); 32010594SGeorge.Wilson@Sun.COM 321789Sahrens child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), 322789Sahrens KM_SLEEP); 323789Sahrens 324*12296SLin.Ling@Sun.COM for (c = 0, idx = 0; c < vd->vdev_children; c++) { 325*12296SLin.Ling@Sun.COM vdev_t *cvd = vd->vdev_child[c]; 326789Sahrens 327*12296SLin.Ling@Sun.COM /* 328*12296SLin.Ling@Sun.COM * If we're generating an nvlist of removing 329*12296SLin.Ling@Sun.COM * vdevs then skip over any device which is 330*12296SLin.Ling@Sun.COM * not being removed. 331*12296SLin.Ling@Sun.COM */ 332*12296SLin.Ling@Sun.COM if ((flags & VDEV_CONFIG_REMOVING) && 333*12296SLin.Ling@Sun.COM !cvd->vdev_removing) 334*12296SLin.Ling@Sun.COM continue; 335789Sahrens 336*12296SLin.Ling@Sun.COM child[idx++] = vdev_config_generate(spa, cvd, 337*12296SLin.Ling@Sun.COM getstats, flags); 338*12296SLin.Ling@Sun.COM } 339*12296SLin.Ling@Sun.COM 340*12296SLin.Ling@Sun.COM if (idx) { 341*12296SLin.Ling@Sun.COM VERIFY(nvlist_add_nvlist_array(nv, 342*12296SLin.Ling@Sun.COM ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 343*12296SLin.Ling@Sun.COM } 344*12296SLin.Ling@Sun.COM 345*12296SLin.Ling@Sun.COM for (c = 0; c < idx; c++) 346789Sahrens nvlist_free(child[c]); 347789Sahrens 348789Sahrens kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); 3491485Slling 3501485Slling } else { 35110817SEric.Schrock@Sun.COM const char *aux = NULL; 35210817SEric.Schrock@Sun.COM 3531732Sbonwick if (vd->vdev_offline && !vd->vdev_tmpoffline) 3541485Slling VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, 3551732Sbonwick B_TRUE) == 0); 3564451Seschrock if (vd->vdev_faulted) 3574451Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, 3584451Seschrock B_TRUE) == 0); 3594451Seschrock if (vd->vdev_degraded) 3604451Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, 3614451Seschrock B_TRUE) == 0); 3624451Seschrock if (vd->vdev_removed) 3634451Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, 3644451Seschrock B_TRUE) == 0); 3654451Seschrock if (vd->vdev_unspare) 3664451Seschrock VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, 3674451Seschrock B_TRUE) == 0); 36810594SGeorge.Wilson@Sun.COM if (vd->vdev_ishole) 36910594SGeorge.Wilson@Sun.COM VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, 37010594SGeorge.Wilson@Sun.COM B_TRUE) == 0); 37110817SEric.Schrock@Sun.COM 37210817SEric.Schrock@Sun.COM switch (vd->vdev_stat.vs_aux) { 37310817SEric.Schrock@Sun.COM case VDEV_AUX_ERR_EXCEEDED: 37410817SEric.Schrock@Sun.COM aux = "err_exceeded"; 37510817SEric.Schrock@Sun.COM break; 37610817SEric.Schrock@Sun.COM 37710817SEric.Schrock@Sun.COM case VDEV_AUX_EXTERNAL: 37810817SEric.Schrock@Sun.COM aux = "external"; 37910817SEric.Schrock@Sun.COM break; 38010817SEric.Schrock@Sun.COM } 38110817SEric.Schrock@Sun.COM 38210817SEric.Schrock@Sun.COM if (aux != NULL) 38310817SEric.Schrock@Sun.COM VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, 38410817SEric.Schrock@Sun.COM aux) == 0); 38511422SMark.Musante@Sun.COM 38611422SMark.Musante@Sun.COM if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { 38711422SMark.Musante@Sun.COM VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, 38811422SMark.Musante@Sun.COM vd->vdev_orig_guid) == 0); 38911422SMark.Musante@Sun.COM } 390789Sahrens } 391789Sahrens 392789Sahrens return (nv); 393789Sahrens } 394789Sahrens 39510594SGeorge.Wilson@Sun.COM /* 39610594SGeorge.Wilson@Sun.COM * Generate a view of the top-level vdevs. If we currently have holes 39710594SGeorge.Wilson@Sun.COM * in the namespace, then generate an array which contains a list of holey 39810594SGeorge.Wilson@Sun.COM * vdevs. Additionally, add the number of top-level children that currently 39910594SGeorge.Wilson@Sun.COM * exist. 40010594SGeorge.Wilson@Sun.COM */ 40110594SGeorge.Wilson@Sun.COM void 40210594SGeorge.Wilson@Sun.COM vdev_top_config_generate(spa_t *spa, nvlist_t *config) 40310594SGeorge.Wilson@Sun.COM { 40410594SGeorge.Wilson@Sun.COM vdev_t *rvd = spa->spa_root_vdev; 40510594SGeorge.Wilson@Sun.COM uint64_t *array; 406*12296SLin.Ling@Sun.COM uint_t c, idx; 40710594SGeorge.Wilson@Sun.COM 40810594SGeorge.Wilson@Sun.COM array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); 40910594SGeorge.Wilson@Sun.COM 410*12296SLin.Ling@Sun.COM for (c = 0, idx = 0; c < rvd->vdev_children; c++) { 41110594SGeorge.Wilson@Sun.COM vdev_t *tvd = rvd->vdev_child[c]; 41210594SGeorge.Wilson@Sun.COM 41310594SGeorge.Wilson@Sun.COM if (tvd->vdev_ishole) 41410594SGeorge.Wilson@Sun.COM array[idx++] = c; 41510594SGeorge.Wilson@Sun.COM } 41610594SGeorge.Wilson@Sun.COM 41710666SGeorge.Wilson@Sun.COM if (idx) { 41810666SGeorge.Wilson@Sun.COM VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, 41910666SGeorge.Wilson@Sun.COM array, idx) == 0); 42010666SGeorge.Wilson@Sun.COM } 42110666SGeorge.Wilson@Sun.COM 42210594SGeorge.Wilson@Sun.COM VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 42310594SGeorge.Wilson@Sun.COM rvd->vdev_children) == 0); 42410594SGeorge.Wilson@Sun.COM 42510594SGeorge.Wilson@Sun.COM kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); 42610594SGeorge.Wilson@Sun.COM } 42710594SGeorge.Wilson@Sun.COM 428789Sahrens nvlist_t * 429789Sahrens vdev_label_read_config(vdev_t *vd) 430789Sahrens { 4311635Sbonwick spa_t *spa = vd->vdev_spa; 432789Sahrens nvlist_t *config = NULL; 433789Sahrens vdev_phys_t *vp; 434789Sahrens zio_t *zio; 4359725SEric.Schrock@Sun.COM int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 4369725SEric.Schrock@Sun.COM ZIO_FLAG_SPECULATIVE; 437789Sahrens 4387754SJeff.Bonwick@Sun.COM ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 4391635Sbonwick 4405329Sgw25295 if (!vdev_readable(vd)) 441789Sahrens return (NULL); 442789Sahrens 443789Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 444789Sahrens 4459725SEric.Schrock@Sun.COM retry: 4467754SJeff.Bonwick@Sun.COM for (int l = 0; l < VDEV_LABELS; l++) { 447789Sahrens 4487754SJeff.Bonwick@Sun.COM zio = zio_root(spa, NULL, NULL, flags); 449789Sahrens 450789Sahrens vdev_label_read(zio, vd, l, vp, 451789Sahrens offsetof(vdev_label_t, vl_vdev_phys), 4527754SJeff.Bonwick@Sun.COM sizeof (vdev_phys_t), NULL, NULL, flags); 453789Sahrens 454789Sahrens if (zio_wait(zio) == 0 && 455789Sahrens nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), 4561544Seschrock &config, 0) == 0) 457789Sahrens break; 458789Sahrens 459789Sahrens if (config != NULL) { 460789Sahrens nvlist_free(config); 461789Sahrens config = NULL; 462789Sahrens } 463789Sahrens } 464789Sahrens 4659725SEric.Schrock@Sun.COM if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { 4669725SEric.Schrock@Sun.COM flags |= ZIO_FLAG_TRYHARD; 4679725SEric.Schrock@Sun.COM goto retry; 4689725SEric.Schrock@Sun.COM } 4699725SEric.Schrock@Sun.COM 470789Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 471789Sahrens 472789Sahrens return (config); 473789Sahrens } 474789Sahrens 4753377Seschrock /* 4763377Seschrock * Determine if a device is in use. The 'spare_guid' parameter will be filled 4773377Seschrock * in with the device guid if this spare is active elsewhere on the system. 4783377Seschrock */ 4793377Seschrock static boolean_t 4803377Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, 4815450Sbrendan uint64_t *spare_guid, uint64_t *l2cache_guid) 4823377Seschrock { 4833377Seschrock spa_t *spa = vd->vdev_spa; 4843377Seschrock uint64_t state, pool_guid, device_guid, txg, spare_pool; 4853377Seschrock uint64_t vdtxg = 0; 4863377Seschrock nvlist_t *label; 4873377Seschrock 4883377Seschrock if (spare_guid) 4893377Seschrock *spare_guid = 0ULL; 4905450Sbrendan if (l2cache_guid) 4915450Sbrendan *l2cache_guid = 0ULL; 4923377Seschrock 4933377Seschrock /* 4943377Seschrock * Read the label, if any, and perform some basic sanity checks. 4953377Seschrock */ 4963377Seschrock if ((label = vdev_label_read_config(vd)) == NULL) 4973377Seschrock return (B_FALSE); 4983377Seschrock 4993377Seschrock (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 5003377Seschrock &vdtxg); 5013377Seschrock 5023377Seschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 5033377Seschrock &state) != 0 || 5043377Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 5053377Seschrock &device_guid) != 0) { 5063377Seschrock nvlist_free(label); 5073377Seschrock return (B_FALSE); 5083377Seschrock } 5093377Seschrock 5105450Sbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 5113377Seschrock (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, 5123377Seschrock &pool_guid) != 0 || 5133377Seschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 5143377Seschrock &txg) != 0)) { 5153377Seschrock nvlist_free(label); 5163377Seschrock return (B_FALSE); 5173377Seschrock } 5183377Seschrock 5193377Seschrock nvlist_free(label); 5203377Seschrock 5213377Seschrock /* 5223377Seschrock * Check to see if this device indeed belongs to the pool it claims to 5233377Seschrock * be a part of. The only way this is allowed is if the device is a hot 5243377Seschrock * spare (which we check for later on). 5253377Seschrock */ 5265450Sbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 5273377Seschrock !spa_guid_exists(pool_guid, device_guid) && 5287214Slling !spa_spare_exists(device_guid, NULL, NULL) && 5295450Sbrendan !spa_l2cache_exists(device_guid, NULL)) 5303377Seschrock return (B_FALSE); 5313377Seschrock 5323377Seschrock /* 5333377Seschrock * If the transaction group is zero, then this an initialized (but 5343377Seschrock * unused) label. This is only an error if the create transaction 5353377Seschrock * on-disk is the same as the one we're using now, in which case the 5363377Seschrock * user has attempted to add the same vdev multiple times in the same 5373377Seschrock * transaction. 5383377Seschrock */ 5395450Sbrendan if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && 5405450Sbrendan txg == 0 && vdtxg == crtxg) 5413377Seschrock return (B_TRUE); 5423377Seschrock 5433377Seschrock /* 5443377Seschrock * Check to see if this is a spare device. We do an explicit check for 5453377Seschrock * spa_has_spare() here because it may be on our pending list of spares 5465450Sbrendan * to add. We also check if it is an l2cache device. 5473377Seschrock */ 5487214Slling if (spa_spare_exists(device_guid, &spare_pool, NULL) || 5493377Seschrock spa_has_spare(spa, device_guid)) { 5503377Seschrock if (spare_guid) 5513377Seschrock *spare_guid = device_guid; 5523377Seschrock 5533377Seschrock switch (reason) { 5543377Seschrock case VDEV_LABEL_CREATE: 5555450Sbrendan case VDEV_LABEL_L2CACHE: 5563377Seschrock return (B_TRUE); 5573377Seschrock 5583377Seschrock case VDEV_LABEL_REPLACE: 5593377Seschrock return (!spa_has_spare(spa, device_guid) || 5603377Seschrock spare_pool != 0ULL); 5613377Seschrock 5623377Seschrock case VDEV_LABEL_SPARE: 5633377Seschrock return (spa_has_spare(spa, device_guid)); 5643377Seschrock } 5653377Seschrock } 5663377Seschrock 5673377Seschrock /* 5685450Sbrendan * Check to see if this is an l2cache device. 5695450Sbrendan */ 5705450Sbrendan if (spa_l2cache_exists(device_guid, NULL)) 5715450Sbrendan return (B_TRUE); 5725450Sbrendan 5735450Sbrendan /* 5743377Seschrock * If the device is marked ACTIVE, then this device is in use by another 5753377Seschrock * pool on the system. 5763377Seschrock */ 5773377Seschrock return (state == POOL_STATE_ACTIVE); 5783377Seschrock } 5793377Seschrock 5803377Seschrock /* 5813377Seschrock * Initialize a vdev label. We check to make sure each leaf device is not in 5823377Seschrock * use, and writable. We put down an initial label which we will later 5833377Seschrock * overwrite with a complete label. Note that it's important to do this 5843377Seschrock * sequentially, not in parallel, so that we catch cases of multiple use of the 5853377Seschrock * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with 5863377Seschrock * itself. 5873377Seschrock */ 5883377Seschrock int 5893377Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) 590789Sahrens { 591789Sahrens spa_t *spa = vd->vdev_spa; 592789Sahrens nvlist_t *label; 593789Sahrens vdev_phys_t *vp; 5949056SLin.Ling@Sun.COM char *pad2; 5951732Sbonwick uberblock_t *ub; 596789Sahrens zio_t *zio; 597789Sahrens char *buf; 598789Sahrens size_t buflen; 599789Sahrens int error; 6005450Sbrendan uint64_t spare_guid, l2cache_guid; 6017754SJeff.Bonwick@Sun.COM int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 602789Sahrens 6037754SJeff.Bonwick@Sun.COM ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 6041635Sbonwick 6057754SJeff.Bonwick@Sun.COM for (int c = 0; c < vd->vdev_children; c++) 6063377Seschrock if ((error = vdev_label_init(vd->vdev_child[c], 6073377Seschrock crtxg, reason)) != 0) 608789Sahrens return (error); 609789Sahrens 61010594SGeorge.Wilson@Sun.COM /* Track the creation time for this vdev */ 61110594SGeorge.Wilson@Sun.COM vd->vdev_crtxg = crtxg; 61210594SGeorge.Wilson@Sun.COM 613789Sahrens if (!vd->vdev_ops->vdev_op_leaf) 614789Sahrens return (0); 615789Sahrens 616789Sahrens /* 6173377Seschrock * Dead vdevs cannot be initialized. 618789Sahrens */ 619789Sahrens if (vdev_is_dead(vd)) 620789Sahrens return (EIO); 621789Sahrens 622789Sahrens /* 6233377Seschrock * Determine if the vdev is in use. 624789Sahrens */ 62511422SMark.Musante@Sun.COM if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && 6265450Sbrendan vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) 6273377Seschrock return (EBUSY); 628789Sahrens 6293377Seschrock /* 6305450Sbrendan * If this is a request to add or replace a spare or l2cache device 6315450Sbrendan * that is in use elsewhere on the system, then we must update the 6325450Sbrendan * guid (which was initialized to a random value) to reflect the 6335450Sbrendan * actual GUID (which is shared between multiple pools). 6343377Seschrock */ 6355450Sbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && 6365450Sbrendan spare_guid != 0ULL) { 6377768SJeff.Bonwick@Sun.COM uint64_t guid_delta = spare_guid - vd->vdev_guid; 6382082Seschrock 6397768SJeff.Bonwick@Sun.COM vd->vdev_guid += guid_delta; 6407768SJeff.Bonwick@Sun.COM 6417768SJeff.Bonwick@Sun.COM for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 6427768SJeff.Bonwick@Sun.COM pvd->vdev_guid_sum += guid_delta; 6432082Seschrock 6443377Seschrock /* 6453377Seschrock * If this is a replacement, then we want to fallthrough to the 6463377Seschrock * rest of the code. If we're adding a spare, then it's already 6474451Seschrock * labeled appropriately and we can just return. 6483377Seschrock */ 6493377Seschrock if (reason == VDEV_LABEL_SPARE) 6503377Seschrock return (0); 65111422SMark.Musante@Sun.COM ASSERT(reason == VDEV_LABEL_REPLACE || 65211422SMark.Musante@Sun.COM reason == VDEV_LABEL_SPLIT); 653789Sahrens } 654789Sahrens 6555450Sbrendan if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && 6565450Sbrendan l2cache_guid != 0ULL) { 6577768SJeff.Bonwick@Sun.COM uint64_t guid_delta = l2cache_guid - vd->vdev_guid; 6585450Sbrendan 6597768SJeff.Bonwick@Sun.COM vd->vdev_guid += guid_delta; 6607768SJeff.Bonwick@Sun.COM 6617768SJeff.Bonwick@Sun.COM for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) 6627768SJeff.Bonwick@Sun.COM pvd->vdev_guid_sum += guid_delta; 6635450Sbrendan 6645450Sbrendan /* 6655450Sbrendan * If this is a replacement, then we want to fallthrough to the 6665450Sbrendan * rest of the code. If we're adding an l2cache, then it's 6675450Sbrendan * already labeled appropriately and we can just return. 6685450Sbrendan */ 6695450Sbrendan if (reason == VDEV_LABEL_L2CACHE) 6705450Sbrendan return (0); 6715450Sbrendan ASSERT(reason == VDEV_LABEL_REPLACE); 6725450Sbrendan } 6735450Sbrendan 674789Sahrens /* 6753377Seschrock * Initialize its label. 676789Sahrens */ 677789Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 678789Sahrens bzero(vp, sizeof (vdev_phys_t)); 679789Sahrens 680789Sahrens /* 681789Sahrens * Generate a label describing the pool and our top-level vdev. 682789Sahrens * We mark it as being from txg 0 to indicate that it's not 683789Sahrens * really part of an active pool just yet. The labels will 684789Sahrens * be written again with a meaningful txg by spa_sync(). 685789Sahrens */ 6863377Seschrock if (reason == VDEV_LABEL_SPARE || 6873377Seschrock (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { 6883377Seschrock /* 6893377Seschrock * For inactive hot spares, we generate a special label that 6903377Seschrock * identifies as a mutually shared hot spare. We write the 6913377Seschrock * label if we are adding a hot spare, or if we are removing an 6923377Seschrock * active hot spare (in which case we want to revert the 6933377Seschrock * labels). 6943377Seschrock */ 6952082Seschrock VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 696789Sahrens 6972082Seschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 6982082Seschrock spa_version(spa)) == 0); 6992082Seschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 7002082Seschrock POOL_STATE_SPARE) == 0); 7012082Seschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 7022082Seschrock vd->vdev_guid) == 0); 7035450Sbrendan } else if (reason == VDEV_LABEL_L2CACHE || 7045450Sbrendan (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { 7055450Sbrendan /* 7065450Sbrendan * For level 2 ARC devices, add a special label. 7075450Sbrendan */ 7085450Sbrendan VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); 7095450Sbrendan 7105450Sbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, 7115450Sbrendan spa_version(spa)) == 0); 7125450Sbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, 7135450Sbrendan POOL_STATE_L2CACHE) == 0); 7145450Sbrendan VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, 7155450Sbrendan vd->vdev_guid) == 0); 7162082Seschrock } else { 71711422SMark.Musante@Sun.COM uint64_t txg = 0ULL; 71811422SMark.Musante@Sun.COM 71911422SMark.Musante@Sun.COM if (reason == VDEV_LABEL_SPLIT) 72011422SMark.Musante@Sun.COM txg = spa->spa_uberblock.ub_txg; 72111422SMark.Musante@Sun.COM label = spa_config_generate(spa, vd, txg, B_FALSE); 7222082Seschrock 7232082Seschrock /* 7242082Seschrock * Add our creation time. This allows us to detect multiple 7252082Seschrock * vdev uses as described above, and automatically expires if we 7262082Seschrock * fail. 7272082Seschrock */ 7282082Seschrock VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, 7292082Seschrock crtxg) == 0); 7302082Seschrock } 731789Sahrens 732789Sahrens buf = vp->vp_nvlist; 733789Sahrens buflen = sizeof (vp->vp_nvlist); 734789Sahrens 7353460Smmusante error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); 7363460Smmusante if (error != 0) { 737789Sahrens nvlist_free(label); 738789Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 7393460Smmusante /* EFAULT means nvlist_pack ran out of room */ 7403460Smmusante return (error == EFAULT ? ENAMETOOLONG : EINVAL); 741789Sahrens } 742789Sahrens 743789Sahrens /* 744789Sahrens * Initialize uberblock template. 745789Sahrens */ 7469846SEric.Taylor@Sun.COM ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); 7479846SEric.Taylor@Sun.COM bzero(ub, VDEV_UBERBLOCK_RING); 7481732Sbonwick *ub = spa->spa_uberblock; 7491732Sbonwick ub->ub_txg = 0; 750789Sahrens 7519056SLin.Ling@Sun.COM /* Initialize the 2nd padding area. */ 7529056SLin.Ling@Sun.COM pad2 = zio_buf_alloc(VDEV_PAD_SIZE); 7539056SLin.Ling@Sun.COM bzero(pad2, VDEV_PAD_SIZE); 7549056SLin.Ling@Sun.COM 755789Sahrens /* 756789Sahrens * Write everything in parallel. 757789Sahrens */ 7589725SEric.Schrock@Sun.COM retry: 7595688Sbonwick zio = zio_root(spa, NULL, NULL, flags); 760789Sahrens 7617754SJeff.Bonwick@Sun.COM for (int l = 0; l < VDEV_LABELS; l++) { 762789Sahrens 763789Sahrens vdev_label_write(zio, vd, l, vp, 764789Sahrens offsetof(vdev_label_t, vl_vdev_phys), 7655688Sbonwick sizeof (vdev_phys_t), NULL, NULL, flags); 766789Sahrens 7679056SLin.Ling@Sun.COM /* 7689056SLin.Ling@Sun.COM * Skip the 1st padding area. 7699056SLin.Ling@Sun.COM * Zero out the 2nd padding area where it might have 7709056SLin.Ling@Sun.COM * left over data from previous filesystem format. 7719056SLin.Ling@Sun.COM */ 7729056SLin.Ling@Sun.COM vdev_label_write(zio, vd, l, pad2, 7739056SLin.Ling@Sun.COM offsetof(vdev_label_t, vl_pad2), 7749056SLin.Ling@Sun.COM VDEV_PAD_SIZE, NULL, NULL, flags); 7759056SLin.Ling@Sun.COM 7769846SEric.Taylor@Sun.COM vdev_label_write(zio, vd, l, ub, 7779846SEric.Taylor@Sun.COM offsetof(vdev_label_t, vl_uberblock), 7789846SEric.Taylor@Sun.COM VDEV_UBERBLOCK_RING, NULL, NULL, flags); 779789Sahrens } 780789Sahrens 781789Sahrens error = zio_wait(zio); 782789Sahrens 7839725SEric.Schrock@Sun.COM if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { 7849725SEric.Schrock@Sun.COM flags |= ZIO_FLAG_TRYHARD; 7859725SEric.Schrock@Sun.COM goto retry; 7869725SEric.Schrock@Sun.COM } 7879725SEric.Schrock@Sun.COM 788789Sahrens nvlist_free(label); 7899056SLin.Ling@Sun.COM zio_buf_free(pad2, VDEV_PAD_SIZE); 7909846SEric.Taylor@Sun.COM zio_buf_free(ub, VDEV_UBERBLOCK_RING); 791789Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 792789Sahrens 7933377Seschrock /* 7943377Seschrock * If this vdev hasn't been previously identified as a spare, then we 7954451Seschrock * mark it as such only if a) we are labeling it as a spare, or b) it 7965450Sbrendan * exists as a spare elsewhere in the system. Do the same for 7975450Sbrendan * level 2 ARC devices. 7983377Seschrock */ 7993377Seschrock if (error == 0 && !vd->vdev_isspare && 8003377Seschrock (reason == VDEV_LABEL_SPARE || 8017214Slling spa_spare_exists(vd->vdev_guid, NULL, NULL))) 8023377Seschrock spa_spare_add(vd); 8032082Seschrock 8045450Sbrendan if (error == 0 && !vd->vdev_isl2cache && 8055450Sbrendan (reason == VDEV_LABEL_L2CACHE || 8065450Sbrendan spa_l2cache_exists(vd->vdev_guid, NULL))) 8075450Sbrendan spa_l2cache_add(vd); 8085450Sbrendan 8093377Seschrock return (error); 8102082Seschrock } 8112082Seschrock 812789Sahrens /* 813789Sahrens * ========================================================================== 814789Sahrens * uberblock load/sync 815789Sahrens * ========================================================================== 816789Sahrens */ 817789Sahrens 818789Sahrens /* 819789Sahrens * Consider the following situation: txg is safely synced to disk. We've 820789Sahrens * written the first uberblock for txg + 1, and then we lose power. When we 821789Sahrens * come back up, we fail to see the uberblock for txg + 1 because, say, 822789Sahrens * it was on a mirrored device and the replica to which we wrote txg + 1 823789Sahrens * is now offline. If we then make some changes and sync txg + 1, and then 824789Sahrens * the missing replica comes back, then for a new seconds we'll have two 825789Sahrens * conflicting uberblocks on disk with the same txg. The solution is simple: 826789Sahrens * among uberblocks with equal txg, choose the one with the latest timestamp. 827789Sahrens */ 828789Sahrens static int 829789Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) 830789Sahrens { 831789Sahrens if (ub1->ub_txg < ub2->ub_txg) 832789Sahrens return (-1); 833789Sahrens if (ub1->ub_txg > ub2->ub_txg) 834789Sahrens return (1); 835789Sahrens 836789Sahrens if (ub1->ub_timestamp < ub2->ub_timestamp) 837789Sahrens return (-1); 838789Sahrens if (ub1->ub_timestamp > ub2->ub_timestamp) 839789Sahrens return (1); 840789Sahrens 841789Sahrens return (0); 842789Sahrens } 843789Sahrens 844789Sahrens static void 845789Sahrens vdev_uberblock_load_done(zio_t *zio) 846789Sahrens { 84710921STim.Haley@Sun.COM spa_t *spa = zio->io_spa; 8487754SJeff.Bonwick@Sun.COM zio_t *rio = zio->io_private; 8491732Sbonwick uberblock_t *ub = zio->io_data; 8507754SJeff.Bonwick@Sun.COM uberblock_t *ubbest = rio->io_private; 851789Sahrens 8521732Sbonwick ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd)); 853789Sahrens 8541544Seschrock if (zio->io_error == 0 && uberblock_verify(ub) == 0) { 8557754SJeff.Bonwick@Sun.COM mutex_enter(&rio->io_lock); 85610921STim.Haley@Sun.COM if (ub->ub_txg <= spa->spa_load_max_txg && 8578188SVictor.Latushkin@Sun.COM vdev_uberblock_compare(ub, ubbest) > 0) 858789Sahrens *ubbest = *ub; 8597754SJeff.Bonwick@Sun.COM mutex_exit(&rio->io_lock); 860789Sahrens } 861789Sahrens 862789Sahrens zio_buf_free(zio->io_data, zio->io_size); 863789Sahrens } 864789Sahrens 865789Sahrens void 866789Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest) 867789Sahrens { 8687754SJeff.Bonwick@Sun.COM spa_t *spa = vd->vdev_spa; 8697754SJeff.Bonwick@Sun.COM vdev_t *rvd = spa->spa_root_vdev; 8709725SEric.Schrock@Sun.COM int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | 8719725SEric.Schrock@Sun.COM ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; 872789Sahrens 8737754SJeff.Bonwick@Sun.COM if (vd == rvd) { 8747754SJeff.Bonwick@Sun.COM ASSERT(zio == NULL); 8757754SJeff.Bonwick@Sun.COM spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 8767754SJeff.Bonwick@Sun.COM zio = zio_root(spa, NULL, ubbest, flags); 8777754SJeff.Bonwick@Sun.COM bzero(ubbest, sizeof (uberblock_t)); 8787754SJeff.Bonwick@Sun.COM } 8797754SJeff.Bonwick@Sun.COM 8807754SJeff.Bonwick@Sun.COM ASSERT(zio != NULL); 8817754SJeff.Bonwick@Sun.COM 8827754SJeff.Bonwick@Sun.COM for (int c = 0; c < vd->vdev_children; c++) 883789Sahrens vdev_uberblock_load(zio, vd->vdev_child[c], ubbest); 884789Sahrens 8857754SJeff.Bonwick@Sun.COM if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 8867754SJeff.Bonwick@Sun.COM for (int l = 0; l < VDEV_LABELS; l++) { 8877754SJeff.Bonwick@Sun.COM for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { 8887754SJeff.Bonwick@Sun.COM vdev_label_read(zio, vd, l, 8897754SJeff.Bonwick@Sun.COM zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), 8907754SJeff.Bonwick@Sun.COM VDEV_UBERBLOCK_OFFSET(vd, n), 8917754SJeff.Bonwick@Sun.COM VDEV_UBERBLOCK_SIZE(vd), 8927754SJeff.Bonwick@Sun.COM vdev_uberblock_load_done, zio, flags); 8937754SJeff.Bonwick@Sun.COM } 8947754SJeff.Bonwick@Sun.COM } 8957754SJeff.Bonwick@Sun.COM } 896789Sahrens 8977754SJeff.Bonwick@Sun.COM if (vd == rvd) { 8987754SJeff.Bonwick@Sun.COM (void) zio_wait(zio); 8997754SJeff.Bonwick@Sun.COM spa_config_exit(spa, SCL_ALL, FTAG); 900789Sahrens } 901789Sahrens } 902789Sahrens 903789Sahrens /* 9045688Sbonwick * On success, increment root zio's count of good writes. 9051635Sbonwick * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). 906789Sahrens */ 907789Sahrens static void 908789Sahrens vdev_uberblock_sync_done(zio_t *zio) 909789Sahrens { 9105688Sbonwick uint64_t *good_writes = zio->io_private; 911789Sahrens 9121635Sbonwick if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) 913789Sahrens atomic_add_64(good_writes, 1); 914789Sahrens } 915789Sahrens 9165688Sbonwick /* 9175688Sbonwick * Write the uberblock to all labels of all leaves of the specified vdev. 9185688Sbonwick */ 919789Sahrens static void 9207754SJeff.Bonwick@Sun.COM vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) 921789Sahrens { 9225688Sbonwick uberblock_t *ubbuf; 9237754SJeff.Bonwick@Sun.COM int n; 924789Sahrens 9257754SJeff.Bonwick@Sun.COM for (int c = 0; c < vd->vdev_children; c++) 9267754SJeff.Bonwick@Sun.COM vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); 927789Sahrens 928789Sahrens if (!vd->vdev_ops->vdev_op_leaf) 929789Sahrens return; 930789Sahrens 9317754SJeff.Bonwick@Sun.COM if (!vdev_writeable(vd)) 932789Sahrens return; 933789Sahrens 9345688Sbonwick n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); 935789Sahrens 9365688Sbonwick ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); 9375688Sbonwick bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 9381732Sbonwick *ubbuf = *ub; 939789Sahrens 9407754SJeff.Bonwick@Sun.COM for (int l = 0; l < VDEV_LABELS; l++) 9415688Sbonwick vdev_label_write(zio, vd, l, ubbuf, 9427754SJeff.Bonwick@Sun.COM VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), 9435688Sbonwick vdev_uberblock_sync_done, zio->io_private, 9447754SJeff.Bonwick@Sun.COM flags | ZIO_FLAG_DONT_PROPAGATE); 945789Sahrens 9465688Sbonwick zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); 9475688Sbonwick } 948789Sahrens 9495688Sbonwick int 9505688Sbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) 9515688Sbonwick { 9525688Sbonwick spa_t *spa = svd[0]->vdev_spa; 9537754SJeff.Bonwick@Sun.COM zio_t *zio; 9545688Sbonwick uint64_t good_writes = 0; 955789Sahrens 9567754SJeff.Bonwick@Sun.COM zio = zio_root(spa, NULL, &good_writes, flags); 9577754SJeff.Bonwick@Sun.COM 9587754SJeff.Bonwick@Sun.COM for (int v = 0; v < svdcount; v++) 9597754SJeff.Bonwick@Sun.COM vdev_uberblock_sync(zio, ub, svd[v], flags); 9607754SJeff.Bonwick@Sun.COM 9615688Sbonwick (void) zio_wait(zio); 962789Sahrens 963789Sahrens /* 9645688Sbonwick * Flush the uberblocks to disk. This ensures that the odd labels 9655688Sbonwick * are no longer needed (because the new uberblocks and the even 9665688Sbonwick * labels are safely on disk), so it is safe to overwrite them. 967789Sahrens */ 9685688Sbonwick zio = zio_root(spa, NULL, NULL, flags); 969789Sahrens 9707754SJeff.Bonwick@Sun.COM for (int v = 0; v < svdcount; v++) 9715688Sbonwick zio_flush(zio, svd[v]); 972789Sahrens 9735688Sbonwick (void) zio_wait(zio); 9745688Sbonwick 9755688Sbonwick return (good_writes >= 1 ? 0 : EIO); 976789Sahrens } 977789Sahrens 978789Sahrens /* 9795688Sbonwick * On success, increment the count of good writes for our top-level vdev. 980789Sahrens */ 981789Sahrens static void 9825688Sbonwick vdev_label_sync_done(zio_t *zio) 983789Sahrens { 9845688Sbonwick uint64_t *good_writes = zio->io_private; 985789Sahrens 986789Sahrens if (zio->io_error == 0) 987789Sahrens atomic_add_64(good_writes, 1); 988789Sahrens } 989789Sahrens 9905688Sbonwick /* 9915688Sbonwick * If there weren't enough good writes, indicate failure to the parent. 9925688Sbonwick */ 993789Sahrens static void 9945688Sbonwick vdev_label_sync_top_done(zio_t *zio) 9955688Sbonwick { 9965688Sbonwick uint64_t *good_writes = zio->io_private; 9975688Sbonwick 9985688Sbonwick if (*good_writes == 0) 9995688Sbonwick zio->io_error = EIO; 10005688Sbonwick 10015688Sbonwick kmem_free(good_writes, sizeof (uint64_t)); 10025688Sbonwick } 10035688Sbonwick 10045688Sbonwick /* 10057041Seschrock * We ignore errors for log and cache devices, simply free the private data. 10066976Seschrock */ 10076976Seschrock static void 10087041Seschrock vdev_label_sync_ignore_done(zio_t *zio) 10096976Seschrock { 10106976Seschrock kmem_free(zio->io_private, sizeof (uint64_t)); 10116976Seschrock } 10126976Seschrock 10136976Seschrock /* 10145688Sbonwick * Write all even or odd labels to all leaves of the specified vdev. 10155688Sbonwick */ 10165688Sbonwick static void 10177754SJeff.Bonwick@Sun.COM vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) 1018789Sahrens { 1019789Sahrens nvlist_t *label; 1020789Sahrens vdev_phys_t *vp; 1021789Sahrens char *buf; 1022789Sahrens size_t buflen; 1023789Sahrens 10247754SJeff.Bonwick@Sun.COM for (int c = 0; c < vd->vdev_children; c++) 10257754SJeff.Bonwick@Sun.COM vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); 1026789Sahrens 1027789Sahrens if (!vd->vdev_ops->vdev_op_leaf) 1028789Sahrens return; 1029789Sahrens 10307754SJeff.Bonwick@Sun.COM if (!vdev_writeable(vd)) 1031789Sahrens return; 1032789Sahrens 1033789Sahrens /* 1034789Sahrens * Generate a label describing the top-level config to which we belong. 1035789Sahrens */ 10361635Sbonwick label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); 1037789Sahrens 1038789Sahrens vp = zio_buf_alloc(sizeof (vdev_phys_t)); 1039789Sahrens bzero(vp, sizeof (vdev_phys_t)); 1040789Sahrens 1041789Sahrens buf = vp->vp_nvlist; 1042789Sahrens buflen = sizeof (vp->vp_nvlist); 1043789Sahrens 10445688Sbonwick if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { 10455688Sbonwick for (; l < VDEV_LABELS; l += 2) { 10465688Sbonwick vdev_label_write(zio, vd, l, vp, 10475688Sbonwick offsetof(vdev_label_t, vl_vdev_phys), 10485688Sbonwick sizeof (vdev_phys_t), 10495688Sbonwick vdev_label_sync_done, zio->io_private, 10507754SJeff.Bonwick@Sun.COM flags | ZIO_FLAG_DONT_PROPAGATE); 10515688Sbonwick } 10525688Sbonwick } 1053789Sahrens 1054789Sahrens zio_buf_free(vp, sizeof (vdev_phys_t)); 1055789Sahrens nvlist_free(label); 1056789Sahrens } 1057789Sahrens 10585688Sbonwick int 10597754SJeff.Bonwick@Sun.COM vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) 1060789Sahrens { 10617754SJeff.Bonwick@Sun.COM list_t *dl = &spa->spa_config_dirty_list; 10625688Sbonwick vdev_t *vd; 10637754SJeff.Bonwick@Sun.COM zio_t *zio; 1064789Sahrens int error; 1065789Sahrens 10665688Sbonwick /* 10677754SJeff.Bonwick@Sun.COM * Write the new labels to disk. 10685688Sbonwick */ 10697754SJeff.Bonwick@Sun.COM zio = zio_root(spa, NULL, NULL, flags); 1070789Sahrens 10715688Sbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { 10725688Sbonwick uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), 10735688Sbonwick KM_SLEEP); 107410594SGeorge.Wilson@Sun.COM 107510594SGeorge.Wilson@Sun.COM ASSERT(!vd->vdev_ishole); 107610594SGeorge.Wilson@Sun.COM 10778632SBill.Moore@Sun.COM zio_t *vio = zio_null(zio, spa, NULL, 10787041Seschrock (vd->vdev_islog || vd->vdev_aux != NULL) ? 10797041Seschrock vdev_label_sync_ignore_done : vdev_label_sync_top_done, 10805688Sbonwick good_writes, flags); 10817754SJeff.Bonwick@Sun.COM vdev_label_sync(vio, vd, l, txg, flags); 10825688Sbonwick zio_nowait(vio); 10835688Sbonwick } 10847754SJeff.Bonwick@Sun.COM 10857754SJeff.Bonwick@Sun.COM error = zio_wait(zio); 1086789Sahrens 10875688Sbonwick /* 10885688Sbonwick * Flush the new labels to disk. 10895688Sbonwick */ 10905688Sbonwick zio = zio_root(spa, NULL, NULL, flags); 1091789Sahrens 10925688Sbonwick for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) 10935688Sbonwick zio_flush(zio, vd); 10944527Sperrin 10955688Sbonwick (void) zio_wait(zio); 1096789Sahrens 1097789Sahrens return (error); 1098789Sahrens } 1099789Sahrens 1100789Sahrens /* 11015688Sbonwick * Sync the uberblock and any changes to the vdev configuration. 1102789Sahrens * 1103789Sahrens * The order of operations is carefully crafted to ensure that 1104789Sahrens * if the system panics or loses power at any time, the state on disk 1105789Sahrens * is still transactionally consistent. The in-line comments below 1106789Sahrens * describe the failure semantics at each stage. 1107789Sahrens * 11085688Sbonwick * Moreover, vdev_config_sync() is designed to be idempotent: if it fails 1109789Sahrens * at any time, you can just call it again, and it will resume its work. 1110789Sahrens */ 11117754SJeff.Bonwick@Sun.COM int 11129725SEric.Schrock@Sun.COM vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) 1113789Sahrens { 11145688Sbonwick spa_t *spa = svd[0]->vdev_spa; 1115789Sahrens uberblock_t *ub = &spa->spa_uberblock; 11161635Sbonwick vdev_t *vd; 1117789Sahrens zio_t *zio; 11187754SJeff.Bonwick@Sun.COM int error; 11197754SJeff.Bonwick@Sun.COM int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; 1120789Sahrens 11219725SEric.Schrock@Sun.COM /* 11229725SEric.Schrock@Sun.COM * Normally, we don't want to try too hard to write every label and 11239725SEric.Schrock@Sun.COM * uberblock. If there is a flaky disk, we don't want the rest of the 11249725SEric.Schrock@Sun.COM * sync process to block while we retry. But if we can't write a 11259725SEric.Schrock@Sun.COM * single label out, we should retry with ZIO_FLAG_TRYHARD before 11269725SEric.Schrock@Sun.COM * bailing out and declaring the pool faulted. 11279725SEric.Schrock@Sun.COM */ 11289725SEric.Schrock@Sun.COM if (tryhard) 11299725SEric.Schrock@Sun.COM flags |= ZIO_FLAG_TRYHARD; 11309725SEric.Schrock@Sun.COM 1131789Sahrens ASSERT(ub->ub_txg <= txg); 1132789Sahrens 1133789Sahrens /* 11345688Sbonwick * If this isn't a resync due to I/O errors, 11355688Sbonwick * and nothing changed in this transaction group, 11365688Sbonwick * and the vdev configuration hasn't changed, 11371635Sbonwick * then there's nothing to do. 1138789Sahrens */ 11395688Sbonwick if (ub->ub_txg < txg && 11405688Sbonwick uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && 11417754SJeff.Bonwick@Sun.COM list_is_empty(&spa->spa_config_dirty_list)) 11427754SJeff.Bonwick@Sun.COM return (0); 1143789Sahrens 1144789Sahrens if (txg > spa_freeze_txg(spa)) 11457754SJeff.Bonwick@Sun.COM return (0); 1146789Sahrens 11471635Sbonwick ASSERT(txg <= spa->spa_final_txg); 11481635Sbonwick 1149789Sahrens /* 1150789Sahrens * Flush the write cache of every disk that's been written to 1151789Sahrens * in this transaction group. This ensures that all blocks 1152789Sahrens * written in this txg will be committed to stable storage 1153789Sahrens * before any uberblock that references them. 1154789Sahrens */ 11555688Sbonwick zio = zio_root(spa, NULL, NULL, flags); 11565688Sbonwick 1157789Sahrens for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; 11585688Sbonwick vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) 11595688Sbonwick zio_flush(zio, vd); 11605688Sbonwick 1161789Sahrens (void) zio_wait(zio); 1162789Sahrens 1163789Sahrens /* 1164789Sahrens * Sync out the even labels (L0, L2) for every dirty vdev. If the 1165789Sahrens * system dies in the middle of this process, that's OK: all of the 1166789Sahrens * even labels that made it to disk will be newer than any uberblock, 1167789Sahrens * and will therefore be considered invalid. The odd labels (L1, L3), 11685688Sbonwick * which have not yet been touched, will still be valid. We flush 11695688Sbonwick * the new labels to disk to ensure that all even-label updates 11705688Sbonwick * are committed to stable storage before the uberblock update. 1171789Sahrens */ 11727754SJeff.Bonwick@Sun.COM if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) 11737754SJeff.Bonwick@Sun.COM return (error); 1174789Sahrens 1175789Sahrens /* 11767754SJeff.Bonwick@Sun.COM * Sync the uberblocks to all vdevs in svd[]. 11771635Sbonwick * If the system dies in the middle of this step, there are two cases 11781635Sbonwick * to consider, and the on-disk state is consistent either way: 1179789Sahrens * 1180789Sahrens * (1) If none of the new uberblocks made it to disk, then the 1181789Sahrens * previous uberblock will be the newest, and the odd labels 1182789Sahrens * (which had not yet been touched) will be valid with respect 1183789Sahrens * to that uberblock. 1184789Sahrens * 1185789Sahrens * (2) If one or more new uberblocks made it to disk, then they 1186789Sahrens * will be the newest, and the even labels (which had all 1187789Sahrens * been successfully committed) will be valid with respect 1188789Sahrens * to the new uberblocks. 1189789Sahrens */ 11907754SJeff.Bonwick@Sun.COM if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) 11917754SJeff.Bonwick@Sun.COM return (error); 1192789Sahrens 1193789Sahrens /* 1194789Sahrens * Sync out odd labels for every dirty vdev. If the system dies 1195789Sahrens * in the middle of this process, the even labels and the new 1196789Sahrens * uberblocks will suffice to open the pool. The next time 1197789Sahrens * the pool is opened, the first thing we'll do -- before any 1198789Sahrens * user data is modified -- is mark every vdev dirty so that 11995688Sbonwick * all labels will be brought up to date. We flush the new labels 12005688Sbonwick * to disk to ensure that all odd-label updates are committed to 12015688Sbonwick * stable storage before the next transaction group begins. 1202789Sahrens */ 12037754SJeff.Bonwick@Sun.COM return (vdev_label_sync_list(spa, 1, txg, flags)); 1204789Sahrens } 1205